Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
The HIV-1 envelope protein gp120 is both the target of neutralizing antibodies and a major focus of vaccine efforts; however how it is delivered to B cells to elicit an antibody response is unknown . Here , we show that following local gp120 injection lymph node ( LN ) SIGN-R1+ sinus macrophages located in interfollicular pockets and underlying SIGN-R1+ macrophages form a cellular network that rapidly captures gp120 from the afferent lymph . In contrast , two other antigens , phycoerythrin and hen egg lysozyme , were not captured by these cells . Intravital imaging of mouse LNs revealed persistent , but transient interactions between gp120 bearing interfollicular network cells and both trafficking and LN follicle resident gp120 specific B cells . The gp120 specific , but not the control B cells repetitively extracted gp120 from the network cells . Our findings reveal a specialized LN antigen delivery system poised to deliver gp120 and likely other pathogen derived glycoproteins to B cells . The human immunodeficiency virus ( HIV-1 ) functional envelope spike is a trimer of non-covalently associated gp120/gp41 heterodimers , which are coated with N-linked carbohydrates that shield vulnerable protein surfaces from antibody recognition ( Bonomelli et al . , 2011; White et al . , 2011 ) . The host cell glycosylation pathways attach these carbohydrates ( Varki et al . , 2009 ) . However , the glycosylation processing of gp120 diverges from typical host glycoproteins resulting in densely packed patches of oligomannose glycans ( Doores et al . , 2010; Bonomelli et al . , 2011 ) . Such clusters do not occur on mammalian glycoproteins and , two such sites on the envelope , one associated with the first/second hypervariable loops ( V1/V2-glycan ) , and the other around the third hypervariable loop ( V3-glycan ) have served as targets for broadly neutralizing antibodies ( Bonomelli et al . , 2011; Raska et al . , 2014 ) . The glycan shield protects additional sites of viral vulnerability including the gp120 CD4 binding site and the envelope membrane proximal region ( Raska et al . , 2014 ) . The impact of the glycan shield on the uptake of gp120 by antigen presenting cells ( APCs ) and its subsequent delivery to B cells in lymph nodes ( LNs ) or the spleen is unknown . For B cells to mount an antibody response to an antigen such as gp120 they must encounter intact antigen . Since most B cells reside inside lymphoid follicles in the spleen , LNs , and at mucosal immune sites , most studies of LN antigen delivery have focused on the transport of antigen to the LN follicle and its subsequent loading onto follicular dendritic cells ( FDCs ) ( Pape et al . , 2007; Phan et al . , 2007; Batista and Harwood , 2009; Roozendaal et al . , 2009; Suzuki et al . , 2009; Cyster , 2010; Yuseff et al . , 2013 ) . FDCs retain antigen and are needed for the clonal selection of B cells with high affinity antigen receptors during germinal center reactions . Following local injection most antigens access the afferent lymph and are rapidly transported into the subcapsular sinus of the regional LN . Hen egg lysozyme ( HEL ) is a low molecular weight protein that can rapidly access LN follicle via the conduits ( Roozendaal et al . , 2009 ) . The conduits are an interconnected network of tubules that function as a molecular sieve allowing fluid and small molecules to enter the LN from the subcapsular sinus ( Gretz et al . , 1997 ) . However , gp120 is too large to enter the conduits as is phycoerythrin ( PE ) , a fluorescent non-glycosylated algae protein , whose delivery to FDCs has been examined as an antigen–antibody complex ( Phan et al . , 2007 ) . PE immune complexes are efficiently trapped by subcapsular sinus macrophages ( SSMs ) and delivered to FDCs in a complement dependent manner . Furthermore , cognate B cells residing in the follicle can acquire the antigen directly from the overlying SSMs . Keyhole limpet hemocyanin ( KLH ) is perhaps a better model for gp120 as it also heavily glycosylated , but similar to PE , KLH has been studied as an immune complex ( Roozendaal et al . , 2009 ) . While these studies have contributed to our understanding of FDC loading and germinal center responses , the kinetics of primary antibody responses do not favor naïve , recirculating B cell encountering high molecular weight antigens on FDCs ( MacLennan , 2007 ) . This suggests that another mechanism tailored to deliver a neo-antigen such as gp120 to cognate , naïve B cells might exist . One possibility is the SSMs that directly overlie the LN follicle . SSMs are CD169+CD11b+F4/80− and besides capturing immune complexes they also retain particulate material such as ferritin and liposomes ( Gray and Cyster , 2012 ) . Perhaps less likely are two other types of LN macrophage , medullary sinus macrophages ( MSMs ) and medullary cord macrophages ( MCMs ) . Like the SSMs , the MSMs are also CD169+CD11b+ , but they also express F4/80 and the pattern recognition receptors SIGN-R1 and MARCO . Their known functions are to clear particulates , pathogens , and dying cells ( Gray and Cyster , 2012 ) . The MCMs are CD169−CD11b+F4/80+April+ and they support plasma cell homeostasis . However , these macrophages are not very motile and localized far from most follicular and trafficking B cells . A better candidate is the interfollicular macrophages ( IFMs ) ( Gray and Cyster , 2012 ) . They are phenotypically similar to the MSMs , but they reside between the LN follicles in the interfollicular channel ( IFC ) , a site where early T-B cell collaboration occurs ( Kerfoot et al . , 2011 ) . High endothelial venules ( HEVs ) and cortical sinus lymphatics are located nearby ( Park et al . , 2009 ) . However , the functional role of IFMs in humoral immunity is poorly defined . LN resident dendritic cells ( DCs ) predominately sample the conduit contents making them an unlikely contender; however DCs in the vicinity of locally administered antigens can capture them , enter the afferent lymphatics , and access local LNs via the IFCs ( Qi et al . , 2006 ) . Yet such a mechanism is slow compared to the rapid antigen delivery via the lymph . Local DC-mediated antigen delivery is likely important for those antigens that do not enter the afferent lymph . To test how gp120 is captured in the LN we injected mice with fluorescently labeled gp120 near the inguinal LN . We followed the label using thick LN sections and confocal microscopy , and by intravital two-photon laser scanning microscopy ( TP-LSM ) . We also developed a gp120 overlay assay that allowed the identification of gp120 binding cells in lymphoid organ sections . To determine how cognate B cells acquire gp120 we adoptively transferred B cells from mice in which the variable portions of the human b12 neutralizing antibody were introduced into endogenous mouse Ig heavy and light chain loci by gene targeting ( Ota et al . , 2013 ) . The b12 antibody recognizes the CD4 binding site in gp120 ( Burton et al . , 1991; Roben et al . , 1994 ) . Following injection of fluorescently labeled gp120 we could track the acquisition of antigen by the gp120 specific B cells using intravital TP-LSM . Together these studies identified a group of macrophages that overlie the IFC and which extend to the cortical ridge and sinuses that bound and delivered gp120 to both re-circulating and follicle B cells . These IFMs are adjacent to , but distinct from the SSMs that overlie the LN follicle . We also identified a SIGN-R1 positive cell located in the splenic marginal zone that rapidly acquired blood borne gp120 . Our studies revealed an efficient mechanism for exposing trafficking naïve B cells to gp120 . For these studies we used an early HIV-1 viral isolate subtype A/C gp120 , R66M , expressed in 293F cells ( Nawaz et al . , 2011 ) , and injected 1 μg of fluorescently labeled gp120 near the base of the mouse tail . Confocal microscopy of thick LN sections prepared 2 hr after gp120 injection revealed that labeled gp120 had been captured by LN macrophages that overlie and extend into IFCs and that localize at the cortical medullary junction ( Figure 1A , top and middle panels ) . The asymmetric gp120 signal results from the gp120 accessing the afferent lymphatics serving the left side of the inguinal LN as it is oriented in the figure . Further immunostaining revealed that gp120 co-localized with SIGN-R1 ( Figure 1A , bottom panel ) , a c-type lectin and functional ortholog of DC-specific ICAM-3-grabbing non-integrin ( DC-SIGN ) , which has been implicated in HIV-1 transmission by human DCs ( Geijtenbeek et al . , 2000; Kang et al . , 2003 ) . Intravital TP-LSM revealed the rapid appearance of gp120 in the subcapsular sinus and identified the same subset of SIGN-R1+ macrophages capturing gp120 ( Figure 1B ) . Together the imaging and flow cytometry identified the gp120 binding cells as SIGN-R1+CD169midCD11bmidCD4+/CD11c−F4/80low sinus macrophages ( SIGN-R1+ subcapsular macrophages ) and SIGN-R1+/CD169midCD11blowCD4+CD11c−F4/80low IFM ( SIGN-R1+ IFC macrophages ) ( Figure 1C ) . These cells are to be distinguished from the SIGN-R1+CD11b+ DCs located in the medullary region , previously identified to uptake inactivated influenza virus ( Gonzalez et al . , 2010 ) . The SIGN-R1+ DCs also bound gp120 and are likely important for T cell priming ( Figure 1D ) . Next , we investigated the role of SIGN-R1 in gp120 binding . To do this we first checked whether gp120 bound in vitro to LN SIGN-R1+CD169midCD11b+ cells and whether unlabeled gp120 competitively inhibited the binding . We found that gp120 bound a phenotypically similar subset of macrophages and that unlabeled gp120 reduced the binding of the labeled material ( Figure 1E ) . When we added a SIGN-R1 blocking antibody with the labeled gp120 , the level of SIGN-R1 on the LN SIGN-R1+CD169midCD11b+ cells declined as did the fluorescent gp120 binding arguing that SIGN-R1 directly participated in the binding ( Figure 1E ) . The percentage of cells that bound gp120 declined by approximately 50% in the presence of the SIGN-R1 antibody . To directly visualize these cells in vitro , we sorted Gr-1−CD11c−SIGN-R1+CD11b+CD169+ cells from mice previously injected with fluorescent gp120 . The sorted cells were imaged ( Figure 1—figure supplement 1 ) . Because of the rarity of the cells in the LN population the sorted cells were contaminated with other cell types yet many gp120+SIGN-R1+ cells could be visualized . We also cultured the sorted cells with M-CSF . At day 7 the cultured cells were incubated with fluorescent gp120 and immunostained for SIGN-R1 . The majority of the cultured cells retained SIGN-R1 expression and most of these cells bound gp120 ( Figure 1—figure supplement 1 ) . To determine whether the uptake of gp120 triggered a biologic response in the IFC macrophages we injected gp120 locally and checked the intracellular interferon-γ levels in these cells ( Figure 1F ) . Some of the LN SIGN-R1+CD169midCD11b+ cells had an elevated level of intracellular interferon-γ compared to control cells . We verified these results using an interferon-γ eYFP reporter mouse ( Reinhardt et al . , 2009 ) . Flow cytometry was used to assess the percent of eYFP positive cells in the gated SIGN-R1+ macrophages ( Figure 1—figure supplement 2 ) , and to examine the induction of eYFP expression in various other cell populations in the immunized LN ( Figure 1—figure supplement 3 ) . To verify that the eYFP signal arose from the SIGN-R1+ macrophages we sorted the Gr-1−CD11c−SIGN-R1+CD11b+CD169+ cells and imaged them . We could readily identify SIGN-R1+eYFP+ cells , while the other contaminating cells present in the sorted population lacked YFP expression . Finally , we injected non-labeled gp120 near the inguinal lymph of the reporter mouse and 6 hr later made thick LN sections from the draining LN node and from a distant LN . Confocal microscopy revealed eYFP positive SIGN-R1+CD169mid cells in the IFC region of the draining LN , but similar cells were not present in the distant LN ( Figure1—figure supplement 3 ) . Together these results identified a subset of mouse subcapsular macrophages that overlie and reside in the IFC , which express SIGN-R1 and rapidly uptake gp120 . In addition , our data indicates that the local injection of gp120 likely elicits interferon-γ production by these cells . 10 . 7554/eLife . 06467 . 003Figure 1 . SIGN-R1 positive interfollicular channel ( IFC ) and cortical medullary junction macrophages rapidly accumulate lymph borne gp120 . ( A ) Confocal microscopy of thick lymph node ( LN ) sections prepared from mice that had received adoptively transferred B cells ( previous day ) , injected with fluorescently labeled gp120 , and immunostained as indicated . LN section image ( sagittal , tiled ) shows gp120 , green; CD169 , pink; CD21/35 , cyan; and B cells , red and blue . Scale bar is 200 μm ( top ) . A zoomed image of the white boxed area is shown . Scale bar is 60 μm ( middle ) . Images of an IFC are shown: gp120 , green; CD169 , red; SIGN-R1 , cyan; CD21/35 , blue; adoptively transferred B cells , pink; and CD169 , red ( bottom , left ) . SIGN-R1 signal removed ( bottom , right ) . Arrows indicate gp120 positive cells . Scale bars is 50 μm . ( B ) Intravital two-photon laser scanning microscopy ( TP-LSM ) images of the inguinal LN from a mouse injected with fluorescent gp120 and the indicated antibodies . The top images over the IFC show gp120 , white; CD169 , red; F4/80 , green; and adoptively transferred B cells , blue , ( left panel ) . SIGN-R1 antibody , red , used instead of CD169 ( right panel ) . Scale bars are 100 μm . The bottom images are from the follicular-medullary junction and show gp120 , white; SIGN-R1 , red; F4/80 , green; and adoptively transferred B cells , blue . Scale bar is 50 μm . ( C , D ) Flow cytometry of LN cells immunostained and gated as indicated using inguinal LN cells from a mouse injected with fluorescent gp120 1 . 5 hr previously , or not . LiveGr-1−CD11c− gated population plotted for F4/80 vs CD11b . Gates ‘a’ , ‘b’ , and ‘c’ as indicated were re-plotted to show SIGN-R1 vs gp120 in right three plots ( top 2 rows ) . ( C ) . LiveGr-1−CD11c+ population is shown plotted for SIGN-R1 vs CD11b . Histogram of indicated three populations ( a , b , and c ) plotted as gp120 signal ( black line ) vs % of maximum intensity compare to gp120 negative control ( shaded ) . Numbers are % gp120 positive cell population in gate ( D ) . ( E ) In vitro binding by LN cells incubated with fluorescent gp120 , or not , and in the presence of non-labeled gp120 or non-labeled SIGN-R1 antibody ( different epitope ) and then analyzed by flow cytometry . LiveGr-1−CD11c−CD11b+ cells were analyzed for gp120 vs SIGN-R1 ( not shown ) and the CD11b+ cells , gray contour; the SIGN-R1+gp120+ cells , green dots; SIGN-R1−gp120− cells , black dots; and SIGN-R1+gp120− cells , gray dots , were plotted to show CD169 vs SIGN-R1 . ( F ) Interferon-γ intracellular flow cytometry of cells prepared from the inguinal LNs of mice administered gp120 near the tail base , or not , 3 hr prior to collection . LiveGr-1−CD11c−CD11b+ cells were analyzed for SIGN-R1 vs CD169 and separated into three populations ( left panels ) . The levels of intracellular interferon-γ are shown as histograms of maximum intensity in cells from the gp120 non-exposed ( gray ) and gp120 injected mice ( white , outlined by black lines ) . Unstained control is delineated by a gray line . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 00310 . 7554/eLife . 06467 . 004Figure 1—figure supplement 1 . Sorted SIGN-R1+ macrophages capture gp120 . ( A ) Confocal microscopy image of FACS sorted SIGN-R1+ macrophage 2 hr after sorting . Differential interference contrast ( DIC ) visualized the cell body and nucleus and was used as a background ( left ) . gp120 ( green ) and SIGN-R1 ( red ) signals were overlapped with DIC ( middle ) . gp120 ( green ) and SIGN-R1 ( red ) signals were visualized without DIC ( right ) . Scale bars are 10 μm . ( B ) Confocal microscopy image of FACS sorted SIGN-R1+ macrophage , which were cultured with 20 ng/ml of M-CSF for 7 days . Cells were fixed with 4% paraformaldehyde for 2 hr and overlaid with fluorescent gp120 ( green ) . The cells were washed and immunostained with SIGN-R1 antibody ( red ) . The fluorescent gp120 signal was overlapped with DIC ( left ) . SIGN-R1 signal was overlapped with DIC ( middle ) . Merged signal ( yellow ) was overlapped with DIC ( right ) . Scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 00410 . 7554/eLife . 06467 . 005Figure 1—figure supplement 2 . The injection of gp120 triggers transcription of interferon- γ in SIGN-R1+ macrophages as assessed by using a interferon-γ-eYFP reporter mouse . Flow cytometric analysis of cells prepared from the inguinal LNs of interferon-γ-eYFP reporter mice administered gp120 near the tail base at 0 , 3 and 6 hr prior to collection . eYFP vs SIGN-R1 expression in cell located in gates a , b and c are shown via a contour plot . WT mice were used as a negative control for eYFP expression in each of the gates . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 00510 . 7554/eLife . 06467 . 006Figure 1—figure supplement 3 . The injection of gp120 triggers transcription of interferon- γ in SIGN-R1+ macrophages . ( A ) Flow cytometric analysis of various cellular populations prepared from the inguinal LNs of interferon-γ-eYFP reporter mice administered gp120 near the tail base , 0 ( shaded ) , 3 ( shaded , darker ) and 6 ( black line ) hs prior to collection . Histograms of indicated populations were plotted as eYFP signal vs % of maximum intensity . ( B ) Histogram of NK cells ( black line , first to third graph from left ) and NKT cells ( black line , forth to sixth graph from left ) were re-plotted with histogram of ‘Gate b’ to compare eYFP signal at indicated time points . ( C ) Confocal microscopy image of FACS sorted SIGN-R1+ macrophage at 0 . 5 hr after sorting . Differential interference contrast ( DIC ) ( gray ) visualized cell body as a background ( left ) . eYFP ( green ) and SIGN-R1 ( red ) signals were overlapped with DIC ( left ) . eYFP ( green ) signals were visualized with DIC ( middle ) . eYFP ( green ) and SIGN-R1 ( red ) signals were visualized without DIC ( right ) . Scale bars are 10 μm . ( D ) Confocal microscopy of a thick LN section from the draining LN of a mouse previously injected with gp120 6 hr previously ( upper panels ) and a thick section of a cervical LN ( far from the site of gp120 injection ) obtained at the same time point ( lower panels ) . Sections were immunostained for SIGN-R1 ( red ) and CD169 ( blue ) . Arrows in panel ( upper , left ) indicate eYFP ( green ) expressed in SIGN-R1+ macrophages . eYFP , SIGN-R1 and CD169 signals were overlapped in the left panels . SIGN-R1 only and eYFP only signals were shown in middle and right panels , respectively . Scale bar is 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 006 The IFC connects the subcapsular sinus to the cortical ridge at the boundary of the B and T cell zones . By 3 hr after injection gp120 labelled processes extended into the IFC making contact with B and T lymphocytes ( Figure 2A ) . In addition , IFC DCs could be found associated with the SIGN-R1+ IFC cells bearing gp120 ( Figure 2—figure supplement 1 ) . The likely involvement of SIGN-R1 in the uptake of gp120 by the LN macrophages prompted us to examine whether the SIGN-R1 positive cell known to reside in the marginal zone region of the spleen could uptake gp120 from the blood ( Kang et al . , 2003 ) . Whether these SIGN-R1 positive marginal zone cells are macrophages or resident DCs has been debated ( Lyszkiewicz et al . , 2011 ) , however they are generally referred to as SIGN-R1+ marginal zone macrophages . Following gp120 injection into the blood we observed a subset of marginal zone cells that rapidly acquired the gp120 signal ( Figure 2B ) . They expressed high levels of SIGN-R1 and lacked CD11c ( data not shown ) . Confocal microscopy of thick spleen sections immunostained with CD169 , CD21 , and SIGN-R1 and overlaid with fluorescent gp120 identified a similar marginal zone SIGN-R1 positive cell ( Figure 2C ) . A tiled confocal image of a portion of the spleen from the gp120 overlay is also shown ( Figure 2D ) , which demonstrates a remarkable overlap between the SIGN-R1 and gp120 signals . These results indicate that there is a network of SIGN-R1 positive macrophages in the LN IFC that provide a platform for nearby B cells and DCs to acquire gp120 and that a subset of SIGN-R1+ cells in the marginal zone of the spleen are also poised to deliver gp120 to splenic marginal zone and trafficking follicular B cells . 10 . 7554/eLife . 06467 . 007Figure 2 . IFC cell processes bearing gp120 directly contact B cells and a subset of splenic marginal zone cells also bind gp120 . ( A ) Confocal microscopy of a thick LN section from a mouse previously injected with fluorescent gp120 and immunostained for B220 and CD3 . Scale bar is 30 μm . Boxed areas in left image were enlarged and shown in the right panels . Scale bars are 10 μm . ( B ) Confocal microscopy of a thick splenic section from a mouse previously injected intravenously with fluorescent gp120 and immunostained with the indicated markers . Zoomed images shown below . Scale bars are 100 μm , above , and 30 μm , below . ( C ) Confocal microscopy of a thick splenic section overlaid with fluorescent gp120 and immunostained for the indicated markers . Zoomed images are shown below . Scale bars are 100 μm , above , and 40 μm , below . ( D ) Tiled confocal microscopy images of a spleen section immunostained with CD169 ( red ) and SIGN-R1 ( cyan ) and overlaid with fluorescent gp120 ( green ) . As indicated the images show CD169 alone , CD169 and SIGN-R1; CD169 and gp120; and overlay of all three . Scale bar is 300 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 00710 . 7554/eLife . 06467 . 008Figure 2—figure supplement 1 . The IFC network macrophages contact CD11c positive cells . Confocal microscopy of a thick LN section from a mouse previously injected with fluorescent gp120 and immunostained for CD169 and CD11c . The image is centered over an IFC . Scale bar is 30 μm . Boxed areas in left image were enlarged and shown in the right panels . Scale bars are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 00810 . 7554/eLife . 06467 . 009Figure 2—figure supplement 2 . Overlay of gp120 visualizes DC-SIGN+/CD163+ macrophages in a human LN section . Confocal microscopy of human frozen LN sections overlaid with fluorescent gp120 and immunostained for the indicated markers . ( A ) Section was stained with CD19 ( blue ) and CD4 ( gray ) . Scale bar is 400 μm . ( B ) Adjacent section from ( A ) section was stained with CD163 ( green ) , DC-specific ICAM-3-grabbing non-integrin ( DC-SIGN ) ( red ) and CD11c ( blue ) . Scale bar is 200 μm . ( C ) Zoomed image of the white box in ( B ) . Arrows indicated overlap ( yellow ) of CD163+ cells ( green ) and DC-SIGN+ cells ( red ) . Scale bar is 100 μm . ( D ) Same area with ( C ) . Arrow indicated overlap ( yellow ) of gp120+ cells ( green ) and DC-SIGN+ cells ( red ) . Scale bar is 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 009 The ability to specifically detect gp120 binding cells using an overlay assay prompted us to determine whether we could detect similar macrophages in human LN . To identify the LN follicles and T cell zone we immunostained a LN section for CD19 and CD4 expression . Using 2 adjacent sections we immunostained one for CD163 , a human macrophage marker ( Martens et al . , 2006 ) , CD11c , and DC-SIGN; and the other for gp120 and DC-SIGN . This allowed the identification of a group of CD163+ , DC-SIGN+ , and CD11c− cells near the LN follicle that bound the overlaid gp120 ( Figure 2—figure supplement 2 ) . These results indicate that in human LN DC-SIGN expressing macrophage near the follicle may uptake gp120 similar to the mouse IFC SIGN-R1+ macrophages . To determine the kinetics of gp120 uptake by SIGN-R1+ subcapsular macrophages and the underlying SIGN-R1+ IFC macrophages , we intravitally imaged for 3 . 5 hr following gp120 injection . The amount of gp120 associated with the SIGN-R1+ subcapsular macrophages gradually increased and then declined , while the underlying network cells incrementally increased their gp120 binding eventually surpassing the sinus macrophages ( Figure 3A , B , Videos 1 , 2 ) . Cellular processes labeled with gp120 were visualized extending into the LN follicle . In one instance an endogenous cell ( weakly fluorescent ) approached and contacted a gp120 labeled cellular process ( Video 3 ) . The gp120 identified IFC cellular network extended to the cortical ridge and even into the cortical sinuses ( Figure 3C ) . These results reveal the rapid gp120 loading of the SIGN-R1+ IFC macrophages and suggest that a transport mechanism may exist to transfer gp120 from the superficial to the underlying cells in the IFC . 10 . 7554/eLife . 06467 . 010Figure 3 . The IFC network of macrophages extends from the subcapsular sinus to the cortical sinus . ( A ) Intravital TP-LSM images of the inguinal LN 8 min after injection of fluorescent gp120 and 30 min after CD169 . Yellow circles indicate sinus macrophages ( upper panel ) or network cells in the IFC ( lower panel ) . Blue circles indicate subcapsular sinus lumen . Distance between two slices is 30 μm . Scale bars are 50 μm . ( B ) The gp120 signal associated with SIGN-R1+ SM ( ○ ) or deeper in the IFC ( ⃞ ) was quantitated over time following gp120 injection: 8 min-1 hr 14 min , upper; 1 hr 30 min–2 hr 30 min , middle; and 3 hr–3 hr 32 min , bottom panel . Calculated slopes ( ○ , ⃞ ) are 1 . 59 ± 0 . 046 and 0 . 45 ± 0 . 013 , upper; −1 . 04 ± 0 . 16 and 0 . 63 ± 0 . 11; middle , and −5 . 61 ± 0 . 23 and 2 . 35 ± 0 . 33; bottom panel . In the same panel the slopes differed by a p value <0 . 0001 . The reference signal of gp120 in subcapsular sinus lumen is shown with dotted lines . Error bars , ±SEM ( B ) . ( C ) Intravital TP-LSM images from deep in the IFC following injection of fluorescent gp120 and LYVE-1 antibody . The left upper panel show an image from a 30 μm z-projection . Scale bar is 100 μm . Dotted box was reconstructed using the LYVE-1 signal , green , and gp120 signal , red ( upper middle panel ) and modified by adding low intensity gp120 signal , white ( upper right panel ) . Scale bars are 50 μm . The left lower panel shows a higher power 20 μm z-projection image . Scale bar is 50 μm . Dotted box was reconstructed using the LYVE-1 signal , green , and gp120 signal , red ( lower middle panel ) and modified by adding the low intensity gp120 signal , white ( lower right panel ) . Scales bars are 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01010 . 7554/eLife . 06467 . 011Video 1 . Intravital two-photon laser scanning microscopy ( TP-LSM ) images of the interfollicular channel ( IFC ) network that extends from the subcapsular sinus to the cortical sinus . Images from two focal planes separated by 30 μm and located over the IFC channel of a mouse inguinal lymph node ( LN ) . The mouse had previously been injected with fluorescent CD169 antibody , red , which delineates the subcapsular sinus macrophages ( SSMs ) . The images were acquired over an hour ( 8 min–1 hr 8 min post fluorescent gp120 , white , injection into the tail base ) . The white lines delineate the subcapsular sinus . Scale bar is 50 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01110 . 7554/eLife . 06467 . 012Video 2 . Intravital TP-LSM images of the gp120 loaded cellular network in the IFC of the inguinal LN . An image sequence of a 30 μm z-projection was acquired from a LysM-EGFP mouse , which had previously received by adoptive transfer both B cells , blue , and CD4 T cells , purple . Host endogenous neutrophils/monocytes , strong green signal , and stromal cells , weak green signal , can be seen on the basis of their expression of LysM-EGFP . Images were acquired for an hr from 1 . 5–2 . 5 hr after fluorescent gp120 , white , injection near the tail base . GFP positive cells can be seen flowing in blood vessels in the IFC . CD169 antibody , red , delineated the SSMs . Second harmonic signal , blue , from collagen delineated LN capsule . Scale bar is 100 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01210 . 7554/eLife . 06467 . 013Video 3 . Intravital TP-LSM images of the dynamic interaction of SIGN-R1+ gp120+ macrophages and a lymphocyte . An image sequence of a 20 μm z-projection was acquired from the inguinal LN following nearby injection of fluorescent gp120 , red , and SIGN-R1 , green , antibody ( left ) . Track and displacement ( yellow arrow ) of lymphocytes ( gray spot ) superimposed with 3D-reconstructed images of cell process is gray ( right ) . Scale bars are 20 μm and 10 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 013 We checked the specificity of this cellular network by comparing gp120 to two other proteins; Alexa-488 labeled HEL , which because of its low lower molecular mass should enter the conduits ( Roozendaal et al . , 2009 ) , and 4-Hydroxy-3-nitrophenylacetyl ( NP ) modified PE , a large non-glycosylated fluorescent algae protein . NP-PE predominately targeted the sinus lining cells while as expected Alexa-488-HEL largely entered the conduit system ( Figure 4A , B ) . The first region of interest ( ROI-1 ) overlies the LN follicle and shows the sinus lining cells have taken up NP-PE , while the same cells have little gp120 ( Figure 4B ) . Conduits filled with Alexa-488-HEL ( blue ) , which lack gp120 and NP-PE , penetrated into the LN follicle . The ROI overlying the IFC ( ROI-2 ) shows strong gp120 positivity , some sinus lining cells are NP-PE positive , but appear distinct from the gp120 positive cells ( Figure 4B ) . This is evident from the analysis of the two ROI defined within ROI-2 . The third , ROI-3 , connects the IFC to a lymphatic sinus . Numerous solely gp120 positive cells are present , while conduits and likely fibroreticular and lymphatic cells are outlined by HEL uptake . Again we detected little overlap between NP-PE bearing cells and those bearing gp120 ( Figure 4B ) . Long cellular processes labeled by the presence of fluorescent gp120 can be visualized extending into the IFC making contact with other cells ( Figure 4C ) . Three-dimensional reconstruction of the imaging data shows that gp120 resides both within and on the surface of the IFC macrophages ( Figure 4C , right panel ) . The IFC macrophage processes and gp120 signal often wrapped around the conduits and fibroreticular cells ( Figure 4D , E ) . These data highlight three different mechanisms of antigen uptake by LN cells , which depend upon the size and composition of the antigen . 10 . 7554/eLife . 06467 . 014Figure 4 . IFC network macrophages do not uptake hen egg lysozyme ( HEL ) or nitrophenylacetyl ( NP ) -phycoerythrin ( PE ) . ( A , B ) Confocal microscopy of a thick LN section immunostained for CD4 , white , following injection of NP-PE , red , fluorescent gp120 , green , and fluorescent HEL , blue , near the inguinal LN . Three regions of interest are shown over LN follicle region of interest ( ROI-1 ) , superficial IFC ( ROI-2 ) , and deep IFC ( ROI-3 ) . Scale bar is 40 μm ( A ) . In part B each ROI is further subdivided as indicated by letters to delineate specific cells or groups of cells . The fluorescent intensity of NP-PE or gp120 in each of these regions was quantitated and is indicated . Numbers in graphs indicate fold difference . ( C ) Confocal microscopy image of a thick LN section from a mouse previously injected with fluorescent gp120 , green , plus NP-PE , red , and immunostained for CD169 , white , and CD4 , blue . CD4 is excluded in the 2nd–4th panels . Electronically zoomed image of boxed area is shown in 3rd panel . A 3-D reconstruction of the 3rd panel image is shown in the 4th panel . Scale bars from left to right are 100 , 100 , 50 , and 50 μm . ( D ) Confocal microscopy image of a thick LN section immunostained for CD4 and CD169 prepared from a mouse injected near the inguinal LN with fluorescent gp120 and NP-PE . The middle image is an electronically zoomed image of the region in the left panel . A portion of the middle image was used to perform a 3-D reconstruction of the imaging data shown in the right panel . The scale bars from left to right are 25 , 10 , and 5 μm . ( E ) Confocal microscopy image of a LN section immunostained for ERTR-7 , white , and CD169 , red , from a mouse previously injected with fluorescent gp120 , green , focusing on the IFC . The indicated portion of left panel was used for the 3-D reconstruction shown in the right panel . The scale bars from left to right are 30 and 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 014 SIGN-R1 selectively recognizes α-2 , 6-sialylated glycoproteins ( Silva-Martin et al . , 2015 ) . Oligosaccharides present on the envelope of various viruses including HIV contain terminal α-2 , 6 sialic acid linkages . To assess whether differences in the carbohydrate structure of gp120 affected its uptake by this cellular network we labeled CHO-S or 293F expressed R66M gp120 with Alexa-488 ( green ) or Alexa-594 ( red ) . CHO-S expressed proteins have a heterogeneous pattern of oligo-mannose and complex carbohydrate type glycans while 293F expressed proteins have mostly complex carbohydrate type glycans . Furthermore , 293F cells sialylate the terminal galactose moieties of complex carbohydrates using −2 , 3 and −2 , 6 linkages while CHO-S cells only use the −2 , 3 linkage ( Nawaz et al . , 2011 ) . By injecting both labeled gp120 simultaneously we could follow and compare the acquisition of gp120 by the network cells . We found both gp120 preparations accessed the previously defined cellular network; however , the 293F derived gp120 bound the SIGN-R1+ IFC macrophages better than did the CHO-S derived gp120 , irrespective of how it was labeled ( Figure 5A–C ) . Perhaps as a consequence the antibody response to 293F derived gp120 exceeded the response to CHO-S gp120 ( Figure 5D ) . 10 . 7554/eLife . 06467 . 015Figure 5 . IFC network macrophages differentially uptake two different R66M gp120 preparations . ( A ) Intravital TP-LSM image of the inguinal lymph following the injection of differentially labeled R66M gp120 expressed in either 293F or CHO-S cells . Scale bars are 100 μm . ( B ) A z-projection ( 50 μm ) of intravital TP-LSM images of the inguinal LN following the injection of differentially labeled R66M gp120 expressed in either 293F cells , red , or CHO-S cells , green , and SIGN-R1 antibody , white . Adoptively transferred B cells ( blue ) and the 2nd harmonic signal delineated the LN follicle and capsule . The various signals shown are indicated . Scale bars are 50 μm . ( C ) Level of gp120 binding to SIGN-R1+ cells was quantitated . The amount of 293 gp120 and CHO-gp120 bound was determined using Imaris . *p < 0 . 001 . ( D ) Results from ELISA assays to analyze gp120 specific antibodies present in the sera of mice at various days following immunization with R66M gp120 expressed in either 293F or CHO-S cells . Error bars , ±SEM . ( E ) A z-projection ( 50 μm ) of intravital TP-LSM images of the inguinal LN following the local injection of R66M gp120 expressed in 293F cells , green , or CHO-S cells , red . The mice had been immunized with 293-gp120 and boosted 4 weeks later . 3 weeks after the boost the mice were injected near the inguinal LN with labeled gp120s . The day prior to the injection naive B cells ( blue ) were adoptively transferred . The LN follicle images are from 4 hr after gp120 injection . Scale bar are 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01510 . 7554/eLife . 06467 . 016Figure 5—figure supplement 1 . Deglycosylated gp120 loses its binding specificity to SIGN-R1+ macrophages . Confocal microscopy of thick LN sections overlaid with fluorescent gp120 , which is treated with PNGase F or not , and immunostained for the indicated markers . ( A ) LN section image ( sagittal , tiled ) shows control gp120 , green; CD169 , cyan; CD21/35 , gray; F4/80 , blue and SIGN-R1 , red . Scale bar is 200 μm ( top ) . ( B ) LN section image ( sagittal , tiled ) shows deglycosylated gp120 , green; CD169 , cyan; CD21/35 , gray; F4/80 , blue and SIGN-R1 , red . Scale bar is 200 μm ( top ) . Zoomed images of the white boxed area in ( A ) and ( B ) were shown in middle and bottom panels . Image in middle left shows gp120 , green; CD169 , cyan; CD21/35 , gray; F4/80 , blue and SIGN-R1 , red . Image in middle right shows gp120 , green and CD169 , cyan . Image in bottom left shows CD169 and SIGN-R1 , red . Image in bottom right shows the generated channel ( yellow ) of colocalization between gp120 and SIGN-R1 . Scale bars are 100 μm . ( C ) Percentage of ROI colocalized area in bottom left panel in ( A ) and ( B ) were compared in graph . ( D ) The strength of fluorochrome after PNGase F treatment was measured by Odyssey CLx Infrared Imaging System . The numbers on top of each lanes indicated the amount of loaded gp120 . The numbers on bottom of each lanes indicated the relative score of strength of fluorochrome on gp120 . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01610 . 7554/eLife . 06467 . 017Figure 5—figure supplement 2 . SIGN-R1+ IFC and cortical medullary junction macrophages rapidly accumulate lymph borne soluble trimeric gp120 . Confocal microscopy of thick LN sections prepared from mice that had been injected with biotinylated soluble trimeric gp120 . Biotinylated gp120 was detected by AlexaFluor 488 conjugated streptavidin . In the LN section image ( sagittal , tiled ) trimeric gp120 , green; CD169 , cyan; CD21/35 , gray; F4/80 , blue and SIGN-R1 , red , are shown . Scale bar is 200 μm ( top ) . Zoomed images of the white boxed areas are shown in the middle and bottom panels . Scale bars are 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 017 To check and compare the FDC network loading of the two gp120 preparations we waited 3 weeks after boosting mice previously immunized with 293F gp120 and injected both 293F and CHO-S expressed gp120 near the inguinal LN . Using intravital TP-LSM we checked both the loading of the IFC macrophages and the FDC network in a nearby follicle . As we had previously noted the IFC macrophages rapidly captured the lymph borne gp120 , however , in contrast to the naïve mouse within 4 hr of injection both gp120 preparations had loaded on to the FDC network of a germinal center ( Figure 5E ) . The germinal center region was outlined by adoptively transferred naïve B cells . The strongly fluorescent cells in the germinal center are likely tingible body macrophages . To more drastically change the carbohydrate structure of gp120 , we treated the purified 293F gp120 with PNGaseF , an amidase that cleaves between the innermost GlcNAc and asparagine residues of high mannose , hybrid , and complex oligosaccharides . The untreated gp120 and the PNGaseF treated gp120 were used in a mouse LN overlay assay . As previously the untreated gp120 bound strongly to the SIGN-R1 positive macrophages , while the PNGase F treated gp120 exhibited much less selectivity , binding to many different cell types . This resulted in a reduction in the co-localization between SIGN-R1 and gp120 . We verified that the PNGase F treatment appropriately altered the molecular mass of gp120 ( Figure 5—figure supplement 1 ) . Since the trimeric version of gp120 offers different antigenic determinants and is rapidly becoming accepted as the preferred vaccine candidate ( Sanders et al . , 2013 ) , we also checked whether a trimeric version of gp120 was captured by the SIGN-R1+ macrophages overlying the IFC . We found a very similar pattern of uptake as we had observed with the monomeric gp120 ( Figure 5—figure supplement 2 ) . Together these results indicate that the carbohydrate structure of gp120 can affect the binding of gp120 to the SIGN-R1 positive macrophages , which may affect the subsequent antibody response . However minor differences in the glycosylation of gp120 did not impact FDC loading in the setting of gp120 specific antibody . Next , we monitored the interaction of B cells with the gp120 bearing cells in the LN by adoptively transferring b12 knock-in B cells that possess gp120 reactive antigen receptors . The knock-in B cells bearing the H & L , H chain , and L chain genes bind gp120 with high , low , and no detectable affinity , respectively ( Burton et al . , 1991; Ota et al . , 2013 ) . Previously 90% of all mature b12 HL B cells bound soluble trimers of HIV Env ( the JRFL isolate ) , as measured by flow cytometry . Similar to the results with the soluble trimers of the JRFL envelope , labeled R66M gp120 bound more than 90% of the splenic follicular and marginal zone b12 HL B cells ( Ota et al . , 2013 ) ( Figure 6A ) . Of note , the average gp120 mean fluorescent intensity on marginal zone B cells exceeded that on follicular B cells by twofold . Next we transferred gp120 specific B cells 2 hr after the injection of gp120 to assess the delivery of gp120 to newly arriving HIV-1 specific B cells . The tracks of the newly arriving b12 HL B cells and b12 L B cells predominately localized in the IFC although the b12 HL B cells focused preferentially on the gp120 bearing network cells ( Figure 6B ) . The motility patterns of the b12 HL and b12 L B cells differed . Although the average velocities were similar , the b12 HL B cells moved less straight with more speed variability and exhibited greater displacements than the b12 L B cells ( Figure 6C ) . The b12 HL B cells interacted vigorously with the gp120 bearing cells extracting gp120 from the network cells , which accumulated in their uropods ( Figure 6D , Video 4 ) . By loading the b12 HL B cells with the Ca2+ sensitive dye , Calcium Orange , we could observe transient increases in intracellular Ca2+ as the b12 HL B cells interacted with , and extracted gp120 from the IFC cells ( Figure 6E , Video 5 ) . The intracellular Ca2+ rise often occurred in conjunction with an increase in gp120 in the b12 HL B cells ( Figure 6E ) . These results indicate that newly arriving recirculating B cells can acquire antigen from the IFC macrophages bearing gp120 . Similarly naïve B cells will have the opportunity to encounter cognate antigen on IFC cells as they exit the LN follicle . 10 . 7554/eLife . 06467 . 018Figure 6 . Recently arrived LN B cells that express the b12 antigen receptor can extract gp120 from IFC network cells . ( A ) Flow cytometry to evaluate the binding of labeled R66M gp120 to follicular and marginal zone B cells from either b12 HL or b12 L B cells . Light gray line is background fluorescence . MFI-mean fluorescence intensity . ( B ) Tracks of b12L and b12 HL B cells in the IFC after adoptive transfer to a mouse previously injected with fluorescent gp120 . The tracks are superimposed on the IFC network delineated by gp120 . Scale bar is 100 μm . ( C ) Comparison of the motility parameters generated from the analysis of b12 L and b12 HL B cells adoptively transferred into mice previously injected with gp120 . Statistics calculated using unpaired t-test , p < 0 . 01 , **p < 0 . 001 . ( D ) Time laps images of a fluorescently labeled b12 HL B cell , red , approaching and departing from a cell in the IFC network that has accumulated gp120 , green . Scale bar is 10 μm . Time stamps in top right , min:s . ( E ) Intravital TP-LSM imaging of b12 HL B cell labeled with eFluor 450 , blue , and Calcium Orange , red , as it approaches and interacts with gp120 expressing cells , green , in the IFC . In graph , signal intensity ratio was plotted as a function of time . The ratio was calculated by dividing the intensity of an individual point by the average intensity of all the time points . Reference ratio calculated using the eFluor 450 b12 HL B cell signal . Arrows from 1 to 10 correspond to the numbered time lapse images shown below . Scale bar is 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01810 . 7554/eLife . 06467 . 019Video 4 . Intravital TP-LSM images of a newly arriving b12 B cell extracting gp120 from IFC network cells . An image sequence of a 20 μm z-projection was acquired from the inguinal LN of a mouse , which had fluorescent gp120 ( green ) delineated IFC network cells . Fluorescently labeled b12 HL B cells ( red ) were adoptively transferred an hour after gp120 injection . Scale bar is 30 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 01910 . 7554/eLife . 06467 . 020Video 5 . Intravital TP-LSM images of in vivo calcium response of a b12 B cell engaging a gp120 bearing IFC cell . An image sequence of a 25 μm z-projection was acquired from the inguinal LN of a mouse , which has fluorescent gp120 ( green ) delineated IFC network cells . b12 HL B cells ( blue ) were labeled with calcium orange ( red ) . Shaded circle indicates the interaction between b12 HL B cells and gp120 loaded IFC network cells . Images were acquired for 40 min beginning 41 min after fluorescent gp120 injection into tail base . Scale bar is 15 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 020 To determine whether B cells resident in LN follicles might also acquire cognate antigen from the IFC network , we adoptively transferred b12 HL and wild type ( WT ) B cells to recipient mice the day prior to gp120 injection . This allowed the B cells to localize in the follicle . Then , we injected gp120 and over the next 2 hr monitored its uptake by the IFC network and the behavior of B cells in the LN follicle . As noted previously the IFC network cells rapidly accumulated gp120 and within 30 min fine cellular processes bearing gp120 became visible along the follicle edge ( Figure 7A ) . Tracking the transferred B cells near the follicle edge revealed that the b12 HL B cells moved slower and tended to remain in the imaging space longer resulting in a longer track lengths and increased displacements ( Figure 7B ) . As a consequence b12 HL B cells accumulated at these sites while the control B cells did not ( Figure 7C ) . The b12 HL B cells made numerous , transient interactions with the gp120 bearing cellular processes ( Video 6 ) . Tracking individual b12 HL B cells revealed that the B cells slowed as they extracted antigen from the network , after which they sped up and the cell associated gp120 signal declined , whereupon they re-engaged the network and extracted more gp120 ( Figure 7D ) . Interestingly , the average velocities of both the WT and the gp120 specific B cells increased over time following the gp120 injection ( Figure 7E ) . Detailed analyses of six tracks from B cells located in the LN follicle and near the intrafollicular channel are shown ( Figure 8 ) . These results show that antigen loaded IFC can provide a source of antigen for cognate B cells traversing the edge of the LN follicle and perhaps provide signals that enhance B cell motility in the follicle . 10 . 7554/eLife . 06467 . 021Figure 7 . LN follicle B cells that express the b12 antigen receptor can extract gp120 from IFC network cells . ( A ) Intravital TP-LSM images of b12 HL , red , and WT B , blue , cells in the inguinal LN at 3 , 35 , and 63 min following injection of fluorescent gp120 , green , near the inguinal LN . Blood vessels were visualized by intravenous injection of Evans blue , white . Scale bar is 100 μm . Bellow each image is an electronic zoomed image from the indicated area . White arrowheads indicate B cells that have accumulated gp120 and green arrowheads gp120 in the LN follicle . Scale bar is 25 μm . ( B ) Motility parameters . Analyses of b12 HL and wild type ( WT ) tracks are shown . Statistics are by unpaired t-test p < 0 . 01 , p < 0 . 001 . ( C ) The ratio between the number of WT and b12 HL B cells at various times points following antigen injection near the IFC or in the center of follicle . Error bar , ±SEM . p < 0 . 002 . ( D ) Tracking a b12 HL B cell located in the LN follicle following injection of gp120 . Displacement and gp120 signal overlying the cell tracked over time . Graph shows the displacement , black line , from the origin and gp120 signal , green peaks , for each individual time point . The reference , thick black , is the fluorescent signal in another channel . ( E ) The average speed of b12 HL and WT B cells near the IFC channel increases over time following gp120 injection . WT and b12 HL B cells were tracked over 20 min intervals . The time shown below is the endpoint of the tracking interval . Average speed is shown as a box- and-whisker blot . The results from the later intervals were compared to the initial interval for each cell type by unpaired t-test . p < 0 . 5; p < 0 . 003; **p < 0 . 00002 . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 02110 . 7554/eLife . 06467 . 022Video 6 . Intravital TP-LSM images of resident b12 B cells that extract gp120 from the IFC network cells . An image sequence of a 20 μm z-projection was acquired from inguinal LN of mouse , which had received by adoptive transfer the previous day b12 HL B cells , red , and wild type B cells , blue . Evans blue delineated the blood vessels , gray . Fluorescent gp120 , green , was injected near the tail base and the image sequence was acquired over the next 2 hr . Scale bar is 50 μm . Time counter shows hr:min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 02210 . 7554/eLife . 06467 . 023Figure 8 . Individual b12 HL B cell tracks from b12 HL B cells located in the LN follicle near the IFC channel following injection of gp120 . ( A ) Six tracks , red lines , and displacements , yellow or cyan arrows , of b12 HL B cells were superimposed on 3-D reconstruction image of gp120 loaded IFC network cells , green . 3-D reconstruction images were generated with 50 μm z-stack volume image . Displacement arrows were generated with fragments of the original track , which disconnected at time points that showed typical turning or discontinues movement from the original single track . Yellow arrows in each track indicate the starting point . Scale bar is 50 μm . Grid spacing in 3-D view is 10 μm . ( B , C ) Each track in ( a ) was visualized from a different angle to see typical tracked cell pattern ( approach , survey , move away ) . Scale bar is 50 μm . Grid spacing in 3-D view is 10 μm . Each individual track is shown as a single track plot of velocity ( solid line ) and gp120 intensity ( dotted line ) . Duration is total time span of the original single tracks in a 2 hr intravital TPLSM imaging . DOI: http://dx . doi . org/10 . 7554/eLife . 06467 . 023 A subset of SSMs that express SIGN-R1 captures monomeric and trimeric gp120 from the afferent lymph following its local injection . These macrophages are distinguishable from the standard subcapsular macrophages , but their localization over the IFC and by their expression of SIGN-R1 and other pattern recognition receptors . The IFC macrophages underlying these cells gradually acquire gp120 . A blocking antibody to SIGN-R1 reduced the interaction of gp120 with these cells as did altering the gp120 glycan shield . Arguing that a similar subset of macrophages is present in human LN , a gp120 overlay assay revealed CD163+DC-SIGN+ cells localized near the LN follicle that bound gp120 . In the immunized mouse LN , gp120 specific B cells arriving via the HEVs could interact with , and extract gp120 from the SIGN-R1+ macrophages . Some of these macrophages extended cellular processes into the LN follicle . This allowed gp120 specific B cells located in the follicle to also acquire gp120 . The B cells did not form long-lasting conjugates with the gp120 bearing IFC macrophages , but rather they repetitively , and transiently , interacted with them . B cells lacking gp120 reactive antigen receptors often wandered away showing little interest in the IFC cells . As observed previously with HEL transgenic B cells ( Suzuki et al . , 2009 ) , the migrating gp120 specific B cells localized the extracted antigen to their uropods . These results suggest that much like germinal center B cells , which extract antigen from FDCs to acquire T cell help , naïve B cells can extract antigen from the IFC macrophages to acquire T cell help to initiate the extra-follicular antibody response and to generate germinal center precursors . The IFC and cortical ridge is a crossroad for cellular traffic in the LN . Blood borne B cells enter nearby HEVs to access the LN parenchyma . The presence of the IFC macrophages and newly arrived antigen laden DCs explains why recent LN B cell entrants spend several hours exploring the IFC before entering the follicle ( Park et al . , 2012 ) . This region likely serves as a testing ground for B cells arriving in the LN from the blood to determine whether their BCRs possess sufficient affinity to acquire antigen . The IFC macrophage antigen repertoire will reflect the specificity of their cell surface receptors for material delivered in the subcapsular sinus lymph . In the case of gp120 , SIGN-R1 is functionally important for its capture by the IFC cells . However , IFC macrophages likely express other receptors that assist in gp120 acquisition since the blocking SIGN-R1 antibody only reduced the uptake of gp120 by 50% . Those B cells that do not find an IFC cell with cognate antigen migrate into the LN follicle where they remain for approximately a day ( Park et al . , 2012 ) . During that time should new antigen arrive in the subcapsular sinus that can be captured by the IFC macrophages , cognate B cells migrating near the follicle edge can acquire it . It will be of interest to determine whether antigen loading of the IFC macrophages results in their production of EBI2 ligands , which would tend to localize LN follicle B cells toward the IFC network cells ( Hannedouche et al . , 2011; Kelly et al . , 2011 ) . We did note that the velocity of both the WT and the gp120 specific B cells along the follicle edge increased during the 2 hr imaging period that followed gp120 injection . The follicle B cell can also scan the FDC network for cognate antigen ( Suzuki et al . , 2009 ) , however , until gp120 specific antibody appears little gp120 will likely be present . Those B cells that do not find cognate antigen eventually leave the follicle to exit the LN via the cortical lymphatics . Again they will have an opportunity to scan the IFC cellular network and cortical lymphatic macrophages for cognate antigen . Although the b12 B cells carry a mutated BCR , they are functionally naïve as they have not been exposed to cognate antigen . The affinity of the b12 HL BCR for the R66M gp120 is likely low as the b12 antibody predominately recognizes B clade HIV-1 gp120 while the R66M gp120 is an A/C clade . In addition , the b12 antibody binds weakly to recombinant R66M gp120 ( J Arthos , unpublished observation ) . As indicated above the gp120 specific B cells did not make a single long lasting synapse with the gp120 bearing IFC macrophages , but rather many transient interactions during which time the specific B cells serially extracted gp120 from the IFC cells , much like bumble bees gathering nectar . The temporal variation in gp120 fluorescence associated with the specific B cells , presumably reflects gp120-b12 HL BCR engagement , BCR internalization , gp120 degradation , and reduced fluorescence . As the B cell moves on to interact with another gp120 loaded macrophage the process is repeated . The reason why naïve B cells spend several hours repetitively engaging cognate antigen bearing cells is unclear . The serial BCR engagements and antigen extractions should provide additional B cell activation signals and saturate class II MHC molecules with peptide fragments , respectively . Those B cells that capture the highest amounts of antigen are most likely to receive help from CD4+ T cells ( Yuseff et al . , 2013; Avalos and Ploegh , 2014 ) . Our study is consistent with a recent study that showed murine B cells can very rapidly extract antigen from plasma membrane sheets decorated with antigen or an antigen surrogate ( Natkanski et al . , 2013 ) . In that study B cells acquired antigen from APCs by invaginating and pinching off the presenting cell membrane using their BCR along with myosin IIa-mediated contractions . Since we visualized the IFC macrophages as a consequence of their gp120 binding , we could not detect whether the gp120 specific B cells also pinched off IFC macrophage membrane although the imaging suggested that might be the case as the B cells appeared to grab aggregates of fluorescent gp120 . Alternatively , B cells can extract immobilized antigen by recruiting MHC class II-containing lysosomes to a B cell synapse ( Yuseff et al . , 2011 ) . Localized lysosome exocytosis acidifies the synapse and releases hydrolases , which promote antigen extraction . In our study this scenario seems less likely due to the transient nature of the interactions . For the gp120 specific B cells to gather gp120 from the SIGN-R1+ macrophages , the gp120 must remain on the surface of the macrophage and not be fully internalized . SIGN-R1 functions as a phagocytic receptor and is known to bind bacterial dextrans and the capsular polysaccharides of Streptococcus pneumonia ( Kang et al . , 2003 , 2004 ) . Our data indicates that SIGN-R1 is also involved in gp120 binding and that sufficient gp120 remains surface bound for B cells to acquire it from the SIGN-R1+ macrophages . Intravital imaging the inguinal LN of a naïve mouse 72 hr after local injection of three micrograms of labeled gp120 revealed its continued presence on IFC macrophages although the fluorescence levels had declined . This suggests that the gp120 remains available for several days following its acquisition by the IFC macrophages . Of note at the same time point gp120 was not found on the LN follicle FDCs ( C Park , unpublished data ) . It will important to assess how long the injected gp120 remains associated with the IFC macrophages and to determine whether at later time points other cell types can acquire and provide gp120 to B cells . The eventual loading of FDCs following gp120 immunization will likely depend upon gp120 persistence , the primary antibody response , gp120 immune complex formation , the capture of immune complexes by subcapsular macrophages , and their delivery to the underlying FDC network . Consistent with that scenario within 4 hr of the injection of gp120 into previously immunized mice , we found gp120 on the FDC network; however , the IFC macrophages also continued to capture it . The LN SIGN-R1+ IFC macrophages preferentially captured two early viral isolate monomeric gp120 preparations and a HIV-1 envelope glycoprotein trimer that adopts a native conformation . In contrast , a monomeric gp120 protein preparation treated with Peptide-N-Glycosidase F to reduce its glycan content , the algae protein PE , or HEL showed little specificity for these cells . As many of the current recombinant vaccine candidates are being produced in HEK 293 they are also likely to be captured in the LN by the SIGN-R1+ IFC macrophages we have described . However , the glycosylation pattern of the HIV-1 recombinant envelope proteins we tested likely differ from the HIV-1 envelope proteins produced in vivo in infected T cells and macrophages . Additional studies with gp120 or HIV virions produced in endogenously infected cells types are certainly warranted . Disruption of the IFC cellular network by pathogens would likely limit early antibody responses to gp120 and other antigens captured by this network of macrophages . As a consequence this would reduce subsequent immune complex formation and ultimately decrease FDC network loading . The IFMs are possible HIV-1 targets as they are CD4 and CCR5 positive ( Gray and Cyster , 2012 ) . Furthermore , cells located within the LN IFC have been noted to be infected in HIV patients ( Schuurman et al . , 1988 ) . Conversely , enhancing the loading and function of the IFC macrophages could improve the early humoral response to gp120 vaccines and other antigens captured by the IFC cells . Since we injected gp120 in the absence of any adjuvant , further experiments will be needed to assess how different adjuvants affect gp120 capture and its delivery to LN B cells . In addition , our study predominately focused on the events that occurred over the first several hours following gp120 injection . Many questions remain to be answered . What are the fates of the B cells that encounter antigen from the macrophages in the IFC ? How does the binding of gp120 to the IFC macrophages functionally affect them ? Additional studies built on the identification of this IFC cellular network should help to optimize the early extra-follicular antibody production and germinal center formation following gp120 immunization . C57BL/6 and C . 129S4 ( B6 ) -Ifngtm3 . 1Lky/J ( GREAT mice ) were obtained from Jackson Laboratory ( Bar Harbor , ME ) . The C57BL/6 , b12 HL , b 12H , and b12 L mice were obtained from Dr David Nemazee and maintained at the NIH . The LysM-enhanced green fluorescent protein ( EGFP ) mice were kindly provided by Ron Germain ( NIAID , NIH ) with permission from Thomas Graf ( Center for Genomic Regulation , Barcelona , Spain ) . All mice were used in this study were 6–14 weeks of age . Mice were housed under specific-pathogen-free conditions . All the animal experiments and protocols used in the study were approved by the NIAID Animal Care and Use Committee ( ACUC ) at the National Institutes of Health . The coding sequences of the R66M ( A/C R66M 7Mar06 3A9env2 ) envelope protein , from +1 to the gp120-gp41 junction was inserted into a mammalian expression vector downstream of a synthetic leader sequence . The coding sequence was provided by Dr Cynthia Derdeyn ( Emory University ) . The vector was transiently transfected into either 293F or CHO-S cells using FreeStyle MAX Reagent ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) per the manufacturer's instructions . Protein-containing supernatants were harvested 72 hr after transfection and were passed over a column of lectin sepharose from Galanthus nivalis ( Vector Laboratories , Burlingame , CA ) , which was diluted 1:5 with sepharose 4B not bound to ligand to minimize avid binding . gp120 was eluted with 20 mM glycine-HCl , pH 3 . 0 , 150 mM NaCl , 500 mM α-methyl-mannopyranoside ( Sigma Aldrich , St . Louis , MO ) , in 5-ml fractions , directly into 1 M Tris-HCl , pH 8 . 0 . Peak fractions were pooled , concentrated with a stirred cell concentrator ( EMD Millipore , Bilerica , MA ) and dialyzed exhaustively against HEPES , pH 7 . 4 , 150 mM NaCl . Higher molecular weight forms were removed by size-exclusion chromatography . To eliminate possible endotoxin contamination from purified proteins , a Triton X114 extraction was done . Proteins were quantified by ultraviolet absorption at a wavelength of 280 nm ( extinction coefficient , 1 . 1 ) and values were confirmed by a bicinchoninic acid protein assay ( Thermo Fisher Scientific ) . Biotinylated soluble HIV-1 Env trimeric gp120 , BG505 SOSIP , was kindly provided by Dr John P Moore ( Sanders et al . , 2013 ) . Recombinant gp120 and HEL were conjugated to fluorescent ( Alexa Fluor 488 , 594 , or 647 ) with the Microscale Protein Labeling Kit ( Molecular probe , Thermo Fisher Scientific ) . Antibodies against CD169 ( 3D6 . 112 , BioLegend , San Diego , CA ) , SIGN-R1 ( eBio22D1 , eBiosciences , San Diego , CA ) , LYVE-1 ( Clone# 223322 , R&D System , Minneapolis , MN ) and ER-TR7 ( ER-TR7 , AbD serotec , Bio-Rad Laboratories , Hercules , CA ) were conjugated to Alexa Fluor 488 , 594 , or 647 with the Antibody Labeling Kits ( Molecular probe ) . Labeling reactions , conjugates purification , and determination of degree of labeling were performed following the company manuals . 6–10-week-old recipient , anesthetized mice were injected with fluorescent labeled materials for intravital imaging or section imaging by tail base injection . Inguinal LNs were prepared for intravital microscopy as described ( Park et al . , 2009 , 2012 ) . Cell populations were labeled for 10 min at 37°C with 2 . 5–5 μM red cell tracker CMTMR ( Molecular probes ) or 2 μM of eFluor450 ( eBioscience ) . 5–10 million labeled cells of each population in 200 ml of PBS were adoptively transferred by tail vein injection into recipient mice . After anesthesia the skin and fatty tissue over inguinal LN were removed . The mouse was placed in a pre-warmed coverglass chamber slide ( Nalgene , Nunc , Thermo Fisher Scientific ) . The chamber slide was then placed into the temperature control chamber on the microscope . The temperature of air was monitored and maintained at 37 . 0 ± 0 . 5°C . Inguinal LN was intravitally imaged from the capsule over a range of depths ( 10–220 μm ) . All two-photon imaging was performed with a Leica SP5 inverted 5 channel confocal microscope ( Leica Microsystems , Wetzlar , Germany ) equipped with 25× water dipping objective , 0 . 95 NA ( immersion medium used distilled water ) . Two-photon excitation was provided by a Mai Tai Ti:Sapphire laser ( Spectra Physics , Newport Research Corporation , Invine , CA ) with a 10 W pump , tuned wavelength ranges from 810 to 910 nm . Emitted fluorescence was collected using a 4 channel non-descanned detector . Wavelength separation was through a dichroic mirror at 560 nm and then separated again through a dichroic mirror at 495 nm followed by 525/50 emission filter for GFP or Alexa Fluor 488 ( Molecular probes ) ; and the eFluor450 ( eBioscience ) or second harmonic signal was collected by 460/50 nm emission filter; a dichroic mirror at 650 nm followed by 610/60 nm emission filter for CMTMR , PE or Alexa Fluor 594; and the Evans blue or Alexa Fluor 647 signal was collected by 680/50 nm emission filter . For four-dimensional analysis of cell behavior , stacks of various numbers of section ( z step = 3 , 4 , or 6 μm ) were acquired every 10–12 s to provide an imaging volume of 30–100 μm in depth . Sequences of image stacks were transformed into volume-rendered four-dimensional videos using Imaris software v . 7 . 7 . 1 64× ( Bitplane AG , Zurich , Switzerland ) , and the tracks analysis was used for semi-automated tracking of cell motility in three dimensions by using the following parameters: autoregressive motion algorithm , estimated diameter 10 μm , background subtraction true , maximum distance 20 μm , and maximum gap size 3 . Tracks acquired that could be tracked for at least 20% of total imaging duration were used for analysis . Some tracks were manually examined and verified . Calculations of the cell motility parameters ( speed , track length , displacement , straightness and speed variability ) were performed using the Imaris software v . 7 . 7 . 1 64× ( Bitplane AG ) . Statistical analysis was performed using Prism software ( GraphPad Software , La Jolla , CA ) . 3D-reconstructions from original images from TP-LSM were generated by the surfaces function of the Surpass view in Imaris software v . 7 . 7 . 1 64× ( Bitplane AG ) , performed with semi-automated creation wizard . Annotations on videos and video editing were performed using Adobe Premiere Pro CS3 ( Adobe Systems Incorporated , McLean , VA ) . Video files were converted to MPEG4 format with Imtoo Video Converter Ultimate 6 . 0 . 2 for Mac ( Imtoo Software Studio ) . Immunohistochemistry was performed using a modified method of a previously published protocol ( Chai et al . , 2013 ) . Briefly , freshly isolated LNs or spleens were fixed in newly prepared 4% paraformaldehyde ( Electron Microscopy Science , Hatfield , PA ) overnight at 4°C on an agitation stage . Spleens or LNs were embedded in 4% low melting agarose ( Invitrogen ) in PBS and sectioned with a vibratome ( Leica VT-1000 S ) at a 30 μm thickness . Thick sections were blocked in PBS containing 10% fetal calf serum , 1 mg/ml anti-Fcγ receptor ( BD Biosciences ) , and 0 . 1% Triton X-100 ( Sigma Aldrich ) for 30 min at room temperature . Sections were stained overnight at 4°C on an agitation stage with the following antibodies: anti-B220 ( RA3-6B2 , BD Biosciences ) , anti-CD3e ( 17A2 , BD Biosciences ) , anti-CD4 ( RM4-5 , BD Biosciences ) , anti-CD11c ( HL3 , BD Biosciences ) , anti-CD169 ( 3D6 . 112 , BioLegend ) , anti-ER-TR7 ( ER-TR7 , AbD serotec ) and anti-CD21/35 ( BioLegend ) and with labeled gp120 . For the human LN analysis , cold acetone fixed frozen sections ( cat# . T1234161 ) were purchased from BioChain Institute , Inc , ( Newark , CA ) . Sections were fixed again with 4% paraformaldehyde for 10 min at room temperature . Then sections were blocked in PBS containing 10% fetal calf serum , 1 mg/ml human IgG ( Sigma Aldrich ) , and 0 . 1% Triton X-100 ( Sigma Aldrich ) for 30 min at room temperature . Sections were stained overnight at 4°C on an agitation stage with the following antibodies: anti-CD19 ( 4G7 , BD Biosciences ) , anti-CD4 ( RPA-T4 , eBiosciences ) , anti-DC-SIGN ( DCN46 , BD Biosciences ) , anti-CD163 ( eBioGHI/61 , eBiosciences ) and anti-CD11c ( S-HCL-3 , BD Biosciences ) or with gp120 . For the trimeric gp120 , biotinylated trimeric gp120 was injected into tail base for 2 hr and detected by AlexaFluor 488 conjugated streptavidin in LN sections . Stained thick sections and human LN sections were microscopically analyzed using a Leica SP5 confocal microscope ( Leica Microsystem , Inc . ) and images were processed with Leica LAS AF software ( Leica Microsystem , Inc . ) and Imaris software v . 7 . 7 . 1 64× ( Bitplane AG ) . Recombinant of gp120 was deglycosylated with Peptide-N-Glycosidase F ( PNGase F , New England Biolabs , Ipswich , MA ) . In order to minimize the denaturation of gp120 , recombinant gp120 was deglycosylated with a company protocol that uses non-denaturing reaction conditions . One µg of fluorescent gp120 ( AlexaFluor 647 conjugated ) was incubated at 37°C for 20 hr in reaction mixture of 1 unit of PNGase F , 2 µl of 10X GlycoBuffer and dH2O to make a 20 µl total reaction volume . In determine whether the degylcosylation affected the fluorescent signal , the deglycosylated gp120 and reaction control ( without PNGase F ) were analyzed on a 10% NuPAGE Bis-Tris gel ( Life technologies , Thermo Fisher Scientific ) , and the strength of fluorochrome was measured by Odyssey CLx Infrared Imaging System ( LI-COR , Inc . , Lincoln , NE ) . To outline blood vessels 50 μl of Evans Blue solution ( 0 . 5 μg/ml in PBS ) was injected into orbital or tail vein prior to imaging . HEVs were delineated via the presence of adherent T-cells previously adoptively transferred into the tail vein . To visualize endothelial cells in the lymphatic sinuses , purified rat anti-mouse LYVE-1 was conjugated with Alexa Fluor 647 . Five μg of AlexFluor 647 conjugated LYVE-1 antibody in 50 μl of PBS was injected into tail base 1 hr prior to imaging . To visualize changes in intracellular calcium in b12 HL B cells , the cells were loaded with Calcium Orange ( Invitrogen ) . A single cell suspension of b12 HL B cells in culture media was incubated with 5 μM of Calcium Orange ( 5 mM stock solution in Fluronic F-127 [20% [wt/vol]solution in DMSO] ) , at room temperature for 30 min . The stained cells were washed five times with culture media prior to being adoptively transferred . The Calcium Orange signal was imaged by TP-LSM . The intensities of fluorescent signals in ROIs were measured by LSA AF Lite software ( Leica Microsystem , Inc . ) . To make the relative mean intensity score , Alexa Fluor 647 conjugated gp120 intensity from SIGN-R1+ sinus macrophages ( 4 cells ) was divided by signal intensity from sinus lumen ( 1 area ) as a reference signal . The gp120 intensity from the network cells ( 4 cells ) was determined in the same manner . The generated mean intensity score was plotted as a function of time . The intensity scores of gp120 and NP-PE were calculated using intensity of HEL as a reference signals in ROIs . The intensity of fluorescent signals in individual tracked cells was measured by Imaris software v . 7 . 7 . 1 64× ( Bitplane AG ) . The cell volume was reconstructed by the surface function in Imaris software , tracked , and evaluated manually . The intensity ratio of each signal at the indicated time points was calculated using mean intensity of each signal over the entire imaging period . Inguinal LNs were carefully collected without fat tissue and gently teased apart with micro-forceps into RPMI 1640 media containing 2 mM L-glutamine , antibiotics ( 100 IU/ml penicillin , 100 μg/ml streptomycin ) , 1 mM sodium pyruvate , and 50 μM 2-mercaptoethanol , pH 7 . 2 . The tissue was then digested with Liberase Blendzyme 2 ( 0 . 2 mg/ml , Roche Applied Science , Penzberg , Germany ) and DNase I ( 20 μg/ml ) for 30 min at 37°C , while rocking vigorously . Proteases were then inactivated with 10% fetal bovine serum and 2 mM EDTA and the cell disaggregated by passing them through a 40 μm nylon sieve ( BD Bioscience ) . Single cells were then washed with 1% BSA/PBS and blocked with anti-Fcγ receptor ( BD Biosciences ) . LIVE/DEAD Fixable Aqua Dead Cell Stain Kit ( Molecular Probes ) was used in all experiments to exclude dead cells . Single cells were re-suspended in PBS , 2% FBS , and stained with fluorochrome-conjugated or biotinylated antibodies against Gr-1 ( RB6-8C5 , eBioscience or BD Bioscience ) , CD169 ( 3D6 . 112 , BioLegend ) , SIGN-R1 ( eBio22D1 , eBiosciences ) , anti-B220 ( RA3-6B2 , BD Bioscience ) , anti-CD4 ( clone RM4-5 , BD Bioscience ) , anti-CD11b ( M1/70 , eBiosciences ) , anti-CD11c ( HL3 , BD Bioscience ) , anti-F4/80 ( BM8 , eBiosciences ) . Data acquisition was done on FACSCanto II ( BD Bioscience ) flow cytometer and analyzed with FlowJo software ( FLOWJO , LLC , Ashland , OR ) . FACS Sorting of gp120 positive cells was performed with LN cells prepared as was done for flow cytometry . The suspended LN cells were immunostained and applied to a FACS-Aria , which was set for 5–6 droplets through 100 μm nozzle ( 20 psi ) . Sorted cells were directly visualized with confocal microscope and cultured with standard media containing 20 ng/ml of M-CSF for 7 days before analysis . The same fluorescently labeled gp120 used in the in vivo experiments was used in the in vitro binding assay . Single cells suspensions were prepared , washed with 1% BSA/PBS , and blocked with anti-Fcγ receptor ( BD Biosciences ) . LIVE/DEAD Fixable Aqua Dead Cell Stain Kit ( Molecular Probes ) was used to exclude dead cells . Single cells were re-suspended in PBS , 2% FBS , and stained on ice with various antibodies and fluorescent gp120 ( 1 μg ) . In some instances non-labeled gp120 ( 2 μg ) or mouse SIGN-R1/CD209b antibody ( 2 μg , R&D System ) was added for 30 min prior to immunostaining and the addition of fluorescent gp120 . Data acquisition was done on FACSCanto II ( BD Bioscience ) flow cytometer and analyzed with FlowJo software ( FLOWJO , LLC ) . 3 hr after gp120 injection the draining inguinal LNs were collected and prepared for flow cytometry . The cells were immunostained using the BD Cytofix/Cytoperm Fixation/Permeabilization Kit protocol ( BD Bioscience ) . Briefly single cell suspensions were fixed and permeabilized with Fixation/Permeabilization solution for 20 min at 4°C . The cells were immunostained with PE-Cy7 conjugated interferon-γ antibody ( 1:100 dilution of XMG1 . 2 , eBiosciences ) overnight . The cells were then pelleted , washed , and resuspended with 1× BD Perm/Wash buffer . All flow cytometry data were collected on a BD FACS CANTO II and analyzed with FlowJo software ( FLOWJO , LLC ) . The level of eYFP expression in C . 129S4 ( B6 ) -Ifngtm3 . 1Lky/J ( GREAT mice ) after gp120 injection was measured with LN cell suspension prepared as outlined for flow cytometry analysis . Each group of C57Bl/6 mice was immunized with gp120 prepared from either 293F or CHO-S cells . The recombinant gp120 ( 50 µg ) was mixed with Imject Alum ( Thermo Fisher Scientific ) and injected subcutaneously . Mice were boosted with same dose of antigen at the indicated days along with Alum . Serum gp120 specific Ig levels in these mice were analyzed by ELISA . Briefly , 96 well ELISA plates ( Nalgene , Nunc ) were coated with gp120 ( 0 . 8 μg/well ) overnight at 4°C , washed and blocked with 1% BSA fraction V ( Sigma–Aldrich ) , serum titers were then added to the plates and incubated 4 hr at 4°C . After washing alkaline phosphatase-labeled goat anti-mouse IgM or IgG isotype specific antibodies were added for 2 hr at room temperature ( SouthernBiotech , Birmingham , AL ) . After washing , PNPP one component substrate ( SouthernBiotech ) was used to detect the amount of antibody bound . All experiments were performed at least three times . Primary data calculated by Imaris ( Bitplane AG ) was acquired and processed with Microsoft Excel software . Error bars with ±SEM , and p values were calculated with GraphPad Prism ( GraphPad software ) as a function of linear regression in XY analyses ( slope , Figure 2B ) , 2way ANOVA ( Figures 2B , 4E ) , unpaired t-test ( Figure 7C ) .	The human immune system contains many different cell types , which play specific roles in defending the body from invading pathogens , such as bacteria and viruses . For example , macrophages engulf and digest foreign material , whereas specialized B cells termed plasma cells produce molecules called antibodies that help to destroy specific pathogens . However , specific antibodies are only produced if naive B cells have already encountered the pathogen or its surface proteins . Attempts to improve how the immune system responds to the human immunodeficiency virus ( HIV-1 ) have failed to control and prevent infection . One of the main components of many prospective HIV-1 vaccines is a protein called gp120 , which is located on the surface of the virus . Specific B cells recognize this protein and can develop into plasma cells that produce antibodies against HIV-1 . However , little is known about how these specific B cells initially get exposed to gp120 . Park et al . injected gp120 into mice , and used sophisticated microscopy to track its movement through the animal . This revealed that gp120 is rapidly transported to nearby lymph nodes—organs that are spread throughout the body , and play an important role in maintaining the immune response . Specialized macrophages can then capture and deliver gp120 to other macrophages in the lymph node . These specialized macrophages serve as a gp120 reservoir and are located in part of the lymph node that is a bit like a traffic hub , in that other immune cells constantly pass through it . As such , B cells that specifically recognize gp120 have a high likelihood of encountering these gp120-bearing macrophages , thereby allowing the specific B cells to extract gp120 , develop into plasma cells , and produce HIV-1 specific antibodies . Manipulating this macrophage network may help to optimize the antibody responses to gp120 and so , in the future , could provide a way of treating or preventing HIV-1 infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2015	The HIV-1 envelope protein gp120 is captured and displayed for B cell recognition by SIGN-R1+ lymph node macrophages
To successfully guide limb movements , the brain takes in sensory information about the limb , internally tracks the state of the limb , and produces appropriate motor commands . It is widely believed that this process uses an internal model , which describes our prior beliefs about how the limb responds to motor commands . Here , we leveraged a brain-machine interface ( BMI ) paradigm in rhesus monkeys and novel statistical analyses of neural population activity to gain insight into moment-by-moment internal model computations . We discovered that a mismatch between subjects’ internal models and the actual BMI explains roughly 65% of movement errors , as well as long-standing deficiencies in BMI speed control . We then used the internal models to characterize how the neural population activity changes during BMI learning . More broadly , this work provides an approach for interpreting neural population activity in the context of how prior beliefs guide the transformation of sensory input to motor output . Even simple movements , like reaching to grasp a glass of water , require dozens of muscles to be activated with precise coordination . This precision is especially impressive in light of sensory feedback delays inherent to neural transmission and processing: when we make a swift arm movement , the brain only knows where the arm was a split second ago , not where it currently is . To generate the desired movement , it is widely believed that we form internal models that enable selection of appropriate motor commands and prediction of the outcomes of motor commands before sensory feedback becomes available ( Crapse and Sommer , 2008; Shadmehr et al . , 2010 ) . Mechanistic studies have made important progress toward identifying the neural circuits that implement internal models in sensory ( Komatsu , 2006; Kennedy et al . , 2014; Schneider et al . , 2014 ) , vestibular ( Laurens et al . , 2013 ) , and motor ( Sommer , 2002; Ghasia et al . , 2008; Keller and Hahnloser , 2009; Azim et al . , 2014 ) systems . In parallel , psychophysical studies have demonstrated the behavioral correlates of these internal models ( Shadmehr and Mussa-Ivaldi , 1994; Wolpert et al . , 1995; Thoroughman and Shadmehr , 2000; Kluzik et al . , 2008; Mischiati et al . , 2015 ) and the behavioral deficits that result from lesions to corresponding brain areas ( Shadmehr and Krakauer , 2008; Bhanpuri et al . , 2013 ) . Together with studies showing neural correlates of internal models ( Sommer , 2002; Gribble and Scott , 2002; Ghasia et al . , 2008; Mulliken et al . , 2008; Keller and Hahnloser , 2009; Green and Angelaki , 2010; Berkes et al . , 2011; Laurens et al . , 2013 ) , these previous studies have provided strong evidence for the brain’s use of internal models . These internal models are presumably rich entities that reflect the multi-dimensional neural processes observed in many brain areas ( Cunningham and Yu , 2014 ) and can drive moment-by-moment decisions and motor output . However , to date , most studies have viewed internal models through the lens of individual neurons or low-dimensional behavioral measurements , which provides a limited view of these multi-dimensional neural processes ( although see Berkes et al . , 2011 ) . Here , we address these limitations by extracting a rich internal model from the activity of tens of neurons recorded simultaneously . The key question that we ask is whether such an internal model can explain behavioral errors that cannot be explained by analyzing low-dimensional behavioral measurements in isolation . We define an internal model to be one’s inner conception of a motor effector , which includes one’s prior beliefs about the physics of the effector as well as how neural commands drive movements of the effector . When we extract a subject’s internal model , we seek a statistical model of the effector dynamics that is most consistent with the subject’s neural commands . Interpreting high-dimensional neural activity through the lens of such an internal model offers insight into how one’s prior beliefs about the effector affect the transformation of sensory inputs into population-level motor commands on a timescale of tens of milliseconds . To date , it has been difficult to identify such an internal model due the complexities of non-linear effector dynamics and multiple sensory feedback modalities , the need to monitor many neurons simultaneously , and the lack of an appropriate statistical algorithm . To overcome these difficulties , we leveraged a closed-loop brain-machine interface ( BMI ) paradigm ( Figure 1A ) in rhesus monkeys , which translates neural activity from the primary motor cortex ( M1 ) into movements of a computer cursor ( Green and Kalaska , 2011 ) . A BMI represents a simplified and well-defined feedback control system , which facilitates the study of internal models ( Golub et al . , 2016 ) . In particular , the BMI mapping from neural activity to movements is completely specified by the experimenter and can be chosen to define linear cursor dynamics , the relevant sensory feedback can be limited to one modality ( in this case , vision ) , and all neural activity that directly drives the cursor is recorded . 10 . 7554/eLife . 10015 . 003Figure 1 . Closed-loop control of a brain-machine interface ( BMI ) cursor . ( A ) Schematic view of the brain-machine interface . Subjects produce neural commands to drive a cursor to hit visual targets under visual feedback . ( B ) Cursor trajectories from the first 10 successful trials to each of 16 instructed targets ( filled circles ) in representative data sets . Target acquisition was initiated when the cursor visibly overlapped the target , or equivalently when the cursor center entered the cursor-target acceptance zone ( dashed circles ) . Trajectories shown begin at the workspace center and proceed until target acquisition . Data are not shown during target holds . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 00310 . 7554/eLife . 10015 . 004Figure 1—figure supplement 1 . Proficient control of the brain-machine interface ( BMI ) . ( A ) Histograms of within-session averaged success rates and ( B ) movement times across all sessions and both monkeys . Red lines denote averages across sessions , and triangles indicate the within-session averages for the example sessions from Figure 1B . Movement times were calculated as the time elapsed between target onset and target acquisition ( i . e . , excluding all hold times , but including reaction times ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 004 During proficient BMI control , as with other behavioral tasks , subjects make movement errors from time to time . One possible explanation for these errors is that they arise due to sensory or motor “noise” that varies randomly from one trial to the next ( Harris and Wolpert , 1998; Osborne et al . , 2005; Faisal et al . , 2008 ) . Another possibility , which is the central hypothesis in this study , is that a substantial component of movement errors is structured and can be explained by a mismatch between the subject’s internal model of the BMI and the actual BMI mapping . Testing this hypothesis required the development of a novel statistical method for estimating the subject’s internal model from the recorded M1 activity , BMI cursor movements , and behavioral task goals . The internal model represents the subject’s prior beliefs about the physics of the BMI cursor , as well as how the subject’s neural activity drives the cursor . To justify the study of internal models in a BMI context , we first asked whether subjects show evidence of internal prediction during BMI control . Next , we asked whether interpreting M1 activity through extracted internal models could explain movement errors that are present throughout proficient BMI control and long-standing deficiencies in control of BMI movement speed . Finally , because a key feature of internal models is their ability to adapt ( Shadmehr et al . , 2010 ) , we altered the BMI mapping and asked whether the internal model adapted in a manner consistent with the new BMI mapping . An important distinction that we make relative to previous work is that we are not asking circuit-level questions about how and where in the brain these internal models operate . Rather , we seek a statistical representation of the subject’s prior beliefs about the BMI mapping ( i . e . , an internal model ) that can be used to explain behavioral errors . Although internal models might not reside in M1 ( Shadmehr , 1997; Pasalar et al . , 2006; Miall et al . , 2007; Mulliken et al . , 2008; Lisberger , 2009 ) , their computations influence activity in M1 . Thus , by examining the moment-by-moment relationship between M1 population activity and task objectives , it may be possible to extract a detailed representation of the subject’s internal model . Because internal models have not previously been studied in a BMI context , we sought evidence of internal prediction . A hallmark of internal prediction is compensation for sensory feedback delays ( Miall et al . , 2007; Shadmehr et al . , 2010; Farshchiansadegh et al . , 2015 ) . To assess the visuomotor latency experienced by a subject in our BMI system , we measured the elapsed time between target onset and the appearance of target-related activity in the recorded neural population ( Figure 2A ) . The delays we measured ( 100 ms , monkey A; 133 ms , monkey C ) are consistent with visuomotor latencies reported in arm reaching studies of single-neurons in primary motor cortex ( Schwartz et al . , 1988 ) . Next , we asked whether subjects produced motor commands consistent with the current cursor position , which was not known to the subject due to visual feedback delay , or whether motor commands were more consistent with a previous , perceived position ( Figure 2B , C and Figure 2—figure supplement 1 ) . If subjects did not compensate for visual feedback delays and aimed from the most recently available visual feedback of cursor position , we would expect errors to be smallest at lags of 100 ms and 133 ms relative to the current cursor position for monkeys A and C , respectively ( dashed red lines in Figure 2C ) . Rather , we found that these error curves had minima at lags close to 0 ms ( dashed black lines in Figure 2C ) , indicating that motor commands through the BMI mapping pointed closer to the targets when originating from the current cursor position than from any previous position . This finding suggests that subjects use an internal model to internally predict the current cursor position . 10 . 7554/eLife . 10015 . 005Figure 2 . Subjects compensate for sensory feedback delays while controlling a BMI . ( A ) The visuomotor latency experienced by a subject in our BMI system was assessed by measuring the elapsed time between target onset and the first significant ( p<0 . 05 ) decrease in angular error . If that first decrease was detected τ+1 timesteps following target onset , we concluded that the visuomotor latency was at least τ timesteps ( red dashed lines ) . For both subjects , the first significant difference was highly significant ( p<10-5 , two-sided Wilcoxon test with Holm-Bonferroni correction for multiple comparisons; n = 5908 trials; monkey C: n = 4578 trials ) . ( B ) Conceptual illustration of a single motor command ( black arrows ) shifted to originate from positions lagged relative to the current cursor position ( open circle ) . In this example , the command points farther from the target as it is shifted to originate from earlier cursor positions . ( C ) Motor commands pointed closer to the target when originating from the current cursor position ( zero lag ) than from outdated ( positive lag ) cursor positions that could be known from visual feedback alone ( p<10-5 , two-sided Wilcoxon test; monkey A: n = 33 , 660 timesteps across 4489 trials; monkey C: n = 31 , 214 timesteps across 3639 trials ) . Red lines indicate subjects’ inherent visual feedback delays from panel A . Shaded regions in panels A and C ( barely visible ) indicate ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 00510 . 7554/eLife . 10015 . 006Figure 2—figure supplement 1 . Error metrics for assessing estimates of movement intent . The primary error metric we used was the absolute angle by which a velocity command would have missed the target , taking into account the cursor and target radii . Because task success requires hitting the target ( i . e . , cursor-target overlap ) , we define all commands that would result in cursor-target overlap as having zero angular error . Mathematically , this corresponds to any velocity command that points within ΘZ=s⁢i⁢n-1⁢ ( ( RC+RT ) /D ) degrees from the target center , where D is the distance between target center and the position from which the velocity command originates , and RC and RT are the cursor and target radii , respectively . A velocity command that would not hit the target is given an error , ΘP , equal to the angle by which the cursor would have missed the target . Equivalently , we can consider the cursor-target overlap zone defined by a target-concentric circle with radius RT+RC , and define angular error , ΘP , to be the smallest angle between the velocity command and the perimeter of the cursor-target overlap zone . ( A ) Consider an example in which we assess the error of velocity commands ( blue and green arrows ) originating from a position D=85 mm from the target center ( the distance between workspace center and target center in a typical experiment ) . Here , the cursor radius , RC , and the target radius , RT , are both 7 mm ( typical values from experiments ) . Any velocity command that points within ΘZ=s⁢i⁢n-1⁢ ( ( RC+RT ) /D ) =9 . 48o of the target center would result in cursor-target overlap and thus would be evaluated as having zero angular error . The green arrow points in the direction farthest from the target center such that movement of the cursor ( dashed blue circle ) in this direction would result in cursor-target overlap . A velocity command ( blue arrow ) pointing ΘC=30o from the target center would miss the cursor-target overlap zone by ΘP=ΘC-ΘZ=20 . 52o . ( B ) Consider a similar example , but with the velocity command originating from a position D=60 mm from the target center . Because the cursor-target distance has decreased , the zero error window increases to ΘZ=13 . 49o . As a result , a velocity command that points ΘC=30o from the target center ( blue arrow; same ΘC as in panel A ) , is now evaluated as having a smaller error , ΘP=16 . 51o . The difference between the error angles , ΘP , in panel A and panel B , reflects the task goals , because a wider range of velocity commands would result in task success in panel B compared to panel A , and thus the same velocity command is more task-appropriate in panel B than in panel A . The ΘP metric was used extensively throughout this work ( Figure 2A , C , Figure 3B , C , Figure 4C , Figure 6A , Figure 3—figure supplement 3 , Figure 3—figure supplement 4 , Figure 3—figure supplement 7 , Figure 3—figure supplement 8 , and Figure 4—figure supplement 2 ) . We repeated those analyses using ΘC as the error metric ( i . e . , ignoring the distance to the target , cursor radius , and target radius ) and found qualitatively similar results . In Figure 2C and Figure 3—figure supplement 8 , velocity commands were evaluated as originating from a range of lagged cursor positions . Since cursor positions later in a trial tend to be closer to the target than earlier positions , velocity commands will tend to have smaller ΘP when originating from these later cursor positions . We controlled for this distance-to-target effect to ensure that it did not influence our results ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 006 Because we have not yet explicitly identified the subject’s internal model , motor commands were defined in this analysis using the BMI mapping , which is external to the subject . If the internal model bears similarities to the BMI mapping , it is reasonable to use the BMI mapping as a proxy for the internal model to assess feedback delay compensation . With evidence that subjects engage an internal model during BMI control , we next asked whether we could explicitly identify an internal model from the recorded neural activity . The BMI mapping , which determines the cursor movements displayed to the subject , provides one relevant , low-dimensional projection of the high-dimensional neural activity . With evidence that subjects use an internal model during closed-loop BMI control , we asked whether mismatch between an internal model and the actual BMI mapping could explain the subject’s moment-by-moment aiming errors . This requires identifying the subject’s internal model , which could reveal a different projection of the high-dimensional neural activity , representing the subject’s internal beliefs about the cursor state . Because of the closed-loop nature of the BMI paradigm , the subject continually updates motor control decisions as new visual feedback of the cursor becomes available . To resolve these effects , the internal model needs to operate on a timescale of tens of milliseconds ( in this case , a single timestep of the BMI system ) on individual experimental trials . The extraction of such a rich internal model has been difficult prior to this study due to the lack of an appropriate statistical framework . To overcome this limitation , we developed an internal model estimation ( IME ) framework , which extracts , from recorded population activity , a fully parameterized internal model along with a moment-by-moment account of the internal prediction process ( Figure 3A ) . In the IME framework , the subject internally predicts the cursor state according to: ( 2 ) x~t=A~⁢x~t-1+B~ut+b~ 10 . 7554/eLife . 10015 . 007Figure 3 . Mismatch between the internal model and the BMI mapping explains the majority of the subjects’ cursor movement errors . ( A ) At each timestep , the subject’s internal state predictions ( x~t-2 , x~t-1 , x~t ) are formed by integrating the visual feedback ( xt-3 ) with the recently issued neural commands ( ut-2 , ut-1 , ut ) using the internal model ( A~ , B~ , b~ ) . We defined cursor states and internal state predictions to include components for position and velocity ( i . e . , xt=[pt;vt] , x~t=[p~t;v~t] ) . ( B ) Cursor trajectory ( black line ) from a BMI trial that was not used in model fitting . Red whisker shows the subject’s internal predictions of cursor state as extracted by IME . The critical comparison is between the actual cursor velocity ( vt; black arrow ) and the subject’s internal prediction of cursor velocity ( v~t; red arrow ) . ( C ) Cross-validated angular aiming errors based on IME-extracted internal models are significantly smaller than cursor errors from the BMI mapping ( p<10-5 , two-sided Wilcoxon test; monkey A: n = 5908 trials; monkey C: n = 4577 trials ) . Errors in panel B are from a single timestep within a single trial . Errors in panel C are averaged across timesteps and trials . Errors in panels B and C incorporate temporal smoothing through the definition of the BMI mapping and the internal model , and are thus not directly comparable to the errors shown in Figure 2C , which are based on single-timestep velocity commands needed for additional temporal resolution . Error bars ( barely visible ) indicate ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 00710 . 7554/eLife . 10015 . 008Figure 3—figure supplement 1 . Full probabilistic graphical model for the internal model estimation ( IME ) framework . At timestep t , the subject generates a set of internal state predictions ( row of x~t variables in the solid box ) based on the most recently available visual feedback ( xt-3 ) and recently issued neural commands ( ut-2raw , … , utraw ) . Because the subject aims straight to the target from the subject’s up-to-date prediction of cursor position , the target position ( Gt ) should lie along the line defined by subject’s up-to-date position and velocity predictions ( p~tt and v~tt , included in x~tt ) . At the next timestep ( t+1 ) , the subject generates a revised set of internal predictions ( row of x~t+1 variables ) based on newly received visual feedback ( xt-2 ) , the most recently issued neural command ( ut+1raw ) , and previously issued neural commands ( ut-1raw and utraw ) . A column of internal state predictions represents the subject’s internal predictions given more and more recent visual feedback ( e . g . , the x~tk variables in the dashed box represent the subject’s internal predictions of the timestep t cursor state xt given visual feedback available through timestep k ) . Once a neural command is issued , it cannot be revised , and as such , the same neural command continues to influence internal predictions until visual feedback becomes available from the corresponding timestep ( utraw affects the x~t , x~t+1 and x~t+2 variables , but not the x~t+3 variables as xt has become available , rendering utraw irrelevant ) . The target position Gt took on the same value for all timesteps within the same trial . Shaded nodes indicate observed data , and red unshaded nodes are latent variables representing the subject’s internal state predictions . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 00810 . 7554/eLife . 10015 . 009Figure 3—figure supplement 2 . A unit-by-unit comparison of the subject’s internal model and the BMI mapping . Both the internal model and the BMI mapping can be visualized as a collection of pushing vectors . ( A ) The BMI mapping parameter B from Equation 1 describes how each neuronal unit actually drove the BMI cursor . Each 2D column in the velocity portion ( lower two rows ) of B corresponds to a particular unit and can be visualized as a pushing vector describing the direction and magnitude by which a single spike from that unit would push the cursor according to the BMI mapping . Each unit’s pushing vector is given a unique color . ( B ) The internal model parameter B~ from Equation 2 describes how the subject believes each neuronal unit drives the cursor . As in panel A , each 2D column of the velocity portion of B~ corresponds to a particular unit . Pushing vectors corresponding to the same unit in panel A and panel B are given the same color . See Visualizing an extracted internal model for additional details on the pushing vectors in panels A and B . ( C ) Unit-by-unit comparison of pushing vectors from the BMI mapping ( circles ) and internal model ( squares ) from panel A . When looking across all units , there was no consistent structure in the differences between pushing vectors through the internal model versus through the BMI mapping . Some units’ pushing vectors were similar through the BMI mapping and the subject’s internal model , whereas other units’ pushing vectors showed substantial differences . Despite these differences , some patterns of neural activity resulted in similar velocities through the internal model and the BMI mapping ( Figure 4A ) , whereas other patterns resulted in different velocities ( Figure 4B ) . Analyzing the high-dimensional population activity enabled the identification of these effects , which could not have been revealed by analyzing the low-dimensional behavior or individual units in isolation . Parameters visualized in panels A–C were taken from representative session A010609 . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 00910 . 7554/eLife . 10015 . 010Figure 3—figure supplement 3 . The explanatory power of IME comes primarily from structure in the high-dimensional neural activity . IME explains cursor errors by identifying two main types of statistical structure in the data . The two types of structure are ( i ) temporal structure in the dynamics of low-dimensional velocities ( v~t ) , and ( ii ) hidden , task-appropriate structure in the high-dimensional neural activity ( utraw ) . Here we asked to what extent each type of structure contributes toward IME’s overall explanatory power . In the internal model of Equation 10 , A~v summarizes temporal dynamics in the low-dimensional velocity predictions , including how velocity feedback informs internal predictions of cursor state and the degree of temporal smoothness across internal velocity predictions within a single whisker . Intended velocity tends to be similar from one timestep to the next , and whiskers with the appropriate temporal structure ( as determined by fitting A~v to the data ) can explain a portion of cursor movement errors . The internal model parameters B~v and b~v reveal features in the high-dimensional neural activity that are consistent with straight-to-target movement intent and that are often not reflected in the BMI cursor movements . To determine the relative contribution of structure in the high-dimensional neural data toward the explanatory power of IME , we devised a constrained variant of IME that relies entirely on the high-dimensional neural activity to generate whiskers . Specifically , we constrained A~v=0 in Equation 6 , removing IME’s ability to leverage velocity dynamics within its whiskers while preserving its ability to identify structure in the high-dimensional neural activity ( since no constraints are placed on B~v ) . In this constrained IME variant , termed “neural-only” IME , the subject’s internal prediction of the velocity resulting from neural command utraw is simply v~tt=B~v⁢utraw+b~v . Because neural-only IME does not incorporate temporal smoothing of velocity predictions , it cannot be directly compared to the BMI mapping via the cursor error presented in Figure 3C , which was computed using smoothed cursor velocities . To enable fair comparison with the BMI mapping , we computed cursor errors using single-timestep ( unsmoothed ) velocity commands , vtraw=Bv⁢utraw+bv ( replicated from Equation 8 ) . Here we compared the angular errors of these v~tt ( “neural only” ) and vtraw ( “unsmoothed cursor” ) . Internal models extracted using neural-only IME explain 54% and 46% of unsmoothed cursor errors in monkeys A and C , respectively , demonstrating that there is considerable structure in the high-dimensional neural activity . For reference , we also include the error from unconstrained IME ( i . e . , by fitting A~v to the data; “neural + dynamics”; replicated from Figure 3C ) . In this view , the large difference between unsmoothed cursor errors and neural-only internal model errors ( solid arrow ) represents the explanatory power of the structured high-dimensional neural activity without applying any temporal smoothing and without leveraging visual feedback of cursor velocity . The smaller difference between errors through the neural-only internal model and the unconstrained internal model ( dashed arrow ) demonstrates the additional explanatory power gained by incorporating velocity feedback and temporal smoothing . As in Figure 3C , angular error was defined as the angle by which the cursor would have missed the cursor-target overlap zone had it continued in the direction of vtraw ( BMI mapping ) or v~tt ( internal model ) from position p~tt or pt , respectively ( i . e . , Θp in Figure 2—figure supplement 1 ) . Absolute angular errors were first averaged within each trial , then averaged across all trials . Error bars indicate ± SEM ( : p<10-5 , two-sided Wilcoxon test; monkey A: n = 5908 trials; monkey C: n = 4577 trials ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01010 . 7554/eLife . 10015 . 011Figure 3—figure supplement 4 . IME does not explain cursor errors when fit to low-dimensional behavior . We fit neural-only IME using low-dimensional behavior ( i . e . , cursor velocity ) in place of the high-dimensional neural activity . Extracted internal models result in errors similar to or even larger than those of the BMI cursor ( i . e . , internal model errors are not substantially smaller than cursor errors , as in Figure 3C ) . This suggests that the explanatory power of the IME framework is largely due to reliable structure in the high-dimensional neural activity , as opposed to structured errors in the low-dimensional behavior . Further , these results suggest that our ability to identify internal model mismatch might not have been possible given only behavioral measurements . Error bars indicate ± SEM ( : p<0 . 05; : p<0 . 001 , two-sided Wilcoxon test; monkey A: n = 5908 trials; monkey C: n = 4577 trials ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01110 . 7554/eLife . 10015 . 012Figure 3—figure supplement 5 . Subjects could readily produce the entire range of movement directions through the BMI mapping . ( A ) Observed movement velocities from representative session A072208 . Velocities were computed using single-timestep spike counts ( i . e . , vtraw from Equation 8 ) . ( B ) The distribution of gaps between adjacent observed movement directions from panel A . ( C ) The distribution of the mean within-session direction gap across all intuitive sessions . Red triangle indicates the mean gap from the data in panel B . Since movement direction can be noisy at low speeds , we also analyzed these distributions after withholding observed velocities with low speed ( i . e . , points near the origin in panel A ) . Results ( as in panels B and C ) were nearly identical ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01210 . 7554/eLife . 10015 . 013Figure 3—figure supplement 6 . Internal model mismatch is not an artifact of correlated spiking variability . To determine whether our finding of internal model mismatch could be an artifact of correlated spiking variability , we simulated noisy neural commands according to the alternate hypothesis that the internal model was well-matched to the BMI mapping . We could then ask whether our main finding of internal model mismatch ( Figure 3C ) could be explained by the properties of the noise in the real data . Under this alternative hypothesis , the subject generates idealized neural commands that , prior to corruption by noise , would produce the desired movement direction through the BMI mapping . We simulated these idealized neural commands and their noisy counterparts by sampling from the recorded neural activity ( see Assessing whether internal model mismatch could appear as a spurious result due to correlated spiking variability ) . ( A ) By construction , the idealized neural commands had near-zero errors through the BMI mapping . When interpreted through the extracted internal model , those patterns produced substantially larger errors . ( B ) After corruption by noise , which was matched to the statistics of the recorded neural activity , simulated neural commands had substantially larger errors through the extracted internal model than through the BMI mapping . Thus when assuming the alternative hypothesis , simulation results contrast with the results from the real data ( Figure 3C ) . Furthermore , errors increased by roughly the same amount through both the BMI mapping and the internal model . Taken together , these simulation results suggest that our main finding of internal model mismatch cannot simply be explained by extracted internal models that were better fit to the noise properties of the real data , relative to the BMI mapping . Error bars ( barely visible ) indicate ± SEM ( p<10-5 , two-sided Wilcoxon test; panel A: n = 1152 ( monkey A ) and n = 576 ( monkey C ) idealized neural commands; panel B: n = 125 , 459 ( monkey A ) and n = 102 , 689 ( monkey C ) simulated neural commands ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01310 . 7554/eLife . 10015 . 014Figure 3—figure supplement 7 . IME does not explain cursor errors when fit to neural commands that do not contain high-dimensional structure . We applied neural-only IME to datasets in which we shuffled the neural activity in the null-space of Bv , while preserving the neural activity in the row-space of Bv . By design , these shuffled datasets result in exactly the same velocities through the BMI mapping ( “unsmoothed cursor” bars here exactly match those in Figure 3—figure supplement 3 ) . However , any remaining structure in the high-dimensional neural activity is scrambled and as such cannot be leveraged by IME to explain errors . As shown , internal models extracted by neural-only IME from the shuffled data result in errors similar to or even larger than those from the BMI cursor . These results further demonstrate that the explanatory power of the IME framework is largely due to reliable structure in the high-dimensional neural activity and not due to overfitting noise in the data . Error bars indicate ± SEM ( n . s . : not significant , p>0 . 05; p<0 . 001 , two-sided Wilcoxon test; monkey A: n = 5908 trials; monkey C: n = 4577 trials ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01410 . 7554/eLife . 10015 . 015Figure 3—figure supplement 8 . A simplified alternative internal model is not consistent with the data . The central principles of the IME framework are that the subject internally predicts the current cursor position based on an internal model , and the subject aims straight to the target from that predicted position through the internal model . We asked whether we could account for the data using a simpler form of the internal model , which incorporates straight-to-target aiming without internal prediction . In this alternative model , the subject aims straight to the target from the most recent cursor position available from visual feedback , rather than from an internal forward prediction of cursor position . As in IME , this “aim-from-feedback” approach involves an internal model that need not match the BMI mapping . We fit this model via linear regression using the same feedback delays determined from the BMI behavior ( τ from Figure 2A ) . At timestep t , the intended aiming direction was assumed to be straight to the target center from the feedback cursor position , pt-τ , and intended speed was taken to match that of the single-timestep velocity command through the BMI mapping , vtr⁢a⁢w . These aiming directions were regressed against single-timestep spike counts , i . e . , the corresponding utraw , to yield an internal model . Internal models were fit and errors were evaluated using the same cross-validation practices used to evaluate the IME framework . Shaded regions ( barely visible ) indicate ± SEM ( monkey A: n = 33 , 660 timesteps across 4489 trials; monkey C: n = 31 , 214 timesteps across 3639 trials ) . If subjects intend to drive the cursor from the position given by the most recent visual feedback , internal models fit according to this aim-from-feedback principle should predict intended velocities that point closest to targets when originating from feedback cursor positions ( i . e . , cursor positions that lag the recorded neural activity by τ timesteps ) . This was not the case . We found that internal models fit according to the aim-from-feedback principle result in cross-validated velocity commands that point closest to targets when originating from cursor positions more recent than those from the most recently available visual feedback ( i . e . , curves do not have clear minima at the red lines ) . This finding is consistent with subjects aiming straight to the target from an up-to-date internal prediction of the current cursor position ( Figure 2C ) . Because of mismatch between the subject’s internal model and the BMI mapping , the subject’s up-to-date estimate of cursor position need not match the actual current cursor position , and as such , we should not necessarily expect minima in these curves at lag = 0 . Given this evidence that subjects perform some sort of internal tracking , we implemented a number of different internal models to dissect the exact form of that tracking process . Specifically , we implemented internal models that perform no tracking ( aim-from-feedback , presented here ) , tracking without previously issued motor commands ( data not shown ) , and tracking with previously issued motor commands ( main results , which use Equations 9–13 ) , among others . Empirically , these internal models all yield similar cross-validated errors . The explanation for this similarity is that the different formulations identify similar high-to-low dimensional mappings that capture the subject’s intent to move straight to the target , and the direction of these intended commands tends to dominate any effect of the position from which those commands originate . This reasoning is consistent with our finding that cursor errors are better explained by structure in the high-dimensional neural activity than by temporal structure in the cursor kinematics ( Figure 3—figure supplement 3 ) , structure in low-dimensional behavior ( Figure 3—figure supplement 4 ) , or low-dimensional structure in the recorded neural activity ( Figure 3—figure supplement 7 ) . We ultimately chose to present results from the internal model of Equations 9–13 because it performs as well as any internal model we tested , and its formulation is consistent with a number of prominent studies evidencing the use of internal copies of motor commands ( Sommer , 2002; Scott , 2004; Miall et al . , 2007; Crapse and Sommer , 2008; Shadmehr and Krakauer , 2008; Sommer and Wurtz , 2008; Huang et al . , 2013; Azim et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 015 where x~t is the subject’s internal prediction about the cursor state ( position and velocity ) , ut is a vector of recorded neural activity , and A~ , B~ , and b~ are the parameters of the subject’s internal model . This form of the internal model was chosen to be analogous to the BMI mapping from Equation 1 so that the actual BMI mapping lies within the family of internal models that we consider . Additionally , this formulation aligns with recent studies of skeletomotor ( Shadmehr and Krakauer , 2008 ) and oculomotor ( Frens , 2009 ) control , and a vast literature of control theory ( Anderson and Moore , 1990 ) . The primary concept of the IME framework is that , at each timestep , the subject internally predicts the current cursor state by recursively applying Equation 2 ( starting from the most recently available sensory feedback ) and generates neural activity consistent with aiming straight to the target relative to this internal prediction ( see the 'Framework for internal model estimation ( IME ) ' subsection in 'Materials and methods' and Figure 3—figure supplement 1 ) . At each timestep , IME extracts the entire time-evolution of the subject’s internal state prediction using Equation 2 as an internal forward model . This evolution can be visualized in the form of a whisker ( Figure 3B ) that begins at the cursor position of the most recently available feedback and unfolds according to the extracted internal model . At each new timestep , the subject forms a new internal prediction that incorporates newly received visual feedback . If the internal model exactly matches the BMI mapping , the subject’s internal predictions would exactly match the cursor trajectory . A visualization of an example internal model and BMI mapping is given in Figure 3—figure supplement 2 . The central hypothesis in this study is that movement errors arise from a mismatch between the subject’s internal model of the BMI and the actual BMI mapping . The alternative to this hypothesis is that the subject’s internal model is well-matched to the BMI mapping , and movement errors result from other factors , such as “noise” in the sensorimotor system , subjects’ inability to produce certain patterns of neural activity , or subjects disengaging from the task . Our key finding is that recorded neural commands were markedly more consistent with the task goals when interpreted through subjects’ internal models than when viewed through the BMI mapping ( Figure 3C ) . Subjects’ internal models deviated from the actual BMI mappings such that control errors evaluated through extracted internal models were substantially smaller than actual cursor errors: extracted internal models explained roughly 65% of cursor movement errors ( 70% , monkey A; 59% , monkey C ) . Although this finding does not preclude other factors ( e . g . , spiking noise or subject disengagement ) from contributing toward movement errors , it does suggest their contribution is substantially smaller than previously thought , due to the large effect of internal model mismatch . We found that the majority of the explanatory power of extracted internal models was in their ability to identify structure in the high-dimensional neural activity ( Figure 3—figure supplement 3 ) . This structure was captured in the internal model by the mapping from high-dimensional neural activity to low-dimensional kinematics ( B~ in Equation 2 ) , which need not match the BMI mapping ( B in Equation 1 ) . Consistent with this finding , internal models fit to low-dimensional behavior rather than high-dimensional neural activity were not able to explain cursor errors ( Figure 3—figure supplement 4 ) . That a majority of cursor errors can be explained by mismatch of the internal model is not to say that control through the BMI mapping was poor–in fact control was proficient and stable ( Figure 1B and Figure 1—figure supplement 1 ) . Rather , extracted internal models predicted movements that consistently pointed straight to the target , regardless of whether the actual cursor movements did ( Figure 4A ) or did not ( Figure 4B and Figure 4—figure supplement 1 ) point straight to the target . On most trials , BMI cursor trajectories proceeded roughly straight to the target ( Figure 4A ) . On these trials , internal model predictions aligned with actual cursor movements , resulting in small errors through both the BMI mapping and the extracted internal model . In a smaller subset of trials , actual cursor movements were more circuitous and thus had relatively large errors . Previously , the reason behind these seemingly incorrect movements was unknown , and one possibility was that the subject simply disengaged from the task . When interpreted through the extracted internal model , however , neural activity during these circuitous trials appears correct , suggesting that the subject was engaged but was acting under an internal model that was mismatched to the BMI mapping ( Figure 4B and Figure 4—figure supplement 1 ) . In other words , when armed with knowledge of the subject’s internal model , outwardly irrational behavior ( i . e . , circuitous cursor movements ) appears remarkably rational . Across all trials , the majority of neural activity patterns had low or zero error as evaluated through extracted internal models , regardless of whether errors of the actual cursor movements ( i . e . , through the BMI mapping ) were large or small ( Figure 4C and Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 10015 . 016Figure 4 . Neural activity appears correct through the internal model , regardless of how the actual cursor moved . ( A ) Typical trial in which the cursor followed a direct path ( black ) to the target . Internal model predictions ( red whiskers ) also point straight to the target . ( B ) Trial with a circuitous cursor trajectory . Internal model predictions point straight to the target throughout the trial , regardless of the cursor movement direction ( same color conventions as in panel A ) . ( C ) Timestep-by-timestep distribution of BMI cursor and internal model errors . Neural activity at most timesteps produced near-zero error through the internal model , despite having a range of errors through the BMI mapping . ( D ) Hypothetical internal model ( red ) and BMI mapping ( black ) relating 2D neural activity to a 1D velocity output . This is a simplified visualization of Equations 1 and 2 , involving only the B and B~ parameters , respectively . Each contour represents activity patterns producing the same velocity through the internal model ( v~ , red ) or BMI mapping ( v , black ) . Because of internal model mismatch , many patterns result in different outputs through the internal model and the BMI . However , some patterns result in the same output through both the internal model and the BMI ( gray line ) . Here we illustrate using a 2D neural space and 1D velocity space . In experiments with q-dimensional neural activity and 2D velocity , activity patterns producing identical velocities through both the internal model and the cursor span a ( q-4 ) -dimensional space . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01610 . 7554/eLife . 10015 . 017Figure 4—figure supplement 1 . IME whiskers consistently point to the target regardless of cursor movement direction . Additional example cursor trajectories ( black ) are overlaid with cross-validated predictions from extracted internal models ( red whiskers ) , as in Figure 4A , B . Each trial was held-out when fitting the internal model used to generate its whiskers . Each whisker shows the subject’s internal belief of how the cursor trajectory evolved , beginning from the most recently available visual feedback of cursor position ( black dots ) to the subject’s up-to-date prediction of the current cursor position ( red dots ) . The final whisker segments ( red line beyond each red dot ) represent the subject’s intended velocity command . Trials were selected to highlight differences between extracted internal models and the BMI mappings . In these trials , black cursor trajectories at times appear irrational with respect to targets , yet internal models reveal whiskers that consistently point toward the targets . Averaged cursor and internal model errors within each of these trials are shown in Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01710 . 7554/eLife . 10015 . 018Figure 4—figure supplement 2 . Errors from trials in Figure 4—figure supplement 1 highlighted on the distribution of errors across trials . Letters correspond to trials from Figure 4—figure supplement 1 . Format is similar to that of Figure 4C , but there histograms were constructed from single-timestep errors . Here , errors were averaged across all timesteps within each trial , allowing for direct correspondence to the trials shown in Figure 4—figure supplement 1 . Data are from monkey A only . Monkey C data are qualitatively similar ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 018 When cursor trajectories were circuitous , it was not uncommon for some internal model predictions ( whiskers ) to match the actual cursor movement while others did not , even within the same trial ( Figure 4B ) . Given a single internal model , how can some patterns of neural activity result in whiskers aligned to the cursor trajectory , while others patterns produce whiskers that deviate from the cursor trajectory ? This is possible due to mathematical operation of mapping from high-dimensional neural activity patterns to low-dimensional cursor states . Figure 4D provides a conceptual illustration of a simplified BMI mapping: ( 3 ) vt=But and a simplified internal model: ( 4 ) vt~=B~⁢ut each of which relies only on a mapping ( B or B~ ) from neural activity ( ut ) to cursor velocity ( vt or v~t ) . We focus on B and B~ here ( without considering A , b , A~ , and b~ from Equations 1 and 2 ) because of the aforementioned finding that the majority of the internal model mismatch effect is captured by differences between B and B~ ( Figure 3—figure supplement 3 ) . Given a mismatched BMI mapping ( black lines ) and internal model ( red lines ) , many neural activity patterns will produce different velocities through the BMI mapping versus the internal model . However , a subset of activity patterns ( gray line ) will produce identical velocities through both the BMI mapping and the internal model . These patterns lie in the nullspace of B-B~ ( i . e . , solutions to the equation But=B~⁢ut ) . In the example trials shown in Figure 4A , B and Figure 4—figure supplement 1 , internal model predictions ( red ) that match the actual cursor movement ( black ) correspond to neural activity patterns along the gray line in Figure 4D . Predictions not matching the cursor movement correspond to neural activity patterns anywhere off the gray line in Figure 4D . The data presented thus far support our central hypothesis that internal model mismatch is a primary source of movement errors . Next we asked whether it might be possible to have arrived at this result under the alternate hypothesis that the internal model is well-matched to the BMI mapping . We address two specific cases of this alternative hypothesis and show that they do not explain the observed effect of internal model mismatch . First , we explored the possibility that the subject might have a well-matched internal model , but has systematic difficulties producing the neural activity patterns required to drive the cursor in all directions in the 2D workspace using the BMI mapping . This could result in an estimated internal model that appears to be mismatched to the BMI mapping . Although M1 cannot readily produce all possible patterns of high-dimensional neural activity ( Sadtler et al . , 2014 ) , we observed that subjects could readily produce the full range of movement directions through the BMI mapping ( Figure 3—figure supplement 5 ) . Gaps between producible movement directions were typically less than 1/4 of a degree , which is substantially smaller than the cursor errors shown in Figure 3C . This suggests that our main finding of internal model mismatch cannot be explained by subjects’ inability to produce particular neural activity patterns . Second , we explored the possibility that the subject intended to produce neural commands that were correct according to the BMI mapping , but that those intended commands were corrupted by “noise” that is oriented such that errors appear smaller through the extracted internal model than through the BMI mapping . Here we define noise as spiking variability not explained by the desired movement direction under the BMI mapping . If spiking variability is correlated across neurons , it is possible to identify a mapping that best attenuates that variability . To determine whether correlated spiking variability could explain the effect of internal model mismatch , we simulated neural activity according to this alternative hypothesis in a manner that preserved the statistics of the real data ( Figure 3—figure supplement 6 ) . If this simulation produced results that match our findings from the real data , it would indicate that our main finding can be explained by the alternate hypothesis . However , this was not the case . Simulated neural activity was more consistent with the BMI mapping than the extracted internal model , which contrasts with our finding from the recorded neural activity . To further validate the main results presented above , we implemented four statistical controls . First , we ensured that our findings were not simply artifacts of overfitting the data . Second , we removed the high-dimensional structure from the neural activity while preserving the cursor movements , and show that resulting extracted internal models no longer provided explanatory power . Third , we ensured that internal model predictions do not trivially point toward the targets . Finally , we explored a variety of forms for the internal model and found that a simplified form does not account for the data . Here we describe each of these four statistical controls in additional detail . One possible concern when interpreting the findings presented above is that internal models might be simply overfitting the data . To rule out this possibility , all findings presented throughout this paper are cross-validated ( see the 'Computing cross-validated internal model predictions' subsection in 'Materials and methods' ) . Internal models were fit using a subset of trials as training data . Then , trials that were held out during fitting were used to evaluate each extracted internal model . If the extracted internal models had overfit the training data , we would expect those internal models to generalize poorly to the held-out data . However , this was not the case . Internal models explained the majority of cursor errors in the held-out data ( Figure 3C ) , demonstrating that extracted internal models captured real , task-relevant structure in the recorded neural activity . In addition to properly cross-validating our results , we performed a control analysis to show that extracted internal models identified reliable , task-appropriate structure in the high-dimensional neural activity . Here we extracted internal models using neural activity that had been shuffled across timesteps in a manner that preserved the cursor movements through the BMI mapping ( Figure 3—figure supplement 7 ) . If our results could be explained by internal models that simply overfit noise in the data , we would expect internal models fit to these shuffled data data sets to again explain a majority of cursor errors . However , internal models extracted from these shuffled data sets could no longer explain cursor errors , indicating that IME does not identify effects when they do not exist in the data . This result is consistent with our findings that the majority of the explanatory power of extracted internal models relies on structure in the high-dimensional neural activity ( Figure 3—figure supplement 3 ) , and that cursor errors cannot be explained by internal models when high-dimensional neural activity is replaced by low-dimensional behavioral measurements during model fitting ( Figure 3—figure supplement 4 ) . If an internal model prediction points toward the target , it is not trivially due to our inclusion of straight-to-target aiming during model fitting ( see the 'Computing cross-validated internal model predictions' subsection in 'Materials and methods' ) . Although target positions were used during model fitting , they were never used when computing internal model predictions from the data ( e . g . , when constructing the whiskers in Figure 3B , Figure 4A , B , and Figure 4—figure supplement 1 ) . Each whisker was constructed in a held-out trial using only visual feedback ( consisting of a single timestep of cursor position and velocity ) , the recorded neural activity up through the current timestep , and the internal model extracted from the training data . Because of our aforementioned cross-validation procedures , when the neural command ut is used to compute the movement error at timestep t , that neural command had not been seen previously ( i . e . , it was not used when fitting the internal model , when estimating the subject’s internal cursor state prediction , when calibrating the BMI mapping , nor when determining the current position of the actual BMI cursor ) . A whisker that points straight to the target in the held-out data thus reveals that , when interpreted through the subject’s internal model , the recorded neural activity would have driven the cursor straight to the target . Finally , we explored a variety of approaches to modeling the subject’s internal tracking process and found that models demonstrated similarly high degrees of explanatory power as long as they could capture high-dimensional structure in the neural activity . However , a simplified internal model that does not account for any form of internal forward prediction was not consistent with our data ( Figure 3—figure supplement 8 ) . A major limitation in BMI performance is the ability to control cursor speed ( Gilja et al . , 2012; Golub et al . , 2014 ) . Gilja et al . ( 2012 ) and Golub et al . ( 2014 ) have proposed solutions to improve control of BMI speed ( in particular , with respect to stopping the BMI cursor at targets ) . However , it is still an open question as to why BMI speed control is deficient in the first place . In addition to explaining the subjects’ aiming errors , we asked whether mismatch between the internal model and BMI mapping could also explain subjects’ difficulty in controlling cursor speed . Using the extracted internal model , we could compare the subject’s intended speed ( from the internal model ) to the speed of the actual BMI cursor at each timestep . We found that low intended speeds were systematically overestimated , and high intended speeds were systematically underestimated by the BMI mapping ( Figure 5A ) . Furthermore , we discovered that the subjects intended to hold the cursor steadier during the initial hold period and move the cursor faster during the movement than what occurred during experiments ( Figure 5B ) . Note that we make no assumptions about movement speed when extracting the internal model ( see the 'Framework for internal model estimation ( IME ) ' subsection in 'Materials and methods' ) . 10 . 7554/eLife . 10015 . 019Figure 5 . Internal model mismatch limits the dynamic range of BMI cursor speeds . ( A ) BMI cursor speeds across the range of intended ( i . e . , internal model ) speeds . At low intended speeds , BMI speeds were higher than intended , whereas for mid-to-high intended speeds , BMI speeds were lower than intended . Shaded regions indicate ± SEM . ( B ) During the hold period prior to target onset , intended speeds were significantly lower than those produced through the BMI mapping . During movement , intended speeds were significantly higher than those produced through the BMI . Error bars indicate ± SEM ( p<10-5 , two-sided Wilcoxon test; monkey A: n={5006 , 5908} trials; monkey C: n={3008 , 4578} trials ) . In panels A and B internal models were used to predict intended speed on trials not used during model fitting . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 01910 . 7554/eLife . 10015 . 020Figure 5—figure supplement 1 . A unit-by-unit example of internal model mismatch limiting cursor speed dynamic range . ( A ) Spike counts during a single example timestep across 22 units . Circles indicate a zero spike count . This example timestep was recorded mid-movement , when intended speed should be at its maximum . ( B ) Pushing vectors from Figure 3—figure supplement 2 scaled by the spike counts shown in panel A . Dashed arrows indicate the direction and magnitude of the velocity components of offset vectors b ( left ) and b~ ( right ) , which are meant to effectively zero-out the velocity expected when neurons fire at their baseline rates . Straight-to-target directions ( green stars ) are shown relative to the current cursor position ( left ) or the internal-model predicted current cursor position ( right ) . ( C ) Actual ( left ) and intended ( right ) cursor velocities corresponding to the spikes counts shown in panel A . Each resultant pushing vector ( arrows ) is the sum of the offset term and all weighted pushing vectors from panel B . These vectors represents the contribution of the single-timestep spike count from panel A toward the cursor ( left ) or internal model-predicted ( right ) velocity ( i . e . , without considering smoothing across previous time steps ) . Here , intended speed ( magnitude of red arrow ) is greater than the actual speed through the BMI mapping ( magnitude of black arrow ) . This difference between intended and actual speed arises because the pattern of activated units in the example spike count vector had similar pushing directions through the internal model , resulting in a coordinated push . Through the BMI mapping , however , the same spike count activated a more diffuse set of pushing directions , resulting in a “co-contraction” of units that push against each other more than they did through the internal model . These spike counts were recorded mid-movement , when intended speed ought to peak . Thus , this example demonstrates the systematic underestimation of movement speed through the BMI mapping that results from internal model mismatch , consistent with aggregate findings in Figure 5B ( movement bars ) . Also , consistent with aggregate findings in Figure 3C , the internal-model predicted velocity ( red arrow ) points closer in direction to the target ( green star ) than does the actual cursor velocity ( black arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 020 To gain insight into this speed mismatch , we can use extracted internal models to examine the discrepancies between intended and actual speeds at the level of individual units and on the timescale of a single 33-ms timestep ( Figure 5—figure supplement 1 ) . These systematic differences between intended and actual cursor speeds indicate that internal model mismatch limits realizable dynamic range of BMI movement speeds . These findings suggest that the longstanding deficiencies in BMI speed control may be a consequence of internal model mismatch . A key feature of an internal model is its ability to adapt . Arm reaching studies have demonstrated behavioral evidence of internal model adaptation ( Shadmehr and Mussa-Ivaldi , 1994; Thoroughman and Shadmehr , 2000; Joiner and Smith , 2008; Taylor et al . , 2014 ) . Behavioral learning has also been demonstrated in the context of BMIs ( Taylor , 2002; Carmena et al . , 2003; Jarosiewicz et al . , 2008; Ganguly and Carmena , 2009; Chase et al . , 2012; Sadtler et al . , 2014 ) . While these BMI studies suggest that subjects adapt their internal models to better match the BMI mapping , a direct assessment has been difficult without access to those internal models . With the ability to extract a subject’s internal model , here we asked whether extracted internal models adapt in accordance with perturbations to the BMI mapping ( Figure 6 ) . In one monkey , an initial block of trials under an intuitive BMI mapping was followed by a block of trials under a perturbed BMI mapping . All data analyzed prior to this section were recorded during intuitive trials . The intuitive and perturbed mappings were of the form of Equation 1 , but each used different values in the matrix B . The perturbed BMI mapping effectively rotated the pushing directions of a subset of the recorded units , such that the global effect resembled a visuomotor rotation ( see the 'Behavioral task' subsection in 'Materials and methods' ) . Previous studies have shown that perturbations of this type can be learned by monkeys ( Wise et al . , 1998; Paz et al . , 2005; Chase et al . , 2012 ) . 10 . 7554/eLife . 10015 . 021Figure 6 . Extracted internal models capture adaptation to perturbations . ( A ) Cross-validated angular errors computed by interpreting monkey A neural activity through BMI mappings and internal models . The intuitive BMI mapping ( blue ) defined cursor behavior during the intuitive and washout trials . The perturbed BMI mapping ( red ) defined cursor behavior during the perturbation trials . The late intuitive internal model ( yellow ) was extracted from the last 48 intuitive trials ( yellow bar ) . A time-varying internal model ( green ) was extracted from a moving window of the 48 preceding trials . Values were smoothed using a causal 24-trial boxcar filter and averaged across 36 experiments . ( B ) Differences between monkey A’s time-varying internal model and the BMI mappings , assessed through the high-dimensional neural activity . For each round of 16 trials , neural activity from those trials was mapped to velocity through the time-varying internal model , the intuitive BMI mapping , and the perturbed BMI mapping . Signed angles were taken between velocities computed through the time-varying internal model and the intuitive BMI mapping ( blue ) and between velocities computed through the time varying internal model and the perturbed BMI mapping ( red ) . Values were averaged across 36 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10015 . 021 For each experiment , we interpreted recorded population activity through the intuitive and perturbed BMI mappings , as well as through two views of the subject’s internal model: a time-varying internal model extracted from a moving window of 48 trials , and a late intuitive internal model extracted from the last 48 intuitive trials . We could then quantify changes in the subject’s internal model and assess which BMI mapping or internal model was most consistent with the neural activity , relative to task goals ( Figure 6A ) . To avoid circularity , trials used to evaluate the BMI mappings and internal models were not used when fitting the internal models nor when calibrating the BMI mappings . Errors through the intuitive BMI mapping describe the actual cursor performance during the intuitive and washout trials ( thick blue traces; analogous to cursor errors in Figure 3C ) , and how that mapping would have performed had it been in effect during the perturbation trials ( thin blue trace ) . Similarly , errors through the perturbed BMI mapping describe the actual cursor performance during the perturbation trials ( thick red trace ) , and how that mapping would have performed had it been in effect during the intuitive and washout trials ( thin red traces ) . Behavioral learning was evident in that errors through the perturbed BMI mapping were large in early perturbation trials and decreased continuously throughout the perturbation trials . A detailed characterization of this behavioral learning can be found in ( Chase et al . , 2012 ) . Our key finding in this analysis is that extracted internal models adapted in a manner consistent with the BMI perturbations ( Figure 6B ) . During the perturbation trials , the time-varying internal model adapted to better match the perturbed BMI mapping ( red trace trends toward zero ) . Similarly , during the washout trials , the time-varying internal model adapted to better match the intuitive BMI mapping ( blue trace trends toward zero ) . Had the subject’s internal model not adapted , or if the adaptation was not reflected in the extracted internal model , we would expect the traces in Figure 6B to be flat . Rather than being static entities , the extracted internal models were dynamic with timescales independent of experimenter-induced changes to the BMI mapping . Consistent with our central hypothesis , internal model mismatch was present throughout the intuitive , perturbation , and washout trials . During intuitive trials , errors through the time-varying internal model were substantially lower than errors through the intuitive BMI mapping ( green trace lower than blue trace in Figure 6A ) , which is consistent with our main findings in Figure 3C . Because the subject’s internal model adapts , errors through the time-varying internal model remained substantially smaller than errors through the BMI mappings across the perturbation and washout trials as well ( green trace remains low across Figure 6A ) . Although behavioral errors decreased over the course of the perturbation and washout trials , internal model mismatch was still present following adaptation ( red and blue traces are nonzero during late perturbation and washout trials , respectively , in Figure 6B ) . It could have been that this internal model mismatch was only substantial during early intuitive trials before the subject had accrued enough experience to form a stable internal model . This was not the case . The subject’s internal model was stable throughout the intuitive session , as evidenced by the nearly constant angular differences between velocities mapped through the time-varying internal model and the BMI mapping ( red and blue traces are roughly flat in Figure 6B ) and the nearly identical errors through the time-varying internal model and the ( static ) late intuitive internal model ( green and yellow traces overlap in Figure 6A wherever cross-validated errors can be computed ) . Consistent with a stable internal model , behavioral performance was stable throughout the intuitive trials ( blue trace is flat in Figure 6A ) , and internal model mismatch explained a steady fraction of behavioral errors ( green trace is also flat , and substantially lower than blue trace ) . During the perturbation session , the subject’s internal model diverges from this stabilized state ( yellow trace diverges from green trace in Figure 6A and both traces are non-constant in Figure 6B ) . In this study , we considered neural population activity recorded in M1 . It is reasonable to ask how it is possible to deduce anything about internal models from M1 activity if we are not recording i ) signals from the neural circuits that implement internal models ( e . g . , cerebellum ) , nor ii ) the internal copy signals that enable internal model computations . The rationale is the following . First , the neural activity recorded in M1 is likely to be downstream of the internal model computations , whether they be in cerebellum ( Shadmehr , 1997; Pasalar et al . , 2006; Miall et al . , 2007; Lisberger , 2009; Huang et al . , 2013 ) , posterior parietal cortex ( Shadmehr , 1997; Mulliken et al . , 2008 ) , dorsal premotor cortex ( Shadmehr , 1997 ) , or elsewhere . Thus , the internal model is likely to influence the neural activity produced in M1 . By relating the neural activity recorded in M1 to the behavioral task on a moment-by-moment basis , we should be able to infer properties of the upstream internal model . Second , previous studies indicate that internal copy signals ( e . g . , efference copy , corollary discharge ) carry information about movement intent , in particular a copy of the movement intent that M1 sends to the motor effector ( Crapse and Sommer , 2008; Huang et al . , 2013; Schneider et al . , 2014; Azim et al . , 2014 ) . Although we are not directly recording the internal copy signal , the information in the internal copy relevant to movement intent is likely also present in the recorded M1 activity , and this is what we leveraged . In short , we make no claims about the neural circuitry implementing internal models , but rather we infer statistical properties of the internal models from their downstream consequences in M1 . Using this rationale , we extracted internal models from M1 population activity . We chose to capture the subject’s internal model using a forward model framework ( Equation 2 and Figure 3A ) because it is both highly interpretable and consistent with a large body of behavioral and computational studies ( Shadmehr and Krakauer , 2008; Frens , 2009 ) . Our results do not preclude the use of other types of internal models , such as an inverse model ( Shadmehr and Mussa-Ivaldi , 1994; Kawato , 1999 ) , whose acquisition and function is believed to be tightly coupled to that of the forward model ( Wolpert and Kawato , 1998 ) . We presented four important lines of evidence that indicate that the extracted internal models are meaningful , and not a result of logical circularity during model fitting or overfitting to noise in the data . First , extracted internal models explain a majority of behavioral errors on trials not seen during model fitting ( Figure 3C ) . Here , extracted internal models identified structure in the high-dimensional neural activity that indicated straight-to-target movement intent , even when the cursor behavior was circuitous . Internal model predictions on held-out trials could not trivially point toward the targets because that held-out neural activity had not been used during model fitting , and because target positions were never used when constructing internal model predictions from held-out trials . Second , the finding that intended speed is better predicted by internal models than the BMI mapping ( Figure 5B ) lends an additional independent validation of those internal models , since no assumptions were made about intended movement speed when fitting internal models . Third , when we perturbed the BMI mapping , extracted internal models revealed adaptation consistent with the particular perturbations ( Figure 6 ) . Finally , we performed a series of scientific and statistical control analyses . We showed that our data are not consistent with two versions of the alternative hypothesis , in which the subject’s internal model is well-matched to the BMI mapping ( Figure 3—figure supplement 5 and Figure 3—figure supplement 6 ) . Further , we asked whether extracted internal models could explain the observed behavioral errors without access to structure in the high-dimensional neural activity beyond that which defined cursor movements . We considered two different alterations to the data from which internal models were extracted: one in which we replaced the high-dimensional neural activity with low-dimensional cursor velocities ( Figure 3—figure supplement 4 ) and another in which we shuffled the neural activity in a manner that preserved cursor velocities through the BMI mapping ( Figure 3—figure supplement 7 ) . In both cases , we found that the extracted internal models no longer offered a consistent explanation for the observed behavioral errors , thereby demonstrating that the explanatory power of the extracted internal models does not arise from logical circularity or overfitting . An extracted internal model and a BMI mapping are closely related . They take a similar mathematical form ( Equations 1 and 2 ) and both project high-dimensional population activity to a low-dimensional kinematic space . A key difference between internal models and BMI mappings is that internal models are dynamic entities whose properties can change during motor adaptation . In contrast , the BMI mappings are chosen by the experimenter or by a computer algorithm . Critically , in experiments in which we abruptly applied a perturbed BMI mapping , we found that extracted internal models dynamically adjusted in a manner appropriate for the task and at a timescale independent of changes to the BMI mapping ( Figure 6 ) . The ability to interpret neural activity through the subject’s internal model , while the subject controls the cursor through some BMI mapping ( e . g . , Figure 4A , B , Figure 6A and Figure 4—figure supplement 1 ) , offers a unique glimpse into the subject’s movement intentions , sensory prediction errors , and motor adaptation . Given the substantial fraction of behavioral errors that are explained by internal model mismatch during control under the intuitive BMI mapping , it is perhaps surprising that we did not find evidence of behavioral or internal model adaptation during those trials ( Figure 6 ) . A way to reconcile these findings is that , in contrast to the frequent movement errors experienced after the BMI mapping was perturbed , there was a relative paucity of errors during the intuitive trials . As a result , there may not have been sufficient pressure to improve upon a “good enough” internal model ( Loeb , 2012 ) . Had the subject been given more experience with the same BMI mapping ( Ganguly and Carmena , 2009 ) , the internal model may have converged to the BMI mapping . Nevertheless , our findings indicate that the subject’s learning process may be a key limitation in BMI performance ( Sadtler et al . , 2014 ) . It may be possible to overcome these limitations in the subject’s neural adaptation process through complementary innovations in designing the BMI mapping ( Shenoy and Carmena , 2014 ) . For example , applying an extracted internal model as the BMI mapping might improve performance during closed-loop BMI control . Indeed , a recent study incorporating the concept of internal tracking has demonstrated substantial gains in closed-loop BMI performance ( Gilja et al . , 2012 ) . Future studies will be required to determine whether further improvements in performance might be possible by using the IME framework toward designing the BMI mapping . The insights gained in this study were made possible because we monitored the subject’s high-dimensional neural activity . Because the BMI mapping and the subject’s internal model are high-to-low dimensional mappings , neural activity that was consistently correct under the internal model sometimes resulted in aberrant behavior through the BMI mapping . We would not have been able to observe or explain this phenomenon by analyzing the BMI cursor movements in isolation ( Figure 3—figure supplement 4 ) . In particular , by replacing all instances of neural activity ( i . e . , the ut in Equation 2 ) with actual cursor velocities ( or analogously , with actual hand velocities from an arm reaching task ) , IME becomes limited to predicting the subject’s velocity intent to be a scaled and rotated ( in two-dimensions ) version of the actual velocity . In contrast , access to the high-dimensional neural activity enabled the identification of the subject’s intended movements without constraining them to have a consistent relationship with actual movements . Prior beliefs , and their role in sensation and behavior , have been the focus of many studies , including those on visual perception ( Kersten et al . , 2004; Komatsu , 2006; Berkes et al . , 2011 ) , perceptual decision-making ( Ma and Jazayeri , 2014 ) , and sensorimotor learning ( Körding and Wolpert , 2004; Turnham et al . , 2011 ) . Our work provides a means for extracting a rich representation of prior beliefs ( i . e . , the internal model ) that can combine past sensory input with multi-dimensional neural processes to drive moment-by-moment motor control decisions . We found that outwardly aberrant behavior and behavioral limitations could be explained by taking into account the subject’s prior beliefs . By recording simultaneously from multiple neurons and developing the appropriate statistical algorithms , it may be possible to extract similarly rich prior beliefs in other systems . Two male rhesus macaques ( Maccaca mulatta ) were each implanted with a 96-channel microelectrode array ( Blackrock Microsystems , Salt Lake City , UT ) targeting proximal arm area of primary motor cortex . Signals were amplified , bandpass filtered ( 250 Hz - 8 kHz ) and manually sorted ( Plexon Sort Client , box sort ) with a 96-channel Plexon MAP system ( Plexon , Dallas , TX ) . Recorded neuronal units were either well-isolated single cells or multiple cells that could not be well separated but as a group were tuned to intended movement direction . In each session , we recorded 26 . 0 ± 3 . 4 ( monkey A ) and 39 . 2 ± 3 . 9 ( monkey C ) neuronal units ( mean ± one standard deviation ) . Spike counts were taken in nonoverlapping 33-ms bins throughout the behavioral task ( see 'Behavioral task' ) . All animal procedures were approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh . Subjects modulated neural activity to drive movements of a virtual cursor in a 2D brain-machine interface ( BMI ) task . The cursor ( radius: 7–8 mm , monkey A; 6 mm monkey C ) and targets ( same radii as cursor ) were displayed to the subject on a frontoparallel stereoscopic display ( Dimension Technologies , Rochester , NY ) with a refresh rate of 60 Hz . Display updates were subject to a latency of up to 2 refresh cycles ( 0–33 . 3 ms ) . Target positions were chosen pseudorandomly from a set of 16 evenly-spaced radial targets ( center-to-target distance: 85 mm , monkey A; 72–73 mm , monkey C ) . Each trial began with the cursor at the workspace center , where the subject was required to hold the cursor to visibly overlap a central target ( center hold requirement randomly selected for each trial: 50–350 ms , monkey A; 50–150 ms , monkey C ) . Following completion of the initial hold , a peripheral target appeared , instructing the subject to initiate a cursor movement . Target acquisition was recorded as the first timestep during which the cursor visibly overlapped the peripheral target . Following target acquisition , the subject was required to hold the cursor steady without losing visible overlap between the cursor and target ( target hold requirement randomly selected for each trial: 50–100 ms , monkey A; 50 ms , monkey C ) . A limit was placed on the time between target onset and target acquisition ( 1 . 5–2 s , monkey A; 1 . 2–2 s , monkey C ) . A trial was deemed failed and terminated if visible overlap between cursor and target was lost before satisfying either hold requirement . If all requirements were met , a trial was deemed successful , and the subject was provided with a water reward ( 120 μ⁢l , monkey A; 120–130 μ⁢l , monkey C ) . Arms were restrained , and little to no hand movements were observed ( although hand positions were not recorded ) . The analyzed data were subsets of data from larger experiments . The experimental details for monkey A have been described previously ( all no invisible zone conditions from Chase et al . , 2012 ) . Briefly , each experiment began with roughly 40 trials that were used to calibrate the intuitive BMI mapping ( see 'Calibration of the BMI mapping' ) . Following calibration was a block of 169 ± 8 . 1 successful trials under this intuitive BMI mapping . Next , the BMI mapping was systematically perturbed and held constant for 365 ± 126 successful trials . Each perturbation effectively rotated a random subset of recorded units’ decoded pushing directions ( DPDs ) , as in Figure 3—figure supplement 2B , by a particular angle ( 5 experiments with 25% of units’ DPDs rotated 90∘; 20 experiments with 50% of units’ DPDs rotated 60°; 11 experiments with 100% of units’ DPDs rotated 30° ) . In 33 of 36 experiments , perturbation trials were followed by 360 ± 237 successful washout trials , during which the perturbation was removed , and the BMI mapping was restored to the intuitive mapping . Unless noted otherwise , analyses of monkey A data refer to intuitive trials . Data from the perturbation and washout trials appear only in Figure 6 . Each of the 36 experiments comprising these data took place on a unique day . For monkey C , BMI cursor control alternated between the 2D task ( described above ) and a 3D task ( described below ) . All monkey C trials analyzed in this work came from the 2D task . Each day began with roughly 40–50 trials to calibrate an intuitive BMI mapping . Following calibration , subsequent blocks alternated between the 2D task and the 3D task , with the first of these tasks chosen randomly each day . The 3D task was similar to the 2D task , except that the cursor was allowed to move in 3D , and targets were distributed about the surface of a workspace-centered sphere . Blocks with the 2D task consisted of 277 ± 70 . 4 trials , and blocks with the 3D task consisted of 527 ± 252 trials . Each day consisted of either 3 or 4 blocks . Monkey C experiments did not include trials under a perturbed BMI mapping . The 18 2D blocks analyzed in this work took place on 12 unique days . BMI cursor position and velocity were determined from recorded spike counts according to a BMI mapping: ( 5 ) pt = pt-1+vt-1⁢Δ ( 6 ) vt = Bv⁢ut+bv where at timestep t , pt∈ℝ2 is the cursor position , vt∈ℝ2 is the cursor velocity , Δ=33 ms is the timestep duration , ut∈ℝq is the mean spike count vector recorded simultaneously across q neuronal units over the past 5 timesteps ( 167 ms ) , and Bv and bv are the parameters that map neural activity to cursor velocity . Note that the BMI mapping ( Equations 5 and 6 ) can be written equivalently in the form of Equation 1: ( 7 ) xt = Axt-1 + But + b = ptvt = II·∆00 pt-1vt-1 + 0Bv ut + 0bv where the cursor state , xt , concatenates cursor position and velocity . In some of the following analyses , we required more precise time resolution than could be achieved by analyzing the 5-timestep smoothed velocity commands that drove the BMI cursor ( Equation 6 ) . For fine-timescale analyses , we defined single-timestep ( i . e . , unsmoothed ) velocity commands as: ( 8 ) vtraw = Bv⁢utraw + bv where utraw is the vector of recorded spike counts during the single timestep t , and Bv and bv are the decoding parameters that were applied online , as in Equation 6 . Note that vt in Equation 6 is the average of single-timestep velocity commands , vt-4r⁢a⁢w , … , vtr⁢a⁢w . Calibration of parameters Bv and bv of the intuitive BMI mapping was done in closed-loop and followed the population vector algorithm ( Georgopoulos et al . , 1983 ) . Details on this closed-loop calibration have been published previously in Chase et al . ( 2012 ) . For monkey A , an initial sequence of 8 evenly-spaced radial targets was presented to the subject while the cursor remained stationary at the workspace center . Then , an initial set of BMI parameters was determined by regressing the average spike rates for each trial in this sequence against the corresponding target directions . A second sequence of 8 trials followed , with cursor movements determined by the initial parameter set , but with assistance provided by attenuating velocities perpendicular to target directions . Following this second sequence of trials , new decoding parameters were determined by regressing spike rates from all previous trials against the corresponding target directions . This process was repeated for typically 5 sequences ( 40 trials ) , with less assistance in each subsequent sequence until no assistance was provided . The schedule of assistance was determined on an ad-hoc basis . The intuitive BMI mapping calibrated from these trials was then used for the subsequent block of analyzed trials ( see 'Behavioral task' ) . For monkey C , the first task of each day was randomly selected between the 2D and 3D tasks . If the first task was 2D , calibration followed the same procedure as with monkey A . If the first task was 3D , each calibration sequence consisted of 10 targets equidistant from the workspace center . Eight of these targets were on the corners of a workspace-centered cube . The remaining 2 targets were nearly straight out and straight in along the z-direction , but slightly offset so that the cursor was not visually obscured at the central start position . Because these target directions were specified in 3D , calibration regressions resulted in parameters Bv3⁢D∈ℝ3×q and bv3⁢D∈ℝ3 that could map neural activity to 3D velocity . When the task switched to 2D , the parameters Bv and bv were set to the first two rows of Bv3⁢D and bv3⁢D , respectively , corresponding to mapped velocities in the frontoparallel plane only . These 3D calibrations typically spanned five 10-trial sequences ( 50 trials ) . BMI subjects experience an inherent visual feedback delay . To assess the visuomotor latency experienced by a subject in our BMI system , we measured the elapsed time between target onset and the appearance of target-related activity in the recorded neural population ( Figure 2A ) . To determine the first timestep at which neural activity contained target information , we found the first significant decrease in angular error relative to baseline error . For each trial , baseline error was defined to be the average of absolute angular errors prior to target onset . Here , the angular error at timestep t was defined to be the angle by which the cursor would have missed the target had it continued from its current position , pt , in the direction of the single-timestep velocity command , vtraw , from Equation 8 . Single-timestep commands ( vtraw ) were analyzed here ( as opposed to smoothed cursor velocities , vt ) for improved temporal resolution . Because absolute angular errors range from 0-180∘ , one might reasonably expect baseline error to be roughly 90∘ . Baseline errors shown are less than 90∘ because angular errors were computed relative to the cursor-target overlap zone ( i . e . , taking into account cursor and target radii; see Figure 2—figure supplement 1 ) . When errors were instead computed relative to the target center , baseline errors were roughly 90∘ , and identified latencies were unaffected ( data not shown ) . Had we introduced an arbitrary additional delay to the display updates ( Willett et al . , 2013 ) , we would expect a commensurate increase in the identified feedback delay . Because of the visual feedback delay ( Figure 2A ) , at timestep t the subject cannot yet directly access the timestep t cursor position . To determine whether subjects compensated for the visual feedback delay , we asked whether neural activity recorded at timestep t was more appropriate for the timestep t cursor position or for a previous cursor position . Across a range of lags , d=[-100 ms , … , 300 ms] , we computed the angular errors of single-timestep velocity commands , vtraw ( as in Equation 8 ) , as if they had originated at lagged positions pt-d ( Figure 2B ) . Here , angular errors were defined to be the absolute angle by which the cursor would have missed the target had it originated at position pt-d and continued in the direction of the single-timestep velocity command vtraw , taking into account the radii of the cursor and the target ( i . e . , ΘP in Figure 2—figure supplement 1 ) . This error metric was chosen because it reflects the task goal , that to succeed in a trial , the subject had to to acquire visible overlap between the cursor and the target ( Figure 2—figure supplement 1 ) . By taking into account cursor and target radii , this error metric is influenced by cursor-to-target distance . Specifically , velocity commands originating from positions close to the target will have smaller errors under this definition than the same velocity commands originating far from the target ( Figure 2—figure supplement 1 ) . Without accounting for this distance-to-target bias , absolute angular errors might appear smaller for lags that are less positive because these lagged cursor positions will tend to be closer to the targets than cursor positions with more positive lags ( e . g . , pt-d tends to be closer to the target when d=0 ms than when d=300 ms ) . To ensure that this distance-to-target bias did not influence our conclusions about feedback delay compensation , errors were computed for the same exact subset of cursor positions across lags . This selection process preserves cursor-to-target distances across lags and thus ensures that the same exact error bias is applied at each lag . To this end , we included in this analysis only cursor positions for which all required lags of neural activity were recorded within the corresponding trial . Further , we only considered cursor positions that were presented at least 100 ms following target onset to ensure that recorded neural activity could plausibly reflect target position given a feedback delay of 100 ms . To determine the error value for a particular lag along the curves in Figure 2C , we first averaged all absolute angular errors for that lag within each trial , and then averaged across trials . A preliminary version of this analysis using different experiments has appeared in conference form ( Golub et al . , 2012 ) . The IME framework is a statistical tool we developed to extract from neural population activity i ) a subject’s internal model of the BMI mapping , and ii ) the subject’s timestep-by-timestep internal predictions about the cursor state . The central concept underlying the IME framework is that at each timestep , the subject internally predicts the current cursor position using outdated visual feedback and a recollection of previously-issued neural commands ( representative of efference copy or corollary discharge [Crapse and Sommer , 2008] ) , and issues the next neural command with the intention of driving the cursor straight toward the target from the up-to-date prediction of the current cursor position ( Figure 3B ) . Formally , the IME framework is a probabilistic model defined by Equations 9–14 . The subject’s internal model , as introduced in Figure 3A , is is represented as follows: for k = {t-τ+1 , … , t}: ( 9 ) p~kt = p~k-1t + v~k-1t⁢Δ ( 10 ) v~kt = A~v⁢v~k-1t + B~v⁢ukraw + b~v + wkt where p~kt∈ℝ2 and v~kt∈ℝ2 are the subject’s internal predictions of the timestep k cursor position and velocity when the subject is sitting at timestep t , Δ is the timestep of the BMI system ( 33 ms ) , ukraw∈ℝq is a vector of the spike counts recorded simultaneously across the q neuronal units at timestep k , A~v∈ℝ2×2 , B~v∈ℝ2×q , and b~v∈ℝ2 are parameters capturing the subject’s internal model , and wkt∈ℝ2 is a Gaussian random variable ( with isotropic noise variance , w ) representing internal predictions not captured by the internal model . More specifically , A~ represents the subject’s internal conception of the physical properties of the cursor , and B~ represents the subject’s internal conception of how neural activity drives movement of the cursor . Note that the subject’s internal model in Equations 9 and 10 can be written in the form of Equation 2: ( 11 ) x~kt = A~⁢x~k-1t + B~⁢ukraw + b~ + noise =p~ktv~kt = II⋅Δ0A~v⁢ p~k-1tv~k-1t + 0B~v⁢ukraw + 0b~v + 0wkt where the subject’s internal state prediction , x~tk , includes the internal prediction of cursor position , p~kt , and velocity , v~kt . For simplicity in Equation 2 , we omitted the noise term , the superscript notation , and the distinction between spike count vectors recorded at a single timestep , utraw , and average spike count vectors across 5 timesteps , ut ( more details on smoothing are given below ) . Visual feedback grounds the subject’s internal predictions with reality . At timestep t , the subject’s internal prediction of the cursor position and velocity at the feedback delay ( τ , as discussed in Parameter fitting for the IME framework ) match the most recently available cursor position and velocity from visual feedback: ( 12 ) p~t-τt = pt-τ ( 13 ) v~t-τt = vt-τ The internal model in Equations 9 and 10 is then applied recursively ( i . e . , across k∈{t-τ+1 , … , t} ) to arrive at up-to-date predictions , p~tt and v~tt , about the current cursor state . The resulting set of internal predictions corresponds to the whiskers shown in Figure 3 , Figure 4 , and Figure 4—figure supplement 1 . Finally , we incorporate the notion of straight-to-target aiming intention with: ( 14 ) Gt = p~tt + αt⁢v~tt + rt where Gt∈ℝ2 is the target position , αt∈ℝ+ is a non-negative distance scale parameter , and rt∈ℝ2 is a Gaussian random variable ( with isotropic noise variance , r ) representing internal velocity predictions that do not point straight to the target . Since the target was held constant within each BMI trial , Gt took on the same value for all timesteps corresponding to a particular trial . Intuitively , Equation 14 says that when the subject internally believed the cursor to be at position p~tt , the intended velocity command , v~tt , ought to point in the direction of the target , Gt . The distance scale parameters , αt , allow the data to determine the intended speed ( i . e . , velocity magnitude ) at each timestep . This parameterization allows us to avoid imposing a-priori assumptions about the subject’s intended speed . During model fitting , larger values of αt tend to be learned for timesteps when the distance to target is large ( from p~tt ) , and smaller values tend to be learned when this distance is small . In this manner , there are no assumptions imposed upon intended speed during model fitting . Rather , the learned internal model determines intended speed from the data . Additionally , the linear form of Equation 14 was chosen so all latent variables , {p~ , v~} , and observed variables , {G , uraw} , are jointly Gaussian . Throughout control , new visual feedback continues to arrive , and new neural commands are issued at each timestep . IME captures this progression by including a new set of internal predictions ( i . e . , a new whisker ) at each timestep . For example , at timestep t+1 , the subject receives new feedback about the cursor state , pt-τ+1 and vt-τ+1 , and accordingly forms a new set of internal predictions {p~kt+1 , v~kt+1} for k∈{t-τ+2 , … , t+1} . The full IME probabilistic graphical model is drawn in Figure 3—figure supplement 1 to visually depict this instantiation of Equations 9–14 at each timestep during control . Through Equation 14 we assume that the subject attempts to move the cursor straight to the target from an internal estimate of the current position . We believe that straight-to-target aiming is a reasonable first-order assumption because the BMI cursor , on average , moves straight to the target during proficient control ( see Figure 1B ) . It may be possible to incorporate other movement objectives , such as minimizing endpoint error ( Harris and Wolpert , 1998 ) or movement jerk ( Flash and Hogan , 1985 ) , in the IME framework , which may yield even greater explanatory power . However , at present , there is not clear evidence that these other movement objectives underlie BMI cursor control , so we apply only the basic straight-to-target movement objective in this work . Both the BMI mapping ( Equations 5–7 ) and the internal model representation ( Equations 9–11 ) implement smoothness across BMI cursor velocities and internal velocity predictions , respectively . The details of this smoothing are subtly different between the BMI and the IME framework . To mitigate the effects of neural spiking noise , the BMI mapping smooths cursor velocities by incorporating neural activity at each timestep through the 5-timestep boxcar filter , as described following Equation 6 . Temporal smoothing in internal velocity predictions is achieved through the subject’s internal prior belief about how the internal velocity prediction at one timestep influences the prediction at the next timestep , as encoded by A~v . In a preliminary IME formulation we presented recently in conference form ( Golub et al . , 2013 ) , the subject’s internal state prediction was modeled using position only , rather than using both position and velocity , as we have here . The inclusion of velocity has several important advantages . First , it allows the model to capture the subject using feedback about cursor velocity to internally predict cursor position and velocities . Second , including velocity in the state enables IME to automatically determine the degree of smoothness in internal velocity predictions , based on the data , by fitting an appropriate A~v . We fit IME models using expectation maximization ( EM ) ( Dempster et al . , 1977 ) , a maximum likelihood estimation technique for latent variable models . Training data for each trial consisted of recorded spike counts and actual cursor positions for timesteps beginning at movement onset and ending at target acquisition , as well as the target position for that trial . Movement onset for a given trial was defined as the first timestep at which the cursor speed , projected in the center-to-target direction , exceeded 15% of its maximum from that trial . During the E-step , posterior distributions , P ( {x~}∣{x , uraw , G} ) , are computed over the internal states given a set of model parameters . Intuitively , these posteriors are distributions over whiskers that compromise between satisfying the internal model ( Equations 9 and 10 ) and straight-to-target aiming ( Equation 14 ) . During the M-step , these posterior distributions are used to update the model parameters , A~ , B~ , b~ , w , {αt} , and r . We typically ran EM for 5000 iterations , but allowed fewer iterations if model parameters converged sooner . Although the feedback delay parameter , τ , can be determined using standard model selection techniques ( Golub et al . , 2013 ) , we fixed this parameter ( τ=3 , corresponding to 100 ms , monkey A; τ=3 , corresponding to 133 ms , monkey C ) for simplicity and to remain consistent with our experimental characterization of the visuomotor latency from Figure 2A . Throughout our results , if an internal state prediction ( whisker ) points toward the target , it is not trivially due to our inclusion of straight-to-target aiming into IME ( Equation 14 ) . Rather , whiskers that point toward targets are evidence of real structure in the data . We ensure that whiskers do not trivially point toward targets by using cross-validation techniques whenever evaluating or visualizing extracted internal models and their corresponding internal state predictions ( whiskers ) . For a given experimental session , trials were randomly assigned to folds such that each fold consisted of one trial to each unique target . We employed K-fold cross-validation , where K was the number of folds in a given experimental session . Internal models were fit to the data in K-1 folds ( training data ) , and the data from the held-out fold ( test data ) were used when evaluating the extracted internal model . Although target positions were used to incorporate the notion of straight-to-target aiming during model fitting ( through Equation 14 ) , neither targets nor Equation 14 were used when evaluating extracted internal models on held-out data ( relevant for Figures 3–6 , Figure 3—figure supplement 3 , Figure 3—figure supplement 4 , Figure 3—figure supplement 7 , Figure 4—figure supplement 1 , and Figure 4—figure supplement 2 ) . Rather , whiskers were defined as the expected value of the internal state predictions given only available visual feedback and previously issued neural activity according to the probabilistic model , using only Equations 9–13 and not Equation 14 ) : ( 15 ) E =x~t-τtx~t-τ+1t⋮x~tt∣xt-τ , ut-τ+1raw , … , utraw=xt-τA~⁢xt-τ+B~⁢ut-τ+1raw+b~⋮A~⁢x~t-1t+B~⁢utraw+b~ We found that cross-validated whiskers consistently pointed straight to targets . This result did not trivially need to be the case , as those targets were not used to construct the whiskers . Rather , given internal models extracted from the training data , the statistical structure underlying the recorded neural activity in the test data was consistent with aiming straight to targets from internal predictions of cursor position . In Figure 3—figure supplement 2 we visualize the parameters of an extracted internal model as “pushing vectors” , and interpret them relative to the corresponding parameters of the BMI mapping . Because of differences in how temporal smoothing is implemented through the BMI mapping and the internal model , magnitudes of pushing vectors are not directly comparable between the BMI mapping and the internal model . In the BMI mapping , temporal smoothing comes from averaging the neural activity across 5 timesteps , as in Equation 6 . In the internal model , temporal smoothing comes from the specification that each velocity prediction includes a contribution from the previous velocity prediction through A~v , as in Equation 10 . To provide visually comparable pushing vectors , we factored out the influence of temporal smoothing by visualizing the pushing vectors from Bv and B~v as follows . Pushing vectors in Figure 3—figure supplement 2A show how the cursor would have moved given a single smoothed spike count from each unit . Analogously in Figure 3—figure supplement 2B , we rescaled the pushing vectors in B~v by 1/ ( 1-12⁢trace⁢ ( A~v ) ) , approximately normalizing by the fraction of the internal velocity prediction that comes from the previous velocity prediction rather than from the current neural activity . The 12⁢trace⁢ ( A~v ) in the scaling factor gives the average value along the diagonal of the 2×2 matrix , A~v . This normalization was only required because of the particular manner by which cursor velocities were smoothed during BMI experiments . If we had instead used a Kalman filter as the BMI mapping during experiments , pushing vectors would be directly comparable without normalization . In Figure 5—figure supplement 1 , we visually interpret an example spike count vector through the internal model shown in Figure 3—figure supplement 2 . This example spike count vector contributed to the monkey A “movement” bar in Figure 5B , as it was the timestep at which cursor-to-target distance first decreased below 50% of the center-to-target distance . The example spike count vector is from the same session as the BMI mapping and internal model parameters shown in Figure 3—figure supplement 2 , and the spike count vector is from a held-out trial not used to fit that internal model . Figure 5—figure supplement 1B , C reflect rescaled B~v and b~v , as described above . Comparisons of the appropriateness of the recorded neural activity through the BMI mapping versus through extracted internal models are shown as angular errors in Figure 3B , C , Figure 4C , Figure 6A , Figure 3—figure supplement 3 , Figure 3—figure supplement 4 , Figure 3—figure supplement 7 , and Figure 4—figure supplement 2 . For a particular timestep , t , we computed the angular error of the neural activity through the BMI mapping as the absolute angle by which the cursor would have missed the target had it continued from cursor position pt in the direction of the cursor velocity , vt ( i . e . , ΘP in Figure 2—figure supplement 1 ) . Similarly , we computed the angular error of the neural activity through the subject’s internal model as the absolute angle by which the cursor would have missed the target had it continued from the subject’s internal position prediction , p~tt , in the direction of the subject’s internal velocity prediction , v~tt . Internal model errors were computed from whiskers that could be constructed given cursor feedback and recorded spike counts beginning at movement onset and through target acquisition . Whiskers were extracted using the cross-validation techniques described in Computing cross-validated internal model predictions . An alternative explanation of our data could be that the subject’s internal model is well-matched to the BMI mapping , but that correlated noise in neural firing leads us to estimate an internal model that rejects noise better than the BMI mapping . To determine whether our finding of internal model mismatch might have been a spurious result of noise in the recorded neural activity , we performed the following simulation , which assumes the alternative hypothesis that there is no internal model mismatch . First , we simulated neural activity under the assumption that the BMI mapping and the internal model are equal ( i . e . , the alternative hypothesis ) . Then , we evaluated that simulated neural activity through the BMI mapping and the extracted internal model ( which were not equal ) . The key insight provided is due to the ability to explicitly define signal versus noise in simulation . Although there are many possible ways to define signal versus noise in the recorded neural activity , here we assume the internal model and the BMI mapping are equal ( the alternative hypothesis ) , and we define signal to be the component of a neural activity pattern that maps to the subject’s desired movement direction through that mapping . We define noise to be the residual neural activity pattern after subtracting out the signal . We began with a set of 32 desired movement directions , di∈{0∘ , 11 . 25∘ , 22 . 5∘ , …} . This set was chosen to align with the 16 target directions with an additional direction halfway between each pair of adjacent targets . We labeled each recorded neural activity pattern , utraw , according to the direction , di* , that it most closely matched after being passed through the BMI mapping ( Equation 8 ) from that experiment . This labeling procedure produces , for each direction , di* , a set of real recorded neural activity patterns , Ui , that reflect the intention to move in direction di* . For each direction , we then defined an idealized neural activity pattern to be the mean of all real neural activity patterns labeled as matching that direction through the BMI mapping: ( 16 ) ui* = 1\|Ui\|⁢∑utraw∈Uiutraw where \|Ui\| is the number of real activity patterns labeled as matching direction di* . We performed this procedure separately for each intuitive session . Idealized neural activity patterns were calculated from sets of 109 ± 24 ( monkey A ) and 178 ± 56 ( monkey C ) real neural activity patterns ( mean ± standard deviation across all experiments and directions ) . We evaluated the error of these idealized neural activity patterns through the BMI mapping and through the extracted internal model , relative to the corresponding desired direction ( i . e . , the average absolute angular error between di* and vi* = Bui+b , and between di and v~i* = B~⁢ui+b~ , respectively ) ( Figure 3—figure supplement 6A ) . By construction , we expect errors through the BMI mapping to be nearly zero ( nonzero errors are due to the discretization of direction ) . To determine the effect of noise in the recorded neural activity , we corrupted these idealized neural activity patterns by combining them with simulated noise patterns drawn from residuals in the recorded neural activity . Residuals , ct , were computed by subtracting the idealized neural activity pattern , ui , from the recorded neural activity patterns , utraw , corresponding to that idealized pattern: ( 17 ) for each⁢ utraw∈Ui , ct = utraw-ui* Simulated noise patterns were then drawn from the across-direction set of residuals: ( 18 ) sk~{ct}∀t Finally , simulated noisy neural activity patterns , ui , ksim , were formed by combining the idealized neural activity patterns with the simulated noise patterns: ( 19 ) ui , ksim = ui* + sk We evaluated the error of the simulated noisy neural activity patterns , ui , ksim , through the BMI mappings and through the extracted internal models , relative to the corresponding desired direction ( that is , the average absolute angular error between di* and vi , ksim=Bui , ksim+b , and between di* and v~i , ksim=B~⁢ui , ksim+b~ , respectively ) ( Figure 3—figure supplement 6B ) . This analysis was fully cross-validated , meaning that we only evaluated a simulated neural activity pattern through an internal model if its simulated noise pattern was not computed from a recorded neural activity pattern used during fitting of that internal model . To further match the statistics of the real data , we ensured that we evaluated the same number of simulated neural activity patterns corresponding to a particular desired direction as the number of recorded neural activity patterns that matched that desired direction through the BMI mapping . In Figure 5 we compared the timestep-by-timestep speeds of the actual cursor to the subject’s intended cursor speed , as determined by extracted internal models . At timestep t , actual cursor speed was taken to be the magnitude of cursor velocity vt ( Equation 6 ) , and intended cursor speed was taken to be the magnitude of the subject’s velocity belief , v~tt . To form the curves in Figure 5A , we selected all timesteps when intended cursor speed was s and computed the distribution of actual cursor speeds at those same timesteps . Curves show the mean actual cursor speed ( and S . E . M . ) as a function of intended cursor speed . In Figure 5B , we included all timesteps preceding target onset to form the speed difference bars labeled “center hold . ” To form the “movement” bars , we included for each trial the single timestep at which cursor-to-target distance first decreased below 50% of the center-to-target distance . In Figure 6A we interpreted monkey A neural activity through several relevant mappings: the intuitive BMI mapping , the perturbed BMI mapping , the time-varying internal model , and the late intuitive internal model . The time-varying internal model was extracted from a moving window of 48 trials and was updated every 16 trials ( 1 trial to each of the 16 targets ) . The late intuitive internal model was extracted from the last 48 trials during the intuitive session . For each mapping , whiskers were constructed at each timestep and angular errors were evaluated relative to the target perimeter . In general , these errors describe how task-appropriate the subject’s neural activity was for a particular mapping at a particular moment during the experiments . In the case of the BMI mappings , each whisker equates to how the cursor position and velocity would have evolved from a particular position on the actual cursor trajectory ( i . e . , the visual feedback ) had that BMI mapping been in effect . In the case of the intuitive BMI mapping during the intuitive and washout trials and in the case of the perturbed BMI mapping during the perturbation trials , these whiskers by definition exactly match the cursor trajectories displayed during the experiments . When a particular BMI mapping was not in effect ( e . g . , the intuitive BMI mapping during the perturbation trials ) , these whiskers describe how the cursor would have moved under that BMI mapping and thus would not match the cursor behavior from the experiments . In Figure 6B we evaluated the differences between the time-varying internal model and the BMI mappings . We interpreted monkey A neural activity through the BMI mappings and the time-varying internal model , and at each timestep , computed the angle between the velocity predicted by the internal model ( i . e . , by constructing a whisker ) and the velocity computed through each BMI mapping . For experiments with a counter-clockwise rotational component to the perturbation , signs of angles were flipped so that nonzero biases would not cancel out when averaging across experiments . Angles were computed after centering the origins of the two velocity vectors . In contrast to the absolute angles computed relative to the target , which are presented through this work , these angles were signed and computed without using the target position . Analyzing signed angles permits quantification of the bias in these angles , whereas absolute angles are better suited for quantifying the variance of errors . During the intuitive trials , angular errors through the subject’s internal model and through the intuitive BMI mapping tend to be unbiased ( i . e . , the average signed angular error is roughly zero for both; data not shown ) but with different variances ( hence the substantial differences in absolute angular errors in Figure 3C and Figure 6A ) . Angular errors relative to the target become biased during the perturbation and washout trials due to the rotational nature of the perturbations . We also analyzed the signed angular errors ( data not shown ) to ensure that these biases were not affecting our interpretation of the errors in Figure 6A . The neural activity used in this time-varying difference metric was taken from 16-trial non-overlapping blocks of trials aligned to the horizontal axis labels ( as in Figure 6A ) . Results were nearly identical when instead using a fixed set of neural activity taken from the first 48 trials of the intuitive session ( data not shown ) . We also evaluated a number of additional difference metrics , including angles between unsmoothed velocities ( i . e . , between ( B~v⁢utraw+b~v ) and ( Butraw+b ) ; see Equation 10 ) , cartesian distances between whisker endpoints ( i . e . , \|\|p~tt-pt\|\|2 ) , and angles between the corresponding columns of the B and B~ matrices from the BMI mappings and internal models , respectively ( data not shown ) . These metrics were consistent with the metric of Figure 6B in showing that the time-varying internal model becomes more similar to the particular perturbed BMI mapping during the perturbation trials and to the particular intuitive BMI mapping during the washout trials .	The human brain is widely hypothesized to construct “inner beliefs” about how the world works . It is thought that we need this conception to coordinate our movements and anticipate rapid events that go on around us . A driver , for example , needs to predict how the car should behave in response to every turn of the steering wheel and every tap on the brake . But on icy roads , these predictions will often not reflect how the car would behave . Applying the brakes sharply in these conditions could send the car skidding uncontrollably rather than stopping . In general , a mismatch between one’s inner beliefs and reality is thought to cause errors and accidents . Yet this compelling hypothesis has not yet been fully investigated . Golub et al . investigated this hypothesis by conducting a “brain-machine interface” experiment . In this experiment , neural signals from the brains of two rhesus macaques were recorded using arrays of electrodes and translated into movements of a cursor on a computer screen . The monkeys were then trained to mentally move the cursor to hit targets on the screen . The monkeys’ cursor movements were remarkably precise . In fact , the experiment showed that the monkeys could internally predict their cursor movements just as a driver predicts how a car will move when turning the steering wheel . These findings indicate that the monkeys have likely developed inner beliefs to predict how their neural signals drive the cursor , and that these beliefs helped coordinate their performance . In addition , when the monkeys did make mistakes , their neural signals were not entirely wrong—in fact they were typically consistent with the monkeys’ inner beliefs about how the cursor moves . A mismatch between these inner beliefs and reality explained most of the monkeys’ mistakes . The brain constructs such inner beliefs over time through experience and learning . To study this learning process , Golub et al . next conducted an experiment in which the cursor moved in a way that was substantially different from the monkey’s inner beliefs . This experiment uncovered that , during the course of learning , the monkey’s inner beliefs realigned to better match the movements of the new cursor . Taken together , this work provides a framework for understanding how the brain transforms sensory information into instructions for movement . The findings could also help improve the performance of brain-machine interfaces and suggest how we can learn new skills more rapidly and proficiently in everyday life .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Internal models for interpreting neural population activity during sensorimotor control
Prior experience accelerates acquisition of novel , related information through processes like assimilation into mental schemas , but the underlying neuronal mechanisms are poorly understood . We investigated the roles that prior experience and hippocampal CA3 N-Methyl-D-aspartate receptor ( NMDAR ) -dependent synaptic plasticity play in CA1 place cell sequence encoding and learning during novel spatial experiences . We found that specific representations of de novo experiences on linear environments were formed on a framework of pre configured network activity expressed in the preceding sleep and were rapidly , flexibly adjusted via NMDAR-dependent activity . This prior experience accelerated encoding of subsequent experiences on contiguous or isolated novel tracks , significantly decreasing their NMDAR-dependence . Similarly , de novo learning of an alternation task was facilitated by CA3 NMDARs; this experience accelerated subsequent learning of related tasks , independent of CA3 NMDARs , consistent with a schema-based learning . These results reveal the existence of distinct neuronal encoding schemes which could explain why hippocampal dysfunction results in anterograde amnesia while sparing recollection of old , schema-based memories . An essential capacity of the brain is to form internal representations of the external world . Whereas representations of past experiences can be stored internally and be rapidly recalled from memory , neuronal representations of novel experiences develop from the interaction between specific external stimuli and the spontaneous internal neuronal dynamics during the novel experience . During the free exploration of spatial environments , individual hippocampal neurons fire at specific spatial locations of the animal and are activated sequentially along the trajectory of the animal ( place cells ) ( O’Keefe and Dostrovsky , 1971; O’Keefe and Nadel , 1978; McNaughton et al . , 1983; O’Keefe and Recce , 1993; Wilson and McNaughton , 1993; Lee and Wilson , 2002 ) . Sequences of place cells with partially overlapping fields fire with compressed temporal delays that correspond to the Euclidian distance between the location of their place field peaks ( Muller et al . , 1996; Dragoi and Buzsaki , 2006 ) . This phenomenon , known as sequence compression ( Skaggs et al . , 1996; Dragoi and Buzsaki , 2006 ) is thought to be an animal model of the internal representation of an external space by the place cell assemblies in the CA1 area of the rodent hippocampus ( Skaggs et al . , 1996; Dragoi et al . , 2003; Dragoi and Buzsaki , 2006; O’Keefe and Nadel , 1978 ) . The upstream auto-associative area , CA3 , has been proposed to orchestrate ( Tsodyks , 1999; Dragoi and Buzsaki , 2006 ) , together with entorhinal cortex and via synaptic plasticity ( Dragoi et al . , 2003 ) , the functional organization of cellular assemblies ( Hebb , 1949; McNaughton et al . , 1996; Harris et al . , 2003; Dragoi and Buzsaki , 2006 ) in the downstream CA1 region , the source of the hippocampal output to the rest of the cortex . According to a prevailing model , novel temporal and spatial place cell sequences emerge rapidly in the hippocampus upon exploration of a novel linear track predominantly or exclusively in response to the complex stimuli from the external environment and with minimal or no contribution from the internal neuronal dynamics around the time of the exploration ( Skaggs and McNaughton , 1996; Lee and Wilson , 2002 ) . Subsequently , these sequences are replayed during periods of resting ( Foster and Wilson , 2006; Diba and Buzsaki , 2007; Davidson et al . , 2009; Karlsson and Frank , 2009; Dragoi and Tonegawa , 2011 ) or sleep ( Nadasdy et al . , 1999; Lee and Wilson , 2002; Ji and Wilson , 2007; Karlsson and Frank , 2009 ) at higher incidences and perhaps facilitate the consolidation of the encoded information ( Girardeau et al . , 2009; Nakashiba et al . , 2009; Ego-Stengel and Wilson , 2010; Jadhav et al . , 2012 ) . Recently , we described that temporal sequences of firing that correlated with place cell sequences formed during the exploration of a novel linear track had been expressed during the resting/sleep period preceding the exploration , a phenomenon called preplay ( Dragoi and Tonegawa , 2011 , 2013 ) . We proposed that spontaneous neuronal activity preceding a novel spatial experience may prime and contribute to the formation of new spatial representations via a neuronal ensemble selection process ( Dragoi and Tonegawa , 2013 ) . The dynamics of the interplay between the external stimuli available ‘online’ during the novel experience and the internal neuronal activity around the time of the experience along with the specific contribution they have to the emergence of a novel spatial representation remain to be elucidated . Here , we investigated the dynamics and contribution of this interplay by studying the development of novel spatial representations in naïve and experienced animals during spatial exploration and their relation to the internal neuronal dynamics preceding and following the new experience . In order to understand the molecular and cellular mechanisms underlying the development of new spatial representations in the CA1 , we studied mice in which experience-dependent N-methyl-D-aspartate receptor ( NMDAR ) -associated activity and synaptic plasticity were genetically blocked specifically in the upstream auto-associative area CA3 ( Nakazawa et al . , 2002 ) , the source of the main excitatory input and sharp-wave/ripple associated activity into CA1 ( Buzsaki , 1989; Nakashiba et al . , 2009 ) . These mutant mice had their CA3 NR1 subunit knocked-out using a Cre/loxP recombination system ( Tsien et al . , 1996; Nakazawa et al . , 2002 ) and are referred to as the CA3 NMDAR KO , or simply KO mice . Throughout this study , they were compared with their control littermates , the floxed NR1 mice ( Tsien et al . , 1996; Nakazawa et al . , 2002 ) , here referred to as control mice ( CT ) . Previous studies using these two groups of mice have shown that deletion of the NR1 subunit in the CA3 area abolished the NMDAR currents in the CA3 pyramidal cells , but not in the downstream CA1 and upstream dentate gyrus area neurons ( Nakazawa et al . , 2002 ) , which resulted in impaired memory acquisition of one-time experiences ( Nakazawa et al . , 2003 ) . The lack of post-synaptic NMDAR function has been shown to block the induction of long-term potentiation of synaptic transmission in the hippocampus ( Bliss and Lomo , 1973; Bliss and Collingridge , 1993; Tsien et al . , 1996 ) and other brain areas , a mechanism proposed to underlie learning and memory ( Morris et al . , 1986; Tsien et al . , 1996; Moser et al . , 1998 ) . In order to explore the role of prior experience on the formation of novel spatial representations and spatial learning , we compared the single and ensemble place cell dynamics as well as the animals’ performance on a delayed alternation task ( Ainge et al . , 2007 ) in naïve and experienced mice . We therefore investigated the temporal development of hippocampal place cell assemblies and the dynamics of behavioral performance in the CT ( Tsien et al . , 1996 ) and KO mice ( Nakazawa et al . , 2002 ) during de novo spatial exposure and the effect of this ( prior ) experience on cell assembly and learning dynamics during subsequent exposures to related novel space and behavioral tasks . De novo neuronal representations of spatial experiences were formed on the framework of the spontaneous network activity preceding the experience ( Dragoi and Tonegawa , 2011 ) and were subsequently modified and rapidly stabilized via CA3 NMDAR-dependent activity . This prior experience on novel linear tracks accelerated the encoding of subsequent experiences on additional , contiguous or isolated novel tracks and eliminated or significantly reduced their CA3 NMDAR dependence , respectively . We evaluated the behavioral relevance of these neuronal dynamics by comparing them with the performance of naïve and experienced CT and KO animals on a T-maze delayed alternation task . The presence of CA3 NMDARs facilitated the de novo learning of the alternation task; this prior experience accelerated subsequent learning of a related alternation task in a CA3 NMDAR-independent manner , consistent with principles of a schema-based learning ( Bannerman et al . , 1995; Tse et al . , 2007; McKenzie et al . , 2013 ) . Moreover , prior experience and CA3 NMDARs modulated the learning rate and the neuronal ensemble dynamics across experimental conditions in a correlated manner , suggesting that these neuronal dynamics , and in particular the expression of stable cell assemblies in CA1 , might be part of the neural mechanisms of learning in naïve and experienced animals . For all experimental conditions we determined the across-sessions changes in spatial tuning and stability of individual CA1 place cells ( Hill , 1978; Wilson and McNaughton , 1993; Nakazawa et al . , 2003; Frank et al . , 2004; Cacucci et al . , 2007 ) and in the lap-by-lap correlation ( co-variation ) of spiking activity of cell pairs ( Dragoi and Buzsaki , 2006 ) during exploration of novel environments . The changes in single cell activity allow the quantification of the temporal development of place fields over successive exploratory sessions ( see Figure 1B–D for individual cell examples for all three conditions ) , and the lap-by-lap correlation reflects the organization of neurons in coordinated cellular assemblies whose member pairs exhibit higher temporal correlation than pairs of independent neurons ( Harris et al . , 2003; Dragoi and Buzsaki , 2006 ) . The stability of place cell firing within the run session on novel tracks increased with experience in the de novo condition similarly in the CT and KO mice ( Figure 1E ) , appeared high from the beginning of the contiguous condition in both genotypes ( Figure 1F ) , and appeared high from beginning in the CT mice and increased with repeated experience in KO mice in the disjunct condition ( Figure 1G ) . Under the de novo condition , the spatial tuning of CA1 place cells was relatively poor in session 1 ( DnRun1 ) in both CT ( Cacucci et al . , 2007 ) and KO mice ( Nakazawa et al . , 2003 ) by the measure of place field length ( Figure 2A ) . Spatial tuning increased faster in CT vs KO mice from session 1 to session 2 ( Figure 2A , left and 2B left , paired difference in place field length DnRun1–DnRun2 , 8 ± 2 . 3 cm , p<0 . 003 , paired t test in CT; 0 . 6 ± 2 . 4 cm , p=0 . 78 in KO; CT vs KO , p<0 . 027 , ranksum test ) and within-session DnRun1 ( Figure 2C , left , spatial tuning , 4 . 8 ± 2 . 1 cm , p<0 . 003 in CT; −0 . 7 ± 1 . 5 cm , p=0 . 3 in KO; CT vs KO , p<0 . 015 , ranksum test ) . An overnight holding in the home cage led to significantly increased spatial tuning in KO mice the next day ( Figure 2A , DnRun1 compared to FamRun1 , 29 . 6 ± 2 . 6 cm compared to 20 . 6 ± 2 . 0 cm , p<0 . 01 ) . The average number of place fields per place cell ( >1 Hz ) was not different across genotypes and experimental sessions ( CT vs KO: 1 . 52 vs 1 . 58 fields/place cell in DnRun1 , p>0 . 5 , ranksum test; 1 . 50 vs 1 . 53 in DnRun2 , p>0 . 7 ) . This result indicates that the experience- and genotype-dependent changes in place field size reported above ( Figure 2B–C ) result from the spatial tuning of initially large place fields rather than from the scattering of large place fields into multiple smaller ones . At the neuronal ensemble level , there was a significant increase in the lap-by-lap correlations with increased experience for both genotypes , although the increase occurred faster in CT compared to KO mice ( Figure 2D ) . In the CT , but not KO mice , the increase reached a plateau by the second exploratory session ( DnRun2 ) under the de novo condition ( Figure 2D , DnRun1 compared to DnRun2 , 0 . 37 ± 0 . 007 compared to 0 . 48 ± 0 . 01 , p<10−13 , ranksum test in CT mice; 0 . 38 ± 0 . 005 compared to 0 . 38 ± 0 . 004 , p=0 . 38 in KO mice; CT vs KO in DnRun2 , p<10−9 , ranksum test ) . Although the correlations eventually reached stable CT levels in KO mice , this did not occur until the next day ( Figure 2D , DnRun1 compared to Fam , 0 . 37 ± 0 . 007 compared to 0 . 48 ± 0 . 01 , p<10−13 in CT and 0 . 38 ± 0 . 005 compared to 0 . 48 ± 0 . 01 in KO , p<10−10 ) . These results suggest that the first-time spatial experience on a linear track leads to a gradual increase in spatial tuning and co-variation of ensembles of CA1 place cells encoding the novel experience , and that the temporal dynamics of these processes are facilitated by NMDAR-dependent activity in the auto-associative area CA3 of the hippocampus . The two run sessions of the de novo condition ( over 1 hr altogether , Table 1 ) may have provided enough time for the relatively slower non-NMDAR-dependent plasticity to occur and underlie the slower place cell dynamics . The results suggest that CA3 NMDAR-dependent activity and plasticity are specifically involved in the rapid within-session changes in single and ensemble place cell activity . 10 . 7554/eLife . 01326 . 006Figure 2 . Prior experience and CA3 NMDARs accelerate tuning and co-variation of place cells in novel environments . ( A ) Dynamics of place field length for CT ( black color ) and KO ( red color ) during de novo , contiguous , and disjunct conditions . ( B ) Paired changes in place field length across the first two sessions for each condition . ( C ) Within-session changes in place field length ( first novel session of each condition ) . ( D ) Dynamics of lap-by-lap correlations of CA1 pyramidal cell spiking activity . The increase in these correlations followed the increase in spatial tuning . ( E ) Comparison between the CA1 place field length in the familiar and novel portion of the L-shape track . ( F ) Comparison of lap-by-lap correlations of spiking activity across place fields in the familiar vs novel portion of the L-shape track . ( G ) Average velocity of mice across different run sessions . ( H ) Average normalized spatial information of place cells during first run sessions on novel track/arm across conditions and genotypes . Dotted line denotes average normalized spatial information during corresponding FamRun sessions used as reference for the spatial information during novel run sessions . ( I–K ) Lap-by-lap co-variation during exploration of the novel track/arm ( DisjRun1 in I and J , ContigRun in K ) as a function of change in either the relative distance between place fields across the two tracks ( disjunct condition in I , contiguous condition in K ) or the distance between place fields in the novel track ( J ) . For all subplots , asterisks mark significant differences between groups . Data are from mice CT1-4 and KO1-4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 006 We next investigated whether the experience acquired during the de novo condition had an effect on a related , but new experience , by evaluating whether the exploration of a novel arm in contiguity with the familiar one ( contiguous condition ) would re-enact the de novo-type responses only in the novel or in both arms . Surprisingly , for both genotypes , the new place fields were as spatially tuned during the first session ( ContigRun ) in the novel arm of the L-shaped track as in the contiguous familiar arm ( Figure 2A , E ) and as in the preceding FamRun1 session ( Figure 2A ) . No additional spatial tuning could be detected within the ContigRun session within each group of mice ( Figure 2B–C; FamRun1 compared to ContigRun: 22 . 1 ± 1 . 7 cm compared to 23 . 9 ± 1 . 7 cm , p=0 . 46 in CT and 20 . 6 ± 2 . 0 cm compared to 21 . 9 ± 1 . 1 cm , p=0 . 34 in KO ) , across genotypes ( CT compared to KO , p=0 . 6 in FamRun1 , p=0 . 46 in ContigRun ) , and for both the familiar and the novel portions of the L-shaped track ( Figure 2E , p>0 . 05 ) . The average number of place fields per place cell on the L-shaped track was similar across genotypes during ContigRun ( CT vs KO , 1 . 90 vs 1 . 89 , p<0 . 98 ) . Similarly , the increased lap-by-lap correlations observed in FamRun1 were preserved in the ContigRun session for both groups of mice ( Figure 2D , F; FamRun1 compared to ContigRun , 0 . 48 ± 0 . 01 compared to 0 . 47 ± 0 . 009 , p=0 . 41 in CT and 0 . 48 ± 0 . 01 compared to 0 . 46 ± 0 . 01 , p=0 . 2 in KO ) , and there were no significant differences in the correlations between genotypes ( CT compared to KO , p=0 . 98 in FamRun1 , p=0 . 42 in ContigRun ) , nor between the familiar and the novel portions of the L-shaped track ( Figure 2F , p>0 . 05 ) . These results suggest that under the contiguous condition , independent of CA3 NMDARs , the spatial information regarding the previously unvisited novel portion of the L-shaped track was rapidly bound with the existing stable representation of the familiar track given the animals’ prior experience on the familiar track . These rapid dynamics on the novel arm during the Contig session are in contrast to the slow dynamics of place field tuning and co-variation during de novo exposure to a novel track ( i . e . , the DnRun1 session for CT and DnRun1–2 sessions for KO ) . This may reflect the prior recruitment at a subthreshold firing level of the new place cells ( Epsztein et al . , 2011 ) into stable , tuned cortico-hippocampal cellular assemblies during pre-exposures to the familiar track , which manifested as preplay of the novel arm during the preceding sleep or resting epochs ( Dragoi and Tonegawa , 2011 , 2013 ) . To determine whether the contiguity between the novel and familiar arms was essential for the rapid spatial tuning and assembly organization and to further assess the effect of experience on novel spatial representations , experienced animals were exposed to an isolated novel linear track in the same general environment ( disjunct condition , Figure 2A ) . In this case , exploration of the novel track induced a significant , albeit transient , increase in the CA1 place field length in KO but not CT mice ( FamRun2 vs DisjRun1: 23 . 9 ± 1 . 4 vs 23 . 9 ± 1 . 5 cm in CT , p=0 . 9; 22 . 6 ± 1 . 3 vs 28 . 9 ± 2 in KO , p<0 . 01 , ranksum test; between genotypes , CT vs KO in DisjRun1 , p<0 . 05; DisjRun1 vs DisjRun2 in KO , p<0 . 04 , Figure 2A; paired difference in place field length FamRun2–DisjRun1 CT vs KO , p<0 . 027 , ranksum test , Figure 2B; within-session DisjRun1 , CT vs KO , p<10−3 , Figure 2C ) , consistent with an earlier report ( Nakazawa et al . , 2003 ) . The average number of place fields per place cell was slightly higher in CT compared to KO mice , but this relationship was not affected by the additional experience of animals on the novel track ( CT vs KO: 1 . 76 vs 1 . 52 in DisjRun1 and 1 . 76 vs 1 . 47 in DisjRun2 , p<0 . 002 ) . The increased lap-by-lap correlations recorded on the familiar track ( FamRun2 ) decreased with novelty ( DisjRun1 ) in KO but not in CT mice ( FamRun2 vs DisjRun1: 0 . 49 ± 0 . 006 vs 0 . 51 ± 0 . 008 , p>0 . 05 in CT; 0 . 47 ± 0 . 005 vs 0 . 41 ± 0 . 004 , p<10−15 in KO; DisjRun1: CT vs KO , p<10−30; DisjRun1 vs DisjRun2 in KO , p<10−4 , Figure 2D ) . The correlations returned to ‘familiar levels’ by the next session ( DisjRun2 ) in the KO mice more rapidly compared to the de novo condition ( compare DisjRun2 to DnRun2 in KO , Figure 2D ) . Overall , these dynamics were not due to differences in the velocity of animal movement across sessions or genotypes ( Figure 2G ) , as place field size and lap-by-lap correlations were not significantly correlated with the animal velocity ( p>0 . 11 and p>0 . 12 , respectively ) . Moreover , the changes in place field tuning occurring within experimental sessions ( i . e . , DnRun1 , DisjRun1 , first vs last four laps ) were not simply due to a change in the behavior of the mice during the corresponding sessions since their velocity and total distance travelled ( path stereotypy ) were similar in the first four laps compared with the last four laps of the sessions for both genotypes ( first vs last four laps , p>0 . 05 , ranksum test; velocity , Figure 1—figure supplement 1A , total distance travelled , Figure 1—figure supplement 1B ) . The changes in lap-by-lap correlations across experimental sessions were not simply a result of changes in animal behavior during the 3s-bin epochs since the distances travelled by the mice on the tracks during the 3s-bins were not correlated with the values of the lap-by-lap correlations calculated over the same timescale ( p>0 . 2 ) . The overall changes in the spatial information of place cells ( Skaggs et al . , 1996 ) between novel and familiar run sessions for individual CA1 place cells across genotypes and conditions were consistent with the changes in place field size and in lap-by-lap correlations of pairs of cells ( Figure 2H ) . To further examine the structure of correlations during DisjRun1 exploration , we grouped pairs of place cells active on both FamRun2 and DisjRun1 sessions based on the change in the relative distance between their place field peaks across the two tracks into those with changes within 5 cm ( stable assemblies , 8% CT , 10% KO pairs ) , those with changes between 6 cm to 20 cm ( 27% CT , 37% KO ) , and those with changes between 21 cm to 35 cm ( 32% CT , 26% KO ) . The remaining pairs , with changes larger than 35 cm were excluded from this analysis . In the CT mice , cells from all three groups displayed relatively high lap-by-lap co-variation during DisjRun1 exploration ( Figure 2I ) . In contrast , in KO mice , only cells of the stable assemblies ( within 5 cm change ) maintained a high lap-by-lap co-variation during DisjRun1 , while groups of pairs with changes larger than 5 cm exhibited significantly poorer co-variation ( Figure 2I , p<0 . 05 , ranksum test ) . These results suggest a modular , higher order organization of cellular assemblies , in which a certain group of cell pairs ( stable assemblies ) counters a complete NMDAR-dependent reorganization of the entire neuronal population in response to novel stimuli by maintaining their co-variation independent of CA3 NMDARs . This modular organization of cell assemblies was not simply based on the spatial proximity between place fields on the novel track as co-variations in the KO mice were significantly poorer than in CT mice for all three regimes of changes in place field distance ( Figure 2J , p<0 . 05 ) . A similar grouping of the cell pairs active in the contiguous condition during FamRun1 and ContigRun revealed that the changes in cellular ensemble organization upon exploration of the contiguous novel arm are CA3 NMDARs-independent ( Figure 2K ) . The large proportion of cell pairs showing >20 cm changes in their spatial relationship from familiar to novel tracks ( >65% in CT , >53% in KO mice ) indicates that the ensemble of place cells formed distinct representations across the two environments , a process called remapping ( Muller and Kubie , 1987; Dragoi et al . , 2003; Leutgeb et al . , 2005; Dragoi and Tonegawa , 2011 , 2013 ) . Consistent with the remapping , the order in which the place cells fired in the familiar track ( place cell sequences ) was not correlated with their order of firing in the novel arm ( contiguous condition , mean RCT2 = 0 . 03 , mean RKO2 = 0 . 12 , p>0 . 05 , both genotypes ) or novel track ( disjunct condition , mean RCT2 = 0 . 11 , mean RKO2 = 0 . 12 , p>0 . 05 , both genotypes ) , whereas it was similar to the order in which they later fired in the familiar arm of the L-shape track ( contiguous condition , RCT2 = 0 . 6 , RKO2 = 0 . 5 , p<0 . 05 , both genotypes ) . For each run session and for each direction of movement , place cells were ordered according to the location of their peak firing ( >1 Hz ) on the corresponding track/arm , resulting in two place cell sequence templates for each condition and session , one for each direction ( Dragoi and Tonegawa , 2011 ) . Spiking events ( Foster and Wilson , 2006; Diba and Buzsaki , 2007; Dragoi and Tonegawa , 2011 , 2013 ) were detected for each sleep/rest session in the sleep/rest box as increases in multiunit activity ( at least five of the place cells active during the corresponding run session ) that were preceded and followed by >100 ms of silence . For each spiking event , a rank-order correlation between the place cell sequence template and the temporal sequence of cell firing during the event was calculated for each direction of runs and for each session and experimental condition in both CT and KO mice ( Figure 3A–F ) . The event was considered significant if the correlation of its firing sequence with the corresponding place cell sequence template exceeded the 97 . 5th percentile of a distribution of correlations when the order of the place cells in the novel track template or novel arm template was shuffled randomly 100 times ( i . e . , p<0 . 025 ) . For each direction of run , preplay ( i . e . , pre-Run play ) refers to an event’s temporal sequence during pre-Run sleep/rest that has a significant correlation with the spatial cell sequence of the subsequent run session . Likewise , post-Run play or replay refers to an event’s temporal firing sequence during post-Run sleep/rest that has a significant correlation with the spatial cell sequence of the preceding run session . Events significant for both running directions were assigned only to the direction with the higher absolute correlation value . 10 . 7554/eLife . 01326 . 007Figure 3 . Display of preplay and replay sequences during sleep/rest in naïve and experienced mice in all experimental conditions . ( A and B ) Examples of preplay spiking events during the sleep/rest session in the sleep/rest box preceding the first run on the novel track during De novo condition ( first three boxes from the left ) , the corresponding place cell sequence during run ( fourth box ) , and replay events during the sleep/rest session in the sleep/rest box following the run ( last three boxes ) , in one control mouse , CT2 ( A ) and one CA3 NMDAR KO mouse , KO2 ( B ) . Arrows indicate the order of the place cell sequence . Corresponding local field potential recordings are shown above the spiking events . ( C and D ) Examples of preplay spiking events during the sleep/rest session in the sleep/rest box ( second to fourth boxes from the left ) following a run session on the familiar track ( leftmost box , place cells in blue ) and preceding the run session on the novel portion of the L-shape track under the contiguous condition ( fifth box , place cell sequence in red ) , and examples of replay events during the post-run sleep/rest ( last three boxes ) , from one control mouse , CT2 ( C ) and one CA3 NMDAR KO mouse , KO2 ( D ) . ( E and F ) Examples of preplay , place cells , and replay sequences before , during , and after run on an isolated novel linear track ( place cells in red ) in the disjunct condition that all followed run on a familiar track ( place cells in blue ) , in one CT ( E ) and one KO ( F ) mouse . Boxes are assigned to experimental sessions like in ( E and F ) . For all subplots , spikes in red during spiking events represent the first spike for each participating cell; all the other spikes are in yellow . The place cell sequence template panels shown in ( C ) for mouse CT2 ( left , blue and right , red ) are reproduced from Figure 1Ea , Ec; Dragoi and Tonegawa ( 2011 ) , Nature; Nature Publishing Group has granted permission to reproduce these images under the terms of the Creative Commons Attribution 3 . 0 Unported License . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 007 For both genotypes and for all experimental conditions and sessions , the absolute correlation values between events occurring during the pre-Run or post-Run sleep/rest session and the original novel track/arm templates ( Figure 4A–H ) were significantly higher than the correlation values obtained using corresponding shuffled templates ( Figure 4 , left panels , ranksum test , see ‘Materials and methods’; all mice passed individual significance for preplay and replay across conditions , ranksum or binomial probability tests ) . At the level of individual animals , with the exception of one CT mouse ( CT2 , preplay > replay , p<10−4 , ranksum test ) , the absolute values of the correlations of the significant events with the spatial templates of the very first run session on the novel track ( i . e . , DnRun1 ) were similar for the spiking events occurring during the pre-DnRun1 sleep/rest session ( i . e . , when preplay occurs ) compared to those occurring during the post-DnRun1 session ( i . e . , when replay occurs ) , both in CT and KO mice ( Figure 5Aa , p>0 . 05 , ranksum test ) . The proportions of significant events out of all events were also similar in the pre-DnRun1 vs post-DnRun1 sleep/rest for each individual mouse and genotype ( p>0 . 05 for each individual animal , Z-test for two proportions , Figure 5Ba; preplay vs replay: 9/132 vs 9/148 in CT1 , 237/2255 vs 438/4456 in CT2 , 2/10 vs 5/45 in CT3 , 7/136 vs 5/71 in CT4 , 2/48 vs 4/74 in KO1 , 104/1310 vs 112/1496 in KO2 , and 80/1146 vs 141/1824 in KO3 ) . Surprisingly , when the absolute correlation values of the significant events were pulled together from all animals , they were higher for the spiking events occurring during the pre-DnRun1 sleep/rest session ( preplay ) compared to those occurring during the post-DnRun1 session ( replay ) in CT but not KO mice ( Figure 4A–B , right panels , Figure 5Ca , empty bars; pre-DnRun1 compared to post-DnRun1 , 0 . 87 ± 0 . 008 ( std . 0 . 12 ) compared to 0 . 80 ± 0 . 007 ( std . 0 . 13 ) , p<10−4 in CT , Figure 4A , right panel , Figure 5Ca , empty bars , left; and 0 . 89 ± 0 . 008 ( std . 0 . 1 ) compared to 0 . 88 ± 0 . 007 ( std . 0 . 1 ) , p>0 . 075 in KO , Figure 4B , right panel , Figure 5Ca , empty bars , right , ranksum test ) . 10 . 7554/eLife . 01326 . 008Figure 4 . Significance of preplay and replay events across experimental conditions and genotypes . ( A–H ) Distribution of absolute values of spiking event-place cell sequence correlations for all events occurring during the pre-Run sleep/rest ( open bars , Pre-Run all; solid black bars , corresponding shuffle correlations ) , all events during post-Run sleep/rest ( open bars , Post-Run all; solid black bars , corresponding shuffle correlations ) , significant pre-Run events ( red bars , Pre-Run sig ) , and significant post-Run events ( blue bars , Post-Run sig ) from all CT and all KO mice corresponding to the sessions and conditions described in Figure 1A . The conditions , sessions , and the genotype are specified above each subplot ( A–H ) . p values reflect differences between corresponding distributions using ranksum test . Data in A and C are from mice CT1-4; data in B and D are from mice KO1-3; data in E are from mice CT2-4; data in F are from mice KO1-4 . Data in G and H are from three CT and three KO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 00810 . 7554/eLife . 01326 . 009Figure 5 . Comparison of hippocampal temporal sequence activity during sleep/rest across multiple novel spatial experiences in individual mice . ( A ) Average correlation values for preplay and replay in individual animals and group data . Small letters , ( a–d ) , represent different experimental conditions and sessions for A-D . Thin lines represent individual animals; thick lines represent grand averages across all animals by condition/session and genotype for A and B . ( B ) Incidence of significant temporal sequences ( preplay and replay ) in individual animals and group data . Group comparison between preplay and replay during sleep/rest in naïve and experienced control and CA3 NMDAR KO mice . ( C ) Group average absolute correlation values of spiking event-place cell sequence correlations during pre-Run ( red bars ) and post-Run ( blue bars ) sleep/rest sessions . Left: correlations in control animals; right: correlations in KO animals . Solid bars: correlations calculated during all spiking events; empty bars: correlations calculated during significant events . Error bars represent SEM . Stars mark significant differences . ( D ) Group proportion of significant preplay and replay events across conditions , sessions , and genotypes . ( E–G ) Flexibility/rigidity of spatial-temporal sequences between sleep/rest and run across genotypes . ( E–F ) Averages of correlations between firing sequences during spiking events in sleep/rest ( preplay and replay ) and place cell sequences ( all events , E; significant events , F ) . ( G ) Comparison of variance in the distribution of all correlation values from E . Error bars are SEM . Stars denote statistical significance . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 00910 . 7554/eLife . 01326 . 010Figure 5—figure supplement 1 . Comparison between preplay of place cell sequences computed from the activity in the early vs late parts of the de novo DnRun1 session in CT and KO mice . Stars mark significant differences between genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 01010 . 7554/eLife . 01326 . 011Figure 5—figure supplement 2 . Similar features of preplay and replay during sleep/rest across multiple novel spatial experiences in individual mice . ( A ) Average number of cells/per significant event ( preplay vs replay ) . ( B ) Average duration of preplay and replay events . ( C ) Proportion of the corresponding track represented by preplay and replay events . Thin lines represent individual animals; thick lines represent grand averages across all animals by condition/session and genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 01110 . 7554/eLife . 01326 . 012Figure 5—figure supplement 3 . Histograms depicting co-occurrence of ripple oscillations and preplay/replay events across session , conditions , and genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 01210 . 7554/eLife . 01326 . 013Figure 5—figure supplement 4 . Immunohistochemistry for NR1 subunit of the NMDA receptor demonstrating the absence of the NMDA receptors specifically in the CA3 area in the KO mice ( left ) and its preservation in all areas of the hippocampus in the floxed NR1 CT mice . CA1 , CA3 , and DG ( dentate gyrus ) denote subfields of the hippocampal formation . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 013 Given the relative instability of the place fields in the very first run session in naïve animals ( Figure 1E ) , we asked whether the initial and the later spatial templates display different spatial-temporal correlation values with the temporal sequences during the preceding sleep . In order to address this question , we constructed spatial templates from the activity of place cells in the first four laps of run ( early templates ) and the last four laps ( late templates ) and correlated them with the temporal sequences recorded during the previous sleep ( i . e . , early and late preplay ) . We found that the population of early template correlations was not different from the population of late ones in both CT and KO mice ( Figure 5—figure supplement 1 ) . However , both early and late spatial-temporal correlations were higher in the KO mice compared with the CT ones ( Figure 5—figure supplement 1 ) . These results indicate that the relative instability of the place fields during DnRun1 session is associated with a process of relatively mild ‘editing’ of the early template , rather than with a process of dramatic ‘remapping’ into a new chart . The relatively high correlation between early and late place fields in the de novo condition ( 0 . 4–0 . 5 ) is consistent with this scenario . Following a second exposure to the novel track , with the exception of one CT mouse ( CT2 , preplay<replay , p<0 . 006 ) , the significant correlations of spiking events with the DnRun2 templates were similar during post-DnRun2 sleep/rest compared to the pre-DnRun2 sleep/rest in CT and KO mice ( Figure 5Ab , p>0 . 05 , ranksum test ) . Moreover , the proportions of significant events were also similar before and after the DnRun2 run experience in all mice ( p>0 . 05 , Z-test for two proportions , Figure 5Bb; proportions preplay vs replay: 15/164 vs 5/69 in CT1 , 465/4567 vs 196/1753 in CT2 , 21/160 vs 196/1753 in CT3 , 3/70 vs 8/131 in CT4 , 3/74 vs 8/66 in KO1 , 186/1942 vs 261/2578 in KO2 , and 223/2504 vs 158/1807 in KO3 ) . When absolute correlation values were pulled together across animals , the pre-Run vs post-Run play relationship reversed in the CT but not in the KO animals ( Figure 4C–D , right panels , Figure 5Cb , empty bars ) . The significant correlations of spiking events with the DnRun2 templates were stronger during post-DnRun2 sleep/rest compared to the pre-DnRun2 sleep/rest in CT animals ( pre-DnRun2 compared to post-DnRun2 , 0 . 77 ± 0 . 006 ( std . 0 . 13 ) compared to 0 . 83 ± 0 . 008 ( std . 0 . 13 ) , p<10−3 , ranksum test , Figure 4C , right panel , Figure 5Cb , empty bars , left ) ; however , they were similar in the KO mice ( Figure 4D , right panel , Figure 5Cb , empty bars , right; pre-DnRun2 compared to post-DnRun2 , 0 . 85 ± 0 . 006 ( std . 0 . 11 ) compared to 0 . 85 ± 0 . 006 ( std . 0 . 12 ) , p>0 . 35 ) . In the contiguous condition , both at the individual animal level ( Figure 5Ac ) and at the group level the correlations with the ContigRun templates in the post-ContigRun ( i . e . , replay ) were similar to the ones in the pre-ContigRun sleep/rest ( i . e . , preplay ) for both genotypes ( group level; Figure 4E , right panel , Figure 5Cc , empty bars , left , pre-ContigRun compared to post-ContigRun , 0 . 90 ± 0 . 01 ( std . 0 . 1 ) compared to 0 . 87 ± 0 . 009 ( std . 0 . 09 ) , p>0 . 55 in CT mice; Figure 4F , right panel , Figure 5Cc , empty bars , right , pre-ContigRun compared to post-ContigRun , 0 . 89 ± 0 . 01 ( std . 0 . 1 ) vs 0 . 86 ± 0 . 007 ( std . 0 . 09 ) , p>0 . 21 in KO mice , ranksum test ) . The proportions of significant events out of all events were also similar before and after the ContigRun experience ( Figure 5Bc , p>0 . 05 , Z-test for two proportions; preplay vs replay incidence: 59/1013 vs 83/1601 in CT2 , 3/23 vs 21/180 in CT3 , 2/51 vs 23/459 in CT4 , 28/381 vs 74/903 in KO2 , 55/788 vs 114/1422 in KO3 , and 7/123 vs 23/418 in KO4 ) . In the disjunct condition , at the level of individual animals , the absolute correlation values of significant events ( Figure 5Ad ) and their incidence were similar before and after the DisjRun1 experience ( Figure 5Bd , p>0 . 05 for all mice , Z-test for two proportions; preplay vs replay incidence: 541/4822 vs 265/2503 in CT2 , 62/824 vs 74/1024 in CT3 , 313/2769 vs 164/1410 in CT4 , 5/102 vs 3/18 in KO1 , 185/1671 vs 33/311 in KO3 , and 41/446 vs 103/869 in KO4 ) . At the genotype group level , the novel experience ( DisjRun1 ) resulted in slightly increased correlations in the post-DisjRun1 vs pre-DisjRun1 sleep/rest in CT ( 0 . 80 ± 0 . 006 ( std . 0 . 14 ) vs 0 . 77 ± 0 . 005 ( std . 0 . 14 ) , p<0 . 05 , ranksum test , Figure 4G , right panel , Figure 5Cd , empty bars , left ) , but not KO animals ( 0 . 86 ± 0 . 01 ( std . 0 . 12 ) vs 0 . 85 ± 0 . 009 ( std . 0 . 11 ) , p>0 . 17 , Figure 4H , right panel , Figure 5Cd , empty bars , right ) . Across all experimental conditions and sessions and for both genotypes there were no overall changes in the number of cells active per significant event ( preplay vs replay , Figure 5—figure supplement 2A , p>0 . 05 , ranksum test ) , in the duration of significant events ( Figure 5—figure supplement 2B , p>0 . 05 , ranksum test ) , and in the extent of the linear track being represented during the significant events ( Figure 5—figure supplement 2C , p>0 . 05 , ranksum test ) . For both genotypes and in all pre- and post-run sleep/rest sessions , the time of occurrence of significant preplay and replay events correlated with the time of occurrence of high-frequency oscillation ripples in the CA1 ( Figure 5—figure supplement 3 ) . Interestingly , the experience- and CA3 NMDAR-dependent changes in replay over preplay described at the animal group level across sessions in the de novo condition was also found when the correlation values between place field templates and all spiking events ( i . e . , not only the significant ones ) were considered ( Figure 5Ca , b , solid bars ) . The correlation values of all spiking events during pre-DnRun1 sleep/rest exceeded the correlation values of all spiking events recorded during post-DnRun1 sleep/rest in CT but not KO mice ( p<10−5 in CT , p>0 . 5 in KO , Figure 5Ca , solid bars ) . Preplay correlations were not different across genotypes ( p>0 . 9 , ranksum test ) , but replay correlations were higher in the KO than CT mice ( p<10−8 , ranksum test ) . Moreover , after a second experience on the novel track ( Figure 5Cb , solid bars ) , the correlation values of all post-Run ( i . e . , replay ) events became higher than the corresponding values of all pre-Run events in CT ( p<10−5 ) but not KO animals ( p>0 . 5 ) . Despite the changes in correlation values from preplay to replay in CT mice , the proportions of significant events out of all of the detected events in all animals were similar in the pre-Run and post-Run sessions under all experimental conditions for both genotypes ( Figure 5D , p>0 . 05 , Z-test for two proportions ) . The incidences of spiking events were similar in the sleep/rest sessions preceding and following the novel run experiences across conditions in both CT ( p>0 . 08 , paired t test ) and KO mice ( p>0 . 9 ) . More importantly , comparison of averages of absolute preplay and replay correlations over all experimental conditions and sessions performed altogether between the two genotypes ( eight sessions/genotype , paired by session type between genotypes ) revealed that correlation values were significantly higher ( p<0 . 006 for the significant correlations; p<0 . 009 for all the correlations; paired t test ) and their variance was significantly lower ( p<0 . 007 , paired t test ) in KO vs CT mice ( Figure 5E–G ) . This finding is consistent with the overall reduced experience-dependent changes in the correlation values in KO ( Figures 4B , D , F , H and 5C ) and indicates a reduced flexibility ( increased rigidity ) of the hippocampal network in the absence of NMDAR-dependent activity in the CA3 region . Moreover , the overall spatial extent ( i . e . , proportion ) of the linear track represented by the significant spiking events during sleep/rest was higher in the CT vs KO mice ( 0 . 94 ± 0 . 01 compared to 0 . 89 ± 0 . 01 of the track length , p<0 . 027 , paired t test ) , indicating a role for the NMDAR-dependent activity in the CA3 area in temporally linking together chunks of spatial sequences ( Dragoi and Buzsaki , 2006; Gupta et al . , 2012 ) . These differences in correlations between genotypes were not due to overall differences in the duration of the significant spiking events ( KO compared to CT across sessions , 0 . 33 ± 0 . 02 s compared to 0 . 33 ± 0 . 02 s , p=0 . 82 , paired t test ) , or in the number of cells participating in these events ( KO compared to CT , 6 . 1 ± 0 . 2 compared to 6 . 5 ± 0 . 2 , p=0 . 65 , paired t test ) across genotypes . We tested the behavioral relevance of the observed differences in neuronal dynamics by measuring the animal performance on a hippocampal-dependent learning task , the delayed T-maze alternation ( Ainge et al . , 2007 ) . In this task , the animals had to retrieve food rewards placed at the end of left vs right arms of the T-maze in alternate trials . The animals self-initiated the trials , moved toward the reward site and returned on the same path to re-initiate the next trial . Delays between trials were counted as the time between returning from the end of the left/right arms and self-initiation of the next trial ( >10 s ) . Both groups of animals reached the criterion level of performance of 70% correct choices during the 10-day training ( Figure 6A ) . Analysis of the data from both genotypes during the 10-day training using a balanced two-way ANOVA test ( five blocks of 2 consecutive days ) revealed a significant effect of the training day ( F = 11 . 83 , p=0 . 006 ) and genotype ( F = 10 . 5 , p=0 . 001 ) and a marginal duration ( day ) × genotype interaction ( F = 3 . 24 , p=0 . 057 ) . Consistent with the difference in the neuronal dynamics across genotypes during the de novo condition ( Figure 2A–D , left ) , the KO subjects required additional sessions to reach the 70% criterion for learning to alternate ( 9 vs 7 days , KO vs CT , Figure 6A; days 7–8 block , CT vs KO , 78 . 1% vs 53 . 7% correct , p=0 . 006 , ranksum test ) indicating the involvement of the intrinsic hippocampal circuitry in the acquisition of this task . Moreover , consistent with the facilitation of neuronal dynamics on the novel arm by prior experience on the familiar track during the contiguous condition ( Figure 2A–D , middle ) , this learning experience greatly accelerated ( Tse et al . , 2007 ) the acquisition of a similar alternation task in a second , novel T-maze configuration , when both groups of animals performed above threshold by the second day of re-training ( Figure 6B , day 10 vs 12 , 78 . 2% vs 77 . 6% correct , p=0 . 9 for CT , and 72 . 2% vs 75 . 7% correct , p=0 . 6 for KO , ranksum test ) . Together , these results establish a correlation between the neuronal activity and the behavioral performances across genotypes and experimental conditions and suggest that the dynamics of cell assembly organization are part of the underlying mechanisms of spatial learning . 10 . 7554/eLife . 01326 . 014Figure 6 . Behavioral performance of CT and KO mice in a T-maze alternation task . ( A ) Performance in the alternation task over the 10 days of training on the first T-maze configuration ( T-maze 1; 70% is the performance criterion for learning ) . Thin lines , learning curves expressed as averages of all subjects per group in the 10 days of training . Bold lines , learning curves expressed as averages of all subjects per group in blocks of two consecutive days of training . ( B ) Performance in the alternation task on exposure to a second T-maze configuration ( T-maze 2 ) . Note that for both genotypes , the performance in T-maze 2 on day 11 dropped below the 70% correct threshold . In the CT group , the performance in T-maze 2 on day 11 was significantly lower compared to the performance in T-maze 1 on day 10 ( p<0 . 05 , paired t-test ) . This indicates the animals detected ( and subsequently learned ) the novel spatial configuration of T-maze 2 . Error bars are SEM . Stars mark significant differences between groups . Insets: cartoons of the T-maze configurations in thick black lines; yellow dots , location of food reward at the ends of the choice arms . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 014 Previous studies on spatial representation in the hippocampus either investigated the process of spatial encoding by evaluating the dynamics of place field formation and neuronal correlation during exploration ( Hill , 1978; Wilson and McNaughton , 1993; Nakazawa et al . , 2003; Frank et al . , 2004; Cacucci et al . , 2007; Brun et al . , 2008 ) or , separately , described the phenomenon of temporal sequence replay during awake resting ( Foster and Wilson , 2006; Diba and Buzsaki , 2007; Davidson et al . , 2009; Karlsson and Frank , 2009 ) or post-Run sleep ( Nadasdy et al . , 1999; Lee and Wilson , 2002; Ji and Wilson , 2007; Karlsson and Frank , 2009 ) . Importantly , all of these studies used experienced animals which were either pre-trained on similar environments or were simply re-exposed to the familiar ones , and whose prior neuronal activity as naïve animals in previous environments was not documented . Overall , the expression of spatially tuned novel place fields in experienced animals as reported in some of the previous studies ( Hill , 1978; Nakazawa et al . , 2003 ) is consistent with our findings in experienced CT mice ( contiguous and disjunct conditions ) , though a faster timescale analysis might reveal additional dynamics in spatial tuning ( Frank et al . , 2004 ) . Our approach is to study the process of internal development of novel spatial representations as a dynamic whole by comparing and correlating the activity of ensembles of neurons during the sleep/rest period prior to first time exploration of a linear track with the one during the exploration , and both of these activities with the one during the post-Run sleep/rest session , in naïve and experienced animals , in the presence and absence of CA3 NMDARs . This approach allowed us to identify and compare three distinct forms of novelty encoding as revealed by prior experience and CA3 NMDAR KO . In the contiguous and disjunct condition paradigms , although the spatial location and orientation of the novel linear track are as new to the animals as in the de novo condition paradigm , both the geometry ( i . e . , linear tracks ) and the behavioral experience ( i . e . , repetitive runs for food rewards ) are common . These parameters may have already been internalized prior to the novel run session , which may have diminished the dynamic interplay between the internal and external drives and facilitated the formation of more stable spatial representations on the novel arm/track . More importantly , the repeated access to the familiar arm in conjunction with the exploration of the novel arm in the contiguous condition accelerated the recruitment and stabilization of neuronal firing sequences independent of CA3 NMDARs; this likely happened through complementary , non-CA3 NMDAR-dependent plasticity or through NMDAR-dependent activity in other brain regions ( Kentros et al . , 1998 ) . The CA3 NMDAR independence in the contiguous condition did not solely result from increased experience with linear tracks , as exploration of isolated novel tracks by the even more experienced animals during the disjunct condition did require , transiently , these receptors for the rapid formation of stable , tuned place cell sequences . In the disjunct condition , since exposure to the novel track occurred in the same general spatial environment , modules of place cells that remapped together , possibly controlled by common external stimuli ( Lee et al . , 2004 ) , were regrouped in a CA3 NMDAR-dependent manner to rapidly form the new representation . We hypothesize that this regrouping process observed in the disjunct condition makes the need for CA3 NMDARs only transient and thereby facilitates the formation of a new spatial representation based on the prior experience . A more drastic change in the external stimuli in the absence of prior animal training ( like in the de novo condition where animals were shifted for the first time from the sleep/rest box to the linear maze ) may lead to a more complete CA3 NMDAR-dependent recruitment and stabilization of neuronal ensemble activity , a slower formation of new spatial representations , and a slower learning . Our data have shown that a novel representation of a first-time experience on linear tracks ( DnRun1 ) is formed in the CA1 area primarily on the framework of the preconfigured hippocampal network ( preplay ) , which is modified , in part , during the experience and is rapidly stabilized primarily via CA3 NMDAR-dependent activity ( Figure 7 ) . The changes in place cell activity occur without affecting the general stability of the hippocampal network , which indicates a homeostatic regulation of its temporal sequence activity in response to novelty: ∼10% of the temporal sequences reflect the novel experience during the subsequent sleep/rest period , a proportion that is not different from the proportion of corresponding preplay events . Subsequent exposures to the same track ( DnRun2 ) and additional contiguous ( ContigRun ) or isolated novel tracks ( DisjRun1 ) resulted in the expression of similar proportions of correlated temporal sequences before and after the corresponding novel spatial experiences , both in control and CA3 NMDAR KO mice . 10 . 7554/eLife . 01326 . 015Figure 7 . Cartoon model of the internal representation of novel experiences based on the organization of neurons in cellular assemblies . ‘Hexagonal’ panels ( panels 1 , 4 , 6 , 8 ) , network of sequentially activated neurons ( cell assemblies ) during sleep/rest under different experimental conditions: pre-DnRun1 sleep/rest ( panel 1 ) , post-DnRun1 sleep/rest ( panel 4 ) , pre-ContigRun sleep/rest ( panel 6 ) , and post-ContigRun sleep/rest ( panel 8 ) . Arrows indicate potential ( thin ) or actual ( bold ) temporal order of activation during Run in CA1 , not anatomical connectivity . All arrows during sleep/rest indicate the temporal order of cell activation during sleep/rest . Black bold arrows during sleep/rest emphasize temporal replay . Upper case letters: corresponding individual cells/assemblies . Colors: sequential cell assemblies co-active on a given linear track . Cells A , C , and D are active on both the familiar and novel arms . White circles: cells with no place field during the corresponding run session . ‘Linear’ panels ( panels 2 , 3 , 5 , 7 ) , sequences of place cells during different sessions of run on linear tracks: DnRun1 ( early , panel 2 and late , panel 3 , of the run session ) , DnRun2 ( panel 5 ) , and ContigRun ( panel 7 ) . Letters , colors , and order of activation correspond to the ones during sleep/rest sessions . Long thin arrows next to the panel represent the direction of the animal’s movement during run . DOI: http://dx . doi . org/10 . 7554/eLife . 01326 . 015 We propose that the overall robust homeostasis of the hippocampal network seen at the temporal sequence level and expressed during offline states of sleep and rest reflects the default sequential cell assembly architecture of the hippocampal network shaped by the multiple unaccounted experiences the animal has had in the past . Our proposal is consistent with previous studies reporting that temporal sequences emitted during onsite resting periods do not specifically reflect the recent spatial experience of the animal ( Karlsson and Frank , 2009; Gupta et al . , 2010; Dragoi and Tonegawa , 2011; Pfeiffer and Foster , 2013 ) , but rather reflect multiple related spatial experiences the animal had experienced or will experience in the near future in that particular environment ( Dragoi and Tonegawa , 2011 , 2013 ) . The overall stability of the hippocampal network during sleep/rest epochs on both sides of the novel spatial experiences does not mean that the novel experiences did not induce more discrete plastic changes which are apparent at the individual cell level in the network ( Dragoi et al . , 2003 ) . More importantly , in the absence of CA3 NMDAR-dependent activity , the CA1 temporal firing sequences appear more rigid and their correlations with the place cell sequences are less modulated by the recent experiences compared to control animals . Overall , in the absence of CA3 NMDARs , the event correlations with future place cell sequences exhibit lower variance and higher values than in the presence of CA3 NMDARs . These results indicate that the organization of cellular assemblies in the CA1 area is influenced by the NMDAR-dependent activity in the upstream CA3 area . In the absence of this type of activity/plasticity , the CA1 cellular assemblies are less affected by the animals’ novel spatial experiences and maintain an increased correlation across different brain states ( i . e . , across sharp-wave/ripples during sleep and theta during run ) , behaviors , and experiences . This could explain why the mutant animals exhibit deficits in one trial learning which involves rapid plastic changes in hippocampal cellular assemblies ( Nakazawa et al . , 2003 ) . A hallmark of neurons’ organization in cellular assemblies is their onsite coordinated activation across similar animal behaviors such as running laps ( Wills et al . , 2005; Dragoi and Buzsaki , 2006 ) . Here we show that upon familiarization with a novel linear track , the lap-by-lap correlations between place cell pairs increase rapidly during the exploratory session in control animals but remain low for several exploratory sessions in the absence of CA3 NMDARs . These results suggest that NMDAR-dependent activity within the CA3 area of the hippocampus is involved in the rapid organization and linking of CA1 place cells in cellular assemblies during the encoding of a first-time and of subsequent isolated novel experiences on linear tracks , but is not necessary for the general expression of temporal sequences of place cells in the CA1 in the form of preplay and replay . Our results have relevance for theories of learning and memory consolidation . Quite often , a single exposure to absolutely novel experiences does not lead to a lasting memory of the experience , whereas repeated exposures to the same experience or prior knowledge with similar kind of experiences result in more rapid learning and memory consolidation ( Tse et al . , 2007 ) . We propose that the increased map stability , tuning , and cell assembly co-variation we find in repeatedly exposed and in experienced , but not in naïve animals underlie the rapid consolidation of episodic memories of repeated or related new experiences , but not of entirely novel ones . The latter may instead require repeated exposure to engage hippocampal replay-dependent mechanisms for memory consolidation . The intact hippocampus is essential for encoding and rapid consolidation of memory ( Scoville and Milner , 1957; Squire , 1992; Eichenbaum et al . , 1999 ) and for associative-novelty detection ( Kumaran and Maguire , 2007 ) . The role of the hippocampus and of NMDA receptors in learning and memory consolidation is manifested primarily in experimentally naïve animals and diminishes with experience ( Bannerman et al . , 1995; Otnaess et al . , 1999 ) , when new information is presumably rapidly integrated into pre-existing neocortical frameworks of knowledge , or schemas ( Morris , 2006; Tse et al . , 2007; McKenzie and Eichenbaum , 2011 ) . The neural substrates of such learning and memory processes are believed to be the formation of stable , finely tuned cellular assemblies ( Hebb , 1949 ) across the neocortex and the hippocampus ( Tse et al . , 2011 ) . We propose that , in addition to changes in the activity of individual CA1 neurons , their ability to rapidly and flexibly organize in stable cellular assemblies underlies the process of learning and memory . Familiarization with the spatial environment and the behavioral task are associated with the formation of cortical mental schemas ( Tse et al . , 2007; McKenzie et al . , 2013 ) that rely on the stability of neocortical-hippocampal cellular assemblies . These function like strong neural attractors ( Tsodyks , 1999; Wills et al . , 2005 ) that will integrate future neuronal representations . The group of CA1 place cell pairs that maintained high levels of lap-by-lap co-variation and preserved the relative distance between their place fields across different environments independent of CA3 NMDARs ( stable assemblies , disjunct condition; contiguous condition ) may represent one neuronal mechanism underlying schema-based accelerated learning . Previously , lesion experiments have argued for a transient , but necessary role of the intact hippocampus in the assimilation and consolidation of new information into schemas ( Tse et al . , 2007 ) . Whereas not speaking for the whole hippocampus , our data indicate that CA3 NMDAR-dependent activity and synaptic plasticity are not necessary for either the rapid assimilation of new contiguous locations into a previously established spatial representation or for learning of a related alternation task . Instead , this type of plasticity is necessary for the rapid formation of new CA1 cell assemblies in the hippocampus of experimentally naïve animals and the development of a new schema associated with the first-time learning of a hippocampal-dependent alternation task ( Figure 7 ) . The existence of a mechanistic dichotomy between these different forms of learning may help explain why hippocampal dysfunction results in anterograde amnesia ( Scoville and Milner , 1957 ) while recollection of old , schema-based memories ( Winocur et al . , 2005 ) is preserved . All animals were implanted under Avertin anesthesia with six independently movable tetrodes aimed at the CA1 area of the right hippocampus ( 1 . 5–2 mm posterior to bregma and 1–2 mm lateral to the midline ) . The reference electrode was implanted posterior to lambda over the cerebellum . During the following week of recovery , the electrodes were advanced daily while animals rested in a small sleep/resting box ( 12 × 20 × 35 [hr] cm ) having opaque walls . The animal position was monitored via two infrared diodes attached to the headstage . The experimental apparatus consisted of a 90 × 65 cm rectangular , walled linear track maze . All tracks were 4 cm wide at the bottom and 8–9 cm at the top; all linear track walls were translucent , 10 cm high , with opaque , uniform color barriers . Recording sessions ( Table 1 ) were conducted while the animals explored for chocolate sprinkle rewards placed always at the ends of the corresponding linear tracks ( one sprinkle at each end of the track on each lap ) . Under the de novo condition , the neuronal activity was recorded in naïve animals ( four CT mice , CT1–4 , and four KO mice , mice KO1–4 ) during the sleep/rest session in the sleep/rest box ( pre-DnRun1 ) immediately preceding the first experience on the linear track and the recordings continued during the first run session on a novel track ( DnRun1 ) . Following their first run experience on the linear track , the animals were placed back in the sleep/rest box and allowed to sleep/rest ( post-DnRun1 or pre-DnRun2 ) , after which they were exposed for a second session of run on the same linear track ( DnRun2 ) . This run was followed by another session in the sleep/rest box ( post-DnRun2 ) . The first two ( KO1 and KO2 ) out of the eight recorded mice were exposed for the third time to the linear track , followed by an additional session in the sleep/rest box . The remaining six mice went through , under the de novo condition , two run sessions ( DnRun1 and DnRun2 ) each preceded and followed by a sleep/rest session . In the two KO mice that were exposed for a third session on the first novel track , the fields were not significantly more tuned and the lap-by-lap correlations were not as high in this session compared with their next day FamRun1 session or with the DnRun2 session in controls . There was no improved spatial tuning during the additional run session in the first two KO mice . In one KO animal ( mouse KO4 ) , no spiking events could be detected during de novo sleep/rest sessions due to the below threshold number of synchronously active cells . In the contiguous condition , following a recording session on the now familiar linear track ( FamRun1 ) , a barrier that was blocking access to a contiguous novel linear track was lifted and the animals explored the L-shaped linear track for the first time ( ContigRun ) . The orientation of the L-shaped track in the room and the room landmarks were kept constant throughout the experiment . Sessions in the sleep/rest box preceded ( pre-ContigRun; before the barrier was lifted ) and followed ContigRun ( post-ContigRun; after the barrier was lifted ) . In one CT animal ( mouse CT1 ) , no spiking events were detected during pre-ContigRun due to the below threshold number of synchronously active cells . In three KO ( mice KO1 , KO2 , and KO4 ) and two CT animals ( mice CT1 and CT4 ) , FamRun1 was recorded after an overnight sleep , whereas FamRun1 was recorded several hours after the de novo exposure to the novel track in the remaining animals . In three animals ( CT4 , KO1 , and KO4 ) the pre-ContigRun sleep/rest session was preceded by a run session on the now familiar track , fRun ( Table 1 ) . In the Disjunct condition , the animals ( three CT and three KO mice , Table 1 ) were re-exposed to the now familiar L-shaped track 2 days later , after which they were allowed to sleep/rest in the sleep/rest box ( pre-DisjRun1 session ) . Subsequently , they explored an additional linear track on the same maze apparatus in isolation , separated by barriers at both ends from any familiar track , for two run sessions ( DisjRun1 and DisjRun2 ) separated by a sleep/rest session in the sleep/rest box ( post-DisjRun1 or pre-DisjRun2 ) . All of the Run data were collected while the animals ran on the tracks ( velocity of animal’s movement higher than 5 cm/s ) , whereas all sleep/rest data were collected while animals were in the sleep/rest box ( velocity less than 1 cm/s , and overwhelmingly 0 cm/s ) . A total of 458 neurons were recorded from the CA1 area of the hippocampus in four CT and four KO mice across the experimental sessions . Of these , 69 neurons in CT ( 13 , 20 , 26 , and 10 in CT1–4 ) and 74 neurons in KO mice ( 20 , 27 , and 27 in KO1–3 ) were recorded in the de novo condition , 75 neurons in CT ( 25 , 23 , and 27 in CT2–4 ) and 88 neurons in KO ( 18 , 25 , 24 , and 21 in KO1–4 ) were recorded in the Contig condition , whereas 76 neurons in CT ( 26 , 23 , and 28 in CT2–4 ) and 77 neurons in KO ( 20 , 29 , and 28 in KO1 , 3–4 ) were recorded in the disjunct condition . Single cells were identified and place fields were computed as described earlier ( Dragoi and Tonegawa , 2011 ) . Spatial information was calculated for each individual cell in non-overlapping 2 cm spatial bins as described earlier ( Skaggs et al . , 1996 ) and the average values within-sessions/conditions were normalized for each genotype by the average values during FamRun1 ( de novo and contiguous conditions ) and FamRun2 ( disjunct condition ) to compare the across session changes in both genotypes . To analyze preplay and replay processes , spiking events were detected during pre- and post-Run sleep/rest periods in the sleep/rest box in all experimental conditions . A spiking event ( Dragoi and Tonegawa , 2011 ) was defined as a transient increase in the firing activity of a population of at least five different place cells within a temporal window preceded and followed by at least 100 ms of silence . For all conditions , the spikes from all the place cells active during run that were emitted during the preceding and following sleep/rest were sorted by time and further used for the detection of the spiking events . For the calculation of the temporal sequence , the times of the first spike emitted by each of the cells participating in the spiking event were sorted to determine the temporal order of neuronal firing ( Diba and Buzsaki , 2007; Dragoi and Tonegawa , 2013 ) . All four CT and three KO animals exhibited a significant number of spiking events in the sleep/rest sessions of the de novo condition , three CT and four KO animals exhibited a significant number of spiking events in the contig condition , whereas all three CT and three KO mice exhibited a significant number of spiking events in the Disjunct condition . The remaining animals had a below threshold number of simultaneously active CA1 place cells . The place cell sequences ( templates ) were calculated for each direction of the animal’s movement and for each run session in all experimental conditions by ordering the spatial location of the place field peaks that were above 1 Hz . For place cells with multiple place fields above 1 Hz on a particular arm or track in the contiguous condition , only the place field corresponding to the peak firing rate of the place cell on that arm or track was considered for the construction of the template of that particular arm or track . This method is consistent with previous studies that employed spatial templates to demonstrate replay ( Lee and Wilson , 2002; Foster and Wilson , 2006; Diba and Buzsaki , 2007 ) and preplay ( Dragoi and Tonegawa , 2011 ) during sleep or awake rest . Place cells with fields on both the novel arm in the ContigRun session and the familiar track in the FamRun1 session participated in the construction of both the novel arm and familiar track templates . Statistical significance was calculated for each event by comparing the rank-order correlation between the sequence of cells’ firing in the event ( i . e . , event sequence ) and the place cell sequence ( template ) and the distribution of correlation values between the event sequence and 100 surrogate templates obtained by shuffling the order of place cells ( Diba and Buzsaki , 2007; Dragoi and Tonegawa , 2011 ) . The significance level was set at 0 . 025 to control for multiple comparisons ( i . e . , the two directions of run ) . The proportions of significant events ( preplay and replay ) were calculated as the ratio between the number of significant events and the total number of spiking events in which at least five corresponding place cells were active ( Diba and Buzsaki , 2007; Dragoi and Tonegawa , 2011 ) . Ripple oscillations were detected during sleep/rest periods in the sleep/rest box . The EEG signal was filtered ( 120–200 Hz ) and ripple-band amplitude was computed using the Hilbert transform . Ripple epochs with maximal amplitude higher than four standard deviations above the mean , beginning and ending at one standard deviation were detected . The overall significance of the preplay or replay process was calculated by comparing the group of correlation values of all events relative to the original template with each of the 100 groups of an equal number of correlation values relative to the shuffled surrogate templates using the ranksum test . The highest p value out of the resulting 100 p values ( the weakest significant level ) is further reported ( Figure 4 ) , except for the contiguous condition in CT where the criterion of p<0 . 025 was applied due to smaller , yet significant , differences between the original data and the shuffles ( Figure 4E ) . The stability of place cell firing on the novel track ( de novo and disjunct conditions ) and novel arm ( contig condition ) in the beginning vs the end of the run session were assessed by calculating , for each place cell and each direction , a correlation between the spatial firing in the corresponding paired situations ( i . e . , the first four laps vs the last four laps of the run on the novel track or arm [Dragoi and Tonegawa , 2011] ) . The place cell activity was not partitioned in place fields , rather the whole activity on the particular track or arm was considered separately for each cell and direction ( average correlations are shown in Figure 1E–G ) . In addition , we performed the same type of correlations while shuffling the identity of the cell in one member of the correlation ( once for each different cell ) . Shuffle results were computed as correlation between spatial tuning of cells on the novel arm ( or novel track ) during the beginning of ContigRun ( or de novo and disjunct runs ) and spatial tuning of all the other simultaneously recorded cells on the novel arm ( or novel track ) during the end of ContigRun ( Novel arm group ) or de novo and disjunct runs . Original and shuffled correlations were compared using the ranksum test . In all cases , the original correlations were significantly higher than the shuffled ones ( shuffled correlations values were ∼0 . 2 across conditions ) . For the calculation of lap-by-lap correlations , spike times of each place field ( velocity >5 cm/s ) were binned at 3 s . After excluding the common zero-value bins , a correlation coefficient was calculated between the binned activities of pair members for all place field pairs ( Dragoi and Buzsaki , 2006 ) . This measure reflects the degree of co-variation of neuron pairs on multiple laps ( trials ) , and its value is strongly correlated with the theta timescale temporal correlation of pairs of neurons ( Dragoi and Buzsaki , 2006 ) . The absolute value of significant correlation values ( p<0 . 05 ) were compared across sessions and genotype . The mice ran 12–18 laps/session , similar numbers across genotypes ( p>0 . 5 , ranksum test ) . The duration of each session ( trial ) is entered in the Table 1 . For the ContigRun session of the contiguous condition only , we could separate the place fields active on the familiar arm from those active on the novel arm . We calculated three sets of correlation values , between pairs of place fields that were: ( 1 ) Both expressed on the familiar arm ( Fam × Fam ) , ( 2 ) Both expressed on the novel arm ( Novel × Novel ) , and ( 3 ) Expressed one on the familiar arm and one on the novel arm ( Fam × Novel ) . Features of temporal sequences were compared across genotypes using a paired t test applied to grand averages of parameters computed on data recorded during eight sleep/rest sessions: pre-DnRun1 , post-DnRun1 , pre-DnRun2 , post-DnRun2 , pre-ContigRun , post-ContigRun , pre-DisjRun1 , and post-DisjRun1 . For each genotype , the data were calculated for each sleep/rest session for each individual mouse as well as by averaging the specific parameters collected from all of the corresponding mice . The behavioral data were collected from a total of 12 mice performing a delayed alternation task in two configurations of a T-maze . The size of the T-maze apparatus was 90 × 90 cm and the 3-D dimensions of the linear tracks were configured as in the neurophysiology part of the experiment . The mice were food deprived over 1 week to 85% of their body weight and were further trained from naïve state for ten days to alternate between the two arms of T-maze 1 ( Figure 6A inset , vertical arms ) for food rewards placed at each arm end . No barriers were ever used in the choice phase throughout training . On the return from the end of the choice arm to the start point located at the free end of the stem arm ( Figure 6A inset , horizontal arm ) , a temporary barrier blocked animal access to the other choice arm . The criterion for learning was set at 70% correct choices per session for 2 consecutive days for each group of mice ( i . e . , KO and CT ) . Each mouse was trained for one session of ∼20 trials each day . At the end of the 10 days over which both groups reached the criterion , all mice were exposed to a second configuration of the T-maze ( T-maze 2 ) for 2 additional days . From the beginning to the end of each trial over the 12 days , mice behave freely in T-mazes: they self-initiated their first trial , made a choice of an arm , returned to the start point , and self-initiated the next trial , for ∼20 trials/session/day . For each day of the experiment and for each genotype , the performance of all animals was averaged and entered as a data point ( Figure 6 ) . For the first 10 days of the experiment , data were grouped in five blocks of two consecutive days and analyzed using a balanced two-way ANOVA test followed by the ranksum test .	Learning is an inherent feature of the living animals . During development and childhood we learn a large repertoire of items that we encounter for the first time in our life , such as the names of things and how to perform certain tasks . This de novo learning process takes a relatively long time and generally requires repeated exposures to the particular features of the external world that are being learned . Later in life , when we are exposed to novel but related features , we acquire this new information much faster . For instance , if a child learns to associate various odors with various locations around the home—such as associating the smell of bread with the kitchen and the smell air freshener with the bathroom—they will find it easier to make new odor-place associations outside the home , such as associating gasoline with gas stations . It is thought that the de novo learning process is achieved when the newly acquired information is being consolidated and transferred into long-term memory storage within networks of neurons in the brain . The process of consolidation is believed to lead to the formation of mental schemas that can accelerate learning of novel but related information . Although the concepts of mental schema and related learning are widely used in psychology and education , their underlying neuronal mechanisms are poorly understood . The formation of new memories depends on a part of the brain called the hippocampus and involves changes in the strength of the connectivity between groups of neurons in a process called synaptic plasticity . In particular , the interaction between a chemical called glutamate , which is released by sender neurons , and proteins called NMDA receptors ( which bind the glutamate molecules ) on receiver neurons have a central role in synaptic plasticity . Recently , based on experiments with rodents , it has been proposed that the hippocampus is also crucial for the formation of the mental schemas that can accelerate the learning of new spatial association tasks , such as the odor-place associations described above . Now , Dragoi and Tonegawa reveal that the NMDA receptor in a key subregion of the hippocampus is also involved in the de novo learning of spatial tasks . Using repeated exposures to novel spatial experiences and genetic techniques to block the NMDA receptors in this subregion in mice , Dragoi and Tonegawa discovered that de novo learning involves synaptic plasticity in the hippocampus and , possibly , other regions of the brain . This de novo learning , in turn , enables subsequent spatial learning to be accelerated , even when the NMDA receptors are absent . These results reveal that de novo learning , and related learning processes such as accelerated learning , are underpinned by a number of different mechanisms in the brain , which could help explain why damage to the hippocampus prevents the formation of new memories while preserving other forms of memory and learning .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2013	Development of schemas revealed by prior experience and NMDA receptor knock-out
T cell expansion and differentiation are critically dependent on the transcription factor c-Myc ( Myc ) . Herein we use quantitative mass-spectrometry to reveal how Myc controls antigen receptor driven cell growth and proteome restructuring in murine T cells . Analysis of copy numbers per cell of >7000 proteins provides new understanding of the selective role of Myc in controlling the protein machinery that govern T cell fate . The data identify both Myc dependent and independent metabolic processes in immune activated T cells . We uncover that a primary function of Myc is to control expression of multiple amino acid transporters and that loss of a single Myc-controlled amino acid transporter effectively phenocopies the impact of Myc deletion . This study provides a comprehensive map of how Myc selectively shapes T cell phenotypes , revealing that Myc induction of amino acid transport is pivotal for subsequent bioenergetic and biosynthetic programs and licences T cell receptor driven proteome reprogramming . Immune activation transcriptionally reprograms T lymphocytes and initiates changes in cell metabolism and protein synthesis that are required for proliferation and effector differentiation . The signalling pathways that control T cell metabolism are not fully characterised but it has been shown that the transcription factor Myc has a necessary role ( Wang et al . , 2011 ) . In T cells , Myc is rapidly induced in response to engagement of the T cell antigen receptor ( TCR ) and Myc expression is then sustained by costimulatory receptors and cytokines such as interleukin-2 ( IL-2 ) ( Au-Yeung et al . , 2017; Heinzel et al . , 2017; Preston et al . , 2015 ) . The TCR acts as a digital switch for Myc mRNA expression , in that the strength of the antigen stimulus determines the frequency of T cells that switch on Myc mRNA expression ( Preston et al . , 2015 ) . Antigen receptor , costimulation and cytokine driven processes also post-transcriptionally control Myc protein: constant phosphorylation on Thr58 by glycogen synthase kinase 3 ( GSK3 ) and subsequent proteasomal degradation results in a short cellular half-life of Myc protein ( Preston et al . , 2015 ) . O-GlcNAcylation of Myc at this same residue ( Chou et al . , 1995 ) , fuelled by the hexosamine biosynthesis pathway , blocks this degradation and allows Myc to accumulate ( Swamy et al . , 2016 ) . In activated lymphocytes the sustained expression of Myc is also dependent on the rate of protein synthesis and availability of amino acids ( Loftus et al . , 2018; Sinclair et al . , 2013; Swamy et al . , 2016; Verbist et al . , 2016 ) . Myc expression is thus tightly controlled at the population and single cell level during immune responses . The expression of Myc is essential for T cell immune responses and mature T cells with Myc alleles deleted cannot respond to antigen receptor engagement to proliferate and differentiate ( Preston et al . , 2015; Trumpp et al . , 2001; Wang et al . , 2011 ) . Myc-deficient T cells have defects in glucose and glutamine metabolism ( Wang et al . , 2011 ) ; however , the full molecular details of how Myc regulates T cell metabolic pathways and other aspects of T cell function is not fully understood . In this context there are different models of how Myc works and divergent opinions as to whether or not Myc acts a general amplifier of active gene transcription ( Lewis et al . , 2018; Lin et al . , 2012; Nie et al . , 2012 ) or has more selective actions ( Sabò et al . , 2014; Tesi et al . , 2019 ) . There is also evidence Myc can act post transcriptionally , controlling mRNA cap methylation and broadly enhancing mRNA translation ( Cowling and Cole , 2007; Ruggero , 2009; Singh et al . , 2019 ) . The salient point is that there appear to be no universal models of Myc action that can be applied to all cell lineages . As an example , it is reported that oncogenic Myc mutants control amino acid transporter expression in tumour cells ( Yue et al . , 2017 ) whereas analysis of endogenous Myc function in immune activated primary B cells found no such role ( Tesi et al . , 2019 ) . These discepancies highlight the neccessity for direct experimental analysis to understand how Myc controls T lymphocyte function rather than simply being able to extrapolate from other cell models . In this context , T lymphocytes are critical cells of the adaptive immune response and understanding the signalling checkpoints that control T cell function is fundamental for any strategy to manipulate T cell function for immunotherapy or immunosuppression . T cell immune activation is associated with increases in mRNA translation , amino acid transport and protein synthesis all of which shape the execution of the T cell transcriptional program and completely reshape the T cell proteome ( Araki et al . , 2017; Geiger et al . , 2016; Howden et al . , 2019; Ricciardi et al . , 2018; Sinclair et al . , 2013 ) . Hence , one way to gain a full and unbiased understanding of how Myc controls T cell metabolism and T cell function is an in-depth analysis of how Myc shapes T cell proteomes . Accordingly , we have used high-resolution mass-spectrometry to perform a quantitative analysis of the impact of Myc deficiency on the proteomes of immune activated CD4+ and CD8+ T cells . These data reveal a selectivity of Myc action in co-ordinating T cell proteomes and identify both Myc dependent and Myc independent remodelling of T cell metabolic programs . The data uncover that a primary function of Myc in T lymphocytes is control of amino acid transporter expression which affords new insight about how Myc controls T cell biosynthetic and bioenergetic programs . To explore how Myc controls T cell function we used a Cd4Cre+Mycfl/fl ( MyccKO ) mouse model in which Myc is conditionally deleted during late thymic development ( Dose et al . , 2009; Mycko et al . , 2009; Trumpp et al . , 2001 ) . As shown previously ( Wang et al . , 2011 ) , Myc-deficient CD4+ and CD8+ T cells do not substantially increase cell size or proliferate in response to immune activation with anti-CD3/anti-CD28 agonist antibodies ( Figure 1A , Figure 1—figure supplement 1A ) . To examine how Myc loss impacts proteome remodelling during immune activation we performed quantitative label-free high-resolution mass spectrometry on 24 hour CD3/CD28 activated wild-type ( Cd4Cre + , MycWT ) and MyccKO CD4+ and CD8+ T cells . This time point was chosen as it is when we observe maximal increase in cell size of the immune activated cells with no difference in survival between MycWT and MyccKO T cells . Moreover , at this time point there is minimal impact of autocrine secreted cytokine IL-2 on Myc expression ( Figure 1—figure supplement 1B ) . >7000 proteins were identified and protein mass and copy number per cell was estimated by the ‘proteomic ruler’ method which uses the mass spectrometry signal of histones as an internal standard ( Supplementary file 1; Wiśniewski et al . , 2014 ) . The data in Figure 1B show that in contrast to Myc T cells , CD3/CD28 activated MyccKO T cells fail to increase protein content above the level of naive ex vivo isolated MycWT T cells . Hence , the increase in cell biomass that accompanies T cell activation is dependent on Myc . Notably , the protein content of immune activated MycWT CD4+ T cells was lower than activated CD8+ T cells and this correlates with higher levels of Myc in immune activated CD8+ versus CD4+ MycWT T cells ( Figure 1C–D ) . The immune activation of T cells is accompanied by complex proteome remodelling ( Geiger et al . , 2016; Howden et al . , 2019; Ron-Harel et al . , 2016; Tan et al . , 2017 ) . A key question is whether the dramatically lower cell mass in CD3/CD28 activated MyccKO T cells reflects a scaled decrease in expression of all proteins or a selective loss of protein expression . In this respect a few hundred very abundant proteins are known to account for most cellular mass ( Howden et al . , 2019; Hukelmann et al . , 2016; Ly et al . , 2014 ) , with 75% of the protein mass of immune activated MycWT CD4+ and CD8+ T cells comprising 344 and 391 proteins respectively . Myc-deficiency reduced the expression of most , but not all of these abundant proteins ( Figure 1E–F ) . To assess the selectivity of Myc control of T cell proteomes we used nearest neighbour analysis and Pearson correlation to group and align the expression profile of ~6400 proteins in naïve and immune activated MycWT and MyccKO CD4+ and CD8+ T cells ( Figure 1G ) . These analyses highlight how CD3/CD28 stimulation dynamically reshapes the proteomic landscape of CD4+ and CD8+ T cells . The impact of Myc loss is striking but clearly selective and not a simple scaled decrease in expression of all proteins . There are a number of proteins expressed at high levels in naïve cells and downregulated by immune activation in both MycWT and MyccKO T cells ( Figure 1G ) , including Kruppel family transcription factors which maintain pluripotency and cell quiescence ( eg Klf2 ) and growth factors receptors such as the IL7 receptor ( Figure 1H–I ) . There is also a subset of ~300–450 proteins that are strongly induced by immune activation irrespective of Myc expression ( Figure 1G ) . These include CD69 , CD44 and transcription factors cRel and JunB ( Figure 1J–M ) . The critical transcription factors T-bet and Irf4 were also upregulated in immune MyccKO T cells , albeit at reduced levels compared with MycWT T cells ( Figure 1N–O ) . The selective effects of Myc-deficiency on protein expression in activated CD8+ and CD4+ T cells appeared qualitatively similar ( Figure 1G ) . There were however some quantitative differences . These differences reflect that some proteins were more highly expressed in activated MycWT CD8+ T cells than in MycWT CD4+ T cells , however , Myc-deficiency reduced protein expression down to a similar level in both CD4+ and CD8+ T cells , therefore giving a larger effect size in CD8+ T cells ( Figure 1—figure supplement 2 ) . When taken in conjunction with the observation that CD8+ T cells expressed a higher level of Myc ( Figure 1C–D ) , associated with increased cell biomass ( Figure 1A–B ) , this suggests a dose-dependent Myc-driven amplification of protein expression . Collectively , these data show that immune activated T cell proteome remodelling comprises both Myc dependent and independent processes and that Myc has a qualitatively similar , but dose-dependent effect on CD4+ and CD8+ T cell proteomes . When examining the selective effects of Myc-deficiency on T cell immune activation we observed that MyccKO T cells increased expression of the glucose transporters Slc2a1 and Slc2a3 ( Glut1 and Glut3 respectively ) equal to , or exceeding the level seen in MycWT T cells in response to T cell activation ( Figure 2A–B ) . The ability of immune activated MyccKO T cells to upregulate expression of Slc2a1 and Slc2a3 glucose transporters was unexpected as it has been reported that Myc-deficient T cells have abnormal glycolytic metabolism and defective induction of glucose transporter mRNA ( Wang et al . , 2011 ) . Moreover , Slc2a1 has been implicated as a direct transcriptional target of Myc ( Osthus et al . , 2000 ) . In this context , CD3/CD28 triggering increases expression of glycolytic enzymes in both MycWT and MyccKO CD4+ and CD8+ T cells ( Figure 2C , left panel ) . Although the cumulative levels of glycolytic enzymes in MyccKO are reduced by 58% and 30% in CD8+ and CD4+ T cells respectively compared with MycWT controls , they still comprise a large percentage of the proteomes of immune activated MyccKO T cells ( Figure 2C , right panel ) . It was however striking that Myc had a large impact on lactate transporter expression , particularly on the numerically dominant lactate transporter Slc16a1 ( Figure 2D ) . Lactate transporters control a critical rate limiting step for glycolytic flux ( Tanner et al . , 2018 ) . Their absence would prevent lactate export and feedback to suppress glycolytic flux ( Doherty et al . , 2014 ) . Slc16a1 expression increases from <10 , 000 copies per naïve T cell to ~140 , 000 and~80 , 000 copies per immune activated CD8+ and CD4+ MycWT T cell respectively . In contrast , Slc16a1 expression in immune activated MyccKO T cells remains equivalent to naive levels ( Figure 2D , Supplementary file 1 ) . These data display the selectivity of Myc importance for expression of key components of the glycolysis machinery and point to control of lactate export as a mechanism whereby Myc controls glycolytic flux in T cells . Another key Myc controlled metabolic process is glutamine catabolism ( Wang et al . , 2011; Wise et al . , 2008 ) . Once imported glutamine can be metabolised in a number of different processes , including the hexosamine pathway , nucleotide biosynthesis processes , and the citric acid cycle ( Figure 2E ) . The present data reveal the selectivity of the Myc requirement for expression of important enzymes for glutamine metabolism . Myc controls expression of glutaminase ( Gls ) , Cad and Ppat , the enzymes that control the first steps in glutaminolysis , and pyrimidine and purine biosynthesis respectively . However , expression of Gfpt1 , the first and rate limiting step in the hexosamine pathway and Glud1 , the enzyme that converts glutamate to a-ketoglutarate are still expressed in MyccKO T cells ( Figure 2F and Supplementary file 1 ) . One major effect of Myc loss on immune activated T cells is failure to increase cell mass ( Figure 1A–B ) . In this context , immune activation of T cells decreases expression of translational repressors and drives increased expression of ribosomes and mRNA translational machinery ( Geiger et al . , 2016; Howden et al . , 2019; Ron-Harel et al . , 2016; Tan et al . , 2017 ) . The data in Figure 3—figure supplement 1A–C shows that Myc loss does not prevent loss of the translational repressor Pdcd4 in activated T cells . Myc-deficiency did however suppress CD3/CD28 mediated increases in expression of ribosomes , eukaryotic initiation factor 4 ( eIF4F ) complexes that translate methyl capped mRNAs and EIF2 complexes which controls tRNA transfer to ribosomes . Although increasing expression of translational machinery is important , an absolutely fundamental requirement for a substantial increase in cell mass is availability of amino acids ( Hosios et al . , 2016 ) . Therefore , it is striking that the loss of Myc prevents the upregulation of expression of multiple amino acid transporters in activated T cells ( Figure 3A–B ) . The most abundant amino acid transporters expressed on CD3/CD28 activated CD4+ and CD8+ T cells are Slc7a5 ( leucine , methionine , tryptophan ) , Slc1a5 ( glutamine , serine , threonine , alanine ) , Slc38a1 and Slc38a2 ( glutamine , methionine ) and Slc7a1 ( arginine , lysine ) ( Figure 3A , Supplementary file 1 ) . Naïve T cells have very low levels of all of these transporters , expressing ~500–2500 copies per cell ( Figure 3A , Supplementary file 1 ) . Upon activation , amino acid transporters are some of the most highly induced proteins in MycWT T cells , exhibiting up to 100-fold increases relative to naïve cells ( Figure 3A–B ) . In contrast , immune activated MyccKO T cells only express amino acid transporters at near naïve levels ( Figure 3A–B , Supplementary file 1 ) . The high levels of protein production in activated T cells would need to be fuelled by amino acid supply ( Hosios et al . , 2016 ) . Moreover , T cells that lack expression of key amino acid transporters such as Slc7a5 and Slc1a5 are defective in their response to T cell activation ( Nakaya et al . , 2014; Sinclair et al . , 2013 ) . We therefore questioned whether the ability of Myc to control T cell growth could be explained by Myc control of amino acid transporter expression . Accordingly , we examined the impact of Myc expression on the functional capacity of T cells to transport amino acids and we assessed whether the loss of amino acid transporter expression could recapitulate the striking impact of Myc deletion on T cell protein production . We focused on the system L transporter Slc7a5 , as this is the most abundant amino acid transporter expressed on immune activated T lymphocytes ( Figures 3A and 4B , Supplementary file 1 ) and mediates transport of many essential amino acids including methionine , leucine , isoleucine , valine , phenylalanine and tryptophan ( Sinclair et al . , 2019; Sinclair et al . , 2018; Sinclair et al . , 2013 ) . Low basal levels of Slc7a5 in naïve T cells mediate amino acid uptake that is not dependent on Myc ( Figure 3C ) . Within 4 hr of T cell activation there is already increased system L transport activity in MycWT T cells and this increase is substantially lower in MyccKO CD4+ and CD8+ T cells ( Figure 3D–E ) . There was also a strong correlation between the levels of Myc protein expressed by activated T cells and system L amino acid transport capacity ( Figure 3F ) and while system L transport increased substantially over the first 24 hr of T cell activation in MycWT T cells this did not occur in MyccKO T cells ( Figure 3—figure supplement 2 ) . Downstream of Slc7a5 amino acid uptake , Myc-deficient T cells also fail to increase expression of several key enzymes in metabolic pathways that utilise branch-chain amino acid ( Leucine , Isoleucine , Valine ) pathways and methionine ( Figure 3—figure supplement 3A–B ) . Collectively , these data show that Myc plays a critical role in regulating system L amino acid transport and amino acid metabolism in immune activated T cells . Could loss of amino acid transport be the mechanism for the loss of protein production in immune activated MyccKO T cells ? To assess this , we examined the impact of Slc7a5 deletion on immune activated T cell proteomes . Figure 3G–H shows that the dramatic increase in cell mass associated with normal T cell activation does not occur in immune activated Slc7a5cKO ( Cd4Cre+ Slc7a5fl/fl ) CD4+ T cells . We then used nearest neighbour analysis and Pearson correlation to group the expression profile of ~6800 proteins from naïve wild-type and immune activated Slc7a5WT and Slc7a5cKO CD4+ T cell proteomes . These data show Slc7a5 deficiency , similar to Myc deficiency , has a profound effect on protein expression in immune activated CD4+ T cells ( Figure 3I ) . Slc7a5cKO T cells still respond to antigen receptor activation to downregulate a subset of naïve T cell proteins and can still upregulate expression of a small subset of proteins ( Figure 3I ) . The data show a striking overlap in proteins that were both Myc and Slc7a5 regulated ( Figure 3J ) . Most of this overlap was in proteins that were reduced in response to Myc or Slc7a5 deficiency ( Figure 3K ) , including translational machinery such as ribosomes ( Figure 3—figure supplement 4A–B ) . Although there is a large degree of overlap in the proteomics data , Slc7a5-deficiency does not completely phenocopy the effects of Myc-deficiency . Induction of proteins such as the glucose transporter Slc2a3 ( Figure 2B , Figure 3—figure supplement 4C ) and effector molecules like Granzyme B and IFNγ ( Figure 3—figure supplement 4D–E ) exhibit a more severe defect in Slc7a5cKO T cells . This is likely due to the lack of basal-level amino acid transport in naïve Slc7a5cKO T cells which is not deficient in naïve MyccKO T cells ( Figure 3C ) . Overall , deficiency in a single Myc controlled amino acid transporter , Slc7a5 , largely does mimic the phenotype of MyccKO T cells , preventing T cell growth and selectively controlling proteome remodelling . To explore the mechanism for Myc control of amino acid transport in activated T cells we examined the relationship between Myc and amino acid transporter mRNA expression . Single cell RNAseq analysis of antigen activated OT1 CD8+ T cells ( Richard et al . , 2018 ) shows a strong correlation at the single cell level of Myc mRNA expression and expression of mRNA for Slc7a5 , Slc7a1 and Slc1a5 ( Figure 4A ) . Expression of Myc mRNA clearly precedes increased expression of mRNA for Slc7a5 and Slc1a5 ( Figure 4A ) . More importantly , in a proteomics time course of OT1 CD8+ T cell activation , expression of Myc protein precedes antigen induced increases in expression of most amino acid transporters ( Figure 4B ) . Proteomics data shows that expression of amino acid transporters increases gradually over time ( Figure 4B ) , and Kyn uptake experiments confirm that this increase in transporter number corresponds with higher system L uptake ( Figure 3—figure supplement 2 ) . CD3/CD28 activation of MycWT CD4+ and CD8+ T cells drives increases in Slc7a5 , Slc1a5 and Slc7a1 mRNA , whereas activated MyccKO CD4+ and CD8+ T cells do not increase expression of Slc7a5 or Slc1a5 mRNA and show reduced expression of Slc7a1 mRNA ( Figure 4C ) . The current data are consistent with a model that Myc controls T cell growth by controlling the upregulation of amino acid transporter expression required for T cell activation . However one possible inconsistency is that previous studies have shown that Slc7a5 is required for expression of Myc protein ( but not mRNA ) in activated CD8+ T cells ( Sinclair et al . , 2013 ) . We considered that an explanation for this discrepancy would be if there were a positive feedforward loop whereby the initial rapid expression of Myc during immune activation is not Slc7a5 dependent but the sustained expression is . To directly interrogate this model we measured Myc expression over time in CD3/CD28 activated WT and Slc7a5cKO T cells . These data ( Figure 4D ) show that Slc7a5 is not required for the immediate and rapid upregulation of Myc expression that accompanies T cell activation but is required for activated T cells to sustain Myc protein . The finding that Slc7a5cKO T cells can induce Myc expression is also surprising in the context of previous work demonstrating an important role for mTORC1 activation ( which is critically dependent on uptake of leucine ) in controlling Myc expression during T cell activation ( Wang et al . , 2011 ) . Therefore , we tested the dependency of Myc protein expression on mTORC1 signalling in our T cell system . The data show that rapamycin treatment had very little impact on expression of Myc protein after 4 hr of T cell activation but did reduce Myc expression after 24 hr ( Figure 4E ) , consistent with the results seen in Slc7a5cKO T cells ( Figure 4D ) . Although Slc7a5 deficiency or mTORC1 inhibition alone was insufficient to prevent Myc induction , the presence of external amino acids is necessary for expression of Myc protein , with T cells activated in amino acid-free media being unable to express Myc protein despite increasing the activation marker CD69 ( Figure 4F ) . Deficiency of a single amino acid from the media , such as glutamine , methionine or leucine leads to a reduction in Myc levels but does not prevent its expression ( Figure 4G ) . Myc driven increase in amino acid transport thus triggers a positive feedforward loop supporting its own continued expression which in turn drives and sustains further increases in amino acid transport . This study has mapped the impact of Myc deletion on antigen driven proteome remodelling of CD4+ and CD8+ T cells to understand how Myc controls T cell activation and metabolic reprogramming . The study uncovers both Myc dependent and independent restructuring of the T cell proteome during immune activation . Myc was required for the increase in expression of important metabolic pathway proteins; for example , glutamine transporters and glutaminase , key proteins controlling the first steps of glutaminolysis ( Newsholme et al . , 1985 ) ; and lactate transporters , a major rate determining step for glycolytic flux ( Tanner et al . , 2018 ) . However , the current data also show that expression of many key metabolic enzymes for both glutaminolysis and glycolysis can still occur in immune activated Myc null T cells . In particular , in the context of glucose metabolism an unexpected observation was that Myc was not required for protein expression of the glucose transporters Slc2a1 and Slc2a3 in activated T cells . This was surprising , given previous observations that Myc deletion reduced Slc2a1 and Slc2a3 mRNA ( Wang et al . , 2011 ) and highlights the value of a proteomics approach to quantify the expression patterns of proteins where there may be a disconnect between mRNA and protein expression due to translational regulation ( Ricciardi et al . , 2018 ) . The expression of both glucose and lactate transporters are key for glycolytic flux ( Tanner et al . , 2018 ) and the fact that these are differentially controlled by Myc reveals that upregulation of metabolic pathways during T cell activation is more complex than a simple activation switch or amplifier mediated by a single transcription factor ( Nie et al . , 2012 ) . The data gives molecular insight into why Myc is so important for T cell glutamine metabolism and glycolysis but they also reveal that T cell metabolic reprogramming requires the coordination of Myc expression with other signalling pathways . In this respect we have shown recently that activation of mTORc1 is not required for Myc expression in activated T cell but does have a substantive effect on the expression of glucose transporter protein ( Howden et al . , 2019 ) . One key conclusion from the present data is that a primary function of Myc is to control expression of the amino acid transporters , inducing a positive feedforward loop to sustain Myc levels in activated T cells . A salient point is that Myc was only necessary for immune activation associated increases in amino acid transporter expression . The absence of Myc did not impinge on the low basal levels of amino acid transport through the system L transporter seen in naïve T cells and it was clear from the proteomic data that Myc null T cells still had some capacity to increase expression of key proteins . The inability of Myc null T cells to increase amino acid transport aligns with previous metabolomic data that Myc null cells have decreased levels of intracellular amino acids ( Wang et al . , 2011 ) . The loss of amino acid transporter induction in Myc null T cells would also prevent the increases in expression of the protein biosynthetic machinery as well as preventing uptake of the raw material required to synthesise protein . The importance of Myc induction of amino acid transporters for T cell activation is particularly highlighted by the large effect of deleting just one of the Myc controlled amino acid transporters , Slc7a5 , on the T cell proteome , which almost phenocopies the effects of Myc deletion itself . The impact of the loss of a single Myc controlled amino acid transporter was remarkable and reflects that Slc7a5 transports multiple large neutral amino acids including Leucine , Phenylalanine and Tryptophan . Myc control of Slc7a5 expression would be particularly important for protein synthesis as Slc7a5 is also the major T cell transporter for Methionine , the predominant ‘start’ amino acid used to initiate polypeptide synthesis during mRNA translation ( Sinclair et al . , 2019; Sinclair et al . , 2013 ) . These data highlight how Myc control of even one amino acid transporter , Slc7a5 , would have indirect consequences for the expression of thousands of proteins in immune activated T cells and could underpin the ability of Myc to regulate multiple biosynthetic , bioenergetic and epigenetic processes in T cells . Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Doreen Cantrell ( d . a . cantrell@dundee . ac . uk ) .	T cells are white blood cells that form an important part of our immune defence , acting to attack disease-causing microbes and cancer and directing other immune cells to help in this fight . T cells spend most of their time in a resting state , small and inactive , but when an infection strikes , they transform into large , active 'effector' cells . This change involves a dramatic increase in protein production , accompanied by high energy demands . To fully activate , T cells need to boost their metabolism and take in extra amino acids , the building blocks of proteins . For this , they depend upon a protein called Myc . The Myc protein works as a genetic switch , controlling several kinds of cell metabolism , but the molecular details of its effects in T cells remain unclear . Most studies looking to understand Myc have focussed on its role in cancer cells . Here its main job is thought to be driving the use of sugar to make energy . However , it has also been shown to control the levels of transporters that carry amino acids into cells and thus provide the raw materials for protein production . It is possible that Myc plays a similar role in T cells as it does in cancer cells , but this might not be the case because cancer cells have strange biology and do not always accurately represent healthy cells . To find out what role Myc plays in T cell activation , Marchingo et al . compared T cells with and without Myc . The cells lacking Myc were much smaller than their normal counterparts and counts of their proteins revealed why . Without Myc , protein production had stalled . In normal T cells , the number of amino acid transporters increased up to 100 times as cells transformed from a resting to an active state . But , without Myc , this did not happen . The loss of Myc cut off the supply of amino acids , halting protein production . For T cells , the most important amino acid transporter is a protein called System-L transporter Slc7a5 . It supplies several essential amino acids , including methionine – the amino acid that starts every single protein . To confirm the role of amino acid transporters in T cell activation , Marchingo et al . deleted the gene for the System-L transporter Slc7a5 directly . This had the same effect as deleting the gene for Myc itself , demonstrating that a key role of Myc in T cell activation is to increase the number of amino acid transporters . Understanding the role of Myc in T cell activation is an important step towards controlling the immune system . At the moment , many research groups are investigating how best to use T cells to fight diseases like cancer . Further analysis of the link between Myc and amino acid transporters could in the future aid the design of such immunotherapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2020	Quantitative analysis of how Myc controls T cell proteomes and metabolic pathways during T cell activation
Plants constantly renew during their life cycle and thus require to shed senescent and damaged organs . Floral abscission is controlled by the leucine-rich repeat receptor kinase ( LRR-RK ) HAESA and the peptide hormone IDA . It is unknown how expression of IDA in the abscission zone leads to HAESA activation . Here we show that IDA is sensed directly by the HAESA ectodomain . Crystal structures of HAESA in complex with IDA reveal a hormone binding pocket that accommodates an active dodecamer peptide . A central hydroxyproline residue anchors IDA to the receptor . The HAESA co-receptor SERK1 , a positive regulator of the floral abscission pathway , allows for high-affinity sensing of the peptide hormone by binding to an Arg-His-Asn motif in IDA . This sequence pattern is conserved among diverse plant peptides , suggesting that plant peptide hormone receptors may share a common ligand binding mode and activation mechanism . During their growth , development and reproduction plants use cell separation processes to detach no-longer required , damaged or senescent organs . Abscission of floral organs in Arabidopsis is a model system to study these cell separation processes in molecular detail ( Aalen et al . , 2013 ) . The LRR-RKs HAESA ( greek: to adhere to ) and HAESA-LIKE 2 ( HSL2 ) redundantly control floral abscission ( Cho et al . , 2008; Jinn et al . , 2000; Stenvik et al . , 2008 ) . Loss-of-function of the secreted small protein INFLORESCENCE DEFICIENT IN ABSCISSION ( IDA ) causes floral organs to remain attached while its over-expression leads to premature shedding ( Butenko et al . , 2003; Stenvik et al . , 2006 ) . Full-length IDA is proteolytically processed and a conserved stretch of 20 amino-acids ( termed EPIP ) can rescue the IDA loss-of-function phenotype ( Figure 1A ) ( Stenvik et al . , 2008 ) . It has been demonstrated that a dodecamer peptide within EPIP is able to activate HAESA and HSL2 in transient assays in tobacco cells ( Butenko et al . , 2014 ) . This sequence motif is highly conserved among IDA family members ( IDA-LIKE PROTEINS , IDLs ) and contains a central Pro residue , presumed to be post-translationally modified to hydroxyproline ( Hyp; Figure 1A ) ( Butenko et al . , 2003; 2014 ) . The available genetic and biochemical evidence suggests that IDA and HAESA together control floral abscission , but it is poorly understood if IDA is directly sensed by the receptor kinase HAESA and how IDA binding at the cell surface would activate the receptor . 10 . 7554/eLife . 15075 . 003Figure 1 . The peptide hormone IDA binds to the HAESA LRR ectodomain . ( A ) Multiple sequence alignment of selected IDA family members . The conserved PIP motif is highlighted in yellow , the central Hyp in blue . The PKGV motif present in our N-terminally extended IDA peptide is highlighted in red . ( B ) Isothermal titration calorimetry of the HAESA ectodomain vs . IDA and including the synthetic peptide sequence . ( C ) Structure of the HAESA – IDA complex with HAESA shown in blue ( ribbon diagram ) . IDA ( in bonds representation , surface view included ) is depicted in yellow . The peptide binding pocket covers HAESA LRRs 2–14 . ( D ) Close-up view of the entire IDA ( in yellow ) peptide binding site in HAESA ( in blue ) . Details of the interactions between the central Hyp anchor in IDA and the C-terminal Arg-His-Asn motif with HAESA are highlighted in ( E ) and ( F ) , respectively . Hydrogren bonds are depicted as dotted lines ( in magenta ) , a water molecule is shown as a red sphere . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 00310 . 7554/eLife . 15075 . 004Figure 1—figure supplement 1 . The HAESA ectodomain folds into a superhelical assembly of 21 leucine-rich repeats . ( A ) SDS PAGE analysis of the purified Arabidopsis thaliana HAESA ectodomain ( residues 20–620 ) obtained by secreted expression in insect cells . The calculated molecular mass is 65 . 7 kDa , the actual molecular mass obtained by mass spectrometry is 74 , 896 Da , accounting for the N-glycans . ( B ) Ribbon diagrams showing front ( left panel ) and side views ( right panel ) of the isolated HAESA LRR domain . The N- ( residues 20–88 ) and C-terminal ( residues 593–615 ) capping domains are shown in yellow , the central 21 LRR motifs are in blue and disulphide bonds are highlighted in green ( in bonds representation ) . ( C ) Structure based sequence alignment of the 21 leucine-rich repeats in HAESA with the plant LRR consensus sequence shown for comparison . Conserved hydrophobic residues are shaded in gray , N-glycosylation sites visible in our structures are highlighted in blue , cysteine residues involved in disulphide bridge formation in green . ( D ) Asn-linked glycans mask the N-terminal portion of the HAESA ectodomain . Oligomannose core structures ( containing two N-actylglucosamines and three terminal mannose units ) as found in Trichoplusia ni cells and in plants were modeled onto the seven glycosylation sites observed in our HAESA structures , to visualize the surface areas potentially not masked by carbohydrate . The HAESA ectodomain is shown in blue ( in surface representation ) , the glycan structures are shown in yellow . Molecular surfaces were calculated with the program MSMS ( Sanner et al . , 1996 ) , with a probe radius of 1 . 5 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 00410 . 7554/eLife . 15075 . 005Figure 1—figure supplement 2 . Hydrophobic contacts and a hydrogen-bond network mediate the interaction between HAESA and the peptide hormone IDA . ( A ) Details of the IDA binding pocket . HAESA is shown in blue ( ribbon diagram ) , the C-terminal Arg-His-Asn motif ( left panel ) , the central Hyp anchor ( center ) and the N-terminal Pro-rich motif in IDA ( right panel ) are shown in yellow ( in bonds representation ) . HAESA interface residues are shown as sticks , selected hydrogen bond interactions are denoted as dotted lines ( in magenta ) . ( B ) View of the complete IDA ( in bonds representation , in yellow ) binding pocket in HAESA ( surface view , in blue ) . Orientation as in ( A ) . ( C ) Structure based sequence alignment of leucine-rich repeats in HAESA with the plant LRR consensus sequence shown for comparison . Residues mediating hydrophobic interactions with the IDA peptide are highlighted in blue , residues contributing to hydrogen bond interactions and/or salt bridges are shown in red . The IDA binding pocket covers LRRs 2–14 and all residues originate from the inner surface of the HAESA superhelix . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 00510 . 7554/eLife . 15075 . 006Figure 1—figure supplement 3 . The IDA-HAESA and SERK1-HAESA complex interfaces are conserved among HAESA and HAESA-like proteins from different plant species . Structure-based sequence alignment of the HAESA family members: Arabidopsis thaliana HAESA ( Uniprot ( http://www . uniprot . org ) ID P47735 ) , Arabidopsis thaliana HSL2 ( Uniprot ID C0LGX3 ) , Capsella rubella HAESA ( Uniprot ID R0F2U6 ) , Citrus clementina HSL2 ( Uniprot ID V4U227 ) , Vitis vinifera HAESA ( Uniprot ID F6HM39 ) . The alignment includes a secondary structure assignment calculated with the program DSSP ( Kabsch and Sander , 1983 ) and colored according to Figure 1 , with the N- and C-terminal caps and the 21 LRR motifs indicated in orange and blue , respectively . Cysteine residues engaged in disulphide bonds are depicted in green . HAESA residues interacting with the IDA peptide and/or the SERK1 co-receptor kinase ectodomain are highlighted in blue and orange , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 006 We purified the HAESA ectodomain ( residues 20–620 ) from baculovirus-infected insect cells ( Figure 1—figure supplement 1A , see Materials and methods ) and quantified the interaction of the ~75 kDa glycoprotein with synthetic IDA peptides using isothermal titration calorimetry ( ITC ) . A Hyp-modified dodecamer comprising the highly conserved PIP motif in IDA ( Figure 1A ) interacts with HAESA with 1:1 stoichiometry ( N ) and with a dissociation constant ( Kd ) of ~20 μM ( Figure 1B ) . We next determined crystal structures of the apo HAESA ectodomain and of a HAESA-IDA complex , at 1 . 74 and 1 . 86 Å resolution , respectively ( Figure 1C; Figure 1—figure supplement 1B–D; Tables 1 , 2 ) . IDA binds in a completely extended conformation along the inner surface of the HAESA ectodomain , covering LRRs 2–14 ( Figure 1C , D , Figure 1—figure supplement 2 ) . The central Hyp64IDA is buried in a specific pocket formed by HAESA LRRs 8–10 , with its hydroxyl group establishing hydrogen bonds with the strictly conserved Glu266HAESA and with a water molecule , which in turn is coordinated by the main chain oxygens of Phe289HAESA and Ser311HAESA ( Figure 1E; Figure 1—figure supplement 3 ) . The restricted size of the Hyp pocket suggests that IDA does not require arabinosylation of Hyp64IDA for activity in vivo , a modification that has been reported for Hyp residues in plant CLE peptide hormones ( Ohyama et al . , 2009 ) . The C-terminal Arg-His-Asn motif in IDA maps to a cavity formed by HAESA LRRs 11–14 ( Figure 1D , F ) . The COO- group of Asn69IDA is in direct contact with Arg407HAESA and Arg409HAESA and HAESA cannot bind a C-terminally extended IDA-SFVN peptide ( Figures 1D , F , 2D ) . This suggests that the conserved Asn69IDA may constitute the very C-terminus of the mature IDA peptide in planta and that active IDA is generated by proteolytic processing from a longer pre-protein ( Stenvik et al . , 2008 ) . Mutation of Arg417HSL2 ( which corresponds to Arg409HAESA ) causes a loss-of-function phenotype in HSL2 , which indicates that the peptide binding pockets in different HAESA receptors have common structural and sequence features ( Niederhuth et al . , 2013 ) . Indeed , we find many of the residues contributing to the formation of the IDA binding surface in HAESA to be conserved in HSL2 and in other HAESA-type receptors in different plant species ( Figure 1—figure supplement 3 ) . A N-terminal Pro-rich motif in IDA makes contacts with LRRs 2–6 of the receptor ( Figure 1D , Figure 1—figure supplement 2A–C ) . Other hydrophobic and polar interactions are mediated by Ser62IDA , Ser65IDA and by backbone atoms along the IDA peptide ( Figure 1D , Figure 1—figure supplement 2A–C ) . 10 . 7554/eLife . 15075 . 007Figure 2 . Active IDA-family peptide hormones are hydroxyprolinated dodecamers . Close-up views of ( A ) IDA , ( B ) the N-terminally extended PKGV-IDA and ( C ) IDL1 bound to the HAESA hormone binding pocket ( in bonds representation , in yellow ) and including simulated annealing 2Fo–Fc omit electron density maps contoured at 1 . 0 σ . Note that Pro58IDA and Leu67IDA are the first residues defined by electron density when bound to the HAESA ectodomain . ( D ) Table summaries for equilibrium dissociation constants ( Kd ) , binding enthalpies ( ΔH ) , binding entropies ( ΔS ) and stoichoimetries ( N ) for different IDA peptides binding to the HAESA ectodomain ( ± fitting errors; n . d . no detectable binding ) . ( E ) Structural superposition of the active IDA ( in bonds representation , in gray ) and IDL1 peptide ( in yellow ) hormones bound to the HAESA ectodomain . Root mean square deviation ( r . m . s . d . ) is 1 . 0 Å comparing 100 corresponding atoms . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 00710 . 7554/eLife . 15075 . 008Figure 3 . The receptor kinase SERK1 acts as a HAESA co-receptor and promotes high-affinity IDA sensing . ( A ) Petal break-strength assays measure the force ( expressed in gram equivalents ) required to remove the petals from the flower of serk mutant plants compared to haesa/hsl2 mutant and Col-0 wild-type flowers . Petal break-strength is measured from positions 1 to 8 along the primary inflorescence where positions 1 is defined as the flower at anthesis ( n=15 , bars=SD ) . This treatment-by-position balanced two-way layout was analyzed separately per position , because of the serious interaction , by means of a Dunnett-type comparison against the Col-0 control , allowing for heterogeneous variances ( Hasler and Hothorn , 2008 ) . Petal break-strength was found significantly increased in almost all positions ( indicated with a * ) for haesa/hsl2 and serk1-1 mutant plants with respect to the Col-0 control . Calculations were performed in R ( R Core Team , 2014 ) ( version 3 . 2 . 3 ) . ( B ) Analytical size-exclusion chromatography . The HAESA LRR domain elutes as a monomer ( black dotted line ) , as does the isolated SERK1 ectodomain ( blue dotted line ) . A HAESA – IDA – SERK1 complex elutes as an apparent heterodimer ( red line ) , while a mixture of HAESA and SERK1 yields two isolated peaks that correspond to monomeric HAESA and SERK1 , respectively ( black line ) . Void ( V0 ) volume and total volume ( Vt ) are shown , together with elution volumes for molecular mass standards ( A , Thyroglobulin , 669 , 000 Da; B , Ferritin , 440 , 00 Da , C , Aldolase , 158 , 000 Da; D , Conalbumin , 75 , 000 Da; E , Ovalbumin , 44 , 000 Da; F , Carbonic anhydrase , 29 , 000 Da ) . A SDS PAGE of the peak fractions is shown alongside . Purified HAESA and SERK1 are ~75 and ~28 kDa , respectively . ( C ) Isothermal titration calorimetry of wild-type and Hyp64→Pro IDA versus the HAESA and SERK1 ectodomains . The titration of IDA wild-type versus the isolated HAESA ectodomain from Figure 1B is shown for comparison ( red line; n . d . no detectable binding ) ( D ) Analytical size-exclusion chromatography in the presence of the IDA Hyp64→Pro mutant peptide reveals no complex formation between HAESA and SERK1 ectodomains . A SDS PAGE of the peak fractions is shown alongside . ( E ) In vitro kinase assays of the HAESA and SERK1 kinase domains . Wild-type HAESA and SERK1 kinase domains ( KDs ) exhibit auto-phosphorylation activities ( lanes 1 + 3 ) . Mutant ( m ) versions , which carry point mutations in their active sites ( Asp837HAESA→Asn , Asp447SERK1→Asn ) possess no autophosphorylation activity ( lanes 2+4 ) . Transphosphorylation activity from the active kinase to the mutated form can be observed in both directions ( lanes 5+6 ) . A coomassie-stained gel loading control is shown below . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 00810 . 7554/eLife . 15075 . 009Table 1 . Crystallographic data collection , phasing and refinement statistics for the isolated A . thaliana HAESA ectodomain . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 009HAESA NaI shortsoakHAESA apoPDB-ID5IXOData collectionSpace groupP31 21P31 21Cell dimensionsa , b , c ( Å ) 148 . 55 , 148 . 55 , 58 . 30149 . 87 , 149 . 87 , 58 . 48α , β , γ ( ° ) 90 , 90 , 12090 , 90 , 120Resolution ( Å ) 48 . 63–2 . 39 ( 2 . 45–2 . 39 ) 45 . 75–1 . 74 ( 1 . 85–1 . 74 ) Rmeas#0 . 096 ( 0 . 866 ) 0 . 038 ( 1 . 02 ) CC ( 1/2 ) #100/86 . 6100/75 . 6I/σ I#27 . 9 ( 4 . 9 ) 18 . 7 ( 1 . 8 ) Completeness ( % ) #99 . 9 ( 98 . 6 ) 99 . 6 ( 97 . 4 ) Redundancy#53 . 1 ( 29 . 9 ) 14 . 4 ( 14 . 0 ) Wilson B-factor ( Å2 ) #84 . 4581 . 10RefinementResolution ( Å ) 45 . 75 – 1 . 74No . reflections71 , 213Rwork/Rfree$0 . 188/0 . 218No . atomsProtein/glycan4 , 533/126Water71Res . B-factors ( Å2 ) $Protein77 . 54Glycan95 . 98Water73 . 20R . m . s deviations$Bond lengths ( Å ) 0 . 0095Bond angles ( ° ) 1 . 51Highest resolution shell is shown in parenthesis . #As defined in XDS ( Kabsch , 1993 ) $As defined in Refmac5 ( Murshudov et al . , 1997 ) 10 . 7554/eLife . 15075 . 010Table 2 . Crystallographic data collection and refinement statistics for the HAESA – IDA , – PKGV-IDA , – IDL1 and – IDA – SERK1 complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 010HAESA – IDAHAESA – PKGV-IDAHAESA – IDL1HAESA – IDA – SERK1PDB-ID5IXQ5IXT5IYN5IYXData collectionSpace groupP31 21P31 21P31 21P212121Cell dimensions a , b , c ( Å ) 148 . 55 , 148 . 55 , 58 . 30148 . 92 , 148 . 92 , 58 . 02150 . 18 , 150 . 18 , 60 . 0774 . 51 , 100 . 46 , 142 . 76α , β , γ ( ° ) 90 , 90 , 12090 , 90 , 12090 , 90 , 12090 , 90 , 90Resolution ( Å ) 48 . 54–1 . 86 ( 1 . 97–1 . 86 ) 48 . 75–1 . 94 ( 2 , 06–1 . 94 ) 49 . 16–2 . 56 ( 2 . 72–2 . 56 ) 47 . 59–2 . 43 ( 2 . 57–2 . 43 ) Rmeas#0 . 057 ( 1 . 35 ) 0 . 037 ( 0 . 97 ) 0 . 056 ( 1 . 27 ) 0 . 113 ( 1 . 37 ) CC ( 1/2 ) #100/77 . 9100/80 . 3100/89 . 5100/77 . 6I/σI#16 . 7 ( 2 . 0 ) 20 . 9 ( 2 . 4 ) 26 . 0 ( 1 . 9 ) 16 . 12 ( 2 . 0 ) Completeness# ( % ) 99 . 8 ( 98 . 6 ) 99 . 4 ( 97 . 9 ) 99 . 5 ( 98 . 8 ) 99 . 4 ( 96 . 4sRedundancy#20 . 3 ( 19 . 1 ) 11 . 2 ( 11 . 1 ) 14 . 7 ( 14 . 7 ) 9 . 7 ( 9 . 3 ) Wilson B-factor ( Å2 ) #80 . 081 . 789 . 559 . 3RefinementResolution ( Å ) 48 . 54–1 . 8648 . 75–1 . 9449 . 16–2 . 5647 . 59–2 . 43No . reflections58 , 55151 , 55723 , 83538 , 969Rwork/Rfree$0 . 190/0 . 2090 . 183/0 . 2080 . 199/0 . 2360 . 199/0 . 235No . atomsProtein/Glycan4 , 541/1764 , 545/1764 , 499/1765 , 965/168Peptide939390112Water39409136Res . B-factors ( Å2 ) $Protein/Glycan79 . 48/109 . 0279 . 63/113 . 24102 . 12/132 . 4960 . 05/73 . 48Peptide87 . 1989 . 50125 . 7451 . 06Water75 . 3271 . 9274 . 6551 . 47R . m . s deviations$Bond lengths ( Å ) 0 . 00870 . 00910 . 00810 . 0074Bond angles ( ° ) 1 . 481 . 471 . 361 . 34Highest resolution shell is shown in parenthesis . #As defined in XDS ( Kabsch , 1993 ) $As defined in Refmac5 ( Murshudov et al . , 1997 ) We next investigated whether HAESA binds N-terminally extended versions of IDA . We obtained a structure of HAESA in complex with a PKGV-IDA peptide at 1 . 94 Å resolution ( Table 2 ) . In this structure , no additional electron density accounts for the PKGV motif at the IDA N-terminus ( Figure 2A , B ) . Consistently , PKGV-IDA and IDA have similar binding affinities in our ITC assays , further indicating that HAESA senses a dodecamer peptide comprising residues 58-69IDA ( Figure 2D ) . We next tested if HAESA binds other IDA peptide family members . IDL1 , which can rescue IDA loss-of-function mutants when introduced in abscission zone cells , can also be sensed by HAESA , albeit with lower affinity ( Figure 2D ) ( Stenvik et al . , 2008 ) . A 2 . 56 Å co-crystal structure with IDL1 reveals that different IDA family members use a common binding mode to interact with HAESA-type receptors ( Stenvik et al . , 2008; Butenko et al . , 2009 ) ( Figure 2A–C , E , Table 2 ) . We do not detect interaction between HAESA and a synthetic peptide missing the C-terminal Asn69IDA ( ΔN69 ) , highlighting the importance of the polar interactions between the IDA carboxy-terminus and Arg407HAESA/Arg409HAESA ( Figures 1F , 2D ) . Replacing Hyp64IDA , which is common to all IDLs , with proline impairs the interaction with the receptor , as does the Lys66IDA/Arg67IDA → Ala double-mutant discussed below ( Figure 1A , 2D ) . Notably , HAESA can discriminate between IDLs and functionally unrelated dodecamer peptides with Hyp modifications , such as CLV3 ( Figures 2D , 7 ) ( Ogawa et al . , 2008 ) . Our binding assays reveal that IDA family peptides are sensed by the isolated HAESA ectodomain with relatively weak binding affinities ( Figures 1B , 2A–D ) . It has been recently reported that SOMATIC EMBRYOGENESIS RECEPTOR KINASES ( SERKs ) are positive regulators of floral abscission and can interact with HAESA and HSL2 in an IDA-dependent manner ( Meng et al . , 2016 ) . As all five SERK family members appear to be expressed in the Arabidopsis abscission zone ( Niederhuth et al . , 2013 ) , we quantified their relative contribution to floral abscission in Arabidopsis using a petal break-strength assay ( Lease et al . , 2006 ) . Our experiments suggest that among the SERK family members , SERK1 is a positive regulator of floral abscission . We found that the force required to remove the petals of serk1-1 mutants is significantly higher than that needed for wild-type plants , as previously observed for haesa/hsl2 mutants ( Stenvik et al . , 2008 ) , and that floral abscission is delayed in serk1-1 ( Figure 3A ) . The serk2-2 , serk3-1 , serk4-1 and serk5-1 mutant lines ( Albrecht et al . , 2008 ) showed a petal break-strength profile not significantly different from wild-type plants . Possibly because SERKs have additional roles in plant development such as in pollen formation ( Albrecht et al . , 2005; Colcombet et al . , 2005 ) and brassinosteroid signaling ( Gou et al . , 2012 ) , we found that higher-order SERK mutants exhibit pleiotropic phenotypes in the flower ( Meng et al . , 2015 ) , rendering their analysis and comparison by quantitative petal break-strength assays difficult . We thus focused on analyzing the contribution of SERK1 to HAESA ligand sensing and receptor activation . In vitro , the LRR ectodomain of SERK1 ( residues 24–213 ) forms stable , IDA-dependent heterodimeric complexes with HAESA in size exclusion chromatography experiments ( Figure 3B ) . We next quantified the contribution of SERK1 to IDA recognition by HAESA . We found that HAESA senses IDA with a ~60 fold higher binding affinity in the presence of SERK1 , suggesting that SERK1 is involved in the specific recognition of the peptide hormone ( Figure 3C ) . We next titrated SERK1 into a solution containing only the HAESA ectodomain . In this case , there was no detectable interaction between receptor and co-receptor , while in the presence of IDA , SERK1 strongly binds HAESA with a dissociation constant in the mid-nanomolar range ( Figure 3C ) . This suggests that IDA itself promotes receptor – co-receptor association , as previously described for the steroid hormone brassinolide ( Santiago et al . , 2013 ) and for other LRR-RK complexes ( Sun et al . , 2013; Wang et al . , 2015 ) . Importantly , hydroxyprolination of IDA is critical for HAESA-IDA-SERK1 complex formation ( Figure 3C , D ) . Our calorimetry experiments now reveal that SERKs may render HAESA , and potentially other receptor kinases , competent for high-affinity sensing of their cognate ligands . Upon IDA binding at the cell surface , the kinase domains of HAESA and SERK1 , which have been shown to be active protein kinases ( Jinn et al . , 2000; Shah et al . , 2001; Taylor et al . , 2016 ) , may interact in the cytoplasm to activate each other . Consistently , the HAESA kinase domain can transphosphorylate SERK1 and vice versa in in vitro transphosphorylation assays ( Figure 3E ) . Together , our genetic and biochemical experiments implicate SERK1 as a HAESA co-receptor in the Arabidopsis abscission zone . To understand in molecular terms how SERK1 contributes to high-affinity IDA recognition , we solved a 2 . 43 Å crystal structure of the ternary HAESA – IDA – SERK1 complex ( Figure 4A , Table 2 ) . HAESA LRRs 16–21 and its C-terminal capping domain undergo a conformational change upon SERK1 binding ( Figure 4B ) . The SERK1 ectodomain interacts with the IDA peptide binding site using a loop region ( residues 51-59SERK1 ) from its N-terminal cap ( Figure 4A , C ) . SERK1 loop residues establish multiple hydrophobic and polar contacts with Lys66IDA and the C-terminal Arg-His-Asn motif in IDA ( Figure 4C ) . SERK1 LRRs 1–5 and its C-terminal capping domain form an additional zipper-like interface with residues originating from HAESA LRRs 15–21 and from the HAESA C-terminal cap ( Figure 4D ) . SERK1 binds HAESA using these two distinct interaction surfaces ( Figure 1—figure supplement 3 ) , with the N-cap of the SERK1 LRR domain partially covering the IDA peptide binding cleft . 10 . 7554/eLife . 15075 . 011Figure 4 . Crystal structure of a HAESA – IDA – SERK1 signaling complex . ( A ) Overview of the ternary complex with HAESA in blue ( surface representation ) , IDA in yellow ( bonds representation ) and SERK1 in orange ( surface view ) . ( B ) The HAESA ectodomain undergoes a conformational change upon SERK1 co-receptor binding . Shown are Cα traces of a structural superposition of the unbound ( yellow ) and SERK1-bound ( blue ) HAESA ectodomains ( r . m . s . d . is 1 . 5 Å between 572 corresponding Cα atoms ) . SERK1 ( in orange ) and IDA ( in red ) are shown alongside . The conformational change in the C-terminal LRRs and capping domain is indicated by an arrow . ( C ) SERK1 forms an integral part of the receptor's peptide binding pocket . The N-terminal capping domain of SERK1 ( in orange ) directly contacts the C-terminal part of IDA ( in yellow , in bonds representation ) and the receptor HAESA ( in blue ) . Polar contacts of SERK1 with IDA are shown in magenta , with the HAESA LRR domain in gray . ( D ) Details of the zipper-like SERK1-HAESA interface . Ribbon diagrams of HAESA ( in blue ) and SERK1 ( in orange ) are shown with selected interface residues ( in bonds representation ) . Polar interactions are highlighted as dotted lines ( in magenta ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 011 The four C-terminal residues in IDA ( Lys66IDA-Asn69IDA ) are conserved among IDA family members and are in direct contact with SERK1 ( Figures 1A , 4C ) . We thus assessed their contribution to HAESA – SERK1 complex formation . Deletion of the buried Asn69IDA completely inhibits receptor – co-receptor complex formation and HSL2 activation ( Figure 5A , B ) ( Butenko et al . , 2014 ) . A synthetic Lys66IDA/Arg67IDA → Ala mutant peptide ( IDA K66A/R66A ) showed a 10 fold reduced binding affinity when titrated in a HAESA/SERK1 protein solution ( Figures 5A , B , 2D ) . We over-expressed full-length wild-type IDA or this Lys66IDA/Arg67IDA → Ala double-mutant to similar levels in Col-0 Arabidopsis plants ( Figure 5D ) . We found that over-expression of wild-type IDA leads to early floral abscission and an enlargement of the abscission zone ( Figure 5C–E ) . In contrast , over-expression of the IDA Lys66IDA/Arg67IDA → Ala double mutant significantly delays floral abscission when compared to wild-type control plants , suggesting that the mutant IDA peptide has reduced activity in planta ( Figure 5C–E ) . Comparison of 35S::IDA wild-type and mutant plants further indicates that mutation of Lys66IDA/Arg67IDA → Ala may cause a weak dominant negative effect ( Figure 5C–E ) . In agreement with our structures and biochemical assays , this experiment suggests a role of the conserved IDA C-terminus in the control of floral abscission . 10 . 7554/eLife . 15075 . 012Figure 5 . The IDA C-terminal motif is required for HAESA-SERK1 complex formation and for IDA bioactivity . ( A ) Size exclusion chromatography experiments similar to Figure 3B , D reveal that IDA mutant peptides targeting the C-terminal motif do not form biochemically stable HAESA-IDA-SERK1 complexes . Deletion of the C-terminal Asn69IDA completely inhibits complex formation . Void ( V0 ) volume and total volume ( Vt ) are shown , together with elution volumes for molecular mass standards ( A , Thyroglobulin , 669 , 000 Da; B , Ferritin , 440 , 00 Da , C , Aldolase , 158 , 000 Da; D , Conalbumin , 75 , 000 Da; E , Ovalbumin , 44 , 000 Da; F , Carbonic anhydrase , 29 , 000 Da ) . Purified HAESA and SERK1 are ~75 and ~28 kDa , respectively . Left panel: IDA K66A/R67A; center: IDA ΔN69 , right panel: SDS-PAGE of peak fractions . Note that the HAESA and SERK1 input lanes have already been shown in Figure 3D . ( B ) Isothermal titration thermographs of wild-type and mutant IDA peptides titrated into a HAESA - SERK1 mixture in the cell . Table summaries for calorimetric binding constants and stoichoimetries for different IDA peptides binding to the HAESA – SERK1 ectodomain mixture ( ± fitting errors; n . d . no detectable binding ) are shown alongside . ( C ) Quantitative petal break-strength assay for Col-0 wild-type flowers and 35S::IDA wild-type and 35S::IDA K66A/R67A mutant flowers . Petal break is measured from positions 1 to 8 along the primary inflorescence where positions 1 is defined as the flower at anthesis ( n=15 , bars=SD ) . The three treatment groups in this unbalanced one-way layout were compared by Tukey’s all-pairs comparison procedure using the package multcomp ( Hothorn et al . , 2008 ) in R ( R Core Team , 2014 ) ( version 3 . 2 . 3 ) . 35S::IDA plants showed significantly increased abscission compared to Col-0 controls in inflorescence positions 2 and 3 ( a ) . Up to inflorescence position 4 , petal break in 35S::IDA K66A/R67A mutant plants was significantly increased compared to both Col-0 control plants ( b ) and 35S::IDA plants ( c ) ( D ) Normalized expression levels ( relative expression ± standard error; ida: -0 . 02 ± 0 . 001; Col-0: 1 ± 0 . 11; 35S::IDA 124 ± 0 . 75; 35S::IDA K66A/R67A: 159 ± 0 . 58 ) of IDA wild-type and mutant transcripts in the 35S promoter over-expression lines analyzed in ( C ) . ( E ) Magnified view of representative abscission zones from 35S::IDA , Col-0 wild-type and 35S::IDA K66A/R67A double-mutant T3 transgenic lines . 15 out of 15 35S::IDA plants , 0 out of 15 Col-0 plants and 0 out of 15 35S::IDA K66A/R67A double-mutant plants , showed an enlarged abscission zone , respectively ( 3 independent lines were analyzed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 012 In contrast to animal LRR receptors , plant LRR-RKs harbor spiral-shaped ectodomains and thus they require shape-complementary co-receptor proteins for receptor activation ( Hothorn et al . , 2011 ) . For a rapidly growing number of plant signaling pathways , SERK proteins act as these essential co-receptors ( Wang et al . , 2015; Meng et al . , 2015; Meng et al . , 2016; Brandt and Hothorn , 2016 ) . SERK1 has been previously reported as a positive regulator in plant embryogenesis ( Hecht et al . , 2001; Salaj et al . , 2008 ) , male sporogenesis ( Albrecht et al . , 2005; Colcombet et al . , 2005 ) , brassinosteroid signaling ( Albrecht et al . , 2008; Gou et al . , 2012; Santiago et al . , 2013 ) and in phytosulfokine perception ( Wang et al . , 2015 ) . Recent findings by Meng et al . , 2016 and our mechanistic studies now also support a positive role for SERK1 in floral abscission . As serk1-1 mutant plants show intermediate abscission phenotypes when compared to haesa/hsl2 mutants , SERK1 likely acts redundantly with other SERKs in the abscission zone ( Figure 3A ) . It has been previously suggested that SERK1 can inhibit cell separation ( Lewis et al . , 2010 ) . However our results show that SERK1 also can activate this process upon IDA sensing , indicating that SERKs may fulfill several different functions in the course of the abscission process . While the sequence of the mature IDA peptide has not been experimentally determined in planta ( Stenvik et al . , 2008 ) , our HAESA-IDA complex structures and calorimetry assays suggest that active IDLs are hydroxyprolinated dodecamers . It will be thus interesting to see if proteolytic processing of full-length IDA in vivo is regulated in a cell-type or tissue-specific manner . The central Hyp residue in IDA is found buried in the HAESA peptide binding surface and thus this post-translational modification may regulate IDA bioactivity . Our comparative structural and biochemical analysis further suggests that IDLs share a common receptor binding mode , but may preferably bind to HAESA , HSL1 or HSL2 in different plant tissues and organs . In our quantitative biochemical assays , the presence of SERK1 dramatically increases the HAESA binding specificity and affinity for IDA . This observation is consistent with our complex structure in which receptor and co-receptor together form the IDA binding pocket . The fact that SERK1 specifically interacts with the very C-terminus of IDLs may allow for the rational design of peptide hormone antagonists , as previously demonstrated for the brassinosteroid pathway ( Muto and Todoroki , 2013; Santiago et al . , 2013 ) . Importantly , our calorimetry assays reveal that the SERK1 ectodomain binds HAESA with nanomolar affinity , but only in the presence of IDA ( Figure 3C ) . This ligand-induced formation of a receptor – co-receptor complex may allow the HAESA and SERK1 kinase domains to efficiently trans-phosphorylate and activate each other in the cytoplasm . It is of note that our reported binding affinities for IDA and SERK1 have been measured using synthetic peptides and the isolated HAESA and SERK1 ectodomains , and thus might differ in the context of the full-length , membrane-embedded signaling complex . Comparison of our HAESA – IDA – SERK1 structure with the brassinosteroid receptor signaling complex , where SERK1 also acts as co-receptor ( Santiago et al . , 2013 ) , reveals an overall conserved mode of SERK1 binding , while the ligand binding pockets map to very different areas in the corresponding receptors ( LRRs 2 – 14; HAESA; LRRs 21 – 25 , BRI1 ) and may involve an island domain ( BRI1 ) or not ( HAESA ) ( Figure 6A ) . Several residues in the SERK1 N-terminal capping domain ( Thr59SERK1 , Phe61SERK1 ) and the LRR inner surface ( Asp75SERK1 , Tyr101SERK1 , SER121SERK1 , Phe145SERK1 ) contribute to the formation of both complexes ( Figures 4C , D , 6B ) ( Santiago et al . , 2013 ) . In addition , residues 53-55SERK1 from the SERK1 N-terminal cap mediate specific interactions with the IDA peptide ( Figures 4C , 6B ) . These residues are not involved in the sensing of the steroid hormone brassinolide ( Santiago et al . , 2013 ) . In both cases however , the co-receptor completes the hormone binding pocket . This fact together with the largely overlapping SERK1 binding surfaces in HAESA and BRI1 allows us to speculate that SERK1 may promote high-affinity peptide hormone and brassinosteroid sensing by simply slowing down dissociation of the ligand from its cognate receptor . 10 . 7554/eLife . 15075 . 013Figure 6 . SERK1 uses partially overlapping surface areas to activate different plant signaling receptors . ( A ) Structural comparison of plant steroid and peptide hormone membrane signaling complexes . Left panel: Ribbon diagram of HAESA ( in blue ) , SERK1 ( in orange ) and IDA ( in bonds and surface represention ) . Right panel: Ribbon diagram of the plant steroid receptor BRI1 ( in blue ) bound to brassinolide ( in gray , in bonds representation ) and to SERK1 , shown in the same orientation ( PDB-ID . 4lsx ) ( Santiago et al . , 2013 ) . ( B ) View of the inner surface of the SERK1 LRR domain ( PDB-ID 4lsc ( Santiago et al . , 2013 ) , surface representation , in gray ) . A ribbon diagram of SERK1 in the same orientation is shown alongside . Residues interacting with the HAESA or BRI1 LRR domains are shown in orange or magenta , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 013 Our experiments reveal that SERK1 recognizes a C-terminal Arg-His-Asn motif in IDA . Importantly , this motif can also be found in other peptide hormone families ( Kondo et al . , 2006; Matsuzaki et al . , 2010; Tang et al . , 2015 ) ( Figure 7 ) . Among these are the CLE peptides regulating stem cell maintenance in the shoot and the root ( Clark et al . , 1995 ) . It is interesting to note , that CLEs in their mature form are also hydroxyprolinated dodecamers , which bind to a surface area in the BARELY ANY MERISTEM 1 receptor that would correspond to part of the IDA binding cleft in HAESA ( Kondo et al . , 2006; Ogawa et al . , 2008; Shinohara et al . , 2012 ) . Diverse plant peptide hormones may thus also bind their LRR-RK receptors in an extended conformation along the inner surface of the LRR domain and may also use small , shape-complementary co-receptors for high-affinity ligand binding and receptor activation . 10 . 7554/eLife . 15075 . 014Figure 7 . Different plant peptide hormone families contain a C-terminal ( Arg ) -His-Asn motif , which in IDA represents the co-receptor recognition site . Structure-guided multiple sequence alignment of IDA and IDA-like peptides with other plant peptide hormone families , including CLAVATA3 – EMBRYO SURROUNDING REGION-RELATED ( CLV3/CLE ) , ROOT GROWTH FACTOR – GOLVEN ( RGF/GLV ) , PRECURSOR GENE PROPEP1 ( PEP1 ) from Arabidopsis thaliana . The conserved ( Arg ) -His-Asn motif is highlighted in red , the central Hyp residue in IDLs and CLEs is marked in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 15075 . 014 Synthetic genes coding for the Arabidopsis thaliana HAESA ( residues 20–620 ) and SERK1 ectodomains ( residues 24–213 , carrying Asn115→Asp and Asn163→Gln mutations ) , codon optimized for expression in Trichoplusia ni ( Geneart , Germany ) , were cloned into a modified pBAC-6 transfer vector ( Novagen , Billerica , MA ) , providing an azurocidin signal peptide and a C-terminal TEV ( tobacco etch virus protease ) cleavable Strep-9xHis tandem affinity tag . Recombinant baculoviruses were generated by co-transfecting transfer vectors with linearised baculovirus DNA ( ProFold-ER1 , AB vector , San Diego , CA ) followed by viral amplification in Spodoptera frugiperda Sf9 cells . The HAESA and SERK1 ectodomains were individually expressed in Trichoplusia ni Tnao38 cells ( Hashimoto et al . , 2010 ) using a multiplicity of infection of 3 , and harvested from the medium 2 days post infection by tangential flow filtration using 30 kDa MWCO and 10 kDa MWCO ( molecular weight cut-off ) filter membranes ( GE Healthcare Life Sciences , Pittsburgh , PA ) , respectively . Proteins were purified separately by sequential Ni2+ ( HisTrap HP , GE Healthcare ) and Strep ( Strep-Tactin Superflow high-capacity , IBA , Germany ) affinity chromatography . Next , affinity tags were removed by incubating the purified proteins with recombinant Strep-tagged TEV protease in 1:100 molar ratio . The cleaved tag and the protease were separated from HAESA and SERK1 by a second Strep affinity step . The purified HAESA ectodomain was incubated with a synthetic IDA peptide ( YVPIPPSA-Hyp-SKRHN , the N-terminal Tyr residue was added to allow for peptide quantification by UV absorbance ) and the SERK1 ectodomain in 1:1:1 . 5 molar ratio . The HAESA-IDA-SERK1 complex was purified by size exclusion chromatography on a Superdex 200 HR10/30 column ( GE Healthcare ) equilibrated in 20 mM citric acid pH 5 . 0 , 100 mM NaCl ) . Peak fractions containing the complex were concentrated to ~10 mg/mL and immediately used for crystallization . About 0 . 2 mg of purified HAESA and 0 . 1 mg of purified SERK1 protein were obtained from 1 L of insect cell culture , respectively . Hexagonal crystals of the isolated HAESA ectodomain developed at room-temperature in hanging drops composed of 1 . 0 μL of protein solution ( 5 . 5 mg/mL ) and 1 . 0 μL of crystallization buffer ( 21% PEG 3 , 350 , 0 . 2 M MgCl2 · 6 H2O , 0 . 1 M citric acid pH 4 . 0 ) , suspended above 1 . 0 mL of crystallization buffer . For structure solution crystals were derivatized and cryo-protected by serial transfer into crystallization buffer supplemented with 0 . 5 M NaI and 15% ethylene glycol and cryo-cooled in liquid nitrogen . Redundant single-wavelength anomalous diffraction ( SAD ) data to 2 . 39 Å resolution were collected at beam-line PXII at the Swiss Light Source ( SLS ) , Villigen , CH with λ=1 . 7 Å . A native data set to 1 . 74 Å resolution was collected on a crystal from the same drop cryo-protected by serial transfer into crystallization buffer supplemented with 15% ( v/v ) ethylene glycol only ( λ=1 . 0 Å; Table 1 ) . HAESA complexes with IDA ( PIPPSA-Hyp-SKRHN ) , PKGV-IDA ( YPKGVPIPPSA-Hyp-SKRHN ) and IDL1 ( LVPPSG-Hyp-SMRHN ) peptide hormones were obtained by soaking apo crystals in crystallization buffer containing the respective synthetic peptide at a final concentration of 15 mM . Soaked crystals diffracted to 1 . 86 Å ( HAESA – IDA ) , 1 . 94 Å ( HAESA-PKGV-IDA ) and 2 . 56 Å resolution ( HAESA – IDL1 ) , respectively ( Table 2 ) . Orthorhombic crystals of the HAESA-IDA-SERK1 complex developed in 18% PEG 8000 , MgCl2 · 6 H2O , 0 . 1 M citric acid and diffracted to 2 . 43 Å resolution ( Table 2 ) . Data processing and scaling was done in XDS ( Kabsch , 1993 ) ( version: Nov 2014 ) . The SAD method was used to determine the structure of the isolated HAESA ectodomain . SHELXD ( Sheldrick , 2008 ) located 32 iodine sites ( CC All/Weak 37 . 7/14 . 9 ) . 20 consistent sites were input into the program SHARP ( Bricogne et al . , 2003 ) for phasing and identification of 8 additional sites at 2 . 39 Å resolution . Refined heavy atom sites and phases were provided to PHENIX . AUTOBUILD ( Terwilliger et al . , 2008 ) for density modification and automated model building . The structure was completed in alternating cycles of model building in COOT ( Emsley and Cowtan , 2004 ) and restrained TLS refinement in REFMAC5 ( Murshudov et al . , 1997 ) ( version 5 . 8 . 0107 ) against an isomorphous high resolution native data set . Crystals contain one HAESA monomer per asymmetric unit with a solvent content of ~55% , the final model comprises residues 20 – 615 . The refined structure has excellent stereochemistry , with 93 . 8% of all residues in the favored region of the Ramachandran plot , no outliers and a PHENIX . MOLPROBITY ( Davis et al . , 2007 ) score of 1 . 34 ( Table 1 ) . The HAESA – IDA – SERK1 complex structure was determined by molecular replacement with the program PHASER ( McCoy et al . , 2007 ) , using the isolated HAESA and SERK1 ( PDB-ID: 4LSC ) ( Santiago et al . , 2013 ) LRR domain structures as search models . The solution comprises one HASEA-IDA-SERK1 complex in the asymmetric unit . The structure was completed in iterative cycles of manual model-building in COOT and restrained TLS refinement in REFMAC5 . Amino acids whose side-chain position could not be modeled with confidence were truncated to alanine ( 0 . 6 – 1% of total residues ) , the stereochemistry of N-linked glycan structures was assessed with the CCP4 program PRIVATEER-VALIDATE . The refined model has 94 . 44% of all residues in the favored region of the Ramachandran plot , no outliers and a PHENIX . MOLPROBITY score of 1 . 17 ( Table 2 ) . Structural visualization was done with POVScript+ ( Fenn et al . , 2003 ) and POV-Ray ( http://www . povray . org ) . Gel filtration experiments were performed using a Superdex 200 HR 10/30 column ( GE Healthcare ) pre-equilibrated in 20 mM citric acid ( pH 5 ) and 100 mM NaCl . 100 μL of the isolated HAESA ectodomain ( 5 . 5 mg/mL ) , of the purified SERK1 LRR domain ( 3 mg/mL ) or of mixtures of HAESA and SERK1 ( either in the presence or absence of synthetic wild-type IDA , wild-type IDL1 or mutant IDA peptides at a concentration of 25 μM; 10 mg/mL; samples contained HAESA and SERK1 in 1:1 molar ratio ) were loaded sequentially onto the column and elution at 0 . 5 mL/min was monitored by ultraviolet absorbance at 280 nm . ITC experiments were performed using a Nano ITC ( TA Instruments , New Castle , DE ) with a 1 . 0 mL standard cell and a 250 μL titration syringe . Proteins were dialyzed extensively against ITC buffer ( 20 mM citric acid pH 5 . 0 , 100 mM NaCl ) and synthetic wild-type or point-mutant peptides ( with wild-type IDA sequence YVPIPPSA-Hyp-SKRHN , PKGV-IDA YPKGVPIPPSA-Hyp-SKRHN , IDA-SFVN YPIPPSA-Hyp-SKRHNSFVN , IDL1 YLVPPSG-Hyp-SMRHN and CLV3 sequence YRTV-Hyp-SG-Hyp-DPLHH ) were dissolved in ITC buffer prior to all titrations . Molar protein concentrations for SERK1 and HAESA were calculated using their molar extinction coefficient and a molecular weight of 27 , 551 and 74 , 896 Da , respectively ( determined by MALDI-TOF mass spectrometry ) . Experiments were performed at 25°C . A typical experiment consisted of injecting 10 μL aliquots of peptide solution ( 250 μM ) into 20 μM HAESA . The concentrations for the complex titrations were 150 μM of ligand ( either wild-type or point-mutant IDA peptides ) in the syringe and 10 μM of a 1:1 HAESA – SERK1 protein mixture in the cell at time intervals of 150 s to ensure that the titration peak returned to the baseline . Binding of SERK1 to HAESA was assessed by titrating SERK1 ( 100 μM ) into a solution containing HAESA ( 10 μM ) in the pre- or absence of 150 μM wild-type IDA peptide . ITC data were corrected for the heat of dilution by subtracting the mixing enthalpies for titrant solution injections into protein free ITC buffer . Data were analyzed using the NanoAnalyze program ( version 2 . 3 . 6 ) as provided by the manufacturer . Coding sequences of SERK1 kinase domain ( SERK1-KD ) ( residues 264–625 ) and HAESA-KD ( residues 671–969 ) were cloned into a modified pET ( Novagen ) vector providing an TEV-cleavable N-terminal 8xHis-StrepII-Thioredoxin tag . Point mutations were introduced into the SERK1 ( Asp447→Asn; mSERK1 ) and HAESA ( Asp837→Asn; mHAESA ) coding sequences by site directed mutagenesis , thereby rendering the kinases inactive ( Bojar et al . , 2014 ) . The plasmids were transformed into E . coli Rosetta 2 ( DE3 ) ( Novagen ) . Protein expression was induced by adding IPTG to final concentration of 0 . 5 mM to cell cultures grown to an OD600 = 0 . 6 . Cells were then incubated at 16°C for 18 hr , pelleted by centrifugation at 5000 x g and 4°C for 15 min , and resuspended in buffer A ( 20 mM Tris-HCl pH 8 , 500 mM NaCl , 4 mM MgCl2 and 2 mM β-Mercaptoethanol ) supplemented with 15 mM Imidazole and 0 . 1% ( v/v ) Igepal . After cell lysis by sonication , cell debris was removed by centrifugation at 35 , 000 x g and 4°C for 30 min . The recombinant proteins were isolated by Co2+ metal affinity purification using a combination of batch and gravity flow approaches ( HIS-Select Cobalt Affinity Gel , Sigma , St . Louis , MO ) . After washing the resin with the wash buffer ( buffer A + 15 mM Imidazole ) proteins were eluted in buffer A supplemented with 250 mM Imidazole . All elutions were then dialyzed against 20 mM Tris-HCl pH 8 , 250 mM NaCl , 4 mM MgCl2 and 0 . 5 mM TCEP . For SERK1-KD and mSERK1-KD the 8xHis-StrepII-Thioredoxin tag was removed with 6xHis tagged TEV protease . TEV and the cleaved tag were removed by a second metal affinity purification step . Subsequently , all proteins were purified by gel filtration on a Superdex 200 10/300 GL column equilibrated in 20 mM Tris pH 8 , 250 mM NaCl , 4 mM MgCl2 and 0 . 5 mM TCEP . Peak fractions were collected and concentrated using Amicon Ultra centrifugation devices ( 10 , 000 MWCO ) . For in vitro kinase assays , 1 μg of HAESA-KD , 0 . 25 μg of SERK1-KD and 2 μg of mSERK1 and mHAESA were used in a final reaction volume of 20 μl . The reaction buffer consisted of 20 mM Tris pH 8 , 250 mM NaCl , 4 mM MgCl2 and 0 . 5 mM TCEP . The reactions were started by the addition of 4 μCi [γ-32P]-ATP ( Perkin-Elmer , Waltham , MA ) , incubated at room temperature for 45 min and stopped by the addition of 6x SDS-loading dye immediately followed by incubating the samples at 95°C . Proteins of the whole reaction were subsequently separated via SDS-PAGE in 4–15% gradient gels ( TGX , Biorad , Hercules , CA ) and stained with Instant Blue ( Expedeon , San Diego , CA ) . After pictures were taken of the stained gel , 32P-derived signals were visualized by exposing the gel to an X-ray film ( Fuji , SuperRX , Valhalla , NY ) . 35S::IDA wild-type and 35S::IDA ( R66 → Ala/K67 → Ala ) over-expressing transgenic lines in Col-0 background were generated as follows: The constructs were introduced in the destination vector pB7m34GW2 and transferred to A . tumefaciens strain pGV2260 . Plants were transformed using the floral dip method ( Clough and Bent , 1998 ) . Transformants were selected in medium supplemented with BASTA up to the T3 generation . For phenotyping , plants were grown at 21°C with 50% humidity and a 16h light: 8 hr dark cycle . Plants were grown on ½ Murashige and Skoog ( MS ) plates supplemented with 1% sucrose . After 7 d , ∼30 to 40 seedlings were collected and frozen in liquid nitrogen . Total RNA was extracted using a RNeasy plant mini kit ( Qiagen , Valencia , CA ) , and 1 μg of the RNA solution obtained was reverse-transcribed using the SuperScritpVILO cDNA synthesis kit ( Invitrogen , Grand Island , NY ) . RT-qPCR amplifications and measurements were performed using a 7900HT Fast Real Time PCR-System by Applied Biosystems ( Carlsbad , CA ) . RT-qPCR amplifications were monitored using SYBR-Green fluorescent stain ( Applied Biosystems ) . Relative quantification of gene expression data was performed using the 2−ΔΔCT ( or comparative CT ) method ( Livak and Schmittgen , 2001 ) . Expression levels were normalized using the CT values obtained for the actin2 gene ( forward: TGCCAATCTACGAGGGTTTC; reverse: TTCTCGATGGAAGAGCTGGT ) . For detection and amplification of IDA sequence we used specific primers ( forward: TCGTACGATGATGGTTCTGC; reverse: GAATGGGAACGCCTTTAGGT ) . The presence of a single PCR product was further verified by dissociation analysis in all amplifications . All quantifications were made in quadruplicates on RNA samples obtained from three independent experiments . serk1-1 , serk2-2 , serk3-1 , serk4-1 and serk5-1 and Col-0 wild-type plants were grown in growth chambers at 22°C under long days ( 16 hr day/8 hr dark ) at a light intensity of 100 µE·m-2·sec-1 . Petal break-strength was quantified as the force in gram equivalents required for removal of a petal from a flower ( Butenko et al . , 2003 ) when the plants had a minimum of twenty flowers and siliques . Measurements were performed using a load transducer as described in Stenvik et al . , 2008 . Break-strength was measured for 15 plants and a minimum of 15 measurements at each position .	Plants can shed their leaves , flowers or other organs when they no longer need them . But how does a leaf or a flower know when to let go ? A receptor protein called HAESA is found on the surface of the cells that surround a future break point on the plant . When its time to shed an organ , a hormone called IDA instructs HAESA to trigger the shedding process . However , the molecular details of how IDA triggers organ shedding are not clear . The shedding of floral organs ( or leaves ) can be easily studied in a model plant called Arabidopsis . Santiago et al . used protein biochemistry , structural biology and genetics to uncover how the IDA hormone activates HAESA . The experiments show that IDA binds directly to a canyon shaped pocket in HAESA that extends out from the surface of the cell . IDA binding to HAESA allows another receptor protein called SERK1 to bind to HAESA , which results in the release of signals inside the cell that trigger the shedding of organs . The next step following on from this work is to understand what signals are produced when IDA activates HAESA . Another challenge will be to find out where IDA is produced in the plant and what causes it to accumulate in specific places in preparation for organ shedding .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Mechanistic insight into a peptide hormone signaling complex mediating floral organ abscission
Disparate redox activities that take place beyond the bounds of the prokaryotic cell cytosol must connect to membrane or cytosolic electron pools . Proteins post-translationally flavinylated by the enzyme ApbE mediate electron transfer in several characterized extracytosolic redox systems but the breadth of functions of this modification remains unknown . Here , we present a comprehensive bioinformatic analysis of 31 , 910 prokaryotic genomes that provides evidence of extracytosolic ApbEs within ~50% of bacteria and the involvement of flavinylation in numerous uncharacterized biochemical processes . By mining flavinylation-associated gene clusters , we identify five protein classes responsible for transmembrane electron transfer and two domains of unknown function ( DUF2271 and DUF3570 ) that are flavinylated by ApbE . We observe flavinylation/iron transporter gene colocalization patterns that implicate functions in iron reduction and assimilation . We find associations with characterized and uncharacterized respiratory oxidoreductases that highlight roles of flavinylation in respiratory electron transport chains . Finally , we identify interspecies gene cluster variability consistent with flavinylation/cytochrome functional redundancies and discover a class of ‘multi-flavinylated proteins’ that may resemble multi-heme cytochromes in facilitating longer distance electron transfer . These findings provide mechanistic insight into an important facet of bacterial physiology and establish flavinylation as a functionally diverse mediator of extracytosolic electron transfer . Essential microbial-environmental interactions take place beyond the bounds of the cytosol . In prokaryotes , extracytosolic biochemical processes can be situated at the extracytosolic side of the inner membrane , periplasm , cell wall , or surrounding environment . Within the extracytosolic space , redox reactions represent an important class of activities with functions in respiration , the maintenance/repair of extracytosolic proteins , and the assimilation of minerals ( Bertini et al . , 2006; Cho and Collet , 2013; Schröder et al . , 2003 ) . Extracytosolic redox processes require electron transfer pathways that connect with electron pools in the cytosol or membrane . Membrane proteins transfer electrons from donors , like reduced nicotinamide adenine dinucleotide ( NADH ) or quinols . On the extracytosolic side of the membrane , electron transfer between membrane and extracytosolic proteins is generally mediated by a redox-active protein , such as a cytochrome or thioredoxin-like protein . Cytochromes use redox-active hemes as cofactors and are often important for transferring electrons between membrane components and respiratory enzymes ( Bertini et al . , 2006 ) . Thioredoxin-like proteins employ pairs of cysteines that cycle between redox states and are often involved in extracytosolic protein maturation and repair ( Cho and Collet , 2013; Collet and Messens , 2010 ) . Flavins are a group of molecules that contain a conserved redox-active isoalloxazine ring system . In a reversible manner , the oxidized state of the flavin isoalloxazine ring system can undergo a one-electron reduction to generate a semiquinone state or a two-electron reduction to generate a hydroquinone state . Many microbes synthesize the flavin-derivative riboflavin ( or vitamin B2 ) , which can be phosphorylated to yield flavin mononucleotide ( FMN ) and further adenylated to flavin adenine dinucleotide ( FAD ) . In part , because of their ability to transition between multiple redox states , FMN and FAD are well suited to act as enzyme cofactors . Proteins that bind flavins ( flavoproteins ) are common throughout nature and function in diverse redox activities ( Fraaije and Mattevi , 2000 ) . In addition to the well-established electron transfer mediators , like cytochromes and thioredoxin-like proteins , an evolutionarily conserved FMN-binding domain has more recently been implicated in extracytosolic electron transfer ( Zhou et al . , 1999 ) . FMN-binding domains are post-translationally flavinylated by the ApbE enzyme , which transfers the FMN portion of a substrate FAD molecule to a conserved [S/T]GA[S/T]-like sequence motif ( flavinylated amino acid in bold ) ( Figure 1A; Bertsova et al . , 2013 ) . The resulting phosphoester bond irreversibly links the FMN ribitylphosphate group to a serine/threonine hydroxyl side chain in the FMN-binding domain . While covalently bound flavins are a relatively common feature of flavoproteins , the ApbE-catalyzed reaction is unusual . In most flavinylated proteins , a covalent bond links the flavin’s isoalloxazine ring to an amino acid side chain ( Macheroux et al . , 2011 ) . This type of linkage induces a change in the flavin’s redox potential that facilitates the protein’s activity ( Macheroux et al . , 2011 ) . By contrast , the ApbE-catalyzed phosphoester linkage is outside the flavin’s isoalloxazine ring and unlikely to impact the cofactor’s redox potential . Since non-flavinylated FMN-binding domains have low flavin-binding affinity , the ApbE-catalyzed reaction may simply function to secure the cofactor to the protein ( Borshchevskiy et al . , 2015 ) . Once the flavin is linked to the FMN-binding domain , interactions with the folded protein stabilize the flavin’s semiquinone state in a fashion that presumably enhances electron transfer ( Barquera , 2014; Backiel et al . , 2008 ) . ApbE-flavinylated FMN-binding domains have been found to be critical for five characterized extracytosolic electron transfer systems . These systems include the cation-pumping NADH:quinone oxidoreductase ( NQR ) and Rhodobacter nitrogen fixation ( RNF ) complexes ( Figure 1B and C ) , nitrous oxide and organohalide respiratory complexes ( Figure 1D and E ) , and a Gram-positive extracellular electron transfer system ( Figure 1F; Backiel et al . , 2008; Buttet et al . , 2018; Light et al . , 2018; Zhang et al . , 2017; Zhou et al . , 1999 ) . In addition , ApbE-flavinylated [S/T]GA[S/T]-like sequence motifs in homologous extracytosolic fumarate and urocanate reductases facilitate transfer from respiratory electron transport chains ( Bogachev et al . , 2012; Kees et al . , 2019; Light et al . , 2019 ) . Notably , each of these characterized activities links AbpE flavinylation to a different aspect of microbial cellular respiration , with the NQR and RNF complexes being particularly noteworthy . These systems are widely distributed throughout microbial life and catalyze key intermediate steps in the energy metabolism of numerous microbes ( Barquera , 2014; Buckel and Thauer , 2018; Reyes-Prieto et al . , 2014 ) . In recent years , large-scale comparative genomic analyses have emerged as a powerful tool for discovering functionally and/or mechanistically related features of prokaryotic biology ( Burstein et al . , 2017; Crits-Christoph et al . , 2018; Doron et al . , 2018 ) . Here , we develop a ‘guilt by association’ approach ( Aravind , 2000 ) ( summarized in Figure 2—figure supplement 1 ) that exploits genomic diversity to contextualize the significance of extracytosolic flavinylation . Our analysis of flavinylation-associated gene clusters provides evidence of widespread flavinylation throughout bacteria and uncovers new connections to respiration and iron assimilation . We further identify uncharacterized aspects of extracytosolic flavinylation , including novel ApbE substrates and a class of multi-flavinylated proteins . These findings place ApbE-flavinylated proteins alongside cytochromes and thioredoxin-like proteins as central mediators of bacterial extracytosolic electron transfer . As all previously characterized flavinylation systems contain genes that encode for an ApbE enzyme and a substrate that contains an FMN-binding domain , we reasoned that these features are indicative of flavinylation-mediated electron transfer . To identify flavinylation-mediated electron transfer systems , we searched for genes with ApbE ( Pfam accession PF02424 ) or FMN-binding domains ( Pfam accession PF04205 ) within a collection of 31 , 910 genomes that are representative of the genetic diversity of the prokaryotes ( Parks et al . , 2018; Parks et al . , 2020 ) . We found that 18 , 965 of bacterial genomes and 238 of archaeal genomes encode an FMN-binding domain-containing protein and/or an ApbE enzyme . To focus the search on extracytosolic electron transfer , we eliminated genes that lacked a computationally predicted signal peptide or lipidation site . This analysis provides evidence that ~50% of the bacterial ( 15 , 095 ) and ~4% of archaeal ( 63 ) genomes possess extracytosolic flavinylation components ( Supplementary file 1 ) . A phylogenetic analysis of the resulting dataset reveals that extracytosolic flavinylation components are broadly distributed across bacterial life – though strikingly underrepresented in Cyanobacteria and the candidate phyla radiation ( Figure 2 ) . We next took advantage of the colocalization of genes for multi-subunit complexes on the genome to determine the prevalence and phylogenetic distribution of previously characterized flavinylated systems . We devised operational definitions in which the close proximity of a key gene in each system to an FMN-binding- or ApbE-containing gene was used to assign clusters to characterized systems ( Supplementary file 2 ) . These analyses revealed characterized systems in 8928 genomes broadly distributed across bacterial life ( Figure 2 ) . In addition to assigning extracytosolic electron transfer functions , our analyses identified 6230 genomes that contained evidence of extracytosolic flavinylation but lacked a characterized extracytosolic electron transfer system . We also found that many genomes have multiple ApbE and/or FMN-binding genes and thus likely possess multiple extracytosolic electron transfer functions ( Figure 1—figure supplement 1 ) . As these observations suggested that a significant proportion of extracytosolic electron transfer systems remain uncharacterized , we next turned to the identification of uncategorized flavinylation-associated gene clusters . Our initial approach only considered FMN-binding domains as potential flavinylation substrates . However , preliminary analyses identified 17% of genomes ( 2571 total ) within our dataset that encoded an ApbE but no FMN-binding domain-containing protein – implying that some ApbE substrates lack an FMN-binding domain ( Figure 3—figure supplement 1 ) . To identify novel ApbE protein substrates , we examined the genomic context of these ‘orphan’ apbE genes , looking for gene cluster patterns conserved across multiple genomes ( Supplementary file 1 ) . We observed that a subset of orphan apbE genes are associated with DUF2271 – an ~135 amino acid domain of unknown function . In some genomes a single gene encodes a protein with both ApbE and DUF2271 domains ( e . g . , NCBI accession RUL87804 . 1 ) , but more commonly a separate DUF2271 gene is part of a gene cluster that includes an apbE . Consistent with DUF2271 serving as a flavinylation substrate , we identified a conserved [S/T]GA[S/T] motif within ApbE-associated DUF2271s ( Figure 3A ) . We expressed the Amantichitinum ursilacus DUF2271 protein in Escherichia coli and found that it was flavinylated in the presence of its cognate ApbE ( Figure 3B ) . Consistent with the conserved [S/T]GA[S/T] motif representing the sole ApbE target , we found that replacing the threonine at the predicted flavinylation site with an alanine abrogated flavinylation ( Figure 3B ) . We observed another subset of orphan apbE genes that are part of gene clusters that contain a gene annotated as DUF3570 – an ~420 amino acid domain of unknown function . Consistent with DUF3570 serving as a flavinylation substrate , we identified two conserved [S/T]GA[S/T] motifs within ApbE-associated DUF3570s ( Figure 3C ) . Using the coexpression approach described above , we confirmed that the Chlorobium luteolum DUF3570 was flavinylated in the presence of its cognate ApbE ( Figure 3D ) . Consistent with both of the identified [S/T]GA[S/T] motifs being modified , we found that replacing serines with alanines at both predicted flavinylation sites was required to abrogate flavinylation ( Figure 3D ) . These results suggest that DUF2271 and DUF3570 are novel ApbE substrates . Including DUF2271 and DUF3570 within our analyses significantly expanded the number and diversity of predicted ApbE substrates and decreased the number of orphan apbE genomes within the dataset from 17% to 4% ( 670 total ) ( Figure 3—figure supplement 1 ) . Subsequent analyses thus likely account for a significant fraction of flavinylation substrates – though the fact that orphan apbE genomes remain suggests that less prevalent ApbE substrates remain unidentified . We next sought to gain insight into novel roles of protein flavinylation . Characterized flavinylation-based electron transfer systems are minimally defined by a cytosolic electron donor , a transmembrane electron transfer apparatus , and the flavinylated extracytosolic electron acceptor – though , in principle , the direction of electron flow could be reversed . We reasoned that identification of cytosolic and membrane components was important for understanding the extracytosolic capability of uncharacterized flavinylated systems . To clarify the role of flavinylation-associated gene clusters , we analyzed the genomic context of a subset of representative genes with DUF3570 , DUF2271 , FMN-binding , and ApbE domains that were not assigned to a characterized system in our initial analyses . Annotations for the five upstream and five downstream genes were manually reviewed . From these gene clusters , we identified five putative transmembrane electron transfer apparatuses that are present in 6183 genomes , including 3635 of the 6230 genomes that lack a characterized extracytosolic electron transfer system ( Figure 4 , Figure 4—figure supplement 1 and Supplementary file 2 ) . The following subsections describe the organization and likely functions of these systems . Annotations of the gene clusters associated with the flavinylated systems are provided in Supplementary file 1 . We identified 2465 flavinylation gene clusters in 2153 genomes that encode an ‘NapH-like’ iron-sulfur cluster-binding protein ( Pfam accession PF12801 or KEGG accession K19339 ) that contains several transmembrane helices ( Figure 2 and Supplementary file 1 ) . These proteins are homologous to NapH , the putative quinone-binding subunit of periplasmic nitrate reductase , and exhibit a broad phylogenetic distribution ( Figure 4A; Brondijk et al . , 2002 ) . The majority of NapH-like proteins identified in our analysis contain an extracytosolic N-terminal FMN-binding domain ( 1926 gene clusters ) whereas the remaining ones ( 539 gene clusters ) are in a genomic locus with a second gene that encodes an FMN-binding domain-containing protein . These proteins likely receive electrons from a donor ( probably a quinol in some cases ) that are transferred via the iron-sulfur cluster across the membrane ( Figure 4A ) . Several lines of evidence suggest that many NapH-like proteins function with respiratory oxidoreductases . Previously characterized NapH-like proteins PceC and NosZ have been shown to be flavinylated and are part of gene clusters for organohalide and nitrous oxide reduction , respectively ( Figure 1; Buttet et al . , 2018; Zhang et al . , 2017 ) . These previously characterized NapH-like systems were identified within 1197 genomes within our dataset ( Figure 2 ) . We also identified NapH-like gene clusters with an extracytosolic nitrite reductase in 133 genomes or an ethanol oxidase in 172 genomes . Thus , while the reactions catalyzed by the majority of NapH-like systems remain unknown , this electron transfer apparatus seems to be modularly employed to facilitate electron transfer to reductases or from oxidases ( Figure 4A ) . This modularity of NapH-oxidoreductase associations is further underscored by a phylogenetic analysis , which suggests that NapH-like associations with nirS and exaA evolved independently multiple times ( Figure 4—figure supplement 2 ) . We identified 1797 flavinylation gene clusters in 1468 genomes that encode an ‘MsrQ-like’ ( Pfam accession PF01794 ) protein ( Supplementary file 1 ) . These clusters are broadly conserved in Actinobacteria and are infrequently identified in other lineages ( Figure 2 and Figure 4B ) . MsrQ-like proteins are predicted to have six transmembrane helices and two heme-binding sites . MsrQ-like proteins are homologous to MsrQ , the quinone-binding subunit of periplasmic methionine sulfoxide reductase ( Gennaris et al . , 2015 ) . MsrQ-like proteins are also distantly related to eukaryotic proteins that function in transmembrane electron transfer , including NADPH oxidase and STEAP iron reductases ( Zhang et al . , 2013 ) . MsrQ-like gene clusters typically include apbE and an FMN-binding domain-containing gene . MsrQ-like proteins often contain a C-terminal NAD-binding domain ( Pfam accession PF00175 , 1437 gene clusters ) . In other cases , the MsrQ-like gene clusters include an NuoF-like protein ( Pfam accession PF10589 , 31 gene clusters ) , homologous to the NAD-binding subunit of complex I . At least a subset of MsrQ-like proteins thus likely use NAD ( P ) H as a cytosolic electron donor ( Figure 4B ) . We also identified 153 MsrQ-like gene clusters that encode a protein homologous to the eukaryotic ferrous iron transporter VIT1 ( Pfam accession PF01988 ) . This association suggests that some MsrQ-like systems function as assimilatory iron reductases that facilitate iron uptake through VIT1 ( Figure 4B ) . We identified 3220 flavinylation gene clusters in 3040 genomes that encode a PepSY-like ( Pfam accessions PF03929 and PF16357 ) protein ( Supplementary file 1 ) . PepSY-like proteins contain three transmembrane helices and are broadly distributed throughout Gram-negative bacteria ( Figures 2 and 4C ) . PepSY-like gene clusters frequently encode an ApbE enzyme , the flavinylation substrate DUF2271 ( Pfam accession PF10029 , 2833 clusters ) , and a secreted DUF4198 ( Pfam accession PF10670 , 1500 clusters ) protein . Little is known about the structure or function of PepSY-like proteins . The only characterized PepSY-like homolog , Vibrio cholerae VciB ( which is not associated with identified flavinylation components ) , was reported to possess extracytosolic iron reductase activity ( Peng and Payne , 2017 ) an — an activity consistent with transmembrane electron transfer activity . We further identified 1077 PepSY-like genes that have a NAD-binding domain ( Pfam accession PF00175 ) , suggesting that a subset of these proteins use cytosolic NADH as an electron donor ( Figure 4C ) . Several observations implicate a role for identified PepSY-like gene clusters in iron reduction and assimilation . First , a functional connection to MsrQ-like proteins is suggested by our observation that 108 gene clusters contain both PepSY-like and MsrQ-like genes . Second , 167 PepSY-like gene clusters contain a VIT1 ( Pfam accession PF01988 ) ferrous iron transporter ( Figure 4C ) . Finally , PepSY-like gene clusters have been shown to be repressed by Fur ( the primary transcription regulator that responds to iron limitation ) in Shewanella oneidensis and Caulobacter crescentus ( da Silva Neto et al . , 2013; Wan et al . , 2004 ) . Moreover , in some cases , PepSY-like proteins may be actively involved in the extraction of siderophore-bound iron , as periplasmic reduction has been shown to be important for the uptake of siderophore-bound iron and 231 PepSY-like gene clusters encode a TonB receptor ( Pfam accession PF03544 ) related to well-characterized outer membrane siderophore transporters ( Figure 4C; Liu et al . , 2018a; Manck et al . , 2020 ) . We identified 285 flavinylation gene clusters in 275 genomes that contain a DsbD protein ( Pfam accession PF02683 ) ( Figure 2 and Supplementary file 1 ) . DsbD is part of a well-studied transmembrane protein family that uses cysteine pairs to transfer electrons across the membrane ( Krupp et al . , 2001; Missiakas et al . , 1995 ) . DsbD family proteins generally transfer electrons onto extracytosolic thioredoxin-like proteins , which in turn use a similar thiol-disulfide exchange chemistry to promote extracytosolic redox-dependent activities , such as oxidative protein folding ( Cho and Collet , 2013 ) . The DsbD gene clusters identified in our analyses typically include genes for ApbE , the flavinylation substrate DUF3570 ( Pfam accession PF12094 , 227 clusters ) , a thioredoxin-like protein ( Pfam accession PF13899 , 183 clusters ) , and a DUF4266 protein ( Pfam accession PF14086 , 221 clusters ) . DUF4266 is a small secreted protein with a highly conserved C-terminal cysteine – ( any amino acid ) – cysteine ( CXC ) sequence motif that may undergo redox cycling . We also detected 316 gene clusters that encode ApbE , DUF3570 , thioredoxin-like , and DUF4266 but do not colocalize on the genome with DsbD – implying that this system may be more common than is revealed by the DsbD-dependent analysis presented in Figure 2 . These observations suggest that a hybrid thioredoxin-like/flavinylation-based system receives electrons from DsbD in some bacteria ( Figure 4D ) . While the function of DUF3570-based electron transfer remains uncertain , 25 gene clusters contain a VIT1 ferrous iron transporter and thus may play a role in iron assimilation ( Supplementary file 1 ) . We identified 127 flavinylation gene clusters in 127 genomes , primarily from the Actinobacteria and candidate phyla radiation , with evidence of partial NQR or RNF systems ( Figure 2 ) . As shown in Figure 1 , the NQR and RNF complexes contain a common core apparatus with two transmembrane electron transfer pathways that together achieve a semicircular electron flow ( Juárez et al . , 2010; Steuber et al . , 2014 ) . A first path takes electrons from the cytosol to the extracytosolic FMN-binding domain , while a second path takes electrons from the FMN-binding domain back to a cytosolic substrate . NQR/RNF-like gene clusters encode for components associated with a single electron transfer pathway and thus likely function for unidirectional electron flow ( Figure 4E ) . NQR/RNF-like gene clusters encode a protein with an N-terminal membrane domain that is homologous to NqrB/RnfD and a cytosolic C-terminal domain homologous to the NAD-binding domain NqrF ( Pfam accession PF00175 ) . NqrB/RnfD is flavinylated by ApbE and is thought to play a role in electron transfer across the membrane ( Juárez et al . , 2010 ) . This system presumably uses NAD ( P ) H as a cytosolic electron donor for electron transfer to an extracytosolic FMN-binding domain . Six NQR/RNF-like gene clusters contain a VIT1 ferrous iron transporter , suggesting that some of these systems function in iron assimilation ( Supplementary file 1 ) . We next asked about the function of flavinylation-associated gene clusters that lack a core transmembrane electron transfer system . This category of extracytosolic proteins presumably relies on proteins encoded elsewhere on the genome to link up with membrane electron pools . Inspection of these gene clusters led to the identification of two noteworthy examples that are described in the following subsections . A gene cluster that includes the ferrous iron-binding protein P19 ( Pfam accession PF10634 ) and the FTR1 ( Pfam accession PF03239 ) iron transporter has previously been shown to encode a mechanistically uncharacterized system involved in iron assimilation ( Chan et al . , 2010 ) . We identified 260 P19 gene clusters , mostly in Gram-positive bacteria , that contain a gene with an FMN-binding domain ( Figure 5—figure supplement 1A and B and Supplementary file 1 ) . Interestingly , many Gram-negative P19 gene clusters lack an FMN-binding gene but contain an additional thioredoxin-like gene ( Liu et al . , 2018b; Figure 5—figure supplement 1B ) . The role of the FMN-binding/thioredoxin-like protein in these systems has not been defined but , similar to other assimilatory iron reductases , could engage in redox chemistry to facilitate iron uptake . These observations thus establish another connection between flavinylation and microbial iron assimilation and highlight functional parallels between flavinylation and thioredoxin-like extracytosolic electron transfer . Fumarate reductase-like enzymes ( members of the Pfam accession PF00890 enzyme superfamily ) are a group of evolutionarily related proteins that catalyze a variety of redox reactions – though phylogenetic analyses suggest that substrates for many members of the superfamily remain unknown ( Jardim-Messeder et al . , 2017; Light et al . , 2019 ) . We observe that fumarate reductase-like enzymes often contain an FMN-binding domain ( 3070 proteins encoded by 2979 distinct gene clusters in 1236 genomes ) or are part of gene clusters ( ±2 genes ) that contain an FMN-binding domain ( 189 gene clusters in 171 genomes ) ( Supplementary file 1 ) . Several characterized fumarate reductase-like enzymes ( including fumarate , urocanate , and methacrylate reductases ) are extracytosolic and function in respiration ( Bogachev et al . , 2012; Light et al . , 2019; Mikoulinskaia et al . , 1999 ) . The Listeria monocytogenes fumarate reductase and the S . oneidensis urocanate reductase have been shown to be flavinylated ( Bogachev et al . , 2012; Light et al . , 2019 ) . In both cases , the flavinylation motif is thought to facilitate electron transfer from electron transport chain components encoded elsewhere on the genome to the enzyme active site ( Kees et al . , 2019; Light et al . , 2019 ) . These observations thus suggest that FMN-binding domains mediate electron transfer from membrane components to a prevalent class of extracytosolic reductases and highlight another connection between flavinylation and respiration . A comparison of extracytosolic electron transfer systems identified in our analyses revealed multiple instances in which cytochrome and flavinylated electron transfer components appear to be performing similar electron transfer roles within related systems . For example , we identified a flavinylated NapH-like protein that is well situated to mediate electron transfer from the extracytosolic alcohol oxidase to the electron transport chain , where a cytochrome c protein has been shown to play this role in other microbes ( Figure 5A; Schobert and Görisch , 1999 ) . Another example of this dynamic is provided by a comparison of fumarate reductases . The extracytosolic S . oneidensis fumarate reductase contains an N-terminal multi-heme cytochrome c domain ( Pfam accession PF14537 ) , whereas the related L . monocytogenes enzyme uses an FMN-binding domain to connect to the electron transport chain ( Figure 5B; DiChristina and DeLong , 1994; Light et al . , 2019 ) . Different microbes thus seem to utilize flavinylation and cytochrome domains in a similar fashion to link respiratory enzymes to electron transport chains . We further found the type of flavinylation/cytochrome substitution observed for fumarate reductase to be indicative of a broader pattern within fumarate reductase-like enzymes . We identified 147 gene clusters within 108 genomes that encode separate fumarate reductase-like and cytochrome proteins and 879 genes within 360 genomes that encode a single protein with both fumarate reductase-like and cytochrome domains ( Supplementary file 3 ) . Many genomes encode multiple fumarate reductase-like paralogs and 99 genomes encode both cytochrome- and flavinylation-associated enzymes ( Supplementary file 3 ) . This dynamic is exemplified by S . oneidensis , which in addition to the mentioned cytochrome-associated fumarate reductase contains a flavinylated urocanate reductase that also exhibits a respiratory function ( Figure 5C; Bogachev et al . , 2012 ) . Broadly similar flavinylation- and cytochrome-based extracytosolic electron transfer mechanisms thus seemingly coexist within some microbes . Multi-heme cytochromes are proteins that bind multiple hemes to achieve longer distance extracytosolic electron transfer ( Blumberger , 2018 ) . Among other functions , multi-heme proteins are important for transferring electrons across the cell envelope to insoluble electron acceptors that are inaccessible within the cytosolic membrane ( El-Naggar et al . , 2010; Wang et al . , 2019 ) . We have observed that extracytosolic proteins with multiple FMN-binding domains are also common ( 2081 proteins in 1530 genomes ) , particularly in Gram-positive bacteria , and contain as many as 13 predicted flavinylation sites ( Figure 6A and B and Supplementary file 1 ) . Multi-cofactor binding properties thus establish another parallel between cytochrome and flavinylation-based electron transfer . Gene cluster analyses provide some insight into the basis of multi-flavinylated protein electron transfer . We find that multi-flavinylated proteins are often associated with established transmembrane electron transfer components and thus likely receive electrons through conventional mechanisms ( Figure 6C ) . We observe that some clusters contain large unannotated proteins with putative cell wall-binding domains – such as SLH ( Pfam accession PF00395 ) , Rib ( Pfam accession PF08428 ) , FIVAR ( Pfam accession PF07554 ) , or LysM ( Pfam accession PF01476 ) – and are thus likely involved in redox chemistry within the cell wall or at the cell surface ( Figure 6D ) . We also identified 157 clusters that encode a multi-flavinylated protein and additional proteins with FMN-binding domains ( Figure 6D and E ) . These clusters encode as many as five proteins with FMN-binding domains and frequently contain multiple multi-flavinylated proteins ( Figure 6D ) . These observations suggest that some multi-flavinylated proteins are part of multi-step electron transfer pathways and may form elaborate multi-subunit complexes that span the cell wall . The role of multi-flavinylated gene clusters is generally unclear , with only a minority of clusters providing limited functional clues . We identified a subset of clusters that encode proteins with multiple fumarate reductase-like domains that likely establish an unusual multi-functional reductase platform ( Figure 6D and G ) . We also identified a number of RNF clusters that contain multi-flavinylated RnfGs ( the extracytosolic flavinylated subunit in RNF complexes ) ( Figure 6C ) . These multi-flavinylated RNF complexes likely provide a second electron transfer pathway that facilitates transfer to alternative extracytosolic acceptors . This type of bifurcated electron transfer would be similar to a multi-heme cytochrome-based transfer mechanism recently suggested in studies of a methanogen RNF ( Holmes et al . , 2019 ) . Interestingly , we also identified a subset of multi-flavinylated RnfGs in the family Christensenellales that contain a multi-heme cytochrome domain and thus may assume additional functions relevant for cytochrome-based electron transfer ( Figure 6H ) . While much remains to be learned , these preliminary observations are consistent with multi-flavinylated proteins establishing elaborate and functionally diverse electron transfer pathways . Comparative analysis of gene clusters within collections of prokaryotic genomes has emerged as a powerful discovery tool in recent years . Our large-scale survey of diverse genomes extends this approach and reveals that ApbE-mediated flavinylation is a prominent feature of bacterial physiology . More granular analyses of gene clusters provide evidence consistent with extracytosolic flavinylation usage in diverse redox activities and suggest that modular properties facilitate the integration of flavinylated components with various biochemical processes . While additional genetic and/or biochemical studies will be required to develop a better understanding of the physiological roles of flavinylation , these preliminary observations are consistent with flavinylated proteins being important components of microbial extracytosolic electron transfer . Our findings suggest that , alongside thioredoxin-like proteins and cytochromes , ApbE-flavinylated proteins represent a third major class of mediators of extracytosolic electron transfer . The existence of three mechanistically distinctive protein classes with some apparent functional interchangeability ( Figure 5 ) stimulates fundamental questions about the environmental or physiological context that favors each system . Unfortunately , the multiple functions and widespread distribution of flavinylated proteins across diverse microbes make it difficult to identify unique features that distinguish microbes that encode flavinylation components . Nevertheless , the relationship between flavinylation- and cytochrome-based electron transfer is interesting . The apparent functional interchangeability in the electron transfer capabilities of iron- and flavin-containing proteins is reminiscent of the relationship between ferredoxins and flavodoxins . Ferredoxins and flavodoxins are redox-active proteins that function in a number of cytosolic redox activities and contain iron and flavin cofactors , respectively ( Yoch and Valentine , 1972 ) . Microbes switch from ferredoxin to flavodoxin usage in iron-poor environments – presumably because minimizing the demand for iron cofactors is important for conserving the cellular iron reserve in this context ( Knight et al . , 1966; Smillie , 1965 ) . Based on this precedent , it seems plausible that flavinylation-based electron transfer mechanisms might be particularly advantageous within iron-poor environments . The coexistence of functionally similar but mechanistically distinctive flavinylation and cytochrome electron transfer components may thus , in part , reflect divergent resource management strategies in distinct environmental contexts . Of the proteins linked to extracytosolic electron transfer by our studies , the class of multi-flavinylated proteins are particularly intriguing . These proteins may resemble multi-heme cytochromes in their use of multiple redox-active cofactors to achieve longer distance electron transfer . A particularly noteworthy aspect of multi-heme cytochromes concerns their ability to establish ‘nanowires’ that have a variety of potential biotechnological applications ( Blumberger , 2018; Liu et al . , 2020 ) . Considering the unique redox properties of flavins , the electron-transferring behavior of multi-flavinylated proteins may provide an interesting juxtaposition to multi-heme cytochromes with implications for the development of novel redox-based biotechnologies . The function and mechanism of electron transfer in multi-flavinylated proteins may thus represent an interesting subject for future studies . The 30 , 238 bacterial and 1672 archaeal genomes from the GTDB ( release 05-RS95 of July 17 , 2020 ) were downloaded with the taxonomy and the predicted protein sequences of the genomes ( Parks et al . , 2020 ) . Protein sequences were functionally annotated based on the accession of their best Hmmsearch match ( version 3 . 3 ) ( E-value cut-off 0 . 001 ) ( Eddy , 1998 ) against the KOfam database ( downloaded on February 18 , 2020 ) ( Aramaki et al . , 2020 ) . Domains were predicted using the same Hmmsearch procedure against the Pfam database ( version 33 . 0 ) ( Mistry et al . , 2021 ) . SIGNALP ( version 5 . 0 ) was run to predict the putative cellular localization of the proteins using the parameters -org arch in archaeal genomes and -org gram +in bacterial genomes ( Almagro Armenteros et al . , 2019 ) . Prediction of transmembrane helices in proteins was performed using TMHMM ( version 2 . 0 ) ( default parameters ) ( Krogh et al . , 2001 ) . The five genes downstream and upstream of an ApbE , FMN-binding domain , DUF3570 or DUF2271 encoding genes were first collected . Only gene clusters with at least one signal peptide or lipidation site predicted in one of the four target genes were considered for further analysis and were referred to as "flavinylation-associated gene clusters . " The flavinylation-associated gene clusters were then assigned to 1 of the 10 flavinylated systems based on the presence of key genes reported in Supplementary file 2 . The RNF system was considered present if a Na+-translocating ferredoxin:NAD+ oxidoreductase subunit B ( RnfB , KEGG accession K03616 ) was encoded within a flavinylation-associated gene cluster . The nitrous oxide reduction system ( Nos ) was considered present if a nitrous oxide reductase ( NosZ , KEGG accession K00376 ) was encoded within a flavinylation-associated gene cluster . The organohalide respiration system was defined by the presence of a PceA enzyme ( Pfam accession PF13486 ) encoded within a flavinylation-associated gene cluster . The extracellular electron transfer system was considered present if a NADH dehydrogenase ( KEGG accession K03885 ) with a transmembrane helix was encoded within a flavinylation-associated gene cluster . The NQR was defined by the Na+-transporting NADH:ubiquinone oxidoreductase subunit F ( NqrF , KEGG accession K00351 ) encoded within a flavinylation-associated gene cluster . The NapH-like was considered present if an ‘NapH-like’ iron-sulfur cluster-binding protein ( Pfam accession PF12801 or KEGG accession K19339 ) was encoded within a flavinylation-associated gene cluster . NapH-like gene clusters that encoded a NirS ( KEGG accession K15864 ) were identified as containing a nitrite reductase . NapH-like gene clusters that encoded an ExaA enzyme ( KEGG accession K00114 ) were identified as containing an alcohol oxidase . The MsrQ-like system was defined by the presence of ‘MsrQ-like’ ( Pfam accession PF01794 ) gene within a flavinylation-associated gene cluster . The PepSY-like system was defined by a PepSY-like ( Pfam accessions PF03929 and PF16357 ) protein encoded within a flavinylation-associated gene cluster . The DsbD system was considered present if a DsbD protein ( Pfam accession PF02683 ) was encoded within a flavinylation-associated gene cluster . Finally , the NQR/RNF-like system was considered present if a flavinylation-associated gene cluster encoded a protein with an N-terminal membrane domain that is homologous to NqrB/RnfD ( Pfam accession PF03116 ) and a cytosolic C-terminal domain homologous to the NAD-binding domain NqrF ( Pfam accession PF00175 ) . The ‘NapH-like’ iron-sulfur cluster-binding protein tree was built as follows . Sequences were aligned using MAFFT ( version 7 . 390 ) ( –auto option ) ( Katoh and Standley , 2016 ) . The alignment was further trimmed using Trimal ( version 1 . 4 . 22 ) ( –gappyout option ) ( Capella-Gutierrez et al . , 2009 ) . Tree reconstruction was performed using IQ-TREE ( version 1 . 6 . 12 ) ( Nguyen et al . , 2015 ) , using ModelFinder to select the best model of evolution ( Kalyaanamoorthy et al . , 2017 ) and with 1000 ultrafast bootstrap ( Hoang et al . , 2018 ) . A maximum-likelihood tree was calculated based on the concatenation of 14 ribosomal proteins ( L2 , L3 , L4 , L5 , L6 , L14 , L15 , L18 , L22 , L24 , S3 , S8 , S17 , and S19 ) . Homologous protein sequences were aligned using MAFFT ( version 7 . 390 ) ( --auto option ) ( Katoh and Standley , 2016 ) and alignments refined to remove gapped regions using Trimal ( version 1 . 4 . 22 ) ( --gappyout 570 option ) ( Capella-Gutierrez et al . , 2009 ) . The protein alignments were concatenated with a final alignment of 9152 genomes and 2850 positions . Tree reconstruction was performed using IQ-TREE ( version 1 . 6 . 12 ) ( Nguyen et al . , 2015 ) . A LG + I + G4 model of evolution was selected using ModelFinder ( Kalyaanamoorthy et al . , 2017 ) and 1000 ultrafast bootstraps were performed ( Hoang et al . , 2018 ) . Sequences of flavinylation-associated DUF3570 and DUF2271 proteins were aligned using EMBL-EBI Clustal Omega Multiple Sequence Alignment ( Madeira et al . , 2019 ) . Sequence logos of the flavinylation sites shown in Figure 3 were generated in R using the ‘ggseqlogo’ package ( Wagih , 2017 ) . A synthetic construct of the signal peptide-truncated A . ursilacus IGB-41 DUF2271 gene ( NCBI accession WP_053936890 . 1 ) was subcloned into the pMCSG53 vector . A second construct contained a ribosome-binding site and the signal peptide-truncated cognate apbE ( NCBI accession WP_053936888 . 1 ) just downstream of the DUF2271 gene . A similar cloning strategy was used for the C . luteolum DSM 273 DUF3570 ( NCBI accession ABB24424 . 1 ) and its cognate apbE ( NCBI accession ABB24423 . 1 ) . Point mutations of the DUF3570- and DUF2271-encoding genes were generated using the NEB Q5 Site-Directed Mutagenesis Kit . Briefly , overlapping primers containing mutated sequences were used in a PCR using pMCSG53::DUF3570 or pMCSG53::DUF2771 as DNA template to generate expression vectors containing respective mutant sequences . Plasmids containing wild-type protein sequences were removed using digestion with the DpnI enzyme , which only acts on methylated DNA sequences . Sequence verified plasmids were transformed in E . coli BL21 cells ( Rosetta DE3 , Novagen ) . A single colony of each expression strain was isolated on Luria-Bertani ( LB ) agar supplemented with carbenicillin ( 100 μg/mL ) and inoculated into 15 mL of LB . Following overnight growth , cultures were diluted in 500 mL of brain heart infusion broth to a final OD600 of 0 . 1 . When the OD600 reached 0 . 7–1 protein overexpression was induced by adding isopropyl β-D-1-thiogalactopyranoside to a final concentration of 1 mM . The culture was then incubated overnight at 25°C with aeration and collected by centrifugation ( 7000 × g for 15 min ) . After removing the supernatant , cells were washed in 30 mL of lysis buffer ( 5:1 v/weight of pellet; 300 mM NaCl , 1 mM dithiothreitol , 10 mM imidazole , 1 mM ethylenediaminetetraacetic acid [EDTA] , and 50 mM Tris-HCl pH = 7 . 5 ) . Pelleted cells were stored at −80°C overnight , resuspended in lysis buffer , lysed by sonication ( 8 × 30 s pulses ) , and cleared by centrifugation ( 40 , 000 × g for 30 min ) . For the purification of A . ursilacus DUF2271 , cell lysate was collected and loaded onto a 5 mL HisTrapTM column ( GE Healthcare ) using the ÄKTA Pure FPLC . Protein was eluted using an imidazole concentration gradient with a maximal concentration of 500 mM . Protein concentrations of eluted fractions containing His6-tagged DUF2271 were measured on a DeNovix DS-11 FX+Spectrophotometer based on protein molar mass and extinction coefficient and standardized to 0 . 4 mg/mL . Initial observations revealed that the majority of expressed C . luteolum DUF3570 was present in the lysed cell pellet . To purify C . luteolum DUF3570 , the cell pellet was washed with wash buffer ( 10:1 v/weight of pellet; 100 mM Tris-HCl pH = 7 . 5 , 300 mM NaCl , 2 mM 2-mercaptoethanol , 1 M guanidine-HCl , 1 mM EDTA , and 2% w/v Triton X-100 ) and centrifuged at 40 , 000 × g for 30 min . The supernatant was then discarded and this washing step was repeated until the supernatant was clear . The cell pellet was then resuspended in wash buffer ( 10:1 v/weight of pellet ) without guanidine-HCl and Triton X-100 and centrifuged ( 40 , 000 × g for 15 min ) to remove guanidine-HCl and Triton X-100 . After supernatant was discarded , the pellet was resuspended in extraction buffer ( 4:1 v/weight of pellet; 100 mM Tris-HCl pH = 7 . 5 , 300 mM NaCl , 10 mM imidazole , 6 M guanidine-HCl , 2 mM 2-mercaptoethanol , and 1 mM EDTA ) and protein was denatured by overnight rotator mixation . Supernatant containing denatured DUF3570 was collected by centrifugation ( 20 , 000 × g for 30 min ) . Subsequent C . luteolum DUF3570 purification steps were conducted on a Ni-NTA column . Specifically , 4 mL of Ni-NTA slurry ( Nuvia IMAC Resin , 25 mL ) were added into a glass chromatography column ( Econo-Column , 1 . 0 × 10 cm2 ) . The column was prepared with 10 mL of column wash buffer and the samples containing denatured DUF3570 were then loaded . Bound protein was eluted using 5 mL of modified column wash buffer containing 50 , 100 , 200 , or 500 mM of imidazole . To prevent guanidine-HCl from forming precipitates with SDS in following steps , eluted samples were mixed with 100% ethanol ( 9:1 v/v ) and incubated at −20°C for 10 min . After centrifugation at 21 , 100 × g for 5 min and removal of the supernatant , the pelleted protein was washed once with 90% ethanol . The sample was centrifuged again and the pellet was resuspended in diH2O . Protein concentrations of eluted fractions containing His6-tagged DUF3570 were measured as described above and standardized to 1 . 2 mg/mL . Purified and normalized DUF2271 and DUF3570 were loaded and separated on a 12% Bis-Tris gel . Prior to gel staining , flavinylated bands were visualized under UV due to the UV resonance of the flavin molecule . Visualizations of Coomassie blue stained protein were captured with an iBright 1500 gel imager .	In bacteria , certain chemical reactions required for life do not take place directly inside the cells . For instance , ‘redox’ reactions essential to gather minerals , repair proteins and obtain energy are localised in the membranes and space that surround a bacterium . These chemical reactions involve electrons being transferred from one molecule to another in a cascade that connects the exterior of a cell to its internal space . The enzyme ApbE allows proteins to perform electron transfer by equipping them with ring-like compounds called flavins , through a process known as flavinylation . Yet , the prevelance of flavinylation in bacteria and the scope of redox reactions it facilitates has remained unclear . To investigate this question , Méheust , Huang et al . analysed over 30 , 000 bacterial genomes , finding genes essential for ApbE flavinylation in about half of all bacterial species across the tree of life . The role of ApbE-flavinylated proteins was then deciphered using a ‘guilt by association’ approach . In bacteria , genes that perform similar roles are often close to each other in the genome , which helps to infer the function of a protein coded by a specific gene . This approach revealed that flavinylation is involved in processes that allow bacteria to acquire iron and to use various energy sources . A number of interesting proteins were also identified , including a group that carry multiple flavins , and could therefore , in theory , transfer electrons over long distances . This discovery could be relevant to bioelectronic applications , which are already considering another class of bacterial electron-carrying molecules as candidates to form minuscule electric wires .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2021	Post-translational flavinylation is associated with diverse extracytosolic redox functionalities throughout bacterial life
Exosomes are small vesicles that are secreted from metazoan cells and may convey selected membrane proteins and small RNAs to target cells for the control of cell migration , development and metastasis . To study the mechanisms of RNA packaging into exosomes , we devised a purification scheme based on the membrane marker CD63 to isolate a single exosome species secreted from HEK293T cells . Using immunoisolated CD63-containing exosomes we identified a set of miRNAs that are highly enriched with respect to their cellular levels . To explore the biochemical requirements for exosome biogenesis and RNA packaging , we devised a cell-free reaction that recapitulates the species-selective enclosure of miR-223 in isolated membranes supplemented with cytosol . We found that the RNA-binding protein Y-box protein I ( YBX1 ) binds to and is required for the sorting of miR-223 in the cell-free reaction . Furthermore , YBX1 serves an important role in the secretion of miRNAs in exosomes by HEK293T cells . In contrast to the normal pathways of protein secretion , the processes by which unconventional cargoes are secreted have proved diverse and enigmatic . Indeed , our understanding of unconventional secretory mechanisms is limited to a few examples of leader-less soluble and transmembrane proteins ( Malhotra , 2013 ) . Unconventionally secreted molecules may be externalized in a soluble form by translocation across various membranes . This may include direct translocation across the plasma membrane , or across an organelle membrane followed by fusion of the organelle with the plasma membrane ( Zhang and Schekman , 2013 ) . Alternatively , proteins and RNAs can be secreted within vesicles that bud from the plasma membrane , as in the budding of enveloped viruses such as HIV , or within vesicles internalized into a multivesicular body ( MVB ) that fuses with the plasma membrane ( Colombo et al . , 2014 ) . RNA is actively secreted into the medium of cultured cells and can be found in all bodily fluids enclosed within vesicles or bound up in ribonucleoprotin complexes , both forms of which are resistant to exogenous ribonuclease ( Colombo et al . , 2014; Arroyo et al . , 2011; Mitchell et al . , 2008 ) . Importantly , extracellular vesicle-bound RNAs appear to be enriched in specific classes of RNAs , including small RNAs and microRNA ( miRNA ) ( Skog et al . , 2008; Valadi et al . , 2007; Kosaka et al . , 2010 ) . Exosomes are a subclass of extracellular vesicle which can be defined as 30–100 nm vesicles with a buoyant density of ~1 . 10–1 . 19 g/ml that are enriched in specific biochemical markers , including tetraspanin proteins ( Colombo et al . , 2014 ) . It is often assumed that vesicles fitting this description are derived from the multivesicular body , but some evidence suggests that physically and biochemically indistinguishable vesicles bud directly from the plasma membrane ( Booth et al . , 2006 ) . Numerous studies have reported the presence of RNAs , especially miRNAs , from fractions containing exosomes , though many of these studies have relied on isolation techniques ( e . g . high speed sedimentation ) that do not resolve vesicles from other cellular debris or RNPs ( Bobrie et al . , 2012 ) . Thus , it is difficult to know in which form RNAs are secreted and even more challenging to determine which miRNAs may be specifically secreted as exosome cargo . The use of many different cell lines , bodily fluids and isolation methods to identify which miRNAs are specifically packaged into exosomes further complicates the establishment of widely accepted exosomal miRNA cargo . Even with the crude preparations that have been characterized , it is clear that RNA profiles from exosomes are distinct from those of the producer cells . Thus RNA capture or stabilization in exosomes is likely to occur through a selective sorting mechanism . RNA packaging may occur by specific interactions with RNA binding proteins that engage the machinery necessary for membrane invagination into the interior of an MVB or by interaction of RNAs with lipid raft microdomains from which exosomes may be derived ( Janas et al . , 2015 ) . In order to probe the mechanism of exosome biogenesis , we developed procedures to refine the analysis of RNA sorting into exosomes . Using traditional means of membrane fractionation and immunoisolation , we identified unique miRNAs highly enriched in exosomes marked by their content of CD63 . This miRNA sorting process was then reproduced with a cell-free reaction reconstituted to measure the packaging of exosome-specific miRNAs into vesicles formed in incubations containing crude membrane and cytosol fractions . Among the requirements for miRNA sorting in vitro , we found one RNA-binding protein , YBX1 , which is a known constituent of exosomes secreted from intact cells ( Ung et al . , 2014; Buschow et al . , 2010 ) . We first sought to purify exosomes from other extracellular vesicles and contaminating particles containing RNA ( e . g . aggregates , ribonucleoprotein complexes ) that sediment at high speed . We define exosomes as ~30–100 nm vesicles with a density of 1 . 08–1 . 18 g/ml and containing the tetraspanin protein CD63 . Based on these criteria , purified exosomes were recovered using a three-stage purification procedure ( Figure 1a ) . First , large contaminating cellular debris was removed during low and medium speed centrifugation and exosomes were concentrated by high-speed sedimentation from conditioned medium . Next , to eliminate non-vesicle contaminants , the high-speed pellet fraction was suspended in 60% sucrose buffer and overlaid with layers of lower concentrations of sucrose buffer followed by centrifugation to float vesicles to an interface between 20 and 40% sucrose . Analysis of this partially purified material by electron microscopy showed vesicles of the expected size and morphology with fewer profiles of larger ( >200 nm ) membranes and reduced appearance of protein aggregates ( Figure 1b , c compared to Figure 1d , e ) . Finally , sucrose gradient fractions were mixed with CD63 antibody-immobilized beads to recover vesicles enriched in this exosome marker protein . 10 . 7554/eLife . 19276 . 003Figure 1 . Purified CD63-positive exosomes contain RNA . ( a ) Exosome purification schematic . ( b–e ) Representative electron micrographs of negative stained samples from the 100 , 000 ×g pellet fraction ( b , d ) and post-flotation fractions ( c , e ) at either 9300X ( a , b ) or 1900X ( c , e ) magnification . Open arrows indicate large ( >200 nm ) vesicle contaminants and closed arrows indicate protein aggregates . ( f ) CD63-luciferase activity in purified exosomes after treatment with 1% Triton X-100 ( TX-100 ) and/or 100 µg/ml trypsin for 30 min at 4°C . Error bars represent standard deviations from 3 independent samples . ( g ) Specific activity of CD63-luciferase ( RLU/µg of total protein ) at each stage of purification ( green: 100 , 000 ×g pellet , purple: post-flotation , red: post-immunoisolation a-CD63 beads ) . ( h ) Total RNA recovered from conditioned medium after immuno-isolation with a-CD63 or an IgG control . B – bound to beads , FT – flow-through not bound to beads . Error bars represent standard deviations from 3 separate purifications ( biological replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 00310 . 7554/eLife . 19276 . 004Figure 1—figure supplement 1 . Sub-cellular localization of C-terminal CD63-luciferase-FLAG fusion . CD63-luciferase-FLAG cells induced for 48 hr were fixed with 4% paraformaldehyde blocked with 5% BSA in PBS and stained with M2-Flag antibody ( 1:500 ) and then Alexa-488 conjugated anti-mouse secondary . Cells were mounted with prolong gold ( containing DAPI stain ) and imaged at 400 X total magnification . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 004 To monitor and quantify the exosome purification , we generated a stable , inducible HEK293 cell line expressing a CD63-luciferase fusion . Tetraspanin proteins share a common topology in which the amino- and carboxyl-termini face the cytoplasm resulting in a predicted orientation inside the lumen of an exosomal vesicle . Although an intact C-terminal sequence is reported to be required for the proper localization of CD63 to the cell surface ( Rous et al . , 2002 ) , we found that our overexpressed CD63-luciferase fusion was localized to a variety of cell surface and intracellular membranes ( Figure 1—figure supplement 1 ) . Using isolated exosome fractions , we confirmed that the CD63-luciferase fusion maintained the expected topology . Luciferase activity was stimulated by the addition of detergent to disrupt the membrane and allow access to the membrane impermeable substrates luciferin and ATP , and to trypsin , which inactivated luciferase activity in the presence but not in the absence of detergent ( Figure 1f ) . The CD63-luciferase cell line was then used to monitor exosome purification . CD63-luciferase specific activity increased at each step of the purification , yielding a 5-fold purification of exosomes from the starting 100 , 000 ×g pellet ( Figure 1g ) . Additionally , following immunoisolation , most of the RNA was found in the CD63 positive bound ( B ) fraction , showing that RNA is associated with purified exosomes from HEK293T cells ( Figure 1f ) . These results established that the RNA is associated with CD63-containing exosomes , but not necessarily enclosed within exosomes . Previous reports indicated the presence of miRNAs in fractions containing exosomes but which also contain contaminating particles ( Skog et al . , 2008; Valadi et al . , 2007; Bobrie et al . , 2012 ) . To identify the specific miRNAs that are enriched in CD63-positive exosomes from 293T cells , we performed Illumina-based small RNA sequencing on libraries prepared from purified exosomes and from cells . We obtained a total of 123 , 679 miRNA reads ( 4 . 4% of total mapped reads – Figure 2—source data 1 ) in the exosome library representing 502 distinct miRNAs and 880 , 093 reads ( 7 . 3% of total mapped reads – Figure 2—source data 1 ) representing 637 miRNAs in the cell library ( Figure 2a ) . To determine if a particular miRNA species was over-represented in exosomes , we analyzed the datasets for reads mapping to miRNA precursors and the targeting or passenger strand of mature miRNAs ( Figure 2b ) . Exosomes were slightly enriched in reads mapping to precursor and passenger strand transcripts , however , the vast majority of miRNAs ( 91% from cells and 88% from exosomes ) mapped to the mature targeting strand . The relative abundance of each miRNA was estimated by normalizing to the total number of miRNA-mapped reads ( i . e . the number of reads mapped to a miRNA locus divided by the total number of miRNA mapped reads for each dataset – RPM ) . Of these , 134 and 269 miRNAs were uniquely found in the exosome and cell datasets respectively ( Figure 2a ) . Most of the miRNAs uniquely found in exosomes were of very low abundance , with only a few counts for each miRNA . A notable exception was miR-223-3p , which was in the 72nd percentile for normalized reads in exosomes ( Figure 2c , d – red ) . The relatively high abundance of miR-223 in exosomes and its low level in the cellular library indicated that miR-223 was very efficiently packaged and secreted via exosomes . 10 . 7554/eLife . 19276 . 005Figure 2 . Enrichment of select miRNAs in exosomes . ( a ) Venn diagram showing the number of total ( above diagram ) , unique ( inside red or green circles ) and shared miRNAs ( inside yellow ) from each library . ( b ) Pie charts showing the relative proportion of reads mapping to each miRNA species ( Precursor - red , passenger strand – green , targeting strand - blue ) in cellular and exosome small RNA libraries ( c ) Scatterplot showing the enrichment ( reads per million miRNA mapped reads ( RPM ) in exosomes/cells ) and relative abundance in exosomes ( RPM ) of all miRNAs found in both libraries . ( d ) Table showing the enrichment and abundance ( RPM ) of relevant miRNAs in exosomes . ( Red - unique to exosomes , Yellow - highly enriched in exosomes , Green - unique to cells ) ( e , f ) Relative miR-223 ( e ) and miR-144 ( f ) per ng of RNA as quantified by qRT-PCR during each stage of the purification . The 100K pellet was set to 1 . ( g ) RNase protection of exosomal miRNAs quantifed by qRT-PCR . Purified exosomes treated with or without RNase I and/or Triton X-100 . Errors bars represent the standard deviation from 3 biological replicates . Statistical significance was performed using Student's t-test ( *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 00510 . 7554/eLife . 19276 . 006Figure 2—source data 1 . Mapping statistics for small RNA-seq libraries to the human genome ( hg19 ) . Reads were processed ( see Materials and methods ) and mapped to the human genome ( hg19 ) using Bowtie 2 . Total counts for reads mapped to the genome , to rRNA and to miRNA ( using miRdeep2 - see Materials and methods ) are shown . Percent of total reads are shown in parenthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 00610 . 7554/eLife . 19276 . 007Figure 2—source data 2 . miRNA counts from cell and exosomes libraries using miRdeep2 . Number of reads mapped to each miRNA annotated in miRBase version 21 using the quantifier program of the miRdeep2 package . Reads per million miRNA mapped reads ( RPM ) were calculated and the quotient was taken to determine enrichment in exosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 00710 . 7554/eLife . 19276 . 008Figure 2—figure supplement 1 . Read frequency distribution along miR-223 and miR-144 precursors . All reads detected for miR-223 ( a ) and miR-144 ( b ) in the exosome small RNA-seq libraries were aligned to the hairpin precursor sequences and the read frequency distribution across the precursor was determined using the quantifier program of the miRdeep2 package . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 008 We also identified miRNAs that were found in both libraries but were highly enriched in exosomes . A total of 368 miRNAs were detected in both the exosome and cell libraries . Most of these miRNAs were more enriched in cells than exosomes; however , some ( e . g . miR144-3p , miR150-5p , miR142-3p ) were highly enriched in exosomes and , like miR223 , were moderately abundant in exosomes ( Figure 2c , d - yellow ) . To summarize , small RNA sequencing from purified exosomes and subsequent miRNA analysis identified several putative exosomal miRNAs . We selected miR-223 and miR-144 from the group of unique and enriched miRNAs for further analysis , as these were the most abundant species with documented functions that mature by the normal pathway of miRNA biogenesis . An analysis of the read frequency distribution of miR-223 and miR-144 from the exosome small RNA-seq dataset showed that the vast majority of reads mapped to the mature guide strand with few reads also mapping to the passenger strand ( Figure 2—figure supplement 1 ) . We performed quantitative reverse transcriptase PCR ( qRT-PCR ) for each miRNA target during the course of exosome purification from conditioned medium . Our results showed a selective enrichment of both miR-223 and miR-144 at each stage of the purification ( Figure 2e and f ) . Thus , these miRNAs are associated with CD63 exosomes . To determine if these exosome associated RNAs are contained within exosomes and not simply bound to the surface , we performed an RNase protection experiment . Both miRNAs were protected from RNase I digestion , unless detergent was added to disrupt the membrane ( Figure 2g ) . These results confirm that miRNAs are selectively packaged into exosomes purified from HEK293T conditioned media and establish miR-223 and miR-144 as specific exosomal miRNAs . Having shown that specific miRNAs are enriched in exosomes produced in vivo , we next determined if selective sorting of miRNAs into vesicles could be reconstituted in vitro . We used the cell-free packaging assay to compare the efficiency of incorporation of synthetic miR-223 and a relatively abundant cellular miRNA that is not found in exosomes ( miR-190a-5p – miR-190 ) . Exosomal miR-223 was more efficiently packaged into vesicles ( 9% ) than cellular miR-190 ( 1 . 5% ) ( Figure 4e ) . Furthermore , the rate of miR-223 packaging mirrored the rate at which luciferase became sequestered in the biogenesis reaction whereas the rate of miR-190 protection in a 30°C incubation reflected the low rate of formation of sequestered luciferase in an incubation held on ice ( Figure 4f ) . Based on these experiments , we conclude that the cell-free packaging assay reconstitutes the selective sorting of exosomal miR-223 over cellular miR-190 into vesicles , possibly exosomes , formed in vitro . To identify proteins that may be involved in miRNA packaging into exosomes , we employed a proteomics approach utilizing the in vitro packaging assay to capture RNA binding proteins . MiRNA sorting may require an RNA binding protein to segregate an RNP into a nascent budded vesicle . Synthetic 3’ biotinylated miR-223 was substituted for unmodified miRNA in the cell-free reaction . Samples were treated with RNase , quenched with RNase inhibitor and solubilized with Triton X-100 . miR-223-biotin was captured on streptavidin-coated beads and interacting proteins were eluted with high salt buffer . Mir-223-interacting proteins were identified by in-solution liquid chromatography/mass spectrometry ( Figure 5a ) . Based on peptide count and coverage , the most highly represented protein was Y-box binding protein I ( YBX1 ) ( Figure 5b ) . Peptides representing >45% YBX1 of the protein were identified stretching from the cold shock domain to the C-terminus ( Figure 5c ) . 10 . 7554/eLife . 19276 . 011Figure 5 . Identification of YBX1 as a candidate exosomal miRNA sorting protein . ( a ) Scheme to identify candidate miRNA sorting proteins ( b ) Proteins identified by tandem mass spectroscopy from the experiment illustrated in ( a ) . ( c ) Schematic of YBX1 protein . The cold-shock domain ( green ) and positively charged low-complexity region ( blue ) are highlighted . Red lines indicate detected unique peptides from mass spectroscopy . ( d ) Immunoblots for the indicated protein markers in the CD63 immuno-isolated ( bound ) or unbound fractions . Exosomes were purified as in Figure 1a . ( e ) Immunoblot for YBX1 following cell-free packaging reactions performed according to the conditions indicated and immobilized with streptavidin beads as shown in ( Figure 5a ) . Bar graph represents densitometry values for the blot shown . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 011 YBX1 is a multi-functional RNA binding protein that shuttles between the nucleus , where it plays a role and splice site selection ( Wang et al . , 2013; Wei et al . , 2012 ) , and the cytoplasm where it is required for the recruitment of RNAs into cytoplasmic ribonucleoprotein granules containing untranslated mRNAs and plays a role in mRNA stability ( Lyabin et al . , 2014 ) . YBX1 also co-localizes with cytoplasmic P-bodies containing members of the RISC complex , including GW182 which can be found in exosomes ( Goodier et al . , 2007; Gallois-Montbrun et al . , 2007 ) . Interestingly , YBX1 is secreted in a form that resists trypsin in the absence but not in the presence of a non-ionic detergent ( Triton X-100 ) , consistent with a location in vesicles , perhaps exosomes ( Frye et al . , 2009; Rauen et al . , 2009 ) . Furthermore , YBX1 has been detected by mass spectrometry in isolated exosomes ( Ung et al . , 2014; Buschow et al . , 2010 ) . We first determined if YBX1 co-purifies with exosomes . We purified exosomes as in Figure 1a and found that YBX1 was primarily associated with the CD63-bound fraction containing known exosome markers ( TSG101 , Alix , CD9 ) , as opposed to flotillin 2 , which was predominantly found in the unbound fraction ( Figure 5d ) . The CD63 positive ( exosome ) fraction contained most of the RNA ( Figure 1f ) . These results show that HEK293T cells release at least two vesicle types ( CD63 positive and negative ) . We next examined the biochemical requirements for co-packaging of miR-223 and YBX1 using the biotin-miR-223 packaging reaction described in Figure 5a . An immunoblot showed YBX1 bound to biotin-mi223 was recovered in exosomes in a complete reaction , while no detectable YBX1 was recovered in exosomes in incubations that lacked cytosol or membranes and much reduced signals in control incubations held at 4C or conducted in the absence of biotin-mi223 ( Figure 5e ) . These conditions mirror those required for the packaging of miR-223 in our cell-free reaction . Because YBX1 is a known RNA binding protein , is secreted by cells in exosomes and physically interacts with miR-223 during the in vitro packaging assay , it met our criteria for a potential exosomal miRNA sorting factor . Given that mature miRNA guide sequences in the cell are bound to an argonaute family protein , we were surprised that we did not detect any of these proteins associated with miR-223 in the streptavidin-bound fraction . Argonaute proteins have been variously described as being inside of extracellular vesicles ( Arroyo et al . , 2011; Turchinovich et al . , 2011 ) or released as free proteins independent of vesicles ( Squadrito et al . , 2014; Gibbings et al . , 2009 ) . We first performed immunoblot on 100 , 000 ×g pellet fractions ( which should contain vesicle associated and non-vesicle associated Ago2 ) and vesicle fractions purified on a buoyant density gradient ( which should eliminate non-vesicle Ago2 ) to determine if we could detect Ago2 in fractions that contain exosomes . As expected , after flotation the exosome marker proteins were enriched compared to the pellet fraction ( Figure 6a ) . In contrast , Ago2 was not detected in the density gradient purified vesicle fraction ( Figure 6a ) . The apparent lack of Ago2 in vesicles could be due to its absence in vesicles or a relatively low chemical abundance relative to other molecules . Regardless , our results show that in HEK293T cells the large majority of extracellular Ago2 exists as a non-vesicle associated species . These results support previously published evidence that Ago is not associated with density gradient purified exosomes isolated from breast cancer ( MCF7 ) cells ( Van Deun et al . , 2014 ) . 10 . 7554/eLife . 19276 . 012Figure 6 . Lack of evidence for a specific role for Ago2 in sorting miR-223 into exosomes . ( a ) Immunoblots for Ago2 and exosome markers TSG101 and CD9 in 100 , 000 ×g ( 100K ) pellet and the 20/40% sucrose interface fractions . ( b ) Schematics showing 3' and internally biotinylated miR-223 duplex and mature guide strand substrates . ( c ) Immunoblots for Ago2 from substrates mixed with cytosol alone for 30 min at 30°C and then absorbed on streptavidin-conjugated beads . ( d ) Percent protected miR-223 ( either guide strand or passenger strand ) from 3' biotinylated or internally biotinylated single strand or duplex substrates . ( e ) Immunoblots for Ago2 and YBX1 from substrates packaged in the complete in vitro reaction and then absorbed on streptavidin-conjugated beads . ( f ) Percent RNAse protected miR-223 and relative level of streptavidin-absorbed YBX1 protein ( normalized to duplex ) . MiR-223 and YBX1 quantification comes from data in d ) and e ) , respectively . All quantifications represent means from 3 independent samples and errors bars represent standard deviations . All quantifications represent means from three independent experiments and error bars represent standard deviations . Statistical significance was performed using Student's t-test ( p<0 . 05 , *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 012 To further investigate the role of Ago2 in sorting miR-223 into exosomes , we employed the cell-free packaging assay . The inability to detect Ago2 in our mass spectrometry results from in vitro packaged miR-223 could be due to technical limitations of our reaction . We considered two potential technical explanations . First , we used the single stranded guide RNA in our reaction rather than a duplex RNA molecule . Although Ago2 associates with guide RNA in a reaction containing purified components ( MacRae et al . , 2008; Gagnon et al . , 2014 ) , single-stranded miRNA may be rapidly degraded in a crude extract ( Gagnon et al . , 2014 ) . Thus , it was possible that the duplex molecule might represent a more relevant packaging substrate . Second , 3' biotinylated RNA was used as the substrate for the packaging reaction . A recent report suggested that , in some cases , 3' biotinylated RNA could prevent the proper loading of the miRNA into complex with Ago2 ( Guo and Steitz , 2014 ) . To address these concerns we synthesized miR-223 passenger strand and guide strands biotinylated at the 3' end or internally at position 13 ( Figure 6b ) . The internal position was chosen because in crystal structures of miRNA-protein complexes , middle positions of guide RNAs do not appear to be in direct contact with Ago2 ( Guo and Steitz , 2014; Elkayam et al . , 2012; Schirle and MacRae , 2012 ) . We then annealed the guide and passenger strands to form the miR-223 duplexes . To determine if the guide or passenger strands can be efficiently loaded into Ago2 in our in vitro reaction conditions , we first mixed the biotinylated substrates with cytosol alone and evaluated complexes that associated with streptavidin beads . In our reaction conditions , both the single stranded guide and duplex oligonucleotides bound apparently equally to Ago2 in cytosol alone and there was no discernible difference in the association comparing 3' or internally biotinylated oligonucleotides ( Figure 6c ) . We then tested the various substrates in our complete packaging reaction including membranes , cytosol and an ATP regenerating system . Duplex substrates were packaged ~2-fold more efficiently than the single stranded guide RNA , irrespective of the position of the biotin group ( Figure 6d ) . Interestingly , in reactions programmed with duplex RNA substrate , only the guide RNA and not the passenger RNA was sorted into a protected compartment ( Figure 6d ) . In similar incubations , the YBX1 protein was ~2X more efficiently packaged in reactions programmed with duplex RNA but Ago2 was not detected associated with miR-223 in any of the complete reactions ( Figure 6e , f ) . This suggests that whereas the RNA substrates are capable of being bound by Ago2 in the cytosol , in the complete reaction containing membranes and ATP , YBX1 is the predominant binding factor . These results explain why Ago2 was not detected in the mass spectrometry data and are consistent with our failure to detect Ago2 in buoyant density-fractionated extracellular vesicles . Similarly , Van Deun et al . ( 2014 ) found no evidence of Ago2 in density gradient-isolated exosomes from MCF7 cells ( Van Deun et al . , 2014 ) . Given the absence of Ago2 in exosomes or associated with miR-223 in our cell-free RNA sorting reaction , we focused on the primary candidate RNA binding protein found in our mass spectrometry results . We next evaluated the requirement for YBX1 in packaging exosomal miRNAs in cells and in the cell-free reaction . To address this question , we generated a YBX1 knockout HEK293T cell line with CRISPR/Cas9 using a guide RNA targeting the YBX1 locus ( Cong et al . , 2013; Jinek et al . , 2013; Mali et al . , 2013 ) . Clones were screened by genomic PCR and immunoblot for YBX1 . We recovered a homozygous mutant clone ( ΔYBX1 ) that had been targeted at the YBX1 locus and no longer expressed YBX1 protein ( Figure 7a ) . The homozygous mutant cells grew normally under the conditions used to propagate HEK293T cells and released an approximately equal number of particles into the medium after 48 hr of growth ( 2 . 38 × 107 and 2 . 42 × 107 particles/ml for wild type and ΔYBX1 ) as determined by Nanosight nanoparticle tracking analysis . 10 . 7554/eLife . 19276 . 013Figure 7 . YBX1 is necessary for exosomal miRNA packaging and secretion . ( a ) Analysis of wild-type and CRISPR/Cas9 genome edited HEK293T clones by PCR flanking the genomic target site ( top ) and immunoblot for YBX1 ( middle ) and GAPDH ( bottom ) . ( b ) In vitro miR-223 packaging into exosomes from ΔYBX1 or WT cytosol transfected with control ( pCAG ) or YBX1 plasmid . ( c ) Cell-free exosome biogenesis with cytosol from ΔYBX1 or WT cells and membranes from CD63-luciferase cells . ( d ) Fold change of miR-223 and miR-144 in cells and media from by ΔYBX1 ( KO ) and WT cells ( KO/WT ) ND = Not detected . All quantifications represent means from three independent experiments and error bars represent standard deviations . Statistical significance was performed using Student's t-test ( p<0 . 05 , p<0 . 01 and NS = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 01310 . 7554/eLife . 19276 . 014Figure 7—figure supplement 1 . Partial redundancy for YBX2 for the secretion of miR-223 in cells . Relative quantity of miR-223 secreted into the medium by WT and ΔYBX1 cells after 24 hr with or without transfection with control or YBX2 siRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 01410 . 7554/eLife . 19276 . 015Figure 7—figure supplement 2 . miR-223 association with Ago2 in WT and ΔYBX1 cells . RNA immunoprecipitation was performed using mouse anti-Ago2 antibody or a mouse IgG isotype control antibody in lysates of WT or ΔYBX1 cells . Ago2-associated miR-223 was detected by qPCR . Statistical analysis was performed using Student's t-test ( p<0 . 01 , NS = not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19276 . 015 To determine if YBX1 was required for miRNA packaging , we prepared cytosol from ΔYBX1 cells and tested miR-223 incorporation using the in vitro packaging assay . Cytosol from ΔYBX1 cells did not support miR-223 protection in vitro but activity was largely restored in reactions containing cytosol from a ΔYBX1 line transfected with plasmid encoding YBX1 ( Figure 7b ) . We also evaluated the role of YBX1 in our biogenesis reaction ( Figure 3a ) and found that cytosol from wt and ΔYBX1 were indistinguishable in the formation of latent luciferase activity ( Figure 7c ) . Thus , YBX1 is required for exosomal miRNA packaging in vitro but is not required for the sorting of an exosome membrane cargo protein into vesicles in our cell-free reaction . In an effort to connect the results of our cell-free reaction to the mechanism of sorting of miRNAs into exosomes secreted by HEK293T cells , we examined the secretion of miR-223 and of another miRNA , miR-144 , which was also highly enriched in our purified exosome fraction ( Figure 2d ) . We measured the amount of miR-144 and miR-223 secreted into the medium and retained in cells by qRT-PCR . ΔYBX1 cells showed a significant decrease in secretion of both miRNAs , though more notably of miR-144 , during a 24 hr incubation in fresh medium ( Figure 7d ) . When the YBX1 paralog YBX2 was knocked down in ΔYBX1 cells , miR-223 secretion was diminished to the baseline level , suggesting partial functional redundancy for YBX paralogs in the secretion of miR-223 ( Figure 7—figure supplement 1 ) . The defect in miR-223 secretion was less substantial in cells compared to the cell-free reaction ( Figure 7b ) , possibly reflecting a rate effect that distinguishes the magnitude of the defect in a 20 min cell-free incubation at 30°C vs . a 24 hr incubation of cells at 37°C . Nonetheless , the secretion defect was accompanied by an ~three–four fold accumulation of each miRNA in cells . No accumulation was observed for another miRNA ( miR-190 ) , which is not released in exosomes ( Figure 7d ) . To determine if Ago2 binds miR-223 that accumulates in ΔYBX1 cells , we performed RNA immunoprecipitation with Ago2 or isotype matched control IgG in lysates of WT and ΔYBX1 cells and quantified the amount of associated miR-223 by qPCR ( Figure 7—figure supplement 2 ) . No miR-223 above background co-immunoprecipitated with Ago2 in WT cells . In contrast , miR-223 accumulated in ΔYBX1 cells was found to be associated with Ago2 . These results suggest that miR-223 is properly processed in HEK293T cells , likely loaded into Ago2 and then efficiently dissociated and bound by YBX1 . Taken together , our results show that YBX1 controls the secretion of select exosomal miRNAs in vitro and in vivo . We find that synthetic miR-223 is sequestered into vesicles more efficiently than miR-190 , consistent with the possibility of a primary RNA sequence or secondary structure , perhaps stabilized by an RNA binding protein such as YBX1 , that directs RNA sorting . One possible sorting motif – GGAG – is enriched in miRNAs secreted in exosomes from T-cells ( Villarroya-Beltri et al . , 2013 ) . This motif is recognized by hRNPA2B1 , a T-cell exosome RNA-binding protein , which requires sumoylation for efficient secretion via exosomes . It was therefore suggested that binding of GGAG containing miRNAs by sumoylated hRNPA2B1 was a sorting mechanism for miRNAs into T-cell-derived exosomes . We were unable to identify any statistically significant primary sequence motifs for miRNAs by either multiple alignment ( ClustalW ) or multiple Em for motif elicitation ( MEME ) in HEK293T-derived exosomes ( Larkin et al . , 2007; Bailey et al . , 2009 ) . Furthermore , the mature targeting strand of miR-223 packaged into exosomes contains no guanine nucleotides and hRNPA2B1 was not detected in our mass spectrometry results for proteins bound to miR-223-biotin isolated from vesicles formed in our cell-free reaction . The human genome encodes more than 1000 experimentally determined and predicted RNA-binding proteins ( Cook et al . , 2011; Gerstberger et al . , 2014 ) . We therefore propose that different cell types may use RNA binding proteins with distinct binding preferences to secrete miRNAs , and perhaps other RNA classes , in exosomes . In addition , some cell types may deploy multiple RNA-binding proteins to sort RNAs into exosomes , in which case motif discovery would be challenging , even in highly purified vesicles , due to diverse motif preferences from distinct proteins . We identified YBX1 as the dominant RNA-binding protein physically interacting with miR-223 in vitro and confirmed its role in miR-223 packaging into exosomes both in vitro and in cultured cells . YBX1 is found within mammalian P-bodies ( GW bodies ) containing untranslated RNAs ( Kedersha and Anderson , 2007 ) and in the nucleus where it plays a role in RNA splice site selection by binding short sequence motifs ( Wang et al . , 2013; Wei et al . , 2012 ) . YBX1 binds RNA via an internal cold shock domain and an inherently disordered , highly charged C-terminus ( Lyabin et al . , 2014 ) . Interestingly , another cold shock domain containing protein , Lin28 , binds pre-miRNAs of the Let7 family via hairpin-loop structures ( Nam et al . , 2011 ) . YBX1 also binds hairpin-loops in a murine retrovirus , leading to stabilization of the viral RNA genome and increased particle production ( Bann et al . , 2014 ) . YBX1 binding of viral RNA also increases production of other retroviruses , including HIV ( Bann et al . , 2014; Mu et al . , 2013; Li et al . , 2012 ) . This raises the possibility that the recognition motif for sorting into exosomes may be based on secondary rather than primary RNA structure and that YBX1 may act as an RNA co-factor to escort exosomal RNAs into exosomes . Our studies focused on miRNAs , however it is possible that YBX1 is responsible for the secretion of other RNA classes in exosomes . Several recent reports indicate a role for YBX1 in binding various small RNAs , including miRNAs , tRNA fragments and snoRNAs ( Blenkiron et al . , 2013; Liu et al . , 2015; Goodarzi et al . , 2015 ) . It is interesting to note that most miRNAs present in exosomes in our study are not highly enriched compared to their relative abundance in cells . This raises the possibility that highly enriched exosomal miRNAs mimic other classes of RNAs that are more efficiently packaged in a YBX1-dependent manner . A surprising finding from our study is the lack of evidence for the argonaute proteins in isolated exosomes or sequestered with miR-223 in our cell-free RNA sorting reaction . This in spite of our observation that cell-free reaction reconstitutes sorting of the mature strand from a duplex RNA . Some recent evidence suggests that Ago2 may be sorted along with miRNAs into exosomes as a result of aberrant KRAS signaling . Analysis of extracellular RNA in isogenic cell lines differing only in KRAS status revealed that secretion of a sub-population of miRNAs is decreased in colorectal cancer cells harboring an activating KRAS mutation whereas other miRNAs are secreted at equivalent levels irrespective of KRAS status ( Cha et al . , 2015 ) . Subsequent experiments showed that KRAS mutation results in phosphorylation of Ago2 causing its re-localization from multivesicular bodies to P-bodies leading to decreased secretion of select miRNAs ( McKenzie et al . , 2016 ) . Notably , the miRNA ( miR-223-3p ) shown here to be dependent on YBX1 is among the cohort of miRNAs that were not affected by KRAS status . These results combined with ours suggest two possible routes for miRNA egress via exosomes , an Ago2-associated pathway and a RNA-binding protein-dependent pathway that we term chaperone-mediated sorting . The chaperone-mediated pathway would include the previously identified hnRNPA2B1 in T-cells ( Villarroya-Beltri et al . , 2013 ) and YBX1 in HEK293T cells . Interestingly , one other highly enriched miRNA ( miR-328-5p ) has been previously shown to conditionally associate with either Ago2 or with the RNA-binding protein hnRNPE2 ( Eiring et al . , 2010 ) , suggesting that these two pathways may not be mutually exclusive . A notable feature of the chaperone-mediated pathway is that the RNA content of exosomes may be manipulated by altering the expression of individual RNA binding proteins involved in RNA export with distinct nucleic acid binding specificities . Further characterization of the chaperone-mediated pathway may then allow for targeted sorting of engineered RNA species into exosomes . The functional role of secreted miRNAs has been a matter of discussion since the first reports of extracellular RNA ( Théry , 2011 ) . Numerous studies have shown that miRNAs can be transferred to neighboring cells in experimental settings ( Skog et al . , 2008; Valadi et al . , 2007; Kosaka et al . , 2010; Cha et al . , 2015; Pegtel et al . , 2010; Mittelbrunn , 2011; Rana et al . , 2013 ) . However , the transfer of miRNAs in biologically significant quantities for function in a physiological context is far from proven . Indeed , a recent study reported a stoichiometry of less than one specific miRNA per exosome , with the caveat that this study characterized crude , high-speed pellet fractions from conditioned medium ( Chevillet et al . , 2014 ) . Functional miR-223 transferred between macrophages and miR-223-containing exosomes can induce macrophage differentiation , however , it has yet to be shown that miR-223 transfer plays a direct role in the differentiation ( Ismail et al . , 2013 ) . Indeed , direct and convincing evidence for a physiological role of miRNAs secreted via exosomes has so far proven elusive . Alternatively , exosomes may be a convenient carrier to purge unnecessary or inhibitory RNAs from cells . A recent report provided evidence for both alternative views with the demonstration that target transcript levels for miRNAs in the cell modulate the abundance of miRNAs in macrophage exosomes , and this in turn dictates which miRNAs are transferred to repress transcripts in recipient cells ( Squadrito et al . , 2014 ) . Because YBX1 and the RISC machinery have both been shown to localize to P-bodies and P-bodies are closely juxtaposed to multivesicular bodies , all of the necessary machinery is poised to efficiently secrete miRNAs in exosomes ( Gibbings et al . , 2009 ) . YBX1 may complex with miRNAs whose mRNA targets are not expressed , and sort them into the intralumenal vesicles of a multivesicular body for export by unconventional secretion . Physiological studies of the function miRNAs that are secreted via the Ago2-associated vs . chaperone-mediated pathways may explain contradictory results for different miRNAs and provide general rules for extracellular miRNA function . HEK293T cells were cultured in DMEM with 10% FBS ( Thermo Fisher Scientific , Waltham , MA ) . HEK293T cell lines were maintained by the UC-Berkeley Cell Culture Facility and were confirmed by short tandem repeat profiling ( STR ) and tested negative for mycoplasma contamination . For exosome production , cells were seeded to ~10% confluency in 150 mm CellBIND tissue culture dishes ( Corning , Corning NY ) containing 30 ml of growth medium and grown to 80% confluency ( ~48 hr ) . We noted that confluency >80% decreased the yield of exosome RNA . Cells grown for exosome production were incubated in exosome-free medium produced by ultracentrifugation at 100 , 000 ×g ( 28 , 000 RPM ) for 18 hr using an SW-28 rotor ( Beckman Coulter , Brea , CA ) in a LE-80 ultracentrifuge ( Beckman Coulter ) . Unless otherwise noted , all chemicals were purchased from Sigma Aldrich ( St . Louis , MO ) . Conditioned medium ( 3 l for small RNA-seq and 420 ml for all other experiments ) was harvested from 80% confluent HEK293T cultured cells . All subsequent manipulations were performed at 4°C . Cells and large debris were removed by centrifugation in a Sorvall R6+ centrifuge ( Thermo Fisher Scientific ) at 1500 ×g for 20 min followed by 10 , 000 ×g for 30 min in 500 ml vessels using a fixed angle FIBERlite F14-6X500y rotor ( Thermo Fisher Scientific ) . The supernatant fraction was then passed through a 0 . 22 µM polystyrene vacuum filter ( Corning ) and centrifuged at ~100 , 000 ×g ( 26 , 500 RPM ) for 1 . 5 hr using two SW-28 rotors . The maximum rotor capacity was 210 ml , thus the small RNA-seq processing required pooling from ~15 independent centrifugations . The pellet material was resuspended by adding 500 µl of phosphate buffered saline , pH 7 . 4 ( PBS ) to the pellet of each tube followed by trituration using a large bore pipette over a 30 min period at 4°C . The resuspended material was washed with ~5 ml of PBS and centrifuged at ~120 , 000 ×g ( 36 , 500 RPM ) in an SW-55 rotor ( Beckman Coulter ) . Washed pellet material was then resuspended in 200 µl PBS as in the first centrifugation step and 1 ml of 60% sucrose buffer ( 20 mM Tris-HCl , pH 7 . 4 , 137 mM NaCl ) was added and mixed with the use of a vortex to mix the sample evenly . The sucrose concentration in the PBS/sucrose mixture was measured by refractometry and , if necessary , additional 60% sucrose buffer as added until the concentration was >50% . Aliquots ( 1 ml ) of 40% , 20% and 0% sucrose buffer were sequentially overlaid and the tubes were centrifuged at ~150 , 000 ×g ( 38 , 500 RPM ) for 16 hr in an SW-55 rotor . The 20/40% interface was harvested , diluted 1:5 with phosphate buffered saline ( pH 7 . 4 ) and 1 µg of rabbit polyclonal anti-CD63 H-193 ( Santa Cruz Biotechnology , Dallas , TX ) was added per liter of original conditioned medium and mixed by rotation for 2 hr at 4°C . Magvigen protein-A/G conjugated magnetic beads ( Nvigen , Sunnyvale , CA ) were then added to the exosome/antibody mixture and mixed by rotation for 2 hr at 4°C . Beads with bound exosomes were washed three times in 1 ml PBS and RNA was extracted using Direct-Zol RNA mini-prep ( Zymo Research , Irvine , CA ) or protein was extracted in 100 µl 1X Laemmli sample buffer and dispersed with the use of a vortex mixer for 2 min . An aliquot ( 4 μl ) of the resuspended 100 , 000 ×g pellet fraction or a sample from the 20/40% interface that was diluted 10-fold with PBS , centrifuged at 100 , 000 ×g in a TLS-55 rotor and then resuspended in 1% glutaraldehyde , was spread onto glow discharged Formvar-coated copper mesh grids ( Electron Microscopy Sciences , Hatfield , PA ) and stained with 2% Uranyl acetate for 2 min . Excess staining solution was blotted off with filter paper . Post drying , grids were imaged at at 120 kV using a Tecnai 12 Transmission Electron Microscope ( FEI , Hillsboro , OR ) housed in the Electron Microscopy Laboratory at UC Berkeley . Conditioned medium ( 1 ml ) from wild type and ΔYBX1 cells was harvested and the supernatant from a 10 , 000 ×g centrifugation was drawn into a 1 ml syringe and inserted into a Nanosight LM10 instrument ( Malvern , UK ) . Particles were tracked for 60 s using Nanosight nanoparticle tracking analysis software . Each sample was analyzed 4 times and the counts were averaged . HEK293 cells expressing doxycycline-inducible CD63-luciferase was generated using the T-REx - 293 cell line according to the manufacturer’s instructions ( Life Technologies , Grand Island , NY ) . The open reading frame for CD63-was amplified from human cell cDNA and firefly luciferase-FLAG was amplified from a plasmid source , both using Phusion DNA Polymerase ( NEB ) . CD63 was fused to luciferase by NotI digestion , ligation and PCR amplification . The CD63-luciferase-FLAG amplicon was then digested and ligated into pcDNA5/FRT/TO ( Life Technologies ) using NdeI and PstI sites . The resulting plasmid was co-transfected with pOG44 ( Life Technologies ) and a stable cell line was selected using hygromycin selection ( 100 µg/ml ) . CD63-luciferase expression was induced with 1 µg/ml doxycycline 48 hr prior to exosome harvesting . Luciferase activity was measured using a Promega Glowmax 20/20 luminometer ( Promega , Madison , WI ) with a signal collection integration time of 1 s . Luciferase reactions contained 50 µl sample , 10 µl 20X luciferase reaction buffer ( 500 mM Tricine , pH 7 . 8 , 100 mM MgSO4 , 2 mM EDTA ) , 10 µl 10 mM D-luciferin dissolved in PBS , 10 µl ATP dissolved in deionized water and 120 µl deionized water . Where indicated , samples were pre-treated with final concentrations of 1% Triton X-100 and/or 100 µg/ml trypsin for 30 min on ice . Total protein concentrations were measured using Pierce BCA protein assay according to the manufacturer's instructions . Exosome and cell lysates were prepared by mixing in lysis buffer ( 10 mM Tris-HCl , pH 7 . 4 , 100 mM NaCl , 0 . 1% sodium dodecyl sulfate , 0 . 5% sodium deoxycholate , 1% Triton X-100 , 10% glycerol ) . Lysates were diluted four-fold with 4X Laemmli sample buffer , heated to 65°C for 5 min and separated on 4–20% acrylamide Tris-Glycine gradient gels ( Life Technologies ) . Proteins were transferred to polyvinylidene difluoride membranes ( EMD Millipore , Darmstadt , Germany ) , blocked with 5% bovine serum albumin in TBST and incubated overnight with primary antibodies . Blots were then washed with TBST , incubated with anti-Rabbit or anti-Mouse secondary antibodies ( GE Healthcare Life Sciences , Pittsbugh , PA ) and detected with ECL-2 reagent ( Thermo Fisher Scientific ) . Primary antibodies used in this study were anti-YBX1 ( Cell Signaling Technology , Danvers , MA ) , anti-GAPDH ( Santa Cruz Biotechnology ) , anti-TSG101 ( Genetex , Irvine , CA ) , anti-CD9 ( Santa Cruz Biotechnology ) , anti-Flotillin 2 ( Abcam , Cambridge , MA ) , anti-Alix ( Santa Cruz Biotechnology ) and anti-Ago2 ( Cell Signaling Technology ) . For quantitative immunoblotting ( in Figure 6 ) , the same procedures were used , but were instead imaged using the LiCOR Odyssey imaging system . RNA was extracted using the Direct-Zol RNA mini-prep and cDNA was synthesized either by oligo-dT priming ( mRNA ) or gene-specific priming ( miRNA ) according to the manufacturer’s instructions . For miRNA , we used Taqman miRNA assays from Life Technologies ( assay numbers: hsa-mir-223-3p: 000526 , hsa-mir-190a-5p: 000489 and hsa-miR-144-3p: 002676 ) . Because there is no well-accepted endogenous control transcript for exosomes , relative quantification was performed from equal amounts of total RNA . Qubit ( Thermo Fisher Scientific ) , was used to quantify total RNA from the medium or cells: 10 ng of RNA was reverse transcribed and qPCR was performed according to manufacturer’s instructions . Relative quantification was calculated from the expression 2^- ( Ctcontrol-Ctexperimental ) . Taqman qPCR master mix with no amperase UNG was obtained from Life Technologies and quantitative real-time PCR was performed using an ABI-7900 real-time PCR system ( Life Technologies ) . The in vitro packaging assay was performed as described above with miR-223-3p with biotin linked either to the 3’ phosphate or internally biotinylated at position 13 ( IDT ) . Samples were heated to 65°C for 20 min to inactivate RNase If and then mixed with 4 . 4 μl 10% Triton X-100 for a final concentration of 1% and kept on ice for 30 min . Novagen MagPrep Streptavidin-coated beads ( 10 µl/reaction ) ( EMD Millipore ) were washed 3 times with 1 ml PBS and then added to the reaction lysate . The suspension was mixed by rotation for 2 hr at 4°C , the beads were immobilized using a magnet and washed 3 times with 1 ml PBS . Proteins were eluted from bead-bound miR-223 with 50 µl 1 M KCl . In-solution liquid chromatography and mass spectrometry were performed according to standard procedures by the Vincent J . Coates Proteomics/Mass Spectroscopy laboratory ( UC Berkeley ) . RNA was prepared from cells and 3 l of HEK293T conditioned media . Sequencing libraries were generated using the Scriptminer Small RNA sequencing kit ( Epicentre Biotechnologies , Madison , WI ) from 1 µg total RNA from cells and 200 ng total RNA from exosomes according to the manufacturer’s protocol . The libraries were amplified and index barcodes were added by 11 cycles of PCR . Libraries were sequenced by 50 bp single read massively parallel sequencing on an Illumina Hi-Seq 2000 System at the Vincent J . Coates Genomic Sequencing Laboratory ( UC Berkeley ) . Preprocessing of the 50 base pair single reads was filtered for read quality ( read quality >20 and percent bases in sequence that must have quality >90 ) and adaptor sequences were clipped using the FASTX toolkit ( http://hannonlab . cshl . edu/fastx_toolkit/ ) implementation on the GALAXY platform ( usegalaxy . org ) ( Blankenberg , 2010; Goecks et al . , 2010; Giardine et al . , 2005 ) . Sequences were mapped to miRbase using miRdeep2 and counts tables were obtained using the quantifier program using default settings ( Friedländer et al . , 2012 ) . Reads were normalized by dividing the number of reads mapping to each miRNA by the number of total reads mapping to all miRNAs and the quotient was then multiplied by one million ( reads per million miRNA mapped reads - RPM ) . To analyze miRNA species , we used the quantifier program of the miRdeep2 software suite ( Friedländer et al . , 2012 ) . Precursor reads were determined by subtracting the number of reads mapping to mature ( either targeting or passenger strand sequences ) from the total number of reads mapping to the full-length precursor transcript for each miRNA . Those miRNAs with described passenger strands ( star strand ) were then analyzed to determine how many mature reads mapped to either targeting or passenger strands . A pX330-based plasmid expressing venus fluorescent protein was kindly provided by Robert Tjian ( Cong et al . , 2013 ) . A CRISPR guide RNA targeting the first exon of the YBX1 open reading frame was selected using the CRISPR design tool ( Hsu et al . , 2013 ) . The YBX1 guide RNA was introduced into pX330-Venus by oligonucleotide cloning as described ( Cong et al . , 2013 ) . HEK293T cells were transfected for 48 hr at low passage number , trypsinized and sorted for single , venus positive cells in a 96 well plate using a BD Influx cell sorter . Wells containing single clones ( 16 clones ) were allowed to expand and were screened by semi-nested PCR using primers targeting the genomic region flanking the guide RNA site . Primers for the first round of PCR ( 10 cycles ) were: YBX1-F1 ( GGTTGTAGGTCGACTGAATTA ) and YBX1-R1 ( ACCGATGACCTTCTTGTCC ) . The PCR primers from the first round were removed using DNA clean and concentrator-5 kit ( Zymo Research ) according to manufacturer's instructions and the second round of PCR ( 25 cycles ) was performed with primers: YBX1-F2 ( CGGCCTAGTTACCATCACA ) and YBX1-R1 ( ACCGATGACCTTCTTGTCC ) . PCR products were separated on a 2 . 5% agarose gel to identify products smaller than the wild type PCR product , indicating a deletion . Clones ( 8 ) showing homozygous or heterozygous mutations were then screened by immunoblot to identify those that did not express YBX1 . A clone containing a single homozygous mutation at the target site and not expressing YBX1 by immunoblot was recovered and designated ΔYBX1 . Predesigned siRNA oligos targeting YBX2 were obtained from Qiagen ( Hs_YBX2_3:AAGCCGGTGCTGGCAATCCA ) . Cells were seeded at 60% confluency and siRNA was transfected using Lipofectamine 2000 reagent ( Life Technologies ) . After 48 hr the media was replaced with exosome depleted media and allowed to incubate another 24 hr . An aliquot ( 1 ml ) of medium was removed and centrifuged at 1500 and 15 , 000 ×g and then extracted using the Zymo RNA Prep Kit . 1 ng of RNA was reversed transcribed and qPCR was performed as described above . RNA immunoprecipitation was performed using the Magna RIP kit ( Millipore ) according to the manufacturer's instructions . Mouse monoclonal anti-Ago antibody ( 5 µg , clone 2A8 ) purchased from Sigma-Aldrich was used to immunoprecipitate from a lysate generated from 2 . 0 × 107 HEK293T cells ( WT or ΔYBX1 ) in 100 µl Magna RIP lysis buffer .	Human cells release molecules into their surroundings via membrane-bound packets called exosomes . These molecules can then circulate throughout the body and are protected from degradation . Among the cargos carried by exosomes are small molecules of RNA known as microRNAs , which are involved in regulating gene activity . Only a select subset of the hundreds of microRNAs in a human cell end up packaged into exosomes . This suggests that there might be a specific mechanism that sorts those microRNAs that are destined for export . However , few proteins or other factors that might be involved in this sorting process had been identified to date . Shurtleff et al . set out to identify these factors and started by purifying exosomes from human cells grown in the laboratory and looking for microRNAs that were more abundant in the exosomes than the cells . One exosome-specific microRNA , called miR-223 , was further studied via a test-tube based system that uses extracts from cells rather than cells themselves . These experiments confirmed that miR-223 is selectively packed into exosomes that formed in the test tube . Using this system , Shurtleff et al . then isolated a protein called Y-box Protein I ( or YBX1 for short ) that binds to RNA molecules and found that it was required for this selective packaging . YBX1 is known to be a constituent of exosomes released from intact cells and may therefore be required to sort other RNA molecules into exosomes . Future studies will explore how YBX1 recognizes those RNA molecules to be exported from cells via exosomes . Also , because exosomes have been implicated in some diseases such as cancer , it will be important to explore what role exosome-specific microRNAs play in both health and disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cancer", "biology" ]	2016	Y-box protein 1 is required to sort microRNAs into exosomes in cells and in a cell-free reaction
How host and microbial factors combine to structure gut microbial communities remains incompletely understood . Redox potential is an important environmental feature affected by both host and microbial actions . We assessed how antibiotics , which can impact host and microbial function , change redox state and how this contributes to post-antibiotic succession . We showed gut redox potential increased within hours of an antibiotic dose in mice . Host and microbial functioning changed under treatment , but shifts in redox potentials could be attributed specifically to bacterial suppression in a host-free ex vivo human gut microbiota model . Redox dynamics were linked to blooms of the bacterial family Enterobacteriaceae . Ecological succession to pre-treatment composition was associated with recovery of gut redox , but also required dispersal from unaffected gut communities . As bacterial competition for electron acceptors can be a key ecological factor structuring gut communities , these results support the potential for manipulating gut microbiota through managing bacterial respiration . Mammalian gut microbial communities are likely to be structured by both host- and microbial-associated factors . Extensive research to date has focused on how host factors like diet ( David et al . , 2014; Claesson et al . , 2012 ) , genetics ( Goodrich et al . , 2014 ) , geography ( De Filippo et al . , 2010; Yatsunenko et al . , 2012; Arumugam et al . , 2011 ) , and immune state ( Hooper et al . , 2012 ) shape the gut microbiota . Yet , work in free-living microbial systems reveals that bacteria typically play active roles in shaping their own environment ( Shi and Norton , 2000; Gobbetti , 1998; Goddard , 2008; Osono , 2005; Rui et al . , 2009; Gerbersdorf et al . , 2009 ) . Identifying how these drivers interact is necessary both for a more complete understanding of the gut microbiota and for developing rational interventions . If ecological forces prove important , then compositional changes will likely be the result of feedbacks between the microbes and their environment . Redox potential , a metric of the environmental capacity for reducing chemical reactions ( i . e . , those involving the gain of electrons ) to occur , is a composite measurement of various factors that influence gut microbiota structure ( Cowley et al . , 2015; Friedman et al . , 2017; Dhall et al . , 2014 ) . Much of our knowledge for how gut redox potentials are determined involves host-associated pathways ( Spees et al . , 2013; Rivera-Chávez et al . , 2016 ) . Passive diffusion of oxygen from the epithelium increases redox potential and stimulates growth of aerobic microbes ( Espey , 2013; Albenberg et al . , 2014 ) . Secretion of redox-active immune molecules such as reactive oxygen species or nitrate is known to be a feature of inflammation that imposes oxidative stress on commensal microbes and can be exploited by select pathogens to colonize the intestine ( Winter et al . , 2010; 2013; Faber et al . , 2016; Spees et al . , 2013; Rivera-Chávez et al . , 2016; Kelly et al . , 2015; David et al . , 2015 ) . Maintenance of redox homeostasis in host tissue can also have spillover effects on luminal redox state ( Circu and Aw , 2011 ) . Yet , redox potential is likely to also be shaped by microbial metabolism . In free-living microbial communities , variation in available electron acceptors dictates where microbes that employ respiration can thrive; the differential microbial metabolism that follows can produce further changes in electron acceptor availability ( Morris and Schmidt , 2013; Chen et al . , 2017; Noll et al . , 2005; Orcutt et al . , 2011 ) . Moreover , microbial metabolism has been proposed as a mechanism for low redox potential states in the lung of cystic fibrosis patients ( Cowley et al . , 2015 ) . Here , we investigated the nature of redox potential dynamics under antibiotic treatment to assess the importance of host and microbial processes in structuring gut bacterial communities . Antibiotics directly disturb the microbiota but are also expected to alter host biology related to redox potential . Specifically , antibiotics have been found to increase gut epithelium oxygenation as a result of altered microbial composition and metabolic signaling to the host ( Kelly et al . , 2015; Rivera-Chávez et al . , 2016 ) . Increases in luminal oxygen due to diffusion from the microvasculature supplying the epithelium would lead to a higher redox potential under antibiotic treatment . In addition , host inflammation responses to antibiotic treatment and antibiotic-associated pathogen colonization have been shown to produce electron acceptors and other redox active molecules that cause oxidative stress ( Faber et al . , 2016; Winter et al . , 2010; 2013; Spees et al . , 2013 ) . While these pathways have been demonstrated previously , their overall impact on redox potential has not been measured . Furthermore , the contribution of microbial metabolism to gut redox potential has not been tested , although it is known that a wide range of resident gut bacteria can respire aerobically and anaerobically ( Ravcheev and Thiele , 2014 ) . Antibiotic-driven bacterial inhibition could increase the availability of electron acceptors and thus serve as another mechanism for antibiotic induced changes in gut redox potential . On the other hand , antibiotic inhibition could limit bacterial production of oxidizing agents thereby resulting in an overall decrease in redox potential under antibiotics . Changes in redox potential under antibiotics would in turn be expected to yield insight into the forces structuring the composition and function of the microbiota . Elevated redox potential due to a host immune response could restrict the microbiota beyond direct antibiotic mortality as inflammation would introduce additional oxidative stress into microbial ecosystems . By contrast , elevated redox potential due to the accumulation of oxygen or anaerobic electron acceptors would foster the growth of respiring bacteria . Antibiotic disturbance produces reproducible community succession in the gut following treatment—most notably , a transient bloom in Enterobacteriaceae ( Antonopoulos et al . , 2009; David et al . , 2015; Young and Schmidt , 2004; Theriot et al . , 2014; Peterfreund et al . , 2012; Dethlefsen et al . , 2008; Jakobsson et al . , 2010; Looft and Allen , 2012 ) , which has been ascribed to an increase in oxygen availability ( David et al . , 2015 ) . But , the role of redox potential during this successional process has not previously been studied . If redox potential changes due to antibiotic treatment , we would predict that redox state recovery would be necessary for community resilience , that is for compositional recovery to a pre-disturbance state to occur ( Shade et al . , 2012 ) . Furthermore , we expect that feedback between the community ( i . e . the biotic component of the system ) and their environment ( i . e . , the abiotic component ) would drive further redox potential changes during the successional period . Such a pattern would highlight the potential of manipulating redox potential to alter community dynamics after disturbance . Here , we combined in vivo and ex vivo antibiotic studies to isolate the effects of host and microbial pathways on redox state in the mammalian gut . We confirmed that redox potential increased during antibiotic treatment in association with some changes in host immune state . However , multiple lines of evidence in both mice and an artificial human gut model suggested that antibiotic-induced changes in microbial metabolism were sufficient to cause an increase in redox potential . After antibiotic treatments ended , we observed redox recovery within a week . A successional return to conventional community composition only occurred , though , when mice were co-housed and shared gut microbiota , indicating that microbial dispersal is necessary above and beyond environmental recovery for the return of normal microbiota community structure . For five days , we orally gavaged a cohort of conventional mice with a cocktail of antibiotics ( ampicillin , vancomycin , metronidazole , and neomycin [Reikvam et al . , 2011] ) to broadly inhibit gut bacteria . We measured redox potential in freshly voided feces with a microelectrode paired with a reference electrode daily . Within sixteen hours after the first dose of antibiotics , redox potential significantly increased from 37 ± 164 mV at baseline to 227 ± 45 mV ( p=0 . 04 Bonferroni-corrected Mann-Whitney U test; Figure 1A ) . Throughout treatment , redox potential differed overall between treated and control mice ( p=0 . 005 linear mixed effects model likelihood tests ) . We examined first whether antibiotics affected redox potential through direct host effects mediated by an immune response . Expression of three host genes in fecal samples were consistent with the hypothesis that redox shifts were associated with immune activation . We measured increases relative to controls in Nos2 , which is linked to reactive nitrogen species levels ( Dedon and Tannenbaum , 2004; Winterbourn , 2008 ) ; increases relative to controls in Rela , which has been found to contribute to IBD type inflammation in mice via a proinflammatory cytokine response ( Waddell et al . , 2013 ) ; and , decreases relative to controls in Apoa4 , which has known anti-inflammatory function ( Broedl et al . , 2007 ) ( p<0 . 05 , Bonferroni-corrected one-sample t-tests; Figure 1B ) . Yet , other biomarkers did not associate antibiotic treatment with an immune response . Nfkb1 expression , which is associated with inflammation suppression ( Cartwright et al . , 2016 ) , increased after antibiotic treatment . Antibiotic treatment was also followed by a small , but significant , decrease in lipocalin-2 levels , which is a protein biomarker of inflammation ( Chassaing et al . , 2012 ) ( p<0 . 001 , linear mixed effects model likelihood tests; Figure 1C ) . Taken together , our biomarker assays provided equivocal evidence for intestinal inflammation in antibiotic-treated mice . To test if inhibition of bacterial populations by antibiotics could directly contribute to redox potential dynamics , we used an ex vivo human gut system based on a continuous-flow bioreactor ( McDonald et al . , 2013 ) , which allowed the propagation of a stable microbial community representative of the human gut microbiota with all major phyla represented ( Figure 2—figure supplement 1 ) . Treating this system with the same antibiotic cocktail as used in the mouse study led to an increase in redox potential relative to an untreated control ( p=0 . 005 , linear mixed effects model likelihood tests; Figure 2A ) . Redox potential increased by 59 ± 47 mV within fifteen hours of the first antibiotic dose , mirroring the rapidity with which redox shifts occurred in vivo ( Figure 1A ) . Redox potential in our ex vivo model increased again by another 141 ± 37 mV after a second antibiotic dose ( Figure 2—figure supplement 1 ) . Thus , in the absence of direct interactions between antibiotics and a host , antibiotic treatment produced shifts in environmental redox potential experienced by a gut microbial community . To understand how antibiotic effects on bacterial populations could lead to shifts in gut redox potential , we investigated the dynamics and metabolism of microbiota across treatment in mice . We observed depressed levels of bacterial load and metabolic activity that occurred within hours of initial redox potential shifts . Fecal bacterial concentrations decreased significantly within twelve hours of antibiotic treatment ( p=0 . 01 , Bonferroni-corrected Mann-Whitney U test; Figure 2B ) , and remained significantly below controls throughout the five days of treatment ( p<0 . 001 , Bonferroni-corrected Mann-Whitney U tests; Figure 2—figure supplement 2 ) . Using NMR-based metabolomics , we found that the short-chain fatty acids propionate and acetate , end products of key microbial metabolic pathways , decreased eight hours after antibiotic treatment ( p<0 . 05 , Bonferroni-corrected Mann-Whitney U tests; Figure 2—figure supplement 3 , Supplementary file 1 ) . Twenty-one of twenty-eight metabolites measured , including other short-chain fatty acids , amino acids , branched-chain amino acids , and a group of bile acids , later decreased significantly under antibiotic treatment ( p<0 . 05 , Bonferroni-corrected Mann-Whitney U tests; Figure 2C , Figure 2—figure supplement 3 , Supplementary file 1 ) . The dynamics of twelve of those metabolites was significantly associated with treatment overall ( p<0 . 05 , linear mixed effects model likelihood tests; Supplementary file 1 ) . Thus , the timing and persistence of decreased gut load and microbial activity coincided with increases in fecal redox potential . Next , we measured the concentration of three major electron acceptors used during microbial respiration to assess whether changes in these pathways contributed to the redox potential increase . First , measuring luminal oxygen with a novel in vivo sensor system—a hydrogel sensor with covalently attached oxygen sensitive Pd-porphyrin derivative was inserted rectally then read optically through the skin—we observed a significant increase in luminal oxygen levels the day after antibiotic treatment ( 3 . 7 ± 3 . 6 Torr at baseline to 26 . 5 ± 26 . 1 Torr; p=0 . 02 , Bonferroni-corrected Mann-Whitney U test; Figure 3A ) . Previous work ( Kelly et al . , 2015; Rivera-Chávez et al . , 2016 ) has shown elevated tissue oxygenation under antibiotics , which could result in higher diffusion and therefore higher luminal oxygen . In addition , reduced aerobic respiration due to antibiotic mortality or stress could also increase gut oxygen levels . However , gut oxygen levels in antibiotic-treated mice returned to control levels by forty-eight hours after the first dose ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U test ) , and antibiotics were not associated with an overall effect on gut oxygen levels ( p>0 . 05 linear mixed effects model likelihood tests ) . Moreover , among treated mice , gut oxygen levels were not correlated with redox potential ( p>0 . 05 , repeated measures correlation; Figure 3—figure supplement 1 ) . These data together suggest that increased luminal oxygenation during antibiotic treatment may contribute to early shifts in redox potential , but were not responsible for persistent redox shifts . By contrast , we observed sustained and significant increases in nitrate levels during antibiotic treatment . Nitrate is one of the most widely used electron acceptors in the gut ( Ravcheev and Thiele , 2014; Fischer and Lindenmayer , 2007 ) and its reduction to nitrogen gas ( at pH 7 ) has a potential of +0 . 75 V making it one of the most potent electron acceptors after oxygen . The increase in Nos2 expression ( Figure 1B ) led us to hypothesize that nitrate would increase as it is a gene which encodes inducible nitric oxide synthase and whose expression has been found to confer a growth advantage to nitrate respiration competent E . coli strain ( Winter et al . , 2013 ) . Lower levels of microbial respiration could also lead to an accumulation in nitrate separate from changes in host expression . We measured fecal nitrate levels with the chromotropic acid method ( West and Ramachandran , 1966 ) for a subset of time points when redox varied , and found a significant increase in nitrate ( p=0 . 03 , Mann-Whitney U test; Figure 3B ) beginning less than 24 hr after the first dose of antibiotics . This increase persisted throughout treatment , and overall antibiotic treatment produced a significant increase in nitrate ( p=0 . 02 , linear mixed effects model likelihood tests ) . Finally , we used a colorimetric enzyme assay to measure fecal fumarate levels for a subset of time points during antibiotic treatment . Fumarate reductases have been found in over one third of gut microbe genomes , and fumarate is the most common terminal electron acceptor for bacterial anaerobic respiration ( Kröger et al . , 1992 ) . Unlike nitrate and oxygen , though , much of the fumarate in the gut is likely produced by the microbiota itself ( Fischbach and Sonnenburg , 2011; El Aidy et al . , 2013 ) . Overall , there was a significant effect of treatment on fumarate levels ( p<0 . 001 , linear mixed effects model likelihood tests; Figure 3C ) . We observed a spike in fumarate levels 4 hr after the first antibiotics dose ( p=0 . 004 , Bonferroni-corrected Mann-Whitney U test ) ; however , this increase had dissipated by 8 hr after the first dose and later during treatment , we observed a significant decrease ( to below the detection limit ) in fumarate levels . Thus , not all electron acceptors increased in abundance during antibiotic treatment . Ecological succession , the ‘somewhat orderly and predictable’ ( Fierer et al . , 2010 ) dynamics of a community after a perturbation , is known to follow antibiotic treatment ( Antonopoulos et al . , 2009; David et al . , 2015 ) . What is less well-known is how abiotic conditions change during post-antibiotic succession and , by extension , how biotic and abiotic dynamics interface during that period . Having shown that both the community and gut environment are not resistant to antibiotic disturbance , we next sought to determine how resilient they were , i . e . how quickly they recovered ( Shade et al . , 2012; Allison and Martiny , 2008 ) . In this study , we indeed observed consistent biotic changes over time in mice that received antibiotics ( Figure 4—figure supplement 1 ) . Gut bacterial load recovered rapidly after treatment across all treated mice , exhibiting no difference relative to control mice by two days ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U tests; Figure 2—figure supplement 2 ) . A reproducible recovery of beta-diversity was observed after one week ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U test; Figure 4—figure supplement 2 ) . To characterize the recovery of specific microbial taxa , we clustered fecal bacterial genera into groups according to their dynamical patterns across antibiotic treatment ( Figure 4A , B ) . We focused on the five clusters with an average abundance of at least 1% across our dataset ( Supplementary file 2 ) . Amongst these five groups , three differed significantly during treatment ( p<0 . 05 , linear mixed effects model likelihood tests; Figure 4A , B ) . Notably , the two clusters elevated during treatment included many facultative anaerobic taxa ( Supplementary file 2 ) . One of these , the cluster primarily composed of members of the Enterobacteriacaeae , remained elevated into the beginning of the recovery period as well ( p<0 . 05 , Bonferroni-corrected Mann-Whitney U tests ) . This enrichment is in line with previous findings that Enterobacteriaceae often bloom following antibiotic treatment ( Antonopoulos et al . , 2009; David et al . , 2015; Young and Schmidt , 2004; Theriot et al . , 2014; Peterfreund et al . , 2012; Dethlefsen et al . , 2008; Jakobsson et al . , 2010; Looft and Allen , 2012 ) . That cluster , as well as one composed primarily of Akkermansia and one of a consortium of typical commensal taxa , exhibited significant residual effects of treatment during the recovery period ( p<0 . 05 , linear mixed effects model likelihood tests; Figure 4A , B ) . However , by the end of the one week recovery period , the abundance of all clusters was indistinguishable between treated and control animals ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U tests ) . Abiotic conditions in the mouse gut were also resilient and recovery was predictable . Redox potential returned to control levels within a week after treatment ended ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U tests; Figure 4C ) . Notably , though , we observed the day after antibiotic therapy ceased , fecal redox potentials in treated mice were significantly decreased relative to control mice ( -138 . 3 ± 149 . 8 mV vs 54 . 6 ± 211 mV; p=0 . 006 , Bonferroni-corrected Mann-Whitney U test ) , and it was not until later in recovery that redox potential returned to control levels across most mice . Oxygen and inflammation biomarker levels remained low throughout the recovery period ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U tests; Figure 3—figure supplement 2 , Figure 1—figure supplement 1 , respectively ) consistent with their minimal changes during treatment . Furthermore , many microbial metabolic products measured by NMR had recovered two days after the last doses , and all had recovered by the end of the recovery week ( Figure 2—figure supplement 3; Supplementary file 1 ) . In concert , a largely reproducible biotic and abiotic recovery took place across antibiotic treated mice in the days following antibiotic treatment indicating the overall resilience of this ecosystem to antibiotic disturbance . In investigating potential interactions between abiotic and biotic processes in antibiotic recovery , we hypothesized that the Enterobacteriaceae could thrive under high redox potential conditions and directly impact gut redox state . Members of this bacterial family were enriched at the end of treatment in our clustering analysis ( Figure 4B ) . Enterobacteriaceae can also employ many terminal electron acceptors to perform aerobic and anaerobic respiration ( Ravcheev and Thiele , 2014 ) and thus could likely thrive under high redox potential conditions and directly impact gut redox state . We found that the relative abundance of the Enterobacteriaceae was marginally correlated with redox potential ( p=0 . 1 , r= , 0 . 18 , repeated measures correlation; Figure 4—figure supplement 3 ) throughout treatment and recovery . Moreover , the change in redox potential during the earlier recovery period was negatively correlated with changes in Enterobacteriaceae absolute abundance ( calculated as total 16S rRNA gene copy number multiplied by relative abundance ) for both treated animals ( p=0 . 03 , r = −0 . 42 , repeated measures correlation ) and for all animals ( p=0 . 02 , r = −0 . 34 , repeated measures correlation; Figure 4D ) . Together , these observations suggest that increased respiration by the Enterobacteriaceae after antibiotic treatment may lead redox potentials to decrease below conventional levels . More broadly , our observations support the hypothesis that Enterobacteriaceae dynamics contribute to natural redox potential variation in the gut . We next investigated if abiotic recovery led to compositional recovery or if dispersal ( i . e . , migration of microbes ) from unimpacted microbial populations was necessary for microbiota to exhibit resilience and return to a pre-treatment state . Previous research has shown that singly housed mice treated with antibiotics will exhibit altered community composition for weeks after treatment ( Antonopoulos et al . , 2009 ) . We therefore co-housed half of the treated mice with control animals during the recovery period and kept half in single housing . Because mice practice coprophagy ( i . e . , ingestion of feces ) , we would expect that co-housing would re-introduce normal commensal microbes to compete with the Enterobacteriaceae and other early successional species thereby reducing their abundance . At the end of the one-week treatment period , we indeed observed differential success in community recovery between the housing groups . Co-housed mice communities had returned to control composition while singly housed mice had not . For co-housed mice , by the end of the recovery period , the gut microbiota of treated animals was equally dissimilar from baseline as it was for control mice indicating that the community had recovered ( p>0 . 05 , Bonferroni-corrected Mann-Whitney U test; Figure 4—figure supplement 4 ) . Both fecal redox potential and Enterobacteriaceae levels also returned to conventional levels more quickly in co-housed mice than singly housed mice ( Figure 4E ) . By contrast , singly-housed treated mice continued to be more dissimilar at the end of recovery than singly-housed control mice ( p<0 . 001 , Bonferroni-corrected Mann-Whitney U test; Figure 4—figure supplement 4 ) . Such drift is qualitatively consistent with arguments that stochastic variation can influence how communities respond to perturbation , as seen in soil microbial communities ( Zhang et al . , 2016 ) . More broadly , the differences between singly- and co-housed mice here highlight that environmental recovery is not sufficient to produce biotic recovery—dispersal is also necessary for a resilient community . Together , our findings suggest new ecological models for how antibiotics reshape the gut microbiota and for how redox shifts could be associated with enteric disease . Antibiotics are triumphs of modern medicine that have dramatically reduced infectious disease mortality ( Armstrong et al . , 1999 ) . But , we are increasingly learning that antibiotics also meaningfully reshape the resident gut microbiota , leaving an imprint that can last for months to years after treatment ( Dethlefsen and Relman , 2011; Jakobsson et al . , 2010 ) and predisposing hosts to obesity ( Cho et al . , 2012 ) , food allergy ( Stefka et al . , 2014 ) , auto-immune disease ( Russell et al . , 2012 ) , and increased infection risk ( Stecher et al . , 2007; Buffie et al . , 2012; Wiström et al . , 2001 ) . While these drugs reduce levels of susceptible organisms ( Keeney et al . , 2014 ) , an additional ecological mechanism of action is decreasing microbial competition and allowing primary metabolites ( e . g . , primary bile acids , sugars ) ( Ng et al . , 2013; Theriot et al . , 2014 ) , as well as host-sourced electron acceptors like oxygen and nitrate to accumulate . This concept complements recent discoveries that electron acceptors facilitate antibiotic-associated enteric pathogen colonization ( Rivera-Chávez et al . , 2016; Winter et al . , 2013; 2010 ) . Such increases in electron acceptor availability likely are not unique to antibiotic treatment and could generalize to various enteric disturbances . Indeed , germ-free animals ( Phillips et al . , 1958; Celesk et al . , 1976 ) and humans suffering from inflammatory diseases ( Circu and Aw , 2011 ) and malnutrition ( Million et al . , 2016 ) exhibit increased gut redox potential . Thus , novel treatments for microbial disorders or preventing antibiotic-associated infections may include chemical alterations of redox potential or introduction of competitors for excess electron acceptors . More broadly , we propose adding redox potential to the list of abiotic conditions frequently assayed and manipulated to improve host well-being . We performed linear mixed effects model analysis to determine the effects of antibiotics on redox potential , oxygen concentration , fecal lipocalin-2 concentration , and Bray-Curtis dissimilarity . As fixed effects , we entered antibiotic treatment and time with an interaction term into the model . We included mouse identity as a random effect . P values were obtained by likelihood ratio tests comparing the full model against a model including only time and mouse identity and were performed with the ‘anova’ function in the ‘lme4’ package . Repeated measures correlations were used to assess correlations where multiple time points from the same mouse were included in the statistical analysis . Repeated measures correlations were calculated with the ‘rmcorr’ function in the ‘rmcorr’ package . These and all other statistical analyses were carried out in R ( R core team , version 3 . 3 ) . All statistical tests performed were non-parametric except where a Shapiro-Wilks test indicated that data were normally distributed , in which case parametric tests were used . All data points were included in analyses and outliers were not treated in any manner .	The gut is home to a large and diverse community of bacteria and other microbes , known as the gut microbiota . The makeup of this community is important for the health of both the host and its residents . For instance , many gut bacteria help to digest food or keep disease-causing bacteria in check . In return , the host provides them with nutrients . When this balance is disturbed , the host is exposed to risks such as infections . In particular , treatments with antibiotics that kill gut bacteria can lead to side effects like diarrhea , because the gut becomes recolonized with harmful bacteria including Clostridium difficile and Salmonella . Reese et al . have now investigated what happens to the gut environment after antibiotic treatment and how the gut microbiota recovers . Mice treated with broad-spectrum antibiotics showed an increase in the “redox potential” of their gut environment . Redox potential captures a number of measures of the chemical makeup of an environment , and provides an estimate for how efficiently some bacteria in that environment can grow . Some of the change in redox potential came from the host’s own immune system releasing chemicals as it reacted to the effects of the treatment . However , Reese et al . found that treating gut bacteria in an artificial gut – which has no immune system – also increased the redox potential . This experiment suggests that bacteria actively shape their chemical environment in the gut . After the treatment , bacteria that thrive under high redox potentials , which include some disease-causing species , recovered first and fastest . This , in turn , helped to bring redox potential back to how it was before the treatment . Although the gut’s chemical environment recovered , some bacterial species were wiped out by the antibiotic treatment . The microbiota only returned to its previous state when the treated mice were housed together with non-treated mice . This was expected because mice that live together commonly exchange microbes , for instance by eating each other’s feces , and the treated mice received new species to replenish their microbiota . These findings are important because they show that the chemical environment shapes and is shaped by the bacterial communities in the gut . Future research may investigate if altering redox potential in the gut could help to keep the microbiota healthier in infections and diseases of the digestive tract .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2018	Antibiotic-induced changes in the microbiota disrupt redox dynamics in the gut
The perception and response to cellular death is an important aspect of multicellular eukaryotic life . For example , damage-associated molecular patterns activate an inflammatory cascade that leads to removal of cellular debris and promotion of healing . We demonstrate that lysis of Pseudomonas aeruginosa cells triggers a program in the remaining population that confers fitness in interspecies co-culture . We find that this program , termed P . aeruginosa response to antagonism ( PARA ) , involves rapid deployment of antibacterial factors and is mediated by the Gac/Rsm global regulatory pathway . Type VI secretion , and , unexpectedly , conjugative type IV secretion within competing bacteria , induce P . aeruginosa lysis and activate PARA , thus providing a mechanism for the enhanced capacity of P . aeruginosa to target bacteria that elaborate these factors . Our finding that bacteria sense damaged kin and respond via a widely distributed pathway to mount a complex response raises the possibility that danger sensing is an evolutionarily conserved process . Bacteria can occupy highly dynamic environments , where survival is linked to the ability to sense and respond to an assortment of threats ( Cornforth and Foster , 2013 ) . It is increasingly clear that in addition to well-understood environmental and nutritive stresses , antagonistic factors elaborated by other bacteria are a common threat that bacteria must cope with ( Little et al . , 2008; Hibbing et al . , 2010 ) . A number of antagonistic strategies have been identified , including the production of diffusible factors such as small molecule antibiotics . It has been suggested that sub-inhibitory concentrations of these molecules induce specific changes in bacteria , including the production and distribution of resistance mechanisms ( Hoffman et al . , 2005; Linares et al . , 2006; Andersson and Hughes , 2014 ) . Other antagonistic pathways , such as contact-dependent inhibition ( CDI ) and the type VI secretion system ( T6SS ) , require cell contact ( Hayes et al . , 2010; Konovalova and Sogaard-Andersen , 2011 ) . Kin cells are protected from the toxic proteins delivered by these pathways by specific cognate immunity proteins; however , escape from or defense against these pathways by non-kin is not well understood ( Hood et al . , 2010; Russell et al . , 2011 ) . The T6SS is a versatile export machinery that can deliver a wide range of proteinaceous effector molecules from donor to recipient Gram-negative bacterial cells ( Hood et al . , 2010; Coulthurst , 2013; Russell et al . , 2014a ) . One of the best characterized bacterial targeting T6SSs is the Hcp Secretion Island I-encoded T6SS ( H1-T6SS ) of Pseudomonas aeruginosa ( Hood et al . , 2010 ) . The H1-T6SS transports a cargo of at least seven effectors , termed type VI secretion exported 1–7 ( Tse1–7 ) ( Hachani et al . , 2014; Whitney et al . , 2014 ) . The outcome of intoxication by these proteins can be lysis or cessation of growth ( LeRoux et al . , 2012; Li et al . , 2012 ) . Like other species with interbacterial T6SSs , P . aeruginosa has the capacity to target cells of its own genotype with the H1 pathway . To inhibit self-intoxication , cognate type VI secretion immunity proteins ( Tsi ) are produced and localized to the cellular compartment that contains the target of the corresponding effector ( Russell et al . , 2013; Benz and Meinhart , 2014; Durand et al . , 2014; Russell et al . , 2014a ) . T6SSs are found in at least three genetically distinct configurations present among multiple bacterial phyla , making it one of the most widespread pathways mediating interbacterial antagonism known ( Russell et al . , 2014b ) . Expression and activity of the H1-T6SS is tightly regulated at multiple levels ( Silverman et al . , 2012 ) . Stringent post-transcriptional regulation of the H1-T6SS is achieved through the global activation of antibiotic and cyanide synthesis/regulator of secondary metabolism ( Gac/Rsm ) pathway ( Goodman et al . , 2004; Mougous et al . , 2006b ) . This global regulatory system impacts protein production via RsmA , a CsrA-type protein that binds to target mRNA molecules and generally acts to repress their translation ( Lapouge et al . , 2008 ) . RsmA is modulated by levels of the small RNA ( sRNA ) molecules rsmY and rsmZ , which bind to and sequester it from its targets . Transcription of the sRNAs is promoted by phosphorylated GacA , the cognate response regulator of the sensor kinase GacS . Finally , two hybrid sensor kinases , RetS and LadS , acting through GacS , repress or stimulate GacA phosphorylation , respectively ( Ventre et al . , 2006; Goodman et al . , 2009 ) . Consistent with regulation of T6S by the Gac/Rsm pathway , many of its targets in P . aeruginosa and related γ-proteobacteria are involved in the production of social or antagonist factors ( Lapouge et al . , 2008 ) . This theme of Gac/Rsm-dependent modulation of antibiotic activity is exemplified by the defect of P . fluorescens gac mutants in bacterial and fungal growth inhibition on plants ( Laville et al . , 1992 ) . Though the precise cues that activate the Gac/Rsm pathway are unknown , Haas et al . have found that one or more signals accumulate in spent bacterial culture supernatants deriving from both self and non-self organisms ( Dubuis and Haas , 2007 ) . Posttranslational regulation by the threonine phosphorylation pathway ( TPP ) constitutes a second level of control over H1-T6SS activity . In this pathway , phosphorylation of a fork head-associated domain-containing protein , Fha1 , triggers apparatus assembly and effector secretion ( Mougous et al . , 2007 ) . PppA , a phosphatase , opposes the activity of PpkA on Fha1 , returning the system to the inactive state . Additional components of the TPP , encoded by type VI secretion associated genes Q-T ( tagQ-T ) , act upstream of these proteins and are thought to be involved in signal transduction ( Hsu et al . , 2009; Casabona et al . , 2013 ) . Two signals of the TPP have been proposed: surface-associated growth and membrane perturbation ( Silverman et al . , 2011; Basler et al . , 2013 ) . The latter signal is thought to underlie the observation that organisms with active T6S or type IV secretion ( T4S ) are more efficiently targeted by the H1-T6SS than those without ( Ho et al . , 2013 ) . It was proposed that the activity of these apparatuses induces local membrane perturbations in P . aeruginosa that are sensed by the TPP , leading to posttranslational activation of the H1-T6SS and enhanced recipient cell death ( Basler et al . , 2013 ) . A caveat of these studies is that the P . aeruginosa strain used bears an inactivating mutation in retS , which constitutively activates the Gac/Rsm pathway , potentially masking the contribution of this major regulatory mechanism to the defense mounted by P . aeruginosa against the antagonistic pathways of competing bacteria . Here , we show that lysed kin cells act as a danger signal that is sensed by the Gac/Rsm pathway of wild-type P . aeruginosa . Our experiments provide a mechanism for T6S-dependent killing of competitor bacteria possessing either the T6 or T4 secretion pathways , as both induce P . aeruginosa lysis , stimulate the Gac/Rsm pathway , and lead to posttranscriptional de-repression of the H1-T6SS . These findings provide a rationale for the regulation of promiscuous antibiotic mechanisms by a pathway that can respond to self-derived signals . Prior work suggests that efficient T6S-dependent effector delivery between P . aeruginosa cells ( self-targeting ) requires deletion of retS , whereas this activating mutation is not required for robust T6S-dependent intoxication of other bacteria by P . aeruginosa ( non-self targeting ) ( Hood et al . , 2010; Russell et al . , 2011; Ho et al . , 2013 ) . To quantify these observations , we performed bacterial growth competition experiments with wild-type ( PAO1 ) and ∆retS with both self- and non-self recipients under comparable conditions . Burkholderia thailandensis ( B . thai ) was used as the non-self competitor for these studies , while a P . aeruginosa background lacking four H1-T6SS effector–immunity pairs ( ∆tse1-4 ∆tsi1-4 ) was employed as the self recipient . T6S-dependent fitness was determined by comparing the competitive index of a wild-type donor to that of a donor strain lacking tssM , which encodes a core structural component of the T6S apparatus ( Felisberto-Rodrigues et al . , 2011 ) . We found that both P . aeruginosa wild-type and ∆retS strains reduced populations of B . thai in a T6S-dependent manner by several orders of magnitude; however , only P . aeruginosa ∆retS displayed T6S-dependent fitness in co-culture with self recipients ( Figure 1 ) . This direct comparison of H1-T6SS-dependent antibiosis by P . aeruginosa wild-type and ∆retS demonstrates that unlike self-intoxication , non-self targeting by the pathway does not require activation of the system by relief of negative regulation . A parsimonious explanation for these data is that the presence of a non-self competitor stimulates H1-T6SS activity . 10 . 7554/eLife . 05701 . 003Figure 1 . Wild-type P . aeruginosa cells display a strong T6S-dependent fitness advantage in co-culture with non-self but not self competitors . Outcome of growth competition experiments measuring fitness of P . aeruginosa PAO1 parental or ∆retS strains in co-cultures with self or non-self recipients under T6SS-promoting conditions . The self recipient was P . aeruginosa ∆tse1-4 ∆tsi1-4 in the strain background corresponding to the donor genotype ( PAO1 or PAO1 ∆retS ) . T6S-dependent fitness was parental donor competitive index ( change [final/initial] in ratio of donor and recipient colony forming units [c . f . u . ] ) normalized to ∆tssM1 competitive index . Error bars represent ±standard deviation ( SD ) ; n = 3 co-cultures . Asterisks denote a fitness advantage significantly >1 ( p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 003 To test the hypothesis that H1-T6SS activity is affected by the presence of a non-self organism , we used time-lapse fluorescence microscopy ( TLFM ) in conjunction with customized cell and protein tracking software to monitor the expression and subcellular localization of the conserved T6S ATPase ClpV1 ( Cascales and Cambillau , 2012; LeRoux et al . , 2012 ) . Previous studies have established that translational fusion of gfp to the 3′ end of P . aeruginosa clpV1 , encoded within the H1-T6SS gene cluster , yields a stable and functional chimera ( Mougous et al . , 2006a ) . The fluorescence intensity of ClpV1–GFP provides readout of H1-T6SS expression and the assembly of ClpV1–GFP into punctate foci correlates with activity of the system ( Mougous et al . , 2007; Kapitein et al . , 2013 ) . Surprisingly , we observed significantly elevated ClpV1–GFP levels in P . aeruginosa–B . thai co-cultures relative to P . aeruginosa monocultures , suggesting that B . thai stimulates H1-T6SS expression ( Figure 2A , B , Figure 2—figure supplement 1 , Video 1 ) . We obtained similar results from a strain bearing a chromosomal , functional translational fusion of gfp to fha1 ( fha1-gfp ) , a locus found on the second major H1-T6SS transcript ( Figure 2A , C , Figure 2—figure supplement 1 , Video 2 ) ( Mougous et al . , 2006a , 2007 ) . To determine whether elevated H1-T6SS arises from increased expression throughout the population or from high expression within a subset of P . aeruginosa , we examined ClpV1-GFP at the level of individual cells . Consistent with a population-wide , graded response , we found that ClpV1-GFP exhibits a normal distribution both in monoculture and when P . aeruginosa is co-cultivated with B . thai ( Figure 2—figure supplement 2 ) . Mirroring the observed trends in expression , after an initial increase in foci frequency—a correlate of T6S activity—associated with growth on a surface ( Silverman et al . , 2011 ) , a larger percentage of P . aeruginosa cells grown in the presence of B . thai contained ClpV1–GFP and Fha1–GFP foci compared to those grown in monoculture ( Figure 2D , E ) . To investigate the generality of these effects on the H1-T6SS , we repeated our experiments using the γ-proteobacterium Enterobacter cloacae as the competing organism . P . aeruginosa co-cultivated with E . cloacae also displayed increased ClpV1–GFP levels and higher foci frequency in comparison to P . aeruginosa in monoculture , indicating that this response is not specific to B . thai ( Figure 2—figure supplement 3; Video 3 ) . 10 . 7554/eLife . 05701 . 004Figure 2 . Non-self competitor bacteria stimulate expression and activity of the H1-T6SS . ( A ) H1-T6SS expression is increased in P . aeruginosa co-cultured with B . thai expressing an active T6SS . Time-lapse fluorescence microscopy ( TLFM ) sequences of P . aeruginosa clpV1-gfp ( upper ) or fha1-gfp ( lower ) in monoculture or in co-culture with the indicated competitor . Cropped regions from representative time-points are displayed . Remaining time points for monoculture and co-culture with B . thai ∆tssM-1 are depicted in Figure 2—figure supplement 1; see also Videos 1 and 2 . Masks colored by cell identity depict automated cell identification generated from the phase image . Scale bar , 6 µm . ( B–C ) Quantification of H1-T6SS expression from the P . aeruginosa clpV1-gfp ( B ) and fha1-gfp ( C ) monoculture and co-culture TLFM experiments described in ( A ) . Average cellular GFP intensity for P . aeruginosa cells was calculated from background-subtracted images . ( D–E ) H1-T6SS activity is increased in the presence of B . thai with an active T6SS . Percentage of P . aeruginosa clpV1-gfp ( D ) or fha1-gfp ( E ) cells with GFP foci for experiments described in ( A ) . Error bars represent ±SD; n = 3 fields . Asterisks indicate significant differences when B . thai was present ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00410 . 7554/eLife . 05701 . 005Figure 2—figure supplement 1 . Competitors require an active T6SS to stimulate the P . aeruginosa H1-T6SS . TLFM sequences of P . aeruginosa clpV1-gfp ( upper ) and fha1-gfp ( lower ) cultivated either without competitor or with B . thai ∆tssM-1 as depicted in Figure 2A . Scale bar , 6 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00510 . 7554/eLife . 05701 . 006Figure 2—figure supplement 2 . Increased H1-T6SS expression occurs throughout the population . Histograms of cellular ClpV1-GFP intensity of P . aeruginosa clpV1-gfp following 90 min of growth in monoculture or in co-culture with B . thai . Histogram bin size is 20 intensity units and is normalized to total cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00610 . 7554/eLife . 05701 . 007Figure 2—figure supplement 3 . E . cloacae stimulates the H1-T6SS of P . aeruginosa in a T6S-dependent manner . Average cellular ClpV1-GFP expression ( A ) and the percentage of cells with ClpV1-GFP foci ( B ) of P . aeruginosa clpV1-gfp in monoculture or in co-culture with the indicated E . cloacae competitor . Data was collected and analyzed as described in Figure 2 . Error bars represent ±SD; n = 3 fields . Asterisks indicate significant differences when E . cloacae was present ( p < 0 . 05 ) . See also Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00710 . 7554/eLife . 05701 . 008Figure 2—figure supplement 4 . P . aeruginosa doubling time is not affected by the presence of B . thai . Histograms depicting P . aeruginosa doubling times during growth in monoculture or co-culture with B . thai under TLFM conditions . Error bars represent ±SD; n = 3 fields . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00810 . 7554/eLife . 05701 . 009Video 1 . ClpV1 expression increases in the presence of B . thai bearing an active T6SS . TLFM sequences of ClpV1-GFP in P . aeruginosa during monoculture or interspecies co-culture . P . aeruginosa clpV1-gfp cells without competitor ( left sequence ) , with B . thai mCherry ( middle sequence ) , or with B . thai ∆tssM-1 mCherry ( right sequence ) were imaged at 15 min intervals . Overlays of GFP and mCherry channels are displayed . The same thresholds were applied to all background-subtracted GFP channels . See Figure 2B , D for quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 00910 . 7554/eLife . 05701 . 010Video 2 . Fha1 expression increases in the presence of B . thai bearing an active T6SS . TLFM sequences depicting expression of Fha1-GFP in P . aeruginosa during monoculture or interspecies co-cultures . P . aeruginosa fha1-gfp cells without competitor ( left sequence ) , with B . thai mCherry ( middle sequence ) , or with B . thai ∆tssM-1 mCherry ( right sequence ) were imaged at 15 min intervals . Overlays of GFP and mCherry channels are displayed . The same thresholds were applied to all background-subtracted GFP channels . Quantification is provided in Figure 2C , E . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01010 . 7554/eLife . 05701 . 011Video 3 . ClpV1 expression increases in the presence of E . cloacae bearing an active T6SS . TLFM sequences depicting expression of ClpV1-GFP in P . aeruginosa during monoculture or interspecies co-cultures . P . aeruginosa clpV1-gfp mCherry cells without competitor ( left sequence ) , with E . cloacae ( middle sequence ) , or with E . cloacae ∆tssM ( right sequence ) were imaged at 15 min intervals . Overlays of GFP and mCherry channels are displayed . P . aeruginosa cells were labeled with constitutive mCherry and therefore the overlay of GFP and mCherry appears yellow . The unlabeled E . cloacae cells are visible ( light green ) due to autofluorescence in the GFP channel . The same thresholds were applied to all background-subtracted GFP channels . Quantification is provided in Figure 2—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 011 Previous studies demonstrated that the H1-T6SS of P . aeruginosa targets recipient cells that possess an active T6SS with greater efficiency than those that do not ( LeRoux et al . , 2012; Basler et al . , 2013 ) . Though this behavior was shown for P . aeruginosa ∆retS , and not wild-type cells , the apparent capacity of the organism to sense the T6SS of a recipient cell prompted us to test whether T6S in non-self competitors is involved in stimulation of the H1 pathway . Indeed , we found that B . thai and E . cloacae strains bearing in-frame deletions of the tssM genes associated with their known antibacterial T6S pathways , are unable to stimulate T6S in P . aeruginosa ( Figure 2 , Figure 2—figure supplement 1 , Figure 2—figure supplement 3 , Videos 1–3 ) ( Schwarz et al . , 2010; Koskiniemi et al . , 2013; Whitney et al . , 2014 ) . Changes in P . aeruginosa growth rate dependent upon recipient T6S could influence gene expression and protein accumulation , potentially accounting for altered expression and activity of H1-T6SS proteins . However , this possibility was excluded by our observation that the doubling time of P . aeruginosa is insensitive to the activity of the bacterial cell-targeting T6SS of B . thai ( T6SBT ) ( Figure 2—figure supplement 4 ) . Taken together , these data demonstrate that co-cultivation of P . aeruginosa with a non-self organism possessing an active T6S apparatus leads to elevated expression and activity of the H1-T6SS . Henceforth , we refer to this response of P . aeruginosa as PARA ( P . aeruginosa response to antagonism ) . A recent report implicated the TPP in the capacity of P . aeruginosa to sense and respond to T6S in target cells ( Basler et al . , 2013 ) . To determine if PARA requires the TPP , we monitored the T6S-dependent response of P . aeruginosa bearing a deletion in pppA , which encodes a TPP-associated phosphatase ( Mougous et al . , 2007 ) . Interestingly , though this deletion was previously shown to abrogate the ability of P . aeruginosa ∆retS to respond differentially to bacteria containing or lacking a functional T6SS , we found that in the wild-type background , ∆pppA retains the capacity to respond to T6SBT ( Figure 3A , Figure 3—figure supplement 1A ) . PppA is a negative regulator of the TPP; therefore , we sought to rule out the possibility that the remaining TPP components , TagQ-T and PpkA , are sufficient to mediate PARA . Inactivation of all known components of the TPP results in a failure to assemble an active T6S apparatus , which would confound our analyses . However , activity can be restored to TPP deficient strains by the deletion of a gene encoding an independent post-translational repressor of the H1-T6SS , TagF ( Silverman et al . , 2011 ) . As observed with ∆pppA , the H1-T6SS of a P . aeruginosa strain lacking tagF and all known TPP components ( ∆TPP ∆tagF ) was activated by B . thai in a T6SBT-dependent manner ( Figure 3B , Figure 3—figure supplement 1B ) . Consistent with measurements of expression and activation , growth competition experiments revealed that wild-type , ∆pppA , and ∆tagF ∆TPP strains of P . aeruginosa intoxicate B . thai and E . cloacae with active interbacterial T6SSs more efficiently than those without ( Figure 3C , Figure 3—figure supplement 2 ) . These data indicate that the TPP is not required for PARA . 10 . 7554/eLife . 05701 . 012Figure 3 . PARA does not require the TPP . ( A–B ) Increased H1-T6SS expression in the presence of B . thai does not require a functional TPP . Average ClpV1-GFP cellular fluorescence intensity in P . aeruginosa clpV1-gfp ∆pppA ( A ) and ∆TPP ∆tagF ( B ) backgrounds during monoculture or co-culture with the indicated competitors . Error bars represent ±SD; n = 3 fields . Asterisks indicate significant differences when B . thai was present ( p < 0 . 05 ) . Corresponding H1-T6SS activity is shown in Figure 3—figure supplement 1 . ( C ) The TPP is not required for preferential targeting of B . thai with an active T6SS . Outcome of growth competition experiments measuring survival of B . thai following co-culture with the indicated P . aeruginosa strain under T6SS-promoting conditions . Error bars represent ±SD; n = 3 co-cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01210 . 7554/eLife . 05701 . 013Figure 3—figure supplement 1 . Elevated H1-T6SS activity in the presence of B . thai does not require the TPP . Percentage of cells containing ClpV1-GFP foci from P . aeruginosa clpV1-gfp strains of ∆pppA ( A ) or ∆TPP ∆tagF ( B ) backgrounds in monoculture or in co-culture with the indicated competitor . Error bars represent ±SD; n = 3 co-cultures . Asterisks indicate significant differences when B . thai was present ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01310 . 7554/eLife . 05701 . 014Figure 3—figure supplement 2 . P . aeruginosa does not require the TPP to differentially target E . cloacae with a T6SS . Outcome of interspecies growth competition experiments measuring competitive index ( C . I . ) of P . aeruginosa–E . cloacae co-cultures grown under T6SS-promoting conditions . PA , P . aeruginosa; EC , E . cloacae . Error bars represent ±SD; n = 3 co-cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 014 It is worth noting that several lines of evidence suggest PARA is a process distinct from the previously characterized intercellular T6-based response referred to as ‘T6SS dueling’ ( Basler and Mekalanos , 2012 ) . First , PARA is accompanied by changes to T6S gene expression , whereas dueling is a posttranslational behavior thought to involve rapid changes in protein localization . Second , the TPP was found to be essential for T6SS dueling ( Basler et al . , 2013 ) ; however , this pathway is dispensable for PARA . Finally , spatiotemporal coordination of ClpV1-GFP-containing foci forms the basis for T6SS dueling , while such paired focus events are not correlated with PARA ( data not shown ) . Motivated by the finding of PARA as a distinct pathway of clear functional relevance to interbacterial interactions , we proceeded to investigate its mechanistic underpinnings . The H1-T6SS is regulated at the transcriptional level by the quorum sensing regulator LasR , and at the posttranscriptional level by RsmA and RsmF , RNA binding proteins that directly mediate the effects of Gac/Rsm signaling ( Figure 4A ) ( Brencic and Lory , 2009; Lesic et al . , 2009; Marden et al . , 2013 ) . As a first step toward defining the pathway through which PARA operates , we measured H1-T6SS transcription and translation in response to co-culture with B . thai using previously characterized chromosomal β-galactosidase reporters ( Brencic and Lory , 2009 ) . In contrast to minor effects on transcription , we found that co-culture of P . aeruginosa with B . thai markedly stimulates H1-T6SS translation ( Figure 4B ) . In control experiments with B . thai ∆tssM-1 as the competitor , we observed no enhancement of H1-T6SS expression . Based on these data , we hypothesized that PARA is driven by the Gac/Rsm pathway . 10 . 7554/eLife . 05701 . 015Figure 4 . The Gac/Rsm pathway is required for PARA . ( A ) Schematic depicting the Gac/Rsm pathway of P . aeruginosa . The orphan sensor kinases RetS and LadS exert opposing activity on a third sensor kinase , GacS , which in turn activates its cognate response regulator , GacA . Once active , GacA promotes increased transcription of the small RNAs rsmY and rsmZ . These molecules bind and sequester RsmA; therefore , when abundant , they prevent RsmA binding and destabilization of target mRNAs , including H1-T6SS transcripts . ( B ) Elevated H1-T6SS expression in the presence of B . thai occurs primarily at the post-transcriptional level . P . aeruginosa strains bearing chromosomally encoded transcriptional or translational fusions to tssA1 ( Brencic and Lory , 2009 ) were incubated with the indicated competitor . Fold H1-T6SS increase in expression was determined by normalizing P . aeruginosa co-cultures to the corresponding strain cultivated in monoculture . n = 3 co-cultures; asterisk indicates significant differences between translational and transcriptional activity ( p < 0 . 05 ) . ( C ) RsmZ expression is elevated in the presence of B . thai containing a T6SS . Cells masks and the GFP fluorescence channel from representative TLFM sequences of the indicated P . aeruginosa p-rsmZ-gfp co-cultures . Additional time points are shown in Figure 4—figure supplement 1; see also Video 4 . ( D ) Average cellular fluorescence intensity from P . aeruginosa p-rsmZ-gfp corresponding to ( C ) . ( E ) Expression of MagA is elevated in the presence of T6SBT . Average cellular GFP intensity of P . aeruginosa p-magA-gfp in co-culture with B . thai or B . thai ∆tssM-1 . ( F ) GacS is required for H1-T6SS activation in response to T6SBT . ClpV1-GFP expression was quantified for co-cultures of P . aeruginosa ∆gacS clpV1-gfp with the indicated B . thai competitor . ( G ) LadS is not required for elevated H1-T6SS expression in the presence of B . thai . Average cellular ClpV1-GFP intensity of P . aeruginosa ∆ladS clpV1-gfp in co-culture with the indicated competitor . See Figure 4—figure supplement 2A for H1-T6SS activity . ( H ) P . aeruginosa cells lacking retS are unable to respond to T6SBT . Average cellular ClpV1-GFP intensity of P . aeruginosa ∆retS clpV1-gfp in co-culture with the indicated competitor . See Figure 4—figure supplement 2B for H1-T6SS activity . ( I ) A conserved residue in the periplasmic domain of RetS is required for P . aeruginosa response to T6SBT . Average cellular ClpV1-GFP intensity of P . aeruginosa clpV1-gfp retSW90A cultivated in the presence of the indicated competitor . See Figure 4—figure supplement 2C for H1-T6SS activity . n = 3 fields; asterisks indicate significant differences when B . thai was present ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01510 . 7554/eLife . 05701 . 016Figure 4—figure supplement 1 . RsmZ expression is not stimulated by B . thai lacking an active T6SS . Representative TLFM sequences of P . aeruginosa p-rsmZ-gfp cultivated with B . thai ∆tssM-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01610 . 7554/eLife . 05701 . 017Figure 4—figure supplement 2 . PARA-associated increases in H1-T6SS activity depend on RetS but not LadS . ( A ) LadS is not required for increased activity of the H1-T6SS in the presence of B . thai . Percentage of P . aeruginosa clpV1-gfp ∆ladS cells with fluorescent foci during co-culture with the indicated B . thai strain . ( B ) Stimulation of H1-T6SS activity by B . thai requires RetS . Percentage of P . aeruginosa clpV1-gfp ∆retS cells with fluorescent foci during co-culture with the indicated B . thai strain . ( C ) A putative signal-binding RetS mutant does not respond to the presence of B . thai . Percentage of P . aeruginosa clpV1-gfp retSW90A cells with fluorescent foci during co-culture with the indicated B . thai strain . Error bars represent ±SD; n = 3 co-cultures . Asterisks indicate significant differences when B . thai was present ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 017 If the Gac/Rsm pathway was involved in PARA-mediated H1-T6SS activation , we would expect transcription of its associated sRNA molecules to be elevated in cells co-cultivated with B . thai ( Figure 4A ) . Indeed , using a chromosomal fluorescent reporter of rsmZ transcription , we found B . thai stimulates expression of this sRNA on a time scale consistent with other effects associated with PARA ( Figure 4C , D , Figure 4—figure supplement 1 , Video 4 ) . Cells exposed to B . thai ∆tssM-1 did not exhibit changes in rsmZ expression . To further explore the link between Gac/Rsm and PARA , we measured the expression of a validated direct target of RsmA that is unrelated to the H1-T6SS , magA ( Brencic and Lory , 2009; Robert-Genthon et al . , 2013 ) . As observed for H1-T6SS reporters , a translational chromosomal fusion of the magA promoter to gfp ( p-magA-gfp ) displayed increased expression in response specifically to co-cultivation with B . thai bearing an active interbacterial T6SS ( Figure 4E ) . Taken together with our β-galactosidase reporter results , these data show that PARA is a Gac/Rsm-mediated posttranscriptional response of P . aeruginosa to T6S in other bacteria . 10 . 7554/eLife . 05701 . 018Video 4 . RsmZ expression is elevated in the presence of B . thai bearing an active T6SS . TLFM sequences depicting expression of rsmZ in P . aeruginosa during interspecies co-cultures . P . aeruginosa p-rsmZ-gfp mCherry cells with B . thai ( left sequence ) and B . thai ∆tssM ( right sequence ) were imaged at 15 min intervals . Overlays of GFP and mCherry channels are displayed . P . aeruginosa cells were labeled with constitutive mCherry , thus cells appear yellow . The unlabeled B . thai cells are visible ( light green ) due to autofluorescence in the GFP channel . The same thresholds were applied to all background-subtracted GFP channels . See Figure 4D for quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 018 Next we examined upstream components of the Gac/Rsm pathway that might mediate PARA . In P . aeruginosa , the output of the Gac/Rsm pathway is modulated by three sensor kinases ( Figure 4A ) ( Jimenez et al . , 2012 ) . We began by investigating GacS , which directly phosphorylates GacA , the response regulator that activates rsmY and rsmZ expression ( Figure 4A ) . Consistent with our hypothesis that PARA requires Gac/Rsm mediated signaling , we found that a strain bearing an in-frame deletion of gacS fails to elevate H1-T6SS expression in response to T6SBT ( Figure 4F ) . The remaining Gac/Rsm-associated sensor kinases , RetS and LadS , play accessory roles in the Gac/Rsm pathway by positively or negatively regulating GacS phosphorylation of GacA , respectively ( Figure 4A ) . Our experiments showed that strains lacking ladS respond to T6SBT similarly to the wild-type ( Figure 4G , Figure 4—figure supplement 2A ) ; however , we found that PARA is abrogated in strains lacking retS ( Figure 4H , Figure 4—figure supplement 2B ) . This finding suggests that the relief of RetS repression of GacS-catalyzed phosphorylation of GacA is important in PARA induction . This might occur through direct binding of RetS to a signal produced by the presence of the competitor organism , or , alternatively , RetS could transduce a signal from an upstream sensor . To investigate these possibilities , we generated a P . aeruginosa strain bearing a chromosomally-encoded RetS variant containing an amino acid substitution of a highly conserved residue in the predicted periplasmic signal binding pocket ( W90A ) ( Jing et al . , 2010; Vincent et al . , 2010 ) . Consistent with RetS acting as the direct sensor for PARA , in the retSW90A background , neither ClpV1 expression levels nor activity were affected by co-cultivation with B . thai ( Figure 4I , Figure 4—figure supplement 2C ) . A general disruption of RetS function cannot be excluded; however , the finding that H1-T6SS expression levels in retSW90A do not approach those detected in a ∆retS mutant demonstrates that this allele retains partial function . Altogether , our findings suggest that RetS functions upstream in the Gac/Rsm pathway to mediate PARA . Gac/Rsm is a pathway generally noted as a regulator of antibiosis; its stimulation in pseudomonads can increase the expression of a variety of antibiotic factors in addition to T6S , including hydrogen cyanide , secreted hydrolytic enzymes , and phenazines ( Lapouge et al . , 2008 ) . We reasoned that the fitness of cells undergoing PARA is derived not only from an increase in expression and activity of the H1-T6SS , but also from increased levels of these co-regulated factors . To test this , we used growth competition assays and TLFM co-cultures to compare the fitness of ∆gacS to a strain lacking only the function of the H1-T6S pathway ( ∆tssM1 ) . Consistent with our hypothesis , the fitness of ∆gacS was reduced beyond that of ∆tssM1 only when in competition with either B . thai ( 34-fold reduction ) or E . cloacae ( 15-fold reduction ) bearing an active interbacterial T6S pathway ( Figure 5A , B ) . TLFM experiments further indicated a substantial increase in lysis of P . aeruginosa ∆gacS relative to wild-type or ∆tssM1 ( Figure 5C , Video 5 ) . These phenotypes are not due to a general growth defect , as neither P . aeruginosa ∆gacS nor ∆tssM1 exhibit a competitive defect when grown in co-culture with the parental strain ( Figure 5D ) . In total , these data indicate that PARA is a complex bacterial defense mechanism comprising the H1-T6SS and other Gac/Rsm-regulated factors of P . aeruginosa . 10 . 7554/eLife . 05701 . 019Figure 5 . Disruption of the Gac/Rsm pathway results in a profound fitness defect in interspecies co-culture . ( A–B ) A P . aeruginosa strain with an inactivated Gac/Rsm pathway displays fitness defects beyond a strain lacking H1-T6S . Outcome of interspecies growth competition experiments between the indicated P . aeruginosa and B . thai ( A ) or E . cloacae ( B ) strains . n = 3 co-cultures . C . I . , competitive index . PA , P . aeruginosa . BT , B . thai . EC , E . cloacae . ( C ) P . aeruginosa lysis promoted by T6SBT is increased in a strain lacking a functional Gac/Rsm pathway . P . aeruginosa lysis events from TLFM sequences were normalized to initial number of contacts with B . thai . See also Video 5 . n = 3 fields . ( D ) A gacS deletion does not alter growth rate . Outcome of intraspecies growth competition experiments between PAO1 and the indicated competitor strains under conditions identical to those used in ( A–B ) . n = 3 co-cultures . ( A–D ) Error bars represent ±SD; asterisks indicate significant differences between indicated groups ( p < 0 . 05 ) . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 01910 . 7554/eLife . 05701 . 020Video 5 . Inactivation of the Gac/Rsm pathway results in a loss of interbacterial fitness . TLFM sequences of the indicated P . aeruginosa mCherry strains cultivated with B . thai GFP . Overlays of GFP and mCherry channels are displayed . Lysing P . aeruginosa cells are outlined in white . Quantification is provided in Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 020 T6S-dependent interactions require direct cell–cell contact; therefore , we asked whether PARA is also contact-dependent . To examine this , we computationally sorted P . aeruginosa into populations contacting or not contacting B . thai during the course of TLFM experiments ( Figure 6A ) . Surprisingly , we found no significant difference in ClpV1-GFP levels between these populations of cells ( Figure 6B ) , suggesting that either PARA does not require immediate cell contact between P . aeruginosa and a competitor , or P . aeruginosa–competitor contacts can be sensed by non-contacting cells . To differentiate between these possibilities , we employed a population-level approach in which we measured PARA-associated phenotypes in P . aeruginosa cells following cultivation on an agar plate ( Figure 6C ) . Consistent with our microscopy experiments , PARA was detected in P . aeruginosa cultivated in the presence of B . thai , but not in monoculture or with B . thai ∆tssM-1 ( Figure 6D , E; Condition 1 ) . We next measured PARA in P . aeruginosa cells separated by a membrane from either a B . thai monoculture or a P . aeruginosa–B . thai co-culture . Strikingly , PARA was detected when P . aeruginosa was adjacent to contacting P . aeruginosa–B . thai mixtures , but not when adjacent to B . thai alone ( Figure 6D , E; Conditions 2 and 3 ) . Combined with our microscopy results , these data strongly suggest that during co-culture with P . aeruginosa , the activity of T6SBT , or T6SBT itself , generates a diffusible molecule that triggers PARA in surrounding cells . 10 . 7554/eLife . 05701 . 021Figure 6 . PARA is induced by a diffusible signal . ( A ) The PARA-associated increase in H1-T6SS expression is not contact-dependent . A representative region from a P . aeruginosa clpV1-gfp–B . thai co-culture following 120 min of growth is depicted . Cells masks are colored by cell identity ( left panel ) ; GFP intensity with B . thai cell positions outlined in white dashed lines ( right panel ) . Arrows indicate P . aeruginosa cells contacting ( C ) or not contacting ( NC ) B . thai . ( B ) Average cellular ClpV1-GFP expression for contacting and non-contacting subpopulations described in ( A ) . ( C ) Schematic depicting the experimental setup for ( D ) . ( D–E ) PARA induction requires proximity to contacting P . aeruginosa–B . thai cells . Bacterial growth was initiated as pictured in ( C ) . ClpV1-GFP was measured in populations on the membrane ( black dashed lines ) . Average cellular ClpV1-GFP expression ( D ) and percentage of cells with foci ( E ) was determined . ClpV1-GFP expression measured in co-cultures was normalized by subtracting P . aeruginosa monoculture measurements . Error bars represent ±SD; n = 3 fields . Asterisks indicate significant differences between indicated groups ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 021 One outcome of T6S-dependent interactions is cell lysis . Thus , we posited that lysed P . aeruginosa—a consequence of T6SBT activity—could be the source of the diffusible factor mediating PARA . In agreement with this hypothesis , when P . aeruginosa is cultivated with B . thai , but not B . thai ∆tssM-1 , there is a significant increase in the number of P . aeruginosa cells that undergo lysis ( Figure 7A ) . Furthermore , concurrent temporal analysis of H1-T6SS expression and cell lysis in TLFM sequences indicated that P . aeruginosa lysis precedes elevation of ClpV1-GFP expression and the initiation of B . thai cell death ( Figure 7B , Video 6 ) . To determine if lysed P . aeruginosa cells are sufficient to induce PARA , we incubated reporter strains with lysate derived from wild-type P . aeruginosa . Lysate derived from P . aeruginosa , but not from B . thai , stimulated the Gac/Rsm pathway and recapitulated downstream PARA-associated phenotypes ( Figure 7C , D ) . 10 . 7554/eLife . 05701 . 022Figure 7 . P . aeruginosa lysis is sufficient to induce PARA . ( A ) T6SBT promotes P . aeruginosa lysis . Lysis of P . aeruginosa was measured under TLFM conditions and data were normalized to contacts with B . thai . n = 3 fields . Asterisk indicates significant difference between B . thai and B . thai ∆tssM-1 ( p < 0 . 05 ) . ( B ) P . aeruginosa lysis precedes induction of H1-T6SS expression and B . thai lysis . Lysis ( left axis ) and fold increase in ClpV1-GFP ( blue line , right axis ) measured concurrently under TLFM conditions . ClpV1-GFP levels from P . aeruginosa–B . thai co-cultures were normalized to P . aeruginosa monocultures . Error bars and light blue shading , ± SD . n = 4 fields . See also Video 6 . ( C ) P . aeruginosa lysate stimulates the Gac/Rsm pathway . Average cellular p-rsmZ-GFP expression in P . aeruginosa cultivated on lysate-infused growth pads . Cellular GFP expression was calculated as described in Figure 2 . ( D ) H1-T6SS expression is stimulated by P . aeruginosa lysate . Average cellular ClpV1-GFP expression of P . aeruginosa cultivated on lysate-infused growth pads . ( C–D ) n = 3 fields; asterisks indicated significant differences between P . aeruginosa lysate and no lysate ( p < 0 . 05 ) . ( E ) Expression of Gac/Rsm-regulated proteins is increased in lysate-treated P . aeruginosa cells . Quantitative mass spectrometry was used to compare the proteome of PBS ( control ) and lysate treated P . aeruginosa . Previously identified Gac/Rsm targets are indicated and H1-T6SS proteins discussed in this study are labeled . Data derive from two biological replicates . ( F ) Lysate stimulates H1-T6SS-mediated killing of B . thai ∆tssM-1 . Outcome of interspecies growth competition experiments between the indicated P . aeruginosa and B . thai in the presence or absence of P . aeruginosa-derived lysate . Error bars represent ±SD; n = 3 co-cultures . Asterisk indicates significant difference between lysate and no lysate treatments ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 02210 . 7554/eLife . 05701 . 023Figure 7—source data 1 . Proteins and corresponding spectral counts identified by quantitative mass spectrometry for P . aeruginosa cells with and without lysate exposure . Filtering criteria are described in ‘Materials and methods’ . Mean and total spectral counts for the combined replicates are provided . Blue shading corresponds to Gac/Rsm targets as defined by Lory et al . ( Goodman et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 02310 . 7554/eLife . 05701 . 024Video 6 . Lysis of P . aeruginosa precedes an increase in H1-T6SS expression and B . thai lysis . TLFM sequences depicting lysis ( left ) and ClpV1-GFP expression ( right ) in a P . aeruginosa–B . thai co-culture . A mixture of P . aeruginosa clpV1-GFP mCherry and B . thai CFP were imaged at 5-min intervals . Right panel displays the background-subtracted GFP channel with P . aeruginosa cells outlined in red and B . thai outlined in blue . Left panel displays an overlay of mCherry ( P . aeruginosa ) and CFP ( B . thai ) channels; lysing P . aeruginosa ( white outlines ) and lysing B . thai ( magenta ) are indicated . See Figure 7B for quantification . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 024 Due to its complex nature , lysate has the potential to lead to non-specific changes in cellular physiology that could confound interpretation of a small set of individual reporters . To gain a more comprehensive view of the effects of self-derived lysate on P . aeruginosa physiology , we used quantitative mass spectrometry to measure changes in protein abundance at the proteome level . To our knowledge , the global impact of Gac/Rsm activation on the P . aeruginosa proteome has not been reported . In lieu of this , we utilized a microarray study published by Lory and colleagues ( ∆retS vs wild-type ) in order to generate a list of proteins under Gac/Rsm control . Given the short time scale of our experiment and the relatively slow rate by which many proteins are recycled , we focused on factors positively regulated by Gac/Rsm . Remarkably , despite constituting only 4 . 5% of the proteins detected in our proteome , known Gac/Rsm targets accounted for 49% of proteins induced greater than twofold by the addition of lysate ( Figure 7E ) . We performed a gene set enrichment analysis and found that both Gac/Rsm regulated proteins and H1-T6SS proteins are significantly enriched in cells treated with lysate ( Gac/Rsm: NES , 2 . 1 , FDR ≤ 0 . 2% , p > 0 . 01; H1-T6SS: NES , 1 . 6 , FDR ≤ 5 . 1% , p < 0 . 01 ) ( Mootha et al . , 2003; Subramanian et al . , 2005 ) . The remaining 51% of induced proteins could include novel Gac/Rsm targets and lysate-responsive factors outside of the Gac/Rsm regulon . Together , these data show that self-derived lysate activates the Gac/Rsm pathway and is sufficient to induce PARA . A predicted consequence of PARA is a diminished capacity of P . aeruginosa to kill competitor organisms that lack an active lytic pathway . Our finding that lysate is sufficient to induce PARA provided an opportunity to test this directly . Specifically , we asked whether artificial induction of PARA by lysate could stimulate P . aeruginosa killing of B . thai lacking an active interbacterial T6SS . To this end , we measured the effect of lysate-induced PARA on P . aeruginosa fitness in growth competition experiments with B . thai ∆tssM-1 . In agreement with our prediction , the fitness of P . aeruginosa increased approximately 2 . 5-fold in the presence of lysate ( Figure 7F , Figure 7—source data 1 ) . In summary , our data suggest that P . aeruginosa cells that undergo lysis as a result of interspecies antagonism serve as a signal for Gac/Rsm-mediated stimulation of antibiosis in the remainder of the population . Ho et al . recently demonstrated that E . coli cells possessing an active IncP-type conjugative type IV secretion ( T4S ) apparatus are targeted more efficiently by the H1-T6SS of wild-type P . aeruginosa than E . coli lacking this system ( Ho et al . , 2013 ) . The authors of this study proposed that—like T6SS dueling—this effect is a result of enhanced T6S activity against E . coli deriving from local membrane perturbations made by the incoming conjugative apparatus . We hypothesized that the effect observed could also be a result of PARA . To test this , we performed interbacterial growth competition experiments between P . aeruginosa and E . coli , E . coli containing the IncP-type RP4 conjugative plasmid used by Ho et al . , or E . coli bearing a mutant form of this plasmid lacking a functional T4S apparatus ( ∆traG ) ( Waters et al . , 1992; Pansegrau et al . , 1994 ) . Similar to earlier findings , we found that the H1-T6SS of P . aeruginosa reduces populations of T4S+ E . coli to a greater extent than T4S−E . coli ( Figure 8A , Figure 8—figure supplement 1 ) . However , contrary to observations made by Ho et al . using the P . aeruginosa ∆retS background , we found that removal of the TPP did not abrogate the ability of P . aeruginosa to differentially target T4S+ and T4S− E . coli with the H1-T6SS ( Figure 8A ) . 10 . 7554/eLife . 05701 . 025Figure 8 . The RP4-encoded IncP-type T4SS induces PARA through lysis of P . aeruginosa . ( A ) The TPP is not required for differential targeting of E . coli with a T4SS . Outcome of growth competition experiments demonstrating increased susceptibility of T4S+ E . coli to the H1-T6SS of P . aeruginosa . See also Figure 8—figure supplement 1 for genetic complementation data of the traG deletion . n = 4 co-cultures . ( B ) H1-T6SS expression is elevated in the presence of E . coli containing a T4SS . Average cellular ClpV1-GFP expression of the P . aeruginosa clpV1-gfp throughout co-culture with the indicated E . coli strains . ( C ) RetS is required for increased H1-T6SS expression . Average cellular ClpV1-GFP expression of the P . aeruginosa clpV1-gfp ∆retS throughout co-culture with the indicated E . coli strains . ( D ) E . coli bearing a T4SS stimulates rsmZ expression . Average cellular GFP levels of P . aeruginosa p-rsmZ-gfp in TLFM co-culture experiments with the indicated E . coli strains . ( B–D ) n = 3 fields; asterisks indicate significant differences between E . coli and E . coli RP4 co-cultures ( p < 0 . 05 ) . ( E ) The T4SS encoded on RP4 promotes P . aeruginosa lysis . Relative P . aeruginosa attB::lacZ lysis was measured following co-cultivation with the indicated E . coli strain by comparing extracellular to total β-galactosidase activity . Error bars represent ±SD; ( A ) and ( E ) n = 3 co-cultures; asterisks indicate significant differences between indicated groups ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 02510 . 7554/eLife . 05701 . 026Figure 8—figure supplement 1 . Genetic complementation of traG restores H1-T6SS-dependent targeting of E . coli RP4 . Outcome of growth competition experiment in which P . aeruginosa and E . coli were co-cultivated under T6SS-promoting conditions . n = 4 co-cultures . Asterisks indicates significant differences between groups ( p < 0 . 05 ) ; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 02610 . 7554/eLife . 05701 . 027Figure 8—figure supplement 2 . The RP4-encoded T4SS induces a PARA-associated increase in H1-T6SS activity . Percentage of cells containing ClpV1-GFP foci in P . aeruginosa clpV1-gfp ( A ) or ∆retS clpV1-gfp ( B ) cultivated with the indicated competitor . Error bars represent ±SD; n = 3 co-cultures . Asterisks indicate significant differences between co-cultures with E . coli and E . coli RP4 ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 027 Having ruled out a requirement for the TPP in T4S sensing by wild-type P . aeruginosa , we next measured PARA-associated phenotypes in P . aeruginosa–E . coli co-cultures . These experiments showed that P . aeruginosa cultivated with T4S+ E . coli exhibits increased rsmZ expression and elevated ClpV1–GFP levels in a manner dependent upon retS ( Figure 8B–D , Figure 8—figure supplement 2 ) . Moreover , these phenomena were observed at a time point similar to T6SBT-induced PARA . Our data suggest that the mechanism underlying PARA is lysis of P . aeruginosa cells by a competitor; however , to our knowledge , T4S has not been noted to cause lysis of recipient cells . To determine whether E . coli lyse P . aeruginosa in a T4S-dependent manner , we quantified extracellular β-galactosidase activity from co-cultures of T4S+ and T4S− E . coli with P . aeruginosa attB::lacZ . Consistent with PARA induction by E . coli involving cell lysis , we observed a strong enhancement of P . aeruginosa lysis by T4S+ E . coli ( Figure 8E ) . Together , these data demonstrate that the T4SS encoded on the RP4 plasmid induces lysis within a subset of P . aeruginosa cells , which in turn induces PARA , leading to H1-T6SS-dependent E . coli cell death . We have shown that self-derived signal ( s ) generated as a consequence of cell lysis activate the Gac/Rsm pathway of P . aeruginosa , and thus stimulate the production of antibiotic factors under its control . In growth competition experiments , the capacity to mount this multifaceted response grants P . aeruginosa a significant fitness benefit . Our results suggest that similar to multicellular organisms , injury to a bacterial colony can trigger the release of danger signals that lead to a coordinated response against the threat ( Matzinger , 1994; Kaczmarek et al . , 2013 ) . In this study , we focused on offensive factors under control of the Gac/Rsm pathway; however , defensive factors are also likely to be elaborated . For example , a consequence of Gac/Rsm activation in P . aeruginosa is the production of c-di-GMP , which activates exopolysaccharide production of P . aeruginosa ( Lee et al . , 2007; Irie et al . , 2012 ) . This increases cellular adhesiveness , which facilitates multicellular aggregates that are more resistant than planktonic cells to an assortment of antibacterial molecules and environmental stresses ( Colvin et al . , 2011; Billings et al . , 2013 ) . In the context of an infection such as the chronic lung infections that occur in cystic fibrosis patients , host-induced cellular lysis could activate Gac/Rsm and inadvertently convert cells to a state that is more resistant to killing by antibiotics and the immune system . If PARA was the sole mechanism contributing to enhanced T6S-dependent killing of bacteria with T6 or T4 systems , it would follow that P . aeruginosa cells lacking the sensor kinase RetS should target bacteria irrespective of these pathways . However , previous studies have shown that P . aeruginosa ∆retS retains some ability to differentially target T6S- and T4S-positive vs -negative cells ( LeRoux et al . , 2012; Basler et al . , 2013 ) . Thus , the response of P . aeruginosa to antagonism is comprised of a global response mediated by the Gac/Rsm pathway and a secondary T6S-specific element that is not fully understood . We speculate that this ability of a ∆retS strain is related to coordinated spatiotemporal localization of the apparatus among adjacent cells , and that these two mechanisms operate in concert to hone the offensive response of P . aeruginosa . PARA may constitute an initial adaptation in which cells perceive a threat in their proximity and increase expression of the H1-T6SS , followed by the orientation of effector translocation specifically toward aggressor cells . We find that RP4-containing E . coli cells induce lysis in P . aeruginosa , trigger PARA , and in turn are subject to increased antagonism by the H1-T6SS . This mechanism differs from the model put forth by the Mekalanos laboratory , which suggested that the TPP is required for the response of wild-type P . aeruginosa to an incoming conjugative apparatus ( Ho et al . , 2013 ) . A key finding in the prior study was that polymyxin B , an outer membrane-disrupting antibiotic , induces clpV1 foci formation in wild-type P . aeruginosa , but not in a strain lacking tagT . This finding , among other data involving strains in the ∆retS background of P . aeruginosa , led the authors to propose that membrane perturbations caused by an incoming T4 apparatus are sensed by the TPP . A tagT deletion strain intrinsically lacks H1-T6SS activity; therefore , interpreting its inability to respond to the antibiotic as evidence of TPP involvement is problematic ( Basler et al . , 2013; Casabona et al . , 2013; Ho et al . , 2013 ) . We found that H1-T6SS-active strains that lack the TPP ( ∆TPP ∆tagF ) display a generalized targeting defect , but retain the ability to discriminate T4S+ and T4S− E . coli . An alternative explanation for the findings of Ho et al . is that the application of polymyxin B promotes cell lysis , leading to PARA induction ( Barrow and Kwon , 2009; Ho et al . , 2013 ) . Attempts to test the validity of this explanation were confounded by pervasive cell death at the antibiotic concentration reported by the authors ( 20 μg/ml , ∼40-fold the minimum inhibitory concentration against P . aeruginosa PAO1 ) ( Barrow and Kwon , 2009 ) . The adaptive significance of RP4-induced lysis and its mechanistic basis remain to be resolved . This process could be an altruistic behavior of P . aeruginosa that both aborts the T4S-dependent transfer event and alerts surrounding kin , thus decreasing the probability of foreign DNA acquisition by the colony . However , a second possibility is that plasmids such as RP4 carry interbacterial antagonistic factors that provide fitness to their hosts under certain conditions . It is interesting to note that Ho et al . identified an RP4 transposon mutant that was not targeted by P . aeruginosa but retained a functional T4SS ( Ho et al . , 2013 ) . This insertion resides in trbN , a gene encoding a periplasmic transglycosylase . It is plausible given the requirement for both T4 structural genes and this peptidoglycan-degrading accessory factor , that the T4S apparatus facilitates the transfer of this protein to recipient cells , where it induces lysis . Alternatively , upon plasmid transfer , the product of trbN may stochastically trigger lysis in a small portion of recipient cells . Efforts to characterize Gac/Rsm-stimulating signal ( s ) have been performed primarily in P . fluorescens ( Lapouge et al . , 2008 ) . This organism is a close relative of P . aeruginosa , and the Gac/Rsm pathways of the two species share a number of characteristics including regulation of hydrogen cyanide production and an H1-T6SS-like pathway . Haas et al . have made a number of intriguing observations pertaining to the production and sensing of Gac/Rsm-stimulating signals in P . fluorescens that are consistent with our findings in P . aeruginosa . Most notably , they found that conditioned media extracts derived from dense cultures of P . fluorescens , and , to a lesser extent , P . aeruginosa and Vibrio cholerae , are sufficient to activate the Gac/Rsm pathway ( Dubuis and Haas , 2007 ) . In agreement with our results , this indicates that the signal can be self-produced; however , it also raises the intriguing possibility that the lysis of non-self bacteria may activate PARA . While we found that B . thai-derived lysate was not sufficient to stimulate PARA , it is possible that other organisms produce the signal . This could lead to a positive feedback loop by which killed competitor cells further stimulate P . aeruginosa to produce antibacterial factors . Despite decades of research , the chemical structure of the molecule ( s ) that stimulate the Gac/Rsm pathway of P . aeruginosa , presumably via interaction with the periplasmic ligand binding domains of its associated sensor kinases , remain unknown . Our own efforts to identify the signaling molecule ( s ) contained within P . aeruginosa lysate , which included assorted enzymatic treatments , have so far been unsuccessful . Structural studies of RetS revealed its periplasmic region bears a fold resembling known carbohydrate interaction domains . A similar domain is predicted in the ecto domain of LadS ( Vincent et al . , 2010 ) . Given that RetS is required for PARA transduction , it is tempting to speculate that cell-associated carbohydrate ( s ) are released upon lysis and serve as a signal that activates Gac/Rsm in P . aeruginosa ( Jing et al . , 2010 ) . The additional observation that extracts derived from multiple bacterial species can stimulate Gac/Rsm in P . fluorescens suggests two non-mutually exclusive hypotheses: that the molecule is broadly conserved or that the pathway has evolved to respond to a number of inputs ( Dubuis and Haas , 2007 ) . The existence of three sensor kinases that operate upstream in the Gac/Rsm pathway of P . aeruginosa supports the latter hypothesis . In what they referred to as ‘competition sensing’ , Cornforth and Foster recently proposed that bacterial stress responses include antagonistic components , and that these pathways have evolved to respond to threats posed by other bacteria ( Cornforth and Foster , 2013 ) . For example , in response to DNA damage , the SOS pathway stimulates colicin production in E . coli , and in P . aeruginosa , exogenous peptidoglycan fragments have been shown to stimulate quorum-regulated toxins ( Cascales et al . , 2007; Korgaonkar et al . , 2013 ) . Our study demonstrates that ‘competition sensing’ includes an antibacterial response to cellular damage in kin cells . While we found that T6S-dependent killing by P . aeruginosa is part of an antagonistic response to lytic threats , Borgeaud et al . reported that in V . cholerae , T6S is co-regulated with competence machinery and utilized for obtaining access to exogenous DNA ( Borgeaud et al . , 2015 ) . Together , these studies demonstrate how functionally conserved machinery can be incorporated into diverse cellular programs exhibited by bacteria . The ‘danger theory’ of eukaryotic immunity proposes that in addition to the foreign substances that they introduce , threats can be sensed by virtue of cellular damage and ensuing mislocalization of host factors ( Matzinger , 1994; Kono and Rock , 2008 ) . For instance , uric acid microcrystals , which form upon release of the molecule to the sodium-rich extracellular milieu , stimulate dendritic cell maturation ( Shi et al . , 2003 ) . Our study shows that bacteria can also recognize threats by sensing self-derived signals associated with cell damage ( Figure 9 ) . Moreover , we find that the response to such signals includes the activation of factors that combat the threat—akin to the stimulation of inflammation in eukaryotes . It remains to be determined whether danger sensing is common among bacteria . The Gac/Rsm pathway is conserved widely among Gram-negative γ-proteobacteria; however , the variability of genes under its control confounds a prediction of its general involvement in danger sensing ( Lapouge et al . , 2008 ) . An intriguing possibility is that bacteria can utilize a diversity of signaling systems to sense and respond to kin cell damage . 10 . 7554/eLife . 05701 . 028Figure 9 . Bacterial danger sensing . The model depicts antagonism between two species of bacteria , represented in green and brown . The green cells possess a danger sensing pathway; specifics of PARA are provided in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 028 All strains used in this study were derived from P . aeruginosa PAO1 ( Stover et al . , 2000 ) , B . thai E264 ( Yu et al . , 2006 ) , E . cloacae ATCC 13047 ( Ren et al . , 2010 ) , and E . coli XK1502 ( Panicker and Minkley , 1985 ) . Routine cultivation of bacteria was performed using Luria broth ( LB ) medium . A low salt formulation of LB in which no additional sodium chloride was added ( LS-LB ) , was used for plate-based co-culture and competition assays . For P . aeruginosa , media were supplemented with 25 μg/ml irgasan , 30 μg/ ml gentamycin , 75 μg/ml tetracycline , or 40 μg/ml X-gal ( 5-bromo-4-chloro-3-indolyl β-D-galactopyranoside ) as necessary , and counter selection for allelic exchange was performed on low-salt LB supplemented with 5% wt/vol sucrose . For B . thai , media were supplemented with 25 μg/ml irgasan and 200 μg/ml trimethoprim as necessary , and counter-selection for allelic exchange was performed on M9 minimal medium agar plates containing 0 . 4% glucose and 0 . 2% ( wt/vol ) p-chlorophenylalanine ( Chandler et al . , 2009 ) . For E . coli , media was supplemented with 15 μg/ml gentamycin , 200 μg/ml trimethoprim , 25 μg/ml chloramphenicol , and 150 μg/ml carbenicillin as necessary . Markerless deletions of genes in P . aeruginosa and B . thai were generated in frame by allelic exchange with the suicide vectors pEXG2 and pEX18gm ( P . aeruginosa ) , or pJRC115 ( B . thai ) ( Rietsch et al . , 2005; Chandler et al . , 2009 ) . SacB and pheS-A304G were used for counterselection in P . aeruginosa and B . thai , respectively . Deletion alleles were constructed by splicing together 500–600 bp regions flanking the gene to be deleted by overlap extension PCR , and cloning them into the appropriate vector . The open reading frame , except the first and last 4–8 codons of the gene , was replaced by the sequence 5′- TTCAGCATGCTTGCGGCTCGAGTT -3′ . A detailed list of strains and vectors used this study are provided in Tables 1 and 2 , respectively . 10 . 7554/eLife . 05701 . 029Table 1 . Strains used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 029OrganismGenotypeReferenceP . aeruginosa PAO1Type strain ( Stover et al . , 2000 ) attB::lacZ ( Mougous et al . , 2006a ) ∆tssM1 attB::lacZThis study∆retS attB::lacZ ( Mougous et al . , 2006a ) ∆retS ∆tssM1 attB::lacZThis study∆tse1 ∆tsi1 ∆tse2 ∆tsi2 ∆tse3 ∆tsi3 ∆tse4 ∆tsi4This study∆retS ∆tse1 ∆tsi1 ∆tse2 ∆tsi2 ∆tse3 ∆tsi3 ∆tse4 ∆tsi4This studyclpV1-gfp ( Mougous et al . , 2006a ) fha1-sfgfpThis studyclpV1-gfp attTn7::mCherryThis study∆pppA clpV1-gfp ( Silverman et al . , 2011 ) ∆TPP ∆tagF clpV1-gfpThis study∆pppA ( Mougous et al . , 2007 ) ∆TPP ∆tagF ( Silverman et al . , 2011 ) ∆tssM1 ( Silverman et al . , 2011 ) attTn7::PA0082-lacZ10-Gm ( Brencic and Lory , 2009 ) attB::PA0082-lacZ-tet ( Brencic and Lory , 2009 ) p-rsmZ-gfp attTn7::Gm-mCherryThis study∆retS clpV1-gfp ( LeRoux et al . , 2012 ) ∆ladS clpV1-gfpThis study∆gacS clpV1-gfp attTn7::Gm-mCherryThis studyp-magA-sfgfpThis studyretSW90AThis study∆gacSThis studyattTn7::Gm-gfp ( LeRoux et al . , 2012 ) attTn7:: Gm-mCherry ( LeRoux et al . , 2012 ) ∆tssM1 attTn7::Gm-mCherryThis study∆gacS attTn7:: Gm-mCherryThis studyB . thailandensis E264Type strain ( Yu et al . , 2006 ) ∆tssM-1This studyattTn7::Tp-PS12-mCherry ( LeRoux et al . , 2012 ) ∆tssM-1 attTn7::Tp-PS12-mCherryThis studyattTn7::Tp-PS12-GFP ( LeRoux et al . , 2012 ) ∆tssM-1 attTn7::Tp-PS12-GFPThis studyattTn7::Tp-CFPThis studyE . cloacae ATCC 13047Type strain ( Ren et al . , 2010 ) ∆tssM ( Whitney et al . , 2014 ) E . coli XK1502Type strain ( Panicker and Minkley , 1985 ) RP4 ( Pansegrau et al . , 1994 ) RP4 ∆traGThis studyRP4 ∆traG pPSV39This studyRP4 ∆traG pPSV39::traGThis study10 . 7554/eLife . 05701 . 030Table 2 . Plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05701 . 030PlasmidUtilityReferenceP . aeruginosa PAO1miniCtx::lacZconstitutive LacZ expression ( Vance et al . , 2005 ) pUC18T-miniTn7T-Gm-gfpconstitutive GFP expression ( Choi et al . , 2005 ) pUC18T-miniTn7T-Gm-mCherryconstitutive mCherry expression ( LeRoux et al . , 2012 ) pEXG2_∆PA4856retS deletion allele ( Mougous et al . , 2006a ) pEXG2_∆PA0077tssM1 deletion allele ( Mougous et al . , 2006a ) pEXG2_∆PA1844-5tse1 tsi1 deletion allele ( Russell et al . , 2011 ) pEXG2_∆PA2702-3tse2 tsi2 deletion allele ( Hood et al . , 2010 ) pEXG2_∆PA3484-5tse3 tsi3 deletion allele ( Russell et al . , 2011 ) pEXG2_∆PA2774-5tse4 tsi4 deletion allele ( Whitney et al . , 2014 ) pEXG2_PA0090-gfpclpV1 functional translational GFP fusion allele ( Mougous et al . , 2006a ) pEXG2_PA0081-sfgfpfha1 functional translational GFP fusion alleleThis studypEXG2_∆PA0075pppA deletion allele ( Mougous et al . , 2007 ) pEXG2_∆PA0070-0076tagQRST ppkA pppA tagF deletion allele ( Silverman et al . , 2011 ) attTn7::PA0082-lacZ10-GmtssA1 translational lacZ reporter ( Brencic and Lory , 2009 ) attB::PA0082-lacZ-tettssA1 transcriptional lacZ reporter ( Brencic and Lory , 2009 ) miniCtx_p-PA3621 . 1-gfprsmZ transcriptional GFP reporterThis studypEXG2_∆PA0928gacS deletion alleleThis studypEXG2_∆PA3974ladS deletion alleleThis studypEXG2_p-PA4492-gfpmagA translational GFP fusion alleleThis studypEXG2_PA4856W90AretSW90A alleleThis studyB . thailandensis E264pJRC115_∆BTH_I2954tssM-1 deletion alleleThis studypUC18T-miniTn7T-Tp-PS12-gfpconstitutive GFP expression ( Schwarz et al . , 2010 ) pUC18T-miniTn7T-Tp-PS12-mCherryconstitutive mCherry expression ( LeRoux et al . , 2012 ) pUC18T-miniTn7T-Tp-ecfpConstitutive CFP expression ( Choi et al . , 2005 ) E . coli XK1502RP4Naturally occurring plasmid encoding IncP-type T4SS ( Pansegrau et al . , 1994 ) E . coli RP4 ∆traGRP4 bearing traG deletionThis studypPSV39Expression vector ( Silverman et al . , 2013 ) pPSV39-traGIPTG-inducible TraG expression for complementationThis study Functional translational fluorescent fusions to ClpV1 and Fha1 at their native promoters were achieved by allelic exchange . Constructs were generated by amplification of 500–600 bps regions flanking the C-terminus of the gene . These regions were spliced together with a BamHI site replacing the stop codon and cloned into pEXG2 . Gfp and superfolder-gfp ( sfgfp ) containing a stop codon was cloned into the BamHI site for clpV1 and fha1 constructs , respectively . The translational fusion to MagA was generated using a similar strategy except that only the first 12 codons of magA were retained . A resulting clone was introduced to P . aeruginosa by allelic exchange . To generate P . aeruginosa attB::p-rsmZ-gfp , the rsmZ promoter was amplified , spliced to an unstable gfp variant , and cloned into mini-CTX2 ( Hoang et al . , 2000 ) . The resulting clone was introduced to P . aeruginosa by conjugation . The retSW90A strain was generated by amplification of 500–600 bps flanking the W90 residue of RetS . The W90A substitution was encoded on the overlap primers and the two regions were spliced together by overlap extension PCR and cloned into pEXG2 . The mutation was introduced to PAO1 by allelic exchange . Transcriptional and translational lacZ fusions to tssA1 , original described by Brencic et al . ( Brencic and Lory , 2009 ) , were introduced to P . aeruginosa by conjugation and transformation , respectively . Integration of constitutive fluorescent reporters at the attTn7 site of B . thai and P . aeruginosa was achieved by four-parental mating or transformation , respectively ( Choi et al . , 2005 ) . TraG on RP4 was replaced with a chloramphenicol resistance cassette by λ Red recombination ( Datsenko and Wanner , 2000 ) in E . coli CC1254 , resulting in RP4 ∆traG::chlmR ( RP4 ∆traG ) . Following PCR confirmation , RP4 ∆traG was transferred to XK1502 by P1 phage transduction . Overnight cultures were diluted 1:50 or 1:100 in LB and incubated with aeration at 37°C until an OD600 of 0 . 5–0 . 7 was reached . Cultures were concentrated fivefold and mixed 1:1 by volume with the indicated competitor . 1–2 µl of the bacterial suspension was spotted onto a 1 . 5% wt/vol agarose growth pads ( prepared using Vogel Bonner minimal media containing 0 . 2% wt/vol sodium nitrate and 0 . 01% wt/vol casamino acids ) and sealed . Microscopy data were acquired using NIS Elements ( Nikon ) acquisition software on a Nikon Ti-E inverted microscope with a 60× oil objective , automated focusing ( Perfect Focus System , Nikon ) , a xenon light source ( Sutter Instruments ) , and a CCD camera ( Clara series , Andor ) . Lysis and growth rate were measured from TLFM sequences acquired at 5-min intervals; expression was measured from 15-min interval TLFM sequences . At least three fields were acquired and analyzed for each experimental group , and experiments were performed independently multiple times . TLFM sequences were analyzed using previously described methods ( LeRoux et al . , 2012 ) . Briefly , cells were identified from phase images using a watershed algorithm . P . aeruginosa and competitor cell populations were distinguished by the constitutive expression of cytoplasmic mCherry in either P . aeruginosa or the competitor organism . Non-cell debris was excluded based on a size threshold and/or fluorescence gating . For display in figures only , the GFP channel was γ-transformed ( ClpV1-GFP , exponent = 1 . 2; Fha1-GFP , exponent = 1 . 3 ) , thresholded , and the Matlab colormap ( jet ) was applied . To calculate average cellular fluorescence , GFP intensity of non-cell regions was subtracted from GFP intensity of cell regions as defined by cell masks . To identify foci within cells , the cytoplasmic fluorescence of each cell was fit to an empirical model for cytoplasmic fluorescence . A cell template image was generated by applying a Gaussian blur ( radius 2 pixels ) to the square root of the distance transform applied to the cell mask . A least-squares fit of a constant times the cell template was then performed to the observed cell image . The noise was computed as the standard deviation of the cell intensity after subtracting the fits of the cytoplasmic fluorescence and the fits of all potential foci . The signal to noise of a potential focus was computed as the intensity of the brightest pixel divided by the noise of that cell , and was defined as a focus if the value exceeded an empirically determined threshold . Candidate foci were identified globally using a watershed algorithm , then assigned to cells whose masks overlapped the focus position . The average and standard deviation of the percentage of P . aeruginosa cells with foci calculated for three fields is plotted . For contact-dependence analyses , cell neighbors were defined as cells with a boundary within a 2-pixel radius of the current cell ( LeRoux et al . , 2012 ) . Growth rate and cellular lysis were determined from datasets in which cells were linked over time based on frame-to-frame overlap of bright-field images . Doubling times—defined as minutes between birth and death—were calculated for cells that arose from a division after the start of the experiment and divided prior to the end of the experiment . A lysis event was defined as an 80% decrease in fluorescence intensity of a single cell between consecutive frames and was measured for in strains expressing a constitutive fluorophore ( mCherry , GFP , or CFP ) . For figures in which end-point P . aeruginosa lysis was presented , the number of P . aeruginosa cells that lysed was normalized to the initial number of P . aeruginosa–B . thai contacts to control for small fluctuations in cell density and P . aeruginosa–competitor ratio . For all end-point growth competition assays , except when noted , overnight cultures were used . Cultures were washed and mixed at the indicated volumetric ratios . Five microliters of the resulting mixture was spotted on a nitrocellulose membrane placed on a LS-LB 3% wt/vol agar plate and incubated at 37°C . For intraspecies growth competition experiments , P . aeruginosa attB::lacZ was used as the donor strain background , experiments were initiated at a 1:1 donor:recipient ratio , and were plated on LB containing X-gal for enumeration . Growth competition experiments with B . thai competitors were initiated at a donor:recipient ratio of 5:1 ( interspecies wild-type and T6S-dependent fitness , Figure 1; TPP strain experiments , Figure 3 ) or 1:1 ( ∆tssM1 and ∆gacS fitness , Figure 5A ) and plated on LB containing gentamycin ( 30 µg/ml ) for B . thai selection and trimethoprim ( 200 µg/ml ) for P . aeruginosa selection . For E . cloacae competitors , which encode lacZ , log-phase cultures were mixed at a ratio of 8:1 ( TPP strain experiments , Figure 3—figure supplement 2 ) or 1:1 ( ∆tssM-1 and ∆gacS fitness , Figure 5B ) and plated on media containing X-gal for P . aeruginosa–E . cloacae enumeration . For growth competition experiments with E . coli competitors , a P . aeruginosa attB::lacZ background was used , experiments were initiated at a donor:recipient ratio of 1:2 , incubated for 3 hr , and plated on LB containing X-gal . P . aeruginosa strains bearing previously validated chromosomally encoded transcriptional or translation fusions to tssA1 ( PA0082 ) were utilized to quantify transcription and translation ( see also Tables 1 and 2 ) ( Brencic and Lory , 2009 ) . Overnight cultures were washed , mixed at a 1:2 ratio with media ( monoculture ) , B . thai , or B . thai ∆tssM-1 , then spotted on a nitrocellulose membrane placed on a LS-LB 3% wt/vol agar plate . Following a 3 hr incubation at 37°C , cells were harvested in PBS , washed , and assayed for relative levels of β-galactosidase activity using the Galacto-Light Plus Reporter Gene Assay System ( Life Technologies ) . Bacterial suspensions composed of overnight cultures , washed once and mixed at a 2:1 P . aeruginosa:B . thai ratio were spotted as depicted in Figure 6C on LS-LB 3% wt/vol agar . Following 3 hr of incubation at 37°C , bacteria growing on top of the filter ( Figure 6C , black dashed lines ) were resuspended in LB and imaged by microscopy to determine PARA induction . To exclude dead P . aeruginosa cells from our analysis , which were prevalent in Condition 1 , a P . aeruginosa strain constitutively expressing mCherry was used and only mCherry-positive cells were considered . Stationary phase P . aeruginosa or B . thai cultures were pelleted and resuspended in growth media or PBS before sonication . Colony forming units ( c . f . u . ) of cultures and lysates were enumerated to verify that >95% of cells were lysed . For TLFM experiments , an agarose pad prepared as described above was infused with either lysate or PBS ( control ) . The concentration of lysate in the agarose pad was approximately 5 × 104 lysed cells/µl . Reporter strains were cultivated to OD600 of 0 . 5–0 . 7 , spotted on the lysate-containing agarose pad , and imaged . For the growth competition assays , lysate or PBS ( control ) was spotted on a LS-LB 3% wt/vol agar growth plate , a nitrocellulose filter was placed on top , and indicated mixtures of cells were spotted directly over the lysate . LB plates were prepared by applying either lysate ( 25-fold concentrated P . aeruginosa at an OD600 of 0 . 5–0 . 7 resuspended in PBS and lysed by sonication ) or PBS ( control ) at 2% ( vol/vol ) . A concentrated P . aeruginosa clpV1-gfp culture ( at OD600 0 . 5–0 . 7 ) was spotted on the nitrocellulose membrane placed on each plate then incubated for 2 . 5 hr at 37°C . Cells were harvested in PBS , washed , then stored at −80°C . Duplicate biological samples were prepared for mass spectrometry analysis as described previously ( Whitney et al . , 2014 ) . The semi-quantitative technique of spectral counting was used to determine the relative abundance of identified proteins in each sample as described in Whitney et al . ( Liu et al . , 2004; Whitney et al . , 2014 ) . Only proteins present in both biological replicates and with a sum of 20 spectral counts or greater across all replicates were considered in our analysis . Overnight cultures of P . aeruginosa attB::lacZ were washed , mixed 1:2 with E . coli strains , and spotted on a nitrocellulose membrane placed on a NS-LB 3% wt/vol agar plate . Following 3 hr of incubation at 37°C , levels of extracellular and total β-galactosidase activity were determined as previously described ( Chou et al . , 2012 ) . The percentage of lysed P . aeruginosa lysis was determined by normalizing extracellular to total β-galactosidase activity . All TLFM datasets were analyzed by two-way ANOVA using a Bonferroni correction for testing multiple hypotheses . End-point assays were analyzed using a two-tailed Student's t test . Asterisks indicate significance at p < 0 . 05 . Proteomic data were analyzed using Gene Set Enrichment Analysis ( GSEA ) ( Mootha et al . , 2003; Subramanian et al . , 2005 ) . Genes positively regulated by the Gac/Rsm pathway were defined as those found differentially regulated in a prior study comparing mRNA levels of ∆retS and wild-type P . aeruginosa ( Goodman et al . , 2004 ) . 1000 permutations of the analysis were performe d across both phenotype and gene set , and the ratio of classes metric was used for ranking genes .	Bacteria live in diverse and changing environments where resources such as nutrients and space are often limited . They have thus evolved many survival strategies , including competitive and cooperative behaviors . In the first case , bacteria antagonize or prevent the growth of other microorganisms competing with them for resources , such as by generating antibiotics that specifically target rivals . During cooperation , bacteria may coordinate the production of compounds that have a shared benefit for members of their community . In multicellular organisms , some cell types sense harmful microorganisms by the injury they cause in neighboring cells . This triggers a process that can lead to the production of molecules that kill the invaders and factors that promote the repair of cellular damage . An equivalent process has so far not been described for single-celled organisms such as bacteria . However , bacteria often live in structured groups containing many different species . In this type of growth environment , the ability of bacteria to sense when others of their species are attacked and to respond by taking measures to defend themselves could improve their chances of survival . Now , LeRoux et al . reveal that the bacterium Pseudomonas aeruginosa is able to detect ‘danger signals’ released when neighboring P . aeruginosa cells are killed by other bacteria . These signals trigger a response in surviving cells by turning on a pathway that controls a number of antibacterial factors . These include the production of the so-called ‘type VI secretion system’ , a molecular machine that delivers a potent cocktail of antibacterial toxins directly into nearby bacteria . This process , which LeRoux et al . have named ‘P . aeruginosa response to antagonism’ , or PARA for short , enables P . aeruginosa to thrive when grown with competing bacterial species . P . aeruginosa is notorious for infecting the lungs of people with the genetic disease cystic fibrosis , as well as chronic wounds often found in people with diabetes . In both cases , when P . aeruginosa is present , the numbers of other , often less harmful organisms , tend to decrease . PARA may be one reason for the success of P . aeruginosa in these multi-species infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2015	Kin cell lysis is a danger signal that activates antibacterial pathways of Pseudomonas aeruginosa
The general transcription factor TFIID comprises the TATA-box-binding protein ( TBP ) and approximately 14 TBP-associated factors ( TAFs ) . Here we find , unexpectedly , that undifferentiated human embryonic stem cells ( hESCs ) contain only six TAFs ( TAFs 2 , 3 , 5 , 6 , 7 and 11 ) , whereas following differentiation all TAFs are expressed . Directed and global chromatin immunoprecipitation analyses reveal an unprecedented promoter occupancy pattern: most active genes are bound by only TAFs 3 and 5 along with TBP , whereas the remaining active genes are bound by TBP and all six hESC TAFs . Consistent with these results , hESCs contain a previously undescribed complex comprising TAFs 2 , 6 , 7 , 11 and TBP . Altering the composition of hESC TAFs , either by depleting TAFs that are present or ectopically expressing TAFs that are absent , results in misregulated expression of pluripotency genes and induction of differentiation . Thus , the selective expression and use of TAFs underlies the ability of hESCs to self-renew . The specification of tissues and organs in development depends upon the spatially and temporally accurate execution of gene expression programs , much of which is regulated at the level of transcription . The factors involved in the accurate transcription of eukaryotic structural genes by RNA polymerase II can be classified into two groups . First , general ( or basic ) transcription factors ( GTFs ) are necessary and can be sufficient for accurate transcription initiation in vitro ( for review , see Thomas and Chiang , 2006 ) . These basic factors include RNA polymerase II itself and at least six GTFs: TFIID , TFIIA , TFIIB , TFIIE , TFIIF and TFIIH . The GTFs assemble on the core promoter in an ordered fashion to form a pre-initiation complex ( PIC ) . Transcriptional activity is greatly stimulated by the second class of factors , promoter-specific activator proteins ( activators ) . In general , activators are sequence-specific DNA-binding proteins whose recognition sites are typically present upstream of the core promoter . Activators work in large part by increasing PIC formation but can also act through other mechanisms , such as accelerating the rate of transcriptional elongation , promoting multiple rounds of transcription and directing chromatin modifications ( reviewed in Green , 2005; Fuda et al . , 2009; Weake and Workman , 2010 ) . A long-held view of transcription activation is that specificity arises from the differential expression and activity of activators , which function through the common basic transcription machinery . However , it is now clear that the differential expression and use of basic transcription factors can also contribute to eukaryotic gene regulation ( reviewed in Davidson , 2003; Hochheimer and Tjian , 2003 ) . This notion is most dramatically illustrated by a variety of studies focused on the GTF TFIID , a multi-subunit complex composed of the TATA-box-binding protein ( TBP ) and a set of ∼14 TBP-associated factors ( TAFs ) . One of the earliest clues about the differential function of TFIID came from studies in yeast demonstrating distinct classes of protein-coding genes that differ by their dependence on and recruitment of TAFs ( Kuras et al . , 2000; Li et al . , 2000 ) . Subsequently , similar classes of TAF-dependent and -independent genes were identified in mammalian cells ( Raha et al . , 2005; Tokusumi et al . , 2007 ) . Consistent with the existence of TAF-independent promoters , more recent studies have found that TAFs are depleted upon terminal differentiation of muscle ( Deato and Tjian , 2007; Deato et al . , 2008 ) and liver ( D'Alessio et al . , 2011 ) . TFIID diversity is also promoted by tissue-specific variants of TAFs as well as TBP derivatives referred to as TBP-related factors ( reviewed in D'Alessio et al . , 2009; Müller et al . , 2010 ) . Human embryonic stem cells ( hESCs ) are a good example of a specialized cell type that is regulated by a unique transcriptional network . Two characteristic properties of hESCs , pluripotency , a capacity to differentiate into all fetal and adult cell lineages , and the ability to undergo symmetrical self-renewing divisions , are largely controlled at the transcriptional level ( reviewed in Chen and Daley , 2008 ) . In undifferentiated hESCs , pluripotency genes such as OCT4 ( also called POU5F1 ) , NANOG and SOX2 are expressed , whereas genes involved in differentiation are transcriptionally inactive ( reviewed in Sun et al . , 2006; Pan and Thomson , 2007 ) . Decreased expression of pluripotency genes induces differentiation ( Niwa et al . , 2000 ) , and thus proper transcriptional regulation is essential for self-renewal of undifferentiated hESCs . Despite intense efforts to identify hESC-specific activators involved in the transcriptional regulatory network of pluripotency , there has been relatively little analysis of GTFs in general and TFIID in particular . Here we find that both the composition and promoter occupancy patterns of hESC TAFs are highly unusual . We go on to show that this selective expression and use of TAFs establishes a transcriptional program required for hESC self-renewal . In a search of published expression datasets ( Abeyta et al . , 2004 ) , we found that several TAFs of the canonical TFIID complex were apparently not expressed in hESCs . To investigate this possibility , we analyzed expression of 13 TAFs by immunoblotting lysates from H9 cells , a well-characterized hESC line . As a control , we also analyzed TAF expression in HeLa cells , which have been extensively used to study TFIID composition and function . The immunoblot of Figure 1A shows , as expected , that all 13 TAFs were expressed in HeLa cells . By contrast , hESCs clearly expressed TAFs 2 , 3 , 5 , 6 , 7 and 11 , whereas expression of TAFs 1 , 4 , 8 , 9 , 10 , 12 , and 13 was undetectable . Interestingly , TAF6 is expressed in both cell types , but the isoform present in H9 cells is predominantly the short delta form , whereas in HeLa cells , the major TAF6 isoform is the larger , alpha/beta form . The specificity of each TAF antibody was confirmed by RNA interference ( RNAi ) -mediated knockdown ( Figure 1—figure supplements 1 and 2 ) . We observed a similar TAF expression pattern in a second hESC line , H1 cells ( Figure 1—figure supplement 3 ) . Quantitative RT-PCR ( qRT-PCR ) analysis comparing TAF mRNA levels in HeLa and H9 cells correlated with the immunoblotting results ( Figure 1B ) . Unlike the TAFs , all other GTFs analyzed were comparably expressed in HeLa and H9 cells ( Figure 1C ) . Based upon these results we conclude that only six of the canonical TFIID TAFs are present in hESCs . 10 . 7554/eLife . 00068 . 003Figure 1 . Undifferentiated hESCs express only a subset of TFIID TAFs . ( A ) Immunoblot analysis showing TAF levels in HeLa cells and H9 hESCs . β-actin ( ACTB ) was monitored as a loading control . ( B ) qRT-PCR analysis monitoring TAF expression in H9 cells relative to HeLa cells . A ratio of 1 ( indicated by the red dotted line ) indicates no difference in expression . Data are represented as mean ± SEM . ( C ) Immunoblot analysis showing levels of GTFs in HeLa cells and H9 hESCs . α-tubulin ( TUBA ) was monitored as a loading control . ( D ) Immunoblots showing TAF and TBP levels in H9 hESCs following induction of differentiation by retinoic acid treatment for 0 , 3 or 6 days . OCT4 and NES were monitored as controls . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 00310 . 7554/eLife . 00068 . 004Figure 1—figure supplement 1 . Confirmation of specificity of TAF antibodies by RNAi-mediated knockdown in H9 hESCs . Immunoblot analysis showing TAF levels in H9 hESCs 48 hr after transfection with a control luciferase ( Luc ) or TAF siRNA . β-actin ( ACTB ) was monitored as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 00410 . 7554/eLife . 00068 . 005Figure 1—figure supplement 2 . Confirmation of specificity of TAF antibodies by RNAi-mediated knockdown in HeLa cells . Immunoblot analysis showing TAF levels in HeLa cells 48 hr after transfection with a control luciferase ( Luc ) or TAF siRNA . β-actin ( ACTB ) was monitored as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 00510 . 7554/eLife . 00068 . 006Figure 1—figure supplement 3 . TAF expression levels in H1 hESCs . Immunoblot analysis showing TAF levels in H1 hESCs and , as a comparison , H9 hESCs and HeLa cells . β-actin ( ACTB ) was monitored as a loading control . The results demonstrate an identical pattern of TAF expression in both H9 and H1 hESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 006 We next asked whether differentiation of hESCs results in a change in TAF composition . Toward this end , H9 cells were treated with retinoic acid to induce differentiation and TAF expression was analyzed by immunoblotting . Figure 1D shows , as expected , that following retinoic acid treatment , expression of the pluripotency factor OCT4 was lost and NES , a neuroectoderm marker , was induced . Significantly , TAFs 1 , 4 , 8 , 9 , 10 , 12 , and 13 , which are not expressed in undifferentiated H9 cells , were induced following retinoic acid treatment . TAFs 2 , 3 , 5 , 6 , 7 and 11 , which are expressed in undifferentiated H9 cells , were also present at a relatively constant level following retinoic acid treatment . To investigate whether the six hESC TAFs were associated in a stable complex , H9 cell nuclear extract was fractionated by sucrose gradient sedimentation and individual fractions analyzed for TAFs 2 , 3 , 5 , 6 , 7 and 11 by immunoblotting . The results of Figure 2A show that TAFs 2 , 6 , 7 and 11 co-sedimented with an apparent native molecular mass of ∼440 kDa . By contrast , TAFs 3 and 5 fractionated heterogeneously , and a substantial portion of both TAFs had an apparent molecular mass consistent with that of the free proteins ( ∼140 and ∼100 kDa , respectively ) . As expected , TBP , which is associated with multiple complexes involved in transcription by all three RNA polymerases , fractionated heterogeneously . Notably , however , a peak of TBP co-sedimented with TAFs 2 , 6 , 7 and 11 . 10 . 7554/eLife . 00068 . 007Figure 2 . hESCs have a non-canonical TBP-containing TAF complex . ( A ) Sucrose gradient sedimentation . H9 cell nuclear extract was fractionated , and individual fractions were analyzed for TAFs by immunoblotting . Arrows indicate elution peaks of protein standards . ( B ) Co-immunoprecipitation analysis . Nuclear extracts from H9 cells were immunoprecipitated with an anti-TBP or control ( anti-RAB2A ) antibody and the immunoprecipitate was analyzed for TAFs and TBP by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 007 To provide additional evidence for a stable , multi-subunit TAF complex , and to determine whether TBP was a component , we performed co-immunoprecipitation experiments . TBP was immunoprecipitated from H9 cell nuclear extracts , and the immunoprecipitate was analyzed by immunoblotting for the six hESC TAFs . The results of Figure 2B show that TBP was stably associated with TAFs 2 , 6 , 7 and 11 but not TAFs 3 and 5 . Collectively , these results indicate that H9 cells contain a non-canonical TAF complex composed of TBP and TAFs 2 , 6 , 7 and 11 but not TAFs 3 and 5 . The results of the expression analysis and biochemical experiments implied that the PIC formed on the promoters of active genes in hESCs would have an atypical TAF composition . To investigate this issue , we performed a series of chromatin immunoprecipitation ( ChIP ) experiments . In the first set of experiments we selected 10 transcriptionally active genes and performed ChIP analysis to monitor promoter occupancy by the six hESC TAFs . As a normalization standard , we also monitored occupancy of TBP and RNA polymerase II large subunit ( POL2 ) on these 10 promoters . As expected , we found that TBP and POL2 were present at comparable levels at each of the 10 promoters ( Figure 3A ) . However , the absolute level of TBP and POL2 bound to each promoter significantly varied among the 10 genes . Therefore , in this experiment and those presented below , TAF recruitment was normalized to the level of TBP occupancy . The ChIP results of Figure 3B revealed two groups of genes with distinct TAF promoter occupancy patterns . The first group , which we refer to as class I genes , were bound by TAFs 3 and 5 but not by TAFs 2 , 6 , 7 and 11 , whereas the second group , class II genes , were bound by all six hESC TAFs . 10 . 7554/eLife . 00068 . 008Figure 3 . Two classes of hESC genes based on TAF promoter occupancy . ( A ) Recruitment of TBP and POL2 to the promoters of 10 transcriptionally active genes ( GAPDH , INTS6 , MORF4L2 , SLC25A3 , TBP , TP53 , FSCN1 , OCT4 , PCNA , SFPQ ) , each represented by a data point , were monitored by ChIP in H9 cells . For each gene , enrichment of TBP or POL2 binding to the promoter was normalized to a no antibody control and for non-specific recruitment at a control locus . ( B ) ChIP analysis monitoring TAF recruitment to the promoters of the 10 genes in H9 cells . TAF recruitment is specified relative to TBP recruitment ( which was set to 1 ) , after normalizing to a no antibody control and for non-specific recruitment to a control gene desert locus . Data are represented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 008 To support this conclusion , we also compared TAF occupancy in HeLa and H9 cells across seven class I genes that are transcriptionally active in both cell types ( Figure 4A ) . As a control , we first analyzed a representative subset of TAFs that are expressed in HeLa but not H9 cells . The ChIP analysis of Figure 4B shows that TAFs 1 , 8 and 9 were readily detected on the promoters of genes in HeLa cells but , as expected , not H9 cells . Next , we analyzed the six TAFs that are expressed in both HeLa and hESCs . Figure 4C shows that in both cell types TAFs 3 and 5 were recruited to the promoters of the seven genes . By contrast we found that TAFs 2 , 6 , 7 and 11 were bound to the promoters of the seven genes in HeLa but not in H9 cells ( Figure 4C ) . These results indicate that on the same transcriptionally active gene the TAF composition is strikingly different in HeLa and H9 cells . 10 . 7554/eLife . 00068 . 009Figure 4 . Comparison of TAF promoter occupancy on an identical set of transcriptionally active genes in HeLa and H9 cells . ( A ) Recruitment of TBP and POL2 to the promoters of seven class I genes ( EEF1A1 , FOS , GAPDH , SLC25A3 , TBP , TLE4 and TP53 ) were monitored by ChIP in HeLa and H9 cells . Data were normalized as described in Figure 3A . ( B ) ChIP analysis monitoring recruitment of TAFs 1 , 8 and 9 to the promoters of the seven class I genes in HeLa and H9 cells , as described in ( A ) . Data are represented as mean ± SD . ( C ) ChIP analysis monitoring recruitment of TAFs 2 , 3 , 5 , 6 , 7 and 11 to the promoters of the seven class I genes in HeLa and H9 cells , as described in ( A ) . Data are represented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 009 We next performed a series of RNAi experiments to determine the relationship between TAF occupancy and transcriptional activity . The qRT-PCR results of Figure 5A show that siRNA-mediated knockdown of TAFs 3 and 5 in H9 cells ( Figure 5—figure supplement 1 ) greatly reduced expression of both class I and II genes . By contrast , siRNA-mediated knockdown of TAFs 2 , 6 , 7 and 11 decreased expression of class II , but did not affect expression of class I genes ( Figure 5B ) . Comparable results were obtained with a second , unrelated siRNA directed against each of the six TAFs ( Figure 5—figure supplement 2 ) . Collectively , these results establish a strong relationship between TAF occupancy and transcriptional activity in hESCs . 10 . 7554/eLife . 00068 . 010Figure 5 . A strong relationship between TAF occupancy and transcriptional activity in hESCs . ( A ) qRT-PCR analysis monitoring expression of class I and II genes in H9 TAF3 or TAF5 knockdown ( KD ) cells . Normalized Ct values were analyzed after subtracting the signal obtained with the control RN18S1 shRNA ( see ‘Materials and methods’ ) . Data are represented as mean ± SEM . ( B ) qRT-PCR analysis monitoring expression of class I and II genes in H9 TAF2 , 6 , 7 , or 11 KD cells . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01010 . 7554/eLife . 00068 . 011Figure 5—figure supplement 1 . siRNA-mediated knockdown efficiency of TAFs in H9 hESCs . qRT-PCR analysis monitoring TAF expression in H9 cells treated with two independent siRNAs ( A ) directed against the indicated TAF . TAF expression is specified relative to that obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01110 . 7554/eLife . 00068 . 012Figure 5—figure supplement 2 . Confirmation of TAF requirement for transcriptional activity using a second , unrelated siRNA . qRT-PCR analysis monitoring expression of class I and class II genes in H9 cells treated with a TAF3 , TAF5 , TAF2 , TAF6 , TAF7 or TAF11 siRNA . Expression of each gene is specified relative to that obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 012 To confirm and extend the ChIP results , we performed global ChIP-chip analyses . In these experiments we monitored , in parallel , promoter occupancy of TAFs 3 and 5 , TAFs 7 and 11 ( as representative members of the TBP/TAF 2 , 6 , 7 , 11 complex ) , TBP and POL2 . The overall results are summarized in Figure 6 . We first defined a group of ∼3600 high-confidence actively transcribed genes based upon co-occupancy of both TBP and POL2 at the transcription start-site ( Figure 6—figure supplements 1–3 ) . The vast majority of active genes had promoter-bound TAF3 and TAF5 ( Figure 6A ) . Significantly , a smaller fraction of active genes had promoter-bound TAF7 or TAF11 , and there was substantial overlap between TAF7- and TAF11-bound genes ( Figure 6B ) . As expected , the vast majority of genes bound by TAFs 7 and 11 were also bound by TAFs 3 and 5 ( Figure 6C ) . Representative examples of promoter occupancy maps for two class I ( SLC25A3 , CCNB2 ) and class II ( SFPQ , UCHL1 ) genes are shown in Figure 6D . 10 . 7554/eLife . 00068 . 013Figure 6 . Global ChIP-chip analysis of TAF occupancy . ( A ) Venn diagram showing the overlap between TBP- , POL2- , TAF3- and TAF5-bound genes . ( B ) Venn diagram showing the overlap between TBP- , POL2- , TAF7- and TAF11-bound genes . ( C ) Venn diagram showing the overlap between TAF3- and TAF5-bound genes and TAF7- and TAF11-bound genes . ( D ) Representative maps showing TAF3 , TAF5 , TAF7 , TAF11 , TBP and POL2 occupancy at the promoters of class I ( SLC25A3 and CCNB2 ) and class II ( SFPQ and UCHL1 ) genes . ( E ) Differences between class I and II genes with respect to promoter H3K4me3 . ( F ) Differences between class I and II genes with respect to average number of alternative promoters per ChIP-enriched site . ( G ) Representative promoter occupancy maps for two class II genes with alternative promoters , IFRD1 and CSF2RA . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01310 . 7554/eLife . 00068 . 014Figure 6—figure supplement 1 . Location of TBP , POL2 and TAF occupancy relative to the transcription start site . Histograms showing binding of TBP , POL2 and TAFs as a function of distance to the nearest transcription start site ( TSS ) . The results show that binding of TBP , POL2 or a TAF occurred predominantly near the TSS . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01410 . 7554/eLife . 00068 . 015Figure 6—figure supplement 2 . ChIP-chip peak overlap in independent replicates . Venn diagrams showing the degree of overlap between two independent replicates of the ChIP experiment . The number of peaks in each group are indicated by brackets . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01510 . 7554/eLife . 00068 . 016Figure 6—figure supplement 3 . Co-occupancy of TBP and POL2 with TAFs . Venn diagrams showing the degree of overlap between the number of genes whose promoters are bound by TBP and POL2 , and the number bound by a given TAF . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 016 To validate the ChIP-chip results , we selected a representative set of 44 genes and performed directed ChIP experiments using promoter-specific primer pairs . These validation experiments , which are shown in Figure 7 , confirmed the predicted TAF occupancy patterns for ≥85% of the genes tested . For example , consistent with the ChIP-chip results , there was no significant binding ( i . e . , no enrichment relative to the no antibody negative control ) of TBP or TAFs 3 , 5 , 7 or 11 to a group of inactive promoters predicted by the ChIP-chip analyses to not be bound by these factors ( Figure 7A ) . Most importantly , Figure 7B shows that TAF7 occupancy validated at 24 of 27 predicted sites ( 88 . 9% ) , TAF11 occupancy at 27 of 28 predicted sites ( 96 . 4% ) , TAF3 occupancy at 36 of 39 predicted sites ( 92 . 3% ) , and TAF5 occupancy at 39 of 39 predicted sites ( 100 . 0% ) . 10 . 7554/eLife . 00068 . 017Figure 7 . Validation of ChIP-chip results by directed ChIP experiments using promoter-specific primer pairs . ( A ) ChIP analysis monitoring binding of TBP and TAFs 3 , 5 , 7 and 11 to a representative set of promoters that , based on ChIP-chip analyses , were predicted not to be bound by these factors . The results were normalized to a no antibody control ( which was set to 1 ) . Data are represented as mean ± SEM . For four genes ( THOC1 , CBWD3 , RBM39 and ZNF260 ) , the inactive promoter ( A2 ) was analyzed in ( A ) , and the active promoter ( A1 ) was analyzed in ( B ) and ( C ) . ( B ) ChIP analysis monitoring binding of TAFs 3 , 5 , 7 and 11 to the promoters of a representative set of 47 genes predicted by the ChIP-chip analyses to be bound by some or all of the factors . Data are normalized to TBP . Cutoff for a ‘positive’ is >0 . 4-fold enrichment vs TBP ( red line ) . Data are represented as mean ± SEM . ( C ) ChIP analysis monitoring binding of TAF2 and TAF6 to 19 promoters that are bound by TAF7 and TAF11 . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 017 Figure 7B also shows that the overlap between TAF7- and TAF11-bound genes was higher than that predicted by the ChIP-chip analyses . Specifically , the results show that TAF7 was present at 7 of 7 predicted TAF11-only sites , and TAF11 was present at 5 of 6 predicted TAF7-only sites , indicating that TAF7 and 11 co-occupancy was 12 of 13 ( 92 . 3% ) . Moreover , as expected , genes bound by TAFs 7 and 11 were also co-occupied by TAFs 2 and 6 ( Figure 7C ) . For example , of the 16 promoters analyzed that were bound by TAF7 and TAF11 , TAF2 was present at 16 and TAF6 was present at 15 of these promoters . Collectively , the ChIP-chip analyses , in conjunction with the results described above , confirm the existence of two groups of genes in hESCs whose promoters are bound either only by TAFs 3 and 5 ( class I genes ) or by all six hESC TAFs ( class II genes ) . Finally , we analyzed the ChIP-chip dataset in relation to previous genome-wide studies in hESCs ( Abeyta et al . , 2004; Boyer et al . , 2005; Guenther et al . , 2007 ) for features that might distinguish class I and class II genes and found two statistically significant differences ( Figure 6E , F; Table 1 ) . First , the promoters of class I genes had greater histone H3 lysine 4 trimethylation ( H3K4me3 ) than those of class II genes ( Figure 6E ) . Second , the fraction of genes with alternative promoters ( identified based upon UCSC Genome Browser annotations; see ‘Materials and methods’ ) was significantly higher for class II than for class I genes ( Figure 6F ) . Representative examples of promoter occupancy maps for two class II genes with alternative promoters are shown in Figure 6G . 10 . 7554/eLife . 00068 . 018Table 1 . Statistical analysis of ChIP-chip dataDOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 018FeatureAll TBP + POL2 sitesClass I genes ( p-value vs TBP + POL2 ) Class II genes ( p-value vs TBP + POL2 ) p-value ( Class I vs Class II ) Alternative promoters ( average number ) 0 . 7170 . 686 ( 0 . 008 ) 0 . 934 ( 2 . 64 × 10−67 ) 1 . 73 × 10−58Bidirectional promoters ( percent ) 23 . 96125 . 315 ( 0 . 005 ) 28 . 186 ( 0 . 007 ) 0 . 110Promoter occupancy ( percent ) H3K4me391 . 48289 . 214 ( 8 . 72 × 10−05 ) 82 . 721 ( 4 . 24 × 10−20 ) 1 . 16 × 10−13OCT43 . 6843 . 919 ( 0 . 728 ) 3 . 064 ( 1 . 000 ) 0 . 293NANOG12 . 43813 . 448 ( 0 . 006 ) 13 . 725 ( 0 . 698 ) 0 . 861SOX210 . 24311 . 471 ( 9 . 64 × 10−05 ) 11 . 642 ( 0 . 474 ) 0 . 901Promoter elements ( percent ) TATA box4 . 8346 . 365 ( 5 . 15 × 10−15 ) 7 . 353 ( 0 . 012 ) 0 . 335BRE88 . 99992 . 413 ( 6 . 27 × 10−24 ) 92 . 892 ( 0 . 001 ) 0 . 650Initiator99 . 60899 . 928 ( 0 . 000 ) 100 . 000 ( 1 . 000 ) 0 . 402MTE0 . 6790 . 611 ( 1 . 000 ) 0 . 735 ( 1 . 000 ) 0 . 626DCE99 . 84399 . 964 ( 0 . 422 ) 99 . 877 ( 1 . 000 ) 0 . 402 Finally , we analyzed whether the unusual composition of TAFs was important for the characteristic ability of hESCs to maintain an undifferentiated state and self-renew . Figure 8A shows that shRNA-mediated knockdown of each of the six hESC TAFs ( Figure 8—figure supplement 1 ) induced differentiation , as evidenced by a decreased percentage of alkaline phosphatase-positive colonies . To confirm this conclusion , we also tested whether knocking down hESC TAFs would induce differentiation by analyzing expression of a diverse set of differentiation markers: AFP ( endoderm ) , CGB7 ( trophoectoderm ) , IGF2 ( mesoderm ) , NES ( ectoderm ) and SOX1 ( neuroectoderm ) . Figure 8B shows that depletion of each hESC TAF resulted in up-regulation , to varying extents , of these differentiation markers . Comparable results were obtained with a second , unrelated shRNA or siRNA directed against each of the six TAFs ( Figure 8—figure supplements 2 and 3 ) . Finally , the induction of differentiation following knockdown of hESC TAFs was also evidenced by decreased expression of the pluripotency genes NANOG ( Figure 8C and Figure 8—figure supplement 4 ) and OCT4 ( Figure 5 and Figure 5—figure supplement 2 ) . Thus , the hESC TAFs are required to maintain the undifferentiated state . 10 . 7554/eLife . 00068 . 019Figure 8The composition of hESC TAFs is required for maintenance of the undifferentiated state . ( A ) Percent of H9 TAF knockdown ( KD ) colonies staining with alkaline phosphatase . Data are represented as mean ± SD . ( B ) qRT-PCR analysis monitoring expression of differentiation markers ( AFP , CGB7 , IGF2 , NES and SOX1 ) in H9 cells treated with a TAF siRNA . Values are relative to those obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SEM . ( C ) qRT-PCR analysis monitoring expression of NANOG in H9 cells treated with a TAF siRNA . Values are relative to those obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SEM . ( D ) qRT-PCR analysis monitoring TAF1 expression in H9 cells transfected with a plasmid expressing TAF1 or , as a control , empty vector . Expression of TAF1 was monitored 48 hr following transfection . TAF1 expression is specified relative to that obtained with the empty vector , which was set to 1 . Data are represented as mean ± SEM . ( E ) Alkaline phosphatase staining of H9 colonies ectopically expressing TAF1 or , as a control , vector . Data are represented as mean ± SD . ( F ) qRT-PCR monitoring expression of differentiation markers ( AFP , CGB7 , IGF2 , NES and SOX1 ) in H9 cells ectopically expressing TAF1 . Values are relative to those obtained in H9 cells expressing vector , which was set to 1 . Data are represented as mean ± SEM . ( G ) qRT-PCR monitoring expression of class I and II genes in H9 cells ectopically expressing TAF1 . Data are represented as mean ± SEM . ( H ) Immunoblot analysis showing OCT4 levels in H9 cells over-expressing TAFs , TBP or vector . ( I ) ChIP analysis monitoring recruitment of TBP , POL2 and TAF1 to the OCT4 promoter in H9 cells ectopically expressing TAF1 or vector . Data are represented as mean ± SD . ( J ) Schematic model . Some of the protein interactions shown are arbitrary . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 01910 . 7554/eLife . 00068 . 020Figure 8—figure supplement 1 . shRNA-mediated knockdown efficiency of TAFs in H9 hESCs . qRT-PCR analysis monitoring TAF expression in H9 cells treated with two independent shRNAs directed against the indicated TAF . TAF expression is specified relative to that obtained with a control non-silencing shRNA , which was set to 1 . Data are represented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 02010 . 7554/eLife . 00068 . 021Figure 8—figure supplement 2 . Validation of results presented in Figure 8A using a second , unrelated shRNA . Percent of H9 TAF knockdown ( KD ) colonies staining with alkaline phosphatase . Data are represented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 02110 . 7554/eLife . 00068 . 022Figure 8—figure supplement 3 . Validation of results presented in Figure 8B using a second , unrelated siRNA . qRT-PCR analysis monitoring expression of differentiation markers in H9 cells treated with a TAF siRNA . Values are given relative to that obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 02210 . 7554/eLife . 00068 . 023Figure 8—figure supplement 4 . Validation of results presented in Figure 8C using a second , unrelated siRNA . qRT-PCR analysis monitoring expression of NANOG in H9 cells treated with a TAF siRNA . Values are given relative to that obtained with a control luciferase siRNA , which was set to 1 . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 023 In a reciprocal set of experiments , we altered TAF composition by ectopically expressing a TAF that is not normally present in hESCs . We found that ectopic expression of TAF1 ( Figure 8D ) resulted in differentiation , as evidenced by a decreased number of alkaline phosphatase-positive colonies ( Figure 8E ) , the induction of differentiation markers ( Figure 8F , left ) , and decreased expression of NANOG ( Figure 8F , right ) . Interestingly , Figure 8G shows that ectopic TAF1 expression decreased expression of several class II genes , including as expected OCT4 , whereas expression of class I genes was either unaffected or in some instances increased modestly . Ectopic expression of several other TAFs not present in hESCs also resulted in loss of expression of OCT4 , a class II gene , but not ACTB , a class I gene ( Figure 8H ) . By contrast , ectopic expression of TBP or TAFs that are present in hESCs did not affect OCT4 levels . To investigate the basis for the decreased OCT4 expression , we performed ChIP analysis . Figure 8I shows that , following ectopic expression of TAF1 , TBP and POL2 were no longer recruited to the OCT4 promoter . Collectively , the results of Figure 8 show that altering the composition of hESC TAFs results in misregulation of gene expression and induction of differentiation . The absence of seven of the conventional TFIID TAFs suggested that hESCs contain an alternative TBP-containing complex . Our ChIP and biochemical experiments confirmed this possibility and revealed a model in which a stable complex containing TBP and TAFs 2 , 6 , 7 and 11 is recruited to active class II genes , and TAFs 3 and 5 are recruited independently to all active genes ( Figure 8J ) . Notably , this novel TBP-containing complex lacks TAF1 , which in the canonical TFIID complex interacts directly with TBP ( Chen et al . , 1994 ) , and TAF4 , which is essential for assembly of Drosophila TFIID ( Wright et al . , 2006 ) . However , consistent with the existence of a TBP/TAF 2 , 6 , 7 , 11 complex , previous studies have described interactions between TBP and TAF2 ( Verrijzer et al . , 1994 ) , TAF6 ( Weinzierl et al . , 1993 ) and TAF11 ( Lavigne et al . , 1996 ) , and between TAF7 and TAF11 ( Lavigne et al . , 1996 ) . Previous studies have shown that nine of the 14 TFIID TAFs contain a sequence motif homologous to histones , called the histone fold domain ( HFD ) , which mediates protein–protein interactions ( reviewed in Cler et al . , 2009; Papai et al . , 2011 ) . These nine TAFs can form five specific heterodimers: TAF3–10 , TAF6–9 , TAF4–12 , TAF8–10 and TAF11–13 . Several of these heterodimers are thought to be important for the assembly and structure of TFIID . hESCs contain only three HFD-containing TAFs ( 3 , 6 and 11 ) , which cannot form any of the five known heterodimers . This observation strongly suggests there are major differences in assembly and structure of TFIID and the TBP/TAF 2 , 6 , 7 , 11 complex . An important question raised by our results is the basis by which TAFs are differentially recruited to class I or class II promoters . Studies in yeast have shown that differential recruitment of TAFs can be due to promoter-bound activators , core promoter elements , or both ( Shen and Green , 1997; Li et al . , 2002 ) . In this regard , TAF2 and TAF6 have been shown to interact with promoter sequence elements ( reviewed in Maston et al . , 2006 ) , which may contribute to differential promoter recognition by the TBP/TAF 2 , 6 , 7 , 11 complex . In addition , TAF3 contains a PHD finger domain that can bind to the H3K4me3 chromatin mark ( Vermeulen et al . , 2007 ) , which is enriched at class I promoters ( Figure 6E ) . Understanding the basis by which basic transcription factors are differentially recruited to promoters on a genome-wide scale appears to be a particularly challenging problem . For example , it is still not understood what distinguishes TAF-dependent and TAF-independent promoters in yeast ( Kuras et al . , 2000; Li et al . , 2000 ) , why some basic transcription factors , such as mediator , are bound at only some promoters ( Fan et al . , 2006 ) , or the basis by which chromatin-modifying complexes are selectively recruited to promoters ( Ng et al . , 2002 ) . Another question arising from our findings is whether there are functional differences that distinguish class I and class II genes . Gene ontology analysis did not reveal any functional category that was differentially enriched in either class I or class II genes ( data not shown ) . For example , although several pluripotency factors including OCT4 ( Figure 3B ) and NANOG ( Figure 10 ) are encoded by class II genes , we found other pluripotency factors , such as SOX2 , DPPA4 and KLF4 , that are encoded by class I genes ( Figure 10 ) . There was , however , a statistically significant increase in alternative promoters at class II genes ( Figure 6F ) . Notably , alternative promoter use has been suggested to play a role in generating tissue-specific transcripts ( Kimura et al . , 2006; Kolle et al . , 2011; Pal et al . , 2011 ) . 10 . 7554/eLife . 00068 . 025Figure 10 . Classification of additional pluripotency genes as either class I or class II . ChIP analysis monitoring TAF recruitment to the promoters of four pluripotency genes in H9 cells . TAF recruitment is specified relative to TBP recruitment ( which was set to 1 ) , after normalizing to a no antibody control and for non-specific recruitment to a control gene desert locus . Data are represented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00068 . 025 Our ChIP experiments show that genes that are broadly expressed , such as housekeeping genes , can be activated by diverse core promoter recognition complexes in different cell types ( Figure 4C ) , revealing a remarkable plasticity of the transcription machinery . The core promoter sequence is identical in every cell , raising the possibility that in hESCs differences in activators or epigenetic signatures may be coordinated with the alterations in core promoter complexes . For example , in the myogenic program , core promoter recognition complex changes correlates with the presence of developmentally regulated activators ( Deato et al . , 2008 ) . We have found that altering the composition of hESC TAFs , either by RNAi-mediated knockdown of TAFs that are present or ectopic expression of TAFs that are absent , results in loss of pluripotency gene expression and induction of differentiation . Thus , the unusual composition of TAFs described here is required for the ability of hESCs to properly regulate gene expression , maintain an undifferentiated state , and self-renew . This conclusion is reinforced by the finding that the TAFs that are absent from undifferentiated hESCs are expressed following differentiation ( Figure 1D ) . The transcriptional induction of several differentiation markers , which are not normally expressed in hESCs , following knockdown of an hESC TAF can be explained either by the dispensability of the TAF for transcription of the marker , or incomplete knockdown enabling transcription to occur at reduced TAF levels . A characteristic feature of the switch of undifferentiated hESCs to the differentiated state is the loss of pluripotency gene expression . We have shown that ectopic expression of TAFs that are not present in undifferentiated hESCs results in transcriptional inactivation of pluripotency genes . Thus , the transcriptional induction of TFIID TAFs that are absent from undifferentiated hESCs may be at least part of the mechanism by which pluripotency genes are silenced following differentiation . As discussed above , previous studies have shown that terminal differentiation of muscle ( Deato and Tjian , 2007 ) and liver ( D'Alessio et al . , 2011 ) is accompanied by loss of TFIID TAFs . These and other findings have prompted speculation that during differentiation the canonical TFIID complex becomes progressively specialized ( reviewed in D'Alessio et al . , 2009 ) . However our results reveal a more complex model . Rather than starting from a complete , canonical TFIID and progressing to more restricted forms , undifferentiated hESCs start with a specialized , highly unique general transcription machinery , then switch to a period of complete TFIID before specialization of the transcription machinery again in some terminal differentiation programs . Collectively , these findings suggest that altering the composition of the basic transcription machinery in general and TAFs specifically may be a particularly powerful mechanism for developmental reprogramming . H9 ( WA09 ) and H1 ( WA01 ) hESCs were obtained from the UMass Human Stem Cell Bank and Registry . Cell lines were cultured on irradiated mouse feeder cells using DMEM/F12 media supplemented with KnockOut SR ( Invitrogen , Carlsbad , CA ) and basic FGF ( R&D Systems , Minneapolis , MN ) , and cultures were karyotyped every 10–15 passages . All hESCs used were between passages 40–65 . hESCs for ChIP experiments were harvested 5–6 days after passage to ensure high density , such that the mouse feeder cells represented less than ∼12% of the total material ( similar to what has been used in previously published hESC ChIP experiments; Boyer et al . , 2005 ) . For Figure 1D , H9 cells were passaged using Accutase ( STEMCELL Technologies , Vancouver , Canada ) and plated without feeders in DMEM containing 10% FBS and 1 µM all-trans retinoic acid ( Sigma , St . Louis , MO ) . For transfections , H9 cells were grown in mTesR1 media ( STEMCELL Technologies ) under feeder-free conditions on plates coated with Matrigel ( BD Biosciences , San Jose , CA ) . HeLa cells were maintained in DMEM supplemented with 10% FCS at 37°C and 5% CO2 . The mouse ESC line PGK12 . 1 , provided by N . Brockdorff , was cultured as previously described ( Penny et al . , 1996 ) . Cells were trypsinized , pelleted , and washed twice in PBS . Nuclei were isolated by incubating the cell pellet in lysis buffer [10 mM Pipes-K+ pH 6 . 9 , 0 . 2% NP-40 , 5 mM KCl , 1 . 5 mM MgCl2 , 2 µM ZnSO4 , 5% glycerol , 2% PEG 2000 , plus 1 mM PMSF; Complete Protease Inhibitor Cocktail ( Roche , Basel , Switzerland ) and phosphatase inhibitors ( Sigma P2850 and P5726 ) ] for 2 min on ice , and pelleted by centrifugation . Nuclei were resuspended in nuclear extract buffer ( NEB1: 25 mM PIPES pH 6 . 9 , 0 . 2% Tween 20 , 0 . 4 M NaCl , 1 . 5 mM MgCl2 , 2 µM ZnSO4 , 5% glycerol , 2% PEG 2000 , plus 1 mM PMSF; Complete Protease Inhibitor Cocktail and phosphatase inhibitors ) for 10 min on ice , followed by centrifugation to remove nuclear debris . Protein content of the nuclear extracts was measured by BCA assay . Nuclear extracts were separated on either 8% or 12% SDS-PAGE gels . Blots were probed with primary antibodies ( listed in Supplementary file 1A ) overnight at 4°C , washed five times in TBP plus 0 . 1% Tween ( TBST ) and then incubated with the appropriate HRP-conjugated secondary antibody for 1 hr at room temperature . Membranes were washed five times in TBST and visualized on autoradiography film after incubating with ECL reagent ( Supersignal West Pico or Supersignal West Femto; Thermo Scientific , Waltham , MA ) . Total RNA was isolated using TRIzol ( Invitrogen ) , then treated with DNase I ( Promega , Fitchburg , WI ) and repurified using RNeasy columns ( Qiagen , Hilden , Germany ) . Reverse transcription was performed using the SuperScript II Reverse Transcription Kit ( Invitrogen ) with random oligo priming , followed by quantitative real-time PCR using Platinum SYBR Green qPCR SuperMix-UDG with Rox ( Invitrogen ) on either an ABI 7500 or StepOne Plus Real-Time PCR System ( Applied Biosystems , Carlsbad , CA ) . Primer sequences are listed in Supplementary file 1B . For all reactions , inputs were normalized and the Ct values of samples were analyzed after subtracting the signal obtained with the non-silencing shRNA ( for RNAi ) or no antibody ( for ChIP ) controls . For TAF knockdown experiments , human 18S rRNA ( RN18S1 ) was used as the endogenous control , because its expression should not be affected by changes in TAF expression . Sucrose gradient sedimentation analysis was performed as described ( Tanese , 1997 ) . Briefly , 10–40% gradients were formed by layering 500 µl NEB1 ( see ‘Immunoblot analysis’ ) containing 10% , 20% , 30% , or 40% sucrose in a 11 × 34-mm centrifuge tube ( Beckman , Brea , CA ) and allowed to equilibrate at room temperature for 2 hr . Gradients were chilled , loaded with either 500 µg H9 nuclear extract ( adjusted to a volume of 200 µl ) or 200 µl molecular weight markers ( Sigma MW-GF-1000 ) , and centrifuged in a Beckman TLS-55 rotor at 50 , 000 rpm ( 214 , 000×g ) for 12 hr . Twenty-three fractions of ∼90 µl were collected . For the markers , 20 µl of each fraction was electrophoresed and Coomassie stained . For the H9 gradient fractions , 25 µl of even-numbered fractions were analyzed by immunoblotting . H9 nuclear extract ( 600 µg ) was incubated with 6 µg of anti-TBP antibody ( Santa Cruz , Santa Cruz , CA ) at overnight at 4°C . Immune complexes were captured on rabbit TrueBlot IP beads ( eBioScience , San Diego , CA ) , washed three times in NEB1 , and eluted by boiling 10 min in 2× SDS sample buffer . IP material was then analyzed for the presence of TAFs by immunoblotting ( see Supplementary file 1A ) , using the Rabbit IgG TrueBlot HRP-conjugated secondary antibody ( eBioScience ) . Cells were dual-cross-linked with ethylene glycolbis[succinimidyl succinate] ( EGS ) , and formaldehyde as described ( Zeng et al . , 2006 ) . Chromatin shearing and ChIP experiments were then performed essentially as reported ( Hart et al . , 2007 ) , with slight modifications . For each ChIP experiment , 500 µg chromatin ( based on BCA assay ) was pre-cleared with BSA- and ytRNA-blocked protein G agarose beads ( Millipore , Billerica , MA ) . The pre-cleared supernatant was then incubated with 5 µg of primary antibody ( see Supplementary file 1A ) at 4°C overnight . Immune complexes were precipitated with protein G-agarose beads , washed , eluted , and purified as described . ChIP products were analyzed by qRT-PCR using primers listed in Supplementary file 1B . Real-time PCR results were analyzed using QBasePlus software ( Biogazelle , Zwijnaarde , Belgium ) . Site-specific relative fold changes of ChIP-enriched samples were calculated by comparing the amplification threshold ( Ct ) value of a given ChIP sample at a specific target locus ( promoter ) with the amplification Ct of a no-antibody control at the same target locus being analyzed , and also with the same ChIP sample and no-antibody control sample Ct values at a non-recruiting control locus found in a gene desert on human chromosome 16 ( primers ‘GDM’ in Supplementary file 1B ) . For ChIP-chip , 200 ng of ChIP-enriched or no-antibody control chip DNA fragments were blunted using End-It DNA End-Repair Kit ( Epicentre Biotechnologies , Madison , WI ) , then ligated to unidirectional linkers ( annealed oligos oJW102: 5′-GCGGTGACCCGGGAGATCTGAATTC and oJW103: 5′-GAATTCCAGATC ) using Fast-link DNA ligation kit ( Epicentre Biotechnologies ) . Linker-adapted DNA was amplified for 18 rounds using high-fidelity Pfu polymerase , then purified . DNA was labeled with either Cy5-dCTP ( chip samples ) or Cy3-dCTP ( no antibody control ) in a second PCR amplification of 18 rounds . The labeled DNAs were purified on QIAquick columns ( Qiagen ) and the incorporation was checked by spectrophotometry . 20 pmol of each labeled DNA ( chip and control ) was combined and used to hybridize to a Human ChIP-chip 3 × 720K RefSeq Promoter Array ( Roche NimbleGen , Madison , WI ) using a hybridization kit , sample tracking controls , wash buffer kit and array processing accessories from NimbleGen . Arrays were scanned on an Agilent Scanner at 5 µm resolution . Data were analyzed to identify peaks of binding using Nimblescan software ( Roche NimbleGen ) with default settings . Further analysis of ChIP data was conducted using ChipPeakAnno ( Zhu et al . , 2010 ) . The list of sites were filtered to remove multiple peaks occurring in the same promoter , which resulted in a set of genes whose promoters are bound by each factor . The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO Series accession number GSE39312 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE39312 ) . Promoter maps shown in Figure 6 were generated using the UCSC Genome Browser ( genome . ucsc . edu ) . For alternative promoter analyses , our ChIP-chip datasets were compared with the AltEvents track on the UCSC Genome Browser to find the frequency of overlap between chip peaks and ‘AltPro’ annotations . For comparison with H3K4me3 , we used previously published data ( Guenther et al . , 2007 ) after converting that dataset to HG18 coordinates using Galaxy ( Goecks et al . , 2010 ) . Statistical significance of differences in occurrence was determined using Fisher's exact test . For transient siRNA transfections , 30 pmol of siRNA duplexes ( see Supplementary file 1C ) was mixed with 5 µl Lipofectamine RNAiMAX and OptiMEM Reduced-Serum media ( Invitrogen ) in a total volume of 500 µl , incubated for 20 min , and added to H9 cells 1 day after plating in a six-well plate ( i . e . , at ∼25% confluency ) along with 2 . 5 ml fresh media . After 48 hr , RNA was isolated or nuclear extract prepared . For stable shRNA knockdowns , H9 cells seeded in a six-well plate to 25% confluency were stably transduced with 200-µl lentiviral particles expressing shRNAs ( obtained from Open Biosystems through the UMMS RNAi Core Facility; see Supplementary file 1C ) in a total volume of 2 ml of mTesR1 media supplemented with 6 µg/ml polybrene . Media was replaced after overnight incubation to remove the polybrene and viral particles . Alkaline phosphatase staining was performed using the Alkaline Phosphatase Staining Kit ( Stemgent , Cambridge , MA ) . To assess differentiation , 500 colonies were evaluated for AP staining and the percent positively stained was calculated . All assays were performed in triplicate . Full-length open reading frames of human TAF genes were PCR-amplified from HeLa cDNA and cloned in-frame into pECFP vector ( Clontech Laboratories , Mountain View , CA ) . Junctions were sequenced to confirm the construction . For transfection , 2 µg plasmid was mixed with 6 µl FuGENE HD Transfection Reagent ( Roche ) and OptiMEM in a total volume of 100 µl , incubated for 15 min , and then added to H9 cells in a six-well plate at 25% confluency . After 48 hr , RNA was isolated or nuclear extract prepared .	Embryonic stem cells have two characteristic properties: they are able to differentiate into any type of cell , a property known as pluripotency , and they are able to replicate themselves indefinitely to produce an endless supply of new stem cells . Different genes code for the various proteins associated with these two properties , and understanding the behaviour and properties of stem cells in detail is a major challenge in developmental biology . In human embryonic stem cells that have not yet differentiated , the genes that code for the transcription factors involved in the self-renewal process are expressed , whereas the genes associated with differentiation are not active . However , if the expression of the genes for self-renewal is reduced , the process of differentiation will begin , and the embryonic stem cells will be able to produce any one of the 200 or so different types of cell found in the human body . All of this activity is orchestrated by proteins that oversee the transcription of specific regions of DNA into messenger RNA . Transcription is the first step in the process by which genes are expressed as proteins , and it cannot start until the relevant transcription factor binds to a stretch of DNA near the gene called the promoter . These transcription factors are complex structures that contain a central protein called TBP , which binds to the promoter , and 14 or so other proteins called TAFs . Maston et al . now report that the transcription machinery that regulates gene expression and self-renewal in human embryonic stem cells is different from that found in other types of cells , including embryonic stem cells taken from mice . In particular , they found that undifferentiated human embryonic stem cells contain only 6 of the 14 TAFs observed in other cells , although all 14 are present after differentiation . Moreover , for many active genes the transcription factors contained only two of these TAFs . There was also evidence for a new complex that contained the other four TAFs plus TBP . Maston et al . also demonstrated that the removal of just one of the six TAFs , or the addition of just one extra TAF , caused the process of differentiation to begin . This shows , they argue , that the unusual transcription machinery they have discovered is essential for the proper workings of human embryonic stem cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2012	Non-canonical TAF complexes regulate active promoters in human embryonic stem cells
Collision avoidance is critical for survival , including in humans , and many species possess visual neurons exquisitely sensitive to objects approaching on a collision course . Here , we demonstrate that a collision-detecting neuron can detect the spatial coherence of a simulated impending object , thereby carrying out a computation akin to object segmentation critical for proper escape behavior . At the cellular level , object segmentation relies on a precise selection of the spatiotemporal pattern of synaptic inputs by dendritic membrane potential-activated channels . One channel type linked to dendritic computations in many neural systems , the hyperpolarization-activated cation channel , HCN , plays a central role in this computation . Pharmacological block of HCN channels abolishes the neuron's spatial selectivity and impairs the generation of visually guided escape behaviors , making it directly relevant to survival . Additionally , our results suggest that the interaction of HCN and inactivating K+ channels within active dendrites produces neuronal and behavioral object specificity by discriminating between complex spatiotemporal synaptic activation patterns . Neurons within the brain receive information about the outside world through a continuous ever-changing stream of synaptic inputs . These inputs can arrive thousands of times a second spread out across tens or even hundreds of thousands of different synaptic locations . Ultimately , the primary task of a neuron is to filter out the irrelevant elements of this dynamic stream and extract from the noisy cascade features meaningful for the animal . While the importance of timing of synaptic inputs is well known , the role of the spatial pattern of dendritic inputs has received less attention . In fact , it is still an unsettled question whether neurons extract information embedded within the broader spatial patterns of ongoing synaptic inputs ( Grienberger et al . , 2015 ) . In support of this hypothesis , recent investigations demonstrate spatial patterning of excitatory and inhibitory synaptic inputs ( Wilms and Häusser , 2015; Bloss et al . , 2016; Gökçe et al . , 2016; Bloss et al . , 2018 ) and dendritic processes capable of discriminating between different such patterns ( Smith et al . , 2013; Weber et al . , 2016; Wilson et al . , 2016 ) . For instance , local synaptic clustering produces supralinear summation which enhances the selectivity of visual neurons ( Smith et al . , 2013; Wilson et al . , 2016 ) . Studies of clustering have focused on fast positive feedback , such as the dendritic spikes and NMDA receptors that amplify local patterns of synaptic inputs , thereby conferring directional selectivity to some retinal ganglion cells ( Sivyer and Williams , 2013; Poleg-Polsky and Diamond , 2016 ) . Recent results also illustrate the functional role of fine scale synaptic patterning in many neurons ( Druckmann et al . , 2014; Kleindienst et al . , 2011; Petreanu et al . , 2009; Takahashi et al . , 2012 ) , but whether neurons also discriminate between broad spatiotemporal patterns embedded across thousands of synaptic inputs remains largely unknown . As many neuron types receive an ongoing stream of tens of thousands of inputs spread across a dendritic arbor , the ability to discriminate between such synaptic patterns would markedly increase their computational power . Additionally , a neuron's computational task likely determines which aspects of the spatiotemporal pattern of synaptic activities are most relevant and constrains the nonlinear dynamics of the membrane potential in its dendrites ( Spruston , 2008; Ujfalussy et al . , 2015 ) . To address these issues , we focus on large-scale processing of synaptic inputs and the dendritic computations required for visual object segmentation in the context of collision avoidance behaviors . The spatiotemporal sequence of synaptic inputs relevant to collision avoidance is determined by the statistics of the approaching object . Objects approaching on a collision course or their simulation on a screen , called looming stimuli , produce a characteristic visual stimulus on the observer's retina , expanding coherently in all directions with increasing angular velocity . Discriminating this retinal pattern from that of optic flow or from that of an object approaching on a miss trajectory requires integrating information across many points in time and space . Among neurons capable of such discrimination ( Sun and Frost , 1998; Nakagawa and Hongjian , 2010; Liu et al . , 2011; de Vries and Clandinin , 2012; Dunn et al . , 2016; Klapoetke et al . , 2017 ) , the lobula giant movement detector ( LGMD , Figure 1A ) has been extensively studied: it is an identified neuron of the grasshopper optic lobe located three synapses away from photoreceptors ( O'Shea and Williams , 1974 ) . The LGMD responds maximally to looming stimuli ( Schlotterer , 1977; Rind and Simmons , 1992; Hatsopoulos et al . , 1995 ) with a characteristic firing rate profile ( Hatsopoulos et al . , 1995; Gabbiani et al . , 1999 ) ( Figure 1B ) that has been tightly linked to initiating escape behaviors ( Fotowat et al . , 2011 ) . This characteristic firing profile is maintained even when an approaching stimulus is embedded in a random motion background , suggesting that the LGMD may be able to effectively segment visual objects ( Silva et al . , 2015; Yakubowski et al . , 2016 ) . In contrast , the LGMD responds only weakly to a stimulus whose angular size increases linearly in time , corresponding to an object decelerating during approach ( Hatsopoulos et al . , 1995; Simmons and Rind , 1992 ) . Synaptic inputs onto the LGMD are physically segregated into three dendritic fields , two of which receive non-retinotopically organized inhibitory inputs ( Strausfeld and Naessel , 1981 ) . The third one , dendritic field A , receives excitatory inputs originating from each ommatidium ( facet ) on the ipsilateral compound eye in a precise retinotopic projection ( Krapp and Gabbiani , 2005; Peron et al . , 2009; Zhu and Gabbiani , 2016 ) ( Figure 1A ) . These excitatory synaptic inputs are segregated by ommatidia and arranged in columnar fashion over an entire visual hemifield , so the LGMD's dendritic arbor has access to the entire spatial visual pattern activated by an approaching stimulus . Like cortical neurons that often receive inputs from tens of thousands of synapses spread across their dendritic arbors , little is known on whether the LGMD detects the spatial patterning of its synaptic input . The precise retinotopy of field A ( Peron et al . , 2009; Zhu and Gabbiani , 2016 ) means that the spatial pattern of synaptic inputs is directly determined by that of the visual stimulus , offering the possibility to experimentally control the synaptic patterning in vivo by changing the spatial aspect of visual stimuli . Thus , we can examine both the LGMD’s ability to discriminate spatial patterns consisting of thousands of synaptic inputs and the dynamic membrane properties of its dendrites . Here , we study how the LGMD discriminates the spatial coherence of approaching objects , how hyperpolarization-activated cyclic nucleotide-gated nonselective cation ( HCN ) channels within the retinotopic dendrites interact with other membrane channels to enhance this discrimination , and how this neural selectivity influences the animal's ability to effectively avoid approaching predators . We started with the question of whether the spatial pattern of approaching objects influences escape behavior . To control the stimulus pattern , as in our earlier work ( Jones and Gabbiani , 2010 ) , we generated stimuli equivalent to standard looming stimuli but pixelated at the spatial resolution of photoreceptors on the retina , called ‘coarse’ looming stimuli ( Figure 1C , D ) . The LGMD receives similar synaptic excitation and responds equally for standard and coarse looming stimuli ( Jones and Gabbiani , 2010 ) . We could then alter the coherence of these stimuli with minimal change to the temporal pattern of activation experienced by individual photoreceptors by adding a random spatial jitter to each ‘coarse pixel’ ( Figure 1C ) . Spatial stimulus coherence was varied from random ( 0% ) to perfectly coherent ( 100% = standard or coarse looming stimulus; see Materials and methods; Videos 1 and 2 ) . Looming stimuli with full and reduced coherence were presented to unrestrained animals while recording the probability of escape jumps ( Figure 1E , Video 3 ) . Locusts showed a strong behavioral selectivity to spatial coherence; stimuli with less than 50% spatial coherence elicited no escape jumps , but jump probability increased rapidly with coherence above 50% ( Figure 1F ) . The firing rate of the LGMD was also highly sensitive to stimulus coherence , with sharply reduced spike count and peak spike rate at 0% coherence ( Figure 1G ) . Thus , the spatial coherence of an approaching object determines both the LGMD's response and the animal's decision of whether to escape . None of the known properties of the LGMD or its presynaptic circuitry could explain this spatial selectivity . Previous experiments showed that the strength of excitatory inputs encodes the temporal characteristics of the approaching object by tracking local changes in luminance independent of their spatial pattern ( Jones and Gabbiani , 2010 ) . Additionally , the spatial clustering of synaptic inputs that occurs with coherent stimuli reduced summation in simulations of passive LGMD dendrites ( Peron et al . , 2009 ) , as further elaborated below . The LGMD's selectivity for the spatial characteristics of an approaching object are therefore likely determined by active processing within the dendrites of field A . No active conductances , however , have yet been characterized within these dendrites . Evidence suggests that neither the fast Na+ nor Ca2+ channels that produce supralinear summation in other neurons are present there ( Jones and Gabbiani , 2010; Peron and Gabbiani , 2009 ) . Since in many cells HCN channels influence dendritic computations , and previous experiments suggested putative HCN channels within the LGMD ( Gabbiani and Krapp , 2006 ) , we hypothesized that HCN channels within field A might be involved in discriminating the spatial coherence of approaching objects . Specifically , HCN channels narrow the membrane's temporal and spatial integration window to excitatory synaptic currents over an extended dendritic harbor . As approaching objects expand toward collision time , the closing of HCN channels could broaden the integration window and thus provide a slow positive feedback mechanism to tune a neuron to the visual stimuli associated with approaching objects . To test for the presence of HCN channels , we used current and voltage steps , as well as application of known channel blockers and modulators ( Robinson and Siegelbaum , 2003 ) , during visually guided recordings from each of LGMD's three dendritic fields and near the spike initiation zone ( SIZ; Figure 2A ) . Hyperpolarization of field A produced a characteristic rectifying sag , which was abolished by the HCN channel blockers ZD7288 ( Figure 2B ) and Cs+ ( Figure 2—figure supplement 1A , B ) . Applying step currents that generated similar peak hyperpolarization ( Figure 2C , Materials and methods ) , produced a larger , faster sag in field A than in either field B or C , or near the SIZ . To quantify these results across our sample population , we measured sag amplitude from the peak hyperpolarization elicited by these step currents and fitted single exponentials to the sag time course ( Figure 2—figure supplement 1C ) . On average , sags recorded from field A without HCN channel block were significantly larger and faster than those recorded from other neuronal regions ( Figure 2D , E; compare Field A control , vs . trunk and Field B , C ) . Similarly , block of HCN channels by ZD7288 in field A removed these sags ( Figure 2D , E; compare Field A control vs . ZD7288 ) . This shows a greater effect of HCN channels’ conductance ( gH ) in field A , consistent with HCN channels being localized there and the current passively propagating to the rest of the LGMD . Within the dendrites , back-propagating action potentials decay with electrotonic distance from the SIZ . Their amplitude was thus used as a measure of electrotonic distance from the SIZ to the recording location . Recordings at different locations within the dendritic trunk and field A revealed an increase in sag with electrotonic distance from the SIZ ( Figure 2F ) suggesting a higher channel density distally in field A . To characterize the channel kinetics , field A dendrites were voltage clamped . As HCN channels are distinctive in their activation range and time course we were able to measure and fit their currents ( see Materials and methods; Figure 2—figure supplement 1D ) , revealing an activation curve ( Figure 2G ) and time constant ( Figure 2H ) similar to that of HCN2 channels ( Robinson and Siegelbaum , 2003 ) . Next , we tested modulation of the HCN channels by cAMP . Application of cAMP shifted the half-activation potential ( v1/2 ) of gH from −77 . 6 ± 3 . 8 to −73 . 4 ± 2 . 2 mV ( mean ±sd; Figure 2G ) and slightly increased activation at rest ( Figure 2I ) . Both changes , however , were not different from controls ( p=0 . 18 and p=0 . 076 , respectively ) . These observations are in agreement with a recent report of a decrease in exogenous cAMP modulation of HCN channels in vivo post-developmentally , likely due to saturation of naturally occurring cAMP levels ( Khurana et al . , 2012 ) . In contrast , ZD7288 application unambiguously abolished resting gH activation ( Figure 2I ) . To examine whether these HCN channels could be responsible for spatial discrimination , we presented visual stimuli before and after their pharmacological blockade . For standard looming stimuli , gH was excitatory with responses reduced by 61% after HCN blockade ( Figure 2J ) . Responses to localized luminance transients , however , were similar before and after blockade of gH ( Figure 2—figure supplement 2A–C ) . Next , we quantified the responses to looming stimuli of varying coherence in control and after HCN blockade by computing the spike counts elicited over each entire trial . Since the LGMD did not exhibit any significant spontaneous activity , changes in spike counts were entirely caused by changes in the stimulus coherence . Under control conditions , there was a large increase in LGMD response with stimulus coherence which was reduced after ZD7288 blockade ( 42% for coarse looming stimuli; Figure 2K ) . For each experiment , we defined coherence preference as the slope of the linear fit to the number of spikes fired by the LGMD as a function of coarse looming stimulus coherence . For every animal tested , the coherence preference was reduced after gH blockade , decreasing from a median of 0 . 24 to 0 . 06 spikes per percent coherence ( Figure 2L; similar results were observed for the peak firing rate , see Figure 2—figure supplement 2D , E ) . To compare this relationship across animals , we normalized responses to the control response to fully coherent coarse stimuli before averaging across animals ( Figure 2M ) . LGMD responses consistently increased less with stimulus coherence after gH block ( Figure 2M ) . After blockade , the mean response to all stimuli fell within ±1 sd of the mean control response to 0% coherence ( p=0 . 10; KW ) . As explained below , this change in selectivity was reproduced by a biophysical model of the LGMD ( Figure 2M , dashed lines ) . To ensure that these effects were intrinsic to the LGMD , we ascertained that blocking gH with intracellular Cs+ application also reduced the coherence selectivity ( Figure 2—figure supplement 2F; Materials and methods ) . Next , we compared the jump probabilities at each coherence level ( Figure 1F ) with the gH-dependent increase in firing for that coherence level ( difference between control and HCN block in Figure 2M ) . This revealed a strong correlation between physiology and behavior ( Figure 2N ) . Furthermore , responses to faster looming stimuli , which fail to produce escape behaviors before the projected time of collision ( Fotowat and Gabbiani , 2007 ) , showed a smaller gH-dependent increase in firing ( Figure 2—figure supplement 2G ) . Therefore , gH increased responses specifically to stimuli which evoke escape , suggesting that the gH-dependent enhancement produced the escape selectivity . Having found that gH-dependent increase in firing is strongly correlated with jump probabilities ( Figure 2N ) , we sought a direct test of the hypothesis that gH within the LGMD played a role in the animals’ escape from approaching objects . So , we blocked gH in the LGMD in freely behaving animals ( Materials and methods ) . As a control , we developed a chronic recording technique allowing us to monitor the descending LGMD output during escape behaviors before and after gH blockade . Blocking gH in the LGMD reduced escape behavior by 53% for standard looming stimuli compared to saline injection ( Figure 3A , left two dots ) . The coherence preference was also removed by blockade of gH: standard looming stimuli no longer produced a higher percentage of escape than reduced coherence stimuli ( Figure 3A , red dots ) . That these behavioral changes were caused by gH blockade within the LGMD was further confirmed by examination of the LGMD's firing pattern . gH blockade by ZD7288 decreased responses to both standard looming stimuli and 86% coherent stimuli ( Figure 3B , C ) . The reduction in firing in the freely moving animals was less than that in the restrained preparation ( 36% and 60% , respectively ) , which might be due to an incomplete block of gH after stereotactic injection compared to visually guided puffing ( see Materials and methods ) or differences in arousal state . To test this , we used the stereotactic injection procedure in restrained animals and saw a 56% reduction in looming responses ( Figure 3D ) suggesting the difference in firing rate change was more likely due to a difference in behavioral state . Our ability to produce a change in behavior of freely moving animals from blockade of gH was confirmed by simultaneous extracellular recordings revealing a LGMD firing rate change resembling that of intracellular drug application , verification that the surgical procedures did not reduce the response , and postmortem anatomical verification that drug application occurred within the region encompassing the LGMD’s dendrites ( Figure 3—figure supplement 1 ) . The precise biophysical mechanisms by which HCN channels could impart the selective enhancement of coherent stimulus responses are not immediately obvious . HCN channels are not known to increase summation of spatially coherent inputs , and often gH has net inhibitory effects ( Robinson and Siegelbaum , 2003; Poolos et al . , 2002 ) . To determine how HCN channels produced the selective enhancement of looming responses , we investigated the effects of gH on membrane excitability within field A . gH increased the resting membrane potential ( RMP ) by ~6 mV in field A , which would bring the neuron closer to spike threshold ( Figure 4A ) . Blockade also revealed gH to decrease input resistance by 50% and the membrane time constant ( τm ) by 30% ( Figure 4B , C ) , which should substantially reduce the temporal summation of excitatory postsynaptic potentials ( EPSPs ) , as occurs in cortical pyramidal neurons ( Magee , 1998; Mishra and Narayanan , 2015 ) . To confirm this point , we injected currents yielding membrane depolarizations with the same time course as EPSPs to generate ‘simulated EPSPs’ ( sEPSPs; Figure 4D ) . After gH blockade , summation from the first to fifth sEPSP increased for all tested delays ( Figure 4E; the dashed lines are from the biophysical model described below ) . Additionally , the integrated sEPSPs normalized by the integrated current increased by 77% ( Figure 4F ) . This normalized integral generates a measure of effective input resistance for the sEPSPs which was similar to the input resistance to step currents , but with a slightly larger increase after HCN blockade ( compare Figure 4B and F ) . Neither before nor after gH blockade was supralinear summation ever seen in LGMD dendrites . Thus , the mix of local excitatory and inhibitory electrotonic effects of gH does not provide any simple explanation for the large enhancement in looming responses or the conveyed coherence selectivity . It may seem counterintuitive that gH increased looming responses twofold despite decreasing sEPSP amplitude and temporal summation by half . To explain this apparent contradiction , we considered interactions between HCN and other dendritic channels . In several systems , HCN channels have indirect excitatory effects through inactivation of co-localized voltage-gated K+ channels ( Mishra and Narayanan , 2015; Khurana et al . , 2011; MacLean et al . , 2005; Amendola et al . , 2012 ) . To test whether this was also the case in dendritic field A of the LGMD , we measured visual responses in the presence of 4-aminopyridine ( 4AP ) , a blocker of inactivating K+ channels ( Storm , 1988 ) . Application of 4AP , either intracellularly or extracellularly , increased the resting membrane potential in field A by 2–5 mV and the spiking response and instantaneous firing rate to standard looming stimuli ( Figure 5A ) . Application of 4AP also increased responses to coarse looming stimuli of varying degree of coherence , but responses to fully coherent looming stimuli increased the least ( Figure 5B ) . A similar result was observed after normalizing responses to the control response to fully coherent coarse stimuli before averaging across animals ( Figure 5C ) . This relative increase in incoherent responses after blocking inactivating K+ channels was also reproduced in a biophysical model ( Figure 5C , dashed lines; see below ) . The complementary effects of HCN and K+ channels was best revealed by plotting their relative changes to looming responses , shown as the percent difference from block to control ( Figure 5D ) . Thus , while HCN channels predominantly boosted responses to coherent stimuli , inactivating K+ channels mainly decreased responses to incoherent ones . The increase in RMP caused by gH ( Figure 4A ) could result in a change in the resting inactivation level of the 4AP sensitive K+ channels . To test whether the effects of gH on coherent stimuli were primarily due to the shift in the RMP , we hyperpolarized the LGMD during visual stimuli to a potential like that achieved by HCN blockade ( ~6 mV , see above ) . However , lowering the RMP without the changes to input resistance and membrane time constant caused by gH blockade ( Figure 4B , C ) only produced a modest reduction in coherence preference , with responses to standard looming stimuli reduced by 20% vs . 61% after ZD7288 blockade ( p=0 . 03 , WRS; see above and Figure 5—figure supplement 1 ) . This reduction in coherence preference was less than that produced by blockade of either dendritic channel and was independent of stimulus coherence ( p=0 . 36 , KW; Figure 5D ) . This result corroborates the idea that dynamic changes of the gH conductance occurring during looming stimulation and their effects on electrical compactness and membrane time constant play a central role in coherence selectivity . To further confirm that the inactivating K+ channels were exerting a spatially dependent effect on synaptic integration , we measured subthreshold activity in dendritic field A before and after 4AP application while presenting looming stimuli with varying degrees of coherence . As illustrated in Figure 6A ( top ) , we measured during a given period of the visual stimulus ( vertical green lines ) the average membrane depolarization before and after application of 4AP ( grey horizontal lines ) . During this period , a distinct group of coarse pixels were decreasing in luminance . As illustrated in the two bottom panels of Figure 6A , we measured the mean angular distance of each currently changing pixel from the nearest previously darkened pixel ( red lines ) . A short distance example is depicted on the left and large one on the right of Figure 6A . In the control condition ( the top panels ) , the membrane potential was closer to the baseline for the larger angular distance ( compare left and right black traces ) . This decrease in depolarization was attenuated after 4AP application ( blue traces ) . Since mean angular distance increased on average with decreasing coherence , we could obtain a broad sample of distances by carrying out this analysis across trials and animals . This revealed that the membrane depolarization systematically decreased with increasing stimulus distance in the control condition , but this effect was abolished after 4AP ( Figure 6B ) . We repeated this process for a total of six distinct time periods for each looming stimulus . Throughout stimulus expansion , the more dispersed excitatory inputs were , the less dendritic depolarization they produced in control conditions , a feature absent after 4AP application ( Figure 6—figure supplement 1 ) . These results are summarized across the six time periods in Figure 6C , by normalizing angular distance and membrane potential depolarization since their ranges vary over the stimulus time course . We further quantified the change in depolarization caused by currently changing coarse pixels as a function of their mean angular distance to fully darkened ones by computing the slope of the linear fits between these two quantities ( Figure 6D ) . Smaller slopes were observed in the earlier time windows when angular distances were larger and only a few coarse pixels were changing and , vice-versa , larger slopes were observed in later time windows when distances were smaller , but more pixels were changing their luminance . For each time window , 4AP increased the depolarization produced in field A by more distant stimuli with an average slope difference of 1 . 05 mV per degree of visual separation ( Figure 6E ) . These experiments confirm that inactivating K+ channels selectively reduce excitation for the spatially dispersed inputs generated in dendritic field A by incoherent looming stimuli and thus contribute to the selectivity of the neuron to coherently expanding looming stimuli . Detailed biophysical modeling was employed to further understand the biophysical mechanisms by which HCN and inactivating K+ channels allow the LGMD to discriminate spatiotemporal input patterns based on coherence . First , we confirmed that a model of the LGMD with passive dendrites generated a smaller response to retinotopically arranged looming inputs than the same inputs impinging on random dendritic locations both in terms of the mean membrane potential and the instantaneous firing rate ( Figure 7A , top and bottom panels , respectively ) . This illustrates why implementing coherence preference is nontrivial: the spatially distributed excitatory inputs that occur during incoherent looming stimuli produce less reduction in driving force , thus generating a larger current from the same synaptic conductance . Adding HCN channels to the dendrites of this model while adjusting the leak conductance to maintain RMP and Ri , also resulted in stronger responses to spatially scrambled inputs ( Figure 7B ) . As suggested by the results of Figures 5 and 6 , the subsequent addition of inactivating K+ channels in dendritic field A reduced responses to the spatially scrambled inputs , bringing the model in broad agreement with experimental findings ( Figure 7C ) . More precisely , the model reproduced key experimental results , including the LGMD's preference for spatially coherent inputs and the reduction of this preference after block of gH ( Figure 2M; Figure 5D ) ; the electrotonic and summation effects of gH ( Figure 4E; Figure 7—figure supplement 1A–C ) ; the coherence-dependent increase in firing caused by blocking the inactivating K+ channels and their role in suppressing responses to incoherent stimuli ( Figure 5C , D ) . In the model , the inactivating K+ channel activity was similar to the KD current that has been hypothesized to influence dendritic integration in pyramidal and Purkinje neurons ( Storm , 1988; Hounsgaard and Midtgaard , 1989; Zagha et al . , 2010 ) . This similarity extended to its activity at rest , its influence on subthreshold integration within field A dendrites , its apparent slow inactivation , and its 4AP sensitivity . We thus call it KD-like . During coherent looming stimuli , inputs continue to impinge on nearby dendritic segments for a prolonged period , spreading slowly ( Figure 7D , top ) . With spatially incoherent stimuli , inputs spread out over a much larger region of the dendritic arbor ( Figure 7D , bottom ) . The dendritic branches receiving the concentrated inputs of a coherent loom depolarize more than the surrounding branches , while the spatially dispersed inputs of an incoherent loom produce a similar level of depolarization across the dendrites ( Figure 7E ) . The prolonged depolarization generated by a coherent loom causes HCN channels to close ( Figure 7F , dashed black line ) and the KD-like channels to inactivate ( Figure 7F , solid black line ) . The deactivation of HCN channels leads to increased spatial compactness and summation ( Figure 4 ) providing a slow positive feedback while the faster negative feedback provided by KD-like decreases as it inactivates . During spatially incoherent stimuli , however , HCN channels close less and the KD-like channels across the arbor undergo less inactivation ( Figure 7F , gray lines ) . In control conditions , KD-like inactivation is due to the resting depolarization from gH and activity-induced depolarization . After HCN blockade , the lower resting membrane potential reduces the baseline KD-like channel inactivation ( Figure 7F , red line ) so that even with spatially coherent inputs the channels never reach the same level of inactivation . The KD-like channel activation was highest for control looming stimuli and lowest for incoherent looming stimuli ( Figure 7—figure supplement 1D ) as it tracked the membrane potential ( Figure 7E ) . However , the overall conductance of the KD-like channels was lowest for coherent looming stimuli contributing to the higher response ( Figure 7—figure supplement 1E ) . Examination of the membrane currents generated by the channels reveals even larger differences . Toward the end of a looming stimulus , the dendrites approach the HCN channel reversal potential , and the net HCN current approaches zero ( Figure 7G , top ) . Conversely , the K+ driving force increases during the stimulus approach . As a result , the KD-like channels that remain activatable produce a larger current ( Figure 7G , bottom ) . For the coherent stimulus , this late depolarization occurs in the same dendritic region activated by the earlier inputs and since the nearby KD-like channels have already inactivated , it yields little increase in K+ current , irrespective of gH block . The incoherent inputs , however , impinge onto branches where the channels have not already inactivated , yielding a much larger current . In addition to these dendritic channels , the model also included low-threshold Ca2+ channels ( CaT ) and Ca2+-dependent K+ channels ( KCa ) near the SIZ that allowed the LGMD model to fire in bursts ( Gabbiani and Krapp , 2006; Peron and Gabbiani , 2009 ) . In both experimental data and simulations , responses to spatially coherent stimuli generated more sustained , non-burst firing than transient burst firing ( Figure 7H; Figure 7—figure supplement 1F , G ) . The model reproduced the trends in these data qualitatively rather than quantitatively ( see Discussion ) . The decrease in bursting for coherent stimuli was also dependent on CaT channel inactivation . Coherent stimuli produced a steady ramp up of membrane potential increasing CaT inactivation , while incoherent stimuli produced more sudden depolarization , producing bursts . To investigate the role played in the smooth ramp up of activity during looming stimuli by dendritic KD-like inactivation vs . SIZ CaT inactivation and the concomitant suppression of bursting , we plotted the two inactivation variables as a function of stimulus coherence . During the last 2 s before collision , when most firing occurred , the average inactivation of both CaT and KD-like channels increased with stimulus coherence ( Figure 7I ) . Yet , the slope of the best fit line for KD-like inactivation vs . stimulus coherence was three times as steep as that of CaT inactivation . This confirms the relative importance of KD-like in coherence selectivity but also shows that interactions of multiple channels underlie the coherence selectivity of the LGMD model . Figure 8 illustrates the interactions of the channels involved in coherence selectivity during object approach . Dendritic field A receives retinotopic inputs across a compartmentalized arbor . The resting gH contributes to this compartmentalization by decreasing the electrical compactness and membrane time constant . In the model , the selectivity arises from fast negative feedback of KD-like activation embedded in the context of two slow positive feedbacks: one from KD-like inactivation and the other from gH deactivation . Spatially incoherent visual stimuli generate spatially dispersed synaptic inputs that depolarize many branches , rapidly increasing negative feedback by activation of KD-like . This reduces its own slow inactivation and further depolarization generated by subsequent synaptic inputs ( Figure 6; Figure 6—figure supplement 1 ) . The spatially dispersed , transient excitation of incoherent stimuli generate transient depolarizations of the SIZ activating burst firing followed by KCa activation that inhibits sustained spiking . For spatially coherent stimuli , in contrast , the synaptic inputs continue to excite the same dendritic branches for a prolonged period . This prolonged local activation eventually causes KD-like inactivation and HCN deactivation , resulting in positive feedback by reducing K+ current and increasing EPSP amplitude and summation . Additionally , the slow increase in depolarization propagates to the SIZ where it inactivates CaT channels , reducing burst spiking and its subsequent negative feedback caused by KCa . Here , we provide , to the best of our knowledge , the first demonstration of selectivity to spatial coherence for an ecologically important escape behavior ( Figure 1 ) . Our results suggest that this spatial discrimination relies upon discrimination of the broad spatial statistics of synaptic inputs within the dendrites of a single neuron . To examine this issue , we characterized active conductances and studied how HCN and inactivating K+ channels produced selectivity for spatial coherence ( Figures 2 , 5 and 6 ) . Although our results suggest that spatial selectivity is in large part implemented within the LGMD’s dendritic arbor , they do not rule out additional presynaptic mechanisms . Furthermore , we blocked HCN channels in freely moving animals , demonstrating that the selectivity of escape behavior depends on HCN channels enhancing spatially coherent responses ( Figure 3 ) . Our experimental data suggest that HCN channels produce a selective enhancement for inputs generated by spatially coherent approaching objects . To the best of our knowledge , there are no previously described mechanisms by which ion channels could produce such spatial selectivity . While fast positive feedback from Na+channels , Ca2+ channels or NMDA receptors can enhance the impact of clustered synaptic inputs ( Takahashi et al . , 2012; Kleindienst et al . , 2011; Poleg-Polsky and Diamond , 2016; Weber et al . , 2016; Sivyer and Williams , 2013; Smith et al . , 2013; Wilson et al . , 2016 ) it remains unclear whether they could also provide a way to select for broad spatiotemporal patterns of synaptic inputs . Based on biophysical modeling , we developed a plausible hypothesis explaining the underlying mechanisms , schematically illustrated in Figure 8 . These mechanisms involve competition between depolarizing and hyperpolarizing conductances within a compartmentalized dendritic arbor , regulation of membrane potential to control levels of K+ and Ca2+ channel inactivation , and regulation of bursting . The model illustrated in Figure 8 was the simplest that reproduced the wide range of our experimental data . Detecting the differences in spatiotemporal patterns of synaptic inputs requires an electrotonically extended arbor . To test whether the dendritic morphology could be reduced without a loss of selectivity we compressed different dendritic regions into electrotonically equivalent cylinders . Despite containing the same conductances and the same passive properties as the full model , the simplified morphology markedly reduced the coherence selectivity ( Figure 7—figure supplement 1H ) . Although the specific kinetics and distributions of several channels in the LGMD remain to be characterized , the model is well grounded ( Material and methods ) . After extensive searches through parameter space we could not find other combinations of mechanisms that reproduced the experimental data as well . Simulations were conducted with altered kinetics of HCN closing and KD-like inactivation , and both faster and slower kinetics reduced the coherence selectivity ( Figure 7—figure supplement 1I ) . Yet , the model did not reproduce quantitatively all our experimental results: for example , it underestimated the amount of transient firing at high coherence and , vice-versa , overestimated its sustained firing ( Figure 7E; Figure 7—figure supplement 1F , G , dashed lines ) . One likely reason is that details of the bursting mechanisms may be imperfectly tuned in the model , due to the absence of a second calcium-sensitive K+ conductance ( Peron and Gabbiani , 2009 ) or still uncharacterized properties of an M current ( unpublished observations ) . Further confirmation of this will require future experimental tests of channel properties predicted by the model , including the precise location of HCN and KD-like channels within field A ( Figure 2F ) , that KD-like and CaT inactivate above −70 mV , and that KD-like inactivates slowly ( in the range of 0 . 3–2 s ) . HCN channels have long been known to influence dendritic integration in hippocampal pyramidal neurons ( Magee , 1998 ) , and KD as well ( Storm , 1988 ) . More recently , dendritic K+ channels have been found to compartmentalize dendrites , and it has been suggested that spatiotemporal interactions between HCN and K+ channels regulate neuronal excitability ( Harnett et al . , 2013; Harnett et al . , 2015; Mishra and Narayanan , 2015 ) . It is thus possible that selectivity for broad spatial synaptic input patterns arises in pyramidal neurons by mechanisms analogous to those described here . In thalamocortical neurons , HCN channels influence K+ and Ca2+ channel inactivation , thereby regulating bursting and excitability ( McCormick and Pape , 1990; McCormick , 1991 ) . HCN regulation of bursting has been tied to a rat model of absence epilepsy ( Ludwig et al . , 2003; Kole et al . , 2007 ) and may also contribute to human epilepsy ( Bender et al . , 2003 ) . In addition to possible disease states , HCN-dependent regulation of persistent or burst firing has also been involved in working memory ( Thuault et al . , 2013 ) . In summary , our results highlight how nonlinear dendritic conductances and their interactions confers the ability to reliably select synaptic patterns appropriate for the generation of visually guided escape behaviors . This highlights the computing power of individual neurons and may help design object segmentation algorithms for bio-inspired collision avoidance systems . As HCN conductances are ubiquitous , they may contribute to implement analogous computations in other species , including our own ( Bender et al . , 2003 ) . All experiments were performed on adult grasshoppers 7–12 weeks of age ( Schistocerca americana ) . Animals were reared in a crowded laboratory colony under 12 hr light/dark conditions . For experiments preference was given to larger females ~ 3 weeks after final molt that were alert and responsive . Animals were selected for health and size without randomization , and investigators were not blinded to experimental conditions . Sample sizes were not predetermined before experiments . For many experiments , a large number of experiments were conducted ( e . g . >100 experiments in Figure 2 ) . For technically difficult experiments ( e . g . Figure 3 ) , smaller sample sizes were used with enough replications to see a clear effect . The surgical procedure for intracellular recordings was described previously ( Peron et al . , 2009; Jones and Gabbiani , 2010 ) . For extracellular DCMD recordings in freely moving animals , we developed a novel chronic implant technique , allowing the same animals to be recorded over many days , based on previous methods ( Fotowat et al . , 2011 ) . Grasshoppers were fixed ventral side up and a rectangular window was opened in their thorax . Air sacs were removed and the trachea were carefully separated to expose the ventral nerve cords . Two teflon-coated stainless steel wires 50 µm in diameter were cut to a length of ~4 cm and fashioned into hooks with the coating removed from the inside edge of the crook ( supplier: California Fine Wire , Grover Beach , CA ) . The electrodes were implanted with the deinsulated region placed against the dorsomedial edge of the left nerve cord between the pro- and meso-thoracic ganglia . Slight tension was applied to the cord to maintain a fixed position against the wires , and the wires were set in place by waxing them to the left side of the thorax . The cuticle window was then closed and sealed with a wax-rosin mixture and Vetbond ( 3M , St . Paul , MN ) . A ground electrode made of the same wire as the hooks was placed outside on the thorax and embedded in the wax . All three wires were routed laterally and fixed to the dorsal pronotum using the wax-rosin mixture with just enough slack to allow normal pronotum movement . The ends of the wires were de-insulated and positioned pointing up to prevent the animal from reaching them . After the surgery , animals were allowed a day to recover , and survived for up to 7 months during which time the animals behaved normally . To connect the electrode wires to the amplifier during an experiment , the animals were held in place with transparent surgical tape ( Dukal Corp , Ronkonkoma , NY ) . The free ends of the implanted electrodes were each attached to polyurethane-coated hook-up wire with a pair of gold-plated miniature connectors ( 0508 and 3061 , Mill-Max , Oyster Bay , NY; wire diameter: 160 µm or 34 AWG , Belden , St . Louis , MO ) . The hook-up wires were braided together and loosely suspended directly above the animal to allow unrestrained movement . Neither the implantation surgery nor the connection of implanted wires to the amplifier caused a significant reduction in escape behavior ( Figure 3—figure supplement 1 ) . Visual stimuli presented during jump experiments were generated with custom software on a personal computer ( PC ) running the real-time operating system QNX 4 ( QNX Software Systems ) , as previously described ( Gabbiani et al . , 1999 ) . Identical visual stimuli for electrophysiological experiments were generated using Matlab and the PsychToolbox ( PTB-3 ) on a PC running Windows XP . In both cases , a conventional cathode ray tube ( CRT ) monitor refreshed at 200 frames per second was used for stimulus display ( LG Electronics , Seoul , Korea ) . Both monitors were calibrated to ensure linear , 6-bit resolution control over luminance levels . Visual stimuli were presented in blocks with each stimulus shown once per block and the order within the block randomized by the stimulus software for all experiments . For wide field stimuli presented to restrained animals , a 90–120 s delay was used between stimuli and grasshoppers were repeatedly brushed and exposed to light flashes and high frequency sounds to decrease habituation . Some animals still exhibited pronounced visual habituation ( >50% reduction in peak firing rate from that animal's average response to the stimulus ) , and these data were excluded from analysis . In escape behavior experiments , a delay of at least 5 min ( and usually ~15 min ) was used between stimuli to prevent habituation . The drug effects were long lasting , so in all cases the control data was collected before the drug condition . Stimuli were randomly interleaved , no fatigue was evident within experimental conditions , and the drug effects reported are stimulus specific , so habituation or fatigue cannot explain the coherence-dependent results described . It cannot be ruled out that the exact change in firing is unaffected by habituation or fatigue , however . Looming stimuli consisted of dark squares simulating the approach of a solid object on a collision course with the animal ( Hatsopoulos et al . , 1995 ) . Briefly , the instantaneous angular size , 2θ ( t ) , subtended at one eye by a square of radius , l , approaching the animal at constant speed , v , is fully characterized by the ratio , l/\|v\| , since θ ( t ) = tan−1 [l/ ( v t ) ] . By convention , v < 0 for approaching stimuli and t < 0 before collision . Stimuli simulated approach with l/\|v\| values of 50 or 80 ms from an initial subtended angle of 1 . 2° until filling the vertical axis of the screen ( 300 mm ) , lasting approximately 4 and 7 s for l/\|v\| = 50 and 80 ms , respectively . The maximum 2θ values reached by the stimuli were either 136° or 80° for the freely behaving or restrained preparations , respectively , due to the differing distances of the eye to the screen . ‘Coarse’ looming stimuli were generated as in our earlier work ( Jones and Gabbiani , 2010 ) . Briefly , the stimulation monitor was first pixelated with a spatial resolution approximating that of the locust eye ( 2–3° x 2–3° ) , referred to as ‘coarse’ pixels . Each coarse pixel’s luminance followed the same time course as that elicited by the edge of the simulated approaching object sweeping over its area . To alter the spatial coherence of these stimuli , a random two-dimensional Gaussian jitter with zero mean was added to each coarse pixel screen location . The jittered positions were rounded to the nearest available coarse pixel location on the screen to prevent any coarse pixels from overlapping . To control the amount of shifting and thus the resulting spatial coherence of the randomized stimulus , the standard deviation , σ , of the Gaussian was altered between 0° and a maximal angular value σmax determined by the procedure described in the next paragraph . For a given Gaussian jitter σ , we determined the corresponding percent spatial coherence by averaging over 30 pseudo-random draws the minimal total angular distance that jittered coarse pixels had to be moved in each movie image to reconstitute the unaltered coarse looming stimulus . This distance was then normalized by the angular distance computed in the same way between a coarse loom and one with uniformly and independently drawn random spatial positions . Subtracting this normalized distance from one yields coherence values ranging from 100% when σ = 0% to 0% when σ reaches a value σmax for which the jittered stimulus is indistinguishable from a totally random one . The value of σmax was different for freely behaving ( 80° ) and restrained preparations ( 40° ) , due to the different distance between the screen and animal and thus the different angular expanse of the stimulus ( see above ) . For localized light flashes , a 1° x 1° luminance increase was presented briefly ( ~1 s ) on a black background in the dark ( Figure 2—figure supplement 1A–C ) . A window of 200 ms following the flash onset was used to quantify LGMD activity . The behavioral experiments were conducted as previously described ( Fotowat and Gabbiani , 2007 ) . They were recorded with a high-speed digital video camera ( GZL-CL-22C5M; Point Grey , Richmond , BC , Canada ) , equipped with a variable zoom lens ( M6Z 1212–3S; Computar , Cary , NC ) . Image frames were recorded at 200 frames per second with the acquisition of each frame synchronized to the vertical refresh of the visual stimulation display ( Xtium-CL PX4; Teledyne Dalsa , Waterloo Canada ) . Videos were made from the images and saved in lossless motion JPEG format using custom Matlab code . Measurements of the stimulus coherence's effect on escape behavior ( Figure 1F ) include a total of 202 trials from 66 animals with 1–9 trials per animal . Animals which did not jump in response to any stimuli were excluded from analysis , as done previously ( Fotowat and Gabbiani , 2007 ) . Electrophysiological experiments were performed as described previously ( Peron et al . , 2009; Jones and Gabbiani , 2010 ) . Briefly , sharp-electrode LGMD intracellular recordings were carried out in both voltage-clamp and current-clamp modes using thin walled borosilicate glass pipettes ( outer/inner diameter: 1 . 2/0 . 9 mm; WPI , Sarasota , FL ) . After amplification , intracellular signals were low-pass filtered ( cutoff frequency: 10 kHz for Vm , and 5 kHz for Im ) and digitized at a sampling rate of at least 20 kHz . We used a single electrode clamp amplifier capable of operating in discontinuous mode at high switching frequencies ( typically ~25 kHz; SEC-10 , NPI , Tamm , Germany ) . Responses to visual stimulation were measured in bridge mode , current injections were applied in discontinuous current clamp mode ( DCC ) , and voltage-clamp recordings in discontinuous single-electrode voltage-clamp mode ( dSEVC ) . All dSEVC electrodes had resistances < 15 MΩ . Electrode resistance and capacitance were fully compensated in the bath immediately prior to tissue penetration and capacitance compensation was readjusted after entering the neuron . If capacitance could not be fully compensated the recording was not used . In addition to previously described methods , a fluorescent dye ( Alexa Fluor 594 hydrazide salt; Invitrogen , Thermo Fisher Scientific , Carlsbad , CA ) was injected intracellularly and the cell was imaged with a CCD camera mounted to a stereomicroscope ( GuppyPro F125B; Allied Vision Technologies , Exton , PA ) . This allowed subsequent visually guided positioning of the recording electrode . Within the LGMD , back-propagating action potentials ( bAPs; measured from RMP to peak ) decay as they spread into the dendrites . Data from dual recordings ( unpublished ) has revealed that the decay in bAPs is a better indicator of electrotonic distance than the path length , and it is also easier to reliably determine . So , we used bAP amplitude as the measure of electrotonic distance from the SIZ ( Figure 2E ) . During voltage clamp recordings , the membrane potential and current were measured simultaneously to ensure the desired membrane potential was maintained at the electrode location . The LGMD neuron is not electrotonically compact ( Peron et al . , 2007 ) and therefore the issue arises of how well its dendritic membrane potential is controlled through voltage-clamping at a single location ( 'space clamp' ) . The quality of the space clamp cannot be measured with a single electrode recording . In pyramidal neurons , the steady-state dendritic membrane potential is largely uncontrolled when voltage clamping originates at the soma ( Williams and Mitchell , 2008 ) . In contrast , in Purkinje cells , which have a dendritic structure more closely resembling that of the LGMD , the steady-state dendritic membrane potential is well controlled from the soma ( Roth and Häusser , 2001 ) . Simulations in NEURON ( details below ) were used to estimate the quality of the space clamp in the LGMD . For electrodes placed at the base of field A the average steady-state change in membrane potential within field A was 95% of the desired change ( i . e . starting at rest , -65 mV , a -30 mV step to -95 mV , yielded an average membrane potential across field A of -65+0 . 95∙ ( -30 ) or -93 . 5 mV ) . For electrodes placed further away from the base of field A , the quality of the space clamp decreased . Across the dendritic region used for voltage clamp recordings , the estimated quality of the space clamp ranged from 83-95% ( average voltage command attenuation of 5-17% ) . We also carried out simulations as described in Roth and Häusser ( 2001 ) to assess the impact of these findings on the characterization of gH activation and kinetics . The effects were found to be mild , suggesting that the activation curve in Figure 2F might be slightly less steep and the time constants in Figure 2G slightly higher than if they were measured with perfectly space-clamped dendrites . For characterizing gH , 1–2 s hyperpolarizing current or voltage steps were injected in DCC or dSEVC mode , respectively , with 5 s between steps . Voltage clamp was needed to calculate the activation and time constant at a given membrane potential ( Figure 2G , H ) , while current clamp was used for all other experiments because it allowed for easier to hold , longer lasting recordings . Different step amplitudes were randomly interleaved and at least six trials per step amplitude per animal were acquired . For each recording , we used at least four step amplitudes , with values selected to cover the activation range of gH . Example recordings are shown in Figure 2—figure supplement 1 . In most experiments , no holding current was applied between steps ( held at the resting membrane potential ) , while in some experiments a positive holding current was applied ( held near −50 mV ) before hyperpolarizing steps and in other experiments a negative holding current was used ( held near −115 mV ) with depolarizing steps . Estimated activation curves ( see below ) were not different for recordings with different holding potentials , and the data were combined for analysis . Extracellular recordings were taken between the two hook electrodes on the nerve cord , differentially amplified and bandpass filtered from 100 to 5000 Hz ( A-M Systems , model 1700 , Carlsborg , WA ) . The amplitude of DCMD spikes was consistently the largest , allowing their identification with a simple threshold . DCMD spikes uniquely identify the LGMD neuron as they are in one-to-one correspondence with those of the LGMD ( O'Shea and Williams , 1974 ) . Experiments with hyperpolarizing current during visual stimulation ( Figure 5—figure supplement 1 ) were conducted by first staining the LGMD and then inserting an electrode near the base of field A . Within our LGMD model , the resting membrane current magnitude generated by HCN channels was ~2 . 5 nA . Experimentally , injecting –2 . 5 nA produced local hyperpolarization to ≦ −75 mV , which is as hyperpolarized as any LGMD neuron became after HCN blockade ( see Figure 4A ) . Visual stimuli of different coherences were presented either with zero current or –2 . 5 nA current injected from 20 s before stimulus onset through the end of the stimulus . Sets of single trials of each stimulus ( randomized ) were presented while alternating between 0 and –2 . 5 nA currents and continued until at least three trials of each stimulus were presented for both conditions . Drugs were prepared in aqueous solution and mixed with physiological saline containing fast green ( 0 . 5% ) to visually monitor the affected region . They were puffed using a pneumatic picopump ( WPI , PV830 , Sarasota , FL ) . For restrained experiments , injection pipettes had tip diameters of ~2 µm and were visually positioned with a micromanipulator against the posterior edge of the lobula , close enough that the ejected solution penetrated the optic lobe . Drugs were gradually applied while monitoring responses of the LGMD to visual inputs , and care was taken to prevent spread into presynaptic neuropils . Additionally , saline in the bath was exchanged immediately after puffing to prevent diffusion to other brain areas . We used drug concentrations of 10 mM for both ZD7288 and 4AP in the extracellular puff pipette . These concentrations were adjusted in pilot experiments to account for the low mobility of the drugs through the tissue in vivo , taking into account their approximate final concentration , as explained below . Due to dilution of the drugs in the saline bath after puffing , the exact drug concentration at the level of the LGMD cannot be determined . However , our best estimate is ~200 µM for both ZD7288 and 4AP . This estimate comes from comparing the effect of the puffed drugs to those observed after bath application of the same drugs . For example , when bath applying ZD7288 , the same level of blockade as from local puffing was achieved by adding 100 µl of 20 mM drug to ~5 . 5 ml of bath saline for a final concentration of ~350 µM . This concentration is an upper bound on the concentration at the level of the LGMD , since it lies ~150 µm deep within the optic lobe . For local puffing , less than 1 µl of drug was used , which would generate a final bath concentration well below 1 µM after exchanging the saline in the bath , as explained above . For intracellular application , the drug concentrations in the pipette were 1–5 mM for ZD7288 and 5 mM for 4AP . The final concentration inside the LGMD cannot be determined but is likely considerably lower , due to the large volume of the cell and the submicron diameter of the pipette . In those experiments , the effects of the drugs were comparable to those observed with extracellular application . Although it cannot be known whether intracellularly applied ZD7288 or 4AP diffused out from within the LGMD , this seems highly unlikely to have affected our results . For example , the effects of intracellular ZD7288 application on the LGMD's membrane properties were consistent from a minute to an hour after application , giving no evidence of a slow diffusion across tissue that may have affected presynaptic sites . Further , the membrane effects on the LGMD were the same whether excitatory synaptic inputs were blocked with mecamylamine or not . To further rule out the possibility that the effects of ZD7288 observed during visual stimulation were caused by diffusion to presynaptic sites after intracellular application , we conducted visual stimulation experiments in which the gH conductance was blocked intracellularly with Cs+ ( Figure 2—figure supplement 2 ) . Similar effects were seen on visual responses compared to ZD7288 application , although Cs+ was not as specific a blocker since there was also evidence of partial block of K+ conductances . For these experiments , a concentration of 150 mM CsCl was used in the recording pipette . We also attempted to block intracellularly the inactivating K+ conductance by using 4-aminopyridine methiodide ( 4APMI ) which is membrane impermeant ( Stephens et al . , 1994 ) . 4APMI reacted strongly with the silver wire in the electrode forming AgI crystals , so a platinum wire was used for the experiments . Unfortunately , 4APMI which is larger than 4AP failed to block the inactivating K+ conductance even at recording pipette concentrations as high as 50 mM . Nonetheless , presynaptic effects are unlikely as we never observed increases in spontaneous EPSPs within the LGMD following 4AP application and the presynaptic neurons have no information about the overall spatial pattern of the stimulus . In all there were no indications of any nonspecific drug effects on presynaptic neurons that might have influenced visual responses . To observe the effects of ZD7288 in freely moving animals , stereotaxic injections were made through a hole in the dorsal rim region of the right eye . The animal was restrained and the head was placed in a small clamp attached to a 3-axis micromanipulator ( Narishige , Tokyo , Japan ) . Head tilt was positioned manually by fixing the animal at the pronotum . After the head was precisely positioned , a ~ 0 . 5 mm hole was made through the dorsal end of the eye with a steel probe . A drop of saline solution was placed covering the hole to prevent drying or coagulation of the hemolymph . A glass pipette with a tip diameter of 1–2 µm and a taper length >2 mm from shoulder to tip was positioned with a Leica ( Wetzlar , Germany ) manual micromanipulator and lowered just above the eye . The ZD7288 solution ( 2 mM in saline with 1% fast green ) was puffed into the saline drop covering the dorsal rim to determine the appropriate air pressure ejection level . The saline droplet was immediately removed and replaced to prevent spread of ZD7288 to photoreceptors . Next , the pipette was lowered through the eye along the dorsal rim of the optic lobe to the lobula ( ~1 . 5 mm ) while enough positive pressure was maintained to prevent clogging . In control experiments , LGMD activity was measured before and after penetration of the pipette in the lobula to ensure that visual inputs were not damaged by the procedure ( Figure 3—figure supplement 1C ) . Ejection volume was estimated from monitoring changes in the meniscus position of the saline within the visible region of the pipette . After pressure ejection of ZD7288 , the pipette was removed and checked for clogs or breaks . The hole in the eye was sealed with a small amount of Vetbond ( 3M , St . Paul , MN ) , carefully ensuring that no glue spread onto the rest of the eye . Following the conclusion of the experiment , the animal was euthanized and the head was dissected ( ~2 hr post injection ) . Fast green staining was used to confirm that the solution was injected into the lobula ( Figure 3—figure supplement 1D ) . In initial experiments , bath application of ZD7288 was found to reduce visual responses as did application of ZD7288 directly to photoreceptors . When puffing ZD7288 within the lobula , however , even if the solution occasionally spread to the medulla or lamina , visual responses remained similar to those observed after intracellular application . This suggests that there are likely HCN channels within the photoreceptor layer , as is the case in mammals ( Barrow and Wu , 2009 ) , but that any HCN channels within the medulla and lamina ( Hu et al . , 2015 ) do not influence LGMD inputs under our experimental conditions . Because ZD7288 was applied extracellularly , it may have affected other descending neurons whose processes are located in the immediate vicinity of the LGMD dendrites . This is , however , unlikely to have affected escape behaviors , since decrease in escape was tightly correlated with a reduction of LGMD firing rate determined in independent experiments ( Figure 2N ) . In addition , earlier selective ablation experiments have shown that under our experimental conditions nearly all escape behaviors depend solely on LGMD firing ( Fotowat et al . , 2011 ) . Data analysis was carried out with custom MATLAB code ( MathWorks , Natick , MA ) . Linear fits were based on Pearson's linear correlation coefficient , denoted by 'r' in figure legends , with corresponding p values testing significant differences from zero . Non-linear fits , including the activation curve and time constant in Figure 2 and all exponential fits described below were made with the Matlab function ‘lsqcurvefit’ , which minimizes the least square error between the data and fitting function . Goodness of fit was denoted by R2 , calculated as one minus the sum squared error of the fit divided by the sum square deviation from the mean of the data . For behavioral experiments and the comparison of membrane depolarization with stimulus angular distance ( Figure 6 and Figure 6—figure supplement 1 ) , individual trials were used as independent sample points for statistical tests . In all other cases , individual trials were averaged and these trial averages were used for statistical tests . The sag amplitude was measured as the difference in membrane potential between the peak hyperpolarization during a current step and the steady-state value at the end of the step . The sag time constant was calculated from fitting a single exponential to the membrane potential for the period starting 15 ms after peak hyperpolarization to the end of the current step ( Figure 2—figure supplement 1C ) . The hyperpolarizing step currents were also used for calculating membrane time constants . The membrane time constant was calculated by fitting a single exponential to the membrane potential for the period from 0 . 5 to 13 ms after the start of hyperpolarizing current injection . The fitted activation curve of the HCN conductance was based on a Boltzmann equation reflected along the vertical axis to produce decreasing gH with increasing v:gHv=gmax1+ev-v1/2s . The steady-state conductance , gH , is a function of the membrane potential , v , depending on three parameters: the maximum conductance , gmax , the half-activation potential , v1/2 , and the steepness , s . The parameters were fitted from voltage-clamp data based on the equationgH ( v2 ) −gH ( v1 ) =ΔIH/ ( v2−EH ) , where v1 and v2 are the starting and ending clamp potentials and EH is the reversal potential of the HCN conductance , −35 mV , used for all animals . ∆IH is the experimentally measured change in membrane current produced by the voltage step after transients have settled . ∆IH was measured by fitting a single exponential to the current time-course 15 ms after the step onset and up to its end ( Figure 2—figure supplement 1D ) . This period captured the slow change in clamp current due to gH and offered clear experimental advantages over other estimations methods . As all experiments were done in vivo , it was not feasible to reliably block other putative voltage-gated channels . Hence , the most reliable measurements of ∆IH were obtained at hyperpolarized membrane potentials where other active conductances can be safely discounted . Voltage clamping the LGMD to depolarized potentials where all HCN channels will be closed ( > −40 mV ) was not technically feasible , and the use of tail currents yielded less reliable measurements due to contamination by other active conductances . The time constant of the HCN conductance ( τH ) was fit using a function symmetric with respect to its maximum , τmax , τHv=τmax0 . 5 ( ev-v1/2s+ev1/2-vs ) +τmin . Here , v1/2 is the membrane potential with the slowest activation , s is the steepness , and τmin the minimum activation time . Fitted points were obtained from the single exponential fits to IH for both hyperpolarizing ( channel opening ) and depolarizing ( channel closing ) voltage steps . Comparisons of sag amplitudes were obtained with current steps yielding a peak hyperpolarization of ~105 mV ( Figure 2D–F ) . For Figure 2D , E , all values to steps within the range of −95 to −115 mV were pooled . For Figure 2F interpolation of values at nearby potentials was used to estimate sag amplitude at −105 mV to have a single common value for all recordings . Statistical comparisons between sag measurements in different subcompartments of the LGMD ( Figure 2D , E ) were carried out using a Kruskall-Wallis analysis of variance ( KW ) corrected for multiple comparisons with Tukey's Honestly Significant Difference Procedure ( KW-MC ) . To determine the correct statistical test for comparison , we used a Lillifors test of normality ( alpha = 0 . 20 ) and comparison of equality of variance . Much of the data was non-normally distributed and variances increased after drug application so most comparisons were made using the Wilcoxon rank sum test ( WRS ) which does not assume normality or equality of variance . For displaying non-normal data , average values were given as median and variance was displayed as median average deviation ( mad ) . Mean and standard deviation were used for normally distributed values , as indicated in figure legends . Before carrying out paired tests , we determined if the paired differences where normally distributed . The changes in slope reported in Figure 6D , E were the only non-normal paired difference , so for this we used the non-parametric signed-rank test . Percent activation at rest ( Figure 2I ) was calculated through bootstrapped activation curves from current clamp data . Unpaired t statistics were calculated from the bootstrapped mean and variance of activation at the resting membrane potential ( −65 mV ) ( Efron and Tibshirani , 1993 ) . Simulated excitatory postsynaptic potentials ( sEPSPs ) were generated by injecting a series of five current waveforms with a set delay between them . Each waveform , I ( t ) , had a time course resembling that of an excitatory synaptic current , It=A ( 1-e-t/τ1 ) e1-t/τ2with peak amplitude A , rising time constant τ1 = 0 . 3 ms , and falling time constant τ2 = 3 . 0 ms . Summation was calculated as the ratio ( p5-p1 ) /p1 , with p1 and p5 being the peak amplitude of the membrane potential relative to rest during the 1st and 5th sEPSP . In Figure 4F , we plotted the integrated membrane potential ( relative to rest ) divided by the integrated input current ( charge ) giving a value in units of mV ms/nA ms=M Ω that is readily comparable to input resistance . Spike counts elicited by looming stimuli were measured from the start of the stimulus until the time of expected collision , and peak firing rates were calculated by convolving the spike rasters with a 20 ms sd Gaussian as has been done in previous studies ( Hatsopoulos et al . , 1995; Gabbiani et al . , 1999; Peron et al . , 2009; Jones and Gabbiani , 2010; Fotowat and Gabbiani , 2007 ) . After statistics were conducted on unnormalized data , firing rates were normalized before averaging across animals to reduce the between animal variability in responses . This allowed for the clearest comparison of the role of stimulus coherence across conditions and with the biophysical model of the LGMD . Normalized firing rates ( Figure 2M and Figure 5C ) were calculated by dividing the response amplitude for each stimulus by that animal's maximal response amplitude under control conditions ( insets of Figures 2M and 5C ) . Dividing by the maximum response was chosen to show that 100% coherent stimuli generated the maximum response and to give an easy indication of the amount of change from a standard/fully coherent looming response . These individually normalized rates were then averaged across animals . The relative change in response due to a drug ( Figure 5D ) was calculated by dividing the difference in response between control and drug conditions by the drug condition response . This produced percentages covering similar ranges , and so allowed for the best comparison and graphical illustration of their relative effects . ‘Sustained firing’ was defined as the longest period in which the instantaneous firing frequency remained above a 20 spk/s threshold . For each trial , the number of spikes within this longest period was considered the ‘sustained response’ and all spikes outside of this period were counted as the ‘transient response’ ( Figure 7—figure supplement 1F , G ) . These 'sustained' and 'transient' measures were used instead of 'burst' and 'non-burst' statistics based on interspike intervals because the LGMD can generate sustained high frequency firing with similar interspike intervals within and outside of bursts . To compare changes in membrane potential and stimulus angular distance ( Figure 6; Figure 6—figure supplement 1 ) , we identified newly changing coarse pixels in a specified stimulus frame from those that had begun to darken from their background luminance value in earlier frames ( ‘earlier changing’ ) . We then computed the mean minimal distance of newly changing coarse pixels with respect to earlier changing ones . In parallel , changes in the membrane potential were averaged from 25 ms following the appearance of the newly changing coarse pixels until a new group of pixels began to darken . More precisely , we identified six time periods during the stimuli when the luminance of newly changing coarse pixels is decreasing for over 50 ms and they typically have mean angular distances larger than 1° from earlier changing ones . For these six different time periods during each trial , we calculated the linear correlation between these mean angular distances and membrane depolarizations , as explained above . Early in the stimulus presentation , there are fewer coarse pixels changing luminance and less resulting depolarization . To better illustrate the relationship between these variables , the angular distances and membrane potentials were normalized independently for each of the six time windows . The normalization consisted of subtracting the minimum control value and then dividing by the range for control data within each time window . The unnormalized data and example stimulus frames from all time periods are shown in Figure 6—figure supplement 1 . To compare jump probabilities between saline- and ZD7288-injected animals ( Figure 3A ) , we computed 95% bootstrap confidence intervals of the population mean in each condition with the help of the built-in Matlab function ‘bootci’ ( using the bias corrected and accelerated method ) . If there was no overlap of the 95% confidence intervals , the groups were considered significantly different . The reported p-values for these comparisons were the ‘achieved significance level’ ( ASL ) statistic for two-sample testing of equality of means with unequal variance ( Algorithm 16 . 2 in Efron and Tibshirani , 1993 ) . Coherence selectivity was calculated as the slope of the relationship between stimulus coherence and spike count and is reported in units of spikes per percent coherence . For control experiments , the median correlation coefficient of this relationship between stimulus coherence and spike count was 0 . 97 , making the regression slope a reliable indicator of the selectivity . For box plots , the center line shows the median , the upper and lower box limits mark the 25th and 75th percentile of the distribution , and the ‘whiskers’ above and below each box extend 1 . 5 times the interquartile range up to the minimum and maximum values . Points beyond the whiskers mark outliers . Notches , when present , have a width of 1 . 57 times the interquartile range divided by the square root of the number of data points . To better understand the mechanisms of the LGMD's remarkable coherence selectivity , we developed a detailed biophysical model using the NEURON simulation environment . We employed the parallel version of NEURON and a Rice University supercomputing cluster for extensive parameter sweeps and simulations . Three-dimensional reconstructions of the LGMD's dendritic morphology were obtained from two-photon scans using the software suite Vaa3D ( vaa3d . org ) . The resulting model contained 2518 compartments , 1266 of which belonged to dendritic field A . To reproduce the active properties of the LGMD several voltage-gated channel types were included . Some of them had been used in previous simulations ( Peron et al . , 2007; Jones and Gabbiani , 2012 ) , including the fast Na+ and delayed rectifier K+ ( KDR ) channels generating action potentials . KDR channels were distributed throughout the cell , but dendritic branches contained no fast Na+ channels as supralinear summation is never seen in LGMD dendrites . HCN channels had kinetics matching experimental data ( Figure 2 ) and were placed in dendritic field A with density increasing towards the distal dendritic endings . Inactivating K+ channels ( KD-like ) were also distributed throughout field A with density increasing toward distal endings . A slow non-inactivating K+ channel ( M ) was distributed throughout the axon , the spike initiation zone ( SIZ ) , and the main neurite connecting the dendritic subfields to the SIZ . Its peak density was at the SIZ . Additionally , low-threshold Ca2+ ( CaT ) and Ca2+-dependent K+ ( KCa ) channels were placed at the SIZ and on half of the neurite connecting the SIZ and the dendritic subfields , matching results from our earlier work ( Peron and Gabbiani , 2009 ) . Effective modeling often relies on keeping things as simple as possible , so we initially tested a previous LGMD model ( Jones and Gabbiani , 2012 ) with additional HCN channels matching experimental kinetics ( Figure 2 ) added to the dendrites . When this failed to reproduce any coherence selectivity , KD-like channels were added . Then we added a complex dendritic morphology , complex presynaptic transforms , and additional active conductances to the model . While a wide range of parameters of this more complex model reproduced responses to current injection data , only a narrow parameter regime was found that reproduced the roles of gH and KD-like in the spatial coherence preference . The resulting model and mechanistic explanation ( Figure 7 and 8 ) , while quite complex , is still the simplest model that reproduced the wide range of LGMD responses tested . To test the impact of the elaborate dendritic structure of field A on coherence selectivity , we simplified its electrotonic structure in successive steps using Rall’s law of electrotonically equivalent cylinders ( Rall , 1959 ) . This was done by iterating through dendritic branches by selecting dendritic segments according to their electrotonic distance from the base of field A in steps of 0 . 04 times the dendritic space constant ( λ ) . An equivalent compartment was created from the group of dendritic segments found at each successive electrotonic distance from the base of field A . For the least reduction ( the 'six branch' case shown in Figure 7—figure supplement 1 ) , the grouping was limited to dendritic segments that shared a common connection with the base of field A . For further reduction , the group of selected compartments was expanded to segments with adjacent connections to the base of field A . The equivalent compartment size was set by Rall’s equivalent 3/2 diameter law . Each channel’s density was set to the surface area weighted mean of its density in the selected dendritic segments , and all synaptic inputs to these segments were transferred . Tests of channel kinetics were run using 10-fold changes to the time constant of HCN activation and KD-like inactivation . For the 'fast' kinetics , the maximal HCN channel activation was set to τmax = 135 ms and the KD-like inactivation to τmax = 105 ms . For the 'slow' kinetics , the maximal HCN channel activation was τmax = 13 . 5 s and the KD-like inactivation was τmax = 10 . 5 s . The values used for all other simulations in the manuscript were 1 . 35 s and 1 . 05 s , respectively . Evaluation of how well this model informs about the actual neural processes requires some review of the experimental data to which it was constrained . The strength and timing of synaptic inputs was generated based on single facet stimulation data ( Jones and Gabbiani , 2010 ) . Excitatory synaptic input locations were based on the retinotopy and synaptic overlap determined by functional imaging ( Zhu and Gabbiani , 2016; Peron et al . , 2009 ) . The time course of synaptic inputs was based on experiments stimulating individual facets ( Jones and Gabbiani , 2010 ) and the pattern of depolarization measured during the current experiments . The presence of standard Hodgkin-Huxley Na+ and K+ currents was assumed , HCN and KD-like channels were based on the current work , the CaT and KCa channels were based on Peron and Gabbiani , 2009 , the M current was based on our own currently unpublished findings . For each of these channels , conductance and kinetic parameters were adjusted to match experimental data with firing frequency vs . injected current curves and spike waveform used to tune fast Na+ and K+ channels , changes in input resistance and resting membrane potential after pharmacological blockade used to adjust HCN , KD-like , and M parameters , while CaT and KCa were adjusted to match intrinsic burst ( currently unpublished ) and spike frequency adaptation data ( Peron and Gabbiani , 2009 ) . Channel distributions were similarly grounded in experimental data , when available , and were manually fit to find working parameters . To estimate space clamp quality ( see Materials and methods: Electrophysiology ) , we used the Impedance object class in NEURON and measured the percent voltage attenuation from an electrode location to each compartment within field A of the model . The average attenuation was calculated by weighting each section by its surface area to calculate the average change of membrane potential within field A . The full model and simulation code are available in the public repository ModelDB , accession number 195666 . Experimental data and code used to generate figures are available as Source Data and Code .	Whether you are a flying insect or a driver on a freeway , your survival will depend on avoiding collisions . Many species have nerve cells or neurons within their visual system that respond to objects headed towards them on a collision course . In locusts , for example , a neuron called the lobula giant movement detector ( LGMD ) triggers an escape response upon detecting an impending collision . But how does it do this ? The answer may also help us understand how neurons process other complex inputs . This is because like most neurons , the LGMD contains thousands of branches called dendrites . The job of the dendrites is to receive input from other neurons and collect that input for processing . In the LGMD’s case , each piece of input reveals what is happening at a single point in the locust’s field of vision . The cell combines all the inputs and uses the end result to decide whether to trigger an escape response . An object on a collision course will generate a specific sequence of images in the locust’s eye . These images will activate LGMD dendrites in a specific pattern . To test whether LGMD neurons use this pattern to detect approaching objects , Dewell and Gabbiani showed locusts two sets of movies . One set featured an object looming towards the insect on a collision course . But in the other set , the same movies had been scrambled . These movies thus activated LGMD dendrites in a different pattern than the movies showing looming objects . Both the LGMD neurons , and the locusts themselves , responded more to the non-scrambled movies . This suggests that they do use the pattern of activity in dendrites to detect impending collisions . Blocking two types of ion channels in the membrane of the dendrites prevented the neurons from distinguishing between scrambled and non-scrambled movies . Both of these ion channels are also present in the dendrites in our own brains . This suggests that many neurons can detect the spatial pattern in which their dendrites become active . By revealing how neurons process complex visual inputs , the results of Dewell and Gabbiani could help improve algorithms for man-made collision avoidance systems . These could be used in self-driving cars , or in technology to help visually impaired people navigate independently .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Biophysics of object segmentation in a collision-detecting neuron
Small heat shock proteins ( sHSPs ) are essential ‘holdase’ chaperones that form large assemblies and respond dynamically to pH and temperature stresses to protect client proteins from aggregation . While the alpha-crystallin domain ( ACD ) dimer of sHSPs is the universal building block , how the ACD transmits structural changes in response to stress to promote holdase activity is unknown . We found that the dimer interface of HSPB5 is destabilized over physiological pHs and a conserved histidine ( His-104 ) controls interface stability and oligomer structure in response to acidosis . Destabilization by pH or His-104 mutation shifts the ACD from dimer to monomer but also results in a large expansion of HSPB5 oligomer states . Remarkably , His-104 mutant-destabilized oligomers are efficient holdases that reorganize into structurally distinct client–bound complexes . Our data support a model for sHSP function wherein cell stress triggers small perturbations that alter the ACD building blocks to unleash a cryptic mode of chaperone action . Cells have numerous strategies to cope with the consequences of stress conditions that lead to protein misfolding and aggregation . Ineffective resolution of protein misfolding and unmitigated protein aggregation can lead to formation of plaques , fibrils , and other aggregated species encountered in neurodegenerative diseases . Generally , ATP-dependent chaperones such as Hsp70 and Hsp90 assist with protein refolding , while ATP-independent chaperones known as small heat shock proteins ( sHSPs ) act as first responders by maintaining proteins in soluble forms to inhibit misfolding and to delay aggregation . Under transcriptional regulation of the heat shock factor , Hsf1 , levels of sHSPs can rise to 1% of total cellular protein when conditions deviate from the norm , e . g . ischemia , hypoxia , oncogene activation , and chemotherapy ( Kampinga and Garrido , 2012 ) . sHSPs are found throughout prokaryotes and eukaryotes . There are ten sHSPs encoded in the human genome ( HSPB1 , HSPB2 , etc ) , ranging in size from 15 to 25 kDa . Most , including the ubiquitously expressed human sHSP HSPB5 , form large assemblies that exist as dynamic distributions of polydisperse oligomers whose properties are both temperature and pH dependent ( Jehle et al . , 2011 ) . Which form or forms of an sHSP are active and how they function are central unanswered questions . In addition , how sHSPs interact with partly unfolded or aggregate-prone proteins ( ‘clients’ ) to prevent formation of aggregates remains enigmatic . Some acute cellular stresses are associated with acidosis . For example , measurements in mouse brain following ischemic stroke showed a pH of 6 . 4 in ischemic tissue compared to a pH of 7 . 0 in normal tissue ( McVicar et al . , 2014 ) . A decrease in cellular pH may give rise to destabilization of some proteins and/or a decrease in solubility for proteins with pI values between pH 6 and 7 . Furthermore , the pH in normal eye lens fiber cells , which contain extremely high concentrations of the sHSPs HSPB4 and HSPB5 is ∼6 . 5 ( Bassnett and Duncan , 1985; Mathias et al . , 1991 ) . Such observations raise the question as to the nature and determinants of pH-dependent properties of sHSPs in general and HSPB5 in particular . Here , we report how pH affects structural , biochemical , and functional properties of HSPB5 ( also known as αB-crystallin ) . Like all sHSPs , HSPB5 has a conserved central α-crystallin domain ( ACD ) flanked by variable N- and C-terminal regions . We carried out investigations both on the HSPB5-ACD , which forms the dimeric building block found in all sHSPs , and on full-length oligomeric HSPB5 . We find that ACD dimer stability decreases over a narrow physiologically relevant pH range and we have identified a conserved histidine residue that is largely responsible for the observed destabilization . Surprisingly , the histidine is not on the dimer interface but its mutation to glutamine or lysine partly or fully recapitulates low pH behavior both in the context of the ACD dimer and in oligomers . The mutant proteins allowed us to probe the consequences of dimer interface destabilization without additional complicating factors that arise when comparing experiments performed at differing pH . Mutations that destabilize the dimer interface produce extremely large oligomers that can rearrange to form long-lived complexes with a model client protein . Our studies unmask a cryptic mode of HSPB5/client interaction not previously detected using the wild-type protein under non-stress conditions . The results highlight mechanistic differences in the ways in which holdases work and suggest that HSPB5 has a repertoire of ways in which it can carry out its function . As ACD dimers are the fundamental , stably folded building blocks of sHSP oligomers , we first sought to define their response to changes in pH over a physiologically relevant range . Although multiple structures have been solved for HSPB5-ACD , there are differences among them as well as differences in the conditions under which the structures were obtained ( i . e . , pH 4 . 6 , 6 . 0 , 7 . 5 , 8 . 5 , 9 . 0 ) ( Clark et al . , 2011 ) . We felt this situation necessitated structure determination under solution conditions relevant to our studies . The solution structure of HSPB5-ACD at pH 7 . 5 , 22°C ( Figure 1A , B ) was calculated from NMR data that included backbone and side chain chemical shifts , NOEs , and RDCs ( Table 1 ) . As in all previously determined HSPB5-ACD dimer structures ( PDB 2WJ7 , 3L1G , 2KLR , 2Y22 , 4M5S , and 4M5T ) , each subunit adopts a six-stranded β-sandwich structure and two ACDs form a dimer through anti-parallel alignment of their long β6+7 strands ( Figure 1A , B , C ) . A notable difference among solved sHSP ACD structures is that dimers appear in different alignments of the two β6+7 strands ( Clark et al . , 2011; Hochberg et al . , 2014 ) . In the solution structure reported here , the dimer interface places Glu117 across from Glu117′ ( Figure 1D ) . This register is observed in all but two of the available HSPB5-ACD structures and is also the predominant alignment observed in other metazoan sHSP ACD structures . 10 . 7554/eLife . 07304 . 003Figure 1 . Solution structure ensemble of HSPB5-ACD at pH 7 . 5 is an anti-parallel dimer . ( A ) Backbone traces of ten ACD structures determined by RosettaOligomer are aligned over all residues ( RMSD of 1 . 4 Å ) . Subunits of the dimer are shown in blue and gray . ( B ) Cartoon representation of one member of the HSPB5-ACD ensemble is shown . The six strands of the β-sandwich structure are labeled using previously defined nomenclature for ACD structures . The dimer is formed by antiparallel arrangement of the β6+7 strands at the interface . Loop 5/6 is highlighted in the blue subunit . ( C ) The sequence of the ACD construct for which the solution structure was solved is shown with elements of secondary structure highlighted . ( D ) The alignment of the β6+7 strands at the dimer interface , shown schematically , places Glu-117 across from Glu-117′ . This alignment is called ‘APII’ in previous reports ( Clark et al . , 2011 ) . ( E ) Overlay of protomers from all available HSPB5-ACD structures . All members of the solution NMR structure ( this paper; PDB 2N0K ) and the solid-state NMR structure ( 2KLR; Jehle et al . , 2010 ) are shown in dark gray and light gray , respectively . Five crystal structures are shown: ( 1 ) 2WJ7 ( yellow ) , ( 2 ) 3L1G ( orange ) , ( 3 ) 4M5S ( green ) , ( 4 ) 4M5T ( red ) , and ( 5 ) 2Y22 ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 00310 . 7554/eLife . 07304 . 004Figure 1—source data 1 . Comparison of the NMR solution structure and published structures . Backbone root mean square deviations ( rmsd ) between the solution structure and published structures and the correlation between experimentally measured residual dipolar coupling ( RDC ) values and those calculated for each published structure ( PALES Correlation Coefficient ) are given . Values in parentheses are calculated without including loop regions between the β-strands . Protomer–protomer and dimer–dimer rmsd between 2KLR and solution structure ensembles are the averages between all pairwise combinations of all structures in each ensemble . Rmsd values between the solution structure ensemble and the crystal structures are the average of all pairwise rmsd between the ten structures in the ensemble and the crystal structure . 2KLR: solid-state NMR structure; 2WJ7:crystal structure at pH 9 . 0; 3L1G:crystal structure at pH 4 . 6; 4M5S: crystal structure at pH 6 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 00410 . 7554/eLife . 07304 . 005Figure 1—source data 2 . Multiple sequence alignment of the ten human sHSPs . ClustalOmega was used to align the sequences . Structural features of HSPB5-ACD are highlighted as follows: ACD is denoted by the blue line; His-104 is identified by the blue arrow; Loop 5/6 is shown in red box; dimer interface is shown in green box . All ACD histidine residues are shown in bold font . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 00510 . 7554/eLife . 07304 . 007Table 1 . NMR data and refinement statistics for HSPB5-ACD structuresDOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 007NMR distance and dihedral constraintsHSPB5-ACDDistance constraints Total NOE838 Intraresidue310 Inter-residue Sequential ( \|i − j\| = 1 ) 255 Medium-range ( \|i − j\| < 4 ) 85 Long-range ( \|i − j\| > 5 ) 188 Inter-molecular36 Total dihedral angle restraints φ ( TALOS ) 72 ψ ( TALOS ) 72 Residual Dipolar Couplings ( RDCs ) 1H-15N RDCs81Structure statistics Violations ( mean ± s . d . ) Distance constraints ( Å ) 0 . 48 ± 0 . 45 Dihedral angle constraints ( ° ) 14 . 4 ± 14 . 7 Average pairwise r . m . s deviation ( Å ) * Heavy2 . 46 ± 0 . 97 Backbone1 . 48 ± 0 . 6Average pairwise r . m . s . d . was calculated among ten refined structures . At the level of the protomers , the HSPB5-ACD structures are remarkably similar despite having been solved under significantly different conditions and by different techniques ( Figure 1E and Figure 1—source data 1 ) . Two NMR ensembles ( solution-state reported here and solid-state PDB 2KLR ) and five crystal structures all overlay well with the exception of the loop that connects β5 and β6+7 ( called Loop 5/6; residues His-104–Gly-112 ) . Loop 5/6 curves upwards in our solution ensemble ( pH 7 . 5 ) and in three crystal structures ( 2WJ7 [pH 9 . 0] , 3L1G [pH 4 . 6] , and 2Y22 [pH 8 . 6] ) . In the ‘loop up’ conformation , Loop 5/6 residues make contacts across the dimer interface to the other subunit . In contrast , Loop 5/6 curves slightly downwards in the solid-state NMR ensemble ( 2KLR; pH 7 . 5 ) and in two crystal structures ( 4M5S [pH 6 . 0]; 4M5T [pH 6 . 5] ) . Thus , among structures solved between pH 6 . 5 and 7 . 5 , both loop conformations have been observed , suggesting that the loop is dynamic in the physiological pH range . To investigate dynamic processes in the HSPB5-ACD , we performed 15N relaxation measurements . A majority of backbone amide nitrogens have 15N T2 values between 30 and 40 ms , but resonances in Loop 5/6 and the dimer interface have T2 < 30 ms , indicating they undergo a change in their environment in the millisecond timescale ( data not shown ) . To better probe this time regime , we performed NMR 15N relaxation-compensated CPMG relaxation dispersion measurements under the solution structure conditions ( 700 μM ACD at pH 7 . 5 , 22°C ) . Examples of relaxation curves obtained from spectra collected at two magnetic field strengths ( 600 and 800 MHz ) are shown in Figure 2A . Twenty resonances gave curves that could be analyzed; of these , half are well fit by a two-state model ( χred2; reduced chi-square < 10; Figure 2—source data 1 ) . This indicates that the ACD dimer is exchanging between a major species ( i . e . , the one revealed in the solution structure ensemble ) and one or more minor species . The nature of the minor species will be discussed below . Values for kex , the rate of exchange , for individual resonances ranged between 760 s−1 and 1119 s−1 . The less than two-fold difference for the range of values obtained from fitting data for individual resonances suggests that the residues that exhibit exchange could be involved in the same dynamic phenomenon . The residues that undergo exchange are along the dimer interface and in Loop 5/6 ( Figure 2B ) . These findings are consistent with the apparent plasticity of the dimer interface and Loop 5/6 inferred from crystal structures of the ACD dimer . Analysis of available HSPB5-ACD structures and NMR relaxation data combined indicate that ( 1 ) the HSPB5 protomer structure is retained over a wide pH range and ( 2 ) there are two regions of plasticity , namely the dimer interface and Loop 5/6 . 10 . 7554/eLife . 07304 . 008Figure 2 . 15N-CPMG relaxation dispersion experiments detect dimer-to-monomer exchange . ( A ) Relaxation dispersion measurements reveal a two-state transition . Representative relaxation dispersion curves of 15N transverse relaxation rate ( R2 , eff ) plotted as a function of field strength , ( νCPMG ) are shown ( see Materials and methods for details ) . Data were recorded at static field strengths of 800 MHz ( red ) and 600 MHz ( black ) at pH 7 . 5 and 22°C . Values of kex , δω , and pb were extracted using the program , GUARDD . ( B ) Backbone representation of HSPB5-ACD dimer is shown with the Cα atoms of exchanging residues that are well fit by a two-state model shown as spheres . Resonances showing relaxation rates in the range 760 s−1 to 1119 s−1 occur mainly at the dimer interface and in Loop 5/6 . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 00810 . 7554/eLife . 07304 . 009Figure 2—source data 1 . Parameters from relaxation dispersion experiments performed on 0 . 7 mM HSPB5-ACD at 22°C and pH 7 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 009 To assess thermodynamic stability of the ACD dimer , dimer–monomer dissociation constants were determined from isothermal titration calorimetry dilution measurements in which highly concentrated ACD was titrated into buffer . The resulting isotherms fit a simple dimer-monomer equilibrium model ( Table 2 ) . We made measurements at two physiological pH values , pH 7 . 5 and pH 6 . 5 ( 25°C ) , and obtained dissociation constants of 2 ± 2 μM and 30 ± 16 μM , respectively . The value at pH 7 . 5 is in excellent agreement with the value of 2 μM inferred from tandem MS/MS measurements performed at the same pH ( Hochberg and Benesch , 2014 ) . The dimer is also destabilized by an increase in temperature , with a KD of 36 ± 2 μM at pH 7 . 5 , 37°C . Thus , the HSPB5-ACD dimer is destabilized by either a decrease in pH from 7 . 5 to 6 . 5 or an increase in temperature from 25°C to 37°C . 10 . 7554/eLife . 07304 . 010Table 2 . Dimer-to-monomer dissociation constants for HSPB5-ACD determined by ITCDOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 010TemperaturepHKd ( mM ) ΔH ( cal/mol ) 25°C7 . 50 . 002 ± 0 . 0028772 ± 396025°C6 . 50 . 030 ± 0 . 0163242 ± 43237°C7 . 50 . 036 ± 0 . 0029328 ± 527*See ‘Methods and materials’ for experimental details and data analysis . 1H , 15N-HSQC spectra of HSPB5-ACD collected at pH 7 . 5 and 6 . 5 show extensive changes ( Figure 3A ) . Furthermore , spectra collected between pH 7 . 0 and pH 6 . 0 had significant peak doubling which is mostly resolved at pH 6 . 0 , making analysis of the pH-induced spectral changes based solely on existing resonance assignments at pH 7 . 5 ( Jehle et al . , 2009 ) challenging . Therefore , we assigned the HSPB5-ACD spectrum at pH 6 . 5 using standard triple resonance spectra collected at both pH 6 . 5 and pH 6 . 0 so that we could follow individual resonances through a pH titration series . An expanded region of the 1H , 15N-HSQC pH series is shown in Figure 3B . In the absence of another pH-dependent process , resonances arising from residues that undergo a protonation/deprotonation event as a function of pH will shift in a continuous manner . Resonances from residues that are proximal to titrating residues will also show similar behavior . Such processes will appear in the so-called ‘fast-exchange’ NMR regime due to the rapid on/off rate of acidic protons . At pH values above 6 . 7 , some resonances in the HSPB5-ACD spectrum exhibit fast-exchange pH behavior , consistent with protonation/deprotonation of ionizable groups . This behavior is illustrated by the resonance for His-119 in spectra collected at pH 6 . 7 and above ( Figure 3B ) . At pH < 6 . 7 , some resonances double and appear in a new position ( for example , resonances labeled 87 , 119 , and 135 in Figure 3B ) . Such behavior exemplifies intermediate-to-slow exchange and is indicative of the existence of two states that interconvert slowly on the NMR timescale . The relative intensities of the peaks reflect the relative population of the two states , so a new species of HSPB5-ACD is increasingly populated as the pH decreases . 10 . 7554/eLife . 07304 . 011Figure 3 . HSPB5-ACD undergoes a conformational transition between pH 7 . 5 and 6 . 5 . ( A ) 1H-15N HSQC spectra acquired on a 200 µM sample of HSPB5-ACD at pH 7 . 5 and 6 . 5 ( 22°C ) reveal two states . Full spectra collected at pH 7 . 5 ( black ) and 6 . 5 ( red ) are overlaid ( left ) . At pH 7 . 5 , 82/85 non-proline residues are observed ( residues G64 , L65 , and S139 are not detected due to fast exchange with H2O ) . At pH 6 . 5 , approximately ∼65 additional peaks appear , indicating the presence of two conformations . Boxed regions shown on the right provide clear examples of peak doubling . Some resonances ( labeled in the panel on the left ) disappear from their original positions at pH 7 . 5 and their new positions could not be determined by inspection of the spectrum at pH 6 . 5 . ( B ) Example of a full pH titration series ( spectra collected at pH 8 . 4 [blue] , 7 . 5 [black] , 7 . 0 [cyan] , 6 . 7 [green] , and 6 . 5 [red] ) . The behavior of the resonance of H119 as a function of pH illustrates two pH-dependent processes ( see text ) . Its chemical shift is in fast exchange from pH 8 . 4 to 7 . 5 ( solid arrow ) and changes direction and is in slow exchange from pH 7 . 0 to 6 . 5 ( dashed arrow ) . The region shown contains several other resonances that undergo slow-exchange transitions over the same pH range . ( C ) Residues that undergo the slow exchange transition are highlighted in color on a surface representation of the ACD dimer . Residues with perturbations > 0 . 2 ppm are red; those with perturbations between 0 . 1 - 0 . 2 ppm are orange . ( D ) Analysis of relaxation dispersion data yields the difference in 15N chemical shift between the major and minor species ( ΔδNcalc ) . The values of ΔδNcalc ( blue ) are compared to the experimental values obtained from the difference of 15N chemical shifts at pH 7 . 5 and 6 . 5 , i . e . , ΔδN ( 7 . 5−6 . 5 ) ( red ) , in the histogram . Concordance between these two parameters supports the notion that the minor form detected by relaxation dispersion experiments is the monomeric form of ACD that is populated at lower pH . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 011 Spectral assignments at pH 7 . 5 and 6 . 5 allowed identification of residues most affected by the pH-dependent transition . The chemical shift difference between the ‘high’ pH form ( pH 7 . 5 ) and the ‘low’ pH form ( pH 6 . 5 ) , Δδ ( pH7 . 5–pH6 . 5 ) was calculated for each NH resonance ( see ‘Materials and methods’ ) . Resonances that exhibit large chemical shift perturbations ( >0 . 2 ppm ) and/or change trajectory during the titration mainly arise from two structural regions ( colored red in Figure 3C ) : the dimer interface ( residues F113–I124 ) and Loop 5/6 . In light of the pH-dependent KD values obtained by ITC , the dissociation of dimers to monomers is a likely source of the slow exchange process observed in NMR spectra below pH 7 . The above findings raised the possibility that the minor species detected by relaxation dispersion at pH 7 . 5 is an ACD monomer . In addition to the exchange rate , the fractional population of the minor species ( pb = 1 − pa ) , and the chemical shift difference between the major and minor species for a given resonance ( δω ) can be extracted from analysis of relaxation dispersion data ( Kleckner and Foster , 2011 ) . Seven resonances provided an estimate for the fractional population of the minor species , pb , as ∼5% , in good agreement with the value of 4% calculated from the KD measured by ITC . We compared the predicted difference in 15N chemical shift ( δω ) to the experimentally determined difference in chemical shifts between the ACD at pH 7 . 5 and at pH 6 . 5 , ΔδN ( pH7 . 5- pH6 . 5 ) . As seen in Figure 3D , there is remarkably close correspondence for dimer interface residues , supporting the notion that the minor species detected by relaxation dispersion at pH 7 . 5 is the same as the monomeric form populated as the pH decreases below 7 . Finally , we performed relaxation dispersion measurements at 37°C ( pH 7 . 5 ) to determine the rates of association/dissociation at physiological temperature . For a dimer-to-monomer transition , the measured exchange rate will be concentration dependent:kex=kmd[monomer]+kdm , where kmd is the rate constant for monomer–dimer association and kdm is the constant for dimer dissociation . We collected relaxation dispersion data at two protein concentrations ( 700 and 200 µM; pH 7 . 5 , 37°C ) . Global fitting of relaxation dispersion data for residues in the dimer interface gave kex values of ∼1500 s−1 and 460 s−1 at 700 and 200 µM HSPB5-ACD , respectively . The concentration dependence confirms that the observed exchange is due to dimer-monomer dissociation/association . Based on a KD ( =kdm/kmd ) of 36 μM , the concentration of ACD monomers at 700 μM ACD subunits is ∼100 μM . This gives kmd = 1 . 1 × 107 M−1 s−1 ( in the range for diffusion-controlled bimolecular association; Northrup and Erickson , 1992 ) and kdm = 400 s−1 . Thus , several NMR parameters plus ITC measurements provide a picture in which the long dimer interface dissociates at a rate of 400 s−1 at pH 7 . 5 , 37°C and is destabilized as pH decreases below 7 . Under the same conditions , full-length HSPB5 subunits in oligomers exchange much more slowly , at a rate of 10−3 s−1 ( Peschek et al . , 2013 ) . Therefore , the rate at which subunits leave an oligomer is not determined by disruption of the ACD dimer and the results suggest that dimer interfaces within oligomers may break and reform many times during the residency of subunits within HSPB5 oligomers . The arrangement of the ACD dimer creates patches of positively and negatively charged surface that cross the dimer interface on both faces of the structure ( Figure 4A ) . At pH 7 . 5 and 22°C where the dimer is more stable , the electrostatic forces are presumably in balance , but the high densities of like charges suggest the stability is tenuous . A protonation event or other rearrangement at or near the dimer interface may tip the balance and trigger dimer dissociation . HSPB5-ACD has a high histidine content , with five His residues out of 89 total residues , ( 5 . 6% , which is more than double the average His frequency across all proteomes ) . In the dimer structure , eight of ten histidine residues ( five in each protomer ) are located in pairs ( two His-83/His-104 pairs and two His-101/His-119 pairs ) and are concentrated towards the center of the dimer ( Figure 4B ) . Overall , the organization and surface electrostatics suggested a possible role for histidine residues in triggering the pH-dependent dimer dissociation . 10 . 7554/eLife . 07304 . 012Figure 4 . His-104 plays a key role in the dimer-monomer transition . ( A ) The electrostatic surface of HSPB5-ACD at pH 7 . 5 ( calculated using experimentally determined histidine pKR values ) reveals patches of positive ( blue ) and negative ( red ) charges that cross the dimer interface . ( B ) The five histidines of HSPB5-ACD ( blue and cyan sticks ) occur as pairs and are located in proximity to the dimer interface . ( C ) Mutation of His-104 shifts the dimer-monomer equilibrium as observed by NMR . Overlays of 1H-15N HSQC spectra at pH 7 . 5 ( black ) and pH 6 . 5 ( red ) are shown for H104K-ACD ( left panel ) and WT-ACD ( right panel ) . A single set of peaks is observed in WT-ACD at pH 7 . 5 , and peaks due to the monomer conformation appear as the pH is lowered . Peaks belonging to the dimer interface , e . g . , R120 and R116 disappear from their original positions in WT-ACD at pH 6 . 5 and in H104K-ACD at both pH conditions ( dotted circles ) . ( D , E ) Expanded regions of 1H-15N HSQC spectra of WT- , H104K- , and H104Q-ACD . The same color scheme is used for WT- and H104K-ACD as in panel C; H104Q-ACD overlay is in blue ( pH 7 . 5 ) and red ( pH 6 . 5 ) . ( D ) Comparison of H104K-ACD ( left panel ) and WT-ACD ( right panel ) . Dotted lines connect resonances in similar positions in H104K-ACD and WT-ACD . The example shows that the peaks for residues 87 , 95 , 119 , and 136 have similar positions in H104K-ACD ( at both pH values ) as in WT-ACD at pH 6 . 5 . ( E ) Comparison of identical regions of spectra of H104K-ACD ( top left panel ) , WT-ACD ( top right panel ) , and H104Q-ACD ( lower panel ) . Both forms ( dimer and monomer ) are observed in the H104Q-ACD spectrum at both pH values , whereas only peaks corresponding to the monomer are observed in H104K-ACD at both pH values . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 01210 . 7554/eLife . 07304 . 013Figure 4—source data 1 . The pKR values and tautomeric states of HSPB5-ACD histidine side-chain imidazole rings ( 22°C ) are listed . n . d . : not determined because resonances broaden below pH 7 . 5 and become undetectable . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 01310 . 7554/eLife . 07304 . 006Figure 4—figure supplement 1 . His-104 is at the center of a dynamic network of charged and H-bonding interactions . Top panel: ten individual members of the solution ensemble are shown . One protomer of the dimer is in cyan; the other protomer is in gray . His-104 is shown as space-filling spheres; carbonyl groups that are in proximity to His-104 are shown in stick representation; side chains that are proximal to His-104 are shown in stick representation; side chains in Loop 5/6 and in the dimer interface as shown as lines; Bottom panel: examples of a loop-up ( shown in cyan; this publication PDB 2N0K ) and loop-down conformation ( shown in violet; from 2KLR ) are shown to illustrate the rearrangement of potential interacting partners in the two conformations . As above , the second subunit in a dimer is shown in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 006 To identify candidate histidines , pKR values and tautomeric states were determined from NMR chemical shifts of imidazole ring nuclei ( Nδ1 , Nε2 , Hδ2 , and Hε1 ) measured as a function of pH . pKR values were obtained for four of five His residues: His-101 and His-111 have low values of <6 and His-83 and His-119 have values of 6 . 6 and 7 . 7 , respectively ( Figure 4—source data 1 ) . A pKR could not be determined for His-104 because its side chain resonances broaden and disappear at pH values below 7 . 5 . Though His-104 is not positioned on the dimer interface , the behavior of its backbone NH resonance is similar to backbone NH resonances on the dimer interface undergoing exchange between dimer and monomer , indicating that it undergoes a change in environment as a result of the dimer-to-monomer transition . At pH 7 . 5 where its imidazole resonances are detectable , His-104 exists as the less common Nδ1H tautomer , while the other four His residues are in the more common Nε2H tautomer . A resonance is observed at a chemical shift of 12 . 4 ppm for the Nδ1H of His-104 , indicating that it is hydrogen bonded . Among members of the solution ensemble , there are several potential H-bonding partners in proximity to His-104: the backbone carbonyls of His-83 , Glu-105 , and Glu-106 and side chain groups of Arg-107 and Arg-116 ( Figure 4—figure supplement 1 ) . Together , the data reveal that His-104 is in a specific conformation in the ACD dimer at pH 7 . 5 that is stabilized through an H-bond . The sensitivity of His-104 resonances to the dimer-to-monomer transition suggested it as a prime candidate for the pH trigger . His-104 was substituted with either Gln or Lys and the mutant ACDs were compared to WT-HSPB5-ACD . As discussed above ( Figure 3A , B ) , peak doubling is observed in the spectrum of WT ACD below pH 7 , with new peaks appearing at lower pH corresponding to ACD monomers . The 1H-15N HSQC spectrum of H104Q-ACD has peak doubling even at pH 7 . 5 and the two sets of peaks correspond to dimer and monomer peaks already assigned in the WT spectrum ( Figure 4E , lower panel ) . The H104K-ACD HSQC spectrum has a single set of resonances under both pH conditions , but the peaks correspond to the monomer , even at pH 7 . 5 ( Figure 4C–E ) . Thus , substitution of His-104 with glutamine destabilizes the dimer so that approximately equal populations are observed at pH 7 . 5 and NMR concentrations . Glutamine at position 104 would have difficulty mimicking the Nδ1H-bond formed by His-104 . Substitution of His-104 with a positively charged lysine yields an ACD that is predominantly monomeric at pH 7 . 5 . We did not determine KD values for the mutants by dilution ITC experiments because the starting ACD concentration must be ten-fold above KD . Based on the relative populations of dimer and monomer peaks in the NMR spectra , we can estimate that the KD for H104Q-ACD is ∼100–500 μM ( approximately equal populations of dimer and monomer at [ACD] = 200 μM ) and the KD for H104K-ACD is >1 mM ( based on our inability to detect dimer resonances in samples containing 200 μM protein ) . Although only order-of-magnitude , these estimates reveal that substitution of His-104 with Gln or Lys destabilizes the dimer more than 100-fold . To ascertain whether the effect of mutating His-104 is specific , the other four His residues were each mutated to Gln . The mutant ACDs retained the pH-dependent spectral behavior of the WT-ACD ( data not shown ) . Altogether , the data implicate His-104 , which sits at the base of Loop 5/6 ( Figure 4B ) , as a hotspot whose conformation , H-bonding capacity , and protonation state strongly affect the stability of the HSPB5-ACD dimer interface . To uncover consequences of dimer interface stability in HSPB5 oligomer structure , we compared properties of full-length H104Q- and H104K-HSPB5 with WT-HSPB5 . Size exclusion chromatography combined with multi-angle light scattering ( SEC-MALS ) provided information on oligomer size and polydispersity and negative-stain electron microscopy ( EM ) allowed observation of single oligomeric particles ( Figure 5 ) . SEC-MALS data of WT-HSPB5 oligomers confirmed previously reported pH-dependent changes in oligomeric dimension ( Jehle et al . , 2011; Baldwin et al . , 2011a ) . At pH 7 . 5 , the molecular weight ( Mw ) determined at intervals across the elution peak ranged from 440 kDa to 500 kDa . The average Mw of 465 kDa corresponds to the mass of ∼24 subunits , consistent with previous reports ( Aquilina et al . , 2003; Peschek et al . , 2009 ) . At pH 6 . 5 , WT-HSPB5 oligomers elute earlier , with an average Mw of 720 kDa , corresponding to ∼36 subunits per oligomer—an increase of 12 subunits on average per oligomer ( Figure 5A ) . Negative-stain EM micrographs and single particle images collected on WT-HSPB5 oligomers at pH 7 . 5 and 6 . 5 reveal large globular structures at both pHs ( Figure 5—figure supplement 1A ) . In 2D projection averages ( generated by reference-free alignment of single particle data sets ) , WT-HSPB5 oligomers are generally spherical and have varied diameters that are on average greater at pH 6 . 5 than at pH 7 . 5 ( Figure 5B , Figure 5—figure supplement 1B ) . Thus , SEC-MALS and EM measurements reveal an expansion in the hydrodynamic radius , particle dimensions , and number of subunits in WT-HSPB5 oligomers as a function of decreasing pH . 10 . 7554/eLife . 07304 . 014Figure 5 . Destabilizing the ACD dimer interface via low pH or His-104 mutation triggers large expansion of HSPB5 oligomers . ( A ) SEC-MALS analysis showing protein elution profile ( refractive index , right Y-axis ) with average Mw ( horizontal trace under peak corresponds to left Y-axis ) for WT-HSPB5 at pH 7 . 5 ( blue ) and pH 6 . 5 ( green ) , and H104Q-HSPB5 ( orange ) and H104K-HSPB5 ( red ) at pH 7 . 5 . ( B ) 2D projection class averages showing representative ensemble of oligomer sizes . ( C ) SEC-MALS analysis showing elution profile and average Mw of mixed oligomers of WT and H104K-HSPB5 incubated together at ratios of 1:0 ( green ) , 2:1 ( light green ) , 1:1 ( yellow ) , 1:2 ( orange ) , and 0:1 ( dark red ) , respectively . ( D ) Histogram of oligomer diameters showing fraction of oligomer particles vs average diameter ( nm ) for WT at pH 7 . 5 ( blue ) and pH 6 . 5 ( green ) , and H104K at pH 7 . 5 determined by negative-stain EM 2D classification . Representative 2D class averages corresponding to measured diameter are shown . Experiments were performed in triplicate . Scale bar ( lower left image ) equals 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 01410 . 7554/eLife . 07304 . 015Figure 5—figure supplement 1 . EM micrograph images , 2D classification , and particle diameter estimation for WT and H104K HSPB5 . ( A ) Representative micrograph images of WT-HSPB5 at pH 7 . 5 and 6 . 5 and H104K-HSPB5 at pH 7 . 5 negatively stained with 0 . 75% uranyl formate ( scale bar equals 20 nm ) . Example individual particle projections are shown below with scale bar equal to 10 nm . ( B ) 2D reference free class averages for WT-HSPB5 pH 7 . 5 , WT-HSPB5 pH 6 . 5 and H104K-HSPB5 pH 7 . 5 from data sets of 5 , 321 , 5713 , and 5145 single particles for respectively . Scale bar equals 10 nm . ( C ) Example class averages , corresponding rotational averages and estimated diameter ( nm ) for HSPB5 oligomers . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 015 To ask if the change in oligomer size observed with pH is a consequence of dimer interface destabilization , the His-104 mutants were analyzed by SEC-MALS and EM at pH 7 . 5 . H104K-HSPB5 forms very large , polydispersed oligomers ranging from 785 to 840 kDa ( Figure 5A ) , with an average Mw of 810 kDa ( ∼41 subunits ) . H104Q-HSPB5 oligomers elute at an intermediate position relative to WT- and H104K-HSPB5 . Thus , more subunits are recruited when oligomers are assembled from destabilized dimers . We propose that oligomers composed of H104K-HSPB5 are made predominantly of monomeric units while oligomers composed of H104Q-HSPB5 contain both dimeric and monomeric units . If this model is accurate , we should be able to recapitulate intermediate-sized H104Q-containing oligomers by mixing together WT- and H104K-containing subunits . As shown in Figure 5C , the elution position and oligomer mass in mixtures of WT-HSPB5 and H104K-HSPB5 titrate as a function of added H104K-HSPB5 subunits , consistent with a model in which oligomers are formed from both dimeric and monomeric units . WT-HSPB5 oligomers at pH 6 . 5 have a mass range that corresponds to a 1:2 ratio of WT:H104K subunits , indicating that the pH-dependent oligomeric assemblies likely arise from a combination of monomeric and dimeric subunits . The results above suggest that H104K-HSPB5 oligomers represent an endpoint in a continuum of oligomer composition and that H104Q- and ( pH 6 . 5 ) WT-HSPB5 are reflective of species that are favored under differing conditions including stress-induced acidosis . To ask if oligomers composed uniquely or partly of monomeric units have distinct structural features , negative-stain EM images of H104K-HSPB5 at pH 7 . 5 and WT-HSPB5 at pH 6 . 5 were collected and analyzed . In a majority of images , defined features , identified as light and dark regions of density , are apparent within the structures , indicating an organized arrangement of subunits . The particles are generally spherical , with some oblong-shaped structures also observed . Quantitative information on the size and distribution of oligomers was obtained by measuring the diameter of circular averages generated for each 2D projection average ( Figure 5—figure supplement 1C ) . Single particles were then binned based on diameter of their corresponding average to generate a distribution ( Figure 5D ) . The size distribution derived in this way is consistent with the SEC-MALS analysis: WT-HSPB5 at pH 7 . 5 has the smallest average diameter , 15 nm , ( a range of 13–17 nm ) , H104K-HSPB5 has the largest , with an average of 17 nm , and WT-HSPB5 at pH 6 . 5 is intermediate , at 16 nm . Oligomers of WT-HSPB5 at pH 6 . 5 cover the entire range of diameters measured , between 14 and >18 nm , while the distribution of H104K-HSPB5 is skewed to the largest dimensions , with a majority of particles ( >70% ) greater than 16 nm . For comparison , the diameter of the WT-HSPB5 cryo-EM model ( Braun et al . , 2011 ) determined similarly from the circular average measures 15 nm , in good agreement with our measurements . Altogether the results reveal that conditions and/or mutations that affect the stability of the antiparallel dimer interface have substantial impact on HSPB5 oligomer size and structure . To ascertain the functional and mechanistic consequences of the expanded oligomer structures described above , standard holdase assays in which aggregation of a model client protein is monitored as a function of time in the absence and presence of sHSP were performed . At pH 7 . 5 , H104Q- and H104K-HSPB5 are more effective than WT-HSPB5 at delaying formation of large aggregates by two model clients , αLactalbumin ( αLac ) destabilized by addition of DTT , and alcohol dehydrogenase ( ADH ) destabilized by addition of DTT and EDTA ( Figure 6 ) . The results indicate that enhanced holdase activity is associated with destabilization of the dimer interface . 10 . 7554/eLife . 07304 . 016Figure 6 . His-104 mutants of HSPB5 are effective at delaying the onset of aggregation of a model client protein . ( Top panel ) Aggregation of DTT-denatured bovine αLactalbumin at 42°C in the absence ( green ) and presence of WT-HSPB5 ( blue ) or HSPB5 mutants H104K ( red ) and H104Q ( orange ) . Light scattering at 360 nm was used to monitor the DTT-induced aggregation of αLac ( 600 µM ) in the presence and absence of HSPB5 ( 40 µM ) . Assays were performed in duplicate and the average scattering curves are shown . ( Bottom panel ) Aggregation of Yeast Alcohol Dehydrogenase in the presence of EDTA and DTT , at 37°C in the absence ( green ) and presence of WT-HSPB5 ( blue ) or HSPB5 mutants H104K ( red ) and H104Q ( orange ) . Light scattering with 360 nm light was used to monitor ADH ( 100 µM ) aggregation in the presence and absence of HSPB5 ( 20 µM ) . Assays were performed in duplicate and the average scattering curves are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 016 To understand how destabilization of the dimer interface yields more effective holdases , we compared how HSPB5 and the mutants interact with clients . We took advantage of the long delay in αLac aggregation afforded by both the WT and mutant HSPB5 species to attempt to detect sHSP-client interactions . Mixtures of αLac and one of the HSPB5 species were analyzed by SEC-MALS following addition of DTT and before the onset of aggregation ( Figure 7 ) . Although WT-HSPB5 maintains αLac in a soluble form during the time frame of the experiment , no interaction between the two proteins was detected by SEC or SDS-PAGE , indicating that its holdase function is achieved through highly transient interactions ( Figure 7A ) . In contrast , a mixture of αLac and H104K-HSPB5 elutes with a dramatically different profile . SDS-PAGE analysis across the broad peak that elutes between 7 . 5 and 9 . 2 ml shows that it contains both proteins ( Figure 7B ) . There is a broad range of molecular weights across this peak , from >550 kDa to ∼250 kDa . A similar but less dramatic change in the elution profile was obtained for a mixture of αLac and H104Q-HSPB5 ( Figure 7C ) . Thus , long-lived complexes are formed between αLac and the H104 mutant HSPB5s , and these are markedly smaller than the oligomers that are populated in the absence of client protein . 10 . 7554/eLife . 07304 . 017Figure 7 . H104K- and H104Q-HSPB5 oligomers reorganize into small , long-lived client-bound complexes in the presence of αLac model client protein . ( A–C ) SEC-MALS analysis and corresponding Mw of WT ( A ) , H104K ( B ) and H104Q ( C ) , HSPB5 oligomers ( 40 μM in subunit concentration ) incubated with αLac ( 120 μM ) in the absence ( blue ) and presence ( red ) of 50 mM DTT to destabilize αLac . SDS-PAGE analysis of peak fractions for corresponding HSPB5-αLac incubations with DTT ( red trace ) is shown ( inset ) . Dashed lines correspond to WT elution at 8 ml and H104K elution at 7 . 4 ml . ( D ) 2D projection class averages of H104K- and H104Q-HSPB5-αLac peak fractions corresponding to peaks I , II , and III ( gray bars ) in B and C are shown with representative averages of WT , H104K- , and H104Q-HSPB5 incubated without αLac substrate for comparison . Scale bar ( lower left panel ) equals 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 01710 . 7554/eLife . 07304 . 018Figure 7—figure supplement 1 . EM micrograph images and 2D classification of HSPB5 incubated with αLac . ( A ) Representative micrograph images and single particles of WT-HSPB5 and H104K-HSPB5 incubated with or without αLac substrate and DTT prior to fractionation by SEC-MALS showing the shift in oligomer size only occurs with H104K-HSPB5 and α-Lac . Scale bars equal 20 nm and 10 nm for the micrographs and single particles , respectively . ( B ) Complete set of 2D projection class averages of H104K-HSPB5-αLac isolated in Peak 1 and Peak II following SEC-MALS fractionation . Class averages were obtained from 11 , 727 particles in fraction I and 9257 particles in fraction II are shown and the scale bars equal 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07304 . 018 Negative-stain EM was performed on fractions collected across the peak containing both H104K-HSPB5 ( or H104Q-HSPB5 ) and α-Lac ( Figure 7D ) . In micrograph and single particle images , the particles appear intact and there is little background from unbound protein , indicating the mutant HSPB5/αLac complexes remain stable following dilution for EM ( Figure 7—figure supplement 1 ) . The particles are much smaller and are structurally distinct from the spherical oligomers observed for WT- , H104Q- , or H104K-HSPB5 on their own . A large single particle data set of fractionated H104K-HSPB5/αLac complex was collected , and 2D projection averages were determined as above . The 2D averages ( Figure 7D ) show clear structural features and reveal that a major structural reorganization occurs in the mutant sHSP upon binding of a model client protein . Among the 2D averages , some smaller spherical oligomers similar to WT-HSPB5 are observed , but the ultra-large ( >16 nm ) oligomers that predominate in H104K-HSPB5 alone are no longer detected . The sizes are highly variable , likely depending on the orientation , stoichiometry , and/or oligomeric state of the complex . To see if similar HSPB5/αLac complexes form with WT-HSPB5 but do not survive the SEC-MALS experiment , unfractionated samples of mixtures of WT-HSPB5 and α-Lac were analyzed by negative stain EM ( data not shown ) . No smaller species were detected and the spherical oligomers remain unchanged for WT-HSPB5 . Overall the results demonstrate that HSPB5 oligomers with destabilized dimer interfaces disassemble and undergo a dramatic reorganization to form stable complexes with a model client protein while the wild-type protein at pH 7 . 5 does not do so and performs its holdase function via highly transient interactions with client . Despite their key roles in maintaining cellular protein solubility under and following stress conditions , the determinants of sHSP structure and function remain ill defined . Stresses such as ischemia and hypoxia are associated with acidosis in which cellular pH can decrease to as low as pH 6 . 4 ( McVicar et al . , 2014 ) . As HSPB5 must function under stress conditions , we sought to investigate effects of pH on its structure and function . sHSPs have three structural regions , each of which plays a role in oligomer assembly . The highly conserved ACD is necessary and sufficient for dimer formation and dimers are thought to be the fundamental building blocks for oligomers . The N- and C-terminal regions appear to drive oligomer assembly , although the details of these interactions are not yet elucidated . We find that HSPB5-ACD dimers dissociate into folded monomers that are favored as the pH decreases . ITC and NMR measurements revealed a modest affinity for the ACD dimer ( KD of 36 μM at pH 7 . 5 , 37°C ) , with dimers dissociating at a rate of 400 s−1 . The long β6+7 strands of two subunits align in an antiparallel fashion to form the dimer interface . This structure with its associated H-bonds might be expected to afford higher affinity . The juxtaposition of Glu-117 and Glu-117′ in the middle of the interface and the positively and negatively charged patches that cross the interface may contribute to the modest affinity and to the dynamics and plasticity we observed in the interface . Dimer affinity decreases 15-fold over 1 pH unit ( pH 7 . 5 to pH 6 . 5 ) and this likely corresponds to a similar fold increase in the rate of dissociation at the dimer interface . A pH-dependent destabilization of the HSPB5-ACD dimer interface was inferred from native MS studies performed on oligomers as a function of pH; our results offer a direct confirmation and quantification of this hypothesis ( Baldwin et al . , 2011b ) . We identified a histidine residue that plays a key role in ACD dimer interface stability . Substitution of His-104 with Gln or Lys decreases dimer stability substantially . NMR spectra reveal that the His-104 ring is in the more unusual of the two possible tautomeric forms , stabilized by serving as an H-bond donor . Analysis of members of the solution structure ensemble and crystal structures with the Loop 5/6-up conformation identifies several backbone carbonyls as potential H-bond acceptors for His-104: His-83 , Glu-105 , and Glu-106 . Alteration of the H-bonding potential or geometry ( as in the H104Q mutant ) or charge ( as in the H104K mutant and low pH ) destabilizes the dimer . Among the ten human sHSPs , His-104 is the most conserved His residue , appearing in eight of ten proteins , with the non-H104-containing sHSPs having a Gln ( HSPB9 ) or a Lys ( HSPB7 ) ( Figure 1—source data 2 ) . Consistent with a conserved role for His-104 , the unusual tautomeric state adopted by His-104 in HSPB5 is also observed for the analogous histidines in HSPB1 ( His-124 ) and HSPB6 ( His-103 ) ( Rajagopal and Klevit , unpublished observation ) . Identification of the role of His-104 in dimer stability was unexpected because while there are histidines on the dimer interface , His-104 is not one . His-104 is located at the beginning of Loop 5/6 , the loop connecting strands β5 and the dimer interface strand , β6+7 ( Figure 4B ) . NMR relaxation dispersion analysis of the ACD dimer revealed that residues in the dimer interface ( 116–123 ) exist in two states consistent with a dimer-monomer equilibrium ( Figure 2B ) . Residues in Loop 5/6 have similar exchange rates suggesting that loop movements and the structural transition may be coupled . However , a rigorous demonstration that the dynamics of the two regions are coupled requires a robust determination of kex values as a function of [ACD] for loop residues as well as interface residues and the data collected did not allow for this determination with sufficiently high confidence . Nevertheless , consistent with this notion , inter-subunit contacts are observed that involve Loop 5/6 residues: His-111 at the apex of Loop 5/6 makes contacts to Arg-120’ and Tyr-122’ across the dimer interface . These inter-subunit contacts help to position Loop 5/6 in a ‘loop up’ conformation in the solution structure at pH 7 . 5 ( Figure 1E , Figure 4—figure supplement 1 ) . A majority of residues in Loop 5/6 are charged ( H104-E-E-R-Q-D-E-H111 ) so a possible consequence of His-104 protonation , which sits at the base of the loop , is to alter the conformation or position of Loop 5/6 , disrupting inter-subunit contacts involving the loop and shifting the dimer–monomer equilibrium of the HSPB5-ACD . In structures with the ‘loop up’ conformation , side chains of Arg-107 and Arg-116 are in proximity to His-104; the ‘loop down’ conformation moves Arg-107 away from His-104 and Glu-106 closer to His-104 ( Figure 4—figure supplement 1 ) . In sum , we propose that in both loop conformations His-104 sits at the center of a dynamic network of electrostatic and H-bonding interactions that is responsible for modulating the stability of the dimer interface . A survey of all 17 available mammalian sHSP-ACD structures ( 9 HSPB5 [WT and mutants] , 3 HSPB1 , and 3 HSPB6 ) reveals that Loop 5/6 is found in two conformations; a ‘loop up’ conformation as seen in the HSPB5-ACD solution structure and a ‘loop down’ conformation . In the ‘loop up’ structures , the details of the interactions between Loop 5/6 residues across the dimer interface vary , again pointing to the plasticity of this region . Nevertheless , the residues corresponding to His-111 , Asp-109 , and Arg-120 are involved in all cases . Furthermore , although Loop 5/6 contains predominantly polar and/or charged residues , it is remarkably well conserved among the human sHSPs ( Figure 1—source data 2 ) . There are several reported examples where Loop 5/6 residues affect the dimer-monomer equilibrium . In HSPB5 , residues His-104 , Asp-109 , and His-111 have recently been implicated in Cu2+ and Zn2+ binding and the binding of divalent cation appears to destabilize the dimer interface ( Mainz et al . , 2012 ) . In HSPB1 , which has an almost identical Loop 5/6 sequence to HSPB5 , substitution of the two glutamate residues that correspond to HSPB5 Glu-105 and Glu-106 to Ala resulted in monomeric HSPB1-ACD ( Baranova et al . , 2011 ) . Also , a Loop 5/6 mutation in HSPB1 , R127W ( corresponding to Arg107 in HSPB5 numbering ) is reported to promote monomer over dimer ( Almeida-Souza et al . , 2010 ) . In each case , a change in the net charge of Loop 5/6 leads to monomer being more favored . Altogether the observations suggest a model in which properties of Loop 5/6 are coupled to the stability of the dimer interface . It is notable that inheritance of single alleles of D109H-HSPB5 or R127W-HSPB1 is associated with cataract , myofibrillar myopathy , and distal hereditary motor neuropathy ( Nefedova et al . , 2013 ) . Both mutations alter the net charge of Loop 5/6 , implying a critical role for structural and/or dynamic properties of Loop 5/6 in sHSP function . Moving into the context of full-length HSPB5 , we found that decreasing pH or substituting His-104 has dramatic effects on oligomer dimensions and subunit stoichiometry . A continuum of oligomer sizes is observed that trends reciprocally with ACD dimer interface stability . H104K-HSPB5 , whose ACD will be predominantly monomeric , assembles into the largest oligomers containing over 40 subunits . While the notion that dimers are the fundamental building blocks of sHSP oligomers may describe the situation under certain conditions , our results indicate that oligomers can also be built from monomeric subunits and from combinations of dimeric and monomeric units and that more monomeric subunits can be incorporated into a given oligomer . This revelation will be important in any future attempts to determine HSPB5 oligomer structures for oligomers larger than the 24-mer previously determined ( Braun et al . , 2011; Jehle et al . , 2011 ) . The His-104 mutants allowed us to ask if and how the dimer interface contributes to holdase function without the confounding complications of comparing aggregation of model denatured client proteins at differing pHs . Both dimer-destabilizing mutations yield an HSPB5 that can delay aggregation of model clients longer than the wild-type protein at pH 7 . 5 . Unexpectedly , the His-104 mutant proteins , which form much larger oligomers than WT-HSPB5 , reorganize to form long-lived complexes with a client protein . EM images reveal particles that are markedly smaller and distinct from the mostly spherical sHSP-alone structures . During the time period in which WT-HSPB5 inhibits client protein aggregation , we were unable to detect a complex with the client , either by SEC or by EM . Thus , the mutant sHSPB5 species perform holdase function via a different mechanism from the wild-type protein at pH 7 . 5 . WT-HSPB5 acts via weak-and-transient interactions while the mutants act via stronger-and-longer interactions . Notably , two inherited missense mutations in HSPB1 associated with Charcot-Marie-Tooth syndrome , R127W-HPSB1 ( in Loop 5/6 ) and S135F-HSPB1 ( on the dimer interface ) , engage in stronger , longer-lived interactions with client proteins than the WT-HSPB1 , as evidenced from tandem affinity purification ( Almeida-Souza et al . , 2010 ) . Two client binding modes have been observed in studies of the highly related HSPB4 ( αA-crystallin ) , dubbed ‘high capacity’ and ‘low capacity’ based on differing client:sHSP stoichiometries ( McHaourab et al . , 2002 ) . It remains for future studies to ascertain whether these reflect the species observed and reported here . Solid-state NMR studies of WT-HSPB5 oligomers at pH 7 . 5 revealed three types of inter-subunit interactions: ( 1 ) ACD-to-ACD ( i . e . , the dimer interface ) , ( 2 ) C-terminal region-to-ACD , and ( 3 ) N-terminal region to either ACD or other N-terminal regions ( Jehle et al . , 2011 ) . The findings in the current study indicate that the relative strength and , perhaps , abundance of these interactions can affect not only the size and structure of oligomers but also the way in which client proteins are recognized . Subunit exchange in WT-HSPB5 is orders of magnitude slower than the exchange rate we measured for the dimer-to-monomer transition ( 10−3 s−1 vs ca . 103 s−1 , respectively , at pH 7 . 5 , 37°C; Peschek et al . , 2013 ) . So , while breaking the dimer interface is not the rate-limiting step in subunit exchange , destabilizing the ACD dimer yields oligomers that can more readily disassemble to form smaller species with clients and these complexes must be more stable than the mutant sHSP oligomers themselves . Long-lived sHSP-client complexes have been detected by SEC for Hsp18 . 1 from peas , but the complexes formed are larger than those of Hsp18 . 1 in the absence of client ( Lee et al . , 1997; Stengel et al . , 2010 ) . Hsp26 from yeast also forms larger complexes with client than in the absence ( Franzmann et al . , 2005 ) . These examples contrast with our findings , which imply that HSPB5 has evolved to perform its holdase function via weak-and-transient interactions and that small perturbations can unleash a second , cryptic mode of client interaction ( long-and-strong ) . The highly charged ACD dimer interface and the proximal Loop 5/6 provide exquisite sensitivity to small changes in electrostatic environment that can be effected by slight changes in pH , temperature , divalent cations , other cellular conditions , and mutations . Of the approximately twenty inherited disease-related missense mutations documented in ACD regions of human sHSPs , more than half are in residues on the dimer interface or in Loop 5/6 and all these involve substitutions of charged residues . Our original intent was to define how HSPB5 functions under pH conditions associated with stress-induced acidosis . Our results show unequivocally that the dimer interface stability decreases over a physiologically relevant pH range and that mutation ( or protonation ) of a single histidine residue is sufficient to destabilize the dimer . The His-104 mutants provide several general insights regarding the modulation of structure and function of sHSPs . First , the stability of an ACD dimer interface relative to other interactions involved in oligomer formation can modulate holdase function . In our study , destabilization of the dimer interface leads to enhanced holdase activity . It remains to be seen if the converse will be true . Second , dimer stability can be modulated by residues other than the dimer interface itself . Such effects are likely achieved through a network of conserved charged residues that ultimately favor or disfavor the dimer over monomer . Third , relatively small changes in pH or single missense mutations are capable of shifting HSPB5 from a weak , transient mode of client interaction to one that involves long-lived co-complexes . We propose that the continuum of oligomeric structures we observe for HSPB5 under differing pH ( or mutation ) may lead to a continuum of client-binding modes that allow the sHSP to ramp its holdase activity up or down as conditions require . A thorough understanding of the ways in which sHSP structure and activity can be modulated under differing cellular conditions will ultimately provide much needed insights into the cellular functions of this important , but previously intractable class of protein chaperones . HSPB5 and HSPB5-ACD ( residues 64-152 ) were expressed and purified as described previously ( Jehle et al . , 2009 ) . Site-directed mutagenesis was performed with Quik-Change mutagenesis kit from Sigma . The growth and purification protocols of mutant proteins were similar to that of wild-type proteins . Resonances in the NMR spectrum of WT-ACD were previously assigned ( Jehle et al . , 2009 ) . To obtain distance restraints from NOES , 15N-edited NOESY and 13C-edited NOESY spectra on aliphatic and aromatic groups were acquired on a 1 mM , 13C , 15N-WT-ACD sample in NMR buffer ( 50 mM sodium phosphate , pH 7 . 5100 mM NaCl , 0 . 1 mM EDTA , and 1 mM PMSF ) . The spectra were acquired on a Bruker 950 MHz US2 ( ultra-shield , ultra-stabilized ) spectrometer equipped with Avance III console and a z-gradient , triple resonance cryoprobe ( David H Murdock Research Institute in Kannapolis , North Carolina ) . 15N-edited NOESY and 13C-edited spectra were acquired with a mixing time of 120 ms in 90%H2O/10%D2O solution at 22°C and 100% D2O solution at 37°C , respectively . Data were processed with NMRPipe ( Delaglio et al . , 1995 ) and analyzed with NMRViewJ ( Johnson , 2004 ) and CcpNmr ( Vranken et al . , 2005 ) . NOEs were binned into short ( 3 Å ) , medium ( 4 Å ) , and long-range constraints ( 5 Å ) and input as distance restraints into RosettaOligomer ( Sgourakis et al . , 2011 ) . Intermolecular NOEs in homodimeric proteins are usually obtained from edited/filtered-type NOESY experiments on a mixed sample containing labeled and unlabeled protein . This method failed in the case of WT-ACD due to signal-to-noise issues . A preliminary structure of the dimer was determined with CS-Rosetta ( Vernon et al . , 2013 ) and RosettaDock ( Schueler-Furman et al . , 2005 ) ( see below ) . The β-sandwich fold was determined from CS-Rosetta using backbone chemical shifts . RosettaDock gave a model of the dimer in the APIII register ( where residue R116 from each subunit is across from each other ) . Using this preliminary model , intra- and inter-molecular NOEs could be parsed out from 15N-edited and 13C-edited NOESY spectra . The Hα-Hα NOEs observed in the 13C-edited NOESY spectra unambiguously confirmed the APII register as the dimer interface ( Glu-117-Glu-117′ across from each other ) . Intra- and inter-molecular NOE restraints , 1H-15N RDCs , and all chemical shifts including the backbone and side-chain were input into RosettaOligomer for the final determination of the WT-ACD dimer structure . 1H-15N residual dipolar couplings ( RDCs ) were measured on a 500-µM protein sample dissolved in 500 µl of NMR buffer containing 10% Pf1 phage obtained from ASLA biotech . IPAP ( In-Phase/Anti-Phase ) 1H-15N HSQC spectra were acquired in-house on a Bruker Avance III 800 MHz spectrometer equipped with a z-gradient , triple resonance cryoprobe . Spectra were analyzed in NMRView to obtain the values of RDCs . The program PALES ( Zweckstetter et al . , 2004 ) was used to calculate RDCs for the different structures published in literature . The Rosetta symmetric fold-and-dock protocol can be used to determine the structure of symmetric homodimers . Starting from an extended chain , this protocol simultaneously explores the folding and docking degrees of freedom . It consists of four low-resolution stages of increasing complexity in the energy function , in which symmetric fragment insertions are interleaved with symmetric rigid-body trials . During the low-resolution step , side chains are represented using a single , residue-specific pseudo-atom , positioned at the Cα carbon . Finally , symmetric repacking of the side chains and gradient-based minimization of the side chain , rigid body , and backbone degrees of freedom are applied . In this high-resolution step , Rosetta's full-atom energy function is used . The conformational search is largely guided by experimental data , including intra- and inter-molecular NOE distance constraints and RDCs . A penalty term that is proportional to the rmsd between experimental and calculated data was used in Rosetta during the Monte Carlo trials and gradient-based minimization . NOE distances are modeled as atom pair constraints in Rosetta . For a structural model , RDCs are fitted using Levenberg–Marquardt non-linear square fitting algorithm . The orientations of alignment tensor are optimized , while keeping the axial component ( Dα ) and Rhombic component ( R ) of the alignment tensor fixed . The values of Dα and R of 21 . 5 and 0 . 35 , respectively , were estimated from a powder pattern distribution of the RDC data . A total of 10 , 000 models are generated , of which 1000 lowest ones are selected for cluster analysis . Ten models with the lowest Rosetta full-atom energy in the best-ranked cluster were selected as the final structural ensemble and deposited in the Protein Databank as PDB 2N0K . 1H-15N HSQC-TROSY spectra were acquired on WT-ACD ( 200 µM ) in NMR buffer at pH values 9 . 3 , 8 . 56 , 7 . 86 , 7 . 44 , 7 . 0 , 6 . 83 , 6 . 7 , 6 . 52 , and 6 . 0 at 22°C on an in-house Avance III 500 MHz spectrometer equipped with a z-gradient , triple resonance probe . The pH of the solution was adjusted by adding small aliquots of 1N HCl or NaOH . The long-range correlations between the ring carbon-bound hydrogens ( Hε1 and Hδ2 ) and the ring nitrogens ( Nε2 and Nδ1 ) give information on the tautomeric states of the histidines ( Pelton et al . , 1993 ) . These correlations were observed in 1H-15N HSQC spectra using WATERGATE for water suppression and by setting the INEPT delay to an integral multiple of 1/JNH where JNH is the value of the single bond N-H coupling constant . pKR values of histidine residues were determined by following the chemical shifts of Hε1 , Nε2 , and Nδ1 atoms as a function of pH . Non-linear regression fitting of chemical shifts vs pH was performed with Prism ( GraphPad ) using a modified version of the Henderson–Haselbach equation:δobs=δHA+δA−10pH−pKr1+10pH−pKr , δobs is the observed chemical shift at a specific pH value , and δHA and δA are the chemical shifts in the fully protonated and deprotonated state , respectively . CSPs ( Δδ ( pH7 . 5–pH6 . 5 ) ) were computed as follows:Δδ=1/2 ( ( ( ΔHN ) 2+ ( ΔN5 ) 2 ) ) , where Δδ is the chemical shift difference of an amide group at pH 7 . 5 and 6 . 5 , ΔHN and ΔN are the amide proton and 15N backbone amide chemical shift differences , respectively . To probe conformational fluctuations in the ms timescale , 15N effective relaxation rates ( R2 , eff which is the sum of the intrinsic relaxation rate , R20 and the chemical exchange rate , Rex ) were measured for WT-ACD using pulse sequences described in literature ( Korzhnev et al . , 2004 ) . Experiments were performed on 200 and 700 µM samples at 22°C and 37°C at two field strengths , 800 and 600 MHz . The values of νcpmg used were 25 , 50 , 75 , 100 , 150 , 175 , 200 , 300 , 600 , and 1000 Hz . A recycle delay of 2 . 2 s between scans and a total CPMG delay ( T ) of 20 ms was used . Spectra were processed and analyzed with NMRPipe , and the fits of intensities vs νcpmg were analyzed with the program , GUARDD ( Kleckner and Foster , 2012 ) . 32 out of 85 observable residues in WT-ACD exhibit values of R2 , eff ( ∞ ) –R2 , eff ( 0 ) > 8 s−1 at 22°C , where R2 , eff ( ∞ ) and R2 , eff ( 0 ) are the effective relaxation rates at νcpmg values of 25 Hz and 1000 Hz , respectively . Of those , 16 residues could be fit to a two-site exchange model with a reduced χ2 < 10 and these exhibit two exchange regimes ( a ) fast exchange ( model-2 , kex >> δω ) and ( b ) slow exchange ( model-3 , kex << δω ) . kex and δω are the exchange rate and the chemical shift difference between the major and minor state , respectively . In the slow exchange regime , the parameters , pb ( the population of the minor state ) and δω can be extracted from the fits in addition to kex . In the fast exchange regime , only the value of kex can be extracted from the fits . For estimation of goodness of fits , the target function , χ2 is determined as follows:χ2=∑allvcpmg ( R2 , effobs ( vcpmg ) −R2 , effCalc ( vcpmg ) σ ( R2 , effobs ( vcpmg ) ) ) 2 , where R2 , effCalc is the calculated value of the effective relaxation rate , and σ is the experimental uncertainty in observed R2 , eff which is estimated from two data sets that are repeat values of νcpmg . In this case , repeat data sets were collected at 25 and 200 Hz . The errors reported in kex , pb , and δω values are estimated from 100 Monte Carlo simulations and are reported as the standard deviation of the optimized fit parameter from its 100-element distribution . The value of the dimer–monomer dissociation equilibrium constant , KD , was determined with isothermal titration calorimetry ( ITC ) . ITC was performed on a MicroCal iTC200TM Calorimeter at the Analytical Biopharmacy Core , Univ of Washington . WT-ACD samples were dialyzed into 50 mM sodium phosphate , 100 mM NaCl at either pH 7 . 5 or pH 6 . 5 . 200 µl of 1 mM WT-ACD and 300 µl buffer were placed in the sample and reference cells , respectively . Forty 1 µl injections were performed and the exothermic heat was measured . Data were fitted using MicroCal Origin software to obtain values of KD and change in enthalpic heat ( ΔH ) . All holdase assays were performed in duplicate using a 96-well plate reader ( BioTek ) with PBS solutions at pH 7 . 5 and 250 μL well volumes . DTT-denatured bovine αLactalbumin ( Sigma L6010 ) was used as a model substrate and light scattering at 360 nm was used to monitor protein aggregation in the presence and absence of WT- and mutant HSPB5 at 42°C . 40 µM sHSP ( subunit concentration ) was added to 600 µM αLac . Aggregation of αLac was induced by the addition of DTT to final concentrations of 50 mM . Aggregation of the model substrate Yeast Alcohol Dehydrogenase ( Sigma A8656 ) was achieved with the addition of EDTA and DTT to final concentrations of 5 mM at 37°C . Light scattering by aggregates was monitored in the presence and absence of WT- and mutant HSPB5 . 20 µM sHSP ( subunit concentration ) was added to 100 µM Alcohol Dehydrogenase . The Mw of HSPB5 oligomers was determined by separation using a WTC-050S5 SEC column ( Wyatt Technology Corporation ) with an Akta micro ( GE Healthcare ) and analysis with a DAWN HELEOS II MALS detector equipped with a WyattQELS DLS , and Optilab rEX differential refractive index detector using ASTRA VI software ( Wyatt Technology Corporation ) . The Mw was determined from the Raleigh ratio calculated by measuring the static light scattering and corresponding protein concentration of a selected peak . Bovine serum albumin served as a calibration standard . Prior to SEC-MALS HSPB5 samples were pre-incubated at 37°C for 30 min at 40 μM monomer concentration in 50 mM sodium phosphate , pH 7 . 5 , 100 mM NaCl , and 1 mM DTT . To examine HSPB5 behavior at pH 6 . 5 , 40 μM of protein was allowed to equilibrate in phosphate buffer ( 50 mM sodium phosphate , pH 6 . 5 , 100 mM NaCl , and 1 mM DTT ) at 37°C for at least 60 min . Substrate binding assays were performed under similar conditions established from αLac holdase assays: HSPB5 ( 40 μM ) was pre-incubated with αLac ( 240 μM ) at 37°C in pH 7 . 5 buffer for 10 min . DTT was subsequently added to a final concentration of 50 mM to trigger αLac aggregation . After 50 min , the sample was filtered and injected for SEC-MALS analysis and fractionation . Light scattering data and calculations were performed using the ASTRA software package ( Wyatt Technology Corporation ) . HSPB5 and HSPB5-αLac samples were diluted to approximately 100–200 nM , applied to a thin carbon-coated copper grid and negatively stained using uranyl formate , pH ∼6 . 0 , essentially as described ( Ohi et al . , 2004 ) . Micrograph images were collected on a Tecnai T12 transmission electron microscope ( FEI ) equipped with a LaB6 filament operated at 120 kV . Images were recorded at 50 , 000X magnification with 2 . 2 Å/pixel spacing and a 1 . 0–1 . 5 μM defocus on a 4k × 4k CCD camera ( Gatan ) . Micrograph images were phase-corrected following CTF estimation and HSPB5 particle projections were selected and excised from micrographs using EMAN2 ( Tang et al . , 2007 ) . 2D reference-free alignment and classification was performed using SPIDER ( Frank et al . , 1996 ) to generate projection averages . For size estimation , rotational averages were generated from class averages using SPIDER . The contrast was normalized for all images and the diameter was measured across the rotational average . The averages were grouped according to size , and the total number of single particles for each group was used to obtain the size distribution . Independent data collection and analysis were performed in triplicate to obtain error bars .	Proteins are composed of one or more long chain-like molecules that must fold into complex three-dimensional shapes in order to work properly . Incorrectly folded proteins cannot function and often aggregate into toxic states that are associated with a number of neurological diseases including Alzheimer's , Huntington's , and Parkinson's . Elevated temperatures , increased acidity , and other stressful conditions in the cell can hinder the folding process and may cause existing proteins to unfold and aggregate . However , when cells experience these stresses , certain proteins—known as small heat shock proteins ( or sHSPs for short ) —act as ‘holdase chaperones’ to protect cells from protein misfolding . HSPB5 is one such chaperone that binds to and stabilizes other proteins ( called ‘clients’ ) to prevent their aggregation . The core structure of HSPB5 and other similar chaperone proteins is well known . But , it is not clear how chaperones sense stressful conditions and respond to increase their activity to help stabilize client proteins . Now , Rajagopal et al . have identified a single amino acid in HSPB5 that is sensitive to pH changes . When the environment inside a cell becomes more acidic , this amino acid ( a histidine ) triggers changes in HSPB5's structure that enhance the chaperone's activity . This histidine was then replaced with another amino acid in an attempt to lock HSPB5 into a low-pH state that mimics an active HSPB5 chaperone inside a stressed cell . Further experiments revealed that this mutant HSPB5 is a super-active holdase chaperone , and that it dramatically changes its structure to bind to a client protein in the holdase state . From this , Rajagopal et al . propose a model to explain how cellular stress triggers small changes in HSPB5 that propagate through the chaperone in a response mechanism that increases its activity . Future studies will investigate whether inherited mutations in HSPB5 and other similar chaperones—which are associated with cardiac , muscle , and nerve disorders—exert their effect by disrupting this response mechanism .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	A conserved histidine modulates HSPB5 structure to trigger chaperone activity in response to stress-related acidosis
The level of copy number alteration ( CNA ) , termed CNA burden , in the tumor genome is associated with recurrence of primary prostate cancer . Whether CNA burden is associated with prostate cancer survival or outcomes in other cancers is unknown . We analyzed the CNA landscape of conservatively treated prostate cancer in a biopsy and transurethral resection cohort , reflecting an increasingly common treatment approach . We find that CNA burden is prognostic for cancer-specific death , independent of standard clinical prognosticators . More broadly , we find CNA burden is significantly associated with disease-free and overall survival in primary breast , endometrial , renal clear cell , thyroid , and colorectal cancer in TCGA cohorts . To assess clinical applicability , we validated these findings in an independent pan-cancer cohort of patients whose tumors were sequenced using a clinically-certified next generation sequencing assay ( MSK-IMPACT ) , where prognostic value varied based on cancer type . This prognostic association was affected by incorporating tumor purity in some cohorts . Overall , CNA burden of primary and metastatic tumors is a prognostic factor , potentially modulated by sample purity and measurable by current clinical sequencing . Somatic copy number alterations ( CNAs ) are nearly ubiquitous in cancer ( Zack et al . , 2013; Heitzer et al . , 2016 ) and alter a greater portion of the cancer genome than any other type of somatic genetic alteration ( Heitzer et al . , 2016 ) . Different cancer types vary in their balance of copy number alterations to somatic point mutations , with prostate cancer having relatively high rates of CNA compared to point mutation . Given the prevalence of CNAs in cancer , significant effort has been directed towards identifying specific CNAs associated with cancer clinical characteristics and prognosis as well as the potential driver genes they contain ( Liang et al . , 2016; Wang et al . , 2016; Nibourel et al . , 2017 ) . There are well demonstrated associations between specific CNAs and CNA signatures to cancer state and characteristics ( Visakorpi et al . , 1995; Williams et al . , 2014; Taylor et al . , 2010 ) . CNV patterns or clusters have been associated with high Gleason prostate cancer ( Gleason 8 + compared to Gleason 6–7 [Williams et al . , 2014] ) and recurrent disease ( compared to primary [Visakorpi et al . , 1995; Cancer Genome Atlas Research Network , 2015; Viswanathan et al . , 2018] ) . Nonetheless , most CNAs are large , ( Zack et al . , 2013; Beroukhim et al . , 2010 ) and their associations with cancer outcome may not be well identified by gene-specific approaches . Increasing evidence indicates that large CNAs harbor multiple drivers ( Tschaharganeh et al . , 2016; Liu et al . , 2016 ) , emphasizing the need to study their biological and clinical significance beyond individual gene-focused standpoints . The CNA burden of a tumor is the degree to which a tumor's genome is altered as a percentage of genome length and represents a fundamental measure of genome copy number alteration level . As such , tumor CNA burden , rather than individual CNAs , may be associated with cancer outcomes . While tumor mutational burden ( TMB ) predicts response to immunotherapy across multiple cancer types ( Bergerot et al . , 2018; Goodman et al . , 2017 ) , tumor CNA burden may be prognostic for outcomes such as recurrence and survival . Indeed , we and others have previously found CNA burden and genome-wide CNA patterns to be associated with biochemical recurrence and metastasis in primary prostate cancer , the most common cancer in men , across multiple cohorts ( Taylor et al . , 2010; Hieronymus et al . , 2014; Camacho et al . , 2017 ) . This prognostic significance of tumor CNA burden extends to low and intermediate risk prostate cancer ( Gleason scores of 7 and less ) ( Hieronymus et al . , 2014 ) and has the potential to better stratify risk in patients who are considering conservative treatment approaches such as active surveillance to reduce overtreatment ( Chen et al . , 2016; Tosoian et al . , 2016 ) . In addition to questions about the prognostic potential and overall landscape of CNA in conservatively treated prostate cancer , it is unknown whether CNA burden is prognostic for prostate cancer survival , rather than only recurrence and metastasis . Nor is it known whether the prognostic significance of tumor CNA burden extends to other cancer types . Here we set out to address these questions , as well as whether tumor CNA burden can be prognostic in a clinical practice setting , including ( i ) in cancers treated conservatively rather than through immediate surgery or radiation , ( ii ) in biopsy or resection samples , and ( iii ) using a clinical targeted sequencing that allows rapid and cost-effective measurement of tumor CNA burden . To address these questions , we first examine the genomic CNA landscape of conservatively treated prostate cancer in more than a hundred diagnostic biopsy and resection specimens from a conservatively treated cohort; this cohort consisted of patients with localized prostate who were not treated with surgery or radiation within six months of diagnosis . We demonstrate that tumor CNA burden is associated with cancer-specific death , independent of standard clinical predictors . To explore the prognostic significance of tumor CNA burden more broadly in other cancer types , we find that tumor CNA burden is also associated with disease-free and overall survival in TCGA cohorts of primary breast , endometrial , renal clear cell , thyroid , and colorectal cancer in addition to prostate cancer , with the degree of association varying in some cancer types . We then establish the clinical feasibility of measuring tumor CNA burden using the FDA-cleared MSK-IMPACT clinical next generation sequencing ( NGS ) assay and confirm that tumor CNA burden is associated with overall and disease-specific survival in both primary and metastatic tumors across cancer types . In all , we demonstrate that tumor CNA burden is a prognostic factor associated with cancer recurrence and death in multiple cancer types , including in conservatively treated prostate cancer which would benefit from increased risk stratification . To explore the genomic copy number landscape of conservatively treated prostate cancer , we set out to analyze copy number alteration ( CNA ) in cancer obtained non-surgically through biopsy and transurethral prostate resection ( TURP ) using a widely studied , conservatively treated primary prostate cancer cohort ( Cuzick et al . , 2006 ) . This retrospective Transatlantic Prostate Group 1 ( TAPG1 ) cohort ( n = 1675 ) consists of men below age 76 with clinically localized prostate cancer and prostate-specific antigen ( PSA ) below 100 ng/ml who did not receive surgery or radiation within 6 months of diagnosis ( Cuzick et al . , 2006 ) . This population-based cohort was drawn from six cancer registries in Great Britain , and the majority of the cohort was followed without treatment , while a subset received hormonal therapy . The original diagnostic samples , either biopsy or TURP , were obtained and centrally reviewed to obtain consistent pathological evaluation to the current standards . Drawing from this cohort , we carried out genome-wide CNA analysis by array-based comparative genomic hybridization ( aCGH ) of 107 biopsies or TURP samples from the TAPG1 cohort , as tissue availability is limited for much of the full cohort . The subset of cases used for CNA analysis , which make up our conservative treatment CNA cohort , have similar clinical characteristics to the full TAPG1 cohort , including median diagnosis age , baseline PSA , hormonal treatment , and clinical stage , with the exception of higher Gleason score distribution , likely due to selection for cases with sufficient DNA for analysis ( Supplementary file 2 ) . As expected for a cohort not subject to PSA screening , the patients are older and have higher grade at diagnosis than is typical for contemporary US cohorts . Among the cohort , 47 patients developed metastasis and 43 died of prostate cancer . The median follow-up time for survivors was 10 . 3 years from diagnosis . The copy number alteration landscape of the conservative treatment cohort revealed canonical copy number alterations of prostate cancer , including gain of chromosome 8q and losses on chromosomes 6 p , 8 p , 13q and 16 p , though with lower frequency than seen in prostate cancer cohorts analyzed by our group ( MSKCC cohort ) ( Taylor et al . , 2010 ) and TCGA ( Cancer Genome Atlas Research Network , 2015 ) ( Figure 1a ) . The percentage of the cancer genome showing copy number changes , termed tumor CNA burden ( TCB ) , is similar between the conservative treatment CNA cohort and other cohorts ( Figure 1b ) , with a mean tumor CNA burden of 5 . 7% ( median 1 . 5% , IQR 0 . 05–8 . 5% ) compared to 5 . 2% ( median 3 . 0% , IQR 0 . 04–6 . 9% ) for the 2010 MSKCC primary prostate cancer cohort ( Taylor et al . , 2010 ) and 4 . 0% ( median 0 . 7% , IQR 0 . 08–5 . 1% ) for the 2014 MSKCC primary prostate cancer cohort ( Hieronymus et al . , 2014 ) . The tumor CNA burden of the conservative treatment CNA cohort is , however , somewhat lower than the 8 . 7% average tumor CNA burden of the TCGA prostate cohort ( Cancer Genome Atlas Research Network , 2015 ) ( mean 8 . 7% , median 6 . 2% , IQR 1 . 7–11 . 9% ) . Since tumor CNA burden is associated with prostate cancer recurrence and metastasis in prostatectomy cohorts ( Taylor et al . , 2010; Hieronymus et al . , 2014 ) , we sought to determine whether tumor CNA burden was prognostic for cancer-specific death in biopsies of conservatively treated prostate cancer . In our conservative treatment CNA cohort , we find that tumor CNA burden as a continuous variable is significantly associated with prostate cancer-specific death ( per 5% tumor CNA burden , HR 1 . 49; 95% CI 1 . 30 , 1 . 70; p<0 . 0001; Table 1 ) . Greater tumor CNA burden correlates with an increase in death from disease compared to a lower tumor CNA burden ( Figure 2a ) . The risk of death due to prostate cancer within 5 years of diagnosis increases with tumor CNA burden over the majority of the tumor CNA burden distribution ( Figure 2b ) . For example , the 5 year risk of death due to prostate cancer would be 13% for patients with a 2% tumor CNA burden and 28% for patients with a 10% tumor CNA burden ( Figure 2b ) . Tumor CNA burden may therefore serve as a prognostic factor for cancer-specific death in patients who undergo increasingly common conservative treatment approaches . We next asked whether tumor CNA burden was associated with outcome after adjusting for established prognostic variables , including Gleason sum score and the UCSF Cancer of the Prostate Risk Assessment ( CAPRA ) score ( Cooperberg et al . , 2005; Brajtbord et al . , 2017 ) which combines PSA , Gleason score , percentage positive biopsy cores , clinical stage , and age ( Figure 2c ) . Tumor CNA burden is significantly associated with cancer-specific death even after adjusting for biopsy Gleason score ( per 5% tumor CNA burden , HR 1 . 44; 95% CI 1 . 24 , 1 . 67; p<0 . 0001 ) or CAPRA score ( per 5% tumor CNA burden , HR 1 . 44; 95% CI 1 . 24 , 1 . 68; p<0 . 0001 ) ( Table 1 , Figure 2c ) . The addition of tumor CNA burden into the model with the CAPRA score increased Harrell’s concordance index from 0 . 756 to 0 . 805 for cancer-specific survival in our cohort of men with conservatively treated prostate cancer . Large , clinically annotated cancer genomic efforts such as TCGA now provide an opportunity to examine whether CNA burden is prognostic for primary cancer outcomes across many cancer types . In the TCGA primary prostate cancer cohort ( Cancer Genome Atlas Research Network , 2015 ) , tumor CNA burden is significantly associated with biochemical recurrence individually ( p<0 . 0001; per 5% tumor CNA burden , HR = 1 . 27; 95% CI , 1 . 13 , 1 . 42 ) and after adjustment for Gleason score and mutation burden ( p=0 . 015; per 5% tumor CNA burden , HR = 1 . 18; 95% CI , 1 . 03 , 1 . 35 ) , validating our findings from other prostate cancer cohorts ( Figure 2c , Figure 2—figure supplement 1 , Table 2 ) . There were insufficient deaths in this cohort to analyze survival . CNA burden was still significantly associated with biochemical recurrence after adjusting for tumor sample purity determined by ABSOLUTE ( p<0 . 003; per 5% CNA burden , HR = 1 . 22; 95% CI , 1 . 07 , 1 . 40; Table 2 ) . Since tumor CNA burden could potentially reflect simply the prognostic significance of aneuploidy as determined by cytometric DNA index in various cancers ( Walther et al . , 2008; Danielsen et al . , 2016 ) , we examined the tumor CNA burden in a multivariable model together with ploidy . Ploidy , generated by CLONET and previously published for this cohort , estimates the average DNA index of the tumor cells ( Carter et al . , 2012; Prandi et al . , 2014 ) . Tumor CNA burden was associated with recurrence independent of tumor ploidy ( p=0 . 002; per 5% tumor CNA burden , HR = 1 . 32; 95% CI 1 . 11 , 1 . 56; Table 2 ) . Moreover , for a multivariable model that includes tumor CNA burden , Gleason grade , and mutation burden , the Harrell’s C-index is 0 . 691 . In contrast , the C-index for a model including ploidy instead of tumor CNA burden is only 0 . 606 , indicating that a model with clinical factors and ploidy does not perform as well as a model with the same clinical factors and tumor CNA burden . The prognostic significance of tumor CNA burden in prostate cancer led us to ask whether tumor CNA burden is prognostic in other cancer types . Towards this end , we examined published TCGA cohorts for multiple cancer types with available disease-free survival and overall survival data , including breast ( Ciriello et al . , 2015 ) , endometrial ( Cancer Genome Atlas Research Network et al . , 2013 ) , renal clear cell ( Cancer Genome Atlas Research Network , 2013 ) , thyroid ( Cancer Genome Atlas Research Network , 2014 ) , and colorectal ( Cancer Genome Atlas Network , 2012 ) cancers . We found that tumor CNA burden is associated with recurrence ( disease-free survival ) in these cancer types ( Figure 2c , Figure 2—figure supplement 2 , Table 2 ) . This association between tumor CNA burden and lower disease-free survival was independent of disease stage in all cancer types except colorectal cancer , where the association was independent of tumor stage but not disease stage ( Table 2 ) . In addition to lower disease-free survival , higher tumor CNA burden was also significantly associated with lower overall survival in breast , endometrial , thyroid , and colorectal cancer ( Table 2 ) . This association with overall survival was independent of disease stage in breast and endometrial cancer and independent of tumor stage in colorectal cancer ( Table 2 ) . There were insufficient cases of thyroid cancer with stage data for this analysis . In summary , tumor CNA burden is prognostic for recurrence and/or overall survival in multiple cancer types beyond prostate cancer , including breast , endometrial , colorectal , renal clear cell , and thyroid cancer . We next wanted to determine whether CNA burden’s prognostic associations could be observed using panel-based targeted sequencing assays that are increasingly entering clinical use , in contrast to CGH array-based determination of tumor CNA burden . The Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets ( MSK-IMPACT ) assay is a clinical laboratory improvement amendments ( CLIA ) -certified sequencing-based assay ( Cheng et al . , 2015 ) of several hundred cancer genes and 1042 common single nucleotide polymorphisms ( SNPs ) that has been used to profile 504 prostate tumors ( Abida et al . , 2017 ) and more than ten thousand tumors across other cancer types ( Zehir et al . , 2017 ) . The IMPACT assay identifies both somatic point mutations and copy number alterations in the genes included in the panel . Overall copy number burden is calculated across the whole genome ( Figure 1a ) using segmentation derived from a combination of the profiled SNPs to provide low resolution copy number data and the genes sequenced in the panel ( Cheng et al . , 2015; Abida et al . , 2017; Zehir et al . , 2017 ) . To address the possibility that CNA burden from the IMPACT panel might differ from that derived from more comprehensive sequencing , we directly compared CNA burden calculations from 1005 tumors that were profiled using both IMPACT and whole exome sequencing . CNA burden determined by the two methods were highly correlated ( p-value<0 . 0001 , rho = 0 . 88 , n = 1005 ) , indicating that CNA burden is not significantly affected by the reduced resolution in moving from whole exome to targeted panel sequencing ( Figure 2—figure supplement 3 ) . We find that tumor CNA burden assayed by targeted clinical sequencing is significantly associated with overall survival in primary prostate tumors ( per 5% tumor CNA burden , HR = 1 . 17; 95% CI , 1 . 04 , 1 . 3; p=0 . 007; Table 3 , Figure 2c , Figure 2—figure supplement 4 ) in the IMPACT prostate cohort ( Abida et al . , 2017 ) . As clinical sequencing assays such as MSK-IMPACT are principally used in the metastatic patient population , the IMPACT cohorts also provide an opportunity to investigate the prognostic significance of tumor CNA burden in late stage disease . We find that tumor CNA burden of metastatic prostate tumors assayed by clinical sequencing is also significantly associated with survival ( per 5% tumor CNA burden , HR = 1 . 07; 95% CI , 1 . 01 , 1 . 14; p=0 . 020; Table 3 , Figure 2c , Figure 2—figure supplement 4 ) . Since clinical sequencing assays also provide point mutation information for several hundred cancer genes , we asked if tumor CNA burden is prognostic after adjusting for known prostate cancer driver alterations . In separate multivariable regression models adjusting for TP53 , RB1 , or PTEN loss and/or mutation , tumor CNA burden is still associated with overall survival independent of these alterations in primary prostate tumors ( Table 3 ) . In metastatic tumors , these specific gene mutations do not reach prognostic significance when combined with tumor CNA burden ( Table 3 ) . Notably , tumor CNA burden remains significant in metastatic tumors after adjusting for overall tumor mutation burden ( per 5% tumor CNA burden , HR = 1 . 08; 95% CI = 1 . 02 , 1 . 15; p=0 . 011; Table 3 ) . As targeted clinical sequencing is applied to a wide range of cancer types , we expanded our survival analysis to a pan-cancer cohort , consisting of 6610 primary tumors and 4864 metastatic tumors across 53 cancer types assayed by MSK-IMPACT sequencing panel ( Materials and methods and Supplementary file 2 ) . We find that tumor CNA burden is prognostic for overall survival pan-cancer in primary tumors ( p<0 . 0001; per 5% tumor CNA burden , HR = 1 . 04; 95% CI , 1 . 02 , 1 . 05 ) and in metastatic tumors ( p=0 . 005; per 5% tumor CNA burden , HR = 1 . 02; 95% CI , 1 . 01 , 1 . 03 ) in a univariate analysis of these pan-cancer cohorts ( Table 3 , Figure 2c ) . Tumor CNA burden is also prognostic for cancer-specific death in the metastatic tumor cohort ( p=0 . 026; per 5% tumor CNA burden , HR = 1 . 05; 95% CI , 1 . 01 , 1 . 10 ) . Adjustment for sample tumor purity determined by FACETS ( Shen and Seshan , 2016 ) found that CNA burden was still significantly associated with overall survival in primary tumors in the pan-cancer analysis and approached significance for metastatic tumors ( p=0 . 06; Supplementary file 3 ) , though purity-adjusted CNA burden was no longer significantly associated with overall survival in the prostate tumor subsets ( Supplementary file 3 ) . Adjustment for sample tumor purity determined by FACETS ( Shen and Seshan , 2016 ) found that CNA burden was still significantly associated with overall survival in primary tumors in the pan-cancer analysis approached significance for metastatic tumors ( p=0 . 06; Supplementary file 3 ) , though purity-adjusted CNA burden was no longer significantly associated with overall survival in the prostate tumor subsets ( Supplementary file 3 ) . Tumor mutation burden ( TMB ) , in contrast to tumor CNA burden , was not associated with overall survival or cancer-specific survival ( p=0 . 4 and p>0 . 9 , respectively; Table 3 ) . Since the pan-cancer prognostic significance of tumor CNA burden is likely to be influenced by the distribution of cancer types within the IMPACT cohorts , we stratified the primary and metastatic pan-cancer IMPACT cohorts by their ten most prevalent cancer types , which make up nearly three-quarters of the cohort ( Supplementary file 2 ) . A multivariable Cox model was used for each cancer type to adjust for mutation burden and extract the effect size , which was then entered into a meta-analysis . After stratifying by cancer type , the CNA burden of primary tumors measured by the MSK-IMPACT assay is still significantly associated with death ( overall fixed effects HR = 1 . 04; 95% CI 1 . 02 , 1 . 05; test of effects size p<0 . 0001; Table 3; Figure 2c ) . Similarly , metastatic tumor CNA burden was associated with death ( overall fixed effects HR = 1 . 02; 95% CI 1 . 01 , 1 . 04; test of effects size p=0 . 005; Table 3; Figure 2c ) . A closer look at the pan-cancer analysis reveals statistically significant heterogeneity in the relationship between tumor CNA burden and survival across tumor types ( p=0 . 003 and p=0 . 024 in primary and metastatic tumor cohorts respectively , Figure 2—figure supplement 4 ) . In primary tumors , heterogeneity appears to be driven by colorectal and pancreatic cancers , where an inverse association between tumor CNA burden and death is seen ( Figure 2—figure supplement 5a ) . After excluding colorectal and pancreatic cancers , heterogeneity is no longer statistically significant ( overall fixed effects HR = 1 . 05; 95% CI 1 . 03 , 1 . 07; test of effects size p<0 . 0001; test for heterogeneity p=0 . 3; Figure 2—figure supplement 5a ) . In metastatic tumors , two outlying cancer types drive this heterogeneity: pancreatic cancer , which shows the same inverse association of tumor CNA burden with death as in primary pancreatic tumors , and prostate , which shows the opposite effect ( Figure 2—figure supplement 5b ) . Exclusion of either cancer type eliminates the significant heterogeneity in effects size , such that higher tumor CNA burden is associated with increased death in the remaining homogenous set of cancer types ( overall fixed effects HR = 1 . 03; 95% CI 1 . 01 , 1 . 04; test of effects size p=0 . 002; test for heterogeneity p=0 . 8 , Figure 2—figure supplement 5b ) . These results indicate that tumor CNA burden can have differing levels of prognostic effect depending on the cancer type , while a core set of cancer types show a statistically similar association between overall survival and tumor CNA burden assayed by targeted sequencing . More generally , we find that tumor CNA burden determined by a clinically-certified sequencing panel is associated with overall and disease-specific mortality in a large multi-cancer population , including in patients with metastatic cancer where clinical sequencing is increasingly applied . Many specific genes altered by CNA have been associated with cancer outcomes ( Liang et al . , 2016; Wang et al . , 2016; Nibourel et al . , 2017 ) , however the relationship between outcome and the overall level of CNA harbored by a tumor is less well studied . Here we expanded on our previous work showing that tumor CNA burden is associated with recurrence in surgically treated primary prostate cancer ( Taylor et al . , 2010; Hieronymus et al . , 2014 ) by showing a significant association with death from prostate cancer , including in conservatively treated patients where the tumor CNA burden measurement was made from biopsies . Importantly , this association remains significant even after adjusting for Gleason score or for CAPRA score , demonstrating that CNA burden is independent of previously identified associations with these measures of cancer pathology or disease state . Thus , tumor CNA burden assessment from prostate biopsies could have a role in deciding between surgery and surveillance for men at the low end of intermediate risk . Conversely , it may also have role in men at high risk where multimodal treatment may be needed . An unanticipated outcome of our analysis beyond prostate cancer is the prognostic role of tumor CNA burden across a range of tumor types . The pan-cancer tumor CNA burden association is significant but also heterogeneous depending on cancer type . Recent work has similarly found that the presence of any CNA , regardless of gene identity , is associated with overall and event-free survival in pediatric AML ( Vujkovic et al . , 2017 ) and that the percentage of subclonal CNAs but not subclonal somatic point mutations is associated with overall survival in non-small cell lung cancer ( Jamal-Hanjani et al . , 2017 ) . Moreover , survival time was associated with a CNA signature derived from supervised analyses in prostate cancer and extended to breast and lung cancer ( Pearlman et al . , 2018 ) . Prognostic individual CNAs or sets of CNAs , as opposed to the broader measure of genome-wide CNA level examined here may be specific to individual cancer types , whereas we have demonstrated the prognostic potential of a generalized measure of overall copy number dysregulation . Further work will be needed to address the trade-offs between generalizability of CNA burden and discriminatory power . In addition , it will be important to investigate whether the prognostic associations of CNA burden from the pan-cancer analysis are independent of known cancer- or subtype-specific prognostic factors , such as ER receptor status in breast cancer , ultra- and hypermutated ( POLE and MSI+ ) status in endometrial cancer and MSI-positive/CIN-negative status in colorectal cancer ( Walther et al . , 2008 ) . We find it notable that tumor CNA burden assessment using a targeted sequencing can serve as a surrogate for tumor CNA burden calculated using more comprehensive genomic assays such as array CGH . With the proliferation of different clinical sequencing panels for mutation detection , it will be of interest to see how much resolution , depth , and coverage can be reduced and still retain the prognostic association of CNA burden; future work in this area will also need to incorporate the predictive clinical utility of the point mutation data to address the multimodal uses of clinical sequencing assays . Another important variable is tumor purity . The prognostic significance of CNA burden can be affected by sample tumor purity , with purity being independently associated with outcome . The effect of purity on the association between CNA burden and outcome appears complex and may be influenced by the analysis platform , cancer type , and outcome type . For example , pan-cancer CNA burden from clinical sequencing panel remained prognostic for survival after purity adjustment in primary tumors and was just below significance for metastatic tumors , though the CNA burden of the prostate tumor subset assayed by IMPACT sequencing panel did not . However , the CNA burden of prostate tumors assayed by SNP array showed continued association with recurrence after adjustment for purity . Tumor purity alone was also independently associated with survival , revealing a complex interaction between these tumor features that will need further exploration . As targeted sequencing moves from tumor samples to liquid biopsy in the form of cell-free DNA ( cfDNA ) ( Heitzer et al . , 2016; Xia et al . , 2015; Hyman et al . , 2017 ) , it will be important to determine whether tumor CNA burden determined by analysis of cfDNA has similar prognostic utility as that determined by direct analysis of tumor DNA . There is already some evidence this may be possible , as the summed CNA level of the most highly copy number altered genes assayed from whole genome sequencing of cfDNA in twenty metastatic prostate cancer patients correlated with overall survival ( Xia et al . , 2015 ) . As sequencing costs continue to drop and computational power improves , it would be interesting to investigate low pass whole genome sequencing as an alternative approach for determining tumor CNA burden that provides complete genome coverage . Another interesting feature of the association of tumor CNA burden with outcome demonstrated here is that it has prognostic significance independent of tumor mutation burden ( TMB ) . This is consistent with recent work in glioblastoma , breast , lung , and ovarian cancer showing that CNA-derived signatures have more prognostic power than somatic point mutation-based signatures , as measured by concordance index ( Gómez-Rueda et al . , 2015 ) . Thus , tumor CNA burden could complement clinical analyses of actionable driver mutations using a single panel-based sequencing assay . The prognostic significance of tumor CNA burden raises intriguing questions regarding the underlying biology . Tumor CNA burden may be a simple measure that correlates with the extent of oncogenic driver alterations . Yet , we show that tumor CNA burden retains its prognostic significance after adjustment for a number of known oncogenic alterations in primary prostate cancer , including PTEN loss associated with increased tumor CNA burden ( Castro et al . , 2015; Williams et al . , 2014 ) . In metastatic tumors , combining tumor CNA burden with TP53 or RB1 loss in multivariable analyses renders both slightly below conventional significance thresholds , raising the possibility of biological interplay between these genes ( particularly TP53 ) and subsequent copy number alteration that develops during tumor evolution . Further , the prognostic associations of tumor CNA burden are independent of tumor ploidy , which suggests that tumor CNA burden may not simply reflect aneuploidy , defined as abnormal DNA content ( Danielsen et al . , 2016 ) . It is also possible that tumor CNA burden captures prognostic information about currently unidentified driver alterations and/or the rate of ongoing CNA within a tumor that may generate additional driver alterations , including those reflecting intratumoral heterogeneity , thereby affecting outcome . Ongoing work by others has begun to develop genomic methods for identifying mechanisms of somatic CNA ( Wala et al . , 2017 ) ; and identify prognostic CNA signatures and the mechanisms underlying the component CNA ( Macintyre et al . , 2018 ) . Ultimately , the biology underlying the significant association of tumor CNA burden with multiple cancer outcomes will be a fruitful area for future investigation . Of the TAPG1 cohort ( Cuzick et al . , 2006 ) , FFPE prostate tumor tissue from 180 patients was macrodissected from slides . DNA was isolated ( Agilent FFPE DNA isolation for aCGH protocol ) and quantified by picogreen-based quantification . 107 cases yielded greater than 500 ug DNA and were analyzed by Agilent 180K human CGH arrays ( Agilent , 4 × 180K G4449A arrays , per manufacturer's instructions ) . Copy number data from patients in the TAPG copy number cohort were quantified , normalized , segmented , and analyzed with RAE , as previously described ( Taylor et al . , 2010; Hieronymus et al . , 2014 ) . The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO under accession number GSE103665 ( Gene Expression Omnibus , RRID:SCR_007303 ) . Tumor CNA burden ( tumor CNA burden ) was analyzed as percent CNA burden , defined as the length of the genome altered by copy number alteration multiplied by 100 . For regression analyses , tumor CNA burden was scaled as per five percent so that the estimates of our hazard ratios were more interpretable . All statistical analyses were performed using Stata 13 ( RRID:SCR_012763 , StataCorp , College Station , TX ) . For Cox regression analyses , the primary aim was to determine whether tumor CNA burden is associated with cancer specific survival ( CSS ) . First , we assessed whether there was an association between tumor CNA burden and CSS by utilizing a univariate Cox model , censoring patients who did not die at the date of their last follow-up and patients who died of other causes at their death date . Secondly , in order to assess whether there is information from tumor CNA burden over and above biopsy Gleason score , we utilized a multivariable Cox model , adjusting for biopsy Gleason sum categorized as ≤6 , 7 , and ≥8 . Finally , to assess whether there is an association between tumor CNA burden and CSS after accounting for the preoperative predictors of CSS , we utilized a multivariable Cox model , adjusting for the UCSF-CAPRA score , a preoperative risk score calculated by incorporating the patient’s age at diagnosis , PSA at diagnosis , primary and secondary Gleason score at biopsy and clinical tumor stage . As percent of positive biopsy cores was not available for the cohort , a modified CAPRA score was utilized not incorporating this information . Among our cohort of 107 patients , 47 patients were missing clinical tumor stage; multiple imputation was used to impute the missing values . Statistical analyses were performed utilizing the measured and imputed values combined across 10 imputations using Rubin’s method . Furthermore , to evaluate the discriminative accuracy of the model including tumor CNA burden , we calculated bootstrap optimism-corrected Harrell’s C-index . It should be noted that the discrimination of the CAPRA score is lower in the TAPG1 conservative treatment CNA cohort than seen in some other prostate cancer cohorts , and this may impact the degree to which tumor CNA burden increases the concordance index . All data used for these analyses are available in Supplementary file 4 . For illustrative purposes , we utilized competing risk methods to estimate the probability of death from prostate cancer in the setting of death from other causes . Cumulative incidence was shown for patients who died from prostate cancer , or died from other causes , stratified on tumor CNA burden in relation to the median tumor CNA burden among the cohort , using the stcompet command in Stata . For analysis of the prostate cancer MSK-IMPACT cohort ( Abida et al . , 2017 ) , the published cases were analyzed by Cox regression for association between overall survival and tumor CNA burden ( Supplementary file 5 and 6 ) . The IMPACT cases were separated into groups consisting of primary tumors or metastatic tumors , including loco-regional , non-resistant to treatment , and treatment resistant , though primary tumor samples include cases sampled after metastatic spread . Among our primary and metastatic IMPACT prostate cancer cohorts , we excluded men with unknown overall survival status and unknown time until overall survival status , leaving us with a final cohort of 261 and 216 men , respectively . Among these two groups of patients , we assessed the association between tumor CNA burden and overall survival using a univariate Cox model . Multivariable Cox models were then used to determine whether the association between tumor CNA burden and overall survival remained after accounting for purity determined by FACETS ( Shen and Seshan , 2016 ) , the overall point mutation burden , or specific somatic gene alterations ( shallow or deep copy number loss or mutation ) occurring in prostate cancer ( BRCA1 , BRCA2 , ATM , TP53 , RB1 , and PTEN ) , using separate models for each alteration . As the overall point mutation burden was not available for all patients , 34 patients with primary prostate cancer and 11 patients with metastatic prostate cancer were excluded from this portion of the analysis in their respective cohorts . For analysis of our pan-cancer IMPACT cohort ( MSK-IMPACT cohort ( Zehir et al . , 2017 ) and additionally accrued IMPACT samples ) , outcome data at time of analysis , mutation burden , and fraction genome altered data used were derived and available in updated form the cBio Portal ( RRID:SCR_002877 , http://www . cbioportal . org/study ? id=msk_impact_2017 , samples and annotation used at time of analysis available as Supplementary file 7 and 8 ) . A cohort of 7305 primary tumor cases across 53 different cancer types and a cohort 5907 metastatic tumor cases , across 47 different cancer types , were identified . Within the primary and metastatic disease cohorts , we excluded patients with unknown tumor CNA burden , overall survival status , unreported follow-up time , death or censoring immediately after treatment , unknown cancer type , and unknown mutation burden . The final cohort used here therefore included 6610 and 4864 patients , respectively . Within both of these cohorts , univariate Cox models were used to determine whether CNA or mutation burden is associated with overall survival . Reported follow-up time was used . As it is probable that the association between tumor CNA burden and survival likely varies based on the particular cancer type , we focused on patients with the ten most prevalent cancer types in both of the respective cohorts ( Supplementary file 2 , 5198 and 3886 patients with primary and metastatic disease respectively ) and proceeded with a meta-analysis in order to stratify by cancer type . In particular , we utilized a multivariable Cox model , adjusting for mutation burden for each cancer type and extracted the effect size . The effect size for each cancer type was then entered into a meta-analysis using the metan command in Stata . Both fixed effects and random effects estimates were calculated . Fixed effects weights were calculated using inverse-variance weighting , metan weights were calculated using the DerSimonian and Laird method . For analyses of TCGA cohorts , the following published cohorts were filtered for only primary , non-neoadjuvantly treated cases and analyzed: TCGA prostate adenocarcinoma ( 2015 ) ( Cancer Genome Atlas Research Network , 2015 ) , breast carcinoma ( Ciriello et al . , 2015 ) , uterine endometriod cancer ( Cancer Genome Atlas Research Network et al . , 2013 ) , renal clear cell carcinoma ( Cancer Genome Atlas Research Network , 2013 ) , papillary thyroid carcinoma ( Cancer Genome Atlas Research Network , 2014 ) , and colorectal adenocarcinoma ( Cancer Genome Atlas Network , 2012 ) . The number of cases and exclusions based on unavailable data are detailed in Supplementary file 9 . Cox regression was used to test the association of tumor CNA burden as a continuous variable with ( i ) cancer free status and ( ii ) overall survival in univariate models and in multivariable models with disease stage . For the TCGA colorectal cancer cohort , tumor stage was also used . For the TCGA prostate adenocarcinoma cohort , multivariable Cox regression models that included Gleason score , mutation count , ploidy , and/or ABSOLUTE purity ( Carter et al . , 2012 ) originally reported with this cohort were also used . Analyses including purity exclude 37 patients without absolute tumor purity measured , resulting in analysis with 243 men , 29 of whom had BCR , and a median followup time for survivors of 20 . 1 ( 7 . 0 , 37 . 9 ) months . Data access . The conservative treatment TAPG copy number cohort array data was deposited in NCBI GEO ( Gene Expression Omnibus , under accession number GSE103665 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE103665 ) .	Cancer cells carry different types of mutations that are associated with the cell starting to multiply uncontrollably . Certain changes only affect one or a few letters of the genetic code . Others , known as copy number alterations , or CNA , involve larger portions of the genome that can either be lost ( deletions ) or duplicated ( amplifications ) . Tumors in different patients carry variable amounts of these deletions or amplifications , which together are known as the CNA burden . New technologies allow scientists to scan the genomes of tumors and examine the type of mutations present in each patient . The results can help to decide on the best course of action . For example , in prostate cancer , patients whose tumors have a high CNA burden are at greater risk of relapse after treatment . However , it has been unclear whether these people also have lower survival rates , and if CNA burden can predict outcome of other types of cancers . Hieronymus et al . conducted genetic analyses on over a hundred samples from prostate cancer patients who were not treated with surgery or radiation . The results showed that a higher CNA burden in the tumors is correlated with more deaths due to the disease . The findings in prostate cancer were also true across different types of cancers . These conclusions also emerged when Hieronymus et al . then looked at genomic data obtained from patients with various cancers using a different DNA sequencing test , which is certified for clinical use . This demonstrates that CNA burden could be a useful marker in clinical settings to help assess risk in cancer patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "cancer", "biology" ]	2018	Tumor copy number alteration burden is a pan-cancer prognostic factor associated with recurrence and death
Fidaxomicin ( Fdx ) is an antimicrobial RNA polymerase ( RNAP ) inhibitor highly effective against Mycobacterium tuberculosis RNAP in vitro , but clinical use of Fdx is limited to treating Clostridium difficile intestinal infections due to poor absorption . To identify the structural determinants of Fdx binding to RNAP , we determined the 3 . 4 Å cryo-electron microscopy structure of a complete M . tuberculosis RNAP holoenzyme in complex with Fdx . We find that the actinobacteria general transcription factor RbpA contacts fidaxomycin , explaining its strong effect on M . tuberculosis . Additional structures define conformational states of M . tuberculosis RNAP between the free apo-holoenzyme and the promoter-engaged open complex ready for transcription . The results establish that Fdx acts like a doorstop to jam the enzyme in an open state , preventing the motions necessary to secure promoter DNA in the active site . Our results provide a structural platform to guide development of anti-tuberculosis antimicrobials based on the Fdx binding pocket . The bacterial RNA polymerase ( RNAP ) is a proven target for antibiotics . The rifamycin ( Rif ) class of antibiotics , which inhibit RNAP function , is a lynchpin of modern tuberculosis ( TB ) treatment ( Chakraborty and Rhee , 2015 ) . TB , caused by the infectious agent Mycobacterium tuberculosis ( Mtb ) , is responsible for almost 2 million deaths a year . It is estimated that one third of the world is infected . Mortality from TB is increasing , partly due to the emergence of strains resistant to Rifs ( RifR ) ( Zumla et al . , 2015 ) . Hence , additional antibiotics against RifR Mtb are needed . Fidaxomicin ( Fdx; also known as Dificimicin , lipiarmycin , OPT-80 , PAR-101 , or tiacumicin ) , an antimicrobial in clinical use against Clostridium difficile ( Cdf ) infection ( Venugopal and Johnson , 2012 ) , functions by inhibiting the bacterial RNAP ( Talpaert et al . , 1975 ) . Fdx targets the RNAP 'switch region' , a determinant for RNAP inhibition that is distinct from the Rif binding pocket ( Srivastava et al . , 2011 ) , and Fdx does not exhibit cross-resistance with Rif ( Gualtieri et al . , 2009 , 2006; Kurabachew et al . , 2008; O'Neill et al . , 2000 ) . The switch region sits at the base of the mobile RNAP clamp domain and , like a hinge , controls motions of the clamp crucial for DNA loading into the RNAP active-site cleft and maintaining the melted DNA in the channel ( Chakraborty et al . , 2012; Feklistov et al . , 2017 ) . Fdx is a narrow spectrum antibiotic that inhibits Gram-positive anaerobes and mycobacteria ( including Mtb ) much more potently than Gram-negative bacteria ( Kurabachew et al . , 2008; Srivastava et al . , 2011 ) , but the clinical use of Fdx is limited to intestinal infections due to poor bioavailability ( Venugopal and Johnson , 2012 ) . Addressing this limitation requires understanding the structural and mechanistic basis for Fdx inhibition , which is heretofore unknown . Here , we used single-particle cryo-electron microscopy ( cryo-EM ) to determine structures of Mtb transcription initiation complexes in three distinct conformational states , including a complex with Fdx at an overall resolution of 3 . 4 Å . The results define the molecular interactions of Mtb RNAP with Fdx as well as the mechanistic basis of inhibition , and establish that RbpA , an Actinobacteria-specific general transcription factor ( GTF ) , is crucial to the sensitivity of Mtb to Fdx . Fdx has potent inhibitory activity against multi-drug-resistant Mtb cells and the in vivo target is the RNAP ( Kurabachew et al . , 2008 ) . To our knowledge , the in vitro activity of Fdx against mycobacterial RNAPs has not been reported . RbpA , essential in Mtb , is a component of transcription initiation complexes ( TICs ) that tightly binds the primary promoter specificity σA subunit of the RNAP holoenzyme ( holo ) ( Bortoluzzi et al . , 2013; Forti et al . , 2011; Hubin et al . , 2017a , 2015; Tabib-Salazar et al . , 2013 ) . We therefore compared Fdx inhibition of mycobacterial RNAPs containing core RNAP combined with σA ( σA-holo ) and RbpA with inhibition of Escherichia coli ( Eco ) σ70-holo using a quantitative abortive initiation assay ( Davis et al . , 2015 ) . Fdx inhibited Mtb and M . smegmatis ( Msm ) transcription at sub-μM concentrations , whereas inhibition of an Mtb TIC containing Fdx-resistant ( FdxR ) RNAP ( βQ1054H ) ( Kurabachew et al . , 2008 ) required a nearly two orders of magnitude higher concentration of Fdx . Eco RNAP was inhibited even less effectively by another order of magnitude ( Figure 1A , Figure 1—figure supplement 1A ) . We used single-particle cryo-EM to examine the complex of Mtb RbpA/σA-holo with and without Fdx ( Figure 1B ) . Preliminary analyses revealed that the particles were prone to oligomerization , which was reduced upon addition of an upstream-fork ( us-fork ) junction promoter DNA fragment ( Figure 1C ) . We sorted nearly 600 , 000 cryo-EM images of individual particles into two distinct classes , each arising from approximately half of the particles ( Figure 1—figure supplement 2 ) . The first class comprised Mtb RbpA/σA-holo with one us-fork promoter fragment and bound to Fdx . The cryo-EM density map was computed to a nominal resolution of 3 . 4 Å ( Figure 1D , Figure 1—figure supplement 3 , Supplementary file 1 ) . The us-fork promoter fragment was bound outside the RNAP active site cleft , as expected , with the −35 and −10 promoter elements engaged with the σA4 and σA2 domains , respectively ( Figure 1D ) . Local resolution calculations ( Cardone et al . , 2013 ) indicated that the central core of the structure , including the Fdx binding determinant and the bound Fdx , was determined to 2 . 9–3 . 4 Å resolution ( Figure 1E ) . The second class comprised Mtb RbpA/σA-holo bound to two us-fork promoter fragments but without Fdx to a nominal resolution of 3 . 3 Å ( Figure 2A , Figure 2—figure supplement 1 , Supplementary file 1 ) . One us-fork promoter fragment bound upstream from the RNAP active site cleft as in the previous class , but a second us-fork promoter fragment bound the RNAP downstream duplex DNA binding channel , with the 5-nucleotide 3'-overhang ( Figure 1C ) engaged with the RNAP active site ( as the template strand ) like previously characterized 3'-tailed templates ( Gnatt et al . , 2001; Kadesch and Chamberlin , 1982 ) . Local resolution calculations ( Cardone et al . , 2013 ) indicated that the central core of the structure was determined to between 2 . 8–3 . 2 Å resolution ( Figure 2B ) . The overall conformation of this protein complex and its engagement with the upstream and downstream DNA fragments was very similar to the crystal structure of a full Msm open promoter complex ( RPo ) ( Hubin et al . , 2017b ) with one exception ( see below ) . We will therefore call this complex an Mtb RbpA/RPo mimic . RbpA comprises four structural elements , the N-terminal tail ( NTT ) , the core domain ( CD ) , the basic linker , and the sigma interacting domain ( SID ) ( Bortoluzzi et al . , 2013; Hubin et al . , 2017a; Tabib-Salazar et al . , 2013 ) . Our previous crystal structures of Msm TICs containing RbpA showed that the RbpASID interacts with the σA2 domain , the RbpABL establishes contacts with the promoter DNA phosphate backbone just upstream of the −10 element , and the RbpACD interacts with the RNAP β' Zinc-Binding-Domain ( ZBD ) ( Hubin et al . , 2017a , 2017b ) . Density for the RbpANTT ( RbpA residues 1–25 ) was never observed in the crystal structures and was presumed to be disordered . In striking contrast to the crystal structures , both cryo-EM structures reveal density for the RbpANTT , which unexpectedly threads into the RNAP active site cleft between the ZBD and σA4 domains and snakes through a narrow channel towards the RNAP active site Mg2+ ( Figure 3 ) . On its path , conserved residues of the RbpANTT interact with conserved residues of the σ-finger ( σ3 . 2-linker ) on one wall of the channel , and with conserved residues of the ZBD and β'lid on the other wall ( Figure 3C ) . The most N-terminal RbpA residues visible in the cryo-EM structures ( A2 in the Fdx complex , R4 in the RPo ) sit near the tip of the σ-finger where it makes its closest approach to the RNAP active site , too far ( 25 Å ) to play a direct role in RNAP catalytic activity or substrate binding . The σ-finger plays an indirect role in transcription initiation , stimulating de novo phosphodiester bond formation by helping to position the t-strand DNA ( Kulbachinskiy and Mustaev , 2006; Zhang et al . , 2012 ) . The σ-finger is also a major determinant of abortive initiation , playing a direct role in initiation and promoter escape by physically blocking the path of the elongating RNA transcript before σ release ( Cashel et al . , 2003; Murakami et al . , 2002 ) . The intimate association of the RbpANTT with the σ-finger ( Figure 3C ) suggests that the RbpANTT also plays a role in these processes of Mtb RNAP initiation . This is consistent with our findings that the RbpANTT does not strongly affect RPo formation but plays a significant role in in vivo gene expression in Msm ( Hubin et al . , 2017a ) . This location of the RbpANTT explains the high Fdx sensitivity of Mtb RNAP ( see below ) . The reconstruction from the Fdx-bound class ( Figure 1D ) reveals unambiguous density for Fdx ( Figure 4A ) and defines Fdx-interacting residues from four protein components of the complex , β , β' , σA , and RbpA , including six water molecules , four of which mediate Fdx/RNAP interactions ( Figure 4A , B ) . Fdx binding to the TIC buries a large accessible surface area of 4 , 800 Å2 ( β , 2 , 100 Å2; β' , 2 , 000 Å2; σA , 300 Å2; RbpA , 330 Å2 ) . Fdx forms direct hydrogen bonds with nine residues ( βQ1054 , βD1094 , βT1096 , βK1101 , β'R84 , β'K86 , β'R89 , β'E323 , and β'R412 ) and water-mediated interactions with four ( β'R89 , β'D404 , β'Q415 , and RbpA-E17 ) . Notably , the Fdx/RNAP interaction is stabilized by two cation-π interactions between β'R84 and the aromatic ring of the Fdx homodichloroorsellinic acid moiety ( Figure 1B ) and β'R89 and the conjugated double-bond system centered between C4 and C5 of the macrolide core ( Figures 1B and 4A , B ) . Fdx interacts with residues from eight distinct structural elements ( Lane and Darst , 2010 ) of the initiation complex ( βSw3 , βSw4 , β residues belonging to the clamp , β'ZBD , β'lid , β'Sw2 , the σ-finger , and the RbpaNTT ( Figure 4A , B ) . Amino-acid substitutions conferring FdxR have been identified in RNAP β or β' subunits from Bacillus subtilis ( Gualtieri et al . , 2006 ) , Cdf ( Kuehne et al . , 2017 ) , Enterococcus faecalis ( Gualtieri et al . , 2009 ) , and Mtb ( Kurabachew et al . , 2008 ) , corresponding to Mtb RNAP β residues Q1054 ( Sw3 ) , V1100 and V1123 ( clamp ) , and β' residues R89 ( ZBD ) , P326 ( lid ) , and R412 ( Sw2 ) . The structure shows that each of these residues makes direct interactions with Fdx ( Figure 4A , B ) . All five chemical moieties of Fdx ( Figure 1B ) interact with at least one RNAP residue that confers FdxR when mutated ( Figure 4B ) , suggesting that each moiety may be important for Fdx action . In addition to the β and β' subunits , Fdx interacts with residues of the σ-finger ( D424 and V445; Figure 4A , B ) . Finally and unexpectedly , Fdx contacts residues from the RbpANTT ( Figure 4A , B ) . To test the functional importance of the RpbANTT for Fdx inhibition in vitro , we compared Fdx inhibition of MtbσA-holo with either RbpA or RbpA with the NTT truncated ( RbpAΔNTT ) in the abortive initiation assay ( Figure 1—figure supplement 1B ) . Truncation of the RbpA-NTT caused a 35-fold increase in resistance to Fdx ( Figure 4C ) . RbpA is essential in Mtb and Msm , but strains carrying RbpAΔNTT are viable ( Hubin et al . , 2017a ) , allowing us to test the role of the RbpANTT in Fdx growth inhibition of Msm cells . We performed zone of inhibition assays on two Msm strains that are isogenic except one harbors wild-type RbpA ( RbpAwt ) and the other RbpAΔNTT ( Hubin et al . , 2017a ) . The Msm RbpA∆NTT strain grew considerably slower on plates , taking approximately twice the time as the wild-type Msm to reach confluency . Despite the growth defect , the RbpAΔNTT strain was significantly less sensitive to Fdx ( Figure 4D ) . Discs soaked with up to 250 μM Fdx did not produce inhibition zones with RbpAΔNTT but inhibition zones were apparent with RbpAwt . At 500 μM Fdx , the inhibition zone for RbpAΔNTT was significantly smaller than for RbpAwt . By contrast , 860 μM streptomycin , a protein synthesis inhibitor , produced equal inhibition zones for the RbpAwt and RbpAΔNTT strains . We conclude that the essential role of RbpA in Mtb transcription is key to the relatively high sensitivity of Mtb cells to Fdx . The RNAP switch regions are thought to act as hinges connecting the mobile clamp domain to the rest of the RNAP ( Gnatt et al . , 2001; Lane and Darst , 2010 ) . Bacterial RNAP inhibitors myxopyronin , corallopyronin , and ripostatin bind Sw1 and Sw2 and stabilize a closed-clamp conformation of the RNAP ( Belogurov et al . , 2009; Mukhopadhyay et al . , 2008 ) . The Fdx binding determinant does not overlap the sites for these other inhibitors , but the Fdx interactions with the Sw2 , Sw3 , and Sw4 regions ( Figure 4A , B ) suggest that Fdx may influence the clamp conformation as well . To understand the role of Fdx in clamp movement without the complication of DNA binding in the RNAP active site cleft , we determined cryo-EM structures of Mtb RbpA/σA-holo without DNA , with Fdx and without Fdx . Although the particles in the original cryo-EM datasets of Mtb RbpA/σA-holo were prone to oligomerization , we used 2D classification to isolate single particles and determined reconstructions of Mtb RbpA/σA-holo without DNA and with Fdx ( overall 6 . 5 Å resolution from 21 , 000 particles , Figure 5A , Figure 5—figure supplement 1A–E ) and without Fdx ( overall 5 . 2 Å resolution from 88 , 000 particles; Figure 5A , Figure 5—figure supplement 1F–I ) . The cryo-EM density maps were of sufficient detail to visualize the bound antibiotic in the Fdx complex ( Figure 5—figure supplement 1E ) and to determine the domain organization ( including the clamp conformation ) by rigid-body refinement ( Figure 5A ) . Thus , we were able to compare the RNAP conformational states from four solution complexes of the same RNAP in the absence of crystal packing forces ( Figure 5B ) . The four structures were superimposed by the structural core module ( Supplementary file 2 ) , comprising the ω subunit and highly conserved β and β′ regions in or near the active center that have not been observed to undergo significant conformational changes in dozens of RNAP structures . Using the RPo structure ( Figure 2A ) as a reference , the structures superimposed with rmsds < 0 . 4 Å over at least 898 aligned α-carbon ( Cα ) atoms of the structural core module , but rmsds > 9 Å for 461 Cα-atoms of the clamp modules ( Supplementary file 2 ) , indicating large shifts of the clamp module with respect to the rest of the RNAP in the different complexes . Alignment of the structures revealed that the clamp conformational changes could be characterized as rigid body rotations about a common rotation axis ( Figure 5B ) . Assigning a clamp rotation angle of 0° ( closed clamp ) to the RPo structure ( blue , Figure 5B ) , the RbpA/σA-holo clamp is rotated open by about 12° ( green , Figure 5B ) . Because this complex is not interacting with any other ligands that might be expected to alter the clamp conformation ( such as Fdx or DNA ) , we will refer to this as the 'relaxed' clamp conformation . The two Fdx-bound complexes , with or without us-fork DNA , show further opening of the clamp ( 14° and 15° , respectively; orange and red in Figure 5B ) . In the high-resolution Fdx/TIC structure ( Figure 1D ) , Fdx binds in a narrow gap between the open clamp module and the rest of the RNAP ( Figure 5C ) . Examination of the high-resolution RPo ( closed clamp ) structure reveals that clamp closure pinches off the Fdx binding pocket ( Figure 5D ) - Fdx can only bind to the open-clamp conformation of RNAP . We thus conclude that Fdx acts like a doorstop , binding and stabilizing the open-clamp conformation . Clamp dynamics play multiple important roles in the transcription cycle . Motions of the clamp module and the role of the switch regions as hinges were first noted by comparing crystal structures of free RNAPs ( Cramer et al . , 2001; Zhang et al . , 1999 ) with the crystal structure of an elongation complex containing template DNA and RNA transcript ( Gnatt et al . , 2001 ) . Binding of the downstream duplex DNA and RNA/DNA hybrid in the RNAP active-site cleft was proposed to close the clamp around the nucleic acids , explaining the high processivity of the transcription elongation complex . Numerous subsequent crystal structures have supported the idea that stable , transcription-competent complexes of RNAP with nucleic acids , either RPo ( Bae et al . , 2015; Hubin et al . , 2017b; Zuo and Steitz , 2015 ) or elongation complexes ( Gnatt et al . , 2001; Kettenberger et al . , 2004; Vassylyev et al . , 2007 ) , correlate with the closed-clamp conformation . Effects of crystal packing forces on clamp conformation , however , cannot always be ruled out . Observations of clamp positions by solution FRET ( Chakraborty et al . , 2012 ) , and more recently in cryo-EM structures ( Bernecky et al . , 2016; Hoffmann et al . , 2015; Kang et al . , 2017; Neyer et al . , 2016 ) ( in the absence of crystal packing forces ) have confirmed the relationship between clamp closure and stable nucleic-acid complexes . Clamp motions have also been shown to play a critical role in the process of promoter melting to form the transcription bubble during RPo formation ( Feklistov et al . , 2017 ) . Thus , the trapping of an open-clamp RNAP conformation by Fdx in unrestrained cryo-EM conditions ( Figure 5C ) suggests that Fdx inhibits transcription initiation by preventing clamp motions required for RPo formation , or by not allowing RNAP to form stable transcription-competent complexes with nucleic acids , or both ( Figure 5C , D ) . These results are broadly consistent with mechanistic analyses of ( Tupin et al . , 2010 ) and ( Morichaud et al . , 2016 ) showing that Fdx blocks promoter melting at an early step but providing RNAP a pre-melted template overcomes the block . These authors proposed that Fdx likely prevented the clamp from closing , again consistent with our structural findings . Our results establish the molecular details of Fdx interactions with the bacterial RNAP ( Figure 4A , B ) and a mechanism of action for Fdx ( Figure 5C , D ) . Crucially , the essential actinobacterial GTF RbpA is responsible for the high sensitivity of mycobacterial RNAP to Fdx both in vitro ( Figure 4C ) and in vivo ( Figure 4D ) . This new knowledge provides a structural platform for the development of antimicrobials that target the Fdx binding determinant and underscores the need to define structure-activity relationships of drug leads using near-native states , in this case using cryo-EM with the RbpA/σA-holo complex to guide development of effective Mtb treatments . Mtb RNAP overexpression plasmid . The overexpression plasmid ( OEP ) for Mtb RNAP was engineered from an existing OEP for M . bovis RNAP ( Czyz et al . , 2014 ) , pAC22 . Four modifications were made . First , a sequence in pAC22 that encodes an N-terminal 6-aa insertion at codon 2 of rpoB was removed . Second , a sequence upstream of rpoZ that included an ATG that potentially allowed an N-terminal extension on ω was removed . Third , to increase protein expression , the ribosome-binding sites ( RBSs ) for rpoA , rpoZ , and rpoB::C were re-engineered to encode stronger predicted RBSs using predicted translation initiation rates calculated using the Salis RBS strength calculator ( https://salislab . net/software/ ) ( Espah Borujeni et al . , 2014 ) . Finally , the single amino-acid difference between Mtb RNAP and Mbo RNAP at position 69 of β was changed from Pro ( Mbo ) to Arg ( Mtb ) ( P69R ) . The resulting plasmid , pMP55 , encodes β S450Y ( RifR ) Mtb RNAP . A wild-type derivative ( RifS ) was engineered by site-direct mutagenesis to give plasmid pMP61 that expresses the wild-type Mtb RNAP . A derivative of pMP55 encoding the β Q1054H substitution that confers resistance to Fidaxomicin ( Fdx ) ( Kurabachew et al . , 2008 ) was constructed by site-directed mutagenesis . Samples for Cryo-EM grid preparation used Mtb His-tagged-σA and RbpA co-expressed and purified as previously described ( Hubin et al . , 2015; 2017a ) . To compare Fdx sensitivity of full-length RbpA and RbpAΔNTT , these proteins , and Mtb His-tagged-σA were expressed separately and purified as described previously ( Hubin et al . , 2015; 2017a ) . Briefly , Rosetta-2 cells ( EMD-Millipore/Novagen ) were co-transformed with pET plasmids expressing Mtb σA ( His-tagged ) and RbpA and induced with 0 . 5 mM IPTG at 30°C for 4 hr . Clarified lysates was subjected to Ni2+ affinity , removal of the His-tag , a second Ni2+ affinity ( collecting the flow through this time ) and size exclusion chromatography . Mtb RNAP was expressed and purified as previously described for Mbo and Msm RNAPs ( Davis et al . , 2015; Hubin et al . , 2017a ) . Eco core RNAP , Eco σ70 , Msm σA , Msm RbpA , and Msm core RNAP were expressed and purified as described ( Davis et al . , 2015; Hubin et al . , 2015; 2017a ) . In vitro abortive initiation transcription assays were performed using the WT AP3 ( −87 to +71 ) promoter at 37°C as described ( Davis et al . , 2015 ) : Assays were performed in KCl assay buffer ( 10 mM Tris-HCl , pH 8 . 0 , 50 mM KCl , 10 mM MgCl2 , 0 . 1 mM EDTA , 0 . 1 mM DTT , 50 μg-/mL BSA ) . The IC50's of Fdx on the different holos were calculated as follows: Mtb and Msm RNAP holo were incubated with the cognate σA and RbpA-FL or RbpA∆NTT , and Eco RNAP ( 50 nM ) was incubated with σ70 , to form holos . Holos were incubated with Fdx for 10 min at 37°C prior to addition of template DNA . DNA template was added ( 10 nM final ) and the samples were incubated for 15 min at 37°C for open complex formation . Transcription was initiated with nucleotide mix , and stopped with a 2X Stop buffer ( 45 mM Tris-HCl , 45 mM Boric acid , 8 M Urea , 30 mM EDTA , 0 . 05% bromophenol blue , 0 . 05% xylene cyanol ) after 10 min at 37°C . Transcription products were denatured by heating at 95°C for two minutes and visualized by polyacrylamide gel electrophoresis using phosphorimagery and quantified using ImageJ ( Schneider et al . , 2012 ) . Msm strains MGM6232 ( ΔrbpA attB::rbpA kan ) and MGM6234 ( ΔrbpA attB::rbpA ( 28-114 ) kan ) ( Hubin et al . , 2017a ) were grown overnight in LBsmeg ( LB with 0 . 5% glycerol , 0 . 5% dextrose and 0 . 05% Tween80 ) and 2 mL were centrifuged and resuspended in 200 µL of residual media and then plated . Filter discs were placed on the plates and stock solutions of Fdx were prepared in 10% DMSO at different concentrations ( 50 μM , 100 μM , 250 μM and 500 μM ) . 10 μl of antibiotic from each stock solution was pipetted onto the disks . Streptomycin ( 0 . 5 mg/ml , 860 μM ) and 10% DMSO were used as positive and negative controls , respectively . Plates were incubated at 37°C for 74 hr and the zone of inhibition around each disk was photographed and measured . Mtb RbpA/σA-holo ( 0 . 5 ml of 5 mg/ml ) was injected into a Superose 6 Increase column ( GE Healthcare Life Sciences , Pittsburgh , PA ) equilibrated with 20 mM Tris-HCl pH 8 . 0 , 150 mM K-Glutamate , 5 mM MgCl2 , 2 . 5 mM DTT . The peak fractions of the eluted protein were concentrated by centrifugal filtration ( EMD-Millipore , Darmstadt , Germany ) to 6 mg/mL protein concentration . Fdx ( when used ) was added at 100 μM and us-fork DNA ( when used ) was added to 20 μM . The samples were incubated on ice for 15 min and then 3- ( [3-cholamidopropyl]dimethylammonio ) −2-hydroxy-1-propanesulfonate ( CHAPSO ) was added to the sample for a final concentration of 8 mM prior to grid preparation . C-flat CF-1 . 2/1 . 3-4Au 400 mesh gold grids ( Protochips , Morrisville , NC ) were glow-discharged for 20 s prior to the application of 3 . 5 μl of the sample ( 4 . 0–6 . 0 mg/ml protein concentration ) . After blotting for 3–4 . 5 s , the grids were plunge-frozen in liquid ethane using an FEI Vitrobot Mark IV ( FEI , Hillsboro , OR ) with 100% chamber humidity at 22°C . Structural biology software was accessed through the SBGrid consortium ( Morin et al . , 2013 ) . Fdx/RbpA/σA-holo/us-fork . The grids were imaged using a 300 keV Titan Krios ( FEI ) equipped with a K2 Summit direct electron detector ( Gatan , Warrendale , PA ) . Images were recorded with Leginon ( Nicholson et al . , 2010 ) in counting mode with a pixel size of 1 . 1 Å and a defocus range of 0 . 8 μm to 1 . 8 μm . Data were collected with a dose of 8 electrons/px/s . Images were recorded over a 10 s exposure with 0 . 2 s frames ( 50 total frames ) to give a total dose of 66 electrons/Å2 . Dose-fractionated subframes were aligned and summed using MotionCor2 ( Zheng et al . , 2017 ) and subsequent dose-weighting was applied to each image . The contrast transfer function was estimated for each summed image using Gctf ( Zhang , 2016 ) . From the summed images , Gautomatch ( developed by K . Zhang , MRC Laboratory of Molecular Biology , Cambridge , UK , http://www . mrc-lmb . cam . ac . uk/kzhang/Gautomatch ) was used to pick particles with an auto-generated template . Autopicked particles were manually inspected , then subjected to 2D classification in cryoSPARC ( Punjani et al . , 2017 ) specifying 50 classes . Poorly populated and dimer classes were removed , resulting in a dataset of 582 , 169 particles . A subset of the dataset was used to generate an initial model of the complex in cryoSPARC ( ab-initio reconstruction ) . Using the ab-initio model ( low-pass filtered to 30 Å-resolution ) , particles were 3D classified into two classes using cryoSPARC heterogenous refinement . CryoSPARC homogenous refinement was performed for each class using the class map and corresponding particles , yielding two structures with different clamp conformations: open ( Fdx/RbpA/σA-holo/us-fork; Figure 1D ) and closed [RbpA/σA-holo/ ( us-fork ) 2; Figure 2A] . Two rounds of heterogenous/homogeneous refinements were performed for each class to achieve the highest resolution . The open class ( Fdx/RbpA/σA-holo/us-fork ) contained 173 , 509 particles with an overall resolution of 3 . 38 Å ( Figure 1—figure supplement 3A ) while the closed class [Mtb RNAP/σA/RbpA/ ( us-fork ) 2] contained 171 , 547 paricles with a overall resolution of 3 . 27 Å ( Figure 2—figure supplement 1A ) . Particle orientations of each class were plotted in cryoSPARC ( Figure 1—figure supplement 1C , Figure 2—figure supplement 1C ) . FSC calculations ( Figure 1—figure supplement 1A , Figure 2—figure supplement 1A ) were performed in cryoSPARC and the half-map FSC ( Figure 1—figure supplement 1B , Figure 2—figure supplement 1B ) was calculated using EMAN2 ( Tang et al . , 2007 ) . Local resolution calculations ( Figures 1E and 2B ) were performed using blocres ( Cardone et al . , 2013 ) . Mtb RbpA/σA-holo . The grids were imaged using a 200 keV Talos Arctica ( FEI ) equipped with a K2 Summit direct electron detector ( Gatan ) . Images were recorded with Serial EM ( Mastronarde , 2005 ) in super-resolution counting mode with a super-resolution pixel size of 0 . 75 Å and a defocus range of 0 . 8 μm to 2 . 4 μm . Data were collected with a dose of 8 electrons/px/s . Images were recorded over a 15 s exposure using 0 . 3 s subframes ( 50 total frames ) to give a total dose of 53 electrons/Å2 . Dose-fractionated subframes were 2 × 2 binned ( giving a pixel size of 1 . 5 Å ) , aligned and summed using Unblur ( Grant and Grigorieff , 2015 ) . The contrast transfer function was estimated for each summed image using Gctf ( Zhang , 2016 ) . From the summed images , Gautomatch ( developed by K . Zhang , MRC Laboratory of Molecular Biology , Cambridge , UK , http://www . mrc-lmb . cam . ac . uk/kzhang/Gautomatch ) was used to pick particles with an auto-generated template . Autopicked particles were manually inspected , then subjected to 2D classification in RELION ( Scheres , 2012 ) specifying 100 classes . Poorly populated classes were removed , resulting in a dataset of 289 , 154 particles . These particles were individually aligned across movie frames and dose-weighted using direct-detector-align_lmbfgs software to generate ‘polished’ particles ( Rubinstein and Brubaker , 2015 ) . A subset of the dataset was used to generate an initial model of the complex in cryoSPARC ( ab-initio reconstruction ) . ‘Polished’ particles were 3D auto-refined in RELION using this ab-initio 3D template ( low-pass filtered to 60 Å-resolution ) . RELION 3D classification into two classes was performed on the particles using the refined map and alignment angles . Among the 3D classes , the best-resolved class , containing 87 , 657 particles , was 3D auto-refined and post-processed in RELION . The overall resolution of this class was 6 . 9 Å ( before post-processing ) and 5 . 2 Å ( after post-processing ) . Subsequent 3D classification did not improve resolution of this class . Fdx/RbpA/σA-holo . The same procedure as described above for Mtb RbpA/σA-holo was used . After RELION 2D classification , poorly populated classes were removed , resulting in a dataset of 63 , 839 particles . In the end , the best-resolved 3D class , containing 21 , 115 particles , was 3D auto-refined and post-processed in RELION . The overall resolution of this class was 8 . 1 Å ( before post-processing ) and 6 . 5 Å ( after post-processing ) . To build initial models of the protein components of the complex , Msm RbpA/σA-holo/us-fork structure ( PDB ID 5TWI ) ( Hubin et al . , 2017a ) was manually fit into the cryo-EM density maps using Chimera ( Pettersen et al . , 2004 ) and real-space refined using Phenix ( Adams et al . , 2010 ) . In the real-space refinement , domains of RNAP were rigid-body refined . For the high-resolution structures , the rigid-body refined models were subsequently refined with secondary structure restraints . A model of Fdx was generated from a crystal structure ( Serra et al . , 2017 ) , edited in Phenix REEL , and refined into the cryo-EM density . Refined models were inspected and modified in Coot ( Emsley and Cowtan , 2004 ) according to cryo-EM maps , followed by further real-space refinement with PHENIX .	Tuberculosis ( TB ) is an infectious disease that affects over ten million people every year . The Mycobacterium tuberculosis bacteria that cause the disease spread through the air from one person to another and mainly infect the lungs . Although curable , TB is difficult to eradicate because it is remarkably widespread , with one third of the world’s population estimated to carry the bacteria . Treatment for TB involves a mix of antibiotics that should be taken for several months to a year . The number of multidrug-resistant TB cases , where the infection is not treatable by the common cocktail of antibiotics , is rapidly increasing . There is therefore a need to discover new drugs that can kill the M . tuberculosis bacteria . An antibiotic called fidaxomicin is used to treat intestinal infections . Although it can kill Mycobacterium tuberculosis cells in culture , it is not absorbed from the intestines to the blood and thus cannot reach the lungs to kill the bacteria . It may be possible to change the structure of the drug so that it can enter the bloodstream . Before this can be done , researchers need to understand exactly how fidaxomicin kills the bacteria so that they know which parts of the drug they can alter without making it less effective . Fidaxomicin kills bacterial cells by binding to an enzyme called RNA polymerase . The antibiotic prevents the enzyme from reading and ‘transcribing’ DNA to form molecules that are essential for life . To learn more about how fidaxomicin has this effect , Boyaci , Chen et al . used cryo-electron microscopy to look at structures of the M . tuberculosis RNA polymerase in different states , including when it was bound to fidaxomicin . The structures reveal the chemical details of the interactions between the RNA polymerase and the antibiotic . The two molecules bind to each other through a region of the RNA polymerase that is unique to M . tuberculosis and closely related bacteria . Fidaxomicin acts like a doorstop to jam the RNA polymerase in an open state that cannot bind to DNA and transcribe genes . Medicinal chemists could now build on these findings to develop new drugs that might treat TB , either by modifying fidaxomicin or designing new antibiotics that bind to the same region of the RNA polymerase . Because the fidaxomicin-binding region of the RNA polymerase is specific to M . tuberculosis new antibiotics could be tailored towards the bacteria that have a minimal effect on a patient’s normal gut bacteria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2018	Fidaxomicin jams Mycobacterium tuberculosis RNA polymerase motions needed for initiation via RbpA contacts
Pioneer transcription factors recognise and bind their target sequences in inaccessible chromatin to establish new transcriptional networks throughout development and cellular reprogramming . During this process , pioneer factors establish an accessible chromatin state to facilitate additional transcription factor binding , yet it remains unclear how different pioneer factors achieve this . Here , we discover that the pluripotency-associated pioneer factor OCT4 binds chromatin to shape accessibility , transcription factor co-binding , and regulatory element function in mouse embryonic stem cells . Chromatin accessibility at OCT4-bound sites requires the chromatin remodeller BRG1 , which is recruited to these sites by OCT4 to support additional transcription factor binding and expression of the pluripotency-associated transcriptome . Furthermore , the requirement for BRG1 in shaping OCT4 binding reflects how these target sites are used during cellular reprogramming and early mouse development . Together this reveals a distinct requirement for a chromatin remodeller in promoting the activity of the pioneer factor OCT4 and regulating the pluripotency network . Transcription factors read DNA-encoded information to control gene transcription and define the complement of proteins a cell can produce . They achieve this by physically binding specific DNA sequence motifs in gene regulatory elements that ultimately control how RNA polymerases engage with and transcribe genes ( Spitz and Furlong , 2012; Voss and Hager , 2014 ) . In prokaryotes , transcription factors tend to function as self-contained units that recognize extended DNA sequences with high specificity and affinity ( Wunderlich and Mirny , 2009 ) . Therefore , simple sequence-based principles appear to dictate their binding in the genome . In contrast , transcription factors in higher eukaryotes recognize shorter DNA sequences and often interface with other DNA-binding transcription factors in a combinatorial manner to achieve affinity and specificity ( Wunderlich and Mirny , 2009; Villar et al . , 2014 ) . In addition to the widespread requirement for combinatorial DNA binding , eukaryotic transcription factors are also confronted with the added complexity that eukaryotic DNA is wrapped around histones to form nucleosomes and chromatin ( Kornberg and Lorch , 1999 ) . The association of DNA with nucleosomes can therefore compete with DNA-binding transcription factors , limiting their ability to engage with the genome ( Spitz and Furlong , 2012; Voss and Hager , 2014 ) . Indeed , in many cases transcription factor binding appears to require pre-existing chromatin accessibility ( Guertin and Lis , 2010; Biddie et al . , 2011; John et al . , 2011 ) . This chromatin barrier therefore poses a significant challenge in establishing new gene regulatory elements when cells differentiate during development . Therefore , to overcome this obstacle , cells have evolved a specialised set of transcription factors that are able to recognise their cognate motifs even when nucleosomes are present ( Zaret and Carroll , 2011 ) . This is exemplified by the forkhead box A1 ( FoxA1 ) transcription factor , which binds to and de-compacts nucleosomal DNA in vitro ( Cirillo et al . , 2002 ) and in vivo ( Holmqvist et al . , 2005; Lupien et al . , 2008; Iwafuchi-Doi et al . , 2016 ) . Based on this type of activity , factors like FoxA1 have been called ‘pioneer’ transcription factors as they appear to play a primary role in recognizing and shaping how new gene regulatory elements are established in previously inaccessible chromatin ( Magnani et al . , 2011; Zaret and Carroll , 2011 ) . Once bound to their target sites , pioneer transcription factors appear to create accessible chromatin ( Raposo et al . , 2015; Schulz et al . , 2015 ) that supports the recruitment of non-pioneer transcription factors and the formation of functional gene regulatory elements ( Theodorou et al . , 2013; Wapinski et al . , 2013; Foo et al . , 2014; Xu et al . , 2014; Schulz et al . , 2015 ) . These unique features of pioneer transcription factors allow them to function as master regulators that underpin developmental transitions , cellular reprogramming , and responses to cellular signalling events . The capacity of pioneer transcription factors to destabilise or reposition nucleosomes to facilitate chromatin accessibility at their target sites appears to be an essential feature of their activity . Initially , it was proposed that pioneer factors might achieve this by interacting specifically with nucleosomes to alter chromatin structure . For example , FoxA1 and FoxO1 structurally resemble histone H1 , which has been proposed to lead to displacement of H1 and destabilisation of neighbouring nucleosomes ( Cirillo et al . , 2002; Hatta and Cirillo , 2007 ) . Alternatively , it has been proposed that ATP-dependent chromatin remodellers may be required to assist pioneer transcription factors in establishing accessible chromatin ( Hu et al . , 2011; Marathe et al . , 2013; Ceballos-Chávez et al . , 2015; Swinstead et al . , 2016 ) . Nevertheless , for most pioneer transcription factors there remains limited understanding of the mechanisms by which they translate their binding within inaccessible and nucleosome-occluded chromatin into effects on chromatin structure that support additional transcription factor binding and gene regulation . One paradigm that has been used to study pioneer transcription factor function is the cellular reprogramming of somatic cells into induced pluripotent stem cells ( iPSC ) by the Yamanaka transcription factors ( OCT4 , SOX2 , KLF4 , and c-MYC ) ( Takahashi and Yamanaka , 2006 ) . With the exception of c-MYC , these transcription factors are proposed to act as pioneers during cellular reprogramming and can bind their targets sites even when they are occupied by nucleosomes ( Soufi et al . , 2012 , 2015; Chen et al . , 2016 ) . It is thought that this pioneering activity may then pave the way for nucleosome displacement and further transcription factor binding that establishes functional regulatory elements ( You et al . , 2011; Shakya et al . , 2015; Simandi et al . , 2016 ) . While it is clear that these transcription factors have the capacity to engage with previously inaccessible regions of chromatin during iPSC reprogramming , it is unknown whether additional mechanisms are required to transition these initial engagement events into functionally mature transcription factor binding as part of the pluripotency-associated regulatory network . To begin addressing this fundamental gap in our understanding , we have used mouse embryonic stem cells , which exist in an established pluripotent state , as a model system to dissect how the core pluripotency transcription factor OCT4 engages with target sites inside cells . In doing so we discover that OCT4 occupies sites that would otherwise be inaccessible and is required to shape the occupancy of additional pluripotency transcription factors . OCT4 achieves this by recruiting the chromatin remodelling factor BRG1 to support not only its own binding but also to stabilize further transcription factor binding events required to support pluripotency-associated gene regulation . This reveals that although OCT4 can engage with its target sites in inaccessible chromatin , the recruitment of a chromatin remodelling enzyme is a fundamental step in pluripotency transcription factor binding and gene expression . Intriguingly , OCT4-bound regulatory sites that require BRG1 are activated more slowly in response to cellular reprogramming and later during early development . Together these observations reveal that the capacity of OCT4 to cooperate with a chromatin remodeller is a key feature of its pioneering activity , and this is required to mature transcription factor binding and create functional gene regulatory elements . The atomic structure of OCT4 and its DNA binding activity in vitro have been characterized in detail ( Klemm et al . , 1994; Esch et al . , 2013 ) , yet the mechanisms which support how OCT4 functions as a pioneer transcription factor in vivo remain poorly defined . Our current understanding is based largely on overexpression studies during iPSC reprogramming where OCT4 binds a large number of its motifs in inaccessible regions of chromatin ( Soufi et al . , 2012 , 2015; Chen et al . , 2016 ) . However , whether these binding events represent functionally relevant pioneering interactions that precede the creation of accessible chromatin and downstream target gene expression remains largely unknown . We therefore set out to examine how OCT4 engages with and functions at its binding sites in a more physiologically relevant situation in mouse embryonic stems cells ( ESCs ) , where OCT4 plays an essential role in shaping distal regulatory element function and controlling the pluripotent transcriptome . In order to achieve this , we first set out to identify a bona fide set of OCT4 binding events in mouse ESCs using chromatin immunoprecipitation coupled with massively parallel sequencing ( ChIP-seq ) . We applied this approach to a conditional mouse ESC line where addition of a small molecule ( doxycycline ) leads to loss of OCT4 expression ( Niwa et al . , 2000 ) . It is known that prolonged removal of OCT4 in ESCs results in loss of pluripotency and cellular differentiation ( Niwa et al . , 2000; Adachi et al . , 2013 ) . We therefore identified an acute treatment condition where following 24 hr of doxycycline treatment cells lacked appreciable OCT4 protein ( Figure 1A ) but retained normal ESC morphology , were alkaline phosphatase positive , and expressed wild type levels of the pluripotency transcription factors SOX2 and NANOG ( Figure 1A , B ) . Analysis of our OCT4 ChIP-seq identified 15 , 920 high-confidence OCT4 binding sites that were lost following doxycycline treatment ( Figure 1C ) and were highly enriched for known OCT4 binding motifs ( Figure 1—figure supplement 1A , B ) . The majority of these binding events ( 75% ) correlated with a histone modification signature usually associated with distal regulatory elements ( high H3K4me1/low H3K4me3 ) , while only a small subset ( 6 . 8% ) corresponded to sites with a promoter associated histone modification signature ( high H3K4me3/low H3K4me1 ) ( Figure 1D; Figure 1—figure supplement 1C , D ) . These observations are consistent with previous reports indicating that OCT4 binds extensively to distal as opposed to promoter proximal regulatory regions in the genome ( Chen et al . , 2008; Göke et al . , 2011 ) . The identification of bona fide OCT4 target sites , and the maintenance of stem cell features under these treatment conditions , provided us with an opportunity to examine in more detail where and how OCT4 normally engages with the ESC genome , and to ask how this is related to underlying chromatin accessibility and transcription factor co-occupancy . 10 . 7554/eLife . 22631 . 003Figure 1 . OCT4 binds distal regulatory sites in mouse embryonic stems cells to shape chromatin accessibility . ( A ) Western blot analysis of OCT4cond ( ZHBTC4 ) mouse ESCs before ( UNT ) and after 24 hr treatment with doxycycline ( DOX ) . ( B ) Alkaline phosphatase staining of OCT4cond mouse ESCs before ( UNT ) and after 24 hr DOX treatment . ( C ) A metaplot of OCT4 ChIP-seq signal in OCT4cond ESCs before ( UNT ) and after 24 hr DOX treatment at OCT4 peaks ( n = 15920 ) . ( D ) Annotation of OCT4 peaks as promoters or distal regulatory elements using the relative enrichment of promoter-associated H3K4me3 or distal regulatory element-associated H3K4me1 . ( E ) A genomic snapshot of ATAC-seq and OCT4 ChIP-seq signal in OCT4cond ESCs before ( UNT ) and after 24 hr DOX treatment at the Utf1 locus . The downstream OCT4-bound regulatory element is highlighted in the grey box . ( F ) A heatmap illustrating OCT4 targets ( n = 15920 ) ranked by their loss of chromatin accessibility ( ATAC-seq ) after 24 hr DOX treatment of OCT4cond ESCs . Normalised read densities for ATAC-seq and OCT4 ChIP-seq are presented , with a heatmap indicating their annotation as either promoters or distal regulatory elements ( right ) . ( G ) A metaplot of OCT4cond ATAC-seq signal before ( UNT ) and after 24 hr DOX treatment at OCT4 binding sites with significant reductions in ATAC-seq signal ( OCT4-dependent; n = 11557 ) and those without significant changes ( OCT4-independent; n = 4362 ) . Tn5 control represents transposition of purified genomic DNA to control for potential sequence bias . ( H ) As in ( G ) , profiling the changes in nucleosome occupancy before ( UNT ) and after ( DOX ) OCT4 depletion . Nucleosome signal was generated using the NucleoATAC package . ( I ) Piecharts identifying the proportion of OCT4-bound distal regulatory elements ( left ) or OCT4-bound promoters ( right ) that display significant changes in chromatin accessibility as measured by ATAC-seq . Changes were deemed to be significant with FDR < 0 . 05 and a fold change greater than 1 . 5-fold . ( J ) A metaplot depicting the OCT4cond ATAC-seq signal before ( UNT ) and after ( DOX ) treatment at the 25% of OCT4 peaks with the greatest changes in ATAC-seq signal following OCT4 depletion . ( K ) Gene ontology analysis for genes closest to OCT4 target sites depicted in ( J ) . This reveals an enrichment for the pluripotency expression network ( left ) and biological processes associated with developmental gene regulation ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 00310 . 7554/eLife . 22631 . 004Figure 1—figure supplement 1 . Annotation and characterisation of OCT4 binding sites in OCT4cond ESCs . ( A ) Motif enrichment analysis for canonical motif sequences ( top ) or de novo motif sequences ( bottom ) reveals high enrichment for OCT4 ( POU5F1 ) and similar motif sequences in the OCT4cond OCT4 binding sites ( n = 15920 ) . % reflects the percentage of OCT4 peaks identified to contain each motif sequence . ( B ) Central enrichment analysis performed by CentriMO reveals that canonical and de novo OCT4 motifs are centrally enriched at OCT4 peak intervals . ( C ) A metaplot of H3K4me1 and H3K4me3 at OCT4 binding sites annotated as distal regulatory elements or promoters . ( D ) A violin plot comparing the distance from OCT4 binding sites annotated as distal regulatory elements or promoters to nearest refGene TSS . p denotes significance value by Wilcoxon ranked-sign test . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 00410 . 7554/eLife . 22631 . 005Figure 1—figure supplement 2 . Changes in chromatin accessibility at OCT4-bound sites following depletion of OCT4 in ESCs . ( A ) Genomic snapshots of OCT4-dependent chromatin accessibility ( ATAC-seq ) before ( UNT ) and after ( DOX ) OCT4 depletion . Distal OCT4 binding sites are highlighted in grey boxes . ( B ) Genomic snapshot of OCT4-independent chromatin accessibility at OCT4-bound promoter . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 00510 . 7554/eLife . 22631 . 006Figure 1—figure supplement 3 . Chromatin accessibility profiling at OCT4 binding sites in somatic cell lines or tissues . ( A ) A metaplot of ENCODE DNase-seq profiles for eight mouse cell lines or tissues ( Yue et al . , 2014 ) at OCT4 targets most dependent upon OCT4 for normal chromatin accessibility ( 25% most affected in OCT4cond ESCs; as in Figure 1J ) . This reveals that these sites are completely inaccessible in somatic cell lines or tissues which lack OCT4 expression . ( B ) A metaplot analysis of ENCODE DNase-seq data in ( A ) profiled at DNase I hypersensitive sites identified from each cell line/tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 006 During somatic cell reprogramming , exogenous OCT4 is proposed to function as a pioneer transcription factor that can bind to its sequence motifs in inaccessible regions of chromatin . However , it remains unclear whether binding to inaccessible chromatin is also a feature of normal OCT4 binding in mouse ESCs . To address this important question we used the assay for transposase-accessible chromatin with massively parallel sequencing ( ATAC-seq ) which provides a genome-wide measure of chromatin accessibility ( Buenrostro et al . , 2013 ) and examined ATAC-seq signal in wild type and OCT4-depleted cells . Although OCT4-bound sites were highly accessible in wild type cells , when we examined ATAC-seq signal in the OCT4-depleted ESCs , 72% of OCT4 targets showed significant reductions in chromatin accessibility ( Figure 1E , F , G and Figure 1—figure supplement 2 ) and increases in nucleosome occupancy ( Figure 1H ) . These observations are in agreement with previous studies describing a role for OCT4 in maintaining nucleosome-depleted regions and/or chromatin accessibility at individual loci in pluripotent cells ( You et al . , 2011; Shakya et al . , 2015 ) or genome-wide ( Chen et al . , 2014; Lu et al . , 2016 ) . Importantly , OCT4-bound distal regulatory elements appeared to be most significantly affected , while OCT4-bound promoters experienced few significant reductions in accessibility ( Figure 1I ) . Consistent with a pioneering-like role for OCT4 in shaping chromatin structure , many OCT4-bound regulatory elements were completely inaccessible in the OCT4-depleted ESCs ( Figure 1F , J ) and lacked any detectable chromatin accessibility in cells and tissues lacking OCT4 expression ( Figure 1—figure supplement 3 ) . Importantly , OCT4 binding sites that displayed reduced accessibility following OCT4 removal were often in close proximity with genes implicated in the pluripotency regulatory network ( Figure 1K ) , suggesting that these OCT4 binding events may be implicated with the maintenance of pluripotency-associated gene expression . These observations therefore establish that the majority of OCT4 binding events in pluripotent stem cells occur at sites that would otherwise be inaccessible and occupied by nucleosomes , indicating that OCT4 is required to maintain accessible chromatin at its target sites not only during cellular reprogramming but also in the established pluripotent state . A defining feature of pioneer transcription factors is their capacity to support additional transcription factor occupancy , potentially through alteration of local chromatin structure . Given that removal of OCT4 had widespread effects on chromatin accessibility at its binding sites , we wondered whether its absence would result in defects in the binding of other pluripotency-associated transcription factors . To address this possibility , we used ChIP-seq to examine the binding of SOX2 and NANOG in the presence or absence of OCT4 ( Figure 2 ) . Interestingly , we observed major reductions in SOX2 and NANOG binding at the OCT4 target sites that experienced the largest reductions in chromatin accessibility ( Figure 2A , B and Figure 2—figure supplement 1A , B ) . Importantly , the reductions in SOX2 and NANOG binding correlated extremely well with the reductions in chromatin accessibility at OCT4-bound sites ( Figure 2—figure supplement 1C ) . Conversely , the smaller subset of OCT4-bound sites that retained chromatin accessibility following OCT4 depletion retained SOX2 or NANOG occupancy ( Figure 2B ) . This suggests that OCT4 plays a central role in supporting combinatorial binding and chromatin accessibility at the majority of OCT4 binding sites . 10 . 7554/eLife . 22631 . 007Figure 2 . Loss of OCT4 leads to reduced pluripotency-associated transcription factor binding and expression of nearby genes . ( A ) A heatmap illustrating SOX2 and NANOG ChIP-seq at OCT4 targets ( n = 15920 ) ranked by their loss of chromatin accessibility ( ATAC-seq ) after 24 hr DOX treatment of OCT4cond ESCs , as in Figure 1F . ( B ) A metaplot of OCT4cond SOX2 and NANOG ChIP-seq signal before ( UNT ) and after 24 hr DOX treatment at OCT4 binding sites with significant reductions in ATAC-seq signal ( OCT4-dependent; n = 11557 ) and those without significant changes ( OCT4-independent; n = 4362 ) . ( C ) A heatmap of the log2 fold change in RNA-seq signal of the closest gene to OCT4 target sites depicted in ( A ) , following 24 hr DOX treatment of the OCT4cond ESCs . ( D ) A genomic snapshot of Utf1 , which is decreased in expression following loss of OCT4 . The distal OCT4 target site is highlighted by a grey box . ( E ) Comparison of log2 fold change ( log2FC ) in RNA-seq signal for genes neighbouring OCT4 target sites that rely on OCT4 for ATAC-seq signal ( OCT4-dependent; n = 11557 ) or those that do not ( OCT4-independent; n = 4362 ) . ( F ) A Venn diagram comparing the OCT4 binding sites for which the nearest gene has significantly reduced expression after OCT4 ablation ( orange; n = 1979 ) and the OCT4 targets with significant reductions in chromatin accessibility ( red; n = 8788 ) . Only OCT4 target sites for which the nearest gene has sufficient RNA-seq coverage are included . ( G ) A violin plot comparing the distance from OCT4 binding sites with OCT4-dependent ( n = 11557 ) or OCT4-independent ( n = 4362 ) chromatin accessibility to nearest TSS with significant reductions in RNA-seq following 24 hr DOX treatment of the OCT4cond ESCs ( n = 1430 ) . ( H ) Gene ontology analysis for genes down-regulated after loss of OCT4 ( n = 1430 ) reveals enrichment of the pluripotency transcriptional network . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 00710 . 7554/eLife . 22631 . 008Figure 2—figure supplement 1 . Loss of SOX2 and NANOG is highly correlated with reductions in chromatin accessibility at OCT4-SOX2-NANOG targets . ( A ) A Venn diagram depicting the relationship between significant reductions in ATAC-seq and SOX2/NANOG ChIP-seq signal at OCT4-SOX2-NANOG co-bound peaks ( n = 5611 ) . % denotes percentage of OCT4-SOX2-NANOG peaks with significant reduction for each factor or assay . ( B ) A violin plot comparing the log2 fold change in SOX2 and NANOG ChIP-seq signal after depletion of OCT4 between OCT4-SOX2-NANOG co-bound peaks with OCT4-dependent ( n = 3941 ) or OCT4-independent ( n = 1644 ) chromatin accessibility . ( C ) Scatterplots depicting log2 fold changes of SOX2 or NANOG against the log2 fold change in chromatin accessibility ( measured by ATAC-seq ) at OCT4-SOX2-NANOG co-bound targets ( n = 5611 ) in OCT4cond ESCs after 24 hr DOX treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 008 To understand whether this loss of combinatorial transcription factor binding was relevant to gene regulation , we carried out nuclear RNA-seq in wild type and OCT4-depleted cells . Loss of OCT4-dependent chromatin accessibility and transcription factor binding was broadly associated with the down-regulation of nearby genes ( Figure 2C–E ) , such as the pluripotency-associated Utf1 gene ( Figure 2D ) . When we examined this relationship across the genome , reductions in the expression of genes near OCT4-bound distal sites was highly coincident ( 82% ) with reductions in chromatin accessibility ( Figure 2F ) . Interestingly , however , reductions in chromatin accessibility at an OCT4-bound site did not always lead to alterations in the expression of neighbouring gene , suggesting that some bound sites may not function in gene regulation or that they may have regulatory capacity that extends beyond the nearest gene . Nevertheless , genes that were down-regulated in OCT4-depleted ESCs ( n = 1430; OCT4-dependent gene expression ) tended to be significantly closer to OCT4 binding sites that displayed reduced chromatin accessibility ( median distance 16 . 2 kb ) than OCT4 binding sites that retained chromatin accessibility in the absence of OCT4 ( median distance 125 . 4 kb ) ( Figure 2G ) . These sites were also characterized by reductions in SOX2 and NANOG binding ( Figure 2A , B ) and their associated genes were enriched for the pluripotency-associated transcriptional network ( Figure 2H ) . Together , these observations reveal that OCT4 binding plays a primary and widespread role in shaping combinatorial binding of transcription factors at otherwise inaccessible regulatory sites in ESCs , and this activity underpins pluripotency-associated gene expression . Several modalities have been proposed to explain how pioneer factors facilitate chromatin accessibility at otherwise inaccessible regions of the genome . For example , pioneer transcription factors may bind their target motifs and evict or exclude nucleosomes through steric mechanisms ( Cirillo et al . , 2002; Hatta and Cirillo , 2007; Voss and Hager , 2014 ) or they may exploit the activity of chromatin remodelling enzymes ( Marathe et al . , 2013; Ceballos-Chávez et al . , 2015; Swinstead et al . , 2016 ) . However , the relative contribution of such mechanisms to pioneer transcription factor function remains poorly understood and unaddressed for OCT4 . What is clear from our analysis in ESCs is that the majority of OCT4-bound sites exist in an inaccessible chromatin state in its absence , and that OCT4 plays a primary role in supporting SOX2 and NANOG binding at these sites . To examine whether this pioneering-like activity could potentially rely on the activity of ATP-dependent chromatin remodelling enzymes , we took advantage of the extensive ChIP-seq analysis of chromatin remodelling enzymes that exists for mouse ESCs ( Wang et al . , 2014; de Dieuleveult et al . , 2016 ) , and simply examined across the complete complement of accessible sites in the mouse ESC genome whether there was a relationship between binding of any these enzymes and the dependency on OCT4 for chromatin accessibility . Remarkably , of the nine individual chromatin remodelling complexes examined , only BRG1 ( SMARCA4 ) , a catalytic ATPase subunit of the vertebrate SWI/SNF chromatin remodelling complex ( also referred to as BRG1-associated factor ( BAF ) remodelling complexes ) , showed an appreciable correlation between occupancy and the reliance on OCT4 for chromatin accessibility ( Figure 3A ) . Indeed , BRG1 was significantly enriched at regulatory elements that relied on OCT4 for chromatin accessibility ( OCT4-dependent ATAC peaks ) and regulatory elements bound by OCT4 ( Figure 3B–D ) . Furthermore , OCT4 and BRG1 occupancy was highly correlated throughout the mouse ESC genome ( Figure 3E and Ho et al . ( 2009 ) , Kidder et al . ( 2009 ) , de Dieuleveult et al . , 2016 ) , supporting the possibility that there may be a functional link between OCT4 binding , BRG1 , and chromatin accessibility . 10 . 7554/eLife . 22631 . 009Figure 3 . The chromatin remodelling enzyme BRG1 is required to create accessible chromatin at OCT4 target sites . ( A ) A Pearson correlation matrix comparing log2 fold change in ATAC-seq signal in OCT4-depleted cells with wild type ESC ChIP-seq signal for nine chromatin remodellers at wild type ATAC hypersensitive peaks ( n = 76 , 642 ) . ( B ) A metaplot of BRG1 ChIP-seq signal at ATAC hypersensitive peaks with ( OCT4-dependent ) or without ( OCT4-independent ) significant reduction in ATAC-seq signal following removal of OCT4 . ( C ) A violin plot quantifying and comparing BRG1 ChIP-seq reads per kilobase per million ( RPKM ) at OCT4-dependent or OCT4-independent ATAC-seq peaks depicted in ( B ) . ( D ) A metaplot of BRG1 ChIP-seq signal at OCT4-bound or OCT4-free ATAC-seq peaks . ( E ) Genome-wide correlation of OCT4 , BRG1 , H3K4me1 and H3K4me3 in 2 kb windows reveals a high degree of co-localization between OCT4 and BRG1 . ( F ) Western blot analysis for the indicated proteins in Brg1fl/fl mouse ESCs before ( UNT ) and after 72 hr tamoxifen ( TAM ) treatment . ( G ) Alkaline phosphatase staining of Brg1fl/fl ESCs before ( UNT ) and after 72 hr TAM treatment . ( H ) A genomic snapshot of BRG1 ChIP-seq and ATAC-seq in Brg1fl/fl ESCs before ( UNT ) and after 72 hr TAM treatment at the distal OCT4 target site downstream of Utf1 ( highlighted in grey ) . The OCT4cond ATAC-seq is included for comparison and reveals a co-dependency on OCT4 and BRG1 for normal chromatin accessibility . ( I ) A heat map of BRG1 ChIP-seq and ATAC-seq at OCT4 target sites ( n = 15920 ) in Brg1fl/fl ESCs before ( UNT ) and after 72 hr TAM treatment . Sites are ranked by loss of ATAC-seq signal following removal of OCT4 , as in Figure 1F , and the OCT4cond ATAC-seq is included for comparison . ( J ) As in ( I ) , changes in nucleosome occupancy before ( UNT ) and after ( TAM ) BRG1 depletion are plotted based on nucleosome signal derived from the NucleoATAC package . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 00910 . 7554/eLife . 22631 . 010Figure 3—figure supplement 1 . OCT4 target sites require BRG1 to maintain chromatin accessibility . ( A ) Genomic snapshots of BRG1 ChIP-seq and ATAC-seq in Brg1fl/fl ESCs before ( UNT ) and after 72 hr TAM treatment at two distal OCT4 target sites ( highlighted in grey ) . The OCT4cond ATAC-seq is included for comparison and reveals dependency on both OCT4 and BRG1 for chromatin accessibility . ( B ) A metaplot of Brg1fl/fl ATAC-seq signal before ( UNT ) and after 72 hr TAM treatment at OCT4 binding sites that rely upon OCT4 for chromatin accessibility ( OCT4-dependent; n = 11557 ) and those that do not ( OCT4-independent; n = 4362 ) , as in Figure 1G . Tn5 control represents transposition of purified genomic DNA to control for potential sequence bias . ( C ) Same as in ( B ) , but profiling the changes in nucleosome occupancy before ( UNT ) and after ( TAM ) BRG1 depletion . Nucleosome signal was generated using the NucleoATAC package . ( D ) Venn diagram overlap of OCT4 targets which significantly lose ATAC-seq signal ( FDR < 0 . 05; fold change > 1 . 5 ) following deletion of OCT4 ( OCT4cond ) or BRG1 ( Brg1fl/fl ) . ( E ) K-means clustering of OCT4 binding sites that significantly lose ATAC-seq signal following deletion of OCT4 ( OCT4-dependent; n = 11557 ) based on changes in Brg1fl/fl ATAC-seq signal . ( F ) Scatterplots of the changes in OCT4cond and Brg1fl/fl ATAC-seq at OCT4 target sites . R2 represents linear regression score , and cor reflects Pearson correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 010 BRG1 , like OCT4 , is essential for maintaining ESC pluripotency ( Ho et al . , 2009; Kidder et al . , 2009; Zhang et al . , 2014 ) , supporting early embryonic development ( Bultman et al . , 2006 ) and improving the efficiency of iPSC reprogramming ( Singhal et al . , 2010 ) . However , the defined molecular mechanisms by which BRG1 and the BAF complex contribute to maintenance of the pluripotent ESC state remain unclear . Based on our observation that BRG1 was enriched at sites that rely on OCT4 binding for chromatin accessibility ( Figure 3A–C ) , we reasoned that OCT4 may require BRG1 to shape chromatin accessibility and gene regulatory capacity of OCT4-bound sites in maintaining the pluripotent state . To address this possibility we performed ATAC-seq on a conditional mouse ESC line in which BRG1 expression can be conditionally ablated following addition of tamoxifen ( Brg1fl/fl ) ( Ho et al . , 2009 ) . Tamoxifen treatment resulted in near complete loss of BRG1 by 72 hr , but importantly , BRG1-depleted cells retained normal ESC morphology , were alkaline phosphatase positive , and expressed wild type levels of OCT4 , SOX2 and NANOG as described previously ( Ho et al . , 2011 ) ( Figure 3F–G ) . When we examined ATAC-seq signal at several OCT4 target sites in the BRG1-depleted ESCs , we observed substantial reductions in chromatin accessibility compared to the untreated control ESCs , and these effects were similar to the reductions observed in the OCT4-depleted ESCs ( Figure 3H; Figure 3—figure supplement 1A ) . When we extended this analysis to all OCT4-bound sites , it was clear that chromatin accessibility was reduced ( Figure 3I; Figure 3—figure supplement 1B ) and nucleosome occupancy increased following BRG1 depletion at the majority of OCT4 target sites ( Figure 3J; Figure 3—figure supplement 1C ) . We then used a stringent threshold ( fold change >1 . 5 and FDR < 0 . 05 ) to identify sites that experienced significant reductions in chromatin accessibility following BRG1 removal . This revealed that 45% of OCT4-bound sites showed significantly reduced ATAC-seq signal and these sites were also highly dependent on OCT4 for their accessibility ( Figure 3—figure supplement 1D ) . However , we suspected that the use of a threshold to identify affected sites might underestimate the extent of BRG1’s contribution . Indeed , when we applied an unbiased clustering approach to examine the relationship between removal of OCT4 and BRG1 in shaping accessibility it was clear that the majority ( 76% ) of OCT4-bound sites that rely on OCT4 for their accessibility were also dependent on BRG1 for their accessibility ( Figure 3—figure supplement 1E ) . Importantly , the loss of either OCT4 or BRG1 appeared to result in similar , although not identical , reductions in chromatin accessibility across nearly all OCT4 peaks ( Figure 3I; Figure 3—figure supplement 1F ) , revealing that both of these factors are required to maintain chromatin accessibility at these loci in mouse ESCs . Together , these observations suggest that BRG1 plays a widespread role in shaping chromatin accessibility at OCT4 target sites in ESCs and raises the interesting possibility that the pioneering-like activity of OCT4 may rely on BRG1 . Our ATAC-seq experiments revealed an essential role for BRG1 in regulating chromatin accessibility at OCT4 target sites in ESCs . One could envisage several possible mechanisms by which BRG1 could cooperate with OCT4 to achieve this . For example , BRG1 could be actively recruited by OCT4 to target sites in order to create accessible chromatin , or it could alternatively function to broadly remodel the genome and in doing so indirectly support access of OCT4 to its target sites . Given that large scale proteomic studies have previously indicated that BRG1 may interact physically with OCT4 ( Pardo et al . , 2010; van den Berg et al . , 2010; Ding et al . , 2012 ) , and BRG1 is enriched at OCT4-bound sites throughout the genome ( Figure 3 ) , we favoured the possibility that OCT4 recruits BRG1 and the associated BAF complex to OCT4 target sites in the genome in order to remodel chromatin and promote chromatin accessibility . In fitting with this possibility , OCT4 , BRG1 and the additional BAF subunit , SS18 , are enriched within the nucleosome-depleted region at OCT4 target sites ( Figure 4A ) . To address whether OCT4 was responsible for recruiting BRG1 and BAF complexes to the distal regulatory sites it binds , we performed ChIP-seq for BRG1 and SS18 in wild type and OCT4-depleted ESCs . Although BRG1 and SS18 protein levels were unaffected following removal of OCT4 ( Figure 1A ) , we observed a dramatic reduction in BRG1 and SS18 binding at OCT4 target sites in the absence of OCT4 ( Figure 4B , C ) . This loss was specific to OCT4-bound regulatory elements as regulatory elements lacking OCT4 were unaffected ( Figure 4D ) . Together , these observations suggested that OCT4 shapes chromatin accessibility at its target sites through the recruitment of BRG1/BAF complex . Indeed , when we compared reductions in chromatin accessibility with reductions in BRG1 and SS18 at OCT4 binding sites , there was a good correlation between the loss of these factors and loss of ATAC-seq signal in OCT4-depleted ESCs ( Figure 4E ) . Interestingly , our observation that OCT4 recruits the BRG1/BAF complex to its target sites in ESCs is in agreement with recent observations that BRG1 is recruited to pluripotency-associated enhancers coincident with the formation of accessible chromatin during iPSC reprogramming ( Chronis et al . , 2017 ) . Together these analyses indicate that OCT4 plays a central role in recruiting BRG1-containing BAF complexes to pluripotency-associated gene regulatory elements and that this is important for creating accessible chromatin at these sites . 10 . 7554/eLife . 22631 . 011Figure 4 . OCT4 is required for normal BRG1 chromatin occupancy at OCT4-bound regulatory elements . ( A ) A high resolution metaplot illustrating nucleosome occupancy , OCT4 , BRG1 and BRG1-associated factor SS18 at OCT4 peaks ( n = 15920; 10 bp windows ) demonstrates that OCT4 and BRG1/BAF signal co-localises to the nucleosome-depleted region at OCT4 peaks . ( B ) Genomic snapshots of BRG1 and SS18 ChIP-seq in OCT4cond before ( UNT ) and after 24 hr doxycycline ( DOX ) treatment reveals loss of BRG1 and SS18 occupancy at distal OCT4 targets ( highlighted in grey ) following OCT4 removal . ( C ) A heatmap of OCT4 peaks ( n = 15920 ) illustrating enrichment of BRG1 and SS18 at OCT4 target sites in wild type cells and subsequent loss of BRG1 following removal of OCT4 . ( D ) A metaplot of BRG1 and SS18 ChIP-seq signal at ATAC hypersensitive sites with ( OCT4-bound ) or without ( OCT4-free ) OCT4 before and after deletion of OCT4 . ( E ) A scatterplot comparing the changes in BRG1 and SS18 occupancy with the changes in chromatin accessibility ( ATAC-seq ) at OCT4 peaks after deletion of OCT4 . R2 represents linear regression score , and cor reflects Pearson correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 011 We have identified a widespread and important role for the chromatin remodeller BRG1 in creating accessibility at gene regulatory elements bound by OCT4 in ESCs . Given that OCT4 recruits BRG1 to shape chromatin accessibility ( Figure 4 ) , we wondered whether BRG1 was also important to sustain the engagement and function of OCT4 , its binding partner SOX2 and the additional pluripotency associated transcription factor NANOG . To examine this interesting possibility , we carried out ChIP-seq for OCT4 , SOX2 and NANOG in the BRG1 conditional cells before and after tamoxifen treatment ( Figure 5 ) . Importantly , wild type OCT4 ChIP-seq signal was highly similar in both Brg1fl/fl ESCs and OCT4cond ESCs ( Figure 5—figure supplement 1 ) . This allowed us to examine the contribution of BRG1 to transcription factor occupancy at bona fide OCT4 binding sites ( Figure 1 ) , focusing on OCT4-bound distal regulatory elements as these were the sites most dependent on OCT4 and BRG1 for normal chromatin accessibility . Strikingly , we observed significant reductions in OCT4 binding at the majority ( 60%; FDR < 0 . 05 and fold change >1 . 5 fold ) of distal OCT4 targets following BRG1 removal ( Figure 5A–C ) , indicating that BRG1 contributes not only to creating accessibility at OCT4 target sites but is also required to sustain normal OCT4 binding in mouse ESCs . Similarly , SOX2 and NANOG binding was also significantly reduced following BRG1 removal in ESCs , although the total number of sites significantly affected was less than for OCT4 ( 16 . 7% and 22 . 9% of distal OCT4 target sites; FDR < 0 . 05 and fold change >1 . 5 fold ) ( Figure 5A–C ) . Importantly , even when we examined the most extremely affected OCT4 target sites ( BRG1-dependent ) , loss of BRG1 did not result in complete loss of transcription factor binding ( Figure 5D ) , yet resulted in an inability of the cells to maintain the nucleosome-depleted state ( Figure 5—figure supplement 2 ) . This suggests that OCT4 , SOX2 and NANOG may still be able to recognise their sequence motifs , but are unable to engage in normal and robust binding without the cooperation of BRG1/BAF remodelling complexes . Ultimately , this is consistent with a pioneering activity of OCT4 being required to initially sample and engage with inaccessible chromatin , potentially through the recognition of partial DNA motifs ( Soufi et al . , 2015 ) . However , OCT4-dependent BRG1/BAF recruitment appears to subsequently be required to functionally mature these previously inaccessible sites such that they can now effectively support robust binding of OCT4 and additional pluripotency-associated transcription factors . 10 . 7554/eLife . 22631 . 012Figure 5 . BRG1 supports pluripotency-associated transcription factor binding to functionally mature distal gene regulatory elements . ( A ) Genomic snapshots illustrating OCT4 , SOX2 and NANOG ChIP-seq signal in Brg1fl/fl ESCs before ( UNT ) and after tamoxifen ( TAM ) treatment for 72 hr . Three examples of distal OCT4 targets are depicted , with affected sites highlighted in grey . ( B ) A heatmap analysis of OCT4 , SOX2 and NANOG ChIP-seq signal at distal OCT4 targets ( n = 11967 ) ranked by their relative change in OCT4 ChIP-seq after deletion of BRG1 . ( C ) Piecharts depicting the significant changes in OCT4 , SOX2 and NANOG ChIP-seq signal at distal OCT4 target sites , identified using the DiffBind package . Changes were deemed significant with FDR < 0 . 05 and a change greater than 1 . 5-fold . ( D ) Metaplot analysis of OCT4 , SOX2 and NANOG binding at distal OCT4 target sites that are the most ( 20% most affected; BRG1-dependent ) and the least ( 20% least affected; BRG1-independent ) dependent on BRG1 for normal OCT4 binding . ( E ) A heatmap illustrating the changes in gene expression ( log2 fold change in RNA-seq ) for genes neighbouring distal OCT4 targets . Genes are sorted according to the change in OCT4 occupancy at neighbouring distal OCT4 target sites following BRG1 removal , as in ( B ) . Gene expression changes for OCT4-depleted ( OCT4cond ) or BRG1-depleted ( Brg1fl/fl ) ESCs relative to wildtype ESCs are shown . ( F ) Quantitation of log2 fold change ( log2FC ) in Brg1fl/fl RNA-seq signal for genes neighbouring distal OCT4 target sites grouped into quintiles based on the change in OCT4 binding at neighbouring distal OCT4 target sites following removal of BRG1 , as in ( B ) . ( G ) Comparison of significant gene expression changes for genes neighbouring distal OCT4 target sites that show the largest ( 20% most affected; BRG1-dependent ) and least ( 20% least affected; BRG1-independent ) reductions in OCT4 ChIP-seq signal following BRG1 removal . Changes were deemed significant with FDR < 0 . 05 and a change greater than 1 . 5-fold , using DESeq2 . ( H ) Quantitation of log2 fold change ( log2FC ) of OCT4 , SOX2 and NANOG ChIP-seq signal at distal OCT4 targets in proximity to OCT4-dependent genes with decreased ( ↓; n = 468 ) , unchanged ( -; n = 639 ) , or increased ( ↑; n = 816 ) RNA-seq signal after deletion of BRG1 ( as identified in Figure 5—figure supplement 3B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 01210 . 7554/eLife . 22631 . 013Figure 5—figure supplement 1 . Comparison of wild type OCT4 ChIP-seq signal between OCT4cond and Brg1fl/fl ESCs . Scatterplot analysis of reads per kilobase per million ( RPKM ) for untreated ( wild type ) OCT4 ChIP-seq in OCT4cond and Brg1fl/fl ESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 01310 . 7554/eLife . 22631 . 014Figure 5—figure supplement 2 . Nucleosome occupancy changes in Brg1fl/fl ESCs at distal OCT4 target sites . Metaplot analysis of nucleosome occupancy signal at distal OCT4 target sites that are the most ( 20% most affected; BRG1-dependent ) and the least ( 20% least affected; BRG1-independent ) dependent on BRG1 for OCT4 binding ( as in Figure 5D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 01410 . 7554/eLife . 22631 . 015Figure 5—figure supplement 3 . Transcriptional regulation of OCT4-dependent target genes by BRG1 . ( A ) Venn diagrams depicting the overlap between genes with significant reductions in RNA-seq following OCT4-depletion in the OCT4cond ESCs ( OCT4-dependent ) and genes with significant decreases or increases in RNA-seq following BRG1-depletion in the Brg1fl/fl ESCs . DESeq2 was used to identify significant changes ( FDR < 0 . 05 and fold change > 1 . 5 fold ) . ( B ) K-means clustering of OCT4-dependent genes based on changes in Brg1fl/fl RNA-seq signal . Clusters were organized into three groups to reflect the direction of gene expression change following removal of BRG1 . ( C ) Boxplot quantitation of log2 fold change in Brg1fl/fl RNA-seq signal following removal of BRG1 for the three groups of OCT4-dependent genes identified in ( B ) illustrates the utility of the clustering approach . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 01510 . 7554/eLife . 22631 . 016Figure 5—figure supplement 4 . Changes in gene expression following depletion of BRG1 in Brg1fl/fl ESCs are associated with altered transcription factor binding . Genomic snapshots of three genes associated with OCT4 binding sites that experience either decreased ( Fgf4 ) , unchanged ( Utf1 ) or increased ( Ankle2 ) RNA-seq signal following depletion of BRG1 in Brg1fl/fl ESCs . OCT4 binding sites are highlighted in grey . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 016 In the absence of BRG1 , the stable binding of pluripotency transcription factors OCT4 , SOX2 and NANOG was disrupted at distal regulatory elements , suggesting that this may affect expression of the pluripotency-associated transcriptional network . Somewhat surprisingly , previous work has not supported a clear correlation between loss of BRG1 and reduced activity of OCT4 target genes ( Ho et al . , 2009 , 2011; Zhang et al . , 2014; Hainer et al . , 2015 ) , despite our observations that BRG1 plays an important role in transcription factor binding at many distal regulatory elements in close proximity to pluripotency-associated genes . However , we also observed that the effect on OCT4 , SOX2 and NANOG binding following BRG1 removal varied in magnitude between individual sites , with some sites actually displaying increases in transcription factor binding ( Figure 5A–D ) . We therefore reasoned that transcriptional effects following loss of BRG1 might not precisely correlate with the expression changes following OCT4 removal but may instead be related to alterations in OCT4 , SOX2 and NANOG binding at individual regulatory sites . To examine this possibility , we carried out nuclear RNA-seq in wild type and BRG1-depleted cells , and compared gene expression changes to those observed following conditional removal of OCT4 ( Figure 5E and Figure 5—figure supplement 3 ) . As described earlier ( Figure 2 ) , genes in close proximity to OCT4 binding sites tended to be down-regulated following loss of OCT4 ( Figure 5E ) . When we simply overlapped the genes that showed significant reductions in gene expression following OCT4 removal ( OCT4-dependent genes; FDR < 0 . 05 and fold change >1 . 5 ) with genes that significantly changed in expression following BRG1 removal , the overlap was low ( Figure 5—figure supplement 3A ) . This was in agreement with previous work that did not identify a clear correlation between gene expression changes following OCT4 and BRG1 removal ( Ho et al . , 2009 , 2011; Zhang et al . , 2014; Hainer et al . , 2015 ) . However , when we examined this relationship in more detail using unbiased clustering of the OCT4-dependent genes based on their expression changes after removal of BRG1 , we identified three separate types of response; genes that showed reduced expression ( 28 . 1% ) , genes whose expression was unchanged ( 35 . 2% ) , and genes that had increased gene expression ( 36 . 7% ) ( Figure 5—figure supplement 3B , C ) . Given that individual distal regulatory elements vary in their requirement for BRG1 in OCT4 , SOX2 and NANOG binding ( Figure 5A–D ) , we examined whether these changes in gene expression corresponded to changes in transcription factor binding at nearby distal regulatory elements . Importantly , this analysis revealed that the expression of genes associated with sites that rely on BRG1 for OCT4 binding tended to be reduced following deletion of BRG1 ( Figure 5E–G , Figure 5—figure supplement 4 ) . In contrast , genes associated with unchanged or increased OCT4 , SOX2 and NANOG binding tended to show increases in expression ( Figure 5E–G , Figure 5—figure supplement 4 ) . Furthermore , distal OCT4 binding sites in close proximity to OCT4/BRG1 dependent genes experienced the largest reductions in transcription factor binding following removal of BRG1 ( Figure 5H ) . Therefore , our new genome-wide analysis suggests that altered transcription factor binding is a major determinant of the gene expression changes in BRG1-depleted ESCs , with the direction and magnitude of expression change being dictated by effects on transcription factor binding at nearby distal regulatory elements . Importantly , these observations support a model where BRG1 is required to stabilize and mature pioneer binding events by OCT4 at inaccessible distal regulatory elements and to transition these sites into functionally active regulatory elements that control the transcription of nearby genes . In contrast , some distal regulatory elements rely less on BRG1 for transcription factor binding and exhibit less pronounced reductions , or even increases , in the expression of their associated genes . Together our new analyses explain why previous studies had failed to identify a simple relationship between gene expression changes following OCT4 and BRG1 removal and demonstrate that reduced expression of a subset of OCT4 target genes results from the inability of OCT4 and other transcription factors to bind their target sites following BRG1 removal . Through studying OCT4 function in iPSC reprogramming it has been proposed that there may be distinct phases of OCT4 binding which are influenced by pre-existing chromatin states in somatic cells ( Soufi et al . , 2012; Buganim et al . , 2013; Chen et al . , 2016 ) . For example , OCT4 occupies a subset of sites during the early stages of reprogramming which are modified by histone H3 lysine 4 dimethylation or histone H3 lysine 27 acetylation and associated with more accessible chromatin . This is followed by OCT4 binding at sites lacking pre-existing chromatin modifications that are thought to become accessible during the later stages of reprogramming when pluripotency is established . This suggests that OCT4 binding may occur via different mechanisms to support regulatory element function during reprogramming and development . In light of our observation in ESCs that some OCT4 binding does not require BRG1 , we wanted to examine whether the requirement for BRG1 at these sites reflected the dynamics of gene expression during cellular reprogramming and early development . We therefore analysed gene expression during the reprogramming of mouse fibroblasts into iPSCs via OCT4/SOX2/KLF4/MYC from two independent studies ( Chen et al . , 2016; Cieply et al . , 2016 ) . Specifically , we examined genes that required OCT4 for their expression in ESCs ( OCT4-dependent ) and separated these into genes associated with BRG1-independent or BRG1-dependent OCT4 binding at neighbouring distal regulatory elements in ESCs ( see Figure 5 ) . We then compared the timing of their expression during the iPSC reprogramming process . This revealed that genes associated with BRG1-independent OCT4 binding were activated early during reprogramming and BRG1-dependent sites later ( Figure 6A ) . This suggests that OCT4 expression in somatic cells leads to a more immediate transcriptional response from genes associated with BRG1-independent OCT4 binding , presumably due to pre-existing chromatin accessibility at these loci ( Chen et al . , 2016 ) . In contrast , genes that require BRG1 for OCT4 binding were expressed later during reprogramming , perhaps because OCT4 requires BRG1 at these sites to remodel chromatin and mature the function of these regulatory elements . In fitting with this possibility , BRG1-dependent OCT4 binding sites were also associated with genes expressed at later developmental stages during early mouse development ( Figure 6B ) , suggesting that chromatin state in the early embryo may also shape how these OCT4 target sites are used during development . To explore this possibility , we examined the activation of BRG1-independent and BRG1-dependent OCT4 targets in the early mouse embryo using chromatin accessibility as a proxy for OCT4 binding and regulatory activity ( Lu et al . , 2016; Wu et al . , 2016 ) . Interestingly , this demonstrated that BRG1-dependent OCT4 target sites became accessible at later stages of embryonic development ( Figure 6C , Figure 6—figure supplement 1 ) . This suggests that developmental transitions may require both OCT4-dependent chromatin binding and BRG1-dependent remodelling activities to overcome the activation barrier set by chromatin . In contrast , the pre-existing chromatin state at a subset of OCT4 target sites ( BRG1-independent ) may allow them to be activated more rapidly . Importantly , these observations suggest that chromatin structure likely plays an important role in regulating how OCT4 engages with and functions in the genome not only in mouse ESCs but also during reprogramming and development . 10 . 7554/eLife . 22631 . 017Figure 6 . BRG1-dependency reveals distinct modes of OCT4 function of distal regulatory elements during reprogramming and development . ( A ) A time course of gene expression changes ( log2 fold change ) during OKSM-mediated reprogramming of mouse embryonic fibroblasts to iPSCs ( Chen et al . , 2016; Cieply et al . , 2016 ) . The expression changes of OCT4-dependent genes neighbouring the 20% of distal OCT4 targets that were most dependent upon BRG1 for normal OCT4 binding ( BRG1-dependent ) or the 20% of distal OCT4 targets that were least dependent ( BRG1-independent ) following BRG1 removal were quantified and visualised as a smoothed trendline ±95% confidence interval . ( B ) Genomic Regions Enrichment of Annotations Tool ( GREAT ) annotation of BRG1-dependent and BRG1-independent distal OCT4 targets . Plotted are the enrichment ( -log10 ( P Value ) ) against the MGI Expression profile , or Thieler development stage , for genes neighbouring distal OCT4 targets . BRG1-independent sites are strongly enriched for gene expression profiles consistent with very early embryonic stages . ( C ) Metaplot profiles of ATAC-seq signal during early mouse embryonic development ( Wu et al . , 2016 ) at BRG1-independent ( upper ) and BRG1-dependent ( lower ) distal OCT4 target sites . BRG1-independent sites gain accessibility earlier than BRG1-dependent sites . ( D ) A quantitation of ATAC-seq read density at distal OCT4 targets depicted in ( C ) during early embryonic development plotted as a smoothed trendline ±95% confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 01710 . 7554/eLife . 22631 . 018Figure 6—figure supplement 1 . BRG1-dependent and BRG1-independent OCT4 binding at distal regulatory elements is associated with distinct developmental timing of chromatin accessibility . ( A ) Metaplot profiles of DNase-seq signal during early embryonic development ( Lu et al . , 2016 ) at BRG1-independent and BRG1-dependent distal OCT4 targets identified in Brg1fl/fl ESCs . BRG1-independent sites gain accessibility earlier than BRG1-dependent sites . ( B ) A quantitation of DNase-seq read density at distal OCT4 targets depicted in ( A ) during early embryonic development depicted as a smoothed trendline ± 95% confidence interval . DOI: http://dx . doi . org/10 . 7554/eLife . 22631 . 018 Although the capacity of pioneer transcription factors to bind their sequence motifs in chromatinized DNA has been studied in detail , the mechanisms that support how individual pioneer transcription factors function to create accessible chromatin and shape regulatory element function remain poorly defined . Through studying the pioneer transcription factor OCT4 , we establish that OCT4 binds to sites in mouse ESCs that would otherwise be inaccessible to shape chromatin accessibility and transcription factor binding ( Figures 1 and 2 ) . This suggests that the pioneering activities of OCT4 are required not only during reprogramming , but also in the established pluripotent state . Interestingly , chromatin accessibility formed at OCT4 binding sites relies on the chromatin remodelling factor BRG1 ( Figure 3 ) which is recruited to these sites by OCT4 ( Figure 4 ) . The occupancy of BRG1 is then required to support efficient OCT4 , SOX2 and NANOG binding and normal expression of the pluripotency-associated transcriptome ( Figure 5 ) . Importantly , this reliance on BRG1 reflects OCT4 binding dynamics during cellular reprogramming and early mouse development ( Figure 6 ) . Together these observations reveal a distinct requirement for the chromatin remodelling factor BRG1 in shaping the activity of the pioneer transcription factor OCT4 and regulating the pluripotency network in embryonic stem cells . Pioneer transcription factors play very defined roles in shaping gene expression in response to cellular reprogramming ( Soufi et al . , 2012; Raposo et al . , 2015; Boller et al . , 2016; Chen et al . , 2016 ) , developmental stimuli , and environmental cues ( Hu et al . , 2011; Wu et al . , 2011; Ballaré et al . , 2013; Schulz et al . , 2015 ) . While considerable effort has focused on characterizing the chromatin binding activity of pioneer transcription factors , both in vitro and in vivo , the mechanisms by which individual pioneer transcription factors shape chromatin structure and support further transcription factor engagement in vivo have remained poorly understood . Our examination of OCT4 binding now provides a potential rationalization for how this pioneer transcription factor functions to form gene regulatory elements in mouse ESCs . A requirement of this process is OCT4's ability to engage with sequences in inaccessible chromatin presumably through its capacity to bind nucleosomal DNA ( Figure 1; Soufi et al . , 2012 , 2015 ) , suggesting that OCT4 may be able to dynamically sample its target sites throughout the genome . However , cooperation with BRG1 appears to be required for more stable OCT4 binding ( Figure 5 ) . This could be achieved through the chromatin remodelling activity of BRG1 creating transiently accessible DNA that facilitates further OCT4 binding and ultimately more stable cooperative binding with other transcription factors like SOX2 and NANOG . This general model is consistent with previous experiments that showed enhanced OCT4 binding during OCT4-SOX2-KLF4 reprogramming when BRG1 was included in the reprogramming cocktail ( Singhal et al . , 2010 ) and is in agreement with our observation that BRG1 is required for normal chromatin accessibility and OCT4 binding at otherwise inaccessible and inactive regulatory elements . The capacity of OCT4 to recognise its target motifs in inaccessible chromatin may allow it to dynamically sample its complement of binding sites in the genome . However , like many transcription factors , OCT4 only binds stably to a subset of these sites . Our genome-wide analyses indicate that BRG1 and transcription factor co-binding events appear to be essential in stabilizing OCT4 binding and the functional maturation of OCT4-dependent regulatory elements in ESCs . Based on these observations , one would predict that co-binding transcription factors might play a central role in shaping where OCT4 stably engages with the genome in different cell types . In fitting with this possibility , when ESCs are transitioned into epiblast-like cells ( EpiLCs ) , OCT4 stably associates with a new set of previously inaccessible sites in a manner that appears to rely on co-binding of the EpiLC-specific transcription factor OTX2 ( Buecker et al . , 2014; Yang et al . , 2014 ) . In the context of these and other observations , it is tempting to speculate that OCT4 , through its association with BRG1 , is exploited as a dynamic pioneering and chromatin remodelling module by distinct co-binding transcription factors to support the formation of cell type-specific regulatory elements during developmental transitions and reprogramming . In agreement with these general ideas , OCT4 appears to be the only Yamanaka transcription factor that cannot be substituted for in iPSC reprogramming experiments , suggesting its pioneering activity is fundamental in establishing and maintaining the pluripotency network in concert with SOX2 , KLF4 and c-Myc ( Nakagawa et al . , 2008; Sterneckert et al . , 2012 ) , perhaps due to its cooperation with BRG1 ( Singhal et al . , 2010; Esch et al . , 2013 ) . This central requirement in reprogramming is also consistent with our observations that OCT4 plays an essential and BRG1-dependent role in shaping the binding of other pluripotency-associated factors at thousands of distal regulatory elements in ESCs . One of the central features of pioneer transcription factors is their ability to alter local chromatin structure , and this is closely linked to their capacity to support further transcription factor binding and enable the functional maturation of distal regulatory elements . Yet , in most instances the defined molecular mechanisms that support these activities have remained undetermined . While some pioneer factors , such as FoxA1 , can stably bind their target sites and directly alter nucleosomal structure through steric disruption of histone:DNA contacts ( Cirillo et al . , 2002; Hatta and Cirillo , 2007; Iwafuchi-Doi et al . , 2016 ) , the extent to which other pioneer factors might exploit such a mechanism has yet to be fully addressed . In the case of OCT4 , our observations establish that its capacity to support transcription factor binding and chromatin accessibility requires the chromatin remodeller BRG1 . This suggests that unlike FoxA1 , OCT4 does not have an intrinsic chromatin opening activity , but requires cooperation with BRG1 to achieve this . Indeed , our work on OCT4 is generally consistent with and supported by recent work studying the pioneer factor GATA3 which appears to have similar dependency on BRG1 in creating accessible chromatin ( Takaku et al . , 2016 ) . Furthermore , chromatin remodelling activity has been widely implicated in shaping nucleosome positioning and chromatin accessibility at target sites bound by other pioneer transcription factors ( Hu et al . , 2011; Sanalkumar et al . , 2014; Ceballos-Chávez et al . , 2015; Hainer and Fazzio , 2015 ) . Together , these observations suggest that many pioneer transcription factors may rely on chromatin remodelling as a key step in supporting transcription factor binding and/or co-binding events in a manner broadly consistent with the previously proposed assisted loading model for transcription factor binding ( Voss and Hager , 2014; Swinstead et al . , 2016 ) . In the context of this model , the dynamic binding , release and re-binding of transcription factors would help to maintain and stabilise transcription factor binding events , but critically , chromatin remodellers appear to be important in many cases to achieve this ( reviewed recently in Swinstead et al . , 2016 ) . Clearly more work is required to understand the extent to which pioneer transcription factors use chromatin remodellers to support their pioneering activities . However , in the case of the developmental pioneer transcription factors OCT4 and GATA3 , chromatin remodellers act as key component necessary for the formation and function of gene regulatory elements . These emerging observations suggest that chromatin remodelling could function as a central feature of pioneer transcription factor activity during the formation and maintenance of cell type-specific transcriptional programs during development and reprogramming . Mouse embryonic stem cell ( ESCs ) containing a doxycycline-sensitive OCT4 transgene ( ZHBTC4; referred to here as OCT4cond [Niwa et al . , 2000] ) were grown on gelatin-coated plates in DMEM ( Gibco , Carlsbad , CA ) supplemented with 15% FBS , 10 ng/mL leukemia-inhibitory factor , penicillin/streptomycin , beta-mercaptoethanol , L-glutamine and non-essential amino-acids . OCT4cond cells were treated with 1 µg/mL doxycycline for 24 hr to ablate OCT4 expression , which was verified by Western blotting . Brg1fl/fl ESCs were previously described ( Ho et al . , 2011 ) and maintained in DMEM KnockOut supplemented with 10% FBS and 5% KnockOut Serum Replacement ( Life Technologies , Carlsbad , CA ) , plus additional factors described for OCT4cond ESCs above . Brg1fl/fl ESCs were treated with 4-hydroxytamoxifen for 72 hr to ablate BRG1 protein levels , which was verified by Western blotting . Cell lines were routinely tested and confirmed to be mycoplasma-free . Alkaline phosphatase staining was performed by incubating cells with freshly prepared AP buffer ( 0 . 4 mg/mL naphthol phosphate N-5000 ( Sigma , St Louis , MO ) , 1 mg/mL Fast Violet B Salt F-161 ( Sigma ) , 100 mM Tris-HCl ( pH 9 . 0 ) , 100 mM NaCl , 5 mM MgCl2 ) . Chromatin accessibility was assayed using an adaptation of the assay for transposase accessible-chromatin ( ATAC ) -seq ( Buenrostro et al . , 2013 ) . Briefly , 5 × 106 cells were harvested , washed with PBS and nuclei were isolated in 1 mL HS Lysis buffer ( 50 mM KCl , 10 mM MgSO4 . 7H20 , 5 mM HEPES , 0 . 05% NP40 [IGEPAL CA630] ) , 1 mM PMSF , 3 mM DTT ) for 1 min at room temperature . Nuclei were centrifuged at 1000×g for 5 min at 4°C , followed by a total of three washes with ice-cold RSB buffer ( 10 mM NaCl , 10 mM Tris ( pH 7 . 4 ) , 3 mM MgCl2 ) , to remove as much cytoplasmic and mitochondrial material as possible . Nuclei were then counted , and 5 × 104 nuclei were resuspended in Tn5 reaction buffer ( 10 mM TAPS , 5 mM MgCl2 , 10% dimethylformamide ) and 2 µl of Tn5 transposase ( 25 µM ) made in house according to the previously described protocol ( Picelli et al . , 2014 ) . Nuclei were then incubated for 30 min at 37°C , before isolation and purification of tagmented DNA using QiaQuick MinElute columns ( Qiagen , Germany ) . To control for sequence bias of the Tn5 transposase , a Tn5 digestion control was performed by tagmenting ESC genomic DNA with Tn5 for 30 min at 55°C . ATAC-seq libraries were prepared by PCR amplification using custom made Illumina barcodes previously described ( Buenrostro et al . , 2013 ) and the NEBNext High-Fidelity 2X PCR Master Mix ( NEB , Ipswich , MA ) with 8–10 cycles . Libraries were purified with two rounds of Agencourt AMPure XP bead cleanup ( 1 . 5X beads:sample; Beckman Coulter , Brea , CA ) , followed by quantification by qPCR using SensiMix SYBR ( Bioline , UK ) and KAPA Library Quantification DNA standards ( KAPA Biosystems , Wilmington , MA ) . ATAC-seq libraries were sequenced on Illumina NextSeq500 using 80 bp paired-end reads in biological triplicate . Chromatin immunoprecipitation ( ChIP ) was performed as described previously ( Farcas et al . , 2012 ) , with minor modifications . Cells were fixed for 1 hr in 2 mM DSG and 12 . 5 min in 1% formaldehyde . Reactions were quenched by the addition of glycine to a final concentration of 125 µM . After cell lysis and chromatin extraction , chromatin was sonicated using a BioRuptor sonicator ( Diagenode , Belgium ) , followed by centrifugation at 16 , 000×g for 20 min at 4°C , and used fresh or stored at −80°C . Chromatin was quantified by denaturing chromatin 1:10 in 0 . 1M NaOH and measuring DNA concentration by NanoDrop . 150 µg chromatin/IP was diluted ten-fold in ChIP dilution buffer ( 1% Triton-X100 , 1 mM EDTA , 20 mM Tris-HCl ( pH 8 ) , 150 mM NaCl ) prior to pre-clearing with prepared protein A magnetic Dynabeads ( Invitrogen , Carlsbad , CA ) which had been blocked for 1 hr with 0 . 2 mg/mL BSA and 50 µg/mL yeast tRNA . Chromatin samples were then incubated overnight with relevant antibodies at 4°C with rotation . Antibodies used for ChIP experiments were anti-OCT4A ( Cell Signaling Technology ( CST , Danvers , MA ) , #5677 ) , anti-SOX2 ( CST , #23064 ) , anti-Nanog ( CST , #8822 ) , anti-SS18 ( CST , #21792 ) and anti-BRG1 ( abcam ( UK ) , ab110641 ) . Antibody-bound chromatin was isolated on protein A magnetic Dynabeads . ChIP washes were performed with low salt buffer ( 0 . 1% SDS , 1% Triton , 2 mM EDTA , 20 mM Tris-HCl ( pH 8 . 1 ) , 150 mM NaCl ) , high salt buffer ( 0 . 1% SDS , 1% Triton , 2 mM EDTA , 20 mM Tris-HCl ( pH 8 . 1 ) , 500 mM NaCl ) , LiCl buffer ( 0 . 25M LiCl , 1% NP40 , 1% sodium deoxycholate , 1 mM EDTA , 10 mM Tris-HCl ( pH 8 . 1 ) ) and TE buffer ( x2 washes ) ( 10 mM Tris-HCl ( pH 8 . 0 ) , 1 mM EDTA ) . To prepare ChIP-seq material , ChIP DNA was eluted using 1% SDS and 100 mM NaHCO3 , and cross-links reversed at 65°C in the presence of 200 mM NaCl . Samples were then treated with RNase and proteinase K before being purified with ChIP DNA Clean and Concentrator kit ( Zymo , Irvine , CA ) . ChIP-seq libraries were prepared using the NEBNext Ultra DNA Library Prep Kit with NEBNext Dual Indices , and sequenced as 38 bp paired-end reads on Illumina NextSeq500 platform . All ChIP-seq experiments were carried out in biological triplicate . To isolate nuclear RNA , cells were subjected to nuclei isolation described for ATAC-seq . Nuclei were then resuspended in TriZOL reagent ( ThermoScientific , Waltham , MA ) and RNA was extracted according to the manufacturer’s protocol . Nuclear RNA was treated with the TURBO DNA-free Kit ( ThermoScientific ) and depleted for rRNA using the NEBNext rRNA Depletion kit and protocol ( NEB ) . RNA-seq libraries were prepared using the NEBNext Ultra Directional RNA-seq kit ( NEB ) and libraries were sequenced on the Illumina NextSeq500 with 80 bp paired-end reads in biological triplicate . For ATAC-seq and ChIP-seq , paired-end reads were aligned to the mouse mm10 genome using bowtie2 ( Langmead and Salzberg , 2012 ) with the ‘--no-mixed’ and ‘--no-discordant’ options . Non-uniquely mapping reads and reads mapping to a custom ‘blacklist’ of artificially high regions of the genome were discarded . For RNA-seq , reads were initially aligned using bowtie2 against the rRNA genomic sequence ( GenBank: BK000964 . 3 ) to filter out rRNA fragments , prior to alignment against the mm10 genome using the STAR RNA-seq aligner ( Dobin et al . , 2013 ) . To improve mapping of nascent , intronic sequences , reads which failed to map using STAR were aligned against the genome using bowtie2 . PCR duplicates were removed using SAMtools ( Li et al . , 2009 ) . Biological replicates were randomly downsampled to contain the same number of reads for each individual replicate , and merged to create a representative genome track using DANPOS2 ( Chen et al . , 2013 ) which was visualised using the UCSC Genome Browser . Peakcalling analyses were performed using the DANPOS2 dpeak function on untreated and treated samples in biological triplicate with matched input . Merged ATAC-seq datasets were used to extract signal corresponding to nucleosome occupancy information with NucleoATAC ( Schep et al . , 2015 ) using a cross correlation model for all regulatory elements ( ATAC hypersensitive sites ) in each cell line . Significant changes in ATAC-seq or ChIP-seq datasets were identified using the DiffBind package ( Stark and Brown , 2011 ) , while for RNA-seq DESeq2 was used with a custom-built , non-redundant mm10 gene set ( Love et al . , 2014 ) . Briefly , mm10 refGene genes were filtered on size ( >200 bp ) , gene body and TSS mappability , unique TSS and TTS , in order to remove poorly mappable and highly similar transcripts . For both DiffBind and DESeq2 , a FDR < 0 . 05 and a fold change >1 . 5 fold was deemed to be a significant change . For distal OCT4 intervals , gene expression changes for the nearest TSS were considered . K-means clustering to identify OCT4 and BRG1 co-dependency of chromatin accessibility or gene expression was performed using the kmeans function in R . Clusters with similar trends ( i . e . decrease , no change , or increase ) were then grouped together for subsequent analysis . Changes in ATAC-seq or ChIP-seq were visualised using heatmaps or metaplots produced using HOMER2 ( Heinz et al . , 2010 ) , with heatmaps made using Java TreeView ( Saldanha , 2004 ) . Log2 fold change values were visualised using R ( v 3 . 2 . 1 ) , with scatterplots coloured by density using stat_density2d . Regression and correlation analyses were also performed in R using standard linear models and Pearson correlation respectively . Log2 fold change or reads per kilobase per million ( RPKM ) values were compared between different classes of transcription factor binding sites either by visualising gam smoothed trendlines with 95% confidence intervals or using the Wilcoxon signed-rank test . OCT4 peaks were identified in the OCT4cond biological triplicate OCT4 ChIP-seq data using the DANPOS2 dpeak function and only peaks with decreased OCT4 ChIP-seq signal were taken for further analysis ( n = 15920 ) . OCT4 motif enrichment analysis was performed using the MEME suite ( Bailey et al . , 2009 ) . Briefly , Analysis of Motif Enrichment ( AME ) for canonical motifs was performed in parallel to de novo motif identification with Discriminative Regular Expression Motif Elicitation ( DREME ) using the central 200 bp of OCT4 peaks . Putative de novo motifs were further subjected to CentriMo analysis to identify motifs that were enriched for the centre of OCT4 peaks . OCT4 peaks were annotated as putative promoters or putative distal regulatory elements in a manner similar to that described previously ( Hay et al . , 2016 ) , using the relative and absolute coverage of H3K4me3 ( a promoter-associated modification; Yue et al . , 2014 ) and H3K4me1 ( associated with distal regulatory elements; Whyte et al . , 2012 ) . Transcription factor peaks with different characteristics were analysed using the GREAT package ( McLean et al . , 2010 ) , in particular to extract information regarding developmental expression timing from MGI Gene eXpression Database . HOMER2 was used to identify the nearest transcription start sites ( TSS ) of genes and to perform gene ontology ( GO ) analysis for differentially regulated genes . Comparison of different remodelling complexes was performed by calculating RPKM for chromatin remodeller ChIP-seq in mouse ESCs across all ATAC peaks ( n = 76 , 642 ) and determining the Pearson correlation with the log2 fold change of OCT4cond ATAC-seq . Genome-wide correlation of BRG1 , OCT4 , H3K4me3 and H3K4me1 was generated using the bamCorrelate function of deepTools ( Ramírez et al . , 2014 ) . ATAC-seq , ChIP-seq and RNA-seq data from the present study are available for download at GSE87822 . Previously published datasets used for analysis include mouse ESC H3K4me1 ( GSE27844; Whyte et al . , 2012 ) and H3K4me3 ChIP-seq ( GSE49847; Yue et al . , 2014 ) , ENCODE DNase-seq ( GSE37074; Yue et al . , 2014 ) , ESC chromatin remodeller ChIP-seq ( GSE49137 , GSE64825; Wang et al . , 2014; de Dieuleveult et al . , 2016 ) , iPSC expression data ( GSE67462 , GSE70022; Chen et al . , 2016; Cieply et al . , 2016 ) , early mouse embryo ATAC-seq ( GSE66581; Wu et al . , 2016 ) and DNase-seq ( GSE76642; Lu et al . , 2016 ) .	All cells in your body contain the same genetic information in the form of genes encoded within DNA . Yet , cells use this information in different ways so that the activities of individual genes within that DNA can vary from cell to cell . This allows identical cells to become different to each other and to adapt to changing circumstances . A group of proteins called transcription factors control the activity of certain genes by binding to specific sites on DNA . However , this isn’t a straightforward process because DNA in human and other animal cells is usually associated with structures called nucleosomes that can block access to the DNA . Pioneer transcription factors , such as OCT4 , are a specific group of transcription factors that can attach to DNA in spite of the nucleosomes , but it’s not clear how this is possible . Once pioneer transcription factors attach to DNA they can help other transcription factors to bind alongside them . King et al . studied OCT4 in stem cells from mouse embryos to investigate how it is able to act as a pioneer transcription factor and control gene activity . The experiments show that several other transcription factors lose the ability to bind to DNA when OCT4 is absent . This leads to widespread changes in gene activity in the cells , which seems to be due to other transcription factors being unable to get past the nucleosomes to attach to the DNA . Further experiments showed that OCT4 needs a protein called BRG1 in order to act as a pioneer transcription factor . BRG1 is an enzyme that is able to move and remove ( remodel ) nucleosomes attached to DNA , suggesting that normal transcription factor binding requires this activity . The next challenge is to investigate whether BRG1 , or similar enzymes , are also needed by other pioneer transcription factors that are required for normal gene activity and cell identity . This will be important because many enzymes that remodel nucleosomes are disrupted in human diseases like cancer where cells lose their normal identity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "chromosomes", "and", "gene", "expression" ]	2017	The pioneer factor OCT4 requires the chromatin remodeller BRG1 to support gene regulatory element function in mouse embryonic stem cells
The simultaneous imaging and manipulating of neural activity could enable the functional dissection of neural circuits . Here we have combined two-photon optogenetics with simultaneous volumetric two-photon calcium imaging to measure and manipulate neural activity in mouse neocortex in vivo in three-dimensions ( 3D ) with cellular resolution . Using a hybrid holographic approach , we simultaneously photostimulate more than 80 neurons over 150 μm in depth in layer 2/3 of the mouse visual cortex , while simultaneously imaging the activity of the surrounding neurons . We validate the usefulness of the method by photoactivating in 3D selected groups of interneurons , suppressing the response of nearby pyramidal neurons to visual stimuli in awake animals . Our all-optical approach could be used as a general platform to read and write neuronal activity . The precise monitoring and control of neuronal activity may be an invaluable tool to decipher the function of neuronal circuits . For reading out neuronal activity in vivo , the combination of calcium imaging of neuronal populations ( Yuste and Katz , 1991 ) with two-photon microscopy ( Denk et al . , 1990 ) , has proved its utility because of its high selectivity , good signal-to-noise ratio , and depth penetration in scattering tissues ( Yuste and Denk , 1995; Cossart et al . , 2003; Zipfel et al . , 2003; Helmchen and Denk , 2005; Svoboda and Yasuda , 2006; Ji et al . , 2016; Yang and Yuste , 2017 ) . Moreover , two-photon imaging can be combined with two-photon optochemistry ( Nikolenko et al . , 2007; Dal Maschio et al . , 2010 ) or two-photon optogenetics ( Packer et al . , 2012; Rickgauer et al . , 2014; Packer et al . , 2015; Emiliani et al . , 2015; Carrillo-Reid et al . , 2016 ) to allow simultaneous readout and manipulation of neural activity with cellular resolution . But so far , the combinations of these optical methods into an all-optical approach have been largely restricted to two-dimensional ( 2D ) planes ( Nikolenko et al . , 2007; Dal Maschio et al . , 2010; Rickgauer et al . , 2014; Packer et al . , 2015; Carrillo-Reid et al . , 2016 ) . At the same time , neural circuits are three dimensional , and neuronal sub-populations are distributed throughout their volume . Therefore , extending these methods to three dimensions ( 3D ) appears essential to enable systematic studies of microcircuit computation and processing . Here we employed wavefront shaping strategies with a customized dual-beam two-photon microscope to simultaneously perform volumetric calcium imaging and 3D patterned photostimulations in mouse cortex in vivo . For phostostimulation , we adopted a hybrid strategy that combines 3D holograms and galvanometer driven spiral scans . Furthermore , we used a pulse-amplified low-repetition-rate ( 200 kHz ~ 1 MHz ) laser , which significantly reduces the average laser power required for photoactivation , and minimizes thermal effects and imaging artifacts . With this system , we photostimulated large groups of cells simultaneously in layer 2/3 of primary visual cortex ( V1 ) in awake mice ( >80 cells distributed within a 480 × 480 × 150 μm3 imaged volume ) , while simultaneously imaging the activity of the surrounding neurons . Compared with other 3D all-optical approaches ( Dal Maschio et al . , 2017; Mardinly et al . , 2017 ) , which used scanless holographic photostimulation , our hybrid approach requires less laser power to stimulate per cell , and can thus simultaneously photostimulate more cells for a given fixed power budget . This all-optical method is useful to analyze the function of neural circuits in 3D , such as studying cell connectivity , ensemble organization , information processing , or excitatory and inhibitory balance . As a demonstration , we photostimulated groups of pyramidal cells in 3D with high specificity , and also targeted a selective population of interneurons in V1 in awake mice , finding that stimulating the interneurons reduced the response of pyramidal cells to visual simuli . We built a holographic microscope with two independent two-photon lasers , one for imaging and the other for photostimulation ( Figure 1A ) . Each laser beam’s axial focal depth could be controlled without mechanical motion of the objective , yielding maximum flexibility while reducing perturbations to the animal . On the imaging path , we coupled a wavelength-tunable Ti:Sapphire laser through an electrically tunable lens ( ETL , EL-10-30-C-NIR-LD-MV , Optotune AG ) ( Grewe et al . , 2011 ) followed by a resonant scanner for high speed volumetric imaging . The ETL , as configured , provided an adjustable axial focus shift up to 90 μm below and 200 μm above the objective’s nominal focal plane . On the photostimulation path , we used a low-repetition-rate ultrafast laser coupled to a spatial light modulator ( SLM , HSP512-1064 , Meadowlark Optics ) to shape the wavefront , allowing flexible 3D beam splitting that simultaneously targets the user defined positions in the sample ( Figure 1B–1E ) . The axial and lateral targeting error was 0 . 59 ± 0 . 54 μm and 0 . 82 ± 0 . 65 μm , respectively , across a 3D field of view ( FOV ) of 240 × 240 × 300 μm3 ( Figure 1—figure supplement 1; Materials and Methods ) . The SLM path was coupled through a pair of standard galvanometers that can allow for fast extension of the targeting FOV beyond that nominal addressable SLM-only range ( Yang et al . , 2015 ) . For optogenetics experiments , we actuated this pair of galvanometric mirrors to scan the beamlets in a spiral over the cell bodies of the targeted neuron ( see Figure 1E for an exemplary 3D pattern with 100 targets on an autofluorescent plastic slide ) . We term this a ‘hybrid’ approach , as it combined holography with mechanical scanning , as opposed to purely holographic approach . For in vivo experiments , we imaged green fluorescence from the genetically encoded calcium indicator GCaMP6s or GCaMP6f ( Chen et al . , 2013 ) and photostimulated a red-shifted opsin , C1V1-mCherry ( Yizhar et al . , 2011 ) . With switchable kinematic mirrors and dichroic mirrors , the lasers could be easily redirected to whichever path , and thus the system could also be utilized for red fluorophores and blue opsins . We co-expressed GCaMP6s or GCaMP6f ( Chen et al . , 2013 ) and C1V1-p2A ( Yizhar et al . , 2011 ) in mouse V1 ( Figure 1F ) , and excited them with 940 nm and 1040 nm light , respectively . The separation of their excitation spectrum allowed for minimal cross-talk between the imaging and photostimulation paths ( Discussion ) . C1V1-expressed cells were identified through a co-expressed mCherry fluorophore . Single spikes can be evoked with very low average laser power ( ~2 . 25 mW with 20 ms spiral , or ~ 4 . 5 mW with 10 ms spiral , 1 MHz pulse train , layer 2/3 in vivo , Figure 1G ) , latency and jitter ( 17 . 0 ± 4 . 2/8 . 5 ± 1 . 6 ms latency , and 2 . 0±1 . 5/0 . 5±0 . 3 ms jitter for the two conditions , Figure 1—figure supplement 2; jitter defined as the standard deviation of the latency ) . With a higher power ( 10 ~ 20 mW ) , neural activity could also be evoked with photostimulation duration as short as 1 ms ( Figure 2 ) . Compared with alternative scanless strategy like temporal focusing ( Rickgauer et al . , 2014; Mardinly et al . , 2017; Hernandez et al . , 2016; Pégard et al . , 2017 ) or pure holographic approaches ( Dal Maschio et al . , 2017 ) , where the laser power is distributed across the whole cell body of each targeted neuron , our hybrid approach is simple , accommodates large numbers of simultaneous targets , and appears to have a better power budget for large population photostimulation in general . To test this , we compared the required power budget for hybrid approach and the scanless ( pure holographic ) approach at different photostimulation durations ( 20 ms , 10 ms , 5 ms and 1 ms ) . On our system , when photostimulation duration was above 5 ms , the hybrid approach required about half of the laser power than the scanless approach to evoke similar response in the neuron; at 1 ms photostimulation duration , the hybrid approach shows a trend with smaller power budget ( but not significant , p=0 . 17 using one-way ANOVA test ) than the scanless approach ( Figure 2 , Figure 2—figure supplement 1 ) . One reason for this difference is that the scanless approach employs a spatial multiplexed strategy , where the two-photon light is spatially distributed across the entire cell body; to maintain the two-photon excitation efficiency ( squared-intensity ) within its coverage area , a larger total power is typically required . The hybrid approach , on the other hand , is a combination of spatial ( across different cells ) and temporal ( within individual cell ) multiplexed strategy . While optimal strategy will depend on opsin photophysics , the opsin typically has a long opsin decay constant ( Mattis et al . , 2011 ) ( 10 s of millisecond ) and this favors the hybrid approach because the opsin channels can stay open during the entire ( multiple ) spiral scans . But at very short duration , the limited number of laser pulses per unit area may contribute to an efficiency drop of the hybrid approach versus scanless approach . We tested our 3D all-optical system by targeting and photoactivating selected groups of pyramidal cells throughout three axial depths of layer 2/3 of V1 in anesthetized mice , while simultaneously monitoring neuronal activity in those three planes ( 240 × 240 μm2 FOV for each plane ) at 6 . 67 vol/s . Neurons were photoactivated one at a time ( Figure 3—figure supplement 1 ) , or as groups/ensembles ( M neurons simultaneously , M = 3 ~ 27 , Figure 3 ) and the majority of the targeted cells ( 86 ± 6% , Materials and Methods ) showed clear calcium transients in response to the photostimulation ( Figure 3C–E ) . We further investigated the reliability of the photoactivation and also its influence on the activation of non-targeted cells – that is , cells within the FOV not explicitly targeted with a beamlet . We performed 8 ~ 11 trials for each stimulation pattern . Cells not responding to photostimulation under any condition were excluded in this analysis ( see Materials and Methods ) . We characterize the response rate at the individual cell ( Figure 3F ) and the ensemble level ( Figure 3G ) . The former characterizes the response rate of individual targeted cells in any stimulation pattern , and the latter characterizes the percentage of responsive cells within a targeted ensemble ( defined here as ensemble response rate ) . As the simulatenously stimulated neurons number M increased , the response rate for both individual cells and ensembles remained high ( both is 82 ± 9% , over all seven stimulation conditions ) . Although we had high targeting accuracy and reliability for exciting targeted cells , we also observed occasional activity in non-targeted cells ( nonspecific activation ) during photostimulation ( Figure 3H ) . This was distance-dependent , and as the distance d between the non-targeted cells and their nearest targeted cells decreased , their probability of activation increased ( Figure 3H ) . And , for the same d , this probability increased with M . The activation of the non-targeted cells may occur through different mechanisms , such as by direct stimulation ( depolarization ) of the cells through their neurites that course through the photostimulation region , or through synaptic activation by targeted cells , or by a combination of the two . In these experiments , we specifically used extremely long stimulation durations ( 480 ~ 962 ms ) to maximally emulate an undesirable photostimulation scenario . The nonspecific activation was confined ( half response rate ) within d < 25 μm in all conditions ( M = 3 ~ 27 across three planes spanning a volume of 240 × 240 × 100 μm3 ) . Nonspecific activation could be reduced by increasing excitation NA ( which is currently limited by the relatively small size of the activation galvanometer mirrors ) , using somatic-restricted expression ( Pégard et al . , 2017; Baker et al . , 2016; Shemesh et al . , 2017 ) , as well as sparse expression . We then aimed to modulate relatively large groups of neurons in 3D . With the low-repetition-rate laser and hybrid scanning strategy ( Discussion ) , the laser beam can be heavily spatially multiplexed to address a large amount of cells while maintaining a low average power . We performed photostimulation of 83 cells across an imaged volume of 480 × 480×150 μm3 in layer 2/3 of V1 in awake mice ( Figure 4 ) . With a total power of 300 mW and an activation time of ~ 95 ms , we were able to activate more than 50 cells . In one experiment , we further sorted target cells into two groups ( 40 and 43 cells respectively ) and photostimulated them separately . More than 30 cells in each group were successfully activated simultaneously with clear evoked calcium transient . In another example , more than 35 cells out of a target group of 50 cells responded ( Figure 4—figure supplement 1 ) . These large scale photostimulations ( >=40 target cells; Figure 4 ) , show that 78 ± 7% of cells in the target ensemble can be successfully activated ( excluding cells that never respond in any of the tested photostimulation pattern , 8 ± 3% , see Materials and Methods ) . Nonspecific photoactivation was more frequent for cells surrounded by target cells , but overall it was confined within 20 μm from the nearest target cell ( Figure 4F ) . We also noted that cells that could be photoactivated individually or in a small ensemble may not get photoactivated when the number of target neurons increases . We hypothesize that this could be due to feed forward inhibition , as targeted pyramidal neurons may activate local interneurons , which then could suppress the firing of neighboring cells . These network interactions will be the subject of future study . Nonspecific excitation can be minimized with sparse stimulation , by simply reducing the likelihood of stimulating directly adjacent cells . One naturally sparse pool of cells are cortical interneurons . Different interneuron classes participate in cortical microcircuits that could serve as gateways for information processing ( Muñoz et al . , 2017; Karnani et al . , 2014 ) . These interneurons are located sparsely in the cortex , yet are highly connected to excitatory populations ( Fino and Yuste , 2011 ) , and are known to strongly modulate cortical activity ( Tsumoto et al . , 1979 ) . However , the effect of simultaneous stimulation of selective subset of interneurons with single cell resolution has not been studied in detail , as previous reports have largely relied on one-photon optogenetics where widespread activation is the norm ( Lee et al . , 2012; Wilson et al . , 2012 ) [but see Ref . ( Karnani et al . , 2016 ) for single cell interneuron stimulations] . To explore this , we used our all-optical approach to examine the effect of photoactivating specific sets of interneurons in 3D on the activity of pyramidal cells that responded to visual stimuli in awake head-fixed mice ( Figure 5 ) . Using viral vectors , we expressed Cre-dependent C1V1 in somatostatin ( SOM ) inhibitory interneurons ( SOM-Cre mice ) , while simultaneously also expressing GCaMP6s in both pyramidal cells and interneurons , in layer 2/3 of mouse V1 . We first imaged the responses of pyramidal cells across three planes ( separated by ~ 45 μm each ) to orthogonal visual stimuli consisting of drifting grating without photostimulation . We then simultaneously photostimulated a group of SOM cells ( M = 9 , with seven showing responses ) across these three planes concurrently with the visual stimuli ( Figure 5A–C; Materials and Methods ) . We observed a significant suppression ( p<0 . 05 , two-sample t-test ) in response among 46% and 35% of the pyramidal cells that originally responded strongly to the horizontal and vertical drifting-grating respectively ( Figure 5A–E ) . Moreover , the orientation selectivity of highly selective cells was largely abolished by SOM cell photoactivation ( Figure 5E ) . This is consistent with reports that SOM cells inhibit nearby pyramidal cells with one-photon optogenetics in vivo ( Lee et al . , 2012; Wilson et al . , 2012 ) or with two-photon glutamate uncaging in vitro ( Fino and Yuste , 2011 ) . Our two-photon approach provides high precision 3D manipulation over groups of cells ( Figure 5D ) , and simultaneous readout of neuronal activity across the network in vivo . Thus , our approach could be useful for dissecting the excitatory and inhibitory interactions in cortical circuits in vivo . To simultaneously photostimulate multiple cells with two-photon excitation , it is becoming common to use holographic approaches ( Nikolenko et al . , 2008; Packer et al . , 2012; Packer et al . , 2015; Dal Maschio et al . , 2017; Mardinly et al . , 2017; Pégard et al . , 2017 ) . Spatial light modulators can generate an ‘arbitrary’ 3D pattern on the sample , limited only by Maxwell’s equations , and the space-bandwidth product of the modulation device . With SLMs , one can independently target a very large number of sites , far in excess of what we demonstrate here , but the number of addressable neurons is limited by the allowable power budget . Moreover , special care has to be taken to minimize the total power deposited on the brain , and avoid direct and indirect thermal effects ( Podgorski and Ranganathan , 2016 ) . We addressed this issue by using a hybrid holographic strategy and a low-repetition-rate laser for photostimulation , with high peak intensities for efficient two-photon excitation , but moderate average power . This allowed us to target a large group of cells with low average power ( e . g . 83 targeted cells across an imaged volume of 480 × 480 × 150 μm3 in awake mice V1 layer 2/3 with 300 mW in total , Figure 4 ) . As these cells generally are not targeted continuously , we do not expect any heat induced effects on cell health under our stimulation conditions ( Podgorski and Ranganathan , 2016 ) . In our hybrid strategy , a group of beamlets is generated by the SLM that target the centroids of the desired neurons . Each discrete focal point in the hologram maintains sufficient axial confinement for typical inter-cell spacing . These beamlets are then rapidly spirally scanned over the neurons’ cell bodies by post-SLM galvanometers . Several alternative scanless approaches exist: pure 3D holograms and another method combining holographic patterning and temporal focusing . The former approach directly generates the full 3D hologram covering the cell bodies of targeted neurons all at once ( Dal Maschio et al . , 2017 ) . Though simplest , the full 3D hologram has a decreased axial resolution as its lateral extend increases ( Papagiakoumou et al . , 2008 ) , and is subject to light contamination to the non-targeted cells , particularly in scattering tissues such as the mammalian brain . In contrast , temporal focusing ( Oron et al . , 2005; Zhu et al . , 2005 ) decouples axial from lateral extent of the hologram by coupling the holographic pattern to a grating ( Papagiakoumou et al . , 2008 ) such that only one axial position in the sample has sufficient spectral content to generate a short laser pulse , thus tightening the axial confinement . Recent reports have extended this method to 3D stimulation ( Hernandez et al . , 2016;Pégard et al . , 2017;Accanto et al . , 2017 ) . Regardless of the exact implementation , these scanless approaches require higher laser powers per cell in general than our hybrid method . For example , with typical photostimulation duration ( ≥5 ms ) , about twice of the power is required using pure hologram compared with our hybrid strategy to achieve similar response in the same cells ( Figure 2 , Figure 2—figure supplement 1 ) . It would likely require even more power for the same excitation with temporal focusing , as its tighter axial confinement would excite less of the membrane . On the other hand , the area-activation of scanless activation generally gives lower latencies and less jitter , compared to scanning strategies . However , as we show in our hybrid scanning approach , even with low powers and longer scan times , we can obtain latencies under 10 ms , with little jitter ( <1 ms , Figure 1—figure supplement 2 ) . Taken together , the spiral scan strategy we adapted requires a lower laser power budget per cell , and is very scalable towards activating large number of simultaneously targeted cells , making it a practical tool to study ensembles in neural circuits . One key strategy we exploited to lower the total average laser power in patterned photostimulation was to employ a low-repetition-rate laser for photostimulation . The average laser power Pave scales with the product of laser peak power Ppeak and pulse repetition rate frep . Since the laser beam is split into M beamlets to target M individual cell , the two-photon excitation for each cell scales with ( Ppeak/M ) 2 ( Denk et al . , 1990 ) . To maintain the required Ppeak for a large M , we reduced frep instead of increasing Pave . The two-photon photostimulation laser we used had a low frep ( 200 kHz ~ 1 MHz ) , leading to a significant increase in Ppeak and thus the number of possible simultaneously targeted cells M , with the same Pave . We note that most opsins open ion channels , the average open time is much longer than the laser’s interpulse interval ( 1/frep ) , and multiple ions can be conducted during each photostimulation . This is in contrast to fluorescence , where at most a single photon is emitted for each absorption , and the lifetime is significantly shorter than the interpulse interval . Thus opsins are ideal targets for low-repetition rate , high peak power excitation . In addition , the repetition rate should be balanced with the photostimulation duration . When the photostimulation duration is very short ( e . g . 1 ms ) , the whole cell body might not be covered well with enough pulses in the spiral scan approach . In these scenarios , a higher repetition rate could be more favorable . The optimal conditions will likely be cell- and opsin-dependent , but would be expected to follow our trends . We choose an ETL for volumetric imaging , because of its low cost and good performance for focusing . Many other options exist including SLM ( Yang et al . , 2016 ) , ultrasound lens ( Kong et al . , 2015 ) , remote focusing ( Botcherby et al . , 2012; Sofroniew et al . , 2016 ) and acousto optic deflector ( Duemani Reddy et al . , 2008; Grewe et al . , 2010; Katona et al . , 2012 ) ; see Ref . ( Yang and Yuste , 2017 ) , for a complete review . One future modification could be replacing the ETL with a second SLM to perform multiplane imaging ( Yang et al . , 2016 ) and adaptive optics ( Ji , 2017 ) , which could increase the frame rate and improve the imaging quality . Another important consideration in our all-optical method was to minimize the cross-talk between imaging and photostimulation . We chose the calcium indicator GCaMP6 and the red-shift opsin C1V1-mCherry , which has a minimized excitation spectrum overlap . Nevertheless , there is still a small cross-talk between the two , as C1V1 has a blue absorption shoulder , and GCaMP6 has a red shifted absorption tail . The first cross-talk affects neuronal excitability , and is the result of photostimulation by the imaging laser . Although the C1V1 we used was red-shifted , it can still be excited at 920 ~ 940 nm , the typical wavelengths used to image GCaMP6 . This cross-talk highly depends on the relative expression of the calcium indicators and opsin ( Rickgauer et al . , 2014; Packer et al . , 2015 ) . For this reason , the imaging laser power was kept as low as possible to values that are just sufficient for imaging . But if the calcium indicator is weakly expressed , hence naturally dim , the increased imaging power may bias the neuronal excitability . Indeed , our cell-attached electrophysiology recording indicates that neuron firing rate has a trend to increase as the imaging laser power increases . However , we found no significant difference of the firing rate under our normal volumetric imaging conditions ( Figure 1—figure supplement 3 ) , where the laser power was typically below 50 mW and could be up to 80 mW for layers deeper than ~ 250 μm . Nevertheless , as red indicators keep improving , a future switch toward ‘blue’ opsins again will be desirable to reduce the spectral overlap between opsin and indicator . The second type of cross-talk affects the high fidelity recording of neural activity , and is caused by fluorescence ( or other interference ) generated by the photostimulation laser directly , which may cause background artifact on the calcium signal recording . To avoid this , in our experiments we use a narrow filter ( passband: 500 nm ~ 520 nm ) for GCaMP6 signal detection . C1V1 is co-expressed with mCherry , which has negligible fluorescence at the filter’s passband . But , in addition , GCaMP6 can still be excited at the photostimulation laser’s wavelength at 1040 nm . Typically this fluorescence is weak and does not impact the data analysis ( e . g . Figure 3 ) . However , if the baseline of GCaMP6 is relatively high or the number of simultaneously targeted neurons is large , it could cause a significant background artifact in the calcium imaging , identified as sharp rise and then sharp decay of fluorescence signals ( Figure 1—figure supplement 4 ) . If the photostimulation duration is short ( e . g . Figure 4 , only one frame appears to have the artifact ) , and stimulation frequency infrequent , the impacted frames could simply be deleted with negligible data loss . But if the photostimulation duration is long ( e . g . Figure 5 ) , the calcium imaging movie can be pre-processed so that the mesh grid shape background is replaced by their adjacent pixel value ( see Materials and Methods ) . The ‘mesh’ arises because the interpulse interval of the laser is greater than the pixel rate , so only selected pixels are compromised . The grid is non-uniform in the image because of the non-uniform resonant scanner speed . This pre-processing significantly suppresses the artifacts while maintaining the original signal . Nevertheless , to avoid any analysis bias , the neuronal response can be further approximated by measuring the ΔF/F signal right after the photostimulation , when there is no background artifact . Also , an alternative method is to gate the PMT , or the PMTs output during the photostimulation pulse , thought this requires dedicated additional electronics . In this case , there will be ‘lost’ signal , and this can be treated similarly by filling in the data with interpolation . Finally , the constrained nonnegative matrix factorization algorithm ( Pnevmatikakis et al . , 2016 ) used to extract the fluorescence signal could also help , as it can identify the photostimulation artifact as part of the background and subtract it from the signal . With these corrections , the photostimulation artifacts can be eliminated from the extracted fluorescence trace in Figure 3~5 . One strategy to reduce nonspecific stimulation is to reduce the size of the PSF by increasing the NA . In our current set of experiments , we use a relatively low excitation NA ( ~0 . 35 ) beam that is limited by the small mirror size ( 3 mm ) of the post-SLM galvanometric scanners . Increasing the mirror size is a straightforward future improvement that would increase this NA , and decrease the axial point spread function . This would also improve the effective axial resolution of photostimulation ( currently ~ 20 μm , measured by displacing the 12 μm diameter spiral pattern relative to the targeted neuron , Figure 1—figure supplement 2 ) , and thus reduce the nonspecific activation of the non-targeted cells . Another approach to reduce the nonspecific activation is to use a somatic-restricted opsin . Somatic-restricted opsins were reported recently ( Baker et al . , 2016; Pégard et al . , 2017; Shemesh et al . , 2017 ) , and showed reduced , but not eliminated , activation of non-targeted cells in vitro . Finally , it remains possible that a significant number of nonspecific activated cells occur through physiological synaptic activation by the photostimulated neurons . Our method could have wide utility in neuroscience . We demonstrate the successful manipulation of the targeted neural microcircuits in awake head-fixed behaving mice by photostimulating a targeted group of interneurons ( Figure 5 ) , and we expect this 3D all-optical method would find its many other applications in dissecting the neural circuits . Though we only targeted neurons in cortical layers 2/3 , the total targetable range of the SLM can be more than 500 µm ( Yang et al . , 2016 ) , thus covering layers 2/3 and 5 simultaneously . Questions such as how neural ensembles are being organized across different cortical layers , and how different neural assemblies across a 3D volume interact with each other can now be directly explored . Indeed , by identifying behavior-related neural ensemble using closed-loop optogenetics ( Grosenick et al . , 2015; Carrillo-Reid et al . , 2017 ) , one may be able to precisely control the animal behavior , which could have a significant impact in attempts to decipher neural codes and also provide an optical method for potential treatment of neurological and mental diseases in human subjects . The optical setup is illustrated in Figure 1A , which is composed of two femtosecond pulse lasers and a custom-modified two-photon laser scanning microscope ( Ultima In Vivo , Bruker Corporation , Billerica , Massachusetts ) . The laser source for imaging is a pulsed Ti:sapphire laser ( Chameleon Ultra II , Coherent , Inc . , Santa Clara , California ) . Its wavelength is tuned to 940 nm for GCaMP6s or GCaMP6f imaging or 750 nm for mCherry imaging respectively . The laser power is controlled with a Pockels cell ( 350–160-BK Pockels cell , 302RM controller , Conoptics , Inc . , Danbury , Connecticut ) . The laser beam is expanded by a 1:3 . 2 telescopes ( f = 125 mm and f = 400 mm ) and coupled to an ETL ( EL-10-30-C-NIR-LD-MV , Optotune AG , Dietikon , Switzerland ) with a clear aperture of 10 mm in diameter . The transmitted beam is rescaled by a 3 . 2:1 telescope ( f = 400 mm and f = 125 mm ) and imaged onto a resonant scanner and galvanometric mirror , both located at the conjugate planes to the microscope’s objective pupil . The beam is further scaled by a 1:1 . 33 telescope before coupled into a scan lens ( f = 75 mm ) , a tube lens ( f = 180 mm ) and the objective lens ( 25x N . A . 1 . 05 XLPlan N , Olympus Corporation , Tokyo , Japan ) , yielding an excitation NA ~ 0 . 45 . The laser can also be directed to a non-resonant scanning path ( without ETL ) where both X and Y scanning are controlled by galvanometric mirrors . The fluorescence signal from the sample is collected through the objective lens and split at a dichroic mirror ( HQ575dcxr , 575 nm long pass , Chroma Technology Corp . , Bellows Falls , Vermont ) to be detected in two bi-alkali photomultiplier tubes , one for each wavelength range . Two different bandpass filters ( 510/20–2P , and 607/45–2P , Chroma Technology Corp . , Bellows Falls , Vermont ) are placed in front of the corresponding PMT respectively . The optical path for the photostimulation is largely independent from the imaging , except that they combine at a dichroic mirror ( T1030SP , 1030 nm short pass , Chroma Technology Corp . , Bellows Falls , Vermont ) just before the scan lens , and then share the same optical path . The laser source for photostimulation is a low repetition rate ( 200 kHz ~ 1 MHz ) pulse-amplified laser ( Spirit 1040–8 , Spectra-physics , Santa Clara , California ) , operating at 1040 nm wavelength . Its power is controlled by a Pockels cell ( 1147-4-1064 Pockels cell , 8025RS-H-2KV controller , FastPulse Technology , Saddle Brook , New Jersey ) . A λ/2 waveplate ( AHWP05M-980 , Thorlabs , Inc . Newton , New Jersey ) is used to rotate the laser polarization so that it is parallel to the active axis of the spatial light modulator ( HSP512-1064 , 7 . 68 × 7 . 68 mm2 active area , 512 × 512 pixels; Meadowlark Optics , Frederick , Colorado ) . The beam is expanded by two telescopes ( 1:1 . 75 , f = 100 mm and f = 175 mm; 1:4 , f = 50 mm and f = 200 mm ) to fill the active area of the SLM . The reflected beam from the SLM is scaled by a 3:1 telescope ( f = 300 mm and f = 100 mm ) and imaged onto a set of close-coupled galvanometer mirrors , located at the conjugate plane to the microscope’s objective pupil . A beam block made of a small metallic mask on a thin pellicle is placed at the intermediate plane of this telescope to remove the zeroth-order beam . The photostimulation laser beam reflected from the galvanometer mirrors are then combined with the imaging laser beam at the 1030 nm short pass dichroic mirror . The imaging and photostimulation is controlled by a combination of PrairieView ( Bruker Corporation , Billerica , Massachusetts ) and custom software ( Yang , 2018 ) running under MATLAB ( The Mathworks , Inc . Natick , Massachusetts ) on a separate computer . The Matlab program was developed to control the ETL through a data acquisition card ( PCIe-6341 , National Instrument , Austin , Texas ) for volumetric imaging , and the SLM through PCIe interface ( Meadowlark Optics , Frederick , Colorado ) for holographic photostimulation ( Yang , 2018 ) . The two computers are synchronized with shared triggers . At the end of each imaging frame , a signal is received to trigger the change of the drive current ( which is converted from a voltage signal from the data acquisition card by a voltage-current converter [LEDD1B , Thorlabs , Inc . Newton , New Jersey] ) of the ETL , so the imaging depth is changed for the following frame . The range of the focal length change on sample is ~+90 μm ~ −200 μm ( ‘+” means longer focal length ) . The intrinsic imaging frame rate is ~ 30 fps with 512 × 512 pixel image . The effective frame rate is lower as we typically wait 10 ~ 17 ms in between frames to let the ETL fully settle down at the new focal length . The control voltage of the Pockels cell is switched between different imaging planes to maintain image brightness . The typical imaging power is < 50 mW , and could be up to 80 mW for layers deeper than ~ 250 μm . The Matlab programs to control the ETL for volumetric imaging and SLM for holographic photostimulation ( Yang , 2018 ) is available at https://github . com/wjyangGithub/Holographic-Photostimulation-System with a GNU General Public License , version 3 ( copy archived at https://github . com/elifesciences-publications/Holographic-Photostimulation-System ) . The phase hologram on the SLM , ϕ ( u , v ) , can be expressed as: ( 1 ) ϕ ( u , v ) =phase{∑i=1MAie2πj{xiu+yiv+[Z20 ( u , v ) C20 ( zi ) +Z40 ( u , v ) C40 ( zi ) +Z60 ( u , v ) C60 ( zi ) ]}}where [xi , yi , zi] ( i = 1 , 2…M ) is the coordinate of the cell body centroid ( M targeted cells in total ) , and Ai is the electrical field weighting coefficient for the ith target ( which controls the laser power it receives ) . Zm0 ( u , v ) and Cm0 ( zi ) are the Zernike polynomials and Zernike coefficients , respectively , which sets the defocusing and compensates the first-order and second-order spherical aberration due to defocusing . Their expressions are shown in Table 1 . The hologram can also be generated by 3D Gerchberg-Saxton algorithm , with additional steps to incorporate spherical aberration compensation . We adapt Equation ( 1 ) as a simpler method . For the experiments in Figure 2 , and Figure 2—figure supplement 1 , the Gerchberg-Saxton algorithm is used to generate a disk with a diameter similar to the neurons . To match the defocusing length set in SLM with the actual defocusing length , we adjusted the ‘effective N . A . ’ in the Zernike coefficients following the calibration procedure described in Ref . ( Yang et al . , 2016 ) . To register the photostimulation beam’s targeting coordinate in lateral directions , we projected 2D holographic patterns to burn spots on the surface of an autofluorescent plastic slide and visualized them by the imaging laser . An affine transformation can be extracted to map the coordinates . We repeated this registration for every 25 μm defocusing depth on the sample , and applied a linear interpolation to the depths in between . An alternative method to register the targeting coordinate is to set the photostimulation laser in imaging mode , actuate the SLM for different lateral deflection , and extract the transform matrix from the acquired images and that acquired from the imaging laser . To characterize the lateral registration error , we actuated the SLM and burned spots on the surface of an autofluorescent plastic slide across a field of view of 240 μm x 240 μm with a 7 × 7 grid pattern . We then imaged the spots pattern with the imaging laser and measured the registration error . This was repeated for different SLM focal depths . To characterize the axial registration error , we used the photostimulation laser to image a slide with quantum dots sample . The SLM was set at different focal depths , and a z-stack was acquired for each setting to measure the actual defocus and thus the axial registration error . In all these registration and characterization procedures , we used water as the media between the objective and the sample , and we kept the focus of the photostimulation laser at the sample surface by translating the microscope stage axially . We note that the refractive index of the brain tissue is slightly different from that of water ( ~2% ) , and this could cause an axial shift of the calibration . This could be corrected in the Zernike coefficients . In practice , we found this effect is negligible , as the typical focal shift by the SLM is relatively small ( <150 μm ) and the axial PSF is large . Due to the chromatic dispersion and finite pixel size of SLM , the SLM’s beam steering efficiency drops with larger angle , leading to a lower beam power for targets further away from the center field of view ( in xy ) , and nominal focus ( in z ) . The characterization result is shown in Figure 1—figure supplement 1 . A linear compensation can be applied in the weighting coefficient Ai in Eq . ( 1 ) to counteract this non-uniformity . In practice , these weighting coefficients can be adjusted such that the targeted neurons show clear response towards photostimulation . Before each set of experiments on animals , we verify the system ( laser power , targeting accuracy , power uniformity among different beamlets from the hologram ) by generating groups of random spots through holograms , burning the spots on an autofluorescent plastic slide , and comparing the resultant image with the desired target . All experimental procedures were carried out in accordance with animal protocols approved by Columbia University Institutional Animal Care and Use Committee . Multiple strains of mice were used in the experiment , including C57BL/6 wild-type and SOM-cre ( Sst-cre ) mice ( stock no . 013044 , The Jackson Laboratory , Bar Harbor , Maine ) at the age of postnatal day ( P ) 45–150 . Virus injection was performed to layer 2/3 of the left V1 of the mouse cortex , 3 ~ 12 weeks prior to the craniotomy surgery . For the C57BL/6 wild-type mice , virus AAV1-syn-GCaMP6s ( or AAV1-syn-GCaMP6f ) and AAVDJ-CaMKII-C1V1- ( E162T ) -TS-p2A-mCherry-WPRE was mixed and injected for calcium imaging and photostimulation; virus AAV8-CaMKII-C1V1-p2A-EYFP was injected for electrophysiology . For the SOM-cre ( Sst-cre ) mice , virus AAV1-syn-GCaMP6s and AAVDJ-EF1a-DIO-C1V1- ( E162T ) -p2A-mCherry-WPRE was mixed and injected . The virus was front-loaded into the beveled glass pipette ( or metal pipette ) and injected at a rate of 80 ~ 100 nl/min . The injection sites were at 2 . 5 mm lateral and 0 . 3 mm anterior from the lambda , putative monocular region at the left hemisphere . After 3 ~ 12 weeks of expression , mice were anesthetized with isoflurane ( 2% by volume , in air for induction and 1–1 . 5% during surgery ) . Before surgery , dexamethasone sodium phosphate ( 2 mg per kg of body weight; to prevent cerebral edema ) were administered subcutaneously , and enrofloxacin ( 4 . 47 mg per kg ) and carprofen ( 5 mg per kg ) were administered intraperitoneally . A circular craniotomy ( 2 mm in diameter ) was made above the injection cite using a dental drill . A 3 mm circular glass coverslip ( Warner instruments , LLC , Hamden , Connecticut ) was placed and sealed using a cyanoacrylate adhesive . A titanium head plate with a 4 mm by 3 . 5 mm imaging well was attached to the skull using dental cement . After surgery , animals received carprofen injections for 2 days as post-operative pain medication . The imaging and photostimulation experiments were performed 1 ~ 21 days after the chronic window implantation . During imaging , the mouse is either anesthetized with isoflurane ( 1–1 . 5% by volume in air ) with a 37°C warming plate underneath or awake and can move freely on a circular treadmill with its head fixed . Visual stimuli were generated using MATLAB and the Psychophysics Toolbox ( Brainard , 1997 ) and displayed on a monitor ( P1914Sf , 19-inch , 60 Hz refresh rate , Dell Inc . , Round Rock , Texas ) positioned 15 cm from the right eye , at 45o to the long axis of the animal . Each visual stimulus session consisted of four different trials , each trial with a 2 s drifting square grating ( 0 . 04 cycles per degree , two cycles per second ) , followed by 18 s of mean luminescence gray screen . Four conditions ( combination of 10% or 100% grating contrast , 0o or 90o drifting grating direction ) were presented in random order in the four trials in each session . The pulse repetition rate of the photostimulation laser used in the experiment is 500 kHz or 1 MHz . The photostimulation laser beam is split into multiple foci , and spirally scanned ( ~12 μm outer spiral diameter , 8 ~ 50 rotations with progressively shrinking radius; the whole spiral can be continuously repeatedly scanned ) by a pair of post-SLM galvanometric mirror over the cell body of each target cell . For neurons in layer 2/3 of mice V1 , the typical average power used for each spot is 2 mW ~ 5 mW . When studying the photostimulation effect on the non-targeted cells ( Figure 3 ) , we specifically used long photostimulation durations ( 480 ms ~ 962 ms ) to emulate an undesirable photostimulation scenario . In the normal condition , the photostimulation duration is < 100 ms , which was composed of multiple continuously repeated spiral scans , each lasting < 20 ms ( Figure 4 ) . In the experiments where short photostimulation duration ( ≤20 ms , Figure 2 ) is used , the stimulation was composed of a single spiral scan which consists of ~ 50 rotations with progressively shrinking radius . In the experiment that the SOM cells were photostimulated when the mouse were receiving visual stimuli ( Figure 5 ) , the photostimulation started 0 . 5 s before the visual stimuli , and ended 0 . 3 s after the visual stimuli finished . Since the visual stimuli lasted for 2 s , the photostimulation lasted for 2 . 8 s . This long photostimulation was composed of 175 continuously repeated spiral scans , each lasting ~ 16 ms . In our experiments , the lateral separation of the simultaneously targeted cell ranges from ~ 10 µm to ~ 315 µm , and the axial separation ranges from 30 µm to 150 µm . The recording from each plane was first extracted from the raw imaging files , followed by motion correction using a pyramid approach ( Thévenaz et al . , 1998 ) or fast Fourier transform-based algorithm ( Dubbs et al . , 2016 ) . A constrained nonnegative matrix factorization ( CNMF ) algorithm ( Pnevmatikakis et al . , 2016 ) was used to extract the fluorescence traces ( ΔF/F ) of the region of interested ( i . e . neuron cell bodies in the field of view ) . The CNMF algorithm also outputs a temporally deconvolved signal , which is related to the firing event probability . The ΔF/F induced by the photostimulation was quantified with the mean fluorescence change during the photostimulation period over the mean fluorescence baseline within a 0 . 5 ~ 2 s window before the photostimulation . To detect the activity events from each recorded neuron , we typically thresholded the temporally deconvolved ΔF/F signal with at least two standard derivations from the mean signal . Independently , a temporal first derivative is applied to the ΔF/F trace . The derivative is then threshold at least two standard derivations from the mean . At each time point , if both are larger than the threshold , an activity event is recorded in binary format . In case the auto-detected activity event has large deviations from manual inspection ( based on typical shapes of calcium transient ) , the thresholding value is adjusted so that the overall auto-detection agrees with manual inspection . A cell is determined as not responding to photostimulation if there is no single activity event detected or no typical action-potential-corresponding calcium transient during photostimulation period for multiple trials . These non-responding cells could be due to a poor expression of C1V1 . Any GCaMP can generate fluorescence background during photostimulation ( Discussion ) . This background would reduce the sensitivity of the calcium imaging . Since the pixel rate ( ~8 . 2 MHz ) of the calcium imaging recording is much faster than the photostimulation laser’s pulse repetition rate ( 200 kHz ~ 1 MHz ) , the fluorescence background appears to be a mesh grid shape in the calcium imaging movie ( Figure 1—figure supplement 4 ) . Typically it is small and does not impact the above data analysis ( e . g . Figure 3 ) . In the case that it is strong , if the photostimulation duration is short ( e . g . Figure 4 , only one frame appears to have the artifact ) , the impacted frames can be deleted with negligible data loss . If the photostimulation duration is long ( e . g . Figure 5 ) , the recorded frames during photostimulation are pre-processed to suppress this background artifact ( Figure 1—figure supplement 4 ) . To detect the pixels having this artifact , we consider both their fluorescence value and their geometry . First we detect candidate pixels by identifying pixels whose value is significantly higher from the average value calculated from a few frames just before and just after the stimulation . Second , these candidate pixels are tested for connectedness within every horizontal and vertical line of each frame , and the width of the connections compared to that expected based on the stimulation condition . If both these conditions hold , these pixels are marked as ‘contaminated’ and the fluorescence value at these pixels during the stimulation are replaced by those in their adjacent ‘clean’ pixels . This pre-processing significantly suppresses the artifacts while maintaining the original signal . Nevertheless , to avoid any analysis bias , we further approximated the neuronal response by using the ΔF/F signal just after the photostimulation , when there was no background artifact . The same analysis procedure was implemented to the control experiment when there was no photostimulation . The orientation selectivity index and preference of the visual stimuli is calculated as the amplitude and sign of ( ΔF/F\|90 - ΔF/F\|0 ) / ( ΔF/F\|90 + ΔF/F\|0 ) respectively , where ΔF/F\|90 and ΔF/F\|0 is the mean ΔF/F during the visual stimuli with 90o and 0o drifting grating respectively . Mice were head-fixed and anaesthetized with isoflurane ( 1 . 5 ~ 2% ) throughout the experiment . Dura was carefully removed in the access point of the recording pipette . 2% agarose gel in HEPES-based artificial cerebrospinal fluid ( ACSF ) ( 150 mM NaCl , 2 . 5 mM KCl , 10 mM HEPES , 2 mM CaCl2 , 1 mM MgCl2 , pH was 7 . 3 ) was added on top of the brain to avoid movement artifacts . Patch pipettes of 5 ~ 7 MΩ pulled with DMZ-Universal puller ( Zeitz-Instrumente Vertriebs GmbH , Planegg , Germany ) were filled with ACSF containing 25 μM Alexa 594 to visualize the tip of the pipettes . C1V1-expressing cells were targeted using two-photon microscopy in vivo . During recordings , the space between the objective and the brain was filled with ACSF . Cell-attached recordings were performed using Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , California ) , in voltage-clamp mode . The sampling rate was 10 kHz , and the data was low-pass filtered at 4 kHz using Bessel filter .	Modern microscopy provides a window into the brain . The first light microscopes were able to magnify cells only in thin slices of tissue . By contrast , today’s light microscopes can image cells below the surface of the brain of a living animal . Even so , this remains challenging for several reasons . One is that the brain is three-dimensional . Another is that brain tissue scatters light . Trying to view neurons deep within the brain is a little like trying to view them through a glass of milk . Most of the light scatters on its way through the tissue with the result that little of the light reaches the target neurons . Yang et al . have now tackled these challenges using a technique called holography . Holography produces 3D images of objects by splitting a beam of light and then recombining the beams in a specific way . Yang et al . applied this technique to an infrared laser beam , opting for infrared because it scatters much less in brain tissue than visible light . Directing each of the infrared beams to a different neuron can produce 3D images of multiple cells within the brain’s outer layer , the cortex , all at the same time . The holographic infrared microscope can be used alongside two techniques called optogenetics and calcium imaging , in which light-sensitive proteins are inserted into neurons . Depending on the proteins introduced , shining light onto the neurons will either change their activity , or cause them to fluoresce whenever they are active . Just as a computer can both “read” and “write” data , the holographic microscope can thus read out existing neuronal activity or write new patterns of activity . By combining these techniques , Yang et al . were able to stimulate more than 80 neurons at the same time – and meanwhile visualize the activity of the surrounding neurons – at multiple depths within the mouse cortex . This new microscopy technique , while a clear advance over existing methods , still cannot image and control neurons throughout the entire cortex . The next goal is to further extend this method across multiple brain areas and manipulate the activity of any subset of neurons at will . Neuroscientists will greatly benefit from the ability to image and alter the activity of living neural circuits in 3D . In the future , clinicians may be able to use this technique to treat brain disorders by adjusting the activity of abnormal neural circuits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2018	Simultaneous two-photon imaging and two-photon optogenetics of cortical circuits in three dimensions
During transcription elongation , RNA polymerase has been assumed to attain equilibrium between pre- and post-translocated states rapidly relative to the subsequent catalysis . Under this assumption , recent single-molecule studies proposed a branched Brownian ratchet mechanism that necessitates a putative secondary nucleotide binding site on the enzyme . By challenging individual yeast RNA polymerase II with a nucleosomal barrier , we separately measured the forward and reverse translocation rates . Surprisingly , we found that the forward translocation rate is comparable to the catalysis rate . This finding reveals a linear , non-branched ratchet mechanism for the nucleotide addition cycle in which translocation is one of the rate-limiting steps . We further determined all the major on- and off-pathway kinetic parameters in the elongation cycle . The resulting translocation energy landscape shows that the off-pathway states are favored thermodynamically but not kinetically over the on-pathway states , conferring the enzyme its propensity to pause and furnishing the physical basis for transcriptional regulation . Transcription constitutes the first and a central regulatory step for gene expression ( Greive and von Hippel , 2005; Coulon et al . , 2013 ) . During the process of RNA synthesis , RNA polymerase ( RNAP ) converts the energy from chemical catalysis of the nucleoside triphosphate ( NTP ) into mechanical translocation along the DNA template . Two classes of mechanisms have been offered to describe the mechanochemical coupling of transcription elongation . The first class , known as the ‘power stroke’ mechanism , suggests that the forward translocation of RNAP is directly driven by a chemical step such as the release of the pyrophosphate ( PPi ) ( Yin and Steitz , 2004 ) . The second class , known as the ‘Brownian ratchet’ mechanism , postulates that the polymerase oscillates back and forth on the DNA template between a pre- and a post-translocated state at the beginning of each nucleotide addition cycle , and that such thermally-driven motions are rectified to the post-translocated state by the incorporation of the incoming NTP ( Guajardo and Sousa , 1997 ) . After extensive structural and biochemical investigations , it is now generally thought that multi-subunit RNAPs , including bacterial and eukaryotic enzymes , function through the Brownian ratchet mechanism ( Komissarova and Kashlev , 1997a; Bai et al . , 2004; Bar-Nahum et al . , 2005; Brueckner and Cramer , 2008 ) . This mechanism received further support from single-molecule studies , which followed the dynamics of individual transcription elongation complexes ( TECs ) ( Abbondanzieri et al . , 2005; Bai et al . , 2007; Larson et al . , 2012 ) . Nonetheless , in order to explain the relationship between the elongation velocity and the external force applied to RNAP obtained from single-molecule experiments , the classical linear ratchet mechanism ( Figure 1 ) had to be modified such that the incoming NTP must also bind to the pre-translocated TEC ( Figure 1—figure supplement 1 ) ( Abbondanzieri et al . , 2005; Larson et al . , 2012 ) . In the pre-translocated TEC , the primary nucleotide binding site is occupied by the 3′-end of the nascent transcript . Thus , the branched Brownian ratchet scheme necessarily requires a secondary NTP binding site on the enzyme . However , the precise location of this secondary site and the mechanism by which the NTP is transferred to the primary site remain poorly defined . 10 . 7554/eLife . 00971 . 003Figure 1 . Nucleotide addition cycle and off-pathway pausing of transcription elongation . The nucleotide addition phase and the pausing phase are colored in green and blue , respectively . At the beginning of a nucleotide addition cycle , the transcription elongation complex ( TEC ) with a transcript length of n thermally fluctuates between the pre-translocated state ( TECn , 0 ) and the post-translocated state ( TECn , 1 ) with a forward rate constant k1 and a reverse rate constant k−1 . After translocation , the incoming NTP binds to the active site with a binding rate constant k2 and a dissociation rate constant k−2 . NTP binding is followed by NTP sequestration , bond formation , and PPi release , which are collectively described by a single catalysis rate constant k3 in our study . Upon the release of the PPi , TEC is reset to the pre-translocated state ( TECn+1 , 0 ) and ready for the next nucleotide addition cycle . From the pre-translocated state , the polymerase can also enter the off-pathway pausing phase by backtracking . The pausing kinetics are determined by the backward stepping rate constants kbn and forward stepping rate constants kfn . The inset shows cartoon configurations of the TEC in a pre-translocated and a post-translocated state . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00310 . 7554/eLife . 00971 . 004Figure 1—figure supplement 1 . A branched Brownian ratchet model for the nucleotide addition cycle . This kinetic model , proposed by Larson et al . ( Larson et al . , 2012 ) , allows NTP to bind to the TEC both before and after translocation , postulating a secondary nucleotide binding site for NTP binding to the pre-translocated TEC ( TECn , 0 ) . Additionally , the model assumes rapid forward and reverse translocation of the polymerase and hence describes the translocation step simply with an equilibrium constant Kδ . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 004 Pausing is an off-pathway process that plays crucial roles in the regulation of transcription elongation ( Landick , 2006; Nudler , 2012 ) . In one view of the mechanisms of transcriptional pausing , RNAP first enters an elemental pause state ( Herbert et al . , 2006; Toulokhonov et al . , 2007; Sydow et al . , 2009 ) , whose structural evidence was recently presented in bacterial RNAP ( Weixlbaumer et al . , 2013 ) . However , similar evidence is lacking for eukaryotic polymerases . These elemental pauses can be subsequently stabilized into longer-lived pauses by the formation of a hairpin structure in the nascent RNA transcript or by RNAP backtracking ( Artsimovitch and Landick , 2000; Herbert et al . , 2008 ) . The backtracking process is caused by upstream movements of the polymerase , displacing the 3′-end of the nascent RNA away from the active site into the secondary channel of the enzyme ( Nudler et al . , 1997; Komissarova and Kashlev , 1997b ) . An alternative view poses that most pauses are attributed to backtracking , which can be described as a one-dimensional random walk of the enzyme along the DNA template ( Galburt et al . , 2007; Mejia et al . , 2008; Depken et al . , 2009; Hodges et al . , 2009 ) . RNA synthesis resumes when the polymerase diffusively realigns its active site with the 3′-end of the transcript . Both the nucleotide addition phase and the pausing phase are closely regulated by conserved structural motifs near the active center of the polymerase , namely the bridge helix and the trigger loop ( TL ) ( Bar-Nahum et al . , 2005; Wang et al . , 2006; Vassylyev et al . , 2007; Brueckner and Cramer , 2008; Kaplan et al . , 2008; Tan et al . , 2008 ) . In order to understand the mechanism of transcription and its regulation , it is important to achieve a detailed description of both on- and off-pathway kinetics of the elongation reaction . Previous efforts to dissect the kinetic scheme of transcription elongation have assumed that the forward and reverse translocation steps of the Brownian ratchet occur in rapid equilibrium relative to the chemical steps in the nucleotide addition cycle ( Guajardo and Sousa , 1997; Bai et al . , 2004; Abbondanzieri et al . , 2005; Tadigotla et al . , 2006 ) . However , the assumption of fast translocation equilibrium has never been experimentally validated . In fact , recent studies suggested that the translocation step may be partially rate-limiting for the nucleotide addition cycle , which gives rise to the heterogeneous elongation rates at different template positions ( Nedialkov et al . , 2003; Kireeva et al . , 2010; Maoiléidigh et al . , 2011; Malinen et al . , 2012; Nedialkov et al . , 2012; Imashimizu et al . , 2013 ) . In this work , we sought to achieve a comprehensive kinetic characterization of transcription elongation without making any assumption about the rate-limiting mechanism of the reaction . We used an optical tweezers assay to follow the transcription trajectories of single yeast RNA polymerase II ( Pol II ) molecules under a variety of conditions , including varying NTP concentrations , assisting and opposing applied forces , and different tracks ( bare and nucleosomal DNA ) . In vivo , eukaryotic DNA is organized around histone octamers to form nucleosomes , which impose physical barriers to transcription elongation . We have previously demonstrated that a transcribing Pol II cannot actively unravel a wrapped nucleosome . Instead , the polymerase pauses and waits until the local nucleosomal DNA spontaneously unwraps and permits Pol II to advance ( Hodges et al . , 2009; Bintu et al . , 2012 ) . Here we used the nucleosomal barrier as a tool to specifically perturb forward translocation of a transcribing Pol II and separately measured the forward and reverse translocation rates . Surprisingly , we found that the forward translocation rate is of the same order of magnitude as the catalysis rate , in contradiction to previous assumptions of fast translocation . This finding reveals that translocation and catalysis together constitute the rate-limiting steps in the nucleotide addition cycle . As a consequence , we were able to rationalize the observed force–velocity relationship of the enzyme with a linear Brownian ratchet scheme in which the incoming NTP only binds to the post-translocated TEC , thus reconciling bulk and single-molecule data and arriving at a unifying view of the transcription elongation process . We further obtained all the major kinetic parameters in the nucleotide addition phase and the pausing phase of the elongation cycle . The energy landscape for transcription elongation derived from these parameters shows that: ( i ) the enzyme thermodynamically favors the pre-translocated state to the post-translocated state; ( ii ) entry into the 1-basepair ( bp ) backtracked state is easier than into further backtracked states; and ( iii ) from the pre-translocated state , the enzyme thermodynamically favors the backtracked states , but kinetically favors forward translocation . We also applied this analysis to a TL mutant Pol II , Rpb1-E1103G ( Malagon et al . , 2006 ) , to quantitatively elucidate the roles of the TL in transcription elongation . Our results indicate that the conformational transitions of the TL control enzyme translocation , catalysis , and pausing , rendering it a vital target element for transcriptional regulation . The Brownian ratchet kinetic scheme for the nucleotide addition cycle of transcription elongation ( Figure 1 ) can be simplified to: TECn , 0⇄k−1k1TECn , 1⇄k−2k2[NTP]TECn , 1⋅NTP→k3TECn+1 , 0where k1 and k−1 are the forward and reverse translocation rate constants , k2 and k−2 are the NTP binding and dissociation rate constants , and k3 is the combined catalysis rate constant that includes NTP sequestration , bond formation , and PPi release . Because of the large equilibrium constant of transcription elongation and the very low PPi concentration ( 1 μM ) in the buffer , k3 was considered essentially irreversible ( Erie et al . , 1992 ) . Using the concept of net rate constants ( Cleland , 1975 ) , we can replace the reversible rate constants between two adjacent states with a single net rate constant and re-write the above scheme as:TECn , 0→k1netTECn , 1→k2netTECn , 1⋅NTP→k3TECn+1 , 0k1net and k2net are the net rate constants for translocation and NTP binding , respectively , which are given by: ( 1 ) k2net=k2[NTP]⋅k3k−2+k3 ( 2 ) k1net=k1⋅k2netk−1+k2net=k1k2k3[NTP]k−1 ( k−2+k3 ) +k2k3[NTP] The time the enzyme takes to finish one nucleotide addition cycle ( τ ) equals the step size of the polymerase ( d = 1 nt ) divided by the pause-free velocity ( v ) , and also equals the sum of the inverse of each net rate: ( 3 ) τ=dv=1k1net+1k2net+1k3 Plugging Equations 1 and 2 into Equation 3 yields the following expression for the pause-free velocity: ( 4 ) v=k1k3k1+k3⋅[NTP] ( k1+k−1 ) ⋅ ( k−2+k3 ) ( k1+k3 ) ⋅k2+[NTP]⋅d We note that this expression is more general than those shown in previous studies ( Abbondanzieri et al . , 2005; Bai et al . , 2007 ) , as it is derived without assuming local equilibration of translocation and NTP binding . In particular , we describe the kinetics of the translocation step with k1 and k−1 , instead of a single equilibrium constant Kδ = k−1/k1 . Such treatment is a prerequisite to explicitly determine the forward and reverse translocation rates . Equation 4 can be simplified to the Michaelis–Menten equation form: ( 5 ) v=Vmax[NTP]KM+[NTP]where Vmax=k1k3k1+k3⋅d , and KM=k1+k−1k1+k3⋅k−2+k3k2 . We followed the transcriptional dynamics of individual Pol II molecules with a dual-trap optical tweezers instrument . One laser trap holds a polystyrene bead attached to a stalled Pol II molecule , while the other trap holds another bead attached to the upstream DNA template ( assisting force geometry; Figure 2A ) or to the downstream template ( opposing force geometry , not shown ) . Upon addition of NTP , transcription restarts , resulting in a change of the DNA tether length and thereby a variation of the force applied to Pol II . Single-molecule transcription trajectories were collected at a range of NTP concentrations ( 35 μM–2 mM ) ( Figure 2B , C ) . The relationship between pause-free velocity ( v ) and [NTP] for the wild-type enzyme fits well to Equation 5 , with Vmax = 25 ± 3 nt/s and KM = 39 ± 12 μM ( errors are SEM ) ( Figure 2D , gray line ) . We also examined the dynamics of the E1103G mutant Pol II , which is known to transcribe DNA at a faster overall velocity than the wild-type ( Figure 2B , C ) ( Malagon et al . , 2006; Kireeva et al . , 2008 ) . We found that the maximum pause-free velocity of the mutant is ∼1 . 5-fold higher than that of the wild-type , with Vmax = 38 ± 5 nt/s and KM = 62 ± 15 μM ( Figure 2D , blue line ) . 10 . 7554/eLife . 00971 . 005Figure 2 . Single-molecule transcription assay . ( A ) Experimental setup for the single-molecule transcription assay . Each of the two optical traps holds a 2 . 1-μm polystyrene bead . Biotinylated Pol II is attached to the streptavidin ( SA ) bead . The upstream DNA is attached to the antibody ( AD ) bead via the digoxigenin–antidigoxigenin linkage . The black arrow indicates the direction of transcription . A nucleosome can be loaded on the downstream DNA as shown . ( B ) Example transcription trajectories of the wild-type Pol II at 50 μM NTP on bare DNA , 1 mM NTP on bare DNA , and 1 mM NTP in the presence of a nucleosome . The nucleosome positioning sequence ( NPS ) is represented by the yellow shaded region . ( C ) Example transcription trajectories of the E1103G mutant Pol II under various conditions . ( D ) Pause-free velocities of the wild-type ( black ) and mutant Pol II ( blue ) at various NTP concentrations . Dashed lines are fits to the Michaelis–Menten equation ( Equation 3; R2 = 0 . 80 for the wild-type; R2 = 0 . 85 for mutant ) . ( E ) The apparent pause densities ( ρpause ) of the wild-type Pol II at different NTP concentrations are plotted against the corresponding pause-free velocities ( v ) . ( F ) ρpause–v relationship for the mutant enzyme . Error bars represent standard error of the mean ( SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00510 . 7554/eLife . 00971 . 006Figure 2—figure supplement 1 . Cumulative distribution of the pause durations for the wild-type Pol II on bare DNA ( black solid line ) and nucleosomal DNA ( red solid line ) . Dashed lines are theoretical fits of the experimental data to the one-dimensional diffusion model for backtracking ( Equation 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00610 . 7554/eLife . 00971 . 007Figure 2—figure supplement 2 . A gel-based time-coursed transcription assay of the wild-type Pol II on bare and nucleosomal DNA . The transcription reactions were carried out in 40 mM KCl and quenched with EDTA after 1 , 2 , 5 , and 20 min incubation with 1 mM NTPs . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00710 . 7554/eLife . 00971 . 008Figure 2—figure supplement 3 . A gel-based transcription assay of the wild-type Pol II and the E1103G mutant Pol II in various KCl concentrations . The experiment was carried out in 40 , 150 , 300 , 450 mM of KCl . The transcription reactions were quenched with EDTA after 10 min incubation with 1 mM NTPs . The run-off length is 612 nt . The nucleosomal dyad is located at nucleotide position 407 . The arrows indicate the salt condition used in the single-molecule experiments ( 300 mM KCl ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00810 . 7554/eLife . 00971 . 009Figure 2—figure supplement 4 . Mean dwell times of the wild-type Pol II at different nucleotide positions . The experiments were conducted at 1 mM NTP concentration . The yellow shade indicates the extended NPS region ( −115 bp to +85 bp relative to the dyad ) . The arrow on the top axis marks the position of the dyad . The same DNA sequence was used in the bulk and single-molecule assays . Both assays show the most predominant pausing about 20 bp before the dyad , consistent with previous results ( Bondarenko et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 00910 . 7554/eLife . 00971 . 010Figure 2—figure supplement 5 . Mean dwell times of the mutant Pol II at different nucleotide positions . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 010 As shown in the example trajectories ( Figure 2B , C ) , transcription elongation is punctuated by pauses of various durations . Pause density , ρpause , is defined as the average number of pauses per bp of template transcribed . As the concentration of NTP goes up , the pause-free velocity increases and the apparent ρpause , which counts pauses lasting longer than 1 s , decreases ( Figure 2E ) . The same trend was also observed for the mutant Pol II ( Figure 2F ) . The inverse relationship between v and ρpause indicates that elongation and pausing are in kinetic competition and that pausing occurs prior to NTP binding ( Artsimovitch and Landick , 2000; Davenport et al . , 2000; Forde et al . , 2002; Landick , 2006; Mejia et al . , 2008 ) . Note that pausing has also been observed to occur after NTP binding at certain sequences for Escherichia coli RNAP; however , yeast Pol II does not seem to employ such mechanism ( Kireeva and Kashlev , 2009 ) . The pause-free velocities and apparent pause densities at various NTP concentrations are summarized in Table 1 . 10 . 7554/eLife . 00971 . 011Table 1 . Summary of pause-free velocities and apparent pause densities measured at various NTP concentrationsDOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 011Pol II[NTP] ( μM ) NPause-free velocity ( nt/s ) Apparent pause density ( bp−1 ) wild-type351012 . 4 ± 0 . 70 . 0721 ± 0 . 0099501111 . 6 ± 1 . 10 . 0526 ± 0 . 008675917 . 8 ± 1 . 20 . 0358 ± 0 . 00791001315 . 7 ± 1 . 40 . 0326 ± 0 . 00772001721 . 0 ± 2 . 40 . 0184 ± 0 . 004610004424 . 7 ± 1 . 80 . 0156 ± 0 . 00312000926 . 7 ± 4 . 30 . 0188 ± 0 . 0076E1103G351016 . 1 ± 0 . 90 . 0374 ± 0 . 0055501315 . 2 ± 1 . 00 . 0266 ± 0 . 0055751318 . 9 ± 1 . 50 . 0290 ± 0 . 00691001324 . 2 ± 1 . 70 . 0106 ± 0 . 00362001327 . 4 ± 3 . 90 . 0100 ± 0 . 00274001035 . 6 ± 1 . 90 . 0094 ± 0 . 006210009637 . 6 ± 4 . 90 . 0051 ± 0 . 000820001542 . 1 ± 4 . 90 . 0083 ± 0 . 0011Data are shown as mean ± SEM . The apparent pause densities are determined by counting pauses that last between 1 s and 120 s . N is the number of single-molecule transcription trajectories at each condition . Backtracking is a major mechanism for transcriptional pauses . We have previously modeled backtracking as a one-dimensional random walk of the enzyme along the DNA template ( Hodges et al . , 2009 ) . In this model , Pol II diffuses back and forth on DNA with a forward stepping rate constant kf and a backward stepping rate constant kb during a backtracked pause . These rate constants are dependent on the applied force ( F , which is positive for assisting forces and negative for opposing forces ) according to: ( 6 ) kf=k0eF⋅Δ/kBT ( 7 ) kb=k0e−F⋅ ( 1−Δ ) /kBTwhere k0 is the intrinsic zero-force stepping rate constant of Pol II diffusing along DNA during backtracking , Δ is the distance to the transition state for each step ( taken to be 0 . 5 bp , or 0 . 17 nm ) , kB is the Boltzmann constant , and T is the temperature ( kBT = 4 . 11 pN·nm at 25ºC ) . The probability density of pause durations , ψ ( t ) , is equivalent to the distribution of first-passage times for a particle diffusing on a one-dimensional lattice to return to the origin ( Depken et al . , 2009 ) , and is given by: ( 8 ) ψ ( t ) =kfkbexp[− ( kf+kb ) t]tI1 ( 2tkfkb ) where I1 is the modified Bessel function of the first kind . We fit the distribution of pause durations for the wild-type enzyme on bare DNA to this model and extracted a characteristic k0 of 1 . 3 ± 0 . 3 s−1 ( Figure 2—figure supplement 1 , gray dashed line ) . Using the values of k0 and the applied force in our experiment ( 6 . 5 pN ) , we calculated kf and kb to be 1 . 7 ± 0 . 4 s−1 and 1 . 0 ± 0 . 3 s−1 , respectively ( Equations 6 and 7 ) . Next , we investigated the transcriptional dynamics of Pol II through the nucleosome by loading a histone octamer on the 601 nucleosome positioning sequence ( NPS ) ( Lowary and Widom , 1998 ) ( Figure 2B , C , Figure 2—figure supplements 2–5 ) . The wild-type enzyme displays a two-fold increase in the apparent pause density upon encountering the nucleosome ( Table 2 ) . The mean pause duration on nucleosomal DNA is significantly longer than that on bare DNA ( Table 2; Figure 2—figure supplement 1 ) . Similarly , the mutant Pol II displays higher pause density and longer pause duration in the presence of a nucleosome ( Table 2 ) . 10 . 7554/eLife . 00971 . 012Table 2 . Apparent pause densities and mean pause durations on bare DNA and nucleosomal DNA in the extended NPS regionDOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 012Pol IIDNA templateNApparent pause density ( bp−1 ) Mean pause duration ( s ) wild-typeBare380 . 0153 ± 0 . 00413 . 9 ± 0 . 6Nucleosomal940 . 0280 ± 0 . 00369 . 4 ± 0 . 8E1103GBare850 . 0046 ± 0 . 00153 . 9 ± 0 . 5Nucleosomal640 . 0202 ± 0 . 00507 . 6 ± 1 . 0Data are shown as mean ± SEM . The extended NPS region spans −115 nt to+85 nt relative to the nucleosomal dyad . It has been shown that the nucleosomal DNA can spontaneously unwrap and rewrap around the histones ( Li et al . , 2005; Koopmans et al . , 2008; Voltz et al . , 2012 ) . The increased pause duration of Pol II on nucleosomal DNA can be explained by rewrapping of the DNA downstream of a backtracked Pol II , which prevents the polymerase from diffusing back to the 3′-end of the nascent RNA to resume transcription ( Hodges et al . , 2009; Bintu et al . , 2012 ) . Because one bp of nucleosomal DNA fluctuates much faster ( >1000 s−1; see ‘Materials and methods’ for the derivation ) than Pol II stepping ( ∼1 s−1 ) , the nucleosomal DNA in front of the polymerase reaches wrapping/unwrapping equilibrium between each backtracking step . It follows that the pause durations on nucleosomal DNA can be drawn from the same distribution as on bare DNA , except that the effective forward stepping rate is reduced by a factor , γu , corresponding to the fraction of time the local nucleosomal DNA is unwrapped ( Hodges et al . , 2009 ) , that is kf ( nucl ) =γu⋅kf . The backward stepping rate kb is not affected by the nucleosome , because little histone transfer occurs in our experimental geometry where the DNA template is under tension ( Hodges et al . , 2009; Bintu et al . , 2011 ) and therefore the polymerase does not encounter any roadblock when it diffuses backward . The distribution of pause durations for wild-type Pol II on nucleosomal DNA can be correctly fit by this model with a γu value of 0 . 6 ± 0 . 2 ( Figure 2—figure supplement 1 , red dashed line ) . Having understood the effect of the nucleosomal barrier on the pausing dynamics , we then turned our attention to its effect on the on-pathway elongation kinetics . Interestingly , we found that the nucleosome also delays the transcribing enzyme by modulating its pause-free velocity . As the wild-type Pol II transcribes through nucleosomal DNA at saturating [NTP] , its mean pause-free velocity decreases by 14% from 26 . 9 ± 0 . 8 nt/s to 23 . 2 ± 0 . 6 nt/s ( Figure 3A ) . The mutant Pol II is even more dramatically slowed down by the nucleosome , with its mean pause-free velocity reduced by 35% from 39 . 8 ± 0 . 6 nt/s to 26 . 0 ± 0 . 7 nt/s ( Figure 3B ) . 10 . 7554/eLife . 00971 . 013Figure 3 . Comparison of pause-free velocities on bare DNA and nucleosomal DNA . ( A ) Pause-free velocities of the wild-type Pol II on bare DNA ( black ) and nucleosomal DNA ( red ) are plotted as a function of the transcript length . The nucleosomal dyad position corresponds to a transcript length of 407 nt . The extended NPS region ( −115 nt to +85 nt relative to the nucleosomal dyad ) is highlighted in yellow . The arrow on the top axis marks the position of the dyad . ( B ) Pause-free velocities of the E1103G mutant Pol II on bare DNA ( blue ) and nucleosomal DNA ( green ) are plotted as a function of the transcript length . These experiments were conducted at 1 mM NTP . Note that after the polymerase exits the nucleosome , the velocity does not return to the same level of that on bare DNA . This observation could be rationalized if the nucleosome rolls along the DNA and remains ahead of the transcribing polymerase in a fraction of the traces . Error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 013 We have previously shown that a transcribing Pol II cannot actively open a wrapped nucleosome; instead , the enzyme passively waits for the DNA immediately in front of it to spontaneously unwrap and then translocates forward through a locally unwrapped nucleosome ( Hodges et al . , 2009 ) . Since the fluctuations of local nucleosomal DNA occur orders of magnitude faster than the translocations of Pol II during backtracking , we assume that they are also much faster than the on-pathway translocation steps of Pol II . Under this assumption , local DNA reaches wrapping/unwrapping equilibrium before Pol II makes a translocation step and the forward translocation rate ( k1 ) is effectively reduced by the fraction of time the local nucleosomal DNA is unwrapped ( γu ) . The reverse translocation rate ( k−1 ) is unlikely to be affected , again due to the lack of a roadblock against reverse translocation . Thus , according to Equation 5 , the maximum pause-free velocity for nucleosomal DNA transcription is: ( 9 ) Vmax ( nucl ) =γu⋅k1⋅k3 ( γu⋅k1 ) +k3⋅d In comparison , the maximum pause-free velocity for bare DNA transcription is: ( 10 ) Vmax=k1k3k1+k3⋅d Using an optimal γu value of 0 . 6 , we solved Equations 9 and 10 and obtained k1 = 112 ± 30 s−1 ( indeed much slower than local DNA wrapping/unwrapping ) and k3 = 35 ± 3 s−1 for the wild-type . Importantly , these values show that the forward translocation rate is only three times faster than the catalysis rate and , therefore , has a significant contribution to the overall elongation velocity . For the mutant Pol II , translocation becomes even slower than catalysis ( k1 = 50 ± 4 s−1 and k3 = 195 ± 65 s−1 ) . The mutant’s higher k3 compensates for its lower k1 , rendering its overall velocity faster than that of the wild type . We note that these numbers were extracted by using the average values of the pause-free velocity and γu over the whole nucleosomal region . Such a simplifying treatment is based on the observations that both the pause-free velocity ( Figure 3 ) and the local DNA wrapping equilibrium ( Bintu et al . , 2012 ) do not change substantially along the NPS . The pause density , ρpause , is governed by the kinetic competition between pause entry and elongation . Previously , an overall elongation rate , which includes translocation , NTP binding , and catalysis , was used in the expression for ρpause ( Herbert et al . , 2006; Hodges et al . , 2009; Zhou et al . , 2011 ) . A more accurate treatment is to use the elementary rate constant in the elongation pathway directly connected to pausing , which is the net rate constant for forward translocation , k1net ( Figure 1; Equation 2 ) : ( 11 ) ρpause=kb1kb1+k1net=kb1kb1+[NTP]k−1 ( k−2+k3 ) k2k3+[NTP]⋅k1where kb1 is the rate constant of entering the 1-bp backtracked pausing state . At saturating NTP concentrations ( [NTP]>>k−1 ( k−2+k3 ) / ( k2k3 ) ) , k1net becomes equivalent to k1 . Hence ( 12 ) ρpause ( sat ) =kb1kb1+k1where ρpause ( sat ) is the pause density at saturating NTP concentration . In order to obtain a true pause density , the apparent ρpause needs to be corrected to include pauses shorter than 1 s that are missed by our pause detection algorithm . After such a correction ( ‘Materials and methods’ ) , the total ρpause ( sat ) is 0 . 045 ± 0 . 012 bp−1 . Solving Equation 12 yields kb1 = 5 . 3 ± 2 . 0 s−1 . This value is approximately five times larger than subsequent backward stepping rates , which are force-biased stepping rates obtained from Equation 7 ( kbn = 1 . 0 ± 0 . 3 s−1 , n≥2 ) . The difference between kb1 and kbn indicates that the first backtracking transition is easier to make than subsequent backtracking transitions . Using this value of kb1 , along with the value of γu obtained above , we can predict a nucleosomal pause density of 0 . 035 ± 0 . 015 bp−1 for pauses longer than 1 s , which agrees with the experimental measurement ( Table 2 ) . We then compared the pausing kinetics between the wild-type and the mutant enzymes . Interestingly , on bare DNA , the mutation only affects the distribution of pauses that are shorter than 2 s ( Figure 4A , p=0 . 003 , Kolmogorov-Smirnov test ) . In contrast , the distributions of longer pauses are indistinguishable between the mutant and the wild-type Pol II ( Figure 4—figure supplement 1 , p=0 . 9 ) . It is possible to rationalize this observation if the mutation selectively influences the kinetics of the first backtracking step ( kb1 and/or kf1 ) without affecting subsequent backtracking steps , given that pauses of short durations involve small backtracking excursions and that entering the 1-bp backtracked state is distinct from entering longer backtracked ones ( kb1 is different from kbn , n≥2 ) . The first backward stepping rate ( kb1 ) only influences the pause density but not the pause duration , while the first forward stepping rate ( kf1 ) does affect the pause duration . Specifically , the increase in short pauses can be explained if the mutation increases kf1 and accelerates the return from a pause to active elongation . Indeed , Monte Carlo kinetic simulations show that setting kf1 to be larger than 4 s−1—2 . 4-fold higher than the wild-type value ( 1 . 7 ± 0 . 4 s−1; Equation 6 ) —can reproduce the experimentally observed pause duration distributions for the mutant Pol II on bare DNA ( Figure 4A , blue dashed line ) and nucleosomal DNA ( Figure 4B , green dashed line , and Figure 4—figure supplement 2 ) . Moreover , by comparing the experimentally measured and simulated pause densities using different kb1 values , we can set a lower bound for the mutant’s kb1 to be 2 . 8 s−1 . 10 . 7554/eLife . 00971 . 014Figure 4 . Pause durations on bare DNA and nucleosomal DNA . ( A ) Cumulative distributions of the pause durations on bare DNA for the wild-type Pol II ( black solid line ) and the mutant enzyme ( blue solid line ) . The wild-type curve is fit to the one-dimensional random walk model for backtracked pausing ( gray dashed line ) . The blue dashed line represents the simulated pause duration distribution for the mutant enzyme , using a kf1 value of 4 s−1 . ( B ) Cumulative distributions of the pause durations in the nucleosome region for the wild-type enzyme ( red solid line ) and the mutant enzyme ( green solid line ) . The wild-type curve is fit to the one-dimensional diffusion model for backtracked pausing , using a γu value of 0 . 6 ( red dashed line ) . The green dashed line is the simulated pause duration distribution for nucleosomal DNA transcription by the mutant enzyme , using a kf1 value of 4 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 01410 . 7554/eLife . 00971 . 015Figure 4—figure supplement 1 . Cumulative pause duration distributions of pauses longer than 3 s . The curves for the wild-type and the E1103G mutant Pol II are statistically indistinguishable ( P = 0 . 9 , Kolmogorov-Smirnov test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 01510 . 7554/eLife . 00971 . 016Figure 4—figure supplement 2 . Comparison between the experimentally obtained distribution of pause durations and the simulated distribution for the nucleosomal DNA transcription by the mutant Pol II . The square of the difference between the experimental and simulated data is plotted as a function of kf1 used in the simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 016 Taken together , we have shown that the rate of entering the 1-bp backtracked state is higher than those of entering further backtracked states , and that the E1103G mutation modulates the transition kinetics between the 1-bp backtracked state and the pre-translocated state . Until now , kf1 and kb1 have been assumed to be identical with the other stepping rates during backtracking ( kfn and kbn , n≥2 ) ( Galburt et al . , 2007; Hodges et al . , 2009; Bintu et al . , 2012 ) . Our data here suggest that the first backtracking step should be treated differently , consistent with published structural data ( Wang et al . , 2009; Cheung and Cramer , 2011 ) ( see ‘Discussion’ ) . We have determined the rates of forward translocation ( k1 ) and catalysis ( k3 ) in the nucleotide addition cycle and shown that they are comparable . What remains unknown is the reverse translocation rate k−1 , which may also affect the elongation velocity under sub-saturating NTP conditions ( Equation 4 ) . To determine k−1 , we examined the pause densities measured at various NTP concentrations . Equation 11 can be re-written as: ( 13 ) ρpause=kb1kb1+[NTP]k−1Kk3+[NTP]⋅k1where K = ( k−2+k3 ) /k2 . The total ρpause as a function of [NTP] fits well to Equation 13 ( Figure 5A , B ) . Using the values of k1 , k3 , and kb1 determined above , we obtained k−1K equal to ( 4 . 7 ± 0 . 5 ) × 103 µM·s−1 and ( 2 . 5 ± 0 . 4 ) × 104 µM·s−1 for the wild-type and the mutant enzymes , respectively . 10 . 7554/eLife . 00971 . 017Figure 5 . Relationship between pause density and NTP concentration . ( A ) The total pause density for the wild-type Pol II ( black circles ) is plotted against the NTP concentration . The gray dashed line is the fit to Equation 13 ( R2 = 0 . 93 ) . ( B ) ρpause–[NTP] relationship ( blue squares ) for the mutant Pol II is fit to Equation 13 ( blue dashed line , R2 = 0 . 89 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 01710 . 7554/eLife . 00971 . 018Figure 5—figure supplement 1 . Constraining the value of K for the mutant Pol II . The apparent pause densities obtained experimentally for the E1103G Pol II are plotted against the NTP concentration ( blue squares ) . Simulated pause densities using a K value of 1 , 20 , 40 , 100 , and 200 μM are shown in blue , purple , red , orange , and green dashed lines , respectively . The simulated curve starts to deviate from the experimental data once K exceeds 100 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 018 We then revisited the relationship between the pause-free velocity and [NTP] ( Figure 2D ) , which follows Michaelis–Menten kinetics . According to Equation 5 , the Michaelis constant KM is expressed as: ( 14 ) KM=k1+k−1k1+k3⋅k−2+k3k2=k1+k−1k1+k3⋅K Plugging the values of KM , k1 , k3 , and k−1K into Equation 14 yields the values of K and k−1 for the wild-type Pol II: K = 9 . 2 µM and k−1 = 510 s−1 . We could further calculate the translocation equilibrium constant , Kδ = [pre-translocated]/[post-translocated] = k−1/k1 = 4 . 6 . This result indicates that the enzyme favors the pre-translocated state to the post-translocated one , in agreement with most previous reports ( Bar-Nahum et al . , 2005; Bai et al . , 2007; Kireeva et al . , 2008; Maoileidigh et al . , 2011 ) . For the mutant enzyme , we could set an upper bound of K to be 100 µM and a lower bound of k−1 to be 210 s−1 ( Figure 5—figure supplement 1 ) . Assuming that the mutant shares a similar K value with the wild-type , we calculated k−1 to be ∼2700 s−1 and Kδ to be ∼54 for the mutant Pol II ( see ‘Materials and methods’ for a discussion of this assumption ) . A central piece of evidence previously used to favor a branched kinetic scheme ( Figure 1—figure supplement 1 ) over a simpler linear scheme ( Figure 1 ) for the nucleotide addition cycle is the relationship between the pause-free velocity ( v ) and the applied force ( F ) ( Abbondanzieri et al . , 2005; Larson et al . , 2012 ) . However , in those studies , translocation was assumed to be in rapid equilibrium relative to catalysis . Having explicitly determined the translocation rates ( k±1 ) and found that the forward translocation rate ( k1 ) is comparable to the catalysis rate ( k3 ) , we went on to examine whether a linear kinetic scheme ( Figure 1 ) is sufficient to explain the F–v relationship , which for such scheme can be expressed as: ( 15 ) v ( F ) =k1 ( F ) ⋅k3k1 ( F ) +k3⋅[NTP] ( k1 ( F ) +k−1 ( F ) ) ⋅Kk1 ( F ) +k3+[NTP]⋅d We assume that only the translocation transitions in the nucleotide addition cycle are force-sensitive and that the translocation rates depend on force according to the Boltzmann-type equation: k1 ( F ) =k1 ( 0 ) ⋅eFδ/kBT , and k−1 ( F ) =k−1 ( 0 ) ⋅e−F⋅ ( 1−δ ) /kBT , where δ is the distance to the transition state for forward translocation , the only unknown variable left in Equation 15 . We measured the pause-free velocity at different applied forces for both wild-type and mutant enzymes and obtained values in good agreement with previously published single-molecule data ( Larson et al . , 2012 ) ( Figure 6A , B ) . The velocity of the wild-type enzyme shows a weak but detectable dependence on force , while the velocity of the mutant displays a much stronger force dependence . The F–v plots can be fit well to Equation 15 with δ of 0 . 46 ± 0 . 09 bp for the wild-type ( Figure 6A ) and 0 . 24 ± 0 . 05 bp for the mutant ( Figure 6B ) . Therefore , it is indeed possible to explain the observed force–velocity relationship of transcription elongation with a classic , non-branched Brownian ratchet mechanism , in which NTP binding occurs after translocation . 10 . 7554/eLife . 00971 . 019Figure 6 . Relationship between transcription velocity and applied force . ( A ) The pause-free velocity of the wild-type Pol II is plotted against the applied force . Experimental data in the present study are shown in solid squares ( error bars indicate SEM ) . Open triangles represent data from a previously published single-molecule study ( Larson et al . , 2012 ) . The combined data are fit to the force-velocity relationship predicted by a linear Brownian ratchet model ( dashed line ) , yielding a characteristic distance to the transition state δ = 0 . 46 ± 0 . 09 bp ( error is SEM , R2 = 0 . 88 ) . Positive and negative force values indicate assisting and opposing forces , respectively . ( B ) The force-velocity relationship for the mutant Pol II . δ = 0 . 24 ± 0 . 05 bp for the mutant ( R2 = 0 . 85 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 019 RNAP transcribes DNA through a multi-step kinetic pathway . The rate-limiting nature of the various steps in the nucleotide addition cycle has so far remained largely conjectural . Almost all the existing kinetic studies of transcription elongation relied on the major assumption that translocation and NTP binding follow rapid equilibrium kinetics . As a result , the catalytic step occurring after NTP binding has been assigned to be rate-limiting of the overall elongation reaction . The linear Brownian ratchet mechanism that assumes fast translocation equilibrium predicts that , as the NTP concentration increases , the force-sensitivity of the elongation velocity decreases and eventually vanishes , because the enzyme spends less time in the load-sensitive translocation steps . However , the F–v relationships of the enzyme obtained from optical tweezers studies have shown significant dependence of elongation velocity on external force even at saturating NTP concentrations ( Abbondanzieri et al . , 2005; Bai et al . , 2007; Larson et al . , 2012 ) , in contradiction to the above prediction . To account for this discrepancy , a modified , branched ratchet model was proposed in which the NTP must also bind to a secondary site on the polymerase in the pre-translocated configuration . Although the existence of such additional binding site may be rationalized by the downstream allosteric site ( Holmes and Erie , 2003; Gong et al . , 2005 ) , the ‘E’ site or pre-insertion site ( Westover et al . , 2004; Temiakov et al . , 2005 ) , or the tilted hybrid structure ( Cheung et al . , 2011 ) , whether it constitutes a significant pathway in the elongation reaction and how it is related to the primary nucleotide binding pathway remain obscure . More importantly , the branched model neglects the possibility that the translocation steps may not be as fast as assumed . In this study , we tested this possibility of slow translocation by placing a nucleosome in the path of the transcribing polymerase and directly determining the rates of forward and reverse translocation . Our analyses show that the forward translocation rate is in fact within the same order of magnitude as the catalysis rate . For the wild-type Pol II , k1 is only 2 . 5 times higher than k3 ( Figure 7A; Table 3 ) . For the E1103G mutant , k1 even becomes the slowest step in the nucleotide addition cycle ( Table 3 ) . Hence , the translocation step is one of the rate-limiting transitions during transcription elongation . Translocation and catalysis together control the overall elongation velocity . These findings naturally explain the observed F–v relationship: because the enzyme always spends a considerable amount of time in the force-sensitive pre-translocated state even at high [NTP] , we should always expect a force-dependence of the velocity . Moreover , a lower k1 renders the velocity more sensitive to force , consistent with the experimental observation that the mutant Pol II shows a steeper F–v curve than the wild-type ( Figure 6A , B ) . Therefore , our results demonstrate that a linear ratchet model can explain the transcriptional kinetics of Pol II and that it is not necessary to invoke a conceptually more complicated branched model , as long as the constraint of fast translocation equilibrium is relieved . Note that although our data argue against rapid oscillation of the ratchet , they still support the notion that the enzyme is able to spontaneously diffuse along the DNA between the pre- and post-translocated states , as suggested by the Brownian ratchet mechanism . 10 . 7554/eLife . 00971 . 020Figure 7 . A quantitative kinetic model for transcription elongation . ( A ) A comprehensive kinetic characterization of the nucleotide addition phase ( highlighted in green ) and the pausing phase ( highlighted in blue ) for transcription by the wild-type Pol II . Inside the yellow box are the transitions affected by the nucleosomal barrier . ( B ) The schematic translocation free energy landscape at a given RNA length for the wild-type Pol II ( solid black ) and the E1103G Pol II ( dashed cyan ) . The on-pathway elongation is highlighted in green and the off-pathway pausing is highlighted in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 02010 . 7554/eLife . 00971 . 021Figure 7—figure supplement 1 . The schematic three-dimensional free energy landscape for transcription elongation by the wild-type Pol II at 1 mM NTP and zero force . The ribbons represent the minimal energy paths of the nucleotide addition cycle ( green ) and the off-pathway processes ( blue ) . The nomenclature for the TECs ( e . g . , n , 1 ) is the same as that used in Figure 7 . Chemical and mechanical transitions are shown in two orthogonal axes . Mechanical perturbations , such as force , affect the mechanical transitions of the enzyme by tilting the landscape around the chemical axis to a first approximation , while chemical perturbations , such as [NTP] and [PPi] , rotate the landscape around the mechanical axis , again to a first approximation . Two-dimensional projections on the grids highlight the relative free energy of each state . DOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 02110 . 7554/eLife . 00971 . 022Table 3 . Summary of kinetic parameters measured in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00971 . 022ParametersWild-type Pol IIE1103G Pol IIk1 ( s−1 ) 88 ± 2344 ± 4k−1 ( s−1 ) ∼680∼4 . 1 × 103Kδ = k−1/k1∼7 . 7∼92K = ( k−2+k3 ) /k2 ( μM ) ∼9 . 2∼9 . 2k3 ( s−1 ) 35 ± 3195 ± 65kb1 ( s−1 ) 6 . 9 ± 2 . 6∼3 . 7kf1 ( s−1 ) 1 . 3 ± 0 . 3∼3 . 1kbn ( s−1 ) , n ≥ 21 . 3 ± 0 . 31 . 3 ± 0 . 3kfn ( s−1 ) , n ≥ 21 . 3 ± 0 . 31 . 3 ± 0 . 3The values reported in the text were measured at 6 . 5 pN of applied assisting force and are normalized to zero force here . The italicized numbers indicate the parameters that are altered by the E1103G mutation . The asterisks indicate lower bounds of the corresponding values . We extracted the values of k1 and k3 by comparing the maximum pause-free velocities on bare DNA and nucleosomal DNA ( Equations 9 and 10 ) . In principle , k1 and k3 can also be determined by examining Vmax as a function of applied force: ( 16 ) Vmax ( F ) =k1 ( F ) ⋅k3k1 ( F ) +k3⋅dwhere k1 ( F ) =k1 ( 0 ) ⋅eFδ/kBT . Using our data and the previously published data ( Larson et al . , 2012 ) collected at saturating [NTP] ( 1 mM ) and various forces ( Figure 6 ) , we fit the Vmax–F dependence to Equation 16 and obtained the values of k1 = 87 ± 61 s−1 , k3 = 33 ± 8 s−1 , and δ = 0 . 64 ± 0 . 58 bp for the wild-type Pol II , and k1 = 65 ± 37 s−1 , k3 = 62 ± 32 s−1 , and δ = 0 . 64 ± 0 . 50 bp for the mutant Pol II . Thus , the same qualitative conclusion that both translocation and catalysis are rate-limiting for the elongation reaction can be drawn from this alternative approach . Compared to the approach of using the nucleosomal barrier as a tool to determine k1 and k3 , fitting the Vmax–F relationship involves one additional free parameter ( δ ) and the values are less constrained ( larger errors ) . In the future , it is worthwhile to use either of these two approaches or both to test whether prokaryotic transcription also employs a linear ratchet mechanism . With the same transcript length , RNAP is able to move back and forth on the DNA template , forming different TEC configurations ( Figure 7A , B ) . Each translocation state corresponds to a local energy minimum ( Yager and von Hippel , 1991; Bai et al . , 2004; Tadigotla et al . , 2006 ) . Transitions between the pre- and post-translocated states , together with NTP binding and catalysis , constitute the active elongation pathway ( Figure 7—figure supplement 1 , green ) . The enzyme can also enter the pausing pathway by transiting from the pre-translocated state to the backtracked states ( Figure 7—figure supplement 1 , blue ) . The hyper-translocated states , in which the enzyme undergoes further forward translocation beyond 1 bp , are energetically unfavorable . The rate constants extracted from our single-molecule experiments translate into a free energy landscape for Pol II’s mechanical translocations and chemical transitions ( Figure 7B , Figure 7—figure supplement 1; ‘Materials and methods’ ) , which reveals many detailed features of the kinetics of Pol II transcription . First , the staircase shape formed by the energy minima of post-translocated , pre-translocated , and 1-bp backtracked states shows that the off-pathway backtracked states are thermodynamically more stable than the on-pathway states ( Figure 7B ) . This feature confers the enzyme its propensity to enter the pausing pathway , which is the central mechanism for various types of transcriptional control , such as arrest , proofreading , co-transcriptional RNA folding , and recruitment of regulators . Second , the energy barrier from the pre-translocated to the 1-bp backtracked state is 2 . 5 kBT higher than the barrier from the pre-translocated to the post-translocated state , causing k1 to be more than 10 times faster than kb1 . Thus , at the beginning of each nucleotide addition cycle , the pre-translocated TEC favors the catalysis-competent post-translocated state kinetically over the 1-bp backtracked state , even though it is thermodynamically more favorable to move in the opposite direction . This property ensures that pausing only occurs sporadically so that the transcript can be synthesized within a reasonable amount of time . In addition , the barriers between neighboring backtracked states are also relatively high , preventing the enzyme from backtracking too far , which could lead to transcriptional arrest . Third , the first backtracking step appears to be unique from further backtracking steps in two aspects . Kinetically , entering the 1-bp backtracked state is easier than entering subsequent backtracked states , as reflected by the difference between kb1 and kbn ( n≥2 ) . Such a difference is supported by structural data: the structure of an arrested Pol II complex suggests that backtracking beyond 1 bp is disfavored as it is sterically hindered by a ‘gating’ tyrosine ( Rpb2-Y769 ) ( Cheung and Cramer , 2011 ) . Thermodynamically , transiting from the pre-translocated state to the 1-bp backtracked state is favorable , while backtracking for more steps yields no additional energetic benefit . This result can also find structural support: the first backtracked nucleotide is stabilized by a binding pocket formed by several Pol II residues , whereas the second or third backtracked nucleotide makes no additional contact to the enzyme ( Wang et al . , 2009 ) . Thus , our model depicts an enzyme with a delicate balance between active elongation and inactive pausing ( von Hippel and Pasman , 2002 ) . This model can serve as a framework to study the effects of DNA sequence and nascent RNA structure on transcriptional dynamics ( Bai et al . , 2004; Tadigotla et al . , 2006; Zamft et al . , 2012 ) . Moreover , this model may improve our understanding of the control of transcription fidelity . The 1-bp backtracked state is closely associated with the proofreading process of Pol II , as the enzyme in this location preferentially cleaves the 3′ dinucleotide of the RNA containing the mismatched base , empting the active site for NTP binding ( Wang et al . , 2009 ) . It is possible that nucleotide misincorporation slows down forward translocation , thereby promoting the entry to the pausing pathway and the removal of the dinucleotide . It is worth noting that we cannot definitively rule out the alternative scenario in which the first unique pausing state corresponds to a non-backtracked intermediate . Nonetheless , no evidence has been found for the universal occurrence of such an intermediate in Pol II transcription . The interpretation that most pauses in Pol II transcription are caused by enzyme backtracking is more parsimonious , especially given the corroborating structural data mentioned above . The kinetic characterization of the E1103G mutant Pol II reveals that this TL mutation results in many modifications to the enzyme dynamics ( Table 3 ) . Between the pre-translocated state and the post-translocated state , the mutant is significantly more biased toward the former than the wild type ( Figure 7B ) . This property , together with its lower forward translocation rate , renders the mutant’s elongation velocity more sensitive to perturbations of its forward translocation , such as an externally applied force ( Figure 6B ) or the presence of a nucleosomal barrier ( Figure 3B ) . It has been shown that the inter-conversion between pre- and post-translocated states involves the transitions of the TL between an open conformation and a wedged conformation ( Brueckner and Cramer , 2008 ) . It is plausible that the mutation modulates the enzyme’s translocation kinetics by altering the rates of transition between these two conformations . Furthermore , our analyses lead to the conclusion that the faster overall elongation velocity of the mutant is due to its much greater catalysis rate despite a slower translocation step . The increase of the catalysis rate is most likely due to a faster NTP sequestration step induced by the closure of the TL ( Kireeva et al . , 2008 ) . The lack of hydrogen bonding between T1095 and the mutated E1103 residue may destabilize the open state of the TL and speed up its closure ( Walmacq et al . , 2012 ) . The E1103G mutation also affects the pausing kinetics . Specifically , a decrease in the activation energy required to return from the first backtracked state to the pre-translocated state accelerates the recovery from a pause ( Figure 7B ) . Consequently , the mutant populates the 1-bp backtracked state less than the wild-type . This property might affect the overall fidelity of transcription . It has been previously shown that E1103G mutation strongly promotes incorporation of non-cognate NMP and mismatch extension ( Kaplan et al . , 2008; Kireeva et al . , 2008 ) . The destabilization of the 1-bp backtracked state relative to the pre-translocated state in the E1103G mutant , established in this work , is consistent with its efficient mismatch extension and suggests that this mutation might also confer a defect in proofreading activity . Together , our results suggest that the dynamics of TL are involved in multiple phases of transcription elongation , including translocation , catalysis , and pausing . In vivo , various transcription factors and small molecules can directly manipulate the TL dynamics and regulate transcription elongation . For example , transcription factor IIS ( TFIIS ) stimulates the endonuclease activity of Pol II by replacing the TL with its zinc finger domain , and thus , rescues transcription elongation by creating a new 3′-end of the transcript at Pol II’s active site ( Kettenberger et al . , 2003 ) . In fact , the viability of yeast cells expressing only the E1103G mutant Pol II is strictly dependent on TFIIS ( Malagon et al . , 2006 ) . It is interesting to investigate how these trans-acting factors modify the rate-limiting mechanism and detailed kinetics of the elongation reaction . Finally , the elementary rate constants extracted from our analyses should provide a reference frame for future computational studies aiming to fully describe the molecular trajectory of a transcribing polymerase . Biotinylated wild-type and E1103G S . cerevisae Pol II ( unphosphorylated C-terminal domain ) were purified as previously described ( Kireeva et al . , 2005 ) . The 3-kb DNA handle was prepared by PCR from Lambda DNA ( NEB , Ipswich , MA ) using a digoxigenin-labeled primer . The 574-bp DNA template was prepared by PCR from a modified pUC19 plasmid ( Zhang et al . , 2006 ) containing the 601 nucleosome positioning sequence ( NPS ) ( Lowary and Widom , 1998 ) . Each histone protein was recombinantly expressed and purified from E . coli , reconstituted to octamers ( Wittmeyer et al . , 2004 ) , and loaded on the NPS-containing DNA using salt gradient dialysis ( Thåström et al . , 2004 ) . The transcription elongation complexes ( TECs ) were assembled by annealing a 9-nt RNA primer ( IDT , Coralville , IA ) to a 93-nt template DNA , incubating the hybrid with a biotinylated Pol II , and subsequently annealing a 96-nt complementary DNA using previously published sequences and procedures ( Hodges et al . , 2009 ) . The TEC was walked to a stall site by addition of ATP , CTP and GTP . In the assisting force geometry , the downstream end of the stalled TEC was ligated to the 574-bp DNA containing the 601 NPS ( with or without a preloaded nucleosome ) , while its upstream end was ligated to the 3-kb DNA handle . In the opposing force geometry , the downstream end of the TEC was ligated to a 4-kb DNA amplified from Lambda DNA ( Zamft et al . , 2012 ) . The complexes were incubated with 2 . 1-μm streptavidin-coated beads ( Spherotech , Lake Forest , IL ) , and DNA tethers were formed in a dual-trap optical tweezers instrument by attaching the digoxigenin-labeled DNA handle to a 2 . 1-μm anti-digoxigenin IgG-coated bead . In the assisting force geometry , Pol II and its upstream DNA were under tension , while no external force was applied to the downstream nucleosome ( Figure 2A ) . The tension in the upstream DNA prevented intra-nucleosomal loop transfer and thus ensured that the nucleosome was always ahead of the transcribing polymerase . Transcription was restarted in optical tweezers by addition of NTPs ( Thermo Fisher Scientific , Waltham , MA ) . The transcription buffer contains 20 mM Tris-HCl ( pH 7 . 9 ) , 5 mM MgCl2 , 10 μM ZnCl2 , 1 mM β-mercaptoethanol , 1 μM pyrophosphate , 300 mM KCl , and NTPs ranging from 35 μM to 2 mM each . Position data were recorded at 2 kHz , averaged and decimated to 50 Hz , and filtered using a second-order Savitzky-Golay filter with a time constant of 1 s . The contour length of the DNA was calculated from the extension and force using the worm-like-chain formula of DNA elasticity ( Bustamante et al . , 1994 ) with a persistent length of 30 nm . This value of persistent length was obtained from pulling 3-kb DNA in our transcription buffer ( data not shown ) . To alleviate calibration error and improve positional accuracy , single-molecule transcription traces that passed 85% of the template were aligned using both the stall site and the expected run-off length ( Bintu et al . , 2012 ) . Shorter traces were also proportionally extended based on the average error from the run-off traces . To identify pauses , we computed the dwell time of Pol II at each nucleotide position . Pauses were identified from dwell times that were longer the average dwell time by at least a factor of two . Due to the limited spatial resolution , we joined pauses that were separated by 3 bp or fewer into a single continuous pause . Pauses longer than 1 s are most likely caused by backtracking ( Maoileidigh et al . , 2011 ) and were counted . Pause-free velocities were calculated from time derivatives of the filtered position data , with a threshold of 2 nt/s to remove pauses . All curve fittings were performed by non-linear regression of the means weighted by the inverse of the variance . From an elongation-competent state , Pol II can either elongate by 1 nt with the net forward translocation rate k1net and incorporate an NMP to the RNA transcript , or enter a backtracked pause by 1 nt . During a pause , Pol II diffuses forward and backward with force-biased rate constants kf and kb , respectively . For each condition , we simulated 100 trajectories and extracted the pause durations and densities to compare with the experimentally measured values . Fluorescence correlation spectroscopy and fluorescence resonance energy transfer experiments showed that the first 20–30 bp of DNA at the nucleosome ends spontaneously unwrap and rewrap on the histone surface every 10–250 ms ( Li et al . , 2005; Koopmans et al . , 2008 ) . The timescale of the 1-bp DNA fluctuations has not been directly reported but can be estimated from the experimental results above for longer DNA fluctuations . Assuming the wrapping/unwrapping kinetics is uniform along the DNA , we can model the unwrapping of a 25-bp DNA segment as:0⇄kwku1⇄kwku2⇄kwku3…⇄kwku24→ku25where ku and kw are the local unwrapping and wrapping rate constants of each basepair , respectively . Since the local wrapping equilibrium constant has been shown to be close to 1 ( Hodges et al . , 2009; Bintu et al . , 2012 ) , we further approximate ku and kw with a single value k:0⇄kk1⇄kk2⇄kk3…⇄kk24→k25 A net rate constant can substitute for each pair of forward and reverse rate constants ( Cleland , 1975 ) :0→k0→1net1→k1→2net2→k2→3net3…→k23→24net24→k24→25net25 The net rate constants are given by:k24→25net=kk23→24net=k⋅k24→25netk24→25net+k=k2k22→23net=k⋅k23→24netk23→24net+k=k3⋮k1→2net=k⋅k2→3netk2→3net+k=k24k0→1net=k⋅k1→2netk1→2net+k=k25 The time required for unwrapping 25 bp of DNA equals the total time of unwrapping each bp of DNA:τ0→25=1k0→1net+1k1→2net+…+1k24→25net=325k≈ ( 10−250 ) ms Thus , the time for 1-bp DNA to unwrap from the nucleosome is expected to be less than 1 ms:τ0→1=1k<1ms In the same way , we can also show that the 1-bp DNA rewrapping occurs on a similar timescale ( τ1→0 < 1 ms ) . In addition , molecular dynamics simulations also suggested that the local nucleosomal DNA fluctuates very fast ( ns–µs timescale ) ( Voltz et al . , 2012 ) . Therefore , we assume that the 1 bp of DNA in front of the polymerase unwraps and rewraps much faster than the translocation of the enzyme . Experimentally we only counted pauses with lifetimes between 1 s and 120 s . The total pause density ρpause , total is given by:ρpause , total=kbkb+k1net=ρpause , 1<t<120∫1120ψ ( t ) dt The correction factor can be solved analytically to be 2 . 9 for the wild-type Pol II . For the mutant enzyme , the values of kf1 and kb1 are different from those for the wild-type ( Table 3 ) . We simulated transcriptional pauses using the lower bounds of kf1 and kb1 and obtained a correction factor of ∼7 for the mutant Pol II . The value of K for the mutant Pol II ( Kmutant ) cannot be constrained by Equation 14 due to the relatively large experimental error . We took a different approach to constrain Kmutant by simulating the ρpause–[NTP] relationship with varying Kmutant values and then comparing it to the experimental data ( Figure 5—figure supplement 1 ) . We found that the simulated curve substantially deviates from the experimental curve when Kmutant becomes larger than 100 µM . Hence , we set the upper bound of Kmutant to be 100 µM . Using the k−1K value of ( 2 . 5 ± 0 . 4 ) × 104 µM·s−1 obtained from Equation 13 , we set the lower bound of k−1 for the mutant to be 210 s−1 . The notion that the NTP dissociation rate is much faster than the catalysis rate ( k−2 >> k3 ) has been widely used in the kinetic studies of RNA and DNA polymerases , and is supported by biochemical evidence ( Rhodes and Chamberlin , 1974; Johnson , 1993; Foster et al . , 2001; Bai et al . , 2004; Maoiléidigh et al . , 2011 ) . It follows from this notion that K = ( k−2+k3 ) /k2 ≈ k−2/k2 . Thus , K becomes virtually identical to KD , the NTP dissociation constant . Because the mutated residue ( Glu1103 ) is located distal from the NTP-interacting part of TL ( Wang et al . , 2006 ) and the E1103G mutation affects TL closure and NTP sequestration after the initial docking step ( Kireeva et al . , 2008 ) , the NTP binding/dissociation kinetics are unlikely to be markedly affected by the mutation . Therefore , it is reasonable to assume that the wild-type and the mutant enzymes share similar KD values ( ∼9 . 2 µM ) . Under this assumption , we could estimate the k−1 value for the mutant to be ∼2700 s−1 . The free energy difference ( ΔΔG ) between two neighboring translocation states was computed using the forward and reverse rate constants between these states ( k+ and k− ) :ΔΔG=−kBT⋅ln ( k−k+ ) The height of the activation energy barrier ( ΔG† ) was calculated according to the Arrhenius equation:k=A⋅exp ( −ΔG†/kBT ) where k is the corresponding rate constant and A is the pre-exponential factor . In this study , for illustration purposes , we made a simplifying assumption that all reaction steps share the same pre-exponential factor . A was calculated using the stepping rate constant during backtracking k0 = 1 . 3 s−1 and the barrier height between neighboring backtracked states ΔGb† . We assumed that one DNA–RNA hybrid basepair and one DNA–DNA basepair in the transcription bubble must be broken before any other bonds are formed and that no other interactions contribute to the barrier ( Tadigotla et al . , 2006 ) . Using the available free energy data for basepairing ( Sugimoto et al . , 1996; Wu et al . , 2002 ) , we estimated ΔGb† to be ∼8 . 5 kBT , which translates to an Arrhenius pre-factor of ∼6 . 4 × 103 s−1 . The catalysis step is essentially irreversible in our experimental condition . The free energy drop after each nucleotide addition cycle is arbitrarily set to be 10 kBT .	The production of a protein inside a cell starts with a region of the DNA inside the cell nucleus being transcribed to form a molecule of messenger RNA . This process involves an enzyme called RNA polymerase that moves along the DNA , reading the bases and making a complementary strand of messenger RNA from molecules called nucleoside triphosphates ( NTPs ) . Just as there are four different bases in DNA , there are four different natural NTPs . In addition to supplying the correct bases for the messenger RNA molecule , these NTPs also provide the energy needed to drive the transcription process . In many species the RNA polymerase oscillates between two neighbouring positions on the DNA , with this back-and-forth motion–which is powered by thermal energy–being converted into forward movement of the enzyme along the DNA when a new NTP binds to the growing messenger RNA molecule . It has long been assumed that the back-and-forth motion occurs much faster than the overall reaction of adding one NTP to the messenger RNA . This assumption has now been tested by using a single-molecule assay to monitor transcription in real time . Dangkulwanich et al . measured the elongation velocities of yeast RNA polymerase II ( Pol II ) on bare DNA and on DNA in which a nucleosome–a structure that consists of a segment of DNA wrapped around histone proteins–had been placed as a “road block” in front of the enzyme . Surprisingly , the rate of the back-and-forth motion was found to be comparable in magnitude to the rate for adding one molecule of NTP . Dangkulwanich et al . also measured the rates associated with a process called backtracking in which the polymerase moves away from the transcription site to “pause” the process . These measurements show that there is a delicate balance between elongation and pausing during transcription . Overall , by revealing the energy landscape associated with transcription , the work of Dangkulwanich et al . will bring us closer to the goal of creating a molecular movie of this extremely important–and complex–process .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2013	Complete dissection of transcription elongation reveals slow translocation of RNA polymerase II in a linear ratchet mechanism
Feeding and oviposition deterrents help phytophagous insects to identify host plants . The taste organs of phytophagous insects contain bitter gustatory receptors ( GRs ) . To explore their function , the GRs in Plutella xylostella were analyzed . Through RNA sequencing and qPCR , we detected abundant PxylGr34 transcripts in the larval head and adult antennae . Functional analyses using the Xenopus oocyte expression system and 24 diverse phytochemicals showed that PxylGr34 is tuned to the canonical plant hormones brassinolide ( BL ) and 24-epibrassinolide ( EBL ) . Electrophysiological analyses revealed that the medial sensilla styloconica of 4th instar larvae are responsive to BL and EBL . Dual-choice bioassays demonstrated that BL inhibits larval feeding and female oviposition . Knock-down of PxylGr34 by RNAi attenuates the taste responses to BL , and abolishes BL-induced feeding inhibition . These results increase our understanding of how herbivorous insects detect compounds that deter feeding and oviposition , and may be useful for designing plant hormone-based pest management strategies . Many phytophagous insects have evolved to select a limited range of host plants . Understanding the ultimate and proximate mechanisms underlying this selection strategy is a crucial issue in the field of insect–plant interactions . Insects’ decisions to feed and oviposit are mainly based on information carried via chemosensory systems ( Schoonhoven et al . , 2005 ) . Insects discriminate among potential hosts after perceiving a combination of lineage-specific and more ubiquitous chemicals synthesized by plants . Some nutrients are ubiquitous across plant taxa . For example , various sugars and amino acids are feeding stimulants for the majority of herbivorous insects . Secondary plant metabolites occur in certain plant taxa at much higher concentrations than in others , and therefore are of greater significance in host-plant selection by insects ( Jermy , 1966 ) . They are usually deterrent or ‘bitter’ compounds for many phytophagous insects , except for some specialized species that use them as token stimuli ( Schoonhoven et al . , 2005 ) . Phytophagous insects have sophisticated taste systems to recognize deterrent or stimulant compounds , which direct their feeding and oviposition behavior ( Yarmolinsky et al . , 2009 ) . The taste sensilla of insects are mainly situated on the mouthparts , tarsi , ovipositor , and antennae ( Bernays and Chapman , 1994 ) . These sensilla take the form of hairs or cones with a terminal pore in the cuticular structure , and often contain the dendrites of three or four gustatory sensory neurons ( GSNs ) . The axons of GSNs synapse directly onto the central nervous system . Analyses using the tip-recording technique for taste sensilla led to the discovery of a ‘deterrent neuron’ in the larvae of Bombyx mori ( Ishikawa , 1966 ) and Pieris brassicae ( Ma , 1969 ) . Since then , GSNs coding for secondary plant metabolites have been identified in maxillary sensilla in larvae and tarsal sensilla of adults of many Lepidopteran species ( Glendinning et al . , 2002; Zhou et al . , 2009 ) . However , the molecular basis of these deterrent neurons remains unclear . Gustatory receptors ( GRs ) expressed in the dendrites of GSNs determine the selectivity of the response of GSNs ( Thorne et al . , 2004; Wang et al . , 2004 ) . Since the first insect GRs were identified in the model organism Drosophila melanogaster ( Clyne et al . , 2000 ) , the function of some of its bitter GRs have been revealed ( Dweck and Carlson , 2020; Freeman and Dahanukar , 2015 ) . Five D . melanogaster GRs ( Gr47a , Gr32a , Gr33a , Gr66a , and Gr22e ) are involved in sensing strychnine ( Lee et al . , 2010; Lee et al . , 2015; Moon et al . , 2009; Poudel et al . , 2017 ) . Nicotine-induced action potentials are dependent on Gr10a , Gr32a , and Gr33a ( Rimal and Lee , 2019 ) . Gr8a , Gr66a , and Gr98b function together in the detection of L-canavanine ( Shim et al . , 2015 ) . Gr33a , Gr66a , and Gr93a participate in the responses to caffeine and umbelliferone ( Lee et al . , 2009; Moon et al . , 2006; Poudel et al . , 2015 ) . Gr28b is necessary for avoiding saponin ( Sang et al . , 2019 ) . However , only a few studies have focused on the function of bitter GRs in phytophagous insects . PxutGr1 , a bitter GR in Papilio xuthus , was found to respond specifically to the oviposition stimulant , synephrine ( Ozaki et al . , 2011 ) . In recent studies , the insect bitter GRs BmorGr16 , BmorGr18 , and BmorGr53 showed response to coumarin and caffeine in vitro , and coumarin was found to have a feeding deterrent effect on B . mori larvae ( Kasubuchi et al . , 2018 ) ; another bitter GR in B . mori , Gr66 , was reported to be responsible for the mulberry-specific feeding preference of silkworms ( Zhang et al . , 2019 ) . Although some progress has been made in determining the functions of such receptors in Drosophila and a few herbivorous insects , we still lack basic mechanistic information about the functions of bitter GRs from most herbivorous insect lineages . Plutella xylostella ( L . ) is the most widespread Lepidopteran pest species , causing losses of US$ 4–5 billion per year ( You et al . , 2020; Zalucki et al . , 2012 ) . It has developed resistance to the usual insecticides because of its short life cycle ( 14 days ) ( Furlong et al . , 2013 ) . P . xylostella mainly selects Brassica species as its host plants , and its females pat the leaf surfaces with their antennae before egg laying ( Qiu et al . , 1998 ) . This behavior may be related to certain chemical components in leaves of Brassica species , including sugars , sugar alcohols , amino acids , amines , glucosinolates , and plant hormones . Among these compounds , sinigrin and brassinolide ( BL ) have relatively higher concentrations in Brassica than in many other plant species ( Fahey et al . , 2001; Lv et al . , 2014 ) . Sinigrin is known to be a feeding/oviposition stimulant for P . xylostella ( Gupta and Thorsteinson , 1960 ) . The medial sensilla styloconica in the maxillary galea of P . xylostella larvae contain a GSN sensitive to sinigrin and other glucosinolates ( van Loon et al . , 2002 ) . BL is a ubiquitous plant hormone that has been widely studied in relation to its role in plant growth and development ( Clouse and Sasse , 1998 ) , but little is known about its effects on the behavior of phytophagous insects . To uncover the molecular basis of the perception of feeding/oviposition stimulants and deterrents by P . xylostella , we re-examined all the GRs reported in previous studies on this insect . Through transcriptome analysis and qPCR , we identified one bitter GR ( PxylGr34 ) highly expressed in the larval head and the adult antennae . We functionally analyzed this GR with the Xenopus oocyte expression system and RNAi , and found that PxylGr34 is tuned to BL as a feeding and oviposition deterrent in P . xylostella . To search for candidate GRs that may be involved in host selection by P . xylostella , we searched for candidates among those that had been annotated in the P . xylostella genome ( Engsontia et al . , 2014; You et al . , 2013 ) , the transcriptome ( Yang et al . , 2017 ) , and the P . xylostella GRs deposited in GenBank ( Supplementary file 1 ) . We obtained 79 annotated GRs ( 69 GRs from the genome , 7 GRs from the transcriptome , and 3 GRs from GenBank ) . After removing repetitive sequences and deleting paired sequences with amino acid identity greater than 99% , we validated 67 of the GR sequences ( 61 GRs from the genome , 3 GRs from the transcriptome , and 3 GRs from GenBank ) ( Figure 1 and Supplementary file 1 ) . On this basis , we first analyzed the transcriptomes of the antennae , forelegs ( only tibia and tarsi ) , and head ( without antennae ) of adults , and the mouthparts of 4th instar larvae although the genomic data and antennae transcriptome of P . xylostella had been reported previously ( Yang et al . , 2017; You et al . , 2013 ) . Next , we looked specifically for candidates that were highly expressed in these chemosensory organs by calculating the transcripts per million ( TPM ) values of these 67 GRs ( Figure 2A ) . Those GRs with high expression levels in both adult and larval taste organs were considered to be candidates for those with key roles in the host-plant selection of this species . Intriguingly , the TPM value of PxylGr34 , which clustered with bitter GRs , was much higher than those of other GR genes in the antennae , head and forelegs of adults , as well as the mouthparts of larvae ( Figure 2A ) . PxylGr34 was originally annotated from genomic data ( Engsontia et al . , 2014 ) , and then detected in the antennal transcriptomic data of P . xylostella ( named as ‘PxylGr2’ in Yang et al . , 2017 ) . However , both studies provided only its partial coding sequences ( Figure 1—figure supplement 1 ) . Based on our transcriptomic data , we obtained the full-length coding sequence of PxylGr34 through gene cloning and Sanger sequencing ( Figure 1—figure supplement 1 ) . The protein encoded by PxylGr34 is typical of most GRs with seven transmembrane domains , and a full open reading frame ( ORF ) of 418 amino acids ( Figure 1—figure supplement 2 ) . To further confirm the expression patterns of PxylGr34 in the larvae and adults , we detected its relative transcript levels in different tissues of adults and the 4th instar larvae of P . xylostella using quantitative real-time PCR ( qPCR ) . The larvae eat the most in the 4th instar and can forage around the plant more easily ( Harcourt , 1957 ) . The high levels of PxylGr34 transcripts were detected in the larval head . They were also detected in the larval thoracic legs and gut ( Figure 2B ) . In the adults , PxylGr34 transcripts were restricted to the antennae ( Figure 2C ) . We used the Xenopus laevis oocyte expression system and two-electrode voltage-clamp recording to study the function of PxylGr34 . Among 24 tested phytochemicals belonging to sugars , sugar alcohols , amino acids , amines , glucosinolates , and plant and insect hormones ( Key Resources Table ) , BL induced a strong response in the oocytes expressing PxylGr34 , as did its racemate EBL at a concentration of 10–4 M ( Figure 3A and Figure 3B ) . The currents induced by BL increased from the lowest threshold concentration of 10–4 M in a dose-dependent manner ( Figure 3C and Figure 3D ) . Oocytes expressing PxylGr34 showed weak responses to methyl jasmonate and allyl isothiocyanate , but no response to 20-hydroxyecdysone and other tested compounds ( Figure 3A and Figure 3B ) . As negative controls , the water-injected oocytes failed to respond to any of the tested chemical stimuli ( Figure 3—figure supplement 1 ) . Next , using the tip-recording technique , we examined whether any gustatory sensilla in the mouthparts of larvae of P . xylostella could respond to BL and EBL . Of the two pairs of sensilla styloconica in the maxillary galea of 4th instar larvae , the lateral sensilla styloconica had no response to BL and EBL ( Figure 4A , B , C and D ) ; the medial sensilla styloconica exhibited vigorous responses to BL and EBL at 3 . 3 × 10−4 M , and the spike amplitudes induced by BL and EBL were about the same ( Figure 4A , B , C and D ) . As previously reported , the medial sensilla styloconica also exhibited vigorous responses to sinigrin . However , the spike amplitudes induced by sinigrin were larger than those induced by BL and EBL ( Figure 4E and F ) . These results suggest that BL and EBL activate the same neuron , while sinigrin activates a different neuron in the sensillum . The medial sensilla styloconica showed a dose-dependent response to BL , although the testing concentrations were limited because of the low solubility of BL in water ( highest concentration approximately 3 . 3 × 10−4 M ) ( Figure 5 ) . We further tested the effects of BL and EBL on the larval feeding behavior of P . xylostella on pea leaves in the dual-choice leaf disc assay . In a dual-choice feeding test with 4th instar larvae , the feeding areas of larvae were significantly smaller on the leaf discs treated with BL and EBL at concentrations of 10−4 M and above than on the control leaf discs . In addition , the feeding preference of larvae to BL and EBL tended to decrease with increasing BL and EBL concentrations ( Figure 6 and Figure 6—figure supplement 1 ) , indicating that both BL and EBL function as feeding deterrents to P . xylostella larvae . We also tested the effects of BL on the female ovipositing behavior of P . xylostella . In a dual-choice oviposition test with mated females , significantly fewer eggs were laid on the sites treated with BL at 10−4 M , 10−3 M , and 10−2 M than on the control sites . In addition , the oviposition preference tended to decrease as the BL concentrations increased ( Figure 7 ) . To clarify whether PxylGr34 mediates the electrophysiological and behavioral responses of P . xylostella larvae to BL in vivo , we tested the effect of siRNA targeting PxylGr34 on the responses of medial sensilla styloconica and the feeding behavior of the 4th instar larvae . The relative transcript level of PxylGr34 in the head of larvae treated with PxylGr34 siRNA was half that in the head of larvae treated with green fluorescent protein ( GFP ) siRNA or ddH2O ( Figure 8A ) . This confirmed that feeding with siRNA is an effective method for RNAi-inhibition of PxylGr34 in the larval head . To test the effects of RNAi-inhibition of PxylGr34 on the taste responses , the medial sensilla styloconica from siRNA-treated larvae were subjected to a tip-recording analysis as described above , with 3 . 3 × 10−4 M BL or water . As shown in Figure 8 , although the medial sensilla styloconica of PxylGr34 siRNA-treated larvae still showed some response to BL , the frequency of spikes to BL elicited in the medial sensilla styloconica of the PxylGr34 siRNA-treated larvae was decreased ( Figure 8B and Figure 8C ) . To test the effects of RNAi-knockdown of PxylGr34 on the feeding behavior of 4th instar larvae , the siRNA-treated larvae were subjected to a dual-choice leaf disc feeding assay as described above , with leaf discs of pea treated with 10−4 M BL or untreated ( control ) . As shown in Figure 9 , both the water-treated larvae and the GFP siRNA-treated larvae preferred control leaf discs over those treated with BL , whereas the PxylGr34 siRNA-treated larvae showed no significant preference ( Figure 9 ) . Thus , the knock-down of PxylGr34 by RNAi attenuated the electrophysiological responses of the medial sensilla styloconica to BL , and alleviated the deterrent effect of BL on the feeding of P . xylostella larvae . In this study , we identified the full-length coding sequence of PxylGr34 from our transcriptome data , and found that this gene is highly expressed in the head of the 4th instar larvae and in the antennae of females . Our results show that PxylGr34 is specifically tuned to BL and its racemate EBL , and that the medial sensilla styloconica of 4th instar larvae have electrophysiological responses to BL and EBL . Our results also show that BL inhibits larval feeding and female oviposition of P . xylostella , and that knock-down of PxylGr34 by RNAi can attenuate the responses of sensilla to BL , and abolish the feeding inhibition effect of BL . This is the first study to show that an insect can detect and react to this steroid plant hormone . The results of the systematic functional analyses of the GR , the electrophysiological responses of the sensilla , behavioral assays , and behavioral regulation in vivo show that PxylGr34 is a bitter GR specifically tuned to BL . This receptor mediates the deterrent effects of BL on feeding and ovipositing behaviors of P . xylostella . The larvae of Lepidopteran species have two pairs of gustatory sensilla ( medial and lateral sensilla styloconica ) located in the maxillae galea; these sensilla play a decisive role in larval food selection ( Dethier , 1937; Schoonhoven and van Loon , 2002 ) . Each sensillum usually contains four GSNs , of which one is often responsive to deterrents . Ishikawa , 1966 described a ‘deterrent neuron’ in the medial sensillum styloconicum of silkworm , Bombyx mori , and showed that it responds to several plant alkaloids and phenolics ( Ishikawa , 1966 ) . Similar neurons have been found in other Lepidopteran species , but their profiles vary . For example , the tobacco hornworm , Manduca sexta , has a deterrent neuron in the medial sensillum styloconicum that responds to aristolochic acid , and another deterrent neuron in the lateral sensillum styloconicum that responds to salicin , caffeine , and aristolochic acid ( Glendinning et al . , 2002 ) . The diversity of GRs facilitates the detection of , and discrimination among , a wide range of diverse taste stimuli , implying that different sets of bitter GRs are expressed in these neurons . In this study , we proved that one GSN in the medial maxillary sensillum styloconicum of the 4th instar larvae of the diamondback moth responds to sinigrin ( van Loon et al . , 2002 ) , and we identified one neuron responding to BL and EBL in the same sensillum . For P . xylostella larvae , sinigrin is a feeding stimulant whereas BL and EBL are feeding deterrents , together with sinigrin and BL/EBL have different spike amplitudes , we speculate that the GRs tuned to sinigrin and BL/EBL are located in different GSNs in the medial sensilla styloconica . The GR family is massively expanded in moth species , and most of the GRs are bitter GRs ( Cheng et al . , 2017 ) . However , few studies have functionally characterized bitter GRs . In D . melanogaster , the loss of bitter GRs was found to eliminate repellent behavior in response to specific noxious compounds . For example , Gr33a mutant flies could not avoid non-volatile repellents like quinine and caffeine ( Moon et al . , 2009 ) , and mutation of Gr98b impaired the detection of L-canavanine ( Shim et al . , 2015 ) . When the bitter GR PxutGr1 of P . xuthus was knocked-down by RNAi , the oviposition behavior in response to synephrine was strongly reduced ( Ozaki et al . , 2011 ) . Knock-out of the bitter GR BmorGr66 in B . mori larvae resulted in a loss of feeding specificity for mulberry ( Zhang et al . , 2019 ) . In this study , BL and EBL induced a strong response in the oocytes expressing PxylGr34 , knock-down of PxylGr34 in the larvae of P . xylostella eliminated the feeding deterrence of BL , indicating that this bitter GR is specifically tuned to BL and EBL , and mediates the aversive response of larvae to BL and related compounds . Although the 24 tested compounds represent a wide range of compound profiles , the ligands of PxylGr34 could be more than BL and EBL . Given the high expression level of PxylGr34 in the taste organs of P . xylostella , we could not rule out the possibility that this gene also functions together with other GRs to perceive other compounds . BL was the first brassinosteroid ( BR ) hormone to be discovered in plants . It was first isolated and identified from a crude extract of pollen from oilseed rape ( Brassica napus L . ) , and was found to induce rapid elongation of pinto bean Phaseolus vulgaris internodes distinct from gibberellin-mediated stem elongation ( Mitchell et al . , 1970; Grove et al . , 1979 ) . Almost all plant tissues contain BRs , and they function to promote elongation and stimulate cell division , participate in vascular differentiation and fertilization , and affect senescence ( Clouse and Sasse , 1998 ) . As a C28 BR , BL exhibits the highest activity among all BRs and is distributed widely in the plant kingdom , along with other biosynthetically related compounds ( Clouse and Sasse , 1998 ) . Exogenous application of BL and its analog 24-epibrassinolide ( EBL ) to plants has been shown to increase their stress resistance ( Clouse and Sasse , 1998 ) . Plant hormones , although generally found in small amounts and rarely toxic , play a key role in regulating plant growth , development , and resistance to biotic and abiotic stresses ( Bari and Jones , 2009; Krouk et al . , 2011; Wu and Baldwin , 2010 ) . Jasmonic acid , salicylic acid , ethylene , and abscisic acid have been shown to be involved in priming plant defense responses against herbivorous insects and plant pathogens by activating related signal transduction pathways and changing the expression of defense-related genes ( Bari and Jones , 2009 ) . Jasmonic acid plays an important role in plant resistance to insects ( Wang et al . , 2019 ) . Plants accumulate jasmonic acid and its derivatives upon wounding . Exogenous treatment with jasmonic acid activates the expression of hundreds of defense-related genes . Application of exogenous jasmonic acid to cabbage plants was shown to indirectly retard the development of P . xylostella larvae and reduce the pupal weight and female fecundity ( Lv and Liu , 2005 ) . Therefore , signaling molecules associated with induced plant defenses may be used as reliable cues by herbivorous insects . Up to now , only one study showed that Helicoverpa zea reacts to jasmonate and salicylate in plants , resulting in the activation of four of its cytochrome P450 genes that are associated with detoxification ( Li et al . , 2002 ) . However , how the caterpillars eavesdrop the hormone signals remains a mystery . The present study provides the first evidence that P . xylostella can detect the plant hormone BL with a bitter GR . This reflects a new adaptation of insects to plant defenses . Herbivorous insects have evolved counter adaptations against the chemical defenses of plants . The perception of bitter substances is an adaptation to avoid potentially toxic secondary plant metabolites . We can speculate that BL and EBL may influence insect development because BL and EBL have strikingly similar structures to that of ecdysteroid hormones in arthropods , such as 20-hydroxyecdysone ( Fujioka and Sakurai , 1997 ) . Their structures are so similar that BL and EBL show agonistic activity with 20-hydroxyecdysone in many insect species ( Zullo and Adam , 2002 ) . It has been shown that injection with 20 µg EBL was fatal to mid last-instar larvae of Spodoptera littoralis ( Smagghe et al . , 2002 ) . The BL content differs widely among plant species; for example , it is 1 . 37 × 10−4 g/kg in Brassica campestris L . leaves and 1 . 25 × 10−6 g/kg in Arabidopsis thaliana leaves ( Lv et al . , 2014 ) . Our results show that BL has the inhibitory effects on feeding and oviposition of P . xylostella , and the threshold concentration of BL for such behavioral inhibitions of P . xylostella is in the range of 10−4–10−3 g/kg , suggesting that BL plays a dual role of plant hormones and insect feeding/oviposition deterrents in plants . There is a rich variety of bitter GRs in phytophagous insects , but only a few have been functionally characterized ( Kasubuchi et al . , 2018; Zhang et al . , 2019 ) . In this study , we showed that PxylGr34 , a bitter GR highly expressed in larval head and adult antennae of P . xylostella , is tuned to the plant hormones BL and EBL , which mediates the aversive feeding/oviposition responses of P . xylostella to these compounds . These findings not only increase our understanding of the gustatory coding mechanisms of feeding/oviposition deterrents in phytophagous insects , but also offer new perspectives for using plant hormones as potential agents to suppress pest insects . P . xylostella was originally collected from the cabbage field of Institute of Plant Protection , Shanxi Academy of Agricultural Sciences , China . It is not specialized on peas . The insects were reared at the Institute of Zoology , Chinese Academy of Sciences , Beijing . The larvae were fed only with cabbage ( Brassica oleracea L . ) and kept at 26 ± 1°C with a 16L:8D photoperiod and 55–65% relative humidity . The diet for adults was a 10% ( v/v ) honey solution . Pea ( Pisum sativum L . ) plants were grown in an artificial climate chamber at 26 ± 1°C with a 16L:8D photoperiod and 55–65% relative humidity . The plants were grown in nutrient soil in pots ( 8 × 8 × 10 cm ) and were 4–5 weeks old when they were used in experiments . All procedures were approved by the Animal Care and Use Committee of the Institute of Zoology , Chinese Academy of Sciences , and followed The Guidelines for the Care and Use of Laboratory Animals ( protocol number IOZ17090-A ) . Female X . laevis were provided by Prof . Qing-Hua Tao ( MOE Key Laboratory of Protein Sciences , Tsinghua University , China ) and reared in our laboratory with pig liver as food . Six healthy naive X . laevis 18–24 months of age were used in these experiments . They were group-housed in a box with purified water at 20 ± 1°C . Before experiments , each X . laevis individual was anesthetized by bathing in an ice–water mixture for 30 min before surgically collecting the oocytes . We conducted transcriptome analyses of the P . xylostella moth antennae , foreleg ( only tibia and tarsi ) , head ( without antenna ) , and the 4th instar larval mouthparts . Total RNA was extracted using QIAzol Lysis Reagent ( Qiagen , Hilden , Germany ) and treated with DNase I following the manufacturer’s protocol . Poly ( A ) mRNA was isolated using oligo dT beads . First-strand complementary DNA was generated using random hexamer-primed reverse transcription , followed by synthesis of second-strand cDNA using RNaseH and DNA polymerase I . Paired-end RNA-seq libraries were prepared following Illumina’s protocols and sequenced on the Illumina HiSeq 2500 platform ( Illumina , San Diego , CA ) . High-quality clean reads were obtained by removing adaptors and low-quality reads , then de novo assembled using the software package Trinity v2 . 8 . 5 ( Haas et al . , 2013 ) . The GR genes were annotated by NCBI BLASTX searches against a pooled insect GR database , including GRs from P . xylostella ( Engsontia et al . , 2014; You et al . , 2013; Yang et al . , 2017 ) and other insect species ( Guo et al . , 2017; Xu et al . , 2016; Robertson et al . , 2003 ) . The translated amino acid sequences of the identified GRs were aligned manually by NCBI BLASTP and tools at the T-Coffee web server ( Notredame et al . , 2000 ) . The TPM values were calculated using the software package RSEM v1 . 2 . 28 ( Li and Dewey , 2011 ) to analyze GR gene transcript levels . Phylogenetic analysis of P . xylostella GRs was performed based on amino acid sequences , together with those of previously reported GRs of Heliconius melpomene ( Briscoe et al . , 2013 ) and B . mori ( Guo et al . , 2017 ) . Amino acid sequences were aligned with MAFFT v7 . 455 ( Rozewicki et al . , 2019 ) , and gap sites were removed with trimAl v1 . 4 ( Capella-Gutiérrez et al . , 2009 ) . The maximum likelihood phylogenetic tree was constructed using RAxML v8 . 2 . 12 ( Stamatakis , 2014 ) with the Jones-Taylor-Thornton amino acid substitution model . Node support was assessed using a bootstrap method based on 1000 replicates . The tree was visualized in FigTree Version 1 . 4 . 4 ( http://tree . bio . ed . ac . uk/software/figtree/ ) . The adult antennae , head ( without antennae ) , forelegs ( only tarsi and tibia ) , and ovipositor , and the larval head , thorax ( without wing disc , thoracic legs , gut , or other internal tissues ) , thoracic leg , abdomen ( without gut or other internal tissues ) and gut were dissected immediately placed in a 1 . 5 mL Eppendorf tube containing liquid nitrogen , and stored at −80°C until use . Total RNA was extracted using QIAzol Lysis Reagent following the manufacturer’s protocol ( including DNase I treatment ) , and RNA quality was checked with a spectrophotometer ( NanoDrop 2000; Thermo Fisher Scientific , Waltham , MA , USA ) . The single-stranded cDNA templates were synthesized using 2 μg total RNAs from various samples with 1 μg oligo ( dT ) 15 primer ( Promega , Madison , WI , USA ) . The mixture was heated to 70°C for 5 min to melt the secondary structure of the template , then M-MLV reverse transcriptase ( Promega ) was added and the mixture was incubated at 42°C for 1 hr . The products were stored at −20°C until use . Based on the candidate full-length nucleotide sequences of PxylGr34 identified from our transcriptome data , we designed specific primers ( Supplementary file 1 ) . All amplification reactions were performed using Q5 High-Fidelity DNA Polymerase ( New England Biolabs , Beverly , MA , USA ) . The PCR conditions for amplification of PxylGr34 were as follows: 98°C for 30 s , followed by 30 cycles of 98°C for 10 s , 60°C for 30 s , and 72°C for 1 min , and final extension at 72°C for 2 min . Templates were obtained from antennae of female P . xylostella . The sequences were further verified by Sanger sequencing . The qPCR analyses were conducted using the QuantStudio 3 Real-Time PCR System ( Thermo Fisher Scientific ) with SYBR Premix Ex Taq ( TaKaRa , Shiga , Japan ) . The gene-specific primers to amplify an 80–150 bp product were designed by Primer-BLAST ( http://www . ncbi . nlm . nih . gov/tools/primer-blast/ ) ( Supplementary file 1 ) . The thermal cycling conditions were as follows: 10 s at 95°C , followed by 40 cycles of 95°C for 5 s and 60°C for 31 s , followed by a melting curve analysis ( 55–95°C ) to detect a single gene-specific peak and confirm the absence of primer dimers . The product was verified by nucleotide sequencing . PxylActin ( GenBank number: AB282645 . 1 ) was used as the control gene ( Teng et al . , 2012 ) . Each reaction was run in triplicate ( technical replicates ) and the means and standard errors were obtained from three biological replicates . The relative copy numbers of PxylGr34 were calculated using the 2–ΔΔCt method ( Livak and Schmittgen , 2001 ) . The full-length coding sequence of PxylGr34 was first cloned into the pGEM-T vector ( Promega ) and then subcloned into the pCS2+ vector . cRNA was synthesized from the linearized modified pCS2+ vector with mMESSAGE mMACHINE SP6 ( Ambion , Austin , TX , USA ) . Mature healthy oocytes were treated with 2 mg mL−1 collagenase type I ( Sigma-Aldrich , St Louis , MO , USA ) in Ca2+-free saline solution ( 82 . 5 mM NaCl , 2 mM KCl , 1 mM MgCl2 , 5 mM HEPES , pH = 7 . 5 ) for 20 min at room temperature . Oocytes were later microinjected with 55 . 2 ng cRNA . Distilled water was microinjected into oocytes as the negative control . Injected oocytes were incubated for 3–5 days at 16°C in a bath solution ( 96 mM NaCl , 2 mM KCl , 1 mM MgCl2 , 1 . 8 mM CaCl2 , 5 mM HEPES , pH = 7 . 5 ) supplemented with 100 mg mL−1 gentamycin and 550 mg mL−1 sodium pyruvate . Whole-cell currents were recorded with a two-electrode voltage clamp . The intracellular glass electrodes were filled with 3 M KCl and had resistances of 0 . 2–2 . 0 MΩ . Signals were amplified with an OC-725C amplifier ( Warner Instruments , Hamden , CT , USA ) at a holding potential of −80 mV , low-pass filtered at 50 Hz , and digitized at 1 kHz . Each of 24 compounds listed in Key Resources Table was diluted and the pH was adjusted to 7 . 5 in Ringer’s solution before being introduced to the oocyte recording chamber using a perfusion system . Data were acquired and analyzed using Digidata 1322A and pCLAMP software ( RRID:SCR_011323 ) ( Axon Instruments Inc , Foster City , CA , USA ) . Dose-response data were analyzed using GraphPad Prism software ( RRID:SCR_002798 6 ) ( GraphPad Software Inc , San Diego , CA , USA ) . The tip-recording technique was used to record action potentials from the lateral and medial sensilla styloconica on the maxillary galea of larvae , following the protocols described elsewhere ( Hodgson et al . , 1955; van Loon , 1990; van Loon et al . , 2002 ) . Distilled water served as the control stimulus , as KCl solutions elicited considerable responses from the galeal styloconic taste sensilla ( van Loon et al . , 2002 ) , and sinigrin served as the positive control stimulus ( van Loon et al . , 2002 ) . Experiments were carried out with larvae that were 1–2 days into their final stadium ( 4th instar ) . The larvae were starved for 15 min before analysis . The larvae were cut at the mesothorax , and then silver wire was placed in contact with the insect tissue . The wire was connected to a preamplifier with a copper miniconnector . A glass capillary filled with the test compound , into which a silver wire was inserted , was placed in contact with the sensilla . Electrophysiological responses were quantified by counting the number of spikes in the first second after the start of stimulation . The interval between two successive stimulations was at least 3 min to avoid adaptation of the tested sensilla . Before each stimulation , a piece of filter paper was used to absorb the solution from the tip of the glass capillary containing the stimulus solution to avoid an increase in concentration due to evaporation of water from the capillary tip . The temperature during recording ranged from 22° to 25°C . Neural activity was sampled with a computer equipped with a Metrabyte DAS16 A/D conversion board . An interface was used ( GO-box ) for signal conditioning . This involved a second order band pass filter ( −3 dB frequencies: 180 and 1700 Hz ) ( van Loon et al . , 2002 ) . The electrophysiological signals were recorded by SAPID Tools software version 16 . 0 ( Smith et al . , 1990 ) , and analyzed using Autospike software version 3 . 7 ( Syntech ) . Dual-choice feeding assays were used to quantify the behavioral responses of P . xylostella larvae to BL and EBL on pea leaves . Pea ( Pisum sativum L . ) is a neutral host plant of P . xylostella and widely used in the feeding preference analysis in this species ( Thorsteinson , 1953; van Loon et al . , 2002 ) . These assays were based on the protocol reported by van Loon et al . , 2002 , with modifications . The axisymmetric pinnate leaf was freshly picked from 4-week-old pea plants grown in a climate-controlled room . One leaf was folded in half , and two leaf discs ( diameter , 7 mm ) were punched from the two halves as the control ( C ) and treated ( T ) discs , respectively . For the treated discs , 5 µL ( 13 µL/cm2 ) of the test compound diluted in 50% ethanol was spread on the upper surface using a paint brush . For the control discs , 5 µL 50% ethanol was applied in the same way . Control ( C ) and treated ( T ) discs were placed in a C-T-C-T sequence around the circumference of the culture dish ( 60 mm diameter ×15 mm depth; Corning , NY , USA ) . After the ethanol had evaporated ( 15 min later ) , a single 4th instar caterpillar ( day 1 ) , which had been starved for 6 hr , was placed in each dish . The dishes were kept for 24 hr at 23°–25°C in the dark , to avoid visual stimuli . Each dish was covered with a circular filter paper disc ( diameter 7 cm ) moistened with 200 µL ddH2O to maintain humidity . At the end of the test , the leaf discs were scanned using a DR-F120 scanner ( Canon , Tokyo , Japan ) and the remaining leaf area was quantified with ImageJ software ( NIH ) . Paired-sample t-test was used to detect differences in the consumed leaf area between control and treated leaf discs . A dual-choice oviposition bioassay was used to quantify the behavioral responses of P . xylostella mated female moths to BL . This assay was modified from the protocol reported by Gupta and Thorsteinson , 1960 and Justus and Mitchell , 1996 . A paper cup ( 10 cm diameter ×8 cm height ) with a transparent plastic lid ( with 36 pinholes for ventilation ) was used for ovipositing of the mated females . Fresh cabbage leaf juice was centrifuged at 3000 rpm for 5 min , and 60 μL of the supernatant was spread with a paintbrush onto polyethylene ( PE ) film from clinical gloves for 10 min to evaporate . Four culture dishes ( 35 mm diameter ) ( Corning , New York , NY , USA ) covered with these PE films were placed on the bottom of each cup . This oviposition system was developed based on the biology of P . xylostella ( Harcourt , 1957 ) . On each of two diagonally positioned treatment films , 125 µL BL ( 13 µL/ cm2 ) diluted in 50% ethanol was evenly spread on the upper surface using a paint brush . On the other two diagonally positioned films , 125 µL 50% ethanol ( control ) was spread in the same way . After the ethanol had evaporated ( 30 min later ) , a small piece of absorbent cotton soaked with 10% honey-water mixture was placed in the center of the cup . The pupae of P . xylostella were selected and newly emerged adults were checked and placed in a mesh cage ( 25 × 25 × 25 cm ) , with a 10% honey-water mixture supplied during the light phase . The female:male ratio was 1:3 to ensure that all the females would be mated . After at least 24 hr of mating time , the mated females were removed from the cage and placed individually into the oviposition cup during the light phase . After 24 hr at 26 ± 1°C with a 16L: 8D photoperiod and 55–65% relative humidity , the number of eggs on each plastic film was counted . Paired-samples t-test was used to detect significant differences in the number of eggs laid between the control and treatment films . siRNA preparation . A unique siRNA region specific to PxylGr34 was selected guided by the siRNA Design Methods and Protocols ( Takasaki , 2013 ) . The siRNA was prepared using the T7 RiboMAX Express RNAi System kit ( Promega , Madison ) following manufacturer’s protocol . The GFP ( GenBank: M62653 . 1 ) siRNA was designed and synthesized using the same methods . We tested three different siRNAs of PxylGr34 based on different sequence regions , and selected the most effective and stable one for further analyses . The oligonucleotides used to prepare siRNAs are listed in Supplementary file 2 . The siRNAs were supplied to the larvae by oral delivery as reported elsewhere ( Chaitanya et al . , 2017; Gong et al . , 2011 ) , with some modifications . Each siRNA was spread onto cabbage leaf discs ( Brassica oleracea ) and fed to 4th-instar larvae . Freshly punched cabbage leaf discs ( diameter 0 . 7 cm ) were placed into 24-well clear multiple well plates ( Corning , NY , USA ) . For each disc , 3 µg siRNA diluted in 6 µL 50% ethanol was evenly distributed on the upper surface using a paint brush . Both 50% ethanol and GFP siRNA were used as negative controls . After the ethanol had evaporated ( 20 min later ) , one freshly molted 4th-instar larva , which had been starved for 6 hr , was carefully transferred onto each disc and then allowed to feed for 12 hr . Each well was covered with dry tissue paper to maintain humidity . The larvae that had consumed the entire disc were selected and starved for another 6 hr , and then these larvae were used in the dual-choice behavioral assay , electrophysiological responses of contact chemosensilla on the maxilla , or for qPCR analyses as described above . The larvae that did not consume the treated discs were discarded . Data were analyzed using GraphPad Prism 8 . 3 . Figures were created using GraphPad Prism 8 . 3 and Adobe illustrator CC 2018 ( Adobe systems , San Jose , CA ) . Two-electrode voltage-clamp recordings , electrophysiological dose-response curves , and the square-root transformed qPCR data were analyzed by one-way ANOVA and Tukey’s HSD tests with two distribution tails . These analyses were performed using GraphPad prism 8 . 3 . Electrophysiological response data and all dual-choice test data were analyzed using the two-tailed paired-samples t-test . Statistical tests and the numbers of replicates are provided in the figure legends . In all statistical analyses , differences were considered significant at p<0 . 05 . Asterisks represent significance: p<0 . 05 , p<0 . 01 , p<0 . 001 , **p<0 . 0001; ns , not significant . Response values are indicated as mean ± SEM; and n represents the number of samples in all cases .	Plant-eating insects use their sense of taste to decide where to feed and where to lay their eggs . They do this using taste sensors called gustatory receptors which reside in the antennae and legs of adults , and in the mouthparts of larvae . Some of these sensors detect sugars which signal to the insect that the plant is a nutritious source of food . While others detect bitter compounds , such as poisons released by plants in self-defense . One of the most widespread plant-eating insects is the diamondback moth , which feeds and lays its eggs on cruciferous vegetable crops , like cabbage , oilseed rape and broccoli . Before laying its eggs , female diamondback moths pat the vegetable’s leaves with their antennae , tasting for the presence of chemicals . But little was known about the identity of these chemicals . Cabbages produce large amounts of a hormone called brassinolide , which is known to play a role in plant growth . To find out whether diamondback moths can taste this hormone , Yang et al . examined all their known gustatory receptors . This revealed that the adult antennae and larval mouthparts of these moths make high levels of a receptor called PxylGr34 . To investigate the role of PxylGr34 , Yang et al . genetically modified frog eggs to produce this receptor . Various tests on these receptors , as well as receptors in the mouthparts of diamondback larvae , showed that PxylGr34 is able to sense the hormone brassinolide . To find out how this affects the behavior of the moths , Yang et al . investigated how adults and larvae responded to different levels of the hormone . This revealed that the presence of brassinolide significantly decreased both larval feeding and the amount of eggs laid by adult moths . Farmers already use brassinolide to enhance plant growth and protect crops from stress . These results suggest that the hormone might also help to shield plants from insect damage . However , more research is needed to understand how this hormone acts as a deterrent . Further studies could improve understanding of insect behavior and potentially identify more chemicals that can be used for pest control .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2020	A gustatory receptor tuned to the steroid plant hormone brassinolide in Plutella xylostella (Lepidoptera: Plutellidae)
Coat protein complex II ( COPII ) mediates formation of the membrane vesicles that export newly synthesised proteins from the endoplasmic reticulum . The inner COPII proteins bind to cargo and membrane , linking them to the outer COPII components that form a cage around the vesicle . Regulated flexibility in coat architecture is essential for transport of a variety of differently sized cargoes , but structural data on the assembled coat has not been available . We have used cryo-electron tomography and subtomogram averaging to determine the structure of the complete , membrane-assembled COPII coat . We describe a novel arrangement of the outer coat and find that the inner coat can assemble into regular lattices . The data reveal how coat subunits interact with one another and with the membrane , suggesting how coordinated assembly of inner and outer coats can mediate and regulate packaging of vesicles ranging from small spheres to large tubular carriers . In eukaryotic cells , newly synthesized proteins are transported from the endoplasmic reticulum ( ER ) to the Golgi apparatus through the action of the coat protein complex II ( COPII ) . Assembly of COPII coat proteins on the membrane leads to generation of coated membrane vesicles carrying cargo molecules . Vesicle formation proceeds via sequential assembly of the coat components . It is initiated by the small GTPase , Sar1 . Upon exchange of GDP for GTP ( catalysed by Sec12 ) , Sar1 exposes an N-terminal amphipathic helix that inserts into the outer ER membrane leaflet , promoting curvature ( Lee et al . , 2005 ) . Sar1 recruits heterodimers of Sec23 and Sec24 to the membrane , thereby forming the inner layer of the COPII coat ( Matsuoka et al . , 1998 ) . Sec23/24 is an adaptor complex: Sec24 binds transport cargo while Sec23 interacts with Sar1 and recruits Sec13/31 ( Miller et al . , 2002; Bi et al . , 2007 ) . Sec13/31 heterotetramers constitute the outer coat layer , thought to polymerise into cages that enclose the budding membrane ( Fath et al . , 2007; Stagg et al . , 2008 ) . GTP hydrolysis on Sar1 , activated by Sec23 and further accelerated by Sec31 , completes the cycle by promoting fission of the bud and coat depolymerization ( Zanetti et al . , 2012 ) . X-ray crystallography has been used to obtain structural models for all the coat subunits ( Bi et al . , 2002; Fath et al . , 2007 ) . Available structural data for the inner coat is limited to isolated subcomplexes . Progress has been made in understanding how the outer coat subunits assemble into a coat by using single-particle cryo-electron microscopy to derive models of Sec13/31 cages formed in vitro under high salt conditions in the absence of membrane ( Stagg et al . , 2006 , 2008; Bhattacharya et al . , 2012; Noble et al . , 2012 ) . A comparison of the vertices and edges in cages of different sizes ( 60–100 nm ) has suggested geometrical rules governing outer coat assembly , and indicated regions of flexibility in Sec13/31 that permit envelopment of vesicles with sizes ranging from 60 to 120 nm ( Fath et al . , 2007; Stagg et al . , 2008; Bhattacharya et al . , 2012 ) . Nevertheless , a higher degree of flexibility in COPII architecture is needed to explain the ability of COPII to mediate secretion of cargoes such as 300 nm pro-collagen fibres , which are much larger than the 60–100 nm in vitro assembled cages . There is increasing evidence that incorporation of pro-collagen into COPII coated vesicles is a highly regulated process , and that modifications in the outer coat proteins—such as ubiquitination—as well as in timing of GTP hydrolysis and coat recycling may be necessary for COPII ability to coat large vesicles ( Saito et al . , 2009; Jin et al . , 2012; Kim et al . , 2012; Venditti et al . , 2012 ) . COPII-dependent defects in collagen transport are linked to a genetic syndrome characterized by late-closing fontanels , sutural cataracts , facial dysmorphisms and skeletal defects: Cranio-lenticulo-sutural dysplasia ( Boyadjiev et al . , 2006 ) . There are currently no structural data on the COPII coat assembled in its functional state on a membrane , leaving important mechanistic questions unanswered . How do the inner and outer coat components arrange to form a membrane coat ? How do they interact with the membrane ? How do they interact with each other ? Are the existing models for the outer COPII coat cages representative of the complete coat assembled on a membrane ? What is the structural basis for COPII ability to transport large or elongated cargoes such as pro-collagen ? Here we address these questions . We provide evidence for coordinated assembly of the two layers of the coat on a lipid membrane and suggest how this interplay may effect shape changes essential to the capture of large and unusually-shaped secretory cargo complexes and particles . To understand the architecture of the Sec13/31 outer coat we applied reference free , contrast-transfer function ( CTF ) corrected , subtomogram averaging to solve the 3D structure of the vertex , and of the connecting rods to resolutions of ∼40 Å ( Figure 2A–B , Figure 2—figure supplement 1 , and ‘Materials and methods’ ) . The vertex structure was a twofold symmetric X-shape , similar to that seen in in vitro assembled Sec13/31 cages ( Stagg et al . , 2006 , 2008 ) ( Figure 2D , Figure 2—figure supplement 2 ) . The connecting rods are consistent in shape and size with Sec13/31 heterotetramers , and are bent in the middle by approximately 15° . This same bend is seen in the solved X-ray crystal structure , which can be fitted into the densities as a rigid body ( Figure 2B , C ) . In contrast , there is a 45° bend in the rods of in vitro-assembled protein cages ( Stagg et al . , 2008; Figure 2E , F ) . These data indicate that the central hinge between Sec31 molecules can adapt to assemble coats of different curvatures . 10 . 7554/eLife . 00951 . 004Figure 2 . Structure of the outer COPII coat . ( A ) Isosurface representation of the outer coat vertex solved by sub-tomogram averaging of tubular membranes . ‘+’ and ‘−’ ends of Sec13/31 and alpha and beta angles are as defined by ( Stagg et al . , 2008 ) ( Figure 2—figure supplement 2 and panel D ) . ( B ) Structures of the rods that interconnect neighbouring vertices in left-handed ( green ) and right-handed ( purple ) helical directions , viewed from the top ( upper panels ) , and the side ( lower panels ) . Left-handed rods have two ‘−’ ends , whereas right-handed rods have two ‘+’ ends . ( C ) Atomic model of the Sec13/31 complex ( Fath et al . , 2007 ) ( PDB 2PM9 and 2PM6 ) fitted as a rigid body into left- and right-handed rods . ( D ) Isosurface representation of the Sec13/31 vertex structure from cryo-electron microscopy of in vitro assembled cuboctahedral cages ( Stagg et al . , 2006 ) ( EMDB ID 1232 ) . ( E ) Structure of the rods segmented from in vitro assembled cuboctahedral cages ( Stagg et al . , 2006 ) ( EMDB ID 1232 ) for comparison . ‘+’ and ‘−’ ends are coloured purple and green respectively . ( F ) The Sec13/31 complex fitted into a rod from the cuboctahedral cage . To adapt to the 45° bend in the rod two equivalent heterodimers were fitted independently . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 00410 . 7554/eLife . 00951 . 005Figure 2—figure supplement 1 . Resolution of outer coat structures . The resolution of the sub-tomogram averaged structures of outer coat ( blue = vertices , purple = right-handed rods , green = left-handed rods ) were estimated by Fourier Shell Correlation ( see ‘Materials and methods’ for details ) . Resolution at the 0 . 5 threshold criterion is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 00510 . 7554/eLife . 00951 . 006Figure 2—figure supplement 2 . Previously published outer coat structures . ( A ) A close up view of the vertex in the cuboctahedral cage ( Stagg et al . , 2006 ) , with the X-ray model for Sec13/31 dimers fitted into the half-rods . The vertex forms upon interaction of four N-terminal Sec31 β-propellers . The β-propellers on two heterotetramer rods ( purple ) face each other at the center of the vertex and are referred to as the ‘plus’ ends of the rods , while the β-propellers on the other two rods ( green ) are further away from the center , and are referred to as the ‘minus’ ends . In the cuboctahedral and icosidodecahedral cage geometries , each rod has a plus and a minus end . When viewing the vertex from outside the cage , alpha is defined as the clockwise angle from ‘+’ to ‘−’ , and beta the clockwise angle from ‘−’ to ‘+’ . These interactions allow assembly of a coat that curves in two directions , appropriate for coating spherical membranes . ( B ) A close up view of the rod segmented from the edge of the cuboctahedral cage . The rod is bent in the middle at the Sec31 dimerisation interface by an angle of 135° . ( C ) A surface representation of the atomic structure of the Sec13/31 heterodimer as solved by X-ray crystallography ( Fath et al . , 2007 ) , filtered to a resolution of 40 Å . The structure is overall very similar to the cuboctahedral cage edge , but the bend in the middle of the rod is less accentuated , with an angle of 165° . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 006 The published EM reconstructions of in vitro assembled COPII outer coat protein cages reveal that the Sec31 β-propeller domains at the ends of two Sec13/31 rods ( referred to in the literature as the plus ends ) contact each other directly at the center of the vertex , whereas the ends of the other two rods ( the minus ends ) contact the plus ends at the side and are more distant from the center of the vertex . The geometrical relationship of the ‘+’ and ‘−’ rod ends at the vertex can be described by two angles: alpha ( clockwise between + and − ends , which is 60° in the cages ) , and beta ( clockwise between − and + ends which is variable at least between 90° and 108° in the cages ) ( Figure 2A , B , Figure 2—figure supplement 2 , Figure 3—figure supplement 1 ) ( Stagg et al . , 2008 ) . Within the in vitro assembled cages each Sec13/31 rod makes a ‘+’ contact at one end and a ‘−’ contact at the other end . This arrangement allows assembly of a coat that curves in two directions , appropriate for coating spherical membranes . ( Figure 3C , Figure 3—figure supplement 1 ) . In our reconstructions , the arrangement of Sec13/31 rods we observed at the rounded tips of tubes and on spherical membranes ( Figure 1D ) was similar to that seen in in vitro assembled cages , consistent with the presence of such ‘+/−’ rods ( Figure 3C ) . When we analysed the distribution of vertices ( Figure 3A ) and rods ( Figure 3B ) on the tubular membranes we found that they assembled regions of rhomboidal lattice . A rhomboidal lattice could be built in two ways: each vertex could be rotated by approximately 90° in the plane of the membrane relative to the adjacent vertices , or each vertex could have the same orientation ( see schematic in Figure 3—figure supplement 2C ) . The final structure of the vertex on the tubes clearly shows the expected twofold features of the vertex previously described by Stagg et al . ( Noble et al . , 2012 ) , indicating that the majority of vertices have the same orientation ( ‘Materials and methods’ ) . In this arrangement of vertices the rods oriented in one direction ( the right handed rods , purple in Figures 2C and 3B ) each make two ‘+’ contacts ( +/+ ) , while the left-handed rods ( green in Figures 2C and 3B ) each make two ‘−’ contacts ( −/− ) . The alpha angle is 79 . 7° ± 5 . 9° and is oriented around the tube circumference while the beta angle is 95 . 7° ± 5 . 8° and is oriented along the tube axis ( Figure 3—figure supplements 2 and 3 ) . Together these data imply that three properties contribute to outer coat adaptability: ( i ) variability of both alpha and beta angles at the vertices , ( ii ) flexibility of the central rod hinge , and ( iii ) the absence of any inherent asymmetry in the Sec13/31 rods , allowing them to make ‘+’ contacts at both ends , ‘−’ contacts at both ends , or a ‘+’ contact at one end and a ‘−’ contact at the other . Together this versatility allows coating of not only spherical , but also of tubular membranes and therefore accommodation of large elongated cargoes such as pro-collagen . 10 . 7554/eLife . 00951 . 007Figure 3 . Arrangement of the outer COPII coat . ( A ) Visualization of the positions and orientations at which vertices were identified for a representative tube . They are arranged to form a rhomboidal lattice . The positions of inner coat subunits are shown in grey ( Figure 4 ) . ( B ) Visualization of the positions of aligned rods , as in panel A . Right- and left-handed rods in purple and green respectively . ( C ) Schematic depiction of how ‘+/−’ rods can coat regions of spherical curvature by arranging to form orthogonal vertices ( left panel ) . Tubular surfaces ( right panel ) are coated with +/+ and −/− rods that form parallel vertices ( alpha and beta are always in the same direction with respect to the tube axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 00710 . 7554/eLife . 00951 . 008Figure 3—figure supplement 1 . Current models for assembly and size variation of outer COPII coat cages . ( A ) The 60 nm cuboctahedral cage is formed by twofold symmetrical vertices linked by rods . The angle alpha is 60° , the angle beta is 90° ( see Figure 2—figure supplement 2A for definition of alpha and beta ) ( Stagg et al . , 2006 ) . ( B ) In the 90 nm icosidodecahedral cage ( Stagg et al . , 2008 ) , alpha is 60° , but beta is 108° , leading to formation of pentagonal faces . ( C ) If beta was sufficiently flexible to form 120° angles , hexagonal faces could form , leading to assembly of a flat coat , as hypothesized by Stagg et al . ( 2008 ) . Combinations of square , pentagonal and hexagonal faces into the same cage could potentially increase the extent of size variations allowed for the COPII coat . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 00810 . 7554/eLife . 00951 . 009Figure 3—figure supplement 2 . Variability and flexibility in the outer coat calculated from the coordinates of aligned subtomograms . Note that the variability around the mean values may result from alignment errors from the subtomogram averaging . ( A ) The distances between neighbouring vertices measured along right-handed +/+ rods ( purple ) , and along left-handed ( −/− ) rods ( green ) . ( B ) Alpha and beta angles measured based on the coordinates of each vertex and the two corresponding neighbouring vertices . ( C ) A rhomboidal lattice could take two forms . In the first , each vertex would have the same orientation relative to the tube ( left-hand panel ) . In this case the alpha and beta angles would always have the same orientation relative to the direction of the tube , and alpha could be smaller than beta . In the second , each vertex would be rotated by approximately 90° in the plane of the membrane relative to the adjacent vertices ( right-hand panel ) . In this second form , alpha and beta would both be close to 90° . Assembling the second form would therefore require more substantial deformation of the vertex structure seen in in vitro assembled cages , in which alpha is smaller than beta . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 00910 . 7554/eLife . 00951 . 010Figure 3—figure supplement 3 . The unit cell dimensions for the inner and outer coat lattices . The unit cell angles for the inner coat lattice are 83 . 8° ± 10° and 95 . 4° ± 9 . 8° , very close to the angles of 79 . 7° ± 5 . 9° ( alpha ) and 95 . 7° ± 5 . 8° ( beta ) measured for the outer coat . The distance between neighbouring outer coat vertices in the right-handed direction ( 30 . 2 ± 2 nm ) is similar to the distance between two rows of inner coat subunits ( 30 . 2 ± 2 nm ) , while the distance between outer coat vertices in the left-handed direction ( 31 . 9 ± 1 . 9 nm ) is similar to the distance between 4 inner coat subunits ( 31 . 4 ± 4 . 5 nm ) . The two lattices are therefore very closely matched . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 010 Proteins able to form both spherical and tubular structures are found in other biological systems , most notably virus capsids . For example , the elongated heads of T-even phages , as well as the capsid cores of retroviruses , are assembled as closed structures built from hexamers and pentamers of the component protein and can have regions with spherical and with tubular curvatures ( Baschong et al . , 1988; Ganser et al . , 1999 ) . This structural flexibility can be understood within the framework of quasi-equivalence: while subunits are found in symmetrically different positions with different local curvatures , the contacts between the subunits are only subtly different ( Caspar and Klug , 1962 ) . In some virus capsids , subtle changes in the contacts are supplemented by structural switches that mediate the change from hexameric to pentameric assembly ( Johnson and Speir , 1997 ) . It seems likely that flexibility in the outer COPII cage geometry is primarily mediated by subtly varying contacts at multiple hinge positions in the rod and vertex ( Noble et al . , 2012 ) , but we cannot rule out the presence of structural switches . The low resolution that can be obtained in cryo-EM reconstructions of Sec23/24 bound to in vitro assembled Sec13/31 cages , and the necessity of implying symmetry which may not be appropriate for both coat layers , has prevented a clear understanding of the structural relationship between the inner and outer coats ( Stagg et al . , 2008; Bhattacharya et al . , 2012 ) . Our structure of the outer coat does not reveal density corresponding to an ordered inner coat , and vice versa , indicating that the relative positions of the two coat layers are not fixed . Nevertheless , the two layers are approximately aligned both in spacing and orientation . The ‘rows’ in the inner coat are aligned to the left-handed rods of the outer coat , with four inner coat heterotrimers spanning one outer coat rod . The inner coat ‘columns’ are aligned to the right-handed outer coat rods , with two inner coat heterotrimers spanning one outer coat rod ( Figure 3A , B , Figure 3—figure supplement 3 ) . This relationship suggests that the unstructured C-terminal region of Sec31 ( Fath et al . , 2007; Noble et al . , 2012 ) , which connects the inner and outer coats , constrains the coat layers but does not fix their absolute positions , much like multiple anchor cables on a ship . This flexibility in the relative positions of the two layers would permit adaptation of the coat to variable curvatures . Gaps and dislocations in both inner and outer coat lattices may further contribute to curvature variability . The spherical tube ends have an outer coat arrangement similar to that previously observed in in vitro cages , in which adjacent vertices are rotated relative to one another . A helical array of the inner coat heterotrimer cannot be present on the spherically-curved tips . Instead , we consider it likely that on the tube tips , the inner coat lattice is arranged in smaller patches rotated relative to one another , each maintaining the same relationship to the overlying outer coat vertex . Gaps between or within lattice patches could accommodate cargoes with large membrane-proximal domains . A comparison of our results with previous hypotheses for the relationship between inner and outer coats ( Stagg et al . , 2008; Bhattacharya et al . , 2012 ) is in Figure 6 . 10 . 7554/eLife . 00951 . 015Figure 6 . Models for the spatial relationship between inner and outer coat . ( A ) Cryo-electron microscopy reconstructions of COPII cages assembled in vitro in the presence of inner coat subunits , but in the absence of a membrane , showed additional density below that of the outer coat . The top panel shows the density for the inner coat obtained in cages formed in the presence of Sec23 alone ( Bhattacharya et al . , 2012 ) . The bottom panel shows the density for the inner layer in cages formed in the presence of Sec23/24 ( Stagg et al . , 2008 ) . The position of the outer coat is marked as a blue outline . Note that in both studies , the twofold symmetry of the outer coat vertex was imposed on the inner coat . ( B ) Based on the comparison between the structures in panel A , Stagg et al . proposed that Sec23 binds at two positions: one copy below each vertex and two copies at its side . Sec24 would be located under the holes in the cage ( Bhattacharya et al . , 2012 ) . According to this model , the inner coat arrangement does not conform to the twofold symmetry applied during the reconstruction procedure , explaining why it is poorly resolved in the cage reconstruction . ( C ) A previous model was proposed based on the reconstruction of icosidodecahedral cages ( bottom panel in A ) . In this model , Sec23/24 subunits are oriented with their long axes in the direction of the +/+ connection of the outer coat vertex . Two arrangements consistent with this model are shown: in the first ( top ) the subunits all have the same orientation , in the second ( bottom ) , subunits have opposite orientations , following the twofold symmetry of the outer coat . ( D ) A schematic representation of the relationship between inner and outer coat as identifed in this study . This arrangement is similar to the model suggested in C ( top ) . The stoichiometry of the two coat layers is 1:2 outer:inner in coated tubes , meaning that only half of the inner coat subunits are bound by the known Sec31 interface . The percentage of bound inner coat subunits would be higher on spherical vesicles . The difficulty in resolving the position of the inner coat in the studies of cages assembled in the absence of a membrane can be explained by our observation that the relative positions of outer and inner coats are constrained but not fixed , and that the two layers do not follow the same symmetry . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 015 The structural data we have presented shows that both inner and outer COPII layers have sufficient structural versatility to coat both spherical and tubular membranes , and thereby accommodate important large cargoes such as pro-collagen . We found that both coats can assemble regular lattices with constrained relative positions . This relationship would permit the inner coat to influence the arrangement of the outer coat and therefore to play a critical structural role in determining vesicle shape . Our data hint at the possibility that the formation of larger arrays of the inner coat and the formation of associated rhomboidal outer coat lattices , could favour the formation of larger tubular carriers able to accommodate cargoes such as pro-collagen . It has previously been proposed that proteins such as TANGO1 and Sedlin ( Saito et al . , 2009; Jin et al . , 2012; Kim et al . , 2012; Venditti et al . , 2012 ) act in concert to stabilize the inner coat , modulating Sar1 GTPase activity and delaying both outer coat recruitment and scission . We suggest that stabilization of the inner coat could promote growth of larger carriers not only by delaying scission , but also by promoting tubular membrane morphology ( Figure 7 ) . 10 . 7554/eLife . 00951 . 016Figure 7 . Cartoon depiction of a model for two COPII assembly modes . ( A ) On a spherically curved membrane the bow-tie shaped inner coat subunits assemble in small patches that may be randomly oriented with respect to each other . Outer coat rods bind to the inner coat patches in a preferred orientation . The outer coat can assemble to form triangles , squares , or pentamers . ( B ) If the inner coat forms large arrays instead of small patches , then the outer coat , interacting with the inner coat in its preferred orientation , will tend to arrange to form a lozenge pattern . This arrangement of inner and outer coats results in coated tubular membranes . This arrangement could simply be promoted by packaging of elongated cargoes . It could also be favoured when external factors intervene to either delay outer coat recruitment , and/or stabilise larger inner coat patches . Tango I and Sedlin ( Saito et al . , 2009; Venditti et al . , 2012 ) have been proposed to facilitate formation of large COPII carriers that are capable of incorporating 300 nm-long procollagen molecules , and to achieve this by stabilizing the inner coat on the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 00951 . 016 Expression and purification of yeast COPII proteins , as well as GUV production by electroformation , were performed essentially as described in Bacia et al . ( 2011 ) . In vitro reconstitution of coat assembly was performed by adding Sar1p ( 2 μM ) , Sec12ΔCp ( 1 µM ) and GMP-PNP ( 1 mM ) , Sec23/24p ( 320 nM ) and Sec13/31p ( 520 nM ) to 2 μl of GUVs in 20 mM HEPES , pH 6 . 8 , 50 mM KOAc , 1 . 2 mM MgCl2 ( final volume 40 μl ) . After incubation at room temperature for 2 hr the undiluted reaction was mixed with 3 μl Protein-A conjugated 10 nm colloidal gold as fiducial markers for tomography , applied to glow discharged C-flat ( Protochips Inc . ) holey carbon coated grids and vitrified by plunge freezing . CET was performed on FEI Titan Krios electron microscope operated at 200 kV at liquid nitrogen temperature equipped with a Gatan GIF 2002 post column energy filter and a 2k × 2k Multiscan charge-coupled device camera . Tilt series were collected with an angular range of −60° to +60° , angular increment of 3° , defocus 1 . 5–6 µm , total electron dose 80 e−/Å2 and a magnification of 19 , 500× , giving a pixel size of 4 . 3 Å at the specimen level . Tomograms were aligned based on the positions of gold fiducials and reconstructed by weighted back-projection , using the IMOD ( Kremer et al . , 1996 ) and raptor ( Amat et al . , 2008 ) software packages . Visual analysis and segmentation of reconstructed tomograms were performed with Amira ( Visualisation Sciences Group , an FEI company ) and Chimera ( Pettersen et al . , 2004 ) . Sub-tomogram averaging was carried out essentially as described previously ( Faini et al . , 2012 ) using Matlab scripts adapted from the TOM/AV3 tomography toolbox ( Förster et al . , 2005 ) and using the Dynamo package ( Castano-Diez et al . , 2012 ) . Two datasets collected on the FEI Titan Krios were used for sub-tomogram averaging . A ‘far from focus’ dataset collected at nominal defoci of 4–6 µm contained 16 tubes ( typically 1 tube in each tomogram ) . A second , close to focus dataset ( nominal defoci 1 . 5–2 . 5 µm ) , contained 26 tubes . Both unbinned datasets had a pixel size of 4 . 3 Å .	Proteins often need to move between different compartments within cells . To do this they are packaged into transport pods called vesicles . Many trafficked proteins are synthesized in an organelle called the endoplasmic reticulum , or ER; these proteins are transported away from the ER in ‘COPII’ vesicles , which are formed when the COPII proteins assemble on the ER membrane and force it to bulge outward . The bulge pinches off from the ER membrane , forming the vesicle , which can then move to , and fuse with , a different compartment in the cell . The COPII proteins assemble in a particular order to form the vesicle—Sar1 inserts into the membrane of the ER; Sec23 and Sec24 form an inner coat and capture the proteins that the vesicle will transport; and Sec13 and Sec31 form an outer coat . Although the structures of the coat proteins are known , how they bind to each other—and to the ER membrane—to form vesicles of many shapes and sizes is less well understood . Now , Zanetti et al . show how the inner and outer coat proteins can interact flexibly to accommodate a variety of cargoes . Zanetti et al . mixed purified Sar1 and COPII coat proteins with artificial membranes in vitro . As in cells , the proteins assembled a coat on the membranes , creating bulges and vesicles of different shapes . These coats were imaged using an electron microscope , and the images were analysed using computational image-analysis methods . In this way , Zanetti et al . produced a detailed 3D view of the assembled coat . It was found that the inner and outer proteins each arranged to form lattice structures—like fishing nets—which showed flexibility and variability in the way the individual proteins interact , as well as imperfections in the arrangement . Both coats may help to reshape the membrane , and the inner-coat and outer-coat lattices were also found to move with respect to each other . These flexible properties could allow the coat to assemble on membranes with different shapes and curvatures , forming COPII vesicles with distinct sizes and shapes that can carry a range of cargoes .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	The structure of the COPII transport-vesicle coat assembled on membranes
Sprouting angiogenesis is a well-coordinated process controlled by multiple extracellular inputs , including vascular endothelial growth factor ( VEGF ) . However , little is known about when and how individual endothelial cell ( EC ) responds to angiogenic inputs in vivo . Here , we visualized endothelial Ca2+ dynamics in zebrafish and found that intracellular Ca2+ oscillations occurred in ECs exhibiting angiogenic behavior . Ca2+ oscillations depended upon VEGF receptor-2 ( Vegfr2 ) and Vegfr3 in ECs budding from the dorsal aorta ( DA ) and posterior cardinal vein , respectively . Thus , visualizing Ca2+ oscillations allowed us to monitor EC responses to angiogenic cues . Vegfr-dependent Ca2+ oscillations occurred in migrating tip cells as well as stalk cells budding from the DA . We investigated how Dll4/Notch signaling regulates endothelial Ca2+ oscillations and found that it was required for the selection of single stalk cell as well as tip cell . Thus , we captured spatio-temporal Ca2+ dynamics during sprouting angiogenesis , as a result of cellular responses to angiogenic inputs . An extensive branched blood vessel network is crucial to deliver oxygen and nutrients to tissues and organs . Such branched structures of blood vessels form mainly by sprouting angiogenesis , which involves the emergence of new vessels from the pre-existing vasculature ( Eilken and Adams , 2010; Herbert and Stainier , 2011; Phng and Gerhardt , 2009 ) . In the early steps of sprouting angiogenesis , specific motile endothelial cells ( ECs ) activated by angiogenic cues become leading tip cells and migrate outward from the parental vessels ( Gerhardt et al . , 2003 ) . As the tip cell migrates from the parental vessel , neighboring ECs migrate by following the tip cell as stalk cells to keep connection between the tip cell and the parental vessel . While tip cells guide the sprouts , stalk cells constitute the base of the sprouts and are , therefore , considered to be less active than tip cells . Sprouting angiogenesis is triggered by extracellular stimuli . Among them , vascular endothelial growth factor ( VEGF ) -A induces EC motility and loosen interendothelial cell junction by activating a tyrosine kinase receptor , VEGF receptor-2 ( VEGFR2 ) , to induce sprouting ( Ferrara , 2009; Lohela et al . , 2009 ) . VEGF-A stimulates proliferation of ECs as well as survival ( Koch and Claesson-Welsh , 2012; Lohela et al . , 2009 ) . VEGFR3 ( also known as Flt4 ) , a receptor of VEGF-C , is also required for developmental or tumor angiogenesis ( Tammela et al . , 2008 ) . Especially , Vegfr3 is essential for venous sprouting from the posterior cardinal vein ( PCV ) in zebrafish ( Hogan et al . , 2009 ) . However , little is known about how VEGF-A/VEGFR2 or VEGF-C/VEGFR3 signaling in ECs induces initial sprouting in vivo . While sprouting angiogenesis is promoted by VEGFR , it is negatively coordinated by delta-like 4 ( Dll4 ) /Notch signaling ( Eilken and Adams , 2010; Lohela et al . , 2009; Phng and Gerhardt , 2009 ) . During sprouting angiogenesis , Dll4/Notch signaling between ECs determines the selection of single tip cells among ECs of the pre-existing vessel by restricting the angiogenic behavior of neighboring ECs ( Hellström , et al . , 2007 ) . In studies using mouse and zebrafish , loss of Dll4/Notch signaling results in an increased number of ECs showing tip cell behavior ( Hellström , et al . , 2007; Leslie et al . , 2007; Siekmann and Lawson , 2007; Suchting et al . , 2007 ) . While VEGF-A/VEGFR2 signaling and Dll4/Notch signaling have opposing roles in regulating blood vessel sprouting , they are tightly linked in a negative feedback loop . VEGFR2 activation in tip cells leads to an up-regulation of the Notch ligand Dll4 and subsequent activation of Notch signaling in the following stalk cells ( Lobov et al . , 2007; Suchting et al . , 2007; Zarkada et al . , 2015 ) . Increased Notch activity leads to the downregulation of VEGFR2 and/or VEGFR3 , thereby restricting VEGFR2- or VEGFR3-dependent signaling in the stalk cells ( Benedito et al . , 2012; Siekmann and Lawson , 2007; Zarkada et al . , 2015 ) . Indeed , VEGFR3 expression is suppressed by Dll4/Notch signaling in zebrafish and mice ( Benedito et al . , 2012; Siekmann and Lawson , 2007; Tammela et al . , 2008; Zygmunt et al . , 2011 ) . On the other hand , it is unclear to what extent VEGFR2 expression is suppressed by Dll4/Notch signaling ( Benedito et al . , 2012; Jakobsson et al . , 2010; Suchting et al . , 2007 ) . VEGFR2 and VEGFR3 are responsible for the aberrant angiogenesis induced by loss of Dll4/Notch signaling ( Hogan et al . , 2009; Leslie et al . , 2007; Siekmann and Lawson , 2007; Suchting et al . , 2007; Zarkada et al . , 2015 ) ; however , it is still unclear how ECs behave in response to VEGFR-dependent angiogenic cues and how their behavior is spatio-temporally regulated by Dll4/Notch signaling . Binding of VEGF-A to VEGFR2 induces dimerization of VEGFR2 and autophosphorylation of tyrosine residues in VEGFR2 , leading to the activation of various downstream signaling pathways . Autophosphorylated VEGFR2 associates with phospholipase C-γ ( PLCγ ) and increases intracellular Ca2+ via accumulation of inositol 1 , 4 , 5-triphosphate ( IP3 ) ( Koch and Claesson-Welsh , 2012; Moccia et al . , 2012 ) and store-operated Ca2+ entry ( Li et al . , 2011 ) . Besides VEGF-A , various chemical and mechanical stimuli induce an increase of intracellular Ca2+ as a common second messenger in ECs ( Ando and Yamamoto , 2013; Moccia et al . , 2012 ) . However , little is known about when and where the Ca2+ responses in ECs occur during vascular development in vivo . We considered that we could precisely monitor Ca2+ responses in vivo by using genetically encoded fluorescent Ca2+ monitoring probes which can be used for detecting rapid Ca2+ responses in living animals ( Muto et al . , 2013; Rose et al . , 2014 ) . We hypothesized that Ca2+ dynamics might represent one of EC responses to angiogenic cues during sprouting angiogenesis . In this study , we succeeded in visualizing the intracellular Ca2+ dynamics in ECs at single-cell resolution in zebrafish by performing high-speed , three-dimensional ( 3D ) time-lapse imaging , and uncovered how Ca2+ dynamics are spatio-temporally regulated during sprouting angiogenesis . Intracellular Ca2+ oscillations occurred in migrating tip and stalk cells , in a manner dependent upon VEGFR signaling . By investigating when and how Dll4/Notch signaling regulates endothelial Ca2+ oscillations , we demonstrated how suppressive Dll4/Notch signaling was involved in the selection of tip cells from the dorsal aorta ( DA ) and that of stalk sells in intersomitic vessels ( ISVs ) . To understand how individual EC responds to angiogenic stimuli , we examined the dynamics of intracellular Ca2+ in ECs during sprouting angiogenesis in vivo . We conducted in vivo Ca2+ imaging by expressing a genetically encoded Ca2+ indicator , GCaMP7a ( GFP-based Ca2+ probe ) ( Muto et al . , 2013 ) in ECs of zebrafish . GCaMP7a is an improved version of GCaMP ( Nakai et al . , 2001 ) , an engineered GFP that increases fluorescence upon the Ca2+ elevation ( Figure 1—figure supplement 1A ) . Firstly , we established a transgenic fish line , Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) , in which GCaMP7a was driven by the endothelial specific promoter fli1 via the Gal4/UAS system ( Asakawa et al . , 2008 ) . This Tg line showed an increase of fluorescence exclusively in ECs in response to Ca2+ elevation ( Figure 1—figure supplement 1B ) . Secondly , to distinguish each EC , we developed a Tg fish line , Tg ( fli1:H2B-mC ) , in which EC nuclei was labeled by H2B-mCherry , and crossed this line with the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) line . We confirmed that almost all ECs expressed GCaMP7a in developing trunk vessels of these triple Tg embryos ( Figure 1—figure supplement 2A ) , although the expression of GCaMP7a varied among ECs . To monitor fast Ca2+ dynamics in ECs ( see Figure 1—figure supplement 2B , C ) , we used a light sheet microscopy , which allows rapid acquisitions in living embryos by illuminating the sample with a focused light sheet perpendicularly to the direction of observation ( Huisken et al . , 2004 ) . We examined intracellular Ca2+ dynamics in budding ECs of the DA close to somite boundaries at 24–27 somite stages ( ss ) . We defined these budding ECs as tip cells , because we confirmed that they eventually became tip cells . These tip cells showed sustained and non-periodic Ca2+ oscillations ( Figure 1A , B , Figure 1—figure supplement 2B , C and Video 1 ) . To avoid missing the fast Ca2+ oscillations by taking z-axis images , we performed the time-lapse 2D imaging and confirmed that Ca2+ oscillations could be observed at more than every min ( Figure 1—figure supplement 2B , C ) . In every oscillation , a Ca2+ spike occurs throughout the cytoplasm ( Figure 1—figure supplement 2B ) . The time to reach the peak of individual oscillations was varied 5 . 6–18 . 7 s ( average , 9 . 0 s ) ( Figure 1C ) . Therefore , hereafter we performed 3D time-lapse imaging analyses at 5 s intervals to capture all Ca2+ oscillations . Intracellular Ca2+ levels of individual ECs were quantified at each time point by measuring fluorescence intensity of GCaMP7a , while tracking H2B-mC-labelled cell nuclei over time ( Figure 1—figure supplement 2D; see Materials and methods ) . We analyzed Ca2+ oscillations by the frequency and average increases in relative fluorescence intensity of GCaMP7a from the base line ( mean ΔF/F0 ) . Frequency of Ca2+ oscillations is elevated by increased levels of agonists in some cases in ECs ( Carter et al . , 1991; Jacob et al . , 1988; Moccia et al . , 2003; Mumtaz et al . , 2011 ) and non-ECs ( Woods et al . , 1986 ) . Meanwhile , the amplitude of Ca2+ rise and total Ca2+ increases may possibly reflect the dose of agonists ( Brock et al . , 1991; Fewtrell , 1993; Sage et al . , 1989 ) . Thus , in this study , we quantified the oscillations to describe the oscillatory activity in individual EC ( see ‘Materials and methods’ ) . Our quantification analyses clearly revealed that budding tip cells exhibited oscillatory activity at 24–27 ss ( Figure 1D , E ) . Repetitive Ca2+ transients were not detected in other ECs within the DA ( Figure 1A , B , D ) . These results indicate that the Ca2+ imaging method we used precisely detects the endogenous intracellular increase or decrease of Ca2+ in vivo . 10 . 7554/eLife . 08817 . 003Figure 1 . Ca2+ oscillations in tip cells during budding from the dorsal aorta ( DA ) . ( A ) 3D-rendered time-sequential images of the trunk regions of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during vessel sprouting from the DA ( 24 somite stage ( ss ) ) . 3D images were acquired using a light sheet microscope . The merged images of GCaMP7a ( green ) and H2B-mC ( red ) images are shown in the following images , unless otherwise described . All the zebrafish images are lateral views and displayed as anterior to the left . A green arrowhead indicates a tip cell outlined by a dashed line . ( i ) - ( v ) and other images are those indicated by the arrowheads indicated at a graph in B . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , light gray , dark gray , and black ) at the left panel are shown as a graph . To measure the fluorescence intensity of GCaMP7a ( green ) in individual EC , the cell nucleus ( red ) was tracked over time ( see ‘Materials and methods’ ) . ( C ) Dot-plot graphs depicting the time to reach the peak of each Ca2+-oscillation in tip cells . Time-lapse 2D slice images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos taken every 100 ms as in Figure 1—figure supplement 2B were analyzed for quantification . Horizontal lines represent mean ± s . d . ( n = 18 ) . ( D ) Quantification of Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells and other ECs within the DA during tip cell budding ( 24–27 ss ) ( see ‘Materials and methods’ ) . Each dot represents the value for a single cell . Horizontal lines represent mean ± s . d . ( n ≥ 8 ) . ( E ) Schematic model of tip cells showing Ca2+ oscillations when they sprout from the DA . Intensity of green reflects the frequency of Ca2+ oscillations . Scale bar , 10 μm in A . *p < 0 . 001 . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00310 . 7554/eLife . 08817 . 004Figure 1—figure supplement 1 . GCaMP7a works as a Ca2+ indicator in endothelial cells ( ECs ) . ( A ) HUVECs transfected with GCaMP7a expression plasmids were treated with DMSO ( upper ) or ionomycin ( lower ) . GCaMP7a images before ( - ) and after the treatment ( 40 s ) are shown . ( B ) Confocal stack fluorescence images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) embryos at 26 ss treated with DMSO ( upper ) or 25 μM ionomycin ( lower ) . GCaMP7a images before ( - ) and after the treatment ( 30 min ) are shown . ( C ) Fluorescence images of HUVECs transfected with GCaMP7a expression plasmids pretreated with DMSO or 25 μM BAPTA-AM for 30 min and treated with 50 ng/ml VEGF-A before ( - ) and after the treatment ( 100 s ) . Note that enhancement of GCaMP7a fluorescence by VEGF-A is blocked by pretreatment with BAPTA-AM . Scale bars , 10 μm in A-C . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00410 . 7554/eLife . 08817 . 005Figure 1—figure supplement 2 . Quantitative analyses of intracellular Ca2+ dynamics in ECs . ( A ) Confocal stack fluorescence images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 28 ss . The merged images of GCaMP7a ( green ) and H2B-mC ( red ) images are shown with enhanced brightness . Note that almost all ECs express GCaMP7a in developing trunk vessels . ( B ) Time-lapse 2D slice images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos taken every 100 ms at 24 ss using a light sheet microscope . A green arrowhead indicates a budding tip cell outlined by a dashed line . The elapsed time ( s ) after starting imaging of an embryo is indicated at the left upper corner . ( C ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from B indicated by arrowheads ( green , light gray , dark gray , and black ) at the left panel are shown as a graph . ( D ) Quantification analyses for intracellular Ca2+ levels in individual EC . Time-lapse 3D images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos were analyzed for quantification using IMARIS software . The merged images of GCaMP7a ( green ) and H2B-mC ( red ) images are shown in the left panel . To quantify intracellular Ca2+ of individual EC at each time point , the cell nucleus was tracked over time . Trajectories of individual nuclei were shown in the right panel . To mark the individual EC which we analyze , we set a spherical region of interest ( ROI ) as shown in the right panel ( see ‘Materials and methods’ ) . We then defined the highest voxel intensity of the GCaMP7a fluorescence ( green ) within the ROI as the fluorescence intensity ( F ) in the EC . Scale bars , 10 μm in A , B , and D . ISV , intersomitic vessel; DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00510 . 7554/eLife . 08817 . 006Video 1 . Ca2+ oscillations in tip cell during budding from the dorsal aorta ( DA ) . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 24 somite stage ( ss ) . Green , GCaMP7a fluorescence; red , H2B-mC fluorescence . Elapsed time from the start point of imaging is in seconds ( s ) . Lateral view , anterior to the left . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 006 Intracellular Ca2+ oscillations are known to occur in response to physiological concentrations of agonists in vitro in many cell types ( Fewtrell , 1993; Woods et al . , 1986 ) including ECs ( Jacob et al . , 1988; Moccia et al . , 2003; Sage et al . , 1989 ) , suggesting that Ca2+ oscillations detected here may represent EC response to angiogenic stimuli . To examine which angiogenic stimuli are responsible for Ca2+ oscillations during vessel sprouting from the DA , we first tested the involvement of Vegfr2 since VEGF-A/VEGFR2 signaling is essential for sprouting angiogenesis ( Koch and Claesson-Welsh , 2012; Lohela et al . , 2009 ) and can increase intracellular Ca2+in vitro ( Figure 1—figure supplement 1C ) ( Brock et al . , 1991 ) . Firstly , we examined the effect of inhibiting Vegfr2 on Ca2+ oscillations by using an inhibitor of VEGFR2 , ki8751 ( Kubo et al . , 2005 ) . When we treated 22 ss embryos with ki8751 and performed Ca2+ imaging analyses at 24–27 ss , the Ca2+ oscillations detected in budding tip cells of control embryos ( Figure 1 ) were completely abolished by the treatment of ki8751 ( Figure 2A–C ) . Tip cell migration stopped in ki8751-treated embryos ( Figure 2—figure supplement 1A ) , confirming that ki8751 inhibits Vegfr2 of zebrafish . ki8751 might also inhibit Vegfr3 activity , because , in our previous results , ki8751 treatment blocked Vegfr3-dependent venous sprout from the PCV in zebrafish ( Kwon et al . , 2013 ) . Therefore , we regarded ki8751 as a zebrafish Vegfr inhibitor in this study . To confirm the specific contribution of Vegfr2 in budding ECs from the DA to Ca2+ oscillation there , we knocked-down vegfr2 ( also termed kdrl ) by injecting antisense oligonucleotides ( MO ) ( Wiley et al . , 2011 ) . Although we observed partial sprouts from the DA in the morphants as reported in vegfr2 mutants ( Covassin et al . , 2006 ) , Ca2+ oscillations were markedly reduced in these budding cells ( Figure 2D–F ) , suggesting that Ca2+ oscillations in ECs budding from the DA are dependent upon Vegfr2 activation . 10 . 7554/eLife . 08817 . 007Figure 2 . The Ca2+ oscillations during tip cell budding depend upon Vegfa/Vegfr2 signaling . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos treated with a Vegfr inhibitor , ki8751 , during tip cell budding . The embryos were treated from 22 ss with ki8751 and time-lapse imaged at 24 ss . A green arrowhead indicates a tip cell outlined by a dashed line . The elapsed time ( s ) after starting imaging of an embryo is indicated at the left upper corner . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , light gray , dark gray , and black ) at the left panel are shown as a graph . ( C ) Quantification of Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) as in Figure 1D in ki8751-treated embryos . The embryos were treated from 22 ss with ki8751 and imaged at 24–27 ss ( n ≥ 10 ) . ( D ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip cell budding ( 24 ss ) injected with vegfr2 ( kdrl ) morpholino ( MO ) . ( E ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from D indicated by arrowheads at the left panel are shown as a graph . ( F ) Quantification of Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells and other ECs within the DA in control MO- or vegfr2 MO-injected embryos during tip cell budding at 24–27 ss ( n ≥ 11 ) . ( G ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) embryos at 24–27 ss injected with control UAS:NLS-mC plasmid ( upper ) or UAS:sFlt1 , NLS-mC plasmid ( lower ) which drives the expression of NLS-mC or both sFlt1 and NLS-mC simultaneously in ECs in a mosaic manner , respectively . Green and red arrowheads indicate NLS-mC-expressing ECs and both sFlt1- and NLS-mC-expressing ECs , respectively . Yellow dashed lines indicate positions of somite boundaries . ( H ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from G indicated by arrowheads at the left panel are shown as a graph . ( I ) Quantification of Ca2+ oscillatory activity in ECs expressing NLS-mC and both sFlt1 and NLS-mC at 24–27 ss . Graphs show Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) of NLS-mC-positive ECs within the DA close to somite boundaries ( NLS-mC , n = 18; sFlt1+ NLS-mC , n = 20 ) . Horizontal lines represent mean ± s . d . . Scale bars , 10 μm in A , D and G . p < 0 . 01 , *p < 0 . 001; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00710 . 7554/eLife . 08817 . 008Figure 2—figure supplement 1 . Defects in blood vessels and lymphatic vessels found in Vegfr2- or Vegfr3-inhibited embryos . ( A ) Confocal 3D images of Tg ( fli1:GFP ) embryos at 24 ss ( first column ) , and the corresponding subsequent time-lapse images at the indicated time after treatment with DMSO or ki8751 . Note that tip cell migration is blocked by treatment with ki8751 . ( B ) Confocal stack fluorescence images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) embryos at 36 ss injected with UAS:sFlt1 , NLS-mC plasmid which drives the expression of both sFlt1 and NLS-mC simultaneously in ECs in a mosaic manner via a Gal4/UAS-based bidirectional expression system . GCaMP7a ( green ) and NLS-mC ( red ) images are shown . ISV sprouting from the DA was inhibited in the region close to sFlt1- and NLS-mC-co-expressing ECs ( arrow ) . ( C ) Confocal stack fluorescence images of Tg ( fli1:Myr-mC ) ; ( UAS:GFP ) ;SAGFF ( LF ) 27C embryos at 50 hr postfertilization ( hpf ) uninjected or injected with vegfr3 morpholino ( MO ) . These Tg embryos express myristoylation signal ( Myr ) -tagged mCherry ( red ) in all ECs and GFP ( green ) in venous ECs . Note that the secondary sprouts ( arrowheads ) and secondary sprout-derived venous vessels ( arrows ) from the PCV were observed in uninjected embryos but not in vegfr3 morphants . Scale bars , 10 μm in A–C . ISV , intersomitic vessel; DA , dorsal aorta; PCV , posterior cardinal vein . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00810 . 7554/eLife . 08817 . 009Figure 2—figure supplement 2 . Vegfr3 is not involved in Ca2+ oscillations in tip cells budding from the dorsal aorta ( DA ) . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos injected with vegfr3 MO at 25 ss . A green arrowhead indicates a budding tip cell . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads at the left panel are shown as a graph . ( C ) Quantification of Ca2+ oscillatory activity of vegfr3 morphants at 24–27 ss . Graphs show Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) of tip cells and other ECs within the DA during tip cell budding ( n ≥ 12 ) . Scale bar , 10 μm in A . p < 0 . 01 , **p < 0 . 001; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 00910 . 7554/eLife . 08817 . 010Figure 2—figure supplement 3 . Ca2+ oscillations in the venous sprouts from the posterior cardinal vein ( PCV ) . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during vessel sprouting from the PCV injected with control MO ( 39 hpf ) . A green arrowhead indicates a tip cell of a venous sprout . A dashed line outlines the PCV and the venous sprout . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , orange , light gray , dark gray , and black ) at the left panel are shown as a graph . ( C ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos injected with vegfr3 MO ( 39 hpf ) . ( D ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from C indicated by arrowheads at the left panel are shown as a graph . Scale bars , 10 μm in A and C . DA , dorsal aorta; PCV , posterior cardinal vein . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 010 We further tried to confirm whether Ca2+ oscillations depend upon Vegfa/Vegfr2 signaling by examining the effect of forced expression of soluble Flt1 ( sFlt1 ) , a potent Vegfa trap , on Ca2+ oscillations . sFlt1 was transiently and specifically expressed in ECs in a mosaic manner by injecting plasmids at one-cell stage . While expression of control NLS-tagged mCherry ( NLS-mC ) did not affect Ca2+ oscillations in budding ECs ( Figure 2G–I; NLS-mC ) , co-expression of sFlt1 and NLS-mC in ECs completely inhibited Ca2+ oscillations of ECs close to somite boundaries ( Figure 2G–I; sFlt1+NLS-mC ) . Tip cell sprouting was also inhibited by expressing sFlt1 ( Figure 2—figure supplement 1B ) , as previously reported ( Zygmunt et al . , 2011 ) . Collectively , our results indicate that Vegfa/Vegfr2 signaling is required for the Ca2+ oscillations during tip cell budding from the DA . VEGFR3 signaling is another signaling pathway that facilitates sprouting angiogenesis ( Covassin et al . , 2006; Tammela et al . , 2008 ) . To investigate the role of Vegfr3 in Ca2+ oscillations in the budding ECs from the DA , we examined Ca2+ oscillation in vegfr3 morphants ( Hogan et al . , 2009 ) . vegfr3 morphants exhibited loss of venous sprout from the PCV ( Figure 2—figure supplement 1C ) and phenocopied vegfr3 null mutant fish ( Hogan et al . , 2009 ) . These vegfr3 morphants did not show any alteration of Ca2+ oscillations during tip cell budding ( Figure 2—figure supplement 2A–C ) . These results suggest that Vegfr3 is not involved in the Ca2+ responses during vessel sprouting from the DA . While ISV sprouts from the DA are mainly driven by Vegfa/Vegfr2 signaling , those from the PCV are driven by Vegfc/Vegfr3 signaling ( Hogan et al . , 2009 ) . We observed that tip cells in the venous sprouts from the PCV exhibited Ca2+ oscillations in a manner dependent upon Vegfr3 ( Figure 2—figure supplement 3A–D ) . Thus , Ca2+ oscillations are not only specific for arterial sprouting , but also occur in venous sprouting that is regulated by Vegfc/Vegfr3 signaling . To know when and how single tip cells are selected , we looked at Ca2+ dynamics in ECs of the DA before ECs sprouted from the DA . Prior to sprouting from the DA , Ca2+ oscillations occurred widely within the DA ( Figure 3A , B and Video 2 ) and were completely dependent upon Vegfr2 ( Figure 3—figure supplement 1A–D ) . Loss of Ca2+ oscillations in vegfr2 morphants ( Figure 3—figure supplement 1C , D ) was rescued in ECs expressing Vegfr2 ( Figure 3—figure supplement 1E , F ) , indicating a specific role of Vegfr2 in Ca2+ oscillations . Although ISV sprouts usually emerge from the DA just anterior to each somite boundary ( Zygmunt et al . , 2011 ) , Ca2+-oscillating cells were not restricted to specific regions within the DA before vessel sprouting ( 17–19 ss , Figure 3C ) . These results suggest that single tip cells are not specified among the ECs of the DA before sprouting as far as Ca2+ oscillation is used as an indicator ( Figure 3D ) . 10 . 7554/eLife . 08817 . 011Figure 3 . Ca2+-oscillating cells were not restricted to specific regions within the DA before vessel sprouting . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos before ISV sprouting ( 18 ss ) . Yellow dashed lines indicate positions of somite boundaries . ( B ) The DA is subdivided into three regions ( Region 1–3 ) between two somite boundaries ( SBs ) as illustrated in the scheme ( upper ) . The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A are shown as separate graphs ( Region 1–3 ) , determining the region to which individual EC belongs by the location based on the position of the nucleus at the start of time-lapse imaging . A representative graph of two ECs at each region is shown . ( C ) Quantification of Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in ECs of the indicated regions within the DA before vessel sprouting ( 17–19 ss ) . Horizontal lines represent mean ± s . d . ( n ≥ 20 ) . ( D ) Schematic illustration of Ca2+ dynamics before tip cell budding . Before tip cells sprout from the DA , Ca2+ oscillations are found broadly in ECs within the DA . Scale bar , 10 μm in A . NS , not significant . SB , somite boundary; DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01110 . 7554/eLife . 08817 . 012Figure 3—figure supplement 1 . Vegfr2 is responsible for Ca2+ responses before ISVs sprouting from the DA . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos treated with ki8751 to inhibit Vegfr before ISV sprouting . The embryos were treated with ki8751 from 17 ss and time-lapse imaged at 18 ss using a light sheet microscope . Yellow dashed lines indicate positions of somite boundaries . ( B ) The DA is subdivided into three regions ( Region 1–3 ) between two somite boundaries ( SBs ) as illustrated in the schematics ( left ) . The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A are shown as separated graphs ( Region 1–3 ) as in Figure 3B . Where individual EC belong to is determined by the location based on the position of the nucleus at the start of time-lapse imaging . A representative graph of two ECs at each region is shown . ( C ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos before ISV sprouting injected with vegfr2 ( kdrl ) morpholino ( MO ) ( 18 ss ) . ( D ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from C are shown as separated graphs ( Region 1–3 ) as in B . ( E ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) embryos before ISV sprouting injected with vegfr2 MO and UAS:Vegfr2 , NES-mC plasmid which drives the expression of full length Vegfr2 and NES-mC simultaneously in ECs in a mosaic manner ( 19 ss ) . Note that Ca2+ oscillations are recovered in NES-mC-positive ECs ( arrowheads ) . ( F ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs expressing NES-mC from E indicated by arrowheads ( green , blue , orange and magenta ) at the left panel are shown as a graph . Scale bars , 10 μm in A , C and E . DA , dorsal aorta; SB , somite boundary . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01210 . 7554/eLife . 08817 . 013Video 2 . Ca2+ oscillations occur widely within the DA before vessel sprouting . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 18 ss . Green , GCaMP7a fluorescence; red , H2B-mC fluorescence . Elapsed time is in seconds ( s ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 013 Next , we examined endothelial Ca2+ dynamics just after vessel sprouting ( 19–22 ss ) . Since ISV sprouts emerge bilaterally from the DA , we analyzed one side of sprouting ISVs . When ECs within the DA started to sprout dorsally ( 19–22 ss ) , Ca2+ oscillations were detected in two neighboring ECs ( 45 . 2% ) or single ECs ( 52 . 4% ) at somite boundaries , either of which sprouted dorsally ( Figure 4A and Figure 4—figure supplement 1 ) . We confirmed that two neighboring ECs were not the result of a recent division ( 86% , n = 7 , data not shown ) . These results suggest that the initial selection of single tip cells is not completed just after the onset of vessel sprouting . Ca2+ response became restricted to single tip cells at later stage ( 24–27 ss ) ( Figure 4A ) , suggesting the completion of tip cell selection . Then , to visualize the process of tip cell selection between two neighboring cells exhibiting Ca2+ oscillations , we performed Ca2+ imaging for a longer period and found that only one cell maintained the oscillations over time ( Figure 4B , C and Video 3 ) . We also noticed that the ECs losing Ca2+ response retracted their sprout ( Figure 4B ) , implying that these cells lost their ability to respond to angiogenic cues . The ECs showing sustained Ca2+ oscillations eventually became tip cells ( 18/18 , Figure 4—figure supplement 2 ) . Thus , our results provide evidence that the initial selection of tip cells is determined even after ECs start to migrate from the DA ( Figure 4D ) . 10 . 7554/eLife . 08817 . 014Figure 4 . Iinitial tip cell selection in the DA . ( A ) The number of Ca2+-oscillating ECs within the DA at each somite boundary of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 19–22 and 24–27 ss . Graph shows percentage of the number of a Ca2+-oscillating cell ( 1 ) , two cells ( 2 ) , and three cells ( 3 ) at a somite boundary among the total number of somite boundaries ( indicated at the top ) observed . Two each representative 3D-rendered images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) at 19–22 and 24–27 ss are shown in the left . Arrowheads indicate Ca2+-oscillating cells . ( B ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos from 22 ss . Green arrowheads indicate an EC which maintained Ca2+ oscillations , whereas red arrowheads indicate an EC which lost Ca2+ oscillations . Similar results were obtained in five independent experiments . ( C ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of the ECs indicated by arrowheads in B and indicated at the left panel are shown as a graph . ( D ) Schematic illustration of Ca2+ dynamics during tip cell budding . Ca2+ oscillations are detected mostly in a single or two-neighboring EC ( s ) at the onset of vessel sprouting . Finally , only single budding tip cell exhibits Ca2+-oscillation at later stages . Scale bars , 10 μm in A and B . SB , somite boundary; DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01410 . 7554/eLife . 08817 . 015Figure 4—figure supplement 1 . ECs close to somite boundaries have potential to sprout . Light-sheet z-stack fluorescence images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ( lower ) and corresponding 2D-slice bright-field ( BF ) images at the level of somite boundary ( upper ) just after vessel sprouting ( 22 ss ) . Yellow arrowheads indicate somite boundaries . Yellow dashed lines indicate positions of somite boundaries . White arrowheads indicate Ca2+-oscillating cells , either of which extends protrusions dorsally . Note that double ( left ) or single ( right ) Ca2+-oscillating cells are located at somite boundary . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01510 . 7554/eLife . 08817 . 016Figure 4—figure supplement 2 . Tip cell selection between two neighboring ECs exhibiting Ca2+ oscillations . 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip selection between two neighboring cells exhibiting Ca2+ oscillations . An EC maintaining Ca2+ oscillations became tip cell ( green arrowheads ) , whereas an EC losing Ca2+ oscillations stayed in the DA ( magenta arrowheads ) . A representative case among those we observed are shown ( n = 18 ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01610 . 7554/eLife . 08817 . 017Figure 4—figure supplement 3 . PlexinD1 is necessary to confine Ca2+-oscillating sprouts in the vicinity of somite boundaries . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip cell budding injected with plxnD1 MO ( 18 ss ) . Arrowheads indicate Ca2+-oscillating ECs budding from the DA . Yellow dashed lines indicate positions of somite boundaries . Note that plxnD1 morphants display ectopic EC sprouts that exhibit Ca2+ oscillations . ( B ) Fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs indicated by arrowheads in A are shown as a graph . Scale bar , 10 mm in A . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01710 . 7554/eLife . 08817 . 018Video 3 . The process of tip cell selection . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos just after vessel sprouting from the DA ( 22 ss ) . Elapsed time is in seconds ( s ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 018 Sema-PlexinD1 ( PlxnD1 ) signaling is required to determine the position of tip cell budding in the vicinity of somite boundaries ( Zygmunt et al . , 2011 ) . We observed ectopically budding ECs exhibiting Ca2+ oscillations in plxnD1 morphants ( Figure 4—figure supplement 3 ) , confirming an essential role of PlxnD1 in the allocation of budding ECs from the DA . We then examined Ca2+ dynamics at later stages of ISV formation . Sustained repetitive Ca2+ oscillations in tip cells were observed when ECs following tip cells were budding from the DA ( Figure 5A–C ) . We defined those following ECs as stalk cells , because they eventually became stalk cells . Interestingly , Ca2+ oscillations also occurred in budding stalk cells ( Figure 5A , B and Video 4 ) , although their oscillation frequency and mean ΔF/F0 were lower than those in tip cells ( Figure 5C , D ) . We further found that the Ca2+ oscillations both in tip cells and budding stalk cells were Vegfr-dependent ( Figure 5C and Figure 5—figure supplement 1 ) . In clear contrast , we did not detect Ca2+ oscillations in other ECs that remained in the DA ( Figure 5A–C ) . 10 . 7554/eLife . 08817 . 019Figure 5 . Ca2+ oscillations in stalk cells during budding from the DA . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during stalk cell budding from the DA ( 29 ss ) . Green and orange arrowheads indicate tip and stalk cells , respectively . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , orange , light gray , dark gray , and black ) at the left panel are shown as a graph . ( C ) Quantification of Ca2+ oscillatory activity in untreated and ki8751-treated embryos during stalk cell budding from the DA as in A and Figure 5—figure supplement 1A , respectively . Graphs show Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells , stalk cells and other ECs within the DA in untreated and ki8751-treated embryos ( Untreated , n ≥ 10; ki8751-treated , n ≥ 13 ) . ( D ) Stalk cells that are budding from the DA have significant Vegfr2 activity , albeit weaker than that in tip cells . Scale bar , 10 μm in A . p < 0 . 05 , p < 0 . 01 , p < 0 . 001; NS , not significant . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 01910 . 7554/eLife . 08817 . 020Figure 5—figure supplement 1 . Ca2+ responses during stalk cell budding from the DA are dependent upon Vegfr . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos treated with ki8751 during stalk cell budding from the DA as in Figure 5A . The embryos were treated from 27 ss with ki8751 and time-lapse imaged at 28 ss . Green and orange arrowheads indicate tip and stalk cells , respectively . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , orange , light gray , dark gray , and black ) at the left panel are shown as a graph . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02010 . 7554/eLife . 08817 . 021Video 4 . Ca2+ oscillations occur in stalk cell during budding from the DA . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 29 ss . Elapsed time is in seconds ( s ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 021 To investigate the significance of the Vegfr2 activity in tip and stalk cells , we examined the behavior of ECs expressing dominant-negative Vegfr2 . We generated a dominant-negative Vegfr2 by deleting the C-terminus of zebrafish Vegfr2 ( Vegfr2-ΔC ) ( Millauer et al . , 1994 ) . Firstly , to validate whether Vegfr2-ΔC can block Vegfa/Vegfr2 signaling in zebrafish ECs , we established a Tg fish line , Tg ( fli1:Gal4ff ) ; ( UAS:Vegfr2-ΔC , NLS-mC ) , that expresses Vegfr2-ΔC simultaneously with NLS-mC in ECs . In primary culture cells dissociated from the Tg embryos , Vegfaa-induced Erk phosphorylation was blocked in the Vegfr2-ΔC-expressing ECs ( Figure 6A , B; Vegfr2-ΔC+NLS-mC ) . In contrast , in the cells dissociated from the control Tg ( fli1:Gal4ff ) ; ( UAS:NLS-mC ) embryos that expressed NLS-mC alone in ECs , the Erk phosphorylation was not blocked ( Figure 6A , B; NLS-mC ) . These results confirmed that Vegfaa/Vegfr2 signaling is blocked by Vegfr2-ΔC expression . Consistently , ISV sprouting was blocked in the Tg embryos expressing Vegfr2-ΔC in ECs . We , then , observed the behavior of ECs expressing Vegfr2-ΔC in a mosaic manner by injecting the plasmids expressing Vegfr2-ΔC at one-cell stage . While expression of NLS-mC alone neither affect tip , stalk , nor DA cells ( Figure 6C , D; NLS-mC ) , the cells expressing both Vegfr2-ΔC and NLS-mC rarely became tip and stalk cells and remained in the DA ( Figure 6C , D; Vegfr2-ΔC+NLS-mC ) . These results suggest that the activation of Vegfr2 in tip cells and stalk cells is crucial for their exit from the DA . 10 . 7554/eLife . 08817 . 022Figure 6 . Vegfr2 activation in stalk cells is crucial for their migration from the DA . ( A ) The cells dissociated from Tg ( fli1:Gal4FF ) ; ( UAS:NLS-mC ) or Tg ( fli1:Gal4FF ) ; ( UAS:Vegfr2-ΔC , NLS-mC ) embryos ( 34 hpf ) cultured on laminin-coated dish were kept untreated ( - ) or treated for 5 min with the supernatants from HEK293T cells transfected with ( + ) or without ( - ) Vegfaa-myc . The cells were immunostained with anti-phospho-Erk ( pErk ) antibody . mC images ( red ) and pErk images ( green ) are shown . Arrows indicate NLS-mC-positive ECs . ( B ) Quantitative analyses by the results of A are shown as dot-plot graphs depicting mean pixel intensity values with ± s . d . of nuclear pErk in NLS-mC-positive ECs . Each dot represents the value of single cell ( n > 20 ) . Similar results were obtained in four independent experiments . ( C ) Confocal stack fluorescence images of Tg ( fli1:GFP ) ; ( fli1:Gal4FF ) embryos at 36 ss injected with control UAS:NLS-mC plasmid ( upper ) or UAS:Vegfr2-ΔC , NLS-mC plasmid ( lower ) which drives the expression of NLS-mC or both Vegfr2-ΔC and NLS-mC simultaneously in ECs in a mosaic manner , respectively . ( D ) By counting the numbers of NLS-mC-positive ECs constituting tip cells , stalk cells , and DA cells as observed in C in an embryo , the percentage of each group among total number of NLS-mC-positive ECs is indicated . The data are derived from five independent experiments , in each of which ≥ 26 NLS-mC-positive cells were measured . Scale bars , 10 μm in A and C . p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 022 We then investigated Ca2+ dynamics after stalk cells completely came out from the DA . Ca2+ oscillations in stalk cells out of the DA were comparable to those in tip cells ( Figure 7A–C and Video 5 ) . The Ca2+ oscillations were also dependent upon Vegfr2 ( Figure 7C and Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 08817 . 023Figure 7 . Ca2+ oscillations in tip and stalk cells that completely come out from the DA . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos after stalk cells migrate out of the DA ( 29 ss ) . Green and orange arrowheads indicate tip and stalk cells , respectively . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , orange , light gray , dark gray , and black ) at the left panel are shown as a graph . ( C ) Quantification of Ca2+ oscillatory activity in untreated and ki8751-treated embryos after stalk cells migrate out of the DA as in A and Figure 7—figure supplement 1A , respectively . Graphs show Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells , stalk cells and other ECs within the DA in untreated and ki8751-treated embryos . ( Untreated , n ≥ 8; ki8751-treated , n ≥ 11 ) . ( D ) Quantification of the number of synchronous and asynchronous Ca2+ rise between tip and stalk cells . We here define the case , in which a Ca2+ rise in one cell occurs within 10 s late behind a Ca2+ rise in the other cell , as synchronous . The number of synchronous and asynchronous Ca2+ rise were quantified in tip and stalk cells . Percentages of synchronous and asynchronous Ca2+ increase to total Ca2+ increase are shown . The total number of Ca2+ rise analyzed in tip cells and stalk cells is indicated at the top . ( E ) Schematic illustration of Ca2+-oscillatory activity in stalk cells . Frequency of Ca2+ oscillations found in stalk cells are comparable to that in tip cells after stalk cells have completely come out from the DA . Scale bar , 10 mm in A . p < 0 . 001; NS , not significant . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02310 . 7554/eLife . 08817 . 024Figure 7—figure supplement 1 . Ca2+ responses after stalk cell budding from the DA are dependent upon Vegfr . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos treated with ki8751 after stalk cells completely migrate out of the DA as in Figure 7A . The embryos were treated from 28 ss with ki8751 and time-lapse imaged at 29 ss . Green and orange arrowheads indicate tip and stalk cells , respectively . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads at the left panel are shown as a graph . Scale bar , 10 μm in A . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02410 . 7554/eLife . 08817 . 025Figure 7—figure supplement 2 . An EC following a stalk cell exhibits significant Ca2+ oscillations . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 31 ss when a EC following tip and stalk cells is budding from the DA . Green and orange arrowheads indicate tip and stalk cells , respectively . A blue arrowhead indicates an EC following a stalk cell . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads ( green , blue , orange , light gray , dark gray , and black ) at the left panel are shown as a graph . Scale bar , 10 μm in A . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02510 . 7554/eLife . 08817 . 026Video 5 . Ca2+ oscillations occur both in tip and stalk cells after stalk cell migrates out of the DA . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos at 29 ss . Elapsed time is in seconds ( s ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 026 Ca2+ increases generally occur by two modes: one is intracellular Ca2+ waves which only propagate within a single cell; the other is intercellular Ca2+ waves ( ICWs ) which transmit from one cell to the next via gap junctions or extracellular messengers . To investigate whether Ca2+ oscillations were due to either mechanism , we measured how often the oscillations in tip or stalk cells were synchronized with adjacent stalk or tip cells , respectively . We found that 78 . 9% of Ca2+ oscillations in stalk cells occurred independently of tip cells , and 91 . 6% of oscillations in tip cells occurred independently of stalk cells ( Figure 7D ) . These results clearly indicate that most Ca2+ oscillations observed in tip and stalk cells occurs independently of each other and not through ICWs . Therefore , our findings suggest that not only tip cells but also stalk cells have potential to respond to angiogenic cues ( Figure 7E ) . Similarly to the earlier stage , Ca2+ oscillations were not detected in the ECs within the DA ( Figure 7A–C ) , except in ECs that were budding from the DA and following stalk cells ( Figure 7—figure supplement 2 ) . To understand how tip cell behavior is spatio-temporally regulated by Dll4/Notch signaling , we examined the effect of the inhibition of Dll4/Notch signaling on Ca2+ dynamics . Previous reports have shown that MO-mediated knockdown of Dll4 leads to an increased number of ECs exhibiting tip cell behavior in zebrafish ISVs , which phenocopies loss of Notch signaling ( Siekmann and Lawson , 2007 ) . Therefore , we analyzed Ca2+ oscillations before and during vessel sprouting in dll4 morphants ( Siekmann and Lawson , 2007 ) . Before sprouting of ISVs , the number of Ca2+-oscillating ECs and their oscillatory activity were significantly increased in the entire DA in dll4 morphants ( Figure 8A–C ) , suggesting that Dll4 suppresses the Vegfr2-dependent angiogenic responses in entire ECs of the DA . When ECs start budding at 19–22 ss , the Ca2+ responses were restricted to single ECs in 44 . 5% at somite boundaries in the control morphants , whereas those were , in 20 . 8% in dll4 morphants ( Figure 9A; 19–22 ss ) . Most dll4 morphants had two or more budding ECs exhibiting Ca2+ oscillations ( 79 . 2%; Figure 9A; 19–22 ss ) . Thus , our findings support the hypothesis that Dll4 restricts the number of ECs showing tip cell behavior when they sprout from the pre-existing vessels . In the later stages ( 24–27 ss ) , while the Ca2+ responses became restricted to single tip cells in the control embryos , two or more neighboring ECs exhibited Ca2+ oscillations in dll4 morphants ( 74 . 1%; Figure 9A; 24–27 ss ) . Consistently in the dll4 morphants , repetitive Ca2+ oscillations were maintained in the two neighboring ECs , both of which were budding from the DA ( Figure 9B , C and Video 6 ) . Similar results were observed in notch1b morphants ( Figure 9—figure supplement 1 ) . These results suggest that Dll4/Notch signaling regulates the selection of single tip cell from two or more ECs showing Ca2+ oscillation . In dll4 morphants , Ca2+ responses were detected exclusively in budding ECs , but not in other ECs within the DA ( Figure 9D ) , suggesting that Dll4 suppresses the tip cell behavior especially in ECs adjacent to budding tip cells , but not in the entire DA after tip cell budding . 10 . 7554/eLife . 08817 . 027Figure 8 . Dll4 attenuates Ca2+ oscillations in the entire DA before ISV sprouting . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos before ISV sprouting injected with dll4 MO ( 17 ss ) . Yellow dashed lines indicate positions of somite boundaries . ( B ) The DA is subdivided into three regions ( Region 1–3 ) as illustrated in the scheme ( left ) . The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A are shown as separated graphs ( Region 1–3 ) as in Figure 3B . A representative graph of two ECs at each region is shown . ( C ) Quantification of Ca2+ oscillation frequency ( upper ) and mean ΔF/F0 ( lower ) in ECs of the indicated regions within the DA in control MO- or dll4 MO-injected embryos before vessel sprouting ( 17–19 ss ) . Horizontal lines represent mean ± s . d . ( n ≥ 16 ) . Scale bar , 10 mm in A . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02710 . 7554/eLife . 08817 . 028Figure 9 . Dll4 is involved in suppressing Ca2+ responses in ECs adjacent to tip cells . ( A ) The number of Ca2+-oscillating ECs at each somite boundary of the embryo injected with control MO or dll4 MO was quantified as in Figure 4A . ( B ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip cell budding injected with dll4 MO ( 24 ss ) . Green and blue arrowheads indicate two neighboring Ca2+-oscillating ECs , both of which are budding from the DA . ( C ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from B indicated by arrowheads at the left panel are shown as a graph . ( D ) Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in budding ECs and other ECs within the DA in dll4 morphants during tip cell budding at 24–27 ss as illustrated at the left panel ( n ≥ 13 ) . Scale bar , 10 mm in B . p < 0 . 001 . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02810 . 7554/eLife . 08817 . 029Figure 9—figure supplement 1 . Ca2+ oscillations were maintained in two neighboring ECs in notch1b morphants during tip cell budding from the DA . 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip cell budding injected with notch1b MO ( 26 ss ) . Green and blue arrowheads indicate the two neighboring Ca2+-oscillating ECs , both of which are budding from the DA . Scale bar , 10 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 02910 . 7554/eLife . 08817 . 030Figure 9—figure supplement 2 . Vegfr is responsible for ectopic Ca2+ oscillations observed in dll4 morphants . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos injected with dll4 MO and treated with ki8751 during tip cell budding ( 26 ss ) . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads at the left panel are shown as a graph . Scale bar , 10 mm in A . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 03010 . 7554/eLife . 08817 . 031Figure 9—figure supplement 3 . Vegfr3 is partially involved in increases in oscillating cells in dll4 morphants . ( A ) The number of Ca2+-oscillating ECs at each somite boundary of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos injected with both dll4 and vegfr3 MOs was quantified at somite boundaries ( total 38 ) at 24–27 ss as in Figure 4A . ( B ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos injected with both dll4 and vegfr3 MOs during tip cell budding ( 24 ss ) . Green and blue arrowheads indicate two neighboring Ca2+-oscillating ECs , both of which are budding from the DA . ( C ) Fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from B indicated by arrowheads at the left panel are shown as a graph . Scale bar , 10 mm in B . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 03110 . 7554/eLife . 08817 . 032Figure 9—figure supplement 4 . The expression of neither vegfr2 nor kdr is not altered in the trunk vessels of dll4 morphants . Whole-mount in situ hybridization ( WISH ) analyses of the embryos ( 26–27 ss ) uninjected or injected with dll4 MO using antisense probe for vegfr2 ( also termed kdrl ) and kdr ( kdrb ) , a paralog of vegfr2 . Note that the expression of vegfr2 and kdr mRNA in the DA and the sprouting ISVs observed in wild-type embryos is not altered in dll4 morphants . A set of representative images of two independent experiments is shown . Scale bar , 100 μm . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 03210 . 7554/eLife . 08817 . 033Video 6 . Ca2+ oscillations are maintained in two neighboring cells in the absence of dll4 during tip cell budding . Time-lapse recording of 3D-rendered light sheet images of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during tip cell budding from the DA injected with dll4 morpholino antisense oligo ( MO ) . The recording started at 24 somite stage . Elapsed time is in seconds ( s ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 033 Next , we investigated the effect of inhibition of Vegfr in the dll4 morphants on Ca2+ dynamics . Treatment with ki8751 completely abolished the Ca2+ oscillations observed in the dll4 morphants ( Figure 9—figure supplement 2 ) , suggesting that an increase in the number of oscillating cells in the dll4 morphants reflects an increase in the number of ECs showing either Vegfr2 or Vegfr3 activation . In zebrafish , arterial overproliferation and hypersprouting in the absence of Dll4/Notch signaling is , at least partially , attributed to ectopically enhanced Vegfr3 activation ( Hogan et al . , 2009; Siekmann and Lawson , 2007 ) . Accordingly , Vegfr3 might be involved in the ectopic Ca2+ oscillations during vessel sprouting in dll4 morphant . To assess the contribution of Vegfr3 to the ectopic Ca2+ oscillations , we examined Ca2+ oscillations in vegfr3/dll4 double morphants at 24–27 ss when the Ca2+ responses became restricted to single tip cells in wild-type embryos ( Figure 4A ) . In vegfr3/dll4 double morphants , Ca2+ oscillations were detected in two or more neighboring cells in 65 . 8% at somite boundaries ( Figure 9—figure supplement 3 ) , although the phenotype in the double morphants was slightly milder than that in the dll4 morphants ( 74 . 1%; Figure 9A; 24–27 ss ) . Thus , Vegfr3 is partially involved in increases in oscillating cells in the absence of Dll4/Notch signaling . On the other hand , these results imply that ectopic Ca2+ oscillations occur even in the absence of vegfr3 after inhibiting Dll4/Notch signaling . Considering that the Ca2+ responses in dll4 morphants were completely abolished by ki8751 treatment ( Figure 9—figure supplement 2 ) , our findings suggest that Dll4/Notch signaling might restrict the tip cell number mainly by suppressing Vegfr2 signaling . Since Vegfr3 is responsible for hypersprouting of ISVs in the dorsal part of the dll4 morphants ( Hogan et al . , 2009 ) , Dll4/Notch signaling might not suppress Vegfr3 signaling in the early stage but suppress it in the later stage . To investigate the mechanism how Dll4/Notch signaling suppresses Vegfr2 signaling in ECs of zebrafish , we tested the effect of inhibiting Dll4/Notch signaling on the expression of vegfr2 during sprouting of the ISVs . However , we could detect any increase in the expression of neither vegfr2 nor its paralog kdr in the trunk vessels of dll4 morphants ( Figure 9—figure supplement 4 ) , suggesting that Dll4/Notch signaling inhibits the Vegfr2 signaling cascade rather than vegfr2 expression . We further investigated the role of Dll4/Notch signaling in stalk cell selection . Of note , when stalk cells began to migrate from the DA , Ca2+ oscillations were mainly detected only in single stalk cells in the DA , but not in adjacent cells apart from tip cells ( Figure 10A ) . Because Dll4/Notch signaling regulates the selection of single tip cells ( Figure 9 ) ( Eilken and Adams , 2010; Lohela et al . , 2009; Phng and Gerhardt , 2009 ) , we hypothesized that it might also regulate the selection of single stalk cell . To test this hypothesis , we examined the effect of knockdown of Dll4 on Ca2+ dynamics during stalk cell budding . While Ca2+ oscillations were mostly detected in single stalk cell in the control embryos , they were often detected in two or more neighboring ECs following tip cells in dll4 morphants ( Figure 10A–D ) . Longer time-lapse imaging confirmed that both Ca2+-oscillating stalk cell in the dll4 morphants eventually came out from the DA to form the ISVs . These results suggest that Dll4/Notch signaling regulates the selection of single stalk cell by restricting angiogenic behavior in the ECs adjacent to the stalk cells . 10 . 7554/eLife . 08817 . 034Figure 10 . Dll4 is involved in the selection of single stalk cells . ( A ) The number of Ca2+-oscillating ECs following tip cells in each ISV of uninjected , or control MO- or dll4 MO-injected Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during stalk cell budding when an EC/ECs following a tip cell is/are budding from the DA . Graph shows the occurrence rate of the indicated numbers of Ca2+-oscillating cells in each ISV among the total number of ISVs observed ( indicated at the top ) . ( B ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos during stalk cell budding from the DA injected with dll4 MO ( 25 ss ) . A green arrowhead indicates tip cell . Orange and red arrowheads indicate budding stalk cells following tip cell . ( C ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from B indicated by arrowheads at the left panel are shown as a graph . ( D ) Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells , stalk cells , and other ECs within the DA in dll4 morphants during stalk cell budding ( n ≥ 10 ) . As illustrated at the left panel , we designated budding ECs that follow tip cells as stalk cells . Scale bar , 10 mm in B . *p < 0 . 001; NS , not significant . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 034 We then investigated the role of Dll4/Notch signaling after stalk cell budding from the DA . Even in the absence of dll4 , we could not detect significant increases in oscillatory activity in both tip and stalk cells that came out of the DA ( Figure 11A–C ) . Thus , Dll4/Notch signaling does not suppress the Ca2+-responses in the initial sprouting from the DA to form ISVs . 10 . 7554/eLife . 08817 . 035Figure 11 . Dll4 does not regulate Ca2+ oscillations in stalk cells after they completely migrate out of the DA . ( A ) 3D-rendered time-sequential images of Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos after stalk cell budding from the DA injected with dll4 MO ( 27 ss ) . Green and orange arrowheads indicate tip and stalk cells , respectively . ( B ) The fluorescence changes in GCaMP7a ( ΔF/F0 ) of individual ECs from A indicated by arrowheads at the left panel are shown as a graph . ( C ) Ca2+ oscillation frequency ( left ) and mean ΔF/F0 ( right ) in tip cells , stalk cells , and other ECs within the DA in control MO- or dll4 MO-injected embryos after stalk cell budding from the DA as illustrated at the left panel . ( n ≥ 9 ) . Scale bar , 10 mm in A . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 035 In the present study , we visualized Ca2+ dynamics , for the first time , in ECs during sprouting angiogenesis in vivo . By quantitatively analyzing Ca2+ dynamics in individual ECs , we have uncovered how Ca2+ responses are spatio-temporally regulated at the single-cell level ( Figure 12 ) . Intracellular Ca2+ oscillations occurred in ECs exhibiting angiogenic behavior , such as sprouting and migration . Therefore , visualizing Ca2+ oscillations allowed us to monitor EC responses to angiogenic cues . While Ca2+ oscillations depended upon Vegfa/Vegfr2 signaling in ECs sprouting from the DA , they were also detected in venous sprouts in response to Vegfc/Vegfr3 . Thus , Ca2+ oscillations in ECs occur during both arterial and venous sprouting and subsequent migration . 10 . 7554/eLife . 08817 . 036Figure 12 . A schematic representation of endothelial Ca2+ responses during sprouting angiogenesis from the DA . Ca2+ oscillations occur widely within the DA before ISV sprouting . Dll4/Notch signaling attenuates Ca2+ responses in the entire DA at this stage . Ca2+ oscillations are restricted to single or two neighboring ECs just after vessel sprouting and then further restricted to single ECs that eventually become tip cells . Dll4/Notch signaling is required for the selection of single tip cells . During stalk cell budding , tip cells and stalk cells exhibit Ca2+ oscillations , although the Ca2+-oscillatory activity in stalk cells is weaker than that in tip cells . Dll4/Notch signaling regulates the selection of single stalk cells . After stalk cells completely come out from the DA , strong Ca2+ oscillations occur both in tip cells and stalk cells . Intensity of green reflects the frequency of Ca2+ oscillations . DA , dorsal aorta . DOI: http://dx . doi . org/10 . 7554/eLife . 08817 . 036 Oscillatory increases of intracellular Ca2+ were dependent upon Vegfa/Vegfr2 signaling in ECs budding from the DA . VEGF-A treatment induces Ca2+ rise in cultured human ECs , although their Ca2+ response is not oscillatory ( Brock et al . , 1991; Favia et al . , 2014; Li et al . , 2011 ) . In vitro , VEGF-A induces Ca2+ rise by the PLCγ-IP3-IP3R pathway ( Koch and Claesson-Welsh , 2012; Moccia et al . , 2012 ) , by store-operated Ca2+ entry involving Orai1 and CRAC channel ( Li et al . , 2011 ) and by lysosomal Ca2+ release involving two-pore channel TPC2 ( Favia et al . , 2014 ) . Since intracellular Ca2+ oscillations generally occur through the concerted action between several Ca2+ transporters ( Smedler and Uhlen , 2014 ) , further studies are needed to delineate whether Vegfr2 signaling directly regulates Ca2+ store or Ca2+ channels in ECs in vivo . Furthermore , it is also required to detect where intracellular Ca2+ increases during sprouting angiogenesis . Although we observed an increase in Ca2+ throughout the cytoplasm , if ECs respond to Vegfa , the initial increase of Ca2+ might be found in a manner dependent upon the distribution of Vegfr2 . In cultured ECs , they exhibit Ca2+ oscillations only in the front of the cells but not in the rear during migration ( Tsai et al . , 2014 ) . These polarized Ca2+ oscillations might be detected if we can improve the time/spatial resolution of imaging of zebrafish in vivo . Ca2+ imaging at single-cell resolution pointed to the new regulatory mechanism underlying the formation of stalk cells following tip cells in the ISV . Lateral inhibitory action of Dll4/Notch signaling is essential for selection of tip cells ( Eilken and Adams , 2010; Lohela et al . , 2009; Phng and Gerhardt , 2009 ) ; however , it is unknown how stalk cells are specified and migrate from the parental vessels . We found that stalk cells have Vegfr-dependent Ca2+ oscillations when they start budding following tip cells and that Dll4/Notch signaling is required for the selection of single stalk cell similarly to tip cell selection ( Figure 12 ) . VEGF-A/VEGFR2 signaling is active in stalk cells and is important for cell proliferation in mouse retina angiogenesis ( Gerhardt et al . , 2003 ) . Here , we propose that Vegfr2 activation in stalk cells is also necessary for the selection of stalk cells and their exit from the parental vessels during ISV formation . Thus , our Ca2+ imaging highlighted a novel role of Vegfr2 in stalk cells . Tip cells and stalk cells are selected during sprouting angiogenesis . In the present study , we focused on the stage of ECs budding from the DA and observed that even two cells showing Ca2+ oscillations budded from the DA and that one of them finally became single tip cell . Jacobsson et al . report that tip cells and stalk cells change their positions in elongating vessels sprouts ( Jakobsson et al . , 2010 ) , whereas in the earlier stage , we did not observe interexchange of tip cells and stalk cells , suggesting that their plasticity of tip and stalk depends upon the stage and type of angiogenesis . Dll4/Notch signaling is required for selection of single tip cell and single stalk cell during budding from the DA . Our findings support the notion that Dll4 in tip cells suppresses tip cell behavior in adjacent cells for the initial selection of tip cell ( Eilken and Adams , 2010; Herbert and Stainier , 2011; Lohela et al . , 2009; Phng and Gerhardt , 2009 ) . Meanwhile , it has not been determined whether tip cells also suppress angiogenic responses in stalk cells via Dll4/Notch signaling during vessel elongation . Our results indicate that inhibitory activity of Dll4 toward Ca2+ responses in stalk cells diminishes after stalk cells come out from the DA . Instead , Dll4/Notch signaling restricts the number of stalk cells . Thus , our results suggest that Dll4/Notch-mediated lateral inhibition leads to suppression of angiogenic responses mainly in ECs within the DA adjacent to budding tip or stalk cells to restrict excess budding from the DA ( Figure 12 ) . Our Ca2+ imaging provide novel insights into how Dll4/Notch signaling spatio-temporally controls angiogenic behavior . Endothelial Ca2+ responses can be used as an indicator of cellular responses to extracellular stimuli . In this study , we show that endothelial Ca2+ oscillations occur during sprouting angiogenesis in response to Vegfa/Vegfr2 and Vegfc/Vegfr3 signaling . We could investigate endothelial Ca2+ responses induced by other chemical or mechanical stimuli . Among them , blood flow is a well-studied input that induces an intracellular Ca2+ increase in ECs ( Ando and Yamamoto , 2013 ) . Blood flow-dependent Ca2+ increases were recently reported in zebrafish embryos ( Goetz et al . , 2014 ) . Our Ca2+ imaging analyses have captured flow-dependent Ca2+ oscillations in various types of vessels ( Yokota et al . , unpublished data ) : therefore , Ca2+ imaging analysis will be a useful system for understanding quantitatively how individual ECs are regulated by blood flow in vivo . Thus , in vivo Ca2+ imaging in ECs presented here will become a powerful tool to investigate the dynamic behavior of individual ECs in vascular development and homeostasis in the future studies . Zebrafish ( Danio rerio ) were maintained and bred under standard conditions . The experiments using zebrafish were approved by the animal committee of National Cerebral and Cardiovascular Center ( No . 14005 ) and performed according to the guidance of the Institute . cDNA fragments encoding zebrafish H2B , sFlt1 , Vegfr2 ( Kdrl ) , Kdr , and Vegfaa were amplified by PCR using a cDNA library derived from zebrafish embryos and cloned into pCR4 Blunt TOPO vector ( Invitrogen , Carlsbad , CA ) . A cDNA encoding mCherry was subcloned into the pTol2-fli1 vector to construct the pTol2-fli1:mC plasmid ( Wakayama et al . , 2015 ) . Then , the cDNA encoding H2B was inserted into the pTol2-fli1-mC vector to generate the pTol2-fli1:H2B-mC plasmid . An oligonucleotide-encoding nuclear localization signal ( NLS ) derived from SV40 ( PKKKRKV ) was inserted into pmCherry-C1 vector ( Clontech , Takara Bio Inc . , Japan ) to generate the plasmid encoding NLS-tagged mCherry ( NLS-mC ) . The NLS-mC cDNA followed by polyA sequece was subcloned into the pTol2-E1b-UAS-E1b vector to construct the pTol2-E1b-UAS-E1b:NLS-mC plasmid ( also used as the UAS:NLS-mC plasmid for microinjection ) ( Wakayama et al . , 2015 ) . The cDNA encoding zebrafish sFlt1 was inserted into the pTol2-E1b-UAS-E1b:NLS-mC vector to construct the pTol2-UAS:sFlt1 , NLS-mC plasmid . The pTol2-UAS:Vegfr2-ΔC , NLS-mC was constructed by inserting a cDNA fragment that encodes Vegfr2 lacking the intracellular domain ( amino acid 783–1181 ) into the pTol2-E1b-UAS-E1b:NLS-mC vector . The nuclear export signal ( NES , derived from a HIV-1Rev protein; LQLPPLERLTLD ) -tagged mC cDNA followed by polyA signal was subcloned into the pTol2-E1b-UAS-E1b vector to construct the pTol2-E1b-UAS-E1b:NES-mC plasmid . The cDNA encoding full length Vegfr2 was then inserted into the pTol2-E1b-UAS-E1b:NES-mC vector to construct the pTol2-UAS:Vegfr2 , NES-mC plasmid . The cDNA encoding zebrafish Vegfaa was inserted into the pcDNA3 . 1 vector ( Invitrogen ) together with 1xMyc tag to construct the pcDNA3 . 1-Vegfaa-myc . To generate the Tg ( fli1:H2B-mC ) , Tg ( UAS:NLS-mC ) , and Tg ( UAS:Vegfr2-ΔC , NLS-mC ) zebrafish lines , the corresponding pTol2-based plasmid DNAs ( 15 pg ) were microinjected along with Tol2 transposase mRNA ( 30 pg ) into one-cell stage embryos of AB or Tg ( fli1:Gal4FF ) zebrafish . Tol2 transposase mRNAs were in vitro transcribed with SP6 RNA polymerase from NotI-linearized pCS-TP vector using the mMESSAGE mMACHINE kit ( Ambion , Thermo Fisher Scientific , Waltham , MA ) . The embryos showing transient mCherry expression in the vasculature were selected , raised to adulthood , and crossed with wild-type AB to identify germline transmitting founder fishes . Tg ( UAS:GCaMP7a ) fish and Tg ( UAS:GFP ) ;SAGFF ( LF ) 27C fish were used ( Bussmann et al . , 2010; Muto et al . , 2013 ) . Tg ( fli1:EGFP ) fish were provided by N Lawson ( University Massachusetts Medical School , USA ) ( Lawson and Weinstein , 2002 ) . Tg ( fli1:Gal4FF ) fish line was a gift from M Affolter ( University of Basel , Switzerland ) ( Totong et al . , 2011; Zygmunt et al . , 2011 ) . For morpholino oligonucleotide ( MO ) -mediated knockdown , embryos were injected at one-cell or two-cell stage with control MO ( Gene Tools , LLC , Philomath , OR ) , 3 ng of vegfr2 MO , 7 ng of dll4 MO , 3 ng of vegfr3 MO , 3 ng of plxnD1 , and 7 ng of notch1b MO . The sequences for the already-validated MOs used in this study are: vegfr2 MO , 5′-CACAAAAAGCGCACACTTACCATGT-3′;dll4 MO , 5′- TAGGGTTTAGTCTTACCTTGGTCAC-3′; vegfr3 MO , 5′-TTAGGAAAATGCGTTCTCACCTGAG-3′; plxnD1 MO , 5′- CACACACACTCACGTTGATGATGAG-3′; notch1b MO , 5′-AATCTCAAACTGACCTCAAACCGAC-3′ ( Leslie et al . , 2007; Siekmann and Lawson , 2007; Torres-Vázquez , et al . , 2004; Wiley et al . , 2011 ) . To express sFlt1 , Vegfr2-ΔC , and full length Vegfr2 transiently using the Tol2 system , we co-injected the plasmids with capped Tol2 transposase mRNA ( 30 pg ) ( Kawakami et al . , 2004 ) . To co-express sFlt1 and NLS-mC transiently in ECs , 30 pg pTol2-UAS:sFlt1 , NLS-mC ( UAS:sFlt1 , NLS-mC ) plasmid was injected into one-cell stage of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) . In these embryos , the expression of sFlt1 and NLS-mC is simultaneously driven in ECs by a Gal4/UAS-based bidirectional expression system . To express both Vegfr2-ΔC and NLS-mC simultaneously in ECs , 30 pg pTol2-UAS:Vegfr2-ΔC , NLS-mC ( UAS:Vegfr2-ΔC , NLS-mC ) plasmid was injected into one-cell stage of the Tg ( fli1:Gal4FF ) ; ( fli1:GFP ) . In these embryos , expression was induced in a mosaic manner ( see Figure 6C ) . As a negative control , 30 pg pTol2-UAS:NLS-mC ( UAS:NLS-mC ) plasmid was injected . To express Vegfr2 and NES-mC simultaneously in ECs , 30 pg pTol2-UAS:Vegfr2 , NES-mC ( UAS:Vegfr2 , NES-mC ) plasmid was injected into one-cell stage of the Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) . The Tg ( fli1:Gal4FF ) ; ( UAS:GCaMP7a ) ; ( fli1:H2B-mC ) embryos were treated with 1 μM ki8751 , an inhibitor of Vegfr , from 40–90 min before starting light sheet imaging , and subsequently during the imaging . For light sheet imaging , zebrafish embryos were dechorionated and anesthetized in 0 . 4 mg/ml rocuronium bromide ( Merk Sharp and Dohme , Kenilworth , NJ ) in E3 embryo medium ( 5 mM NaCl , 0 . 17 mM KCl , 0 . 33 mM CaCl2 , 0 . 33 mM MgSO4 ) . The anesthetized embryos were pulled into the small glass capillaries containing 1% low-melting agarose in E3 medium , using a metal plunger . After the gel set , the fish were slightly extruded from the capillary and immersed in the chamber containing E3 medium . Light sheet imaging was performed with a Lightsheet Z . 1 system ( Carl Zeiss , Germany ) equipped with a water immersion 20x detection objective lens ( W Plan Apochromat , NA 1 . 0 ) , dual sided 10x illumination objective lenses ( LSMF , NA 0 . 2 ) , a pco . edge scientific CMOS camera ( PCO ) and ZEN software . For all 3D time-lapse datasets , a z-interval of 4 μm , and a time interval of 5 s were used . z-stack images were 3D volume rendered and analyzed with IMARIS 7 . 7 . 1 software ( Bitplane AG , Switzerland ) . For 2D time-lapse imaging , a time interval of 100 ms was used . For confocal imaging , zebrafish embryos were dechorionated , anesthetized in 0 . 016% tricaine ( Sigma-Aldrich , St . Louis , MO ) in E3 medium , and mounted in 1% low-melting agarose poured onto a 35-mm diameter glass-based dish ( Asahi Techno Glass , Japan ) as previously described ( Fukuhara et al . , 2014 ) . Confocal images were taken with a FluoView FV1000 confocal uplight microscope system ( Olympus , Japan ) equipped with a water-immersion 20x lens ( XLUMPlanFL , NA 1 . 0 ) . Images were analyzed with FV10-ASW3 . 1 viewer ( Olympus ) . In our Ca2+ imaging using light sheet microscopy , we analyzed the ISVs and DA located in middle trunk region of embryos at somite levels 8 to 14 . Time-lapse images were collected every 5 s for 1000–2000 s with a z-interval of 4 μm . Z-stack images were 3D volume-rendered and analyzed with IMARIS 7 . 7 . 1 software ( Bitplane AG ) . ISV sprouts emerge bilaterally from the DA . Thus , to focus on sprouting angiogenesis occurring on one side , our analyses were performed at either side ( left or right side ) of the embryos by cropping . To quantify intracellular Ca2+ levels of individual ECs at each time point , the cell nuclei labeled with H2B-mC were automatically ( or manually in some cases ) tracked over time using IMARIS software . Then , for each individually tracked EC , we set a spherical region of interest ( ROI ) of 4–11 μm in diameters that is slightly larger than the nucleus ( see Figure 1—figure supplement 2D ) . The diameters of every spherical ROI were carefully set to cover the nucleus and part of the cytoplasm but do not overlap with adjacent ECs . We then defined the highest voxel intensity of the GCaMP7a fluorescence within the ROI as the florescent intensity ( F ) of GCaMP7a in the EC . Because intracellular Ca2+ waves spread rapidly and uniformly throughout the cytoplasm of each EC ( see Figure 1—figure supplement 2B ) , the spherical ROIs including part of the cytoplasmic region are enough to detect intracellular Ca2+ waves . ΔF/F0 was calculated as ( F − F0 ) / F0 , where F0 is the baseline florescent intensity of GCaMP7a averaged over a 50-s period . Fluorescence changes in GCaMP7a of individual ECs are represented as ΔF/F0 traces in the graphs , and mean ΔF/F0 was calculated by taking the average of every ΔF/F0 . The frequency of intracellular Ca2+ oscillations was counted as the number of Ca2+ oscillations per min . ΔF/F0 increase of 20% from the baseline was defined as an oscillation , while 100% is the average of the three highest ΔF/F0 peaks in oscillating ECs of wild-type embryos . The ECs , which underwent mitosis through the time-lapse images , were removed from our quantification analyses because we found that ECs never exhibit any Ca2+ rise during M phase . HUVECs were purchased from Kurabo ( Japan ) , maintained on a collagen-coated dish in endothelial cell growth medium ( EGM-2 , Lonza , Switzeland ) , and used for the experiments before passage 7 . HEK293T cells were cultured in DMEM ( Nacalai Tesque , Japan ) with 10% FBS and antibiotics ( 100 mg/ml streptomycin and 100 U/ml penicillin ) . HUVECs and 293T cells were transfected with plasmid DNA using ViaFect transfection reagent ( Promega , Madison , WI ) and 293fectin transfection reagent ( Invitrogen ) , respectively . HUVECs expressing GCaMP7a were time-lapse imaged with an inverted fluorescence microscope ( IX-81; Olympus ) equipped with a Plan-Apochromat 40x/1 . 00 NA oil immersion objective lens ( Olympus ) and with a pE-1 LED excitation system ( CoolLED ) with a cooled charge-coupled device camera ( Neo5 . 5 sCMOS; Andor Technology , UK ) . The cells were imaged at 37 oC with 5% CO2 using a heating chamber ( Tokai Hit , Japan ) . Tg ( fli1:Gal4FF ) ; ( UAS:NLS-mC ) or Tg ( fli1:Gal4FF ) ; ( UAS:Vegfr2-ΔC , NLS-mC ) embryos at 34 hpf were dechorionated and collected into a 1 . 5-ml tube after removing the yolk , washed in phosphate buffered saline ( PBS ) , and digested with 500 μl of protease solution ( PBS with 1 mg/ml trypsin , 2 . 7 mg/ml Collagenase P and 1mM EDTA , pH 8 . 0 ) for 15 min at 28oC under occasional pipetting . Digestion of the embryos was terminated with 50 μl of stop solution ( PBS with 30% fetal bovine serum [FBS] and 6 mM calcium chloride ) . The dissociated cells were filtered with 35-μm cell strainers ( BD Falcon , Thermo Fisher Scientific ) and subsequently cultured on glass-base dish coated with laminin in L15 medium ( Life Technologies , Carlsbad , CA ) with 50 U/ml penicillin and 0 . 05 mg/ml streptomycin at 28oC . The cells spread on the culture dishes for 3 . 5 hr were used for experiments . The supernatants of HEK293T cells transfected with or without the plasmid DNA encoding Vegfaa-myc were collected 36 hr after the transfection and added to zebrafish primary cells . After stimulation , the cells were fixed with 4% paraformaldehyde ( PFA ) in PBS for 15 min at RT , permiabilized with 0 . 2% Triton X-100 for 30 min at RT and blocked with 3% BSA in PBS for 30 min at RT . The cells were immunostained with anti-phospho-ERK ( pERK ) antibody ( #4370 , Cell Signaling Technology , Danvers , MA ) in 3% BSA in Can Get Signal Immunostain solution ( TOYOBO , Japan ) at 4°C overnight . Protein reacting with the antibody was visualized with species-matched Alexa-Fluor 488-labeled secondary antibody ( Invitrogen ) . Fluorescence images were taken with an inverted fluorescence microscope ( IX-81 ) equipped with a Plan-Apochromat 60x/1 . 40 NA oil immersion objective lens , an X-cite 110LED excitation system ( Excelitas Technologies , Waltham , MA ) and a cooled charge-coupled device camera ( CoolSNAP HQ; Roper Scientific , Photometrics , Tucson , AZ ) . The microscope and image acquisition were controlled by MetaMorph software ( Molecular Devices , Sunnyvale , CA ) . Quantification of the fluorescence intensity of pErk staining was performed with MetaMorph software . We defined a region of interest ( ROI ) kept at constant size ( 4 μm in diameter ) inside the nucleus of NLS-mC-expressing ECs , and measured the mean fluorescence intensity of pErk staining within the ROI after subtracting the background . Whole-mount in situ hybridization ( WISH ) of zebrafish embryos was performed as described previously ( Fukuhara et al . , 2014 ) . Data were analyzed using GraphPad Prism software . Statistical significance for paired samples was determined using Student’s t test . Data were considered statistically significant at p < 0 . 05 .	Throughout life , new blood vessels grow out like branches from existing vessels in a process called “sprouting angiogenesis” . This involves some of the endothelial cells that line the inner surface of the blood vessel migrating outwards , creating a vessel sprout made up of tip cells and stalk cells . Sprouting is controlled by two opposing signaling systems . One pathway is triggered by a molecule called vascular endothelial growth factor ( VEGF ) . This molecule binds to receptor proteins to activate a range of signaling processes that stimulate endothelial cells to become tip cells , and so encourage the formation of new sprouts . However , it was not known exactly when or how the endothelial cells respond to these signals . By contrast , the Notch signaling pathway inhibits sprouting angiogenesis . The two signaling pathways interact with each other: VEGF signaling in tip cells activates Notch signaling in neighboring cells , which then prevents VEGF signaling in these cells . This feedback mechanism helps a new sprout to form by suppressing tip-like activity in the cells surrounding a new tip cell , forcing these cells to become stalk cells . Activating VEGF receptors also causes brief increases , or oscillations , in the level of calcium ions inside the endothelial cells . Now , Yokota , Nakajima et al . have investigated VEGF activity by genetically engineering zebrafish embryos so that fluorescent proteins inside their endothelial cells emit more light when calcium ion levels inside the cell increase . As zebrafish embryos are transparent , this change in fluorescence can be seen in the living animal . Imaging the embryos revealed that calcium ion oscillations occur in both tip and stalk cells in response to VEGF signaling as they bud from vessels . Notch signaling can also regulate the calcium ion oscillations; this controls whether an individual cell becomes a tip or a stalk cell , and restricts the number of stalk cells in the sprout . The flow of blood through the vessels is also thought to influence calcium ion oscillations in endothelial cells . Future studies could therefore use the imaging technique developed by Yokota , Nakajima et al . to investigate how blood flow influences the development of new blood vessels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Endothelial Ca2+ oscillations reflect VEGFR signaling-regulated angiogenic capacity in vivo
Neutralizing antibodies elicited by prior infection or vaccination are likely to be key for future protection of individuals and populations against SARS-CoV-2 . Moreover , passively administered antibodies are among the most promising therapeutic and prophylactic anti-SARS-CoV-2 agents . However , the degree to which SARS-CoV-2 will adapt to evade neutralizing antibodies is unclear . Using a recombinant chimeric VSV/SARS-CoV-2 reporter virus , we show that functional SARS-CoV-2 S protein variants with mutations in the receptor-binding domain ( RBD ) and N-terminal domain that confer resistance to monoclonal antibodies or convalescent plasma can be readily selected . Notably , SARS-CoV-2 S variants that resist commonly elicited neutralizing antibodies are now present at low frequencies in circulating SARS-CoV-2 populations . Finally , the emergence of antibody-resistant SARS-CoV-2 variants that might limit the therapeutic usefulness of monoclonal antibodies can be mitigated by the use of antibody combinations that target distinct neutralizing epitopes . Neutralizing antibodies are a key component of adaptive immunity against many viruses that can be elicited by natural infection or vaccination ( Plotkin , 2010 ) . Antibodies can also be administered as recombinantly produced proteins or as convalescent plasma to confer a state of passive immunity in prophylactic or therapeutic settings . These paradigms are of particular importance given the emergence of SARS-CoV-2 , and the devastating COVID19 pandemic that has ensued . Indeed , interventions to interrupt SARS-CoV-2 replication and spread are urgently sought , and passively administered antibodies are currently among the most promising therapeutic and prophylactic antiviral agents . Moreover , an understanding of the neutralizing antibody response to SARS-CoV-2 is critical for the elicitation of effective and durable immunity by vaccination ( Kellam and Barclay , 2020 ) . Recent studies have shown that related , potently neutralizing monoclonal antibodies that recognize the SARS-CoV-2 receptor-binding domain ( RBD ) are often elicited in SARS-CoV-2 infection ( Robbiani et al . , 2020; Brouwer et al . , 2020; Cao et al . , 2020; Chen et al . , 2020; Chi et al . , 2020; Rogers et al . , 2020; Shi et al . , 2020; Wu et al . , 2020a; Wec et al . , 2020; Kreer et al . , 2020; Hansen et al . , 2020; Ju et al . , 2020; Seydoux et al . , 2020; Liu et al . , 2020; Zost et al . , 2020 ) . These antibodies have great potential to be clinically impactful in the treatment and prevention of SARS-CoV-2 infection . The low levels of somatic hypermutation and repetitive manner in which similar antibodies ( e . g . those based on IGHV3-53 Robbiani et al . , 2020; Barnes et al . , 2020; Yuan et al . , 2020 ) have been isolated from COVID19 convalescents suggests that potently neutralizing responses should be readily elicited . Paradoxically , a significant fraction of COVID19 convalescents , including some from whom potent neutralizing antibodies have been cloned , exhibit low levels of plasma neutralizing activity ( Robbiani et al . , 2020; Wu et al . , 2020b; Luchsinger et al . , 2020 ) . Together , these findings suggest that natural SARS-CoV-2 infection may often fail to induce sufficient B-cell expansion and maturation to generate high-titer neutralizing antibodies . The degree to , and pace at which SARS-CoV-2 might evolve to escape neutralizing antibodies is unclear . The aforementioned considerations raise the possibility that SARS-CoV-2 evolution might be influenced by frequent encounters with sub-optimal concentrations of potently neutralizing antibodies during natural infection . Moreover , the widespread use of convalescent plasma containing unknown , and often suboptimal , levels of neutralizing antibodies might also increase the acquisition of neutralizing antibody resistance by circulating SARS-CoV-2 populations ( Bloch et al . , 2020; Al‐Riyami et al . , 2020 ) . Reinfection of previously infected individuals who have incomplete or waning serological immunity might similarly drive emergence of antibody escape variants . As human neutralizing antibodies are discovered and move into clinical development as therapeutics and prophylactics , and immunogens based on prototype SARS-CoV-2 spike protein sequences are deployed as vaccines , it is important to anticipate patterns of antibody resistance that might arise . Here , we describe a recombinant chimeric virus approach that can rapidly generate and evaluate SARS-CoV-2 S mutants that escape antibody neutralization . We show that mutations conferring resistance to convalescent plasma or RBD-specific monoclonal antibodies can be readily generated in vitro . Notably , these antibody resistance mutations are present at low frequency in natural populations . Importantly , the use of candidate monoclonal antibody combinations that target distinct epitopes on the RBD ( and therefore have non-overlapping resistance mutations ) can suppress the emergence of antibody resistance . To select SARS-CoV-2 S variants that escape neutralization by antibodies , we used a recently described replication-competent chimeric virus based on vesicular stomatitis virus that encodes the SARS-CoV-2 spike ( S ) protein and green fluorescent protein ( rVSV/SARS-CoV-2/GFP ) ( Schmidt et al . , 2020 ) . Notably , rVSV/SARS-CoV-2/GFP replicates rapidly and to high-titers ( 107 to 108 PFU/ml within 48 hr ) , mimics the SARS-CoV-2 requirement for ACE-2 as a receptor , and is neutralized by COVID19 convalescent plasma and SARS-CoV-2 S-specific human monoclonal antibodies ( Schmidt et al . , 2020 ) . The replication of rVSV/SARS-CoV-2/GFP can be readily monitored and measured by GFP fluorescence and the absence of proof-reading activity in the viral polymerase ( VSV-L ) results in the generation of virus stocks with greater diversity than authentic SARS-CoV-2 , for an equivalent viral population size . These features facilitate experiments to investigate the ability S protein variants to escape antibody neutralization . We used two adapted , high-titer variants of rVSV/SARS-CoV-2/GFP , ( namely rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 ) ( Schmidt et al . , 2020 ) in attempts to derive antibody-resistant mutants . Virus populations containing 1 × 106 infectious particles were generated following three passages to generated sequence diversity . On the third passage , cells were infected at an MOI of ~ 0 . 5 and progeny harvested after as short a time as possible so as to minimize phenotypic mixing in the viral population and to maximize the concordance between the genome sequence and the S protein sequence represented in a given virion particle . Because the mutation rate of VSV is ~ 10−4 to 10−5/base per replication cycle ( Steinhauer and Holland , 1986; Steinhauer et al . , 1989; Combe and Sanjuán , 2014 ) , we estimated that this procedure should generate a large fraction of the possible replication-competent mutants within a population size of 1 × 106 . The viral populations were then incubated with antibodies to neutralize susceptible variants ( Figure 1A ) . For monoclonal antibodies , viral populations were incubated with antibodies at 5 μg/ml or 10 μg/ml , ( ~1000 to 10 , 000 x IC50 ) so as to minimize the number of infection events by antibody sensitive variants , and enable rapid selection of the most resistant rVSV/SARS-CoV-2/GFP variants from the starting population . For plasma samples , the possibility existed that multiple different antibody specificities could be present , that might interfere with the outgrowth of rVSV/SARS-CoV-2/GFP variants that were resistant to the most prevalent or potent antibodies in the plasma . Therefore , in these selection experiments , viruses were incubated with a range of plasma dilutions ( see materials and methods ) . Neutralized viral populations were then applied to 293T/ACE2 ( B ) cells ( Schmidt et al . , 2020 ) , which support robust rVSV/SARS-CoV-2/GFP replication , and incubated for 48 hr . We used three potent human monoclonal antibodies C121 , C135 , and C144 ( Robbiani et al . , 2020 ) , that are candidates for clinical development ( Table 1 ) . In addition , we used four convalescent plasma samples , three of which were from the same donors from which C121 , C135 , and C144 , were obtained ( Robbiani et al . , 2020; Table 1 ) . Two of these plasmas ( COV-47 and COV-72 ) were potently neutralizing while the third ( COV-107 ) had low neutralizing activity . A fourth convalescent plasma sample ( COV-NY ) was potently neutralizing but did not have a corresponding monoclonal antibody ( Table 1 ) . Infection with rVSV/SARS-CoV-2/GFP in the presence of the monoclonal antibodies C121 or C144 reduced the number of infectious units from 106 to a few hundred , as estimated by the frequency of GFP-positive cells ( Figure 1B ) a reduction of > 1000 fold . C135 reduced infection by ~ 100 fold . Imaging of wells infected with rVSV/SARS-CoV-2/GFP in the presence of C121 or C144 revealed a small number of foci ( ~10 to 20/well ) , that suggested viral spread following initial infection ( Figure 1B ) . In the case of C135 , a greater number of GFP-positive cells were detected , obscuring the visualization of focal viral spread following initial infection . Aliquots of supernatants from these passage-1 ( p1 ) cultures were collected 48 hr after infection , diluted in the same concentrations of monoclonal antibodies that were initially employed , and used to infect fresh ( p2 ) cultures ( Figure 1A ) . For p2 cultures , almost all cells became infected within 48 hr , suggesting the possible outgrowth of monoclonal antibody escape variants that were present in the original viral populations . For selection in the presence of plasma , p1 supernatants were harvested at 48 hr after infection in the presence of the highest concentrations of plasma that permitted infection of reasonable numbers ( approximately 10% ) of cells . Then , p2 cultures were established using p1 supernatants , diluted in the same concentrations of plasma used in p1 . This approach led to clear ‘escape’ for the COV-NY plasma with prolific viral growth in p2 as evidenced by a large increase in the number of GFP-positive cells . For COV-47 , COV-72 , and CO107 , plasma clearly retained at least some inhibitory activity in p2 . Thereafter , p3 cultures and p4 cultures were established for COV-47 , COV-72 , and COV-107 plasmas at 5-fold higher concentrations of plasma than were used in p1 and p2 cultures ( Figure 1A ) . RNA was extracted from p2 supernatants ( monoclonal antibodies and COV-NY plasma ) as well as later passages for the COV-47 , COV-72 , and COV-107 plasma selections . Sequences encoding either the RBD or the complete S protein were amplified using PCR and analyzed by Sanger and/or Illumina sequencing . For all three monoclonal antibodies and two of the four plasmas , sequence analyses revealed clear evidence for selection , with similar or identical mutants emerging in the presence of monoclonal antibodies or plasma in both rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 cultures ( Figure 2A–D , Figure 2—figure supplement 1A , B , Figure 2—figure supplement 2A , B , Table 2 ) . In the case of C121 , mutations E484K and Q493K/R within the RBD were present at high frequencies in both p2 selected populations , with mutation at a proximal position ( F490L ) present in one p2 population ( Figure 2A , Table 2 ) . Viruses passaged in the presence of monoclonal antibody C144 also had mutations at positions E484 and Q493 , but not at F490 ( Figure 2C , Table 2 ) . In contrast , virus populations passaged in the presence of monoclonal antibody C135 lacked mutations at E484 or Q493 , and instead had mutations R346K/S/L and N440K at high frequency ( Figure 2C , Table 2 ) . Mutations at specific positions were enriched in viruses passaged in the presence of convalescent plasma , in two out of four cases ( Figure 2—figure supplement 1A , B , Figure 2—figure supplement 2A , B , Table 2 ) . Specifically , virus populations passaged in the presence of COV-NY plasma had mutations within RBD encoding sequence ( K444R/N/Q and V445E ) that were abundant at p2 ( Figure 2—figure supplement 2A , B , Table 2 ) . Conversely , mutations outside the RBD , specifically at N148S , K150R/E/T/Q and S151P in the N-terminal domain ( NTD ) were present at modest frequency in COV-47 p2 cultures and became more abundant at p3 and p4 ( Figure 2—figure supplement 1A , B , Table 2 ) . Replication in the presence of COV72 or COV107 plasma did not lead to the clear emergence of escape mutations , suggesting that the neutralization by these plasmas was not due to one dominant antibody specificity . In the case of COV107 , the failure of escape mutants to emerge may simply be due to the lack of potency of that plasma ( Table 1 ) . However , in the case of COV-72 , combinations of antibodies may be responsible for the potent neutralizing properties of the plasma in that case . Based on the aforementioned analyses , supernatants from C121 , C144 , and C135 and COV-NY plasma p2 cultures , or COV47 p4 cultures , contained mixtures of putative rVSV/SARS-CoV-2/GFP neutralization escape mutants . To isolate individual mutants , the supernatants were serially diluted and individual viral foci isolated by limiting dilution in 96-well plates . Numerous individual rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 derivatives were harvested from wells containing a single virus plaque , expanded on 293T/ACE2 ( B ) cells , then RNA was extracted and subjected to Sanger-sequencing ( Figure 3—figure supplement 1 ) . This process verified the purity of the individual rVSV/SARS-CoV-2/GFP variants and yielded a number of viral mutants for further analysis ( Figure 3—figure supplement 1 ) . These plaque-purified viral mutants all encoded single amino-acid substitutions in S-coding sequences that corresponded to variants found at varying frequencies ( determined by Illumina sequencing ) in the antibody-selected viral populations . Notably , each of the isolated rVSV/SARS-CoV-2/GFP mutants replicated with similar kinetics to the parental rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 viruses ( Figure 3A ) , suggesting that the mutations that emerged during replication in the presence of monoclonal antibodies or plasma did not confer a substantial loss of fitness , at least in the context of rVSV/SARS-CoV-2/GFP . Moreover , for mutants in RBD sequences that arose in the C121 , C135 , C144 , and COV-NY cultures , each of the viral mutants retained approximately equivalent sensitivity to neutralization by an ACE2-Fc fusion protein , suggesting little or no change in interaction with ACE2 ( Figure 3B ) . We next determined the sensitivity of the isolated RBD mutants to neutralization by the three monoclonal antibodies . The E484K and Q493R mutants that emerged during replication in the presence of C121 or C144 , both caused apparently complete , or near complete , resistance to both antibodies ( IC50 > 10 μg/ml , Figure 4A , B ) . However , both of these mutants retained full sensitivity ( IC50 < 10 ng/ml ) to C135 . Conversely , the R346S and N440K mutants that emerged during replication in the presence of C135 were resistant to C135 , but retained full sensitivity to both C121 and C144 ( Figure 4A , B ) . The K444N , K444T , V445G , V445E , and V445L mutants that arose during replication in the presence of COV-NY plasma conferred partial resistance to C135 , with IC50 values ranging from 25 to 700 ng/ml , but these mutants retained full sensitivity to both C121 and C144 ( Figure 4A , B ) . The spatial distribution of these resistance-conferring mutations on the SARS-CoV-2 S RBD surface suggested the existence of both distinct and partly overlapping neutralizing epitopes on the RBD ( Figure 4C ) . The C121 and C144 neutralizing epitopes appear to be similar , and include E484 and Q493 , while a clearly distinct conformational epitope seems to be recognized by C135 , that includes R346 and N440 residues . Antibodies that constitute at least part of the neutralizing activity evident in COV-NY plasma appear to recognize an epitope that includes and K444 and V445 . The ability of mutations at these residues to confer partial resistance to C135 is consistent with their spatial proximity to the C135 conformational epitope ( Figure 4C ) . To test whether neutralization escape mutations conferred loss of binding to the monoclonal antibodies , we expressed conformationally prefusion-stabilized S-trimers ( Hsieh et al . , 2020 ) , appended at their C-termini with NanoLuc luciferase ( Figure 5A ) . The S-trimers were incubated in solution with the monoclonal antibodies , complexes were captured using protein G magnetic beads , and the amount of S-trimer captured was measured using NanoLuc luciferase assays ( Figure 5A ) . As expected , C121 , C135 , and C144 monoclonal antibodies all bound the WT S-trimer ( Figure 5B ) . The E484K and Q493R trimers exhibited complete , or near complete loss of binding to C121 and C144 antibodies but retained WT levels of binding to C135 ( Figure 5B ) . Conversely , the R346S and N440K mutants exhibited complete loss of binding to C135 , but retained WT levels of binding to C121 and C144 . The K444N and V445E mutants retained near WT levels of binding to all three antibodies , despite exhibiting partial resistance to C135 ( Figure 5A , B ) . Presumably the loss of affinity of these mutants for C135 was sufficient to impart partial neutralization resistance but insufficient to abolish binding in the solution binding assay . Analysis of mutants that were isolated from the virus population that emerged during rVSV/SARS-CoV-2/GFP replication in the presence of COV-47 plasma ( specifically N148S , K150R , K150E , S151P ) revealed that these mutants exhibited specific resistance to COV-47 plasma . Indeed , the COV-47 plasma NT50 for these mutants was reduced by 8- to 10-fold ( Figure 6A ) . This finding indicates that the antibody or antibodies responsible for majority of neutralizing activity in COV-47 plasma target an NTD epitope that includes amino acids 148–151 , even though the highly potent monoclonal antibody ( C144 ) isolated from COV-47 targets the RBD . Mutants in the 148–151 NTD epitope exhibited marginal reductions in sensitivity to other plasmas ( Figure 6A ) , indicating that different epitopes are primarily targeted by plasmas from the other donors . The viral population that emerged during replication in COV-NY plasma yielded mutants K444N or T and V445G , E or L . Each of these mutations conferred substantial resistance to neutralization by COV-NY plasma , with ~ 10 to 30-fold reduction in NT50 ( Figure 6B ) . Thus , the dominant neutralizing activity in COV-NY plasma is represented by an antibody or antibodies recognizing an RBD epitope that includes K444 and V445 . As was the case with COV-47 resistant mutants , viruses encoding the mutations conferring resistance to COV-NY plasma retained almost full sensitivity to neutralization by other plasmas ( Figure 6B ) . Interestingly , the mutations that conferred complete or near complete resistance to the potent RBD-specific monoclonal antibodies C144 , C135 , and C121 conferred little or no resistance to neutralization by plasma from the same individual , or other individuals ( Figure 6C ) . These RBD-specific antibodies represent the most potent monoclonal antibodies isolated from COV-47 , COV-72 , and COV-107 , respectively , but the retention of plasma sensitivity by the monoclonal antibody-resistant mutants suggests that these antibodies contribute little to the overall neutralization activity of plasma from the same individual . This finding is consistent with the observation that memory B cells producing these antibodies are rare ( Robbiani et al . , 2020 ) , and with the results of the selection experiments in which rVSV/SARS-CoV-2/GFP replication in the presence of COV-47 , COV-72 , and COV-107 plasma did not enrich for mutations that correspond to the neutralization epitopes targeted by the monoclonal antibodies obtained from these individuals ( Figure 2—figure supplement 1A , B , Figure 2—figure supplement 2A , B . Table 2 ) . Overall , analysis of even this limited set of monoclonal antibodies and plasmas shows that potent neutralization can be conferred by antibodies that target diverse SARS-CoV-2 epitopes . Moreover , the most potently neutralizing antibodies generated in a given COVID19 convalescent individual may contribute in only a minor way to the overall neutralizing antibody response in that same individual ( see discussion ) . The aforementioned neutralizing antibody escape mutations were artificially generated during in vitro replication of a recombinant virus . However , as monoclonal antibodies are developed for therapeutic and prophylactic applications , and vaccine candidates are deployed , and the possibility of SARS-CoV-2 reinfection becomes greater , it is important both to understand pathways of antibody resistance and to monitor the prevalence of resistance-conferring mutations in naturally circulating SARS-CoV-2 populations . To survey the natural occurrence of mutations that might confer resistance to the monoclonal and plasma antibodies used in our experiments we used the GISAID ( Elbe et al . , 2017 ) and CoV-Glue ( Singer et al . , 2020 ) SARS-CoV-2 databases . Among the 55 , 189 SARS-CoV-2 sequences in the CoV2-Glue database at the time of writing , 2175 different non-synonymous mutations were present in natural populations of SARS-CoV-2 S protein sequences . Consistent with the finding that none of the mutations that arose in our selection experiments gave an obvious fitness deficit ( in the context of rVSV/SARS-CoV-2/GFP ) , most were also present in natural viral populations . For phenotypic analysis of naturally occurring SARS-CoV-2 S mutations , we focused on the ACE2 interface of the RBD , as it is the target of most therapeutic antibodies entering clinical development ( Robbiani et al . , 2020; Hansen et al . , 2020; Baum et al . , 2020 ) , and is also the target of at least some antibodies present in convalescent plasmas . In addition to the mutations that arose in our antibody selection experiments , inspection of circulating RBD sequences revealed numerous naturally occurring mutations in the vicinity of the ACE2 binding site and the epitopes targeted by the antibodies ( https://www . gisaid . org , http://cov-glue . cvr . gla . ac . uk ) ( Figure 7A–D ) . We tested nearly all of the mutations that are present in the GISAID database as of June 2020 , in the proximity of the ACE2 binding site and neutralizing epitopes , for their ability to confer resistance to the monoclonal antibodies , using an HIV-1-based pseudotyped virus-based assay ( Figure 7A–C ) . Consistent with , and extending our findings with rVSV/SARS-CoV-2/GFP , naturally occurring mutations at positions E484 , F490 , Q493 , and S494 conferred complete or partial resistance to C121 and C144 ( Figure 7A , C ) . While there was substantial overlap in the mutations that caused resistance to C121 and C144 , there were also clear differences in the degree to which certain mutations ( e . g . G446 , L455R/I/F , F490S/L ) affected sensitivity to the two antibodies . Naturally occurring mutations that conferred complete or partial resistance to C135 were at positions R346 , N439 , N440 , K444 , V445 and G446 . In contrast to the C121/C144 epitope , these amino acids are peripheral to the ACE2 binding site on the RBD ( Figure 7D ) . Indeed , in experiments where the binding of a conformationally stabilized trimeric S-NanoLuc fusion protein to 293T/ACE2cl . 22 cells was measured , preincubation of S-NanoLuc with a molar excess of C121 or C144 completely blocked binding ( Figure 7E ) . Conversely , preincubation with C135 only partly blocked binding to 293T/ACE2cl . 22 cells , consistent with the finding that the C135 conformational epitope does not overlap the ACE2 binding site ( Figure 7D ) . C135 might inhibit S-ACE-2 binding by steric interference with access to the ACE two binding site . These results are also consistent with experiments which indicated that C135 does not compete with C121 and C144 for binding to the RBD ( Robbiani et al . , 2020 ) . All of the mutations that were selected in our rVSV/SARS-CoV-2/GFP antibody selection experiments as well as other mutations that confer resistance to C121 , C144 , or C135 are found in naturally circulating SARS-CoV-2 populations at very low frequencies ( Figure 8 ) . With one exception ( N439K ) that is circulating nearly exclusively in Scotland and is present in ~ 1% of COV-Glue database sequences , ( and whose frequency may be overestimated due to regional oversampling ) all antibody resistance mutations uncovered herein are present in global SARS-CoV-2 at frequencies of < 1 in 1000 sequences ( Figure 8 ) . The frequency with which the resistance mutations are present in naturally occurring SARS-CoV-2 sequences appeared rather typical compared to other S mutations , with the caveat that sampling of global SARS-CoV-2 is nonrandom . Therefore , these observations do not provide evidence that the neutralizing activities exhibited by the monoclonal antibodies or plasma samples used herein have driven strong selection of naturally circulating SARS-CoV-2 sequences thus far ( Figure 8 ) . The ability of SARS-CoV-2 monoclonal antibodies and plasma to select variants that are apparently fit and that naturally occur at low frequencies in circulating viral populations suggests that therapeutic use of single antibodies might select for escape mutants . To mitigate against the emergence or selection of escape mutations during therapy , or during population-based prophylaxis , we tested whether combinations of monoclonal antibodies could suppress the emergence of resistant variants during in vitro selection experiments . Specifically , we repeated antibody selection experiments in which rVSV/SARS-CoV-2/GFP populations containing 106 infectious virions were incubated with 10 μg/ml of each individual monoclonal antibody , or mixtures containing 5 μg/ml of each of two antibodies ( Figure 9 ) . C121 and C144 target largely overlapping epitopes , and mutations conferring resistance to one of these antibodies generally conferred resistance to the other ( Figure 7A–D ) . Therefore , we used mixtures of antibodies targeting clearly distinct epitopes ( C121+C135 and C144+C135 ) . As previously , replication of rVSV/SARS-CoV-2/GFP in the presence of a single monoclonal antibody enabled the formation of infected foci in p1 cultures ( Figure 9A–C ) , that rapidly expanded and enabled the emergence of apparently resistant virus populations . Indeed , rVSV/SARS-CoV-2/GFP yields from p2 cultures established with one antibody ( C121 , C135 or C144 ) were indistinguishable from those established with no antibody ( Figure 9D ) . Conversely , rVSV/SARS-CoV-2/GFP replication in the presence of mixtures of C121+C135 or C144+C135 led to sparse infection of individual cells in p1 cultures , but there was little or no formation of foci that would suggest propagation of infection from these infected cells ( Figure 9A–C ) , Therefore , it is likely that infected cells arose from rare , non-neutralized , virions that retained sensitivity to at least one of the antibodies in mixture . Consequently , viral spread was apparently completely suppressed and no replication-competent rVSV/SARS-CoV-2/GFP was detected in p2 cultures established with mixtures of the two antibodies ( Figure 9D ) . The degree to which resistance will impact effectiveness of antibodies in SARS-CoV-2 therapeutic and vaccine settings is currently unclear ( Baum et al . , 2020 ) . Notably , the inter-individual variation in SARS-CoV-2 sequences is low compared to many other RNA viruses ( van Dorp et al . , 2020; Rambaut et al . , 2020; Dearlove et al . , 2020; Rausch et al . , 2020 ) , in part because coronaviruses encode a 3’−5’ exonuclease activity . The exonuclease activity provides a proofreading function that enhances replication fidelity and limits viral sequence diversification ( Denison et al . , 2011 ) . However , replication fidelity is but one of several variables that affect viral population diversity ( Duffy et al . , 2008; Moya et al . , 2000 ) . One determinant of total viral diversity is population size . Many millions of individuals have been infected by SARS-CoV-2 , and a single swab from an infected individual can contain in excess of 109 copies of viral RNA ( Wölfel et al . , 2020 ) . It follows that SARS-CoV-2 genomes encoding every possible single amino-acid substitution are present in the global population , and perhaps in a significant fraction of individual COVID19 patients . Thus , the frequency with which particular variants occur in the global SARS-CoV-2 population is strongly influenced by the frequency with which negative and positive selection pressures that favor their propagation are encountered , as well as founder effects at the individual patient and population levels ( Korber et al . , 2020 ) . Fitness effects of mutations will obviously vary and will suppress the prevalence of deleterious mutations ( Dolan et al . , 2018; Andino and Domingo , 2015 ) . However , otherwise neutral , or even modestly deleterious , mutations will rise in prevalence if they confer escape from selective pressures , such as immune responses . The prevalence of neutralizing antibody escape mutations will also be strongly influenced by the frequency with which SARS-CoV-2 encounters neutralizing antibodies . Peak viral burden in swabs and sputum , which likely corresponds to peak infectiousness and frequency of transmission events , appears to approximately correspond with the onset of symptoms , and clearly occurs before seroconversion ( Wölfel et al . , 2020 ) . Thus , it is quite plausible that most transmission events involve virus populations that are yet to experience antibody-imposed selective pressure in the transmitting individual . Such a scenario would reduce the occurrence of antibody escape mutations in natural viral populations . It will be interesting to determine whether viral sequences obtained late in infection are more diverse or have evidence of immunological escape mutations . There are situations that are anticipated to increase the frequency of encounters between SARS-CoV-2 and antibodies that could impact the emergence of antibody resistance . Millions of individuals have already been infected with SARS-CoV-2 and among them , neutralizing antibody titers are extremely variable ( Robbiani et al . , 2020; Wu et al . , 2020b; Luchsinger et al . , 2020 ) . Those with weak immune responses or waning immunity could become re-infected , and if so , that encounters between SARS-CoV-2 and pre-existing but incompletely protective neutralizing antibodies might drive the selection of escape variants ( Kk et al . , 2020a; Van Elslande et al . , 2020; Larson et al . , 2020; Kk et al . , 2020b ) . In a similar manner , poorly immunogenic vaccine candidates , convalescent plasma therapy , and suboptimal monoclonal antibody treatment , particularly monotherapy ( Baum et al . , 2020 ) , could create conditions to drive the acquisition of resistance to commonly occurring antibodies in circulating virus populations . The extent to which SARS-CoV-2 evasion of individual antibody responses would have pervasive effects on the efficacy of vaccines and monoclonal antibody treatment/therapy will also be influenced by the diversity of neutralizing antibody responses within and between individuals . Analysis of potent neutralizing antibodies cloned by several groups indicates that potent neutralizing antibodies are commonly elicited , and very similar antibodies , such as those containing IGHV3-53 and IGHV3-66 can be found in different individuals ( Robbiani et al . , 2020; Barnes et al . , 2020; Yuan et al . , 2020 ) . These findings imply a degree of homogeneity the among neutralizing antibodies that are generated in different individuals . Nevertheless , each of the four convalescent plasma tested herein had distinct neutralizing characteristics . In two of the four plasma tested , selection experiments suggested that a dominant antibody specificity was responsible for a significant fraction of the neutralizing capacity of the plasma . However , the failure of single amino-acid substitutions to confer complete resistance to any plasma strongly suggests the existence of multiple neutralizing specificities in each donor . Indeed , in one example ( COV47 ) , viral mutants that were completely resistant to a potent monoclonal antibody from that donor ( C144 ) , retained near complete sensitivity to plasma from that same individual , Thus , in that individual other antibodies in the plasma , not the most potent monoclonal antibody , must dominate the neutralizing activity of the plasma . That COV47 plasma selected mutations at a different site in S ( NTD ) to that selected by C144 ( RBD ) , is orthogonal supportive evidence that the C144 monoclonal antibody does not constitute the major neutralizing activity in the plasma of COV47 . The techniques described herein could be adapted to broadly survey the diversity of SARS-CoV-2 neutralizing specificities in many plasma samples following natural infection or vaccination , enabling a more complete picture of the diversity of SARS-CoV-2 neutralizing antibody responses to be developed . Indeed , the approach described herein can be used to map epitopes of potent neutralizing antibodies rapidly and precisely . It has an advantage over other epitope mapping approaches ( such as array-based oligo-peptide scanning or random site-directed mutagenesis ) ( Greaney et al . , 2020 ) , in that selective pressure acts solely on the naturally formed , fusion-competent viral spike . While mutations outside the antibody binding sites might lead to resistance , the functional requirement will prohibit mutations that simply disrupt the native conformation . Indeed , we found that the neutralizing antibody escape mutations described herein did not detectably alter rVSV/SARS-CoV-2/GFP replication , did not affect ACE2-Fc sensitivity and were found in natural populations at unexceptional frequencies , consistent with the notion that they do not have large effects on fitness in the absence of neutralizing antibodies . That said , the selection/viral evolution scheme described herein approximates to but does not precisely recapitulate the evolutionary dynamics that would play out in natural SARS-CoV-2 infection . Differences in viral populations sizes , replication fidelity and fitness effects of mutations in SARS-CoV-2 versus rVSV/SARS-CoV-2/GFP , and the complexity and changing nature of evolving antibody responses could all affect the nature of , and response to , selection pressures . For example , one caveat is the utilization of plasma for selection experiments . Immunoglobulin subtypes ( e . g . IgA versus IgG ) are differentially represented in plasma versus the respiratory tract , and the concentrations of each immunoglobulin subtype or specificity precisely at the sites of SARS-CoV-2 replication is unknown . However , it is known that IgG that dominates plasma immunoglobulins is also present in lung secretions , albeit at lower levels than IgA ( Burnett , 1986 ) . Moreover , our recent work has indicated that at least some of the neutralizing activity in plasma is contributed by IgA and several of the antibody lineages that we have cloned from SARS-CoV-2 convalescents are class switched to both IgA and IgG ( Wang et al . , 2020 ) . Overall , it is likely that the antibody specificities present in the respiratory tract broadly reflect , but perhaps do not precisely recapitulate , those present in the plasma used for the selection experiments described herein . Human monoclonal antibodies targeting both the NTD and RBD of SARS-CoV-2 have been isolated , with those targeting RBD being especially potent . As these antibodies are used clinically ( Hansen et al . , 2020; Baum et al . , 2020 ) , in therapeutic and prophylactic modes , it will be important to identify resistance mutations and monitor their prevalence in a way that is analogous to antiviral and antibiotic resistance monitoring in other infectious diseases . Moreover , as is shown herein , the selection of antibody mixtures with non-overlapping escape mutations should reduce the emergence of resistance and prolong the utility of antibody therapies in SARS-CoV-2 infection . A replication-competent rVSV/SARS-CoV-2/GFP chimeric virus clone , encoding the SARS-CoV-2 spike protein lacking the C-terminal 18 codons in place of G , as well as GFP immediately upstream of the L ( polymerase ) has been previously described ( Schmidt et al . , 2020 ) . The pHIV-1NLGagPol and pCCNG/nLuc constructs that were used to generate SARS-CoV-2 pseudotyped particles have been previously described ( Schmidt et al . , 2020 ) . The pSARS-CoV-2 protein expression plasmid containing a C-terminally truncated SARS-CoV-2 S protein ( pSARS-CoV-2Δ19 ) containing a synthetic human-codon-optimized cDNA ( Geneart ) has been previously described ( Schmidt et al . , 2020 ) and was engineered to include BamHI , MfeI , BlpI and AgeI restriction enzyme sites flanking sequences encoding the RBD . Gibson assembly was used to introduce mutant RBD sequences into this plasmid , that were generated synthetically ( g/eBlocks IDT ) or by overlap extension PCR with primers that incorporated the relevant nucleotide substitutions . HEK-293T cells and derivatives were cultured in Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) at 37°C and 5% CO2 . All cell lines have been tested negative for contamination with mycoplasma . Derivatives expressing ACE2 were generated by transducing 293T cells with CSIB ( ACE2 ) vector and the uncloned bulk population 293T/ACE2 ( B ) or a single-cell clone 293T/ACE2 . cl22 ( Schmidt et al . , 2020 ) were used . The generation of infectious rVSV/SARS-CoV-2/GFP chimeric viruses stocks has been previously described ( Schmidt et al . , 2020 ) . Two plaque-purified variants designated rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 that encode F157S/R685M ( 1D7 ) and D215G/R683G ( 2E1 ) substitutions were used in these studies . The rVSV/SARS-CoV-2/GFP chimeric virus was used under enhanced BSL-2 conditions . i . e in a BLS-2 laboratory with BSL-3 like precautions . The HIV-1/NanoLuc2AEGFP-SARS-CoV-2 pseudotyped virions were generated as previously described ( Schmidt et al . , 2020 ) . Briefly , 293T cells were transfected with pHIVNLGagPol , pCCNanoLuc2AEGFP and a WT or mutant SARS-CoV-2 expression plasmid ( pSARS-CoV-2Δ19 ) using polyethyleneimine . At 48 hr after transfection , the supernatant was harvested , clarified , filtered , aliquoted and stored at −80°C . To measure the infectivity of pseudotyped or chimeric viral particles , viral stocks were serially diluted and 100 µl of each dilution added to 293T/ACE2cl . 22 target cells plated at 1 × 104 cells/well in 100 µl medium in 96-well plates the previous day . Cells were then cultured for 48 hr ( HIV-1 pseudotyped viruses ) or 16 hr ( replication-competent rVSV/SARS-CoV-2/GFP ) , unless otherwise indicated , and then photographed or harvested for NanoLuc luciferase or flow cytometry assays . For selection of viruses resistant to plasma or monoclonal antibodies , rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 populations containing 106 infectious particles were used . To generated the viral populations for selection experiments , , rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 were passaged generate diversity , incubated with dilutions of monoclonal antibodies ( 10 μg/ml , 5 μg/ml ) or COVID19 plasma ( 1:50 , 1:250 , 1:500 ) for 1 hr at 37 °C . Then , the virus-antibody mixtures were incubated with 2 × 105 293T/ACE2 ( B ) cells in 12-well plates . Two days later , the cells were imaged and supernatant from the wells containing the highest concentration of plasma or monoclonal antibodies that showed evidence of viral replication ( GFP-positive foci ) or large numbers of GFP-positive cells was harvested . A 100 μl of the cleared supernatant was incubated with the same dilution of plasma or monoclonal antibody and then used to infect 2 × 105 293T/ACE2 ( B ) cells in 12-well plates , as before . rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 were passaged in the presence of C121 or C144 two times before complete escape was apparent . rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 were passaged with C135 or plasma samples up to five times . To isolate individual mutant viruses , selected rVSV/SARS-CoV-2/GFP1D7 and rVSV/SARS-CoV-2/GFP2E1 populations were serially diluted in medium without antibodies and individual viral variants isolated by visualizing single GFP-positive plaques at limiting dilutions in 96-well plates containing 1 × 104 293T/ACE2 ( B ) cells . These plaque-purified viruses were expanded , and further characterized using sequencing , spreading replication and neutralization assays . For the identification of putative antibody resistance mutations , RNA was isolated from aliquots of supernatant containing selected viral populations or individual plaque-purified variants using Trizol-LS . The purified RNA was subjected to reverse transcription using random hexamer primers and Superscript III reverse transcriptase ( Thermo Fisher Scientific , US ) . The cDNA was amplified using Phusion ( NEB , US ) and primers flanking RBD encoding sequences . Alternatively , a fragment including the entire S-encoding sequence was amplified using primers targeting VSV-M and VSV-L . The PCR products were gel-purified and sequenced either using Sanger-sequencing or NGS as previously described ( Gaebler et al . , 2019 ) . Briefly , 1 µl of diluted DNA was used for the tagmentation reactions with 0 . 25 µl Nextera TDE1 Tagment DNA enzyme ( catalog no . 15027865 ) , and 1 . 25 µl TD Tagment DNA buffer ( catalog no . 15027866; Illumina ) . Subsequently , the DNA was ligated to unique i5/i7 barcoded primer combinations using the Illumina Nextera XT Index Kit v2 and KAPA HiFi HotStart ReadyMix ( 2X; KAPA Biosystems ) and purified using AmPure Beads XP ( Agencourt ) , after which the samples were pooled into one library and subjected to paired-end sequencing using Illumina MiSeq Nano 300 V2 cycle kits ( Illumina ) at a concentration of 12pM . For analysis of NGS data , the raw paired-end reads were pre-processed to remove adapter sequences and trim low-quality reads ( Phred quality score < 20 ) using BBDuk . Filtered reads were mapped to the codon-optimized SARS-CoV-2 S sequence in rVSV/SARS-CoV-2/GFP using Geneious Prime ( Version 2020 . 1 . 2 ) . Mutations were annotated using Geneious Prime , with a P-value cutoff of 10−6 . Information regarding RBD-specific variant frequencies , their corresponding P-values , and read depth were compiled using the Python programming language ( version 3 . 7 ) running pandas ( 1 . 0 . 5 ) , numpy ( 1 . 18 . 5 ) , and matplotlib ( 3 . 2 . 2 ) . To measure neutralizing antibody activity in plasma , serial dilutions of plasma beginning with a 1:12 . 5 or a 1:100 ( for plasma COV-NY ) initial dilution were five-fold serially diluted in 96-well plates over six or eight dilutions . For monoclonal antibodies , or an ACE2-IgG1Fc fusion protein the initial dilution started at 40 µg/ml . Thereafter , approximately 5 × 104 infectious units of rVSV/SARS-CoV-2/GFP or 5 × 103 infectious units of HIV/CCNG/nLuc/SARS-CoV-2 were mixed with the plasma or mAb at a 1:1 ratio and incubated for 1 hr at 37°C in a 96-well plate . The mixture was then added to 293T/ACE2cl . 22 target cells plated at 1 × 104 cells/well in 100 µl medium in 96-well plates the previous day . Thus , the final starting dilutions were 1:50 or 1:400 ( for COV-NY ) for plasma and 10 µg/ml for monoclonal antibodies . Cells were then cultured for 16 hr ( for rVSV/SARS-CoV-2/GFP ) or 48 hr ( for HIV/CCNG/nLuc/SARS-CoV-2 ) . Thereafter , cells were harvested for flow cytometry or NanoLuc luciferase assays . A conformationally stabilized ( 6P ) version of the SARS-CoV-2 S protein ( Hsieh et al . , 2020 ) , appended at its C-terminus with a trimerization domain , a GGSGGn spacer sequence , NanoLuc luciferase , Strep-tag , HRV 3C protease cleavage site and 8XHis ( S-6P-NanoLuc ) was expressed and purified from the supernatant of 293T Expi cells . Mutants thereof were also expressed and purifies following substitution of sequences encoding the RBD that originated from the unmodified S-expression plasmids . For antibody-binding assays , 20 , 40 , or 80 ng S-6P-NanoLuc ( or mutants thereof ) were mixed with 100 ng of antibodies , C121 , C135 , or C144 , diluted in LI-COR Intercept blocking buffer , in a total volume of 60 μl/well in 96-well plate . After a 30 min incubation , 10 µl protein G magnetic beads was added to each well and incubated for 1 . 5 hr . The beads were then washed three times and incubated with 30 µl lysis buffer ( Promega ) . Then 15 μl of the lysate was used to measure bound NanoLuc activity . For ACE2-binding inhibition assays , 20 ng of S-6P-NanoLuc was mixed with 100 ng of antibodies , C121 , C135 , or C144 , diluted in 3% goat serum/PBS , in a total volume of 50 μl . After 30 min incubation , the mixture was incubated with 1 × 105 293 T cells , or 293T/ACE2cl . 22 cells for 2 hr at 4°C . The cells were then washed three times and lysed with 30 μl lysis buffer and 15 μl of the lysate was used to measure bound NanoLuc activity . For the NanoLuc luciferase assays , cells were washed gently , twice with PBS and lysed in Lucifersase Cell culture Lysis reagent ( Promega ) . NanoLuc luciferase activity in the lysates was measured using the Nano-Glo Luciferase Assay System ( Promega ) and a Modulus II Microplate Multimode reader ( Turner BioSystem ) or a Glowmax Navigator luminometer ( Promega ) , as described previously ( Schmidt et al . , 2020 ) . To record GFP+ cells , 12-well plates were photographed using an EVOS M7000 automated microscope . For flow cytometry , cells were trypsinized , fixed and enumerated using an Attune NxT flow cytometer . The half maximal inhibitory concentrations for plasma ( NT50 ) , and monoclonal antibodies ( IC50 ) was calculated using 4-parameter nonlinear regression curve fit to raw or normalized infectivity data ( GraphPad Prism ) . Top values were unconstrained , the bottom values were set to zero . The human plasma samples COV-47 , COV-72 and COV-107 and monoclonal antibodies C144 , C135 and C121 used in this study were previously reported ( Robbiani et al . , 2020 ) . The human plasma sample COV-NY was obtained from the New York Blood Center ( Luchsinger et al . , 2020 ) . All plasma samples were obtained under protocols approved by Institutional Review Boards at both institutions .	The new coronavirus , SARS-CoV-2 , which causes the disease COVID-19 , has had a serious worldwide impact on human health . The virus was virtually unknown at the beginning of 2020 . Since then , intense research efforts have resulted in sequencing the coronavirus genome , identifying the structures of its proteins , and creating a wide range of tools to search for effective vaccines and therapies . Antibodies , which are immune molecules produced by the body that target specific segments of viral proteins can neutralize virus particles and trigger the immune system to kill cells infected with the virus . Several technologies are currently under development to exploit antibodies , including vaccines , blood plasma from patients who were previously infected , manufactured antibodies and more . The spike proteins on the surface of SARS-CoV-2 are considered to be prime antibody targets as they are accessible and have an essential role in allowing the virus to attach to and infect host cells . Antibodies bind to spike proteins and some can block the virus’ ability to infect new cells . But some viruses , such as HIV and influenza , are able to mutate their equivalent of the spike protein to evade antibodies . It is unknown whether SARS-CoV-2 is able to efficiently evolve to evade antibodies in the same way . Weisblum , Schmidt et al . addressed this question using an artificial system that mimics natural infection in human populations . Human cells grown in the laboratory were infected with a hybrid virus created by modifying an innocuous animal virus to contain the SARS-CoV-2 spike protein , and treated with either manufactured antibodies or antibodies present in the blood of recovered COVID-19 patients . In this situation , only viruses that had mutated in a way that allowed them to escape the antibodies were able to survive . Several of the virus mutants that emerged had evolved spike proteins in which the segments targeted by the antibodies had changed , allowing these mutant viruses to remain undetected . An analysis of more than 50 , 000 real-life SARS-CoV-2 genomes isolated from patient samples further showed that most of these virus mutations were already circulating , albeit at very low levels in the infected human populations . These results show that SARS-CoV-2 can mutate its spike proteins to evade antibodies , and that these mutations are already present in some virus mutants circulating in the human population . This suggests that any vaccines that are deployed on a large scale should be designed to activate the strongest possible immune response against more than one target region on the spike protein . Additionally , antibody-based therapies that use two antibodies in combination should prevent the rise of viruses that are resistant to the antibodies and maintain the long-term effectiveness of vaccines and therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2020	Escape from neutralizing antibodies by SARS-CoV-2 spike protein variants
Domesticated animals experienced profound changes in diet , environment , and social interactions that likely shaped their gut microbiota and were potentially analogous to ecological changes experienced by humans during industrialization . Comparing the gut microbiota of wild and domesticated mammals plus chimpanzees and humans , we found a strong signal of domestication in overall gut microbial community composition and similar changes in composition with domestication and industrialization . Reciprocal diet switches within mouse and canid dyads demonstrated the critical role of diet in shaping the domesticated gut microbiota . Notably , we succeeded in recovering wild-like microbiota in domesticated mice through experimental colonization . Although fundamentally different processes , we conclude that domestication and industrialization have impacted the gut microbiota in related ways , likely through shared ecological change . Our findings highlight the utility , and limitations , of domesticated animal models for human research and the importance of studying wild animals and non-industrialized humans for interrogating signals of host–microbial coevolution . Industrialized , agrarian , and foraging human populations differ along numerous ecological dimensions , including diet , physical activity , the size and nature of social networks , pathogen exposure , types and intensities of medical intervention , and reproductive patterns . Such changes have resulted in large shifts in the gut microbiota in industrialized populations relative to non-industrialized populations or closely related primates ( De Filippo et al . , 2010; Moeller et al . , 2014; Moeller , 2017; Smits et al . , 2017 ) , including reductions in alpha-diversity and changes in composition that have been implicated in the rise of various metabolic and immunological diseases ( Ley et al . , 2006; Cox et al . , 2014; Kamada et al . , 2013 ) . Many aspects of ecology that now differ between industrialized and non-industrialized human populations were similarly altered during the process of animal domestication ( Zeder , 2012 ) . For example , domestic animals often consume less diverse , more easily digestible diets than their wild relatives , expend less energy to achieve adequate ( or excess ) caloric intake , live in comparatively static and high-density groups , and can be subject to modern medical interventions including antibiotic treatment ( McClure , 2013 ) . Although industrialization and domestication are fundamentally different processes , the ecological parallels between human industrialization and animal domestication suggest that the gut microbiota of diverse domesticated animals may differ in consistent ways from those of their wild progenitors , and further , that their differences may resemble those observed between industrialized and non-industrialized human populations . Many of the altered ecological features experienced by industrialized humans and domesticated animals have been independently observed to impact the gut microbiota , including diet ( David et al . , 2014; Carmody et al . , 2015 ) , physical activity ( Allen et al . , 2018; Lamoureux et al . , 2017 ) , the size and nature of social networks ( Dill-McFarland et al . , 2019; Antwis et al . , 2018 ) , antibiotic use ( Bokulich et al . , 2016; Cho et al . , 2012 ) , and changes in birthing and lactation practices ( Bokulich et al . , 2016; Li et al . , 2018 ) . The effects of these features on gut microbiota composition are often found to match or exceed the effects of genetic variation ( Carmody et al . , 2015; Rothschild et al . , 2018 ) , which is also routinely modified by domestication . As such , ecological shifts under domestication might be expected to lead to gut microbial differentiation between domesticated animals and their wild counterparts . To this end , wild mice have been shown to differ from laboratory mice in gut microbial composition ( Kreisinger et al . , 2014; Rosshart et al . , 2017 ) . Similarly , a comparison of domesticated horses and wild Przewalski’s horses in adjacent Mongolian grasslands found that the wild animals harbored compositionally distinct , and overall more diverse , gut microbial communities ( Metcalf et al . , 2017 ) . However , to date , no general survey has been conducted to characterize the global effects of domestication on the gut microbiota . Apart from the pressures of ecological change that domestic animals experience in human environments , animal domestication has also entailed strong artificial selection for phenotypes desirable to humans , such as rapid growth and docility in agricultural animals , reliable reproduction and stress resistance in laboratory animals , and unique physical and/or behavioral attributes in companion animals . Although targeted phenotypes differ based on the species under domestication , all domesticated mammals share the legacy of having been intentionally or indirectly selected for tameness ( Wilkins et al . , 2014 ) . This selection has been argued to have resulted in convergent morphological and physiological features across domesticated mammals that are collectively referred to as ‘domestication syndrome’ – including , for instance , reductions in brain size and tooth size , depigmentation , altered production of hormones and neurotransmitters , and retention of juvenile behaviors into adulthood – with the pleiotropic nature of these effects thought to be mediated by changes in neural crest cells ( Wilkins et al . , 2014 ) . Therefore , to the extent that gut microbiota is dependent on host biology , we might additionally expect domestication to have shaped the gut microbiome in similar , potentially convergent , ways across diverse mammalian lineages . Such microbiota-structuring contributions ascribable to evolutionary rather than ecological forces have the potential to be much greater in and/or unique to domesticated animals relative to industrialized human populations since the process of domestication has been advancing for much longer than industrialization . Here , we assess the effects of domestication on the mammalian gut microbiota , perform controlled dietary experiments that attempt to distinguish between the relative roles of ecology and genetics in driving these patterns , and compare the effects of domestication to those of human industrialization . While we focus primarily on the impacts of domestication on the mammalian gut microbiota , we include analyses of industrialized and non-industrialized human populations because much is known about the effects of industrialization on the gut microbiota and as such it can serve as a benchmark ecological process for domestication . In addition , to explore the extent to which deeper evolutionary history affects these patterns , we also compare humans to chimpanzees ( Pan troglodytes ) , one of our two closest living relatives and arguably the better referential model for the last common ancestor between Pan and Homo ( Muller et al . , 2017 ) . Early Homo sapiens is thought to have undergone a form of self-domestication as a result of selection against aggression ( Wrangham , 2018; Theofanopoulou et al . , 2017 ) , suggesting that there could likewise be parallels between the gut microbial signatures of animal domestication and Pan–Homo speciation . We predict that ( i ) gut microbial communities will differ between domesticated animals and their wild counterparts , ( ii ) gut microbial communities of diverse domesticated animals may exhibit convergent characteristics in a microbial counterpart to the physiological domestication syndrome ( Wilkins et al . , 2014 ) , and ( iii ) gut microbial changes observed with domestication may parallel contrasts observed between chimpanzees and humans . In addition , to the extent that domestication effects are driven by ecology rather than host phylogenetic distance , we should expect ( iv ) experimental manipulation of ecology to overcome differences in the gut microbiota between closely related hosts , and ( v ) the gut microbiota of domesticated animals will share more features with industrialized human populations than with non-industrialized human populations . Identifying the factors shaping the gut microbiota of domesticated animals will provide insights into the ecology of host-associated microbial communities and their impact on health . Domesticated animals serve as reservoirs for zoonotic diseases ( Morand et al . , 2014; Wolfe et al . , 2007; Cleaveland et al . , 2001; Han et al . , 2016 ) and carriers of antibiotic-resistant bacteria ( Sayah et al . , 2005; EFFORT Group et al . , 2018 ) . Furthermore , the ecological impacts of domestication on the gut microbiota could conceivably contribute to the unique health problems experienced by captive ( Hosey et al . , 2009 ) and domesticated animals ( Timoney et al . , 1988 ) . Differences between domesticated and wild animal microbiota may also manifest in poor translatability between laboratory studies and the real world ( Leung et al . , 2018; Beura et al . , 2016 ) . Finally , the convergent nature of many ecological shifts experienced by domesticated animals and industrialized human populations suggests that domesticated animals may provide a uniquely useful model for studying the microbially mediated health impacts of rapid environmental change that results in mismatch between host , microbiota , and/or environment , a situation thought to apply to humans in industrialized settings ( Sonnenburg and Sonnenburg , 2019 ) . Understanding what shapes the domesticated microbiota may therefore identify routes to improve experimental models , animal condition , and human health . First , we characterized the fecal microbiota of wild and domesticated populations of nine pairs of artiodactyl , carnivore , lagomorph , and rodent species ( Figure 1A ) using 16S rRNA gene amplicon sequencing and quantitative PCR ( qPCR ) . Despite observing no single convergent ‘domesticated microbiota’ profile , our analysis detected a global signal of domestication status on gut microbiota composition . Across the combined dataset , the factor that explained the largest proportion of variation was the host dyad ( e . g . , pig/boar; p<0 . 001 , R2 = 0 . 39 , F = 17 . 086 , permutational multivariate analysis of variance [PERMANOVA]; Figure 1B ) . However , correcting for host dyad , domestication status also contributed significantly to variation in microbial communities ( p<0 . 001 , R2 = 0 . 15 , F = 6 . 081 ) , and these results were robust to the distance metric analyzed ( Supplementary file 1 ) . Furthermore , analyses of individual dyads found a significant effect of domestication status for all groups except canids ( p<0 . 05 , R2 = 0 . 18–0 . 41 , PERMANOVAs; Supplementary file 1 ) . Diet and digestive physiology were also primary determinants of the gut microbiota ( diet: p<0 . 001 , R2 = 0 . 12 , F = 21 . 216; physiology: p<0 . 001 , R2 = 0 . 14 , F = 23 . 938; Figure 1—figure supplement 1 ) , as seen in other surveys of mammals ( Muegge et al . , 2011 ) , with effect sizes comparable to that of domestication status . Consistent with the idea that higher ecological homogeneity in domesticates may beget greater gut microbial homogeneity , we found that there was greater between-conspecific variability in wild gut communities than in domesticated gut communities ( p=0 . 002 , F = 8 . 838; permutation test for F ) . To determine whether there was a consistent change in microbial composition with domestication , we calculated the difference between an individual’s ordination coordinates and the average ordination coordinates of its host dyad along the first nonmetric multidimensional scaling ( NMDS ) axis . Quantifying this ordination shift allowed us to consider overall changes in composition while correcting for host dyad and retaining information on the directionality of changes . We found that domesticated individuals were typically further right relative to the average of their host dyad ( p=0 . 006 , Mann–Whitney U test; Figure 1C ) and that this difference was significant regardless of the distance metric analyzed ( Supplementary file 1 ) . Most domesticated species displayed similar trends in these ordination shifts ( Figure 1—figure supplement 2 ) with laboratory and companion animals showing significant differences when analyzed collectively ( p<0 . 05 , Mann–Whitney U tests; Figure 1—figure supplement 2 ) . Free-ranging wild animal populations representing the progenitor species were not sampled for all pairs , potentially limiting the scope of our analysis . To assess whether the patterns described held for more stringent groupings , we analyzed the subset of wild–domesticated dyads for which the wild member was from a free-ranging population ( i . e . , ‘truly wild’ ) as well as the subset for which the wild member was the known progenitor ( i . e . , ‘perfect pair’ ) ( Supplementary file 2 ) . In both cases , we still found that domestication status explained a meaningful portion of variation in gut microbial community composition , regardless of the distance metric used ( all p<0 . 001 , all R2 > 0 . 11 , PERMANOVA; Supplementary file 1 ) . Domestication of mammalian species occurred at different times , so the evolutionary relationships between members of a host dyad are not all equal , even in cases where we sampled the known progenitors . Supporting an underlying influence of host evolutionary history on the gut microbiota , we found that host species that were more closely related ( i . e . , had a shorter time since divergence ) had more similar microbial community compositions ( p<0 . 001 , r = 0 . 157 , Mantel test ) . Similarly , the magnitude of the ordination shifts along NMDS axis 1 were smaller for animals from host dyads that were more closely related ( p=0 . 012 , rho = 0 . 19 , Spearman correlation; Figure 1—figure supplement 3 ) . Nevertheless , supporting the idea that ecology plays a dominant role in shaping the gut microbiota , average dissimilarity between members of a dyad was not lower in species pairs with more recent dates of domestication ( p=0 . 854 ) or more recent divergence times ( p=0 . 380; Figure 1—figure supplement 3 ) . Moreover , differences in the ordination shifts along NMDS axis 1 associated with domestication remained significant even when correcting for host dyad and divergence time ( p<0 . 001 , likelihood test linear mixed effects models ) . Overall , dissimilarity between conspecifics was lowest , but dissimilarity between wild–domesticated dyads was significantly lower than for unrelated pairs ( p<0 . 001 , bootstrapped Kruskal–Wallis tests; Figure 1D ) . We also tested for differences in specific features of the gut microbiota between domesticated and wild mammals . Domestication status did not affect microbial density quantified as copies of the 16S rRNA gene per gram of feces ( p=0 . 089 , Mann–Whitney U test ) , Shannon index ( p=0 . 2017 ) , or operational taxonomic unit ( OTU ) richness ( p=0 . 3506; Figure 1—figure supplement 4 ) , indicating that the domestication signal overall was not primarily driven by microbial species loss . Consistent with experiencing heightened environmental exposure , wild animals generally harbored potential pathogen communities that were more diverse ( p=0 . 001 , Mann–Whitney U test ) and marginally more abundant ( p=0 . 092; Figure 1—figure supplement 4 ) . Among laboratory animals specifically , potential pathogen abundance ( p<0 . 001 ) and pathogen richness ( p<0 . 001 ) were substantially lower than among wild relatives , while total microbial density was higher ( p=0 . 006; Figure 1—figure supplement 2 ) . Companion animals did not differ significantly by domestication status for microbial density , diversity , or pathogen metrics . By contrast , agricultural animals had higher Shannon index and richness values ( p≤0 . 001 , Mann–Whitney U tests ) as well as marginally higher pathogen abundances ( p=0 . 067; Figure 1—figure supplement 2 ) compared with their wild counterparts . Domestication has had profound effects on both ecology and host genetics . To begin to tease apart the relative roles of ecological change and genetic change in shaping the gut microbiota in domesticates , we performed a series of reciprocal diet experiments that tested the extent to which gut microbial signatures of wild–domesticated dyads could be recapitulated and reversed solely by the administration of ecologically relevant diets . We first conducted a fully factorial experiment in which wild-caught and laboratory mice ( Mus musculus ) were maintained for 28 days on wild or domesticate diets ( Figure 2A , Supplementary file 3 ) . Overall , we found that host taxon explained the largest amount of variation in composition ( p<0 . 001 , R2 = 0 . 173 , F = 64 . 255 , PERMANOVA ) , but that diet ( p<0 . 001 , R2 = 0 . 042 , F = 15 . 427 ) and a host taxon by diet interaction term ( p<0 . 001 , R2 = 0 . 020 , F = 7 . 557 ) were also significant ( Figure 2B , Figure 2—figure supplement 1 ) . Ordination shifts describing changes in the gut microbial community over the course of the experiment depended on the experimental group ( axis 1: p<0 . 001 , linear mixed effects model likelihood test; Figure 2D ) . Confirming prior reports that diet plays a dominant role in shaping the murine gut microbiota ( Carmody et al . , 2015 ) , the gut microbiota of wild mice fed a domesticate diet ( WildH/DomD ) moved toward the average microbial community of domesticated mice fed a domesticate diet ( DomH/DomD ) , the microbiota of domesticated mice fed a wild diet ( DomH/WildD ) moved in the opposite direction , and those of control wild or domesticated mice consuming their habitual diets ( WildH/WildD and DomH/DomD ) did not shift ( Figure 2B ) . Over the course of the experiment , alpha-diversity as measured by Shannon index also changed significantly across treatment groups ( p=0 . 025 , Kruskal–Wallis test; Figure 2—figure supplement 2 ) , with DomH/WildD mice becoming significantly more diverse ( p=0 . 004 , one-sample Wilcoxon test ) despite lower baseline levels of alpha-diversity in domesticated versus wild mice ( p=0 . 011 , Mann–Whitney U test; Figure 2C ) . Neither host taxon nor diet was associated with differences in gut microbial density over the experiment ( p=0 . 272 , Kruskal–Wallis test; Figure 2—figure supplement 2 ) , but it is notable that the total amount of feces produced , and thus likely the total number of bacteria , was lower in both host taxon groups when consuming the wild diet ( p<0 . 001 , Kruskal–Wallis test; Figure 2—figure supplement 2 ) . Despite similar trends in fecal production between wild and domesticated mice in response to diet treatment , wild and domesticated mice differed markedly in their ability to harvest energy from experimental diets ( p<0 . 001 , Kruskal–Wallis test; Figure 2—figure supplement 2 ) , as indexed by bomb calorimetry of feces . While wild mice were equally efficient digesters of the wild and domesticated diets , laboratory mice captured 15% fewer calories when consuming the wild versus domesticated diet . Interestingly , asymmetries were also observed between wild and domesticated mice in their gut microbial responses to reciprocal diets . Whereas the microbial communities of WildH/DomD mice grew to resemble those of untreated DomH/DomD mice , the microbial communities of DomH/WildD mice remained distinct from untreated WildH/WildD mice throughout the experiment ( p=0 . 042 , Mann–Whitney U test; Figure 2B , E ) . It is possible that the asymmetry in energy harvest between wild and domesticated mice was rooted in differential microbial responses to reciprocal diets and the inability of DomH/WildD mice to harbor a wild-type microbiota . Based on the lower alpha-diversity in domesticated versus wild mice ( Figure 2C ) , we hypothesized that the asymmetries between domesticated and wild mouse responses to altered diets were due to past extinction of relevant strains from laboratory microbial communities and no dispersal source of replacement strains ( Sonnenburg et al . , 2016 ) . Therefore , we next tested whether experimental dispersal from a wild microbial community in conjunction with feeding a wild diet could support a fully wild microbial community in laboratory mice ( Figure 3A ) . A single colonization treatment with a wild mouse cecal community ( via gavage ) led to significant shifts in the gut microbial community ( Figure 3B , Figure 3—figure supplement 1 ) , resulting in closer resemblance to the wild donor ( p<0 . 001 , Mann–Whitney U test; Figure 3C ) . Shifts in NMDS axis 1 varied across experimental treatment groups ( p<0 . 001 , linear mixed effects model likelihood test; Figure 3D ) . While laboratory mice fed a wild diet but given a control gavage ( PBS ) also moved toward the donor along NMDS axis 1 ( p=0 . 002 , one-sample Wilcoxon test; Figure 3D ) , reflecting the influence of diet , we observed a substantially greater shift following experimental colonization ( p<0 . 001 , Kruskal–Wallis test ) . Surprisingly , among colonized mice , movement of the microbial community toward the wild donor profile was profound even without reinforcement from the wild diet ( p=0 . 182 , Mann–Whitney U test ) . Although all mice exhibited an increase in microbial density over the course of the experiment ( p<0 . 01 , one-sample Wilcoxon tests ) , colonization with a wild community did not lead to higher microbial density overall ( p=0 . 449 , Kruskal–Wallis test; Figure 3—figure supplement 1 ) nor to an increase in alpha-diversity relative to baseline ( p=0 . 258 , one-sample Wilcoxon test ) . As in the original reciprocal diet experiment , wild diet treatment led to lesser fecal production ( p<0 . 001 , Kruskal–Wallis test; Figure 3—figure supplement 1 ) . No differences in fecal production were observed between mice colonized with a wild community and PBS-treated controls ( p=0 . 79; Figure 3—figure supplement 1 ) , suggesting that lower fecal output on the wild diet was not a direct consequence of harboring a wild microbiota . Together , these results suggest that differences observed with experimental colonization reflected shifts in gut microbial community structure rather than simple augmentation of microbial load . To test if our findings were generalizable beyond mice , we conducted an analogous reciprocal diet experiment in captive sympatric populations of wolves and dogs ( Figure 4A ) . We tracked gut microbial dynamics in these canids for 1 week on their standard diet ( raw carcasses or commercial dog food , respectively ) and 1 week on the reciprocal diet . As in the mouse experiment , we found that host taxon ( wild or domesticated ) explained the largest amount of variation in gut microbiota composition ( p<0 . 001 , R2 = 0 . 098 , F = 13 . 730 , PERMANOVA ) , but that diet ( p<0 . 001 , R2 = 0 . 058 , F = 8 . 151 ) and a host taxon by diet interaction term ( p<0 . 001 , R2 = 0 . 028 , F = 3 . 934 ) were also significant ( Figure 4B , Figure 4—figure supplement 1 ) . There were significant differences among experimental groups in the magnitude of their ordination shifts along the first NMDS axis over the experimental periods ( p<0 . 001 , linear mixed effects model likelihood test; Figure 4D ) . As in the mouse experiments , we observed that animals on reciprocal diet treatments ( DomH/WildD; WildH/DomD ) moved significantly toward the habitual gut microbial profile of the other species ( p<0 . 05 , one-sample Wilcoxon tests; Figure 4D ) , while the microbiota of animals consuming their habitual diet ( DomH/DomD; WildH/WildD ) did not shift predictably ( p>0 . 100 ) . In addition , we again observed an asymmetry between domesticated and wild animals in the degree to which the gut microbiota responded to diet . On experimental diets , dogs and wolves differed significantly in their dissimilarity to diet controls ( p<0 . 001 , Kruskal–Wallis test; Figure 4E ) , with the gut microbial communities of dogs fed raw carcasses resembling those of wolves at baseline but the gut microbial communities of wolves fed dog food remaining distinct from those of dogs at baseline ( p=0 . 001 , Mann–Whitney U test ) . The difference in the direction of asymmetry between the canid and mouse experiments may be explained by the different trends in dietary ecology between carnivores and omnivores during domestication . Carnivores , through the addition of extensive carbohydrates to their diet ( Wolfe et al . , 2007 ) , likely encounter more diverse diets in captivity than in the wild , whereas captive herbivores and omnivores typically eat a lesser number of plant species or are maintained on a single feed mix . Supporting this , we found that dogs initially had significantly higher OTU richness ( p<0 . 001 , Figure 4—figure supplement 2 ) and Shannon index ( p=0 . 003 , Figure 4C ) than wolves , but that reciprocal diets led to a switch in diversity ( richness: p=0 . 002 , Mann–Whitney U tests ) , with wolves becoming more diverse when fed dog food while dogs lost diversity when fed raw carcasses ( Figure 4—figure supplement 2 ) . We next explored the extent to which humans harbor gut microbial signatures analogous to those of domestication . Given evidence that the gut microbiota of domesticated animals is shaped by both ecology and speciation , we began by comparing humans to chimpanzees , one of our two closest living relatives . Humans may have undergone a form of self-domestication as a result of selection against aggression ( Wrangham , 2018; Theofanopoulou et al . , 2017 ) in addition to significant ecological change since our divergence from Pan , suggesting that the gut microbial signatures of animal domestication and Pan–Homo speciation could share features in common . We first compared samples that we collected from industrialized humans and wild chimpanzees , finding that the gut microbial communities of these humans and chimpanzees exhibited differences that paralleled those observed between domesticated animals and their wild counterparts when compared in the same ordination space ( p<0 . 001 , Mann–Whitney U test; Figure 5A , B ) . Microbial density ( p=0 . 002 , Mann–Whitney U test ) and Shannon index ( p=0 . 018; Figure 1—figure supplement 4 ) also differed between industrialized humans and wild chimpanzees , confirming prior reports that industrialized humans harbor microbial communities with substantially lower alpha-diversity ( Smits et al . , 2017 ) . We found only a marginal difference in between-conspecific variability in the gut microbiota of industrialized humans and wild chimpanzees ( p=0 . 092 , F = 3 . 0987; permutation test for F ) . Including these human–chimpanzee comparisons in our analysis of the relationship between gut microbiota dissimilarity and the time since dyad divergence strengthened the observed relationship ( p<0 . 001 , r = 0 . 5251; Mantel test ) , with a conservative divergence time of 6 . 5 million years assumed for Pan–Homo in this analysis . However , given the vast ecological differences between wild chimpanzees and industrialized humans , it remained unclear the extent to which these Pan–Homo differences reflected host phylogenetic distance as opposed to ecology . To better gauge the divergence attributable to phylogenetic distance versus ecology , we proceeded to compare the gut microbial communities of humans living in industrialized versus non-industrialized subsistence or agricultural societies , who are all equidistantly related to chimpanzees . Reanalysis of our cross-species comparison including published data on human populations in rural Nepal and Tanzania pursuing various non-industrialized lifestyles Jha et al . , 2018 found that the gut microbial communities of these non-industrialized populations differed substantially from those of two independent U . S . samples , instead clustering more closely to those of chimpanzees in this ordination space ( Figure 5A ) . Only the gut microbial communities of our industrialized populations show the rightward ordination shift along the Bray–Curtis NMDS axis 1 that is also exhibited by the gut microbial communities of domesticated animals ( p<0 . 05 , Mann–Whitney U tests; Figure 5B ) . Moreover , bacterial taxa previously found to distinguish among human lifestyles ( Smits et al . , 2017 ) typically had relative abundances that varied in the same direction between wild and domesticated animals as among wild chimpanzees , non-industrialized human populations , and industrialized human populations ( Figure 5—figure supplement 1 ) . These trends were clearest in the bacterial family Bacteroidaceae , which exhibited a continuous increase from chimpanzees to non-industrialized populations to industrialized populations as well as an increase in domesticated relative to wild animals ( p=0 . 008 , Mann–Whitney U test ) . Together , these data indicate that the human gut microbiota does not carry a global signal of domestication , as would be predicted under a hypothesis of human gut microbial self-domestication . Rather , the corresponding gut microbial responses to domestication and industrialization suggest that these responses are more likely driven by common ecological factors , a conclusion further supported by the observation that the gut microbial communities of domesticated animals were more similar to those of industrialized humans than were those of their wild animal counterparts ( p<0 . 001 , bootstrapped Mann–Whitney U test; Figure 5C ) . Notably , the gut microbial communities of domesticated animals and industrialized humans most closely resembled one another for companion animals ( p<0 . 001 , Kruskal–Wallis test; Figure 1—figure supplement 2 ) , presumably reflecting their greater ecological convergence and degree of physical contact ( Song et al . , 2013 ) . Also supporting the role of ecology driving these trends , we found that an independent sample of captive chimpanzees did not cluster exclusively with the wild chimpanzee samples; indeed , their gut microbial communities were more similar to those of non-industrialized humans than to those of wild chimpanzees ( p<0 . 001 , Mann–Whitney U tests; Figure 5—figure supplement 2 ) . Our data demonstrate that while domestication has not led to a convergent ‘domesticated microbiota , ’ there is nevertheless a significant signal of domestication on gut community composition across diverse mammalian hosts . Furthermore , our experimental and cross-sectional analyses suggest that the domestication effect can , in large part , be ascribed to environmental rather than genetic changes . As in many comparative microbiota studies , host taxonomy was responsible for the largest component of variation in our cross-species analyses , but the contribution of domestication status was comparable to those of diet type and host physiology , factors previously identified as key drivers of the mammalian gut microbiota ( Ley et al . , 2008; McKenzie et al . , 2017 ) . Experimental diet intervention in wild/domesticated pairs reduced gut microbial dissimilarity and alpha-diversity differences between members of the dyad over short time scales . However , differences due to loss or gain of taxa during domestication could not be overcome by diet shifts alone , necessitating experimental recolonization . Together , these results indicate that domestication has played a large role in shaping the microbiota and , through husbandry practices , likely continues to do so today , suggesting that studying variations in animal husbandry practices may illuminate new levers for manipulating the mammalian gut microbiota ( Velazquez et al . , 2019; Villarino et al . , 2016; Schmidt et al . , 2019 ) . Although there are many shared features of contemporary ecology and historic artificial selection on domestic animals , it is perhaps unsurprising that domestication has not produced a single , convergent domesticated gut microbiota . The animals we characterized represent a diverse set of lineages in the mammalian clade , and thus their microbiota also differ due to variation in factors such as gut structure and size , passage rate , diet , and biogeography ( Ley et al . , 2008; Youngblut et al . , 2019 ) . While the domestication signal was comparable in magnitude to those of gut physiology and diet type , it does not mask those fundamental structuring forces . Furthermore , the particulars of a domesticated lineage can help clarify what aspects of ecology are most salient to the domestication effect . Cases where domestication effects are weaker in our comparative study generally consist of animals where the ecological change associated with domestication has been small – for example , sheep and pigs , whose diets may be quite similar to their wild progenitors , at least when kept in the non-industrialized agricultural settings that were sampled ( McClure , 2013 ) – or where ecological changes are in the opposite direction from the domesticated norm – for example , canids , where the domesticate diet typically involves lower protein and higher carbohydrate levels than wild diets , instead of the higher protein levels seen in most laboratory or farm animals ( Axelsson et al . , 2013 ) . Our reciprocal diet experiments in mice and canids substantiate our claim that ecology plays a predominant role in shaping the domesticated gut microbiota . However , they do not pinpoint the mechanism ( s ) for these effects . Variability in diet or other aspects of ecology and their concomitant effects on host physiology ( e . g . , passage rate ) can alter microbial composition or abundance through changes in the selective landscape that microbes experience and changes in environmental exposure ( David et al . , 2014; Carmody et al . , 2019 ) . Animals may experience altered bacterial colonization , leading directly to changes in composition , and/or viral colonization , which could then alter the bacterial community if new bacteriophages target gut bacteria or if eukaryotic viruses activate the host immune system , leading to transformations in the gut environment . That the gut microbial impacts of change in a single ecological variable like diet were sufficiently profound to balance those of host taxon identity suggests that suites of ecological variables changing together , such as during domestication or industrialization , may have collectively exerted an even larger influence ( Jha et al . , 2018 ) . Of course , microbiota changes were not the only pathway for animals undergoing domestication to respond to changing ecological factors . For example , genetic changes have enhanced the capacity for starch digestion in dogs ( Axelsson et al . , 2013; Reiter et al . , 2016 ) . Nevertheless , the increased microbial diversity and shifts in microbial composition that we observed in dogs may have additionally contributed to carbohydrate digestion . Indeed , dogs fed conventional diets have greater representation of carbohydrate metabolism genes in their gut metagenomes than do dogs fed meat-based diets ( Alessandri et al . , 2019 ) . Notably , the microbiota has been found to supplement evolutionary responses during dietary niche expansion in wild animals that consume plants high in toxins ( Kohl et al . , 2014 ) . As such , although hosts and their various gut microbial taxa are each expected to pursue their own fitness interests , gut microbial disparities observed between domesticated and wild animals , and more generally in other organisms under rapid environmental change , could potentially be adaptive for the host ( Alberdi et al . , 2016 ) . Beyond host support of a gut microbiota that can better digest a domesticate diet , humans may have selected for animals harboring a microbiota that helped them grow and reproduce well on such diets , thereby applying unconscious selection on the microbiota ( Zohary et al . , 1998 ) . Changes in microbial function that enhanced host dietary energy harvest , survivorship , or reproduction may have been particularly important early in domestication , before host evolution occurred , although that hypothesis remains to be tested empirically . Regardless , specialization of microbial performance on domesticate diets could conceivably have come at the cost of broader digestive capacity , as seen in the laboratory mouse microbiota , which was better at harvesting energy from domesticate diets than from wild diets ( Figure 2—figure supplement 2 ) . It may also have impacted the immunological functions of the gut microbiota . The elevated pathogen abundances found in wild populations overall may largely be ascribed to low pathogen abundances in laboratory animals ( Figure 1—figure supplement 2 ) , which are maintained under specific pathogen-free conditions that minimize the likelihood of infection . Under natural conditions , though , the domesticated microbiota may exhibit reduced colonization resistance or immune system functioning ( Rosshart et al . , 2017; Beura et al . , 2016; Rosshart et al . , 2019 ) , resulting in higher pathogen colonization , as observed here in agricultural animals . Future studies examining the trade-offs among microbially mediated functions , like digestive capacity , reproduction , and immunity , will help to illuminate the complex selection pressures shaping domesticated animals and their gut microbiota ( Reese and Kearney , 2019 ) . We observed some correspondence between the gut microbial signatures of animal domestication and human industrialization that is most likely attributable to convergent ecological changes . The observation that gut microbial divergence among Pan and Homo primarily affects industrialized populations specifically implicates recent ecological changes as opposed to either ecological changes with deeper roots in human evolution or host evolutionary changes . Many recent human ecological changes involve accelerations of basic patterns established during the evolution of Homo , including increased proportion of calories from fat and protein , increased dependence on animal source foods , and extensive food processing involving both chemical and physical changes to food ( Carmody , 2017 ) . Other ecological changes are likely specific to industrialization , including reduced physical activity , high population density , and antibiotic use . These factors would be absent even in populations currently transitioning from subsistence to industrialized lifestyles ( Jha et al . , 2018 ) , but may overlap with changes experienced by domesticated animals in their diets , habitats , and social milieu . While we limited our analysis to human–chimpanzee comparisons because Pan is the closest sister clade to Homo , recent work has indicated that the human gut microbiota is more similar to that of baboons ( Gomez et al . , 2019; Amato et al . , 2019 ) . Baboons are more distantly related to humans but have been argued to be closer in terms of diet and dietary physiology ( Codron et al . , 2008; Lambert , 1998 ) , accentuating our finding of the importance of ecological factors in shaping the microbiota . Further work will be required to assess the specific combination of ecological factors driving similarities between domesticated and industrialized gut microbial signatures . Because laboratory animals demonstrate some of the largest overall differences relative to their wild counterparts , they might be expected to have high translational potential as models for studying the gut microbiota of industrialized human populations . However , recent findings show that laboratory mice are poorer immunological models for humans in industrialized settings than are wild mice or laboratory mice harboring a wild microbiota ( Beura et al . , 2016; Rosshart et al . , 2019 ) . While the industrialized human gut microbiota exhibits parallels to those of domesticated animals , it may experience a broader array of environments and greater temporal variability; for example , greater ecological variability may explain the elevated gut microbial Shannon diversity seen in humans as compared to laboratory animals ( Figure 1—figure supplement 2 ) . Alternatively , it may be that domesticated laboratory animals are strong models for some aspects of host–microbe biology other than immunology . Certainly , studies of non-domesticated animals will be necessary to understand the natural history of host–microbe interactions ( Reese and Kearney , 2019; Hird , 2017 ) , as well as to determine the most appropriate models for translational research . The fact that laboratory mice were permissive of recolonization by wild strains indicates that the local extinctions that occurred during domestication and/or generations in captivity can potentially be mitigated , thereby potentially improving the utility of these animals for research . Previous work has relied on laboratory mice colonized with a wild microbiota but fed standard laboratory chow ( Rosshart et al . , 2017; Rosshart et al . , 2019 ) or on wild mice fed wild diets ( Martínez-Mota et al . , 2020 ) . A combination of these approaches – adding wild gut microbial community members and feeding wild diet – would be expected to best support a wild gut microbiota in laboratory mice . A wild-microbiota laboratory-genotype model could be especially useful for studying infection challenges , disentangling host gene versus microbiota contributions to disease phenotypes , and testing for host–microbiota coevolution ( Rosshart et al . , 2019 ) . More generally , our data add to growing evidence that the gut microbiota is finely tuned to variations in the environment , affording at once expanded opportunities for biological mismatch to arise between the host and microbiota and for rapid microbiota-mediated host adaptation to novel environments . Further work to characterize the ecological significance of gut microbial plasticity will help reveal the fundamental nature of the host–microbial relationship , the conditions under which plasticity is beneficial versus detrimental , and the ecological conditions promoting cooperative , commensal , and competitive dynamics . The answers will have profound implications for our definition and pursuit of a healthy gut microbiome . Distal gut microbiota samples from a range of non-human species were collected by authors or collaborators . Fecal samples from non-human mammals were collected from the ground within seconds to a few hours ( <6 ) of production over the course of 2017 and 2018 . In the case of artiodactyl , carnivore , lagomorph , and rodent feces , this approach precluded the need for institutional approval . Wild chimpanzee fecal samples were collected by field assistants under the approval of the University of New Mexico IACUC ( protocol 18-200739-MC ) and with permission of the Uganda Wildlife Authority and Uganda National Council for Science and Technology . Captive chimpanzee fecal samples were collected passively by keepers at Southwick’s Zoo , Mendon , MA . Human samples were self-collected by healthy study participants after providing written informed consent under the approval of the Harvard University IRB ( protocol 17-1016 ) ( Carmody et al . , 2019 ) . Samples were immediately frozen prior to permanent storage at –80°C . The only exceptions were wild vicuña and wild chimpanzee samples , which were preserved in RNAlater stabilization solution ( Invitrogen ) due to logistical issues in transportation from remote sampling locales . RNAlater was removed from these samples with centrifugation prior to further processing , and while sample preservation method was significantly associated with microbiota composition ( p<0 . 001 , PERMANOVA ) it explained only a minor portion ( R2 = 0 . 01 ) of the variation in beta-diversity . Sample sizes were chosen based on animal availability with a N > 5 for all species . To compare the microbial differences observed between wild and domesticated animals and between humans and chimpanzees with differences linked specifically to industrialization , we also performed analyses including all of the samples outlined above plus a subset of published data from Jha and colleagues ( Jha et al . , 2018 ) . To match sample sizes used in our human–chimpanzee contrast , we subsampled seven adults from their Chepang ( Nepalese foragers ) , Raji ( Nepalese foragers transitioning to subsistence farming ) , Raute ( Nepalese foragers transitioning to subsistence farming ) , Tharu ( Nepalese subsistence farmers ) , and American populations , as well as seven adults from the Hadza ( Tanzanian hunter gatherers ) population they analyze , which were originally described in another study ( Smits et al . , 2017 ) . All data were downloaded from the European Nucleotide Archive . These populations represent extremes of industrialized and non-industrialized human lifestyles with the variation among the non-industrialized groups not covering the full breadth of intermediate lifestyles ( e . g . , modern agricultural or recent urban transplants ) . We believe that these extremes enable us to test how the human gut microbial communities respond to major ecological change of a magnitude that could be argued to approximate that experienced by gut microbial communities of animals undergoing domestication . These samples were not necessarily collected or processed in an identical manner to each other or to the new data collected in this paper – namely , the Chepang , Raji , Raute , and Tharu samples were collected and preserved using OMNIgene kits while the American and Hadza samples were frozen , while all samples were extracted with the MoBio PowerSoil kit and sequenced on an Illumina MiSeq . Unfortunately , no existing published data on non-industrialized populations have been generated using exactly the same methods employed here . However , we reprocessed the sequences using the 16S rRNA gene amplicon QIIME pipeline described below and rarefied all samples to 10 , 000 reads depth to make the data as comparable as possible . Importantly , to the extent that these discrepancies introduce biases to our analyses , we expect they would do so in a manner agnostic to the comparison with our chimpanzee and American human samples . The high similarity between the US samples that we collected and those collected by Jha and colleagues supports this expectation . Furthermore , the fact that the non-industrialized Hadza samples were not stored with OMNIgene kits precludes conflating any non-industrialized signal with a sample-processing signal . Specific taxa chosen for targeted analyses were identified from the human lifestyle analyses by Smits and colleagues ( Smits et al . , 2017 ) ; only taxa that had a non-zero abundance in wild chimpanzees were analyzed here . Time since Pan–Homo divergence was drawn from http://timetree . org ( Kumar et al . , 2017 ) to be consistent with domestication analyses . All statistical analyses were carried out in R ( R Core Team , version 3 . 3 ) . Alpha-diversity ( Shannon index , OTU richness ) were calculated for rarefied OTU tables ( rarefaction limit of 17 , 500 for cross-species dataset; 27 , 000 for wild mouse study; 15 , 500 for the mouse colonization study; 7 , 500 for canid experiment ) . Beta-diversity ( Bray–Curtis , Weighted UniFrac , Unweighted UniFrac ) metrics were calculated using the vegan package ( Oksanen et al . , 2017 ) or QIIME on unrarefied data . All statistical tests performed were non-parametric . PERMANOVA was carried out with the adonis2 function in vegan with the domestication status variable nested within the species pair to correct for known relationships within dyads . To test how beta-diversity varied based on relatedness ( within species , between wild–domesticated pairs , or among unrelated pairs ) , domestication type ( relative to US human samples ) , or human/primate population ( relative to zoo chimpanzee samples ) , we used a bootstrapping approach , thus correcting for the non-independence of dissimilarity measurements that include the same individuals in multiple comparisons . In short , we permuted the Mann-Whitney U test statistics and p values , resampling ( 25 , 000 permutations ) with stratification specified by individual identity , using the boot package ( Canty and Ripley , 2020 ) . Variability in a species’ microbial community composition was calculated with the permutest and betadisper functions in vegan . For changes in family-level abundance , a Bonferroni correction for multiple hypothesis correction was then applied to all test results . Potential human pathogens were identified following published methods ( Kembel et al . , 2012; Reese et al . , 2016 ) . In short , we obtained a list of potential human pathogens , compiled by Kembel and colleagues ( Kembel et al . , 2012 ) , then manually compared that list to the taxa identified to the genus or species level in our analysis . A subset of the data containing only these species was then analyzed for diversity with the same methods used for the total dataset . To determine the consistency of gut microbial differences across ordination space due to domestication , Pan–Homo divergence , or industrialization in the observational study , we calculated the average position of the host dyad ( e . g . , pig/boar ) or all primates ( humans and chimpanzees ) for axis 1 of the NMDS , then measured the displacement along each axis for an individual sample relative to that mean position . We tested for differences in these ordination shifts by domestication status or primate host taxonomy ( e . g . , chimpanzee versus US human ) . To estimate the direction and magnitude of changes in beta-diversity during the experimental studies , we tested whether inclusion of a treatment group term improved the performance of a linear mixed effects model relative to a model with only time and animal ID terms for predicting the NMDS1 axis value for an individual . These analyses allowed us to consider the direction of beta-diversity changes in addition to the magnitude . We estimated the direction and magnitude of dissimilarity from the expected community composition ( donor microbial community in gavage experiment; baseline species average for DomH/DomD or WildH/WildD in diet experiments ) as the length of the vector through the first axis of ordination space . In analyzing the experimental diet study data , we used the lmer and anova functions in the package lme4 ( Bates et al . , 2015 ) to perform likelihood tests comparing a linear mixed effects model that included the variable of interest ( i . e . , treatment group ) to a model that included only time variables . In both models , individual identity was included as random effects . We explored the role of relatedness in structuring the cross-species dataset by ( i ) performing a Mantel test to compare divergence times and Bray–Curtis dissimilarities; ( ii ) testing for Spearman correlations between the NMDS shifts and the time since divergence and performing likelihood tests to compare a linear mixed effects model that included both domestication status and dyad as fixed effects and divergence time as a random effect with a model that only included the dyad and divergence time terms; and ( iii ) testing for Spearman correlations between the average dissimilarity within a wild–domesticated dyad ( e . g . , the average dissimilarity for all combinations of boar–pig pairs ) and the time since domestication and time since divergence . We also used Mann–Whitney U tests to determine if dissimilarity between unrelated pairs was higher than for wild–domesticated dyads or within sets of conspecifics .	Living inside our gastrointestinal tracts is a large and diverse community of bacteria called the gut microbiota that plays an active role in basic body processes like metabolism and immunity . Much of our current understanding of the gut microbiota has come from laboratory animals like mice , which have very different gut bacteria to mice living in the wild . However , it was unclear whether this difference in microbes was due to domestication , and if it could also be seen in other domesticated-wild pairs , like pigs and wild boars or dogs and wolves . A few existing studies have compared the gut bacteria of two species in a domesticated-wild pair . But , studies of isolated pairs cannot distinguish which factors are responsible for altering the microbiota of domesticated animals . To overcome this barrier , Reese et al . sequenced microbial DNA taken from fecal samples of 18 species of wild and related domesticated mammals . The results showed that while domesticated animals have different sets of bacteria in their guts , leaving the wild has changed the gut microbiota of these diverse animals in similar ways . To explore what causes these shared patterns , Reese et al . swapped the diets of two domesticated-wild pairs: laboratory and wild mice , and dogs and wolves . They found this change in diet shifted the gut bacteria of the domesticated species to be more similar to that of their wild counterparts , and vice versa . This suggests that altered eating habits helped drive the changes domestication has had on the gut microbiota . To find out whether these differences also occur in humans , Reese et al . compared the gut microbes of chimpanzees with the microbiota of people living in different environments . The gut microbial communities of individuals from industrialized populations had more in common with those of domesticated animals than did the microbes found in chimpanzees or humans from non-industrialized populations . This suggests that industrialization and domestication have had similar effects on the gut microbiota , likely due to similar kinds of environmental change . Domesticated animals are critical for the economy and health , and understanding the central role gut microbes play in their biology could help improve their well-being . Given the parallels between domestication and industrialization , knowledge gained from animal pairs could also shed light on the human gut microbiota . In the future , these insights could help identify new ways to alter the gut microbiota to improve animal or human health .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2021	Effects of domestication on the gut microbiota parallel those of human industrialization
Pneumococcal whole cell vaccines ( WCVs ) could cost-effectively protect against a greater strain diversity than current capsule-based vaccines . Immunoglobulin G ( IgG ) responses to a WCV were characterised by applying longitudinally-sampled sera , available from 35 adult placebo-controlled phase I trial participants , to a panproteome microarray . Despite individuals maintaining distinctive antibody ‘fingerprints’ , responses were consistent across vaccinated cohorts . Seventy-two functionally distinct proteins were associated with WCV-induced increases in IgG binding . These shared characteristics with naturally immunogenic proteins , being enriched for transporters and cell wall metabolism enzymes , likely unusually exposed on the unencapsulated WCV’s surface . Vaccine-induced responses were specific to variants of the diverse PclA , PspC and ZmpB proteins , whereas PspA- and ZmpA-induced antibodies recognised a broader set of alleles . Temporal variation in IgG levels suggested a mixture of anamnestic and novel responses . These reproducible increases in IgG binding to a limited , but functionally diverse , set of conserved proteins indicate WCV could provide species-wide immunity . Clinical trial registration: The trial was registered with ClinicalTrials . gov with Identifier NCT01537185; the results are available from https://clinicaltrials . gov/ct2/show/results/NCT01537185 . Streptococcus pneumoniae ( the pneumococcus ) , commonly carried in the nasopharynx , is an important respiratory pathogen capable of causing pneumonia , bacteraemia and meningitis . The earliest recorded pneumococcal vaccinations consisted of two doses of heat-killed pneumococci cultured from the sputum of pneumonia patients , which resulted in limited protection against pneumococcal infections for a few months after inoculation ( Maynard , 1915 ) . Later pneumococcal vaccines used purified capsule polysaccharides , of which there are almost 100 immunologically distinguishable variants ( Bentley et al . , 2006 ) , termed ‘serotypes’ . These formulations expanded from a bivalent formulation in the 1930s ( Ekwurzel et al . , 1938 ) to include 23 capsular polysaccharides by the 1980s ( Mufson et al . , 1985 ) . Such formulations afford little protection to infants , however , as polysaccharides are T-cell-independent antigens that are not efficiently recognised by the immature adaptive immune system ( Stein , 1992 ) . Therefore , the most commonly used pneumococcal vaccines at present are protein-polysaccharide conjugate vaccines ( PCVs ) , currently containing up to 13 different polysaccharides , each attached to a carrier protein ( Nunes and Madhi , 2011 ) . These vaccines elicit protective immune responses , even in young children , which prevent nasopharyngeal carriage as well as disease ( Lee et al . , 2014 ) . The intrinsic disadvantage of PCVs is the limited number of serotypes against which they provide protection , resulting in serotype replacement disease that reduces their impact ( Weinberger et al . , 2011 ) . The further expansion of their valency is limited by the complexity of their manufacture , which also makes them costly ( Miller et al . , 2011; Ray , 2002 ) . Therefore , efforts have continued to develop alternative vaccines that are cheaper , generate T-cell-dependent responses to non-capsular antigens , and afford protection against all pneumococci . Whole cell-based vaccines ( WCVs ) present a possible solution , as they can be relatively inexpensively manufactured and present an almost full complement of antigens to the recipient’s immune system . Rather than the historical precedent of killed clinical isolates , the WCV used in this study is a specifically engineered version of the unencapsulated strain S . pneumoniae RM200 ( Lu et al . , 2010b ) . A randomised double-blind phase I safety and immunogenicity trial of this WCV in 42 healthy U . S . adults was designed to compare four cohorts ( Figure 1 ) . One received placebo saline injections , while the other three received 100 μg , 300 μg or 600 μg doses of the pneumococcal WCV adsorbed to aluminium hydroxide adjuvant . Each individual was given three injections 28 days apart , and serum samples were taken pre-vaccination and at 28 , 56 and 84 days subsequent to the first injection to assay for immunoglobulin G ( IgG ) responses . To quantify the potential multiplicity of immune responses , a set of available samples were analysed with a panproteome microarray including over 2100 probes ( Croucher et al . , 2017 ) , most of which corresponded to full-length proteins , with others representing fragments of larger polypeptides . This provides information on responses to proteins encoded by the core and accessory genome , as well as multiple variants of diverse core loci ( DCL ) , corresponding to genes that can be identified in almost all isolates based on their location in the chromosome and the domain structure of the translated protein , but which exhibit little detectable sequence similarity across the species . In S . pneumoniae , four such loci encode pneumococcal surface proteins A ( PspA ) and C ( PspC ) , and the zinc metalloproteases A ( ZmpA ) and B ( ZmpB ) . This study therefore aimed to identify the range and types of antigens to which IgG responses were mounted following WCV administration , and how these might vary between individuals . The availability of longitudinal samples from multiple trial participants also offered the opportunity to study differences between individuals’ antibody repertoires . Finally , the trial structure allowed the kinetics of responses to successive doses of different concentrations to be to quantified . These analyses should provide information on whether the WCV is likely to consistently induce antibodies capable of recognising antigens conserved across S . pneumoniae isolates . The WCV formulation contains chemically-killed S . pneumoniae RM200 cells ( Malley et al . , 2001 ) , the genome sequence of which was aligned against the original progenitor , S . pneumoniae D39 ( Ravin , 1959 ) . Twenty recombinations distinguishing the pair were identified , one of which represented reversible inversion at a phase variable locus ( Croucher et al . , 2014 ) . The generation of S . pneumoniae SIII-N , an intermediate genotype expressing the mucoid serotype three capsule ( Ravin , 1959 ) , likely accounts for eighteen of these recombinations , which span a total of 101 kb in the RM200 genome and include a recombination importing the serotype three capsule polysaccharide synthesis ( cps ) locus ( Figure 1—figure supplement 1 ) . RM200 was derived from Rx1 , a spontaneous mutant of SIII-N that no longer expressed a capsule . Correspondingly , the cps locus contains two candidate mutations potentially responsible for this phenotype . An Arg320Cys substitution in the 6-phosphogluconate dehydrogenase protein replaces a catalytically-important arginine , which binds a pyrophosphate moiety , with a cysteine , which could interfere with the active site thiol group ( Campbell et al . , 2000 ) . Similarly , the phosphoglucomutase protein has an Asn146Thr substitution that disrupts a Ser-His-Asn motif involved in divalent cation coordination , which is conserved across many orthologues ( NCBI conserved domain cd05799 ) . A subsequently-introduced recombination represents further engineering to replace the lytA gene with the Janus cassette ( Sung et al . , 2001 ) , to reduce virulence and improve the yield of cells from culture ( Berry et al . , 1989; Lu et al . , 2010b ) ( Figure 1—figure supplement 1 ) . Alteration of the pneumolysin toxin gene was not associated with an inferred recombination , as only three bases were substituted , in order to remove the protein’s cytolytic and complement-activating activity ( Lu et al . , 2010b ) . This change also resulted in insertion of the pDP28 shuttle vector ( accession code KJ395591 ) at the adjacent site ( Figure 1—figure supplement 1 ) . Other than these known alterations , there were no large structural changes to the RM200 genome . The cryptic pDP1 plasmid of S . pneumoniae D39 was retained ( Oggioni et al . , 1999 ) , and the large adhesin PclA was still present , but the glycoprotein PsrP , both known pneumococcal pili , and degradative zinc metalloprotease ZmpC were absent ( Croucher et al . , 2017 ) . None of the DCL were affected by the recombinations , and consequently they were all similar to the alleles in D39 . However , there was a nonsense mutation in the pspC gene which removed six of the eight choline-binding domains ( CBDs ) , which may reduce the proportion of this protein attached to the cell surface . Overall , 130 samples were studied from 35 of the 42 trial participants: all four timepoints were analysed for 27 individuals , at least the initial and final samples were available for a further two people , and at least one timepoint was analysed for a further six people ( Figure 1 ) . Of these , 35 samples were from the placebo group , 32 from cohort 1 ( 100 μg dose ) , 29 from cohort 2 ( 300 μg dose ) , and 34 from cohort 3 ( 600 μg dose ) . Each sample was treated as a biological replicate . No technical replicates were included , as analysis of pilot data obtained using a smaller version of the array found reproducible differences between individuals ( Figure 2—figure supplement 1 ) . This suggested results were consistent between array assays of samples from the same individual , and that the study should span the maximal number of trial participants , as between-individual variation might be an important factor in understanding the response to the WCV . To visualise the relative importance of these differences between trial participants , the IgG binding across all probes for each sample was projected in two dimensions using t-distributed stochastic neighbor embedding ( t-SNE; Figure 2 ) , which clusters together similar sets of multidimensional data . This revealed individuals had a distinctive antibody ‘fingerprint’ that was generally preserved over the course of the trial , independent of vaccination dose . A similar pattern was observed when the DCL were excluded from the analysis ( Figure 2—figure supplement 2 ) , or when only probes from core proteins conserved across related streptococcal species were considered ( Figure 2—figure supplement 3 ) . Therefore , these fingerprints do not appear to reflect individuals’ distinctive histories of exposure to variable proteins , but instead unique patterns of IgG responses to common antigens . These were sufficiently robust not to be disrupted by WCV administration . Despite this starting variation between trial participants , there was no overall significant difference in individuals’ median IgG binding to S . pneumoniae proteins between cohorts before vaccination ( Figure 4—figure supplement 1A; Kruskal-Wallis test , N = 29 , χ2 = 0 . 20 , df = 3 , p = 0 . 98 ) . Changes in IgG binding between the start of the trial , day 0 , and the end , day 84 , were quantified as Δ0→84 ( Figure 4—figure supplement 1B ) . The cohorts did not differ in the distribution of median Δ0→84 per individual ( Kruskal-Wallis test , N = 29 , χ2 = 3 . 17 , df = 3 , p = 0 . 37 ) , indicating that the WCV did not raise IgG binding to a high proportion of S . pneumoniae proteins across all members of the vaccinated cohorts . Restricting this analysis to the antibody-binding targets ( ABTs ) , defined as those proteins associated with high IgG binding in the pre-vaccination sample ( Croucher et al . , 2017 ) , also failed to detect overall heterogeneity in the distribution of individuals’ median Δ0→84 between the groups ( Kruskal-Wallis test , N = 29 , χ2 = 4 . 63 , df = 3 , p = 0 . 20; Figure 4—figure supplement 1C ) . However , an ANOVA test of the fits of linear mixed effects models of Δ0→84 across probes ( considered fixed effects ) and individuals ( considered random effects ) found a significant improvement when cohort was also included as a fixed effect ( χ2 = 8 . 96 , df = 3 , p = 0 . 030; see Materials and methods ) . To test whether this reflected biologically uninformative changes in the responses to poorly immunogenic proteins , this analysis was repeated using only the 1 , 584 ‘immunoreactive’ probes , to which an IgG binding of at least one ( i . e . twice the background level ) was detected across the dataset . This found a more significant effect of including cohort as a fixed effect ( χ2 = 11 . 8 , df = 3 , p = 0 . 0081 ) , indicating there was a statistically , and potentially biologically , significant difference in the distribution of Δ0→84 between cohorts . Hence the WCV did have a detectable effect on vaccinated cohorts when considering per-probe , rather than per-individual , data . Combined with the observation that unique antibody fingerprints were maintained throughout the trial , this suggests the WCV either elevated responses to only a subset of proteins , or uniformly boosted multiple anamnestic responses in only a subset of individuals . Comparing the within-cohort medians of pre-vaccine IgG binding and Δ0→84 across all probes provided a simple approach to judging the relative contributions of these alternative explanations . If there were a broad response in only some individuals , the median Δ0→84 values would not be expected to vary much across probes , except for a general rise with higher pre-vaccine antibody binding , if this correlated with the strength of immunological memory . Alternatively , a strong response to only a subset of probes across individuals should result in much greater variation in within-cohort median Δ0→84 . The plots revealed a statistically significant positive correlation between pre-vaccine-binding strength and post-vaccine rise for the three vaccinated groups ( Pearson correlation coefficient , p < 10−16 for each ) , as expected for a uniform rise in IgG binding if many anamnestic responses were upregulated . By contrast , a negative correlation was evident in the placebo recipients , possibly representing regression to the mean ( Figure 3A–D ) . Yet these general increasing trends for the vaccinated cohorts were small compared to the large increases in IgG binding to a particular subset of antigens . This indicates the post-vaccine changes primarily represented strong responses to a subset of proteins . These plots indicated many probes associated with larger increases in IgG binding were already strongly recognised by adaptive immunity pre-WCV . This suggested WCV antigens , which were defined as being associated with Δ0→84 above the empirically-derived threshold Δ0→84 of 0 . 2 ( Figure 4—figure supplement 2 ) , would be enriched for ABTs , those proteins to which IgG binding was already high pre-WCV . ABT probes accounted for 23 of 47 probes rising above the threshold in cohort 1 ( 48 . 9% , a 5 . 95-fold enrichment; Fisher’s exact test , N = 2 , 343 , p = 6 . 11×10−9 ) , 17 of 43 probes in cohort 2 ( 39 . 5% , a 4 . 06-fold enrichment; Fisher’s exact test , N = 2 , 343 , p = 2 . 37×10−5 ) and 60 of 129 probes in cohort 3 ( 46 . 5% , a 5 . 40-fold enrichment; Fisher’s exact test , N = 2 , 343 , p < 10−16 ) . These tests consistently showed ABTs were disproportionately associated with increased IgG binding post-vaccination , although the absolute numbers indicate many ABTs did not trigger a substantial response . The distinction between those proteins to which there was little reaction , and those eliciting large increases in IgG binding post-WCV , was generally consistent both within cohorts and between vaccinated cohorts ( Figure 3E–F ) , suggesting these specific reactions to ABTs were not stochastic , nor driven by variation between individuals’ pre-existing antibody profiles . These results may represent WCV-induced responses being limited to those components of the panproteome that are similar to proteins expressed by RM200 . To test this hypothesis , an amino acid sequence identity threshold of at least 90% between array and RM200 proteins was empirically determined from the distribution of pairwise sequence identities ( Figure 4—figure supplement 4 ) . Excluding all DCL variants , 1602 proteins ( corresponding to 1647 probes ) could be matched to RM200 proteins with a similarity above the 90% threshold , while 455 proteins ( corresponding to 476 probes ) lacked a close orthologue in RM200 . Using these two categories to represent epitopes present and absent from the WCV respectively , a comparison of the median Δ0→84 using a Wilcoxon rank sum test found no significant difference in the post-WCV changes in IgG binding in cohort three between these sets of probes ( Figure 4A , left panel; Wilcoxon rank sum test , N = 2 , 123 , W = 403523 , p = 0 . 33 ) . However , an analogous test restricted to ABTs found significantly stronger antibody rises ( Δ0→84 ) associated with the 70 ABTs with a close orthologue in RM200 ( corresponding to 87 probes ) relative to the 25 ABTs absent from the strain ( corresponding to 40 probes; Figure 4A , middle panel; Wilcoxon rank sum test , N = 127 , W = 970 , p = 6 . 5×10−5 ) . This suggests a vaccine-induced IgG response that primarily targets naturally immunogenic proteins expressed by the WCV , although some ABTs present in the WCV still did not elicit substantially increased antibody binding . A possible confounding factor in this analysis is the higher level of sequence conservation expected for those proteins that are similar in RM200 and the array . This may mean these proteins’ greater Δ0→84 values could be the result of the array being more sensitive to IgG adapted to recognising comparatively invariant antigens . However , a comparison of Δ0→84 and sequence divergence between the WCV and proteins on the array failed to find evidence of a general correlation between Δ0→84 and lower sequence divergence between the array proteins and RM200 ( Figure 4—figure supplement 5 ) . Immune responses to more conserved proteins could also result from encounters with the related commensal streptococci Streptococcus mitis and Streptococcus pseudopneumoniae . However , the overall distribution of Δ0→84 for ABTs conserved in these other species was not significantly higher than to those ABTs exhibiting greater interspecies divergence ( Figure 4A , right panel ) . There was also no significant difference in the pre-vaccination IgG binding to ABTs in the WCV ( Wilcoxon rank sum test , W = 1455 , p = 0 . 20 ) , nor to those conserved in S . mitis and S . pseudopneumoniae ( Wilcoxon rank sum test , W = 937 , p = 0 . 62; Figure 4—figure supplement 3 ) , suggesting recent immune priming through asymptomatic carriage of naturally unencapsulated streptococci is unlikely to play a major role in determining the immunity induced by WCV . Rather than relying on a threshold value of Δ0→84 , two approaches commonly used in the analysis of transcriptome data were employed to identify which proteins elicited statistically significant IgG responses . Only the 1584 immunoreactive probes were included in each , to avoid analysis of changes unlikely to have biological relevance . Firstly , an empirical Bayes ( eBayes ) analysis , which calculates probabilities relative to a prior distribution estimated from the dataset itself , was applied to pairwise contrasts of Δ0→84 across immunoreactive proteins in each of the vaccinated cohorts against the same metric in the placebo group . By only using two timepoints , data from 29 individuals could be included . A Benjamini-Hochberg correction adjusting for 1584 tests , with an expected false discovery rate of 0 . 05 ( Supplementary file 2 ) , identified 88 probes as having significant Δ0→84 values in the comparison of the placebo group with cohort 3 , who received the highest dose of the WCV . The magnitudes of Δ0→84 in this cohort are compared with the statistical significance of this change relative to the placebo group in Figure 4B . After the correction for multiple testing , no probes were associated with significant changes in IgG binding in the comparisons of the placebo group with cohorts 1 or 2 . Secondly , a linear mixed effects model ( LMM ) was used to interpret the IgG binding at each timepoint as a noisy linear response to random effects , corresponding to the different trial participants , and two sets of fixed effects , corresponding to the dose of vaccine and number of injections received . This maximised the information extracted from the complete longitudinal time series , but was limited to the 20 individual trial participants in the three vaccinated cohorts for whom these data were available . Likelihood ratio tests were then used to identify proteins that elicited IgG responses that significantly increased with dose or with number of immunisations . After a Benjamini-Hochberg correction for multiple testing , none of the proteins showed a significant dose-response . This is likely due to the anomalous behaviour of cohort 2 , in which the overall changes in IgG were smaller than those for cohort 1 ( Figure 3 ) . By contrast , 127 probes showed a significant change in IgG binding with repeated immunisations , with all but one of these probes exhibiting increased IgG binding over the course of the trial ( Supplementary file 2 ) . Of these , 77 were also significant in the eBayes analysis of Δ0→84 values ( Figure 4—figure supplement 7 ) . Therefore , this combination of using all individuals for whom initial and final paired samples were available , as well as the full longitudinal data for vaccinated individuals , identified a total of 138 probes to which IgG binding changed significantly ( Table 1 ) . These were highly consistent with the 129 probes identified by applying the empirically-derived threshold to Δ0→84 values from cohort three in Figure 3D ( Figure 4—figure supplement 7 ) . Sixty of the 138 probes represented DCL variants; to test whether the inclusion of this allelic variation affected the model fitting or multiple correction testing , the eBayes and LMM analyses were repeated using the 1384 immunoreactive non-DCL probes . This found 74 of the 78 non-DCL probes were still associated with significant rises in IgG binding if the DCL were excluded from the analyses , with no extra hits being detected , indicating these results were generally robust to changes in the set of probes assayed on the array ( Supplementary file 2 and Figure 4—figure supplement 6 ) . Accounting for those probes that represented different regions of large proteins , the 138 probes came from 112 distinct protein sequences , of which 59 . 8% were identified by both eBayes and LMM approaches ( Figure 4—figure supplement 7 ) . Hence the statistical analyses provide a consistent view of the most immunogenic proteins in the WCV . Seventy-one of the 112 putative antigens were from the 208 ABTs previously defined using the pre-vaccination timepoint ( 63 . 4%; a 6 . 16-fold enrichment ) ( Croucher et al . , 2017 ) . This represented a significant enrichment of ABTs relative to the other immunoreactive proteins ( Fisher’s exact test , N = 1 , 443 , p < 10−16 ) . When limited to the 1062 immunoreactive proteins exhibiting at least 90% similarity between the WCV and the array , ABTs accounted for 39 of 71 significant increases in IgG binding ( 54 . 9%; a 16 . 01-fold enrichment ) , and 34 of the 991 proteins not associated with such an increase ( 3 . 43%; Fisher’s exact test , N = 1 , 062 , p < 10−16 ) . Hence this highlights not just the importance of ABTs in the immune response to the WCV , but also the failure of some ABTs to trigger significantly elevated IgG binding , despite the array being appropriately constructed to detect such a response ( Supplementary file 2 ) . Therefore , the subset of proteins exhibiting significant and reproducible increases in binding following WCV administration were functionally characterised to identify the properties associated with immunogenic proteins . The most statistically significant increase in IgG binding identified by the eBayes analysis was the only probe that did not correspond to a particular protein , but instead was an oligomer of choline-binding domains ( CBDs ) . This common motif , by which multiple pneumococcal proteins adhere to the cell surface via the cell wall polysaccharide , was found to be a protein domain common in ABTs ( Croucher et al . , 2017 ) ( Figure 4B ) . Previous data suggested such domains could be immunogenic ( Giefing et al . , 2008 ) , but the consistent rise in IgG binding observed in this trial corresponded to a small increase on a low baseline . Excluding the CBD oligomer and grouping together orthologous variants , the 112 WCV antigens identified by the eBayes and LMM analyses were found to correspond to 72 functionally distinct proteins , of which all but four could be attributed to sequences in the S . pneumoniae RM200 genome ( Supplementary file 1 and 2 ) . A multivariable analysis of the functional characteristics distinguishing the immunogenic proteins from those in the WCV not provoking an elevated IgG response was used to test whether these antigens were enriched for particular functional or structural characteristics . This did not identify CBDs as a marker of antigenicity ( Supplementary file 3 ) , suggesting CBD-binding IgG did not elevate the overall antibody response to all proteins containing this motif . Nevertheless , such antibodies could potentially explain one inconsistency , in which the LMM identified elevated IgG binding to the lytic amidase LytA , despite only a 15 aa N-terminal fragment remaining in the RM200 strain used for immunization ( Figure 1—figure supplement 1 ) . This could be attributed to the multiple CBDs of LytA on the array being the epitope recognised by vaccine-induced antibodies . Another protein associated with an elevated IgG response , pneumolysin ( CLS01670 ) , seems to have caused a measureable response despite small modifications to remove its cytolytic activity in the WCV ( Lu et al . , 2010b ) . The multivariable analysis of protein characteristics found a significant association between increased IgG binding and signal peptides , which direct proteins for export across the cell membrane ( odds ratio = 6 . 76 , p = 1 . 38×10−5; Supplementary file 3 ) . This corresponded with the strong immune responses to many surface-associated proteins . Excepting the transcriptional regulator , FrlR ( CLS00137 ) , the greatest Δ0→84 values were measured for three conserved surface-exposed proteins: a pre-protein translocase YajC ( CLS01753 ) , a protease regulator HflC ( CLS01867 ) , and the peptidyl-prolyl isomerase PpiA ( CLS00702 ) , the only one of these four proteins to register as an ABT in the pre-vaccination sample ( Figure 4B ) . Another ABT involved in maintaining surface proteins’ conformation , the foldase PpmA ( CLS00885 ) , was identified as eliciting an IgG response by the LMM analysis . Other protein motifs showing similarly strong associations in the multivariable analysis included the Transpeptidase domain , associated with peptidoglycan remodelling , and the SBP_bac_3 domain , associated with ATP-binding cassette ( ABC ) transporter solute-binding proteins ( SBPs ) , which bind exogenous substrates and deliver them to cognate permeases for import into the cell . Accordingly , a broad functional categorisation found cell wall metabolism proteins and SBPs were similarly represented in the WCV antigens as in pre-vaccination ABTs ( Croucher et al . , 2017 ) , whereas relatively few vaccine-induced responses were observed to adhesins and surface-associated degradative enzymes ( Figure 4C ) . Of the 27 SBPs in the RM200 genome , 25 were among the 1062 proteins that were immunoreactive and highly similar between the WCV strain and the panproteome array . Eleven of the 25 ( 44 . 0% ) provoked a substantial increase in IgG binding , relative to 60 of the 1037 non-SBPs , an 7 . 60-fold enrichment representing their significant contribution to the WCV response ( Fisher’s exact test , N = 1 , 062 , p = 1 . 16×10−7 ) . SBPs were enriched in the subset of ABTs triggering significant further IgG rises , accounting for eleven of the 71 ABTs ( 15 . 5%; a 3 . 54-fold enrichment ) identified as immunogenic by the eBayes or LMM analyses , but only six of the 137 ABTs ( 4 . 38% ) not associated with a significant increase ( Fisher’s exact test , N = 208 , p = 0 . 014 ) . Two of the ABT SBPs triggering a post-WCV rise in IgG binding were the siderophore transporters PiaA and PiuA , which have been considered as potential protein vaccine candidates ( Jomaa et al . , 2006 ) . Seven other immunogenic SBPs bound amino acids or peptides for import by ABC transporters . These included a large increase in IgG recognising TcyA ( CLS00206; Figure 4B ) , and rises to the glutamine-binding proteins GlnH ( CLS01210 ) , GlnPH4 ( CLS01088 ) and GlnPH1 ( CLS00459 ) , only the former two of which were classified as ABTs in the pre-vaccination data . Nevertheless , other SBPs strongly recognised by natural immunity showed no sign of increased IgG binding following vaccination ( Figure 4—figure supplement 8 ) . Other transporter proteins eliciting a significant elevation in IgG binding were FruA ( CLS00796 ) , the substrate-recognising IIC subunit of a fructose-specific phosphotransferase system importer , and protein CLS02831 ( Supplementary file 2 ) , a bacteriocin exporter present in RM200 but only found in 29% of the pneumococcal population of Massachusetts ( Croucher et al . , 2013a ) . Another bacteriocin protein to induce elevated IgG binding was the secreted CibA competence-induced bacteriocin , suggesting at least some of the pneumococci were in the ‘X state’ when killed prior to inoculation ( Claverys and Håvarstein , 2007 ) . Sixteen of the 25 previously-defined ABTs ( 64 . 0% ) involved in cell wall metabolism were among the 71 ABTs provoking a significant increase in IgG binding by at least one of the statistical analyses ( Figure 4—figure supplement 9 ) , corresponding to a 3 . 43-fold enrichment ( Fisher’s exact test , N = 208 , p = 0 . 0014 ) . Additionally , four functionally similar proteins that were not classified as ABTs on the basis of pre-vaccination immune responses elicited significantly elevated IgG binding ( Supplementary file 2 ) . As well as Pbp1B ( CLS01817 ) , these included variants of the Pbp1A protein , which , along with Pbp2B and Pbp2X , have diverged under selection for resistance to β-lactams ( Croucher et al . , 2013a ) . The penicillin-binding protein sequences in RM200 are most similar to the ancestral , β-lactam sensitive variants ( Figure 4—figure supplement 10 ) . While there were subtle differences in the immune response to different variants of all three of these proteins , both the dose response and maximum increase in IgG binding were similar for each ( Figure 4—figure supplement 11 ) . Loci exhibiting higher levels of divergence correspondingly provided greater evidence of variant-specific immune responses . Two of the adhesins to which elevated IgG binding was detected were variants of the PclA protein ( Paterson et al . , 2008 ) ( Figure 4—figure supplement 12 and Supplementary file 2 ) . These were the most similar functional variants to the PclA sequence in the S . pneumoniae RM200 genome ( Figure 4—figure supplement 13 ) , with a truncated version ( CLS99466 ) and two divergent full-length proteins ( CLS03178 and CLS03616 ) not showing similar increases ( Figure 4—figure supplement 14 ) . No increased IgG binding to pilus subunits or serine-rich glycoproteins was observed , in keeping with their absence from the RM200 genome ( Figure 4—figure supplement 14 ) . These data are consistent with both allelic divergence , and absence of antigens encoded by genomic islands , being an effective means of pneumococcal strains evading adaptive immune responses induced by colonisation with genetically-distinct strains . The proteins represented by the greatest number of variants on the array were pneumococcal surface proteins A and C , encoded by the DCL pspA and pspC . The array contained no variant highly similar to RM200’s PspA or PspC sequences . Nevertheless , a subset of the variants of each showed a significant increase in IgG binding that was most pronounced in cohort 3 , and absent from the placebo group ( Figure 4B and Figure 5A , B ) . These changes are unlikely to be non-specific responses caused by IgG recognising the CBDs of these proteins , as the variants identified as immunogenic contained a lower mean number of CBDs ( 5 . 58 for PspA , 6 . 00 for PspC ) than the average number of such domains across all variants on the array ( 6 . 97 for PspA , 6 . 17 for PspC ) . However , this induced response was also not simply related to sequence divergence across the rest of the protein . For PspA , IgG binding rose against multiple variants , consistent with previous data suggesting broadly protective anti-PspA immunity can be induced by immunisation with a single variant , despite the diversity of this protein ( Nabors et al . , 2000; Tart et al . , 1996 ) . By contrast , the response to PspC was only strongly evident to variant 42 , which had an intermediate level of sequence similarity to the PspC of RM200 compared to other variants on the array . This observation could not be attributed to cross-reactivity with PspA ( Brooks-Walter et al . , 1999 ) , as this PspC allele is not closely related to the WCV PspA . Hence this may represent a relatively poor immune response to PspC , owing to its weakened association with the cell surface following the loss of some of its CBDs , that triggers recognition of a specific epitope unique to variant 42 . Surface-exposed degradative enzymes generally elicited little increase in IgG binding ( Figure 4—figure supplement 15 ) . Two exceptions were proteins ubiquitous across encapsulated pneumococcal lineages ( Croucher et al . , 2014 ) : the β-galactosidase BgaA ( CLS00596 ) and serine protease HtrA ( CLS00066 ) ( Croucher et al . , 2017 ) ( Figure 4—figure supplement 15 ) . There was also a significant response to the zinc metalloproteases encoded by DCL , ZmpA and ZmpB ( Figure 4B and Figure 5C , D ) . The eBayes analysis only identified a significant IgG binding increase to a single ZmpB variant ( Supplementary file 2 ) , which was the sequence most closely related to that found in RM200; all other alleles exhibited >20% sequence divergence , suggesting a variant-specific immune response . By contrast , IgG binding to many ZmpA variants was increased by WCV administration ( Figure 4B and 5C ) . Consistent with the results with ZmpB , however , the strongest responses were generally associated with those variants with <20% protein sequence divergence with RM200’s ZmpA , again suggesting a simpler relationship between antibody cross-reactivity and immune responses than for PspA and PspC ( Figure 5 ) . Nevertheless , there were some highly divergent ZmpA variants associated with elevated IgG binding post-vaccination , demonstrating some epitopes had a complex distribution across the population . Plotting the changes in IgG binding between each sampling point for proteins found to elicit significantly elevated responses by either the LMM or eBayes analyses showed differences in the timing of responses between functionally-defined categories ( Figure 6 ) . For SBPs , cell wall metabolism proteins , and ZmpA variants , responses were generally maximal after a single injection , particularly for the cohort receiving the highest dose of WCV . The empirically-determined threshold of 0 . 2 was used to identify substantial changes in IgG binding between consecutive timepoints among the probes to which there was a significant overall vaccine response ( Figure 4—figure supplement 2 ) . One hundred and twelve probes increased by more than this cutoff in the 28 days after the first vaccine dose ( Figure 7 ) . These probes had a high median pre-vaccination IgG binding of 2 . 53 , and likely represent many cases of anamnestic responses being triggered . In contrast , only 27 probes detected a change of similar magnitude after the third WCV dose ( Figure 7 ) . These were associated with comparatively low pre-WCV antibody responses . The pre-vaccination IgG binding to these probes , and to the 112 probes to which a substantial rise was detected after the first dose , was compared using a Wilcoxon rank sum test . This found these late-responding probes to have a significantly lower level of pre-vaccination IgG binding ( median 1 . 43; N = 139 , W = 2116 . 5 , p = 0 . 0013 ) . These kinetics probably represent the slower development of a primary immune response to proteins not eliciting antibodies prior to WCV administration . Such reactivity was typically associated with the set of proteins with diverse functional annotations ( Supplementary file 2 ) , not falling into the main functional categories associated with natural immunity . The largest changes were for the pre-protein translocase subunit YajC and the intracellular transcriptional regulator FrlR , neither of which showed high levels of IgG binding in the pre-vaccination samples ( Supplementary file 2; Figure 6 ) , as well as the surface-exposed peptidyl-prolyl isomerase PpiA . Hence multiple doses of the WCV can elicit antibody responses to proteins not commonly recognised by pre-existing immunity . This panproteome-wide analysis of the changes in IgG binding following administration of a WCV provides a comprehensive view of the antibodies generated in response to systemic inoculation with an unencapsulated pneumococcus . The phase I clinical trial of WCV was primarily designed to measure safety outcomes , thus statistical power to assess immunogenicity at the proteome scale may have been limited . Nevertheless , complementary statistical analyses of changes in antibody response found significantly elevated IgG binding to 137 probes , corresponding to 112 protein sequences . Accounting for identification of orthologous variants , these responses spanned 72 functionally-distinct proteins , including multiple conserved antigens . These changes in immunity appear to represent a combination of boosting anamnestic responses to previously-recognised antigens , as well as the induction of novel antibodies to previously-unrecognised antigens . This complexity likely at least partially reflects the participants in this trial being healthy adults , with mature pre-existing immunity to pneumococci that is likely to limit the WCV-induced IgG response . Administration of WCV to more immunologically-naïve infants , the intended recipients of the vaccine , would likely result in a much higher proportion of novel antibody responses . The distinctive antibody fingerprints of the adults in the trial suggest humans’ mature antibody repertoire is shaped by multiple encounters with diverse pneumococci , as the overall modifications caused by three WCV inoculations were relatively small . However , these alterations were remarkably consistent between individuals , despite the pre-existing differences in their adaptive immune responses . Only a subset of proteins , enriched for naturally immunogenic ABTs , provoked a strong post-WCV response . The reasons for this heterogeneity are unclear . These differences in Δ0→84 measurements were evident even between proteins with similar functional and structural characteristics ( Figure 4—figure supplement 8 & Figure 4—figure supplement 9 ) , or of similar immunogenicities , based on the pre-WCV samples ( Figure 3 ) . This pattern is unlikely to represent a lack of sensitivity due to saturation of the system , occurring at IgG-binding levels above six in this study , which were not frequently reached in these data . Nor is it likely attributable to sequence divergence between RM200 and protein representatives on the array , given probes’ high sequence similarity to some ABTs encoded in the RM200 genome that failed to elicit a substantial IgG response ( Supplementary file 1 ) . This suggests relatively subtle differences in protein characteristics , or antibody response kinetics , could affect which proteins are recognised most strongly when presented in the WCV . Alternatively , it may relate to differences in protein concentrations between the WCV and live pneumococci encountered by the host . However , it would seem unlikely that chemical killing , and subsequent processing , of the RM200 cells would retain a subset of each of the immunogenic SBPs , attached to the membrane through a lipid moiety , as well as surface proteins PspA and PspC , attached to cell wall polysaccharide through choline-binding domains , and ZmpA , covalently attached to the cell wall via sortase activity . It is more conceivable that many ABTs may not have been expressed by the RM200 cells during culture , although the strong and apparently novel response to some predominantly intracellular proteins , such as FlrR , suggests expression of surface-associated proteins would have to be very low for lack of exposure to explain the absence of an immune response . The surface-associated proteins enriched among WCV antigens may be important components of a protective IgG response , as recognition of these structures enables intact S . pneumoniae cells to be agglutinated ( Mitsi et al . , 2017 ) or targeted for opsonophagocytosis ( Hyams et al . , 2010 ) . However , the high-throughput nature of the array measurements of antibody binding cannot currently substitute for functional assays in determining whether these IgG responses are sufficient for protection . It is not clear whether a threshold correlate of protection , as determined for anti-capsular antibodies induced by PCVs ( Andrews et al . , 2014 ) , could be applicable to a broad range of distinct protein antigens . One potentially confounding factor would be the accessibility of the different surface antigens on encapsulated strains , which suggests another explanation for the pattern of immune responses . Antibody binding of SBPs and components of the cell wall synthesis machinery is impeded by the capsule of most S . pneumoniae ( Gor et al . , 2005 ) . These proteins accounted for much of the increased IgG binding following administration of the WCV . By contrast , adhesins and surface-associated degradative enzymes must extend beyond the capsule to function , and were comparatively absent from the antigens triggering post-WCV increases in IgG binding . Hence the observed enrichment of the former functional categories in the immune response to the vaccine could represent IgG recognition of the latter , larger surface proteins already being maximal in healthy adults , a situation that would be less likely in children receiving the same vaccine . The opportunity for the adaptive immune system to strengthen responses to SBPs and the cell wall machinery may be particularly valuable , as many of these antigens are conserved across the population ( Croucher et al . , 2017 ) . Other subcapsular , conserved surface ABTs , such as the PpmA and PpiA foldases , were also associated with strong IgG responses . This is consistent with previous work with sera from animals inoculated with the RM200 WCV showing antibody-binding pneumococci of different serotypes ( Campos et al . , 2017; Gonçalves et al . , 2014 ) . Therefore , these antibodies could afford broad protection against a diverse set of strains , as long as these antibodies can penetrate the capsular envelope to bind their cognate proteins . They would likely be more effective against transparent phase variants , in which lower levels of capsule are expressed ( Weiser et al . , 1994 ) . These are more commonly associated with colonising isolates adhering to the nasopharyngeal epithelium , rather than those causing disease . The few strong responses that were observed to adhesins , such as PclA and PspA , and degradative enzymes , such as ZmpA , may reflect exposure to new variants of diverse proteins that an individual has not encountered previously . In each of these cases , there was evidence of some variant-specificity in the response , although this was not simply related to sequence similarity between the proteins in the WCV and the variants on the array , suggesting these proteins contain multiple epitopes with a complex distribution across orthologues . For ZmpA , ZmpB and PclA , heterogeneity in the pattern of IgG binding suggests the evident sequence divergence is enough to at least weaken , if not completely evade , IgG binding induced by other variants , consistent with adaptive immunity being an important driver of the divergence of these antigenic loci ( Croucher et al . , 2017; Li et al . , 2012 ) . The contrasting binding of multiple PspA variants by WCV-induced antibodies could be an artefact of the absence of a variant on the array that was closely-related to the PspA of RM200 , such that the measured responses are all similarly high because all representatives on the array are relatively distant from the variant that stimulated the response . Some previous work has found antibodies induced by individual PspA or PspC representatives to be broadly protective against a wide spectrum of variants ( Brooks-Walter et al . , 1999; Nabors et al . , 2000; Tart et al . , 1996 ) , consistent with induction of genuinely cross-reactive IgG , although other studies have found evidence for greater specificity in anti-PspC responses ( Georgieva et al . , 2018 ) . Therefore , inducing antibodies capable of recognising multiple variants of DCL could contribute to the protection induced by this WCV formulation being effective across the species . Although this study focuses on the humoral response to WCV , cellular responses mediated by CD4+ T cells and interleukin 17A are important in the vaccine’s role in inhibiting nasopharyngeal colonisation in a mouse model ( Campos et al . , 2017; Lu et al . , 2010a ) . Nevertheless , in the same animal model , WCV-induced protection against invasive disease was more dependent on protective antibodies than cellular immunity ( Lu et al . , 2010a ) . In humans , individuals with agammaglobulinemia , who have low levels of circulating antibodies , are highly susceptible to pneumococcal disease , atesting to the importance of protective antibody-mediated humoral immunity ( Conley et al . , 2000 ) . While anticapsular antibodies are important in protecting against pneumococcal colonisation and disease , there is evidence that antibodies recognising proteins may be effective in these roles as well , based on their diversification that indicates immune selection ( Croucher et al . , 2017; Li et al . , 2012 ) , the effects of protein vaccines in various animal models ( Briles et al . , 2000; Gor et al . , 2005; Jomaa et al . , 2005 ) , and the clinical effectiveness of protein-binding antibodies in intravenous immunoglobulin preparations used to treat pneumococcal disease ( Wilson et al . , 2017 ) . These data provide new information on how this critically important pneumococcal antibody repertoire develops , both specifically relating to systemic immunisation with WCV , but likely also serving as a model of how humoral immunity responds to natural exposure to pneumococci . Such routine contacts with the bacterium are common and likely to have been somewhat confounding in this study , as the data in Figure 5 suggest some individuals in the placebo group may have encountered a naturally-circulating pneumococcus with different DCL alleles , and thus mounted a different pattern of IgG responses to the panel of variants . Despite such strain-specificity , there are many strong responses to the WCV that are consistent across individuals , congruent with the generally similar patterns of pre-vaccination immunity in this study ( Croucher et al . , 2017 ) and the comparable profiles evident in pooled immunoglobulins collected from different countries ( Wilson et al . , 2017 ) . This suggests the development of adaptive immunity is similar between individuals , and therefore these data are not just informative about the mechanism by which WCV may provide protection against pneumococcal disease , but also help build a more general understanding of the development of the natural immune repertoire , and what consequences this has for pneumococcal biology . A draft genome of S . pneumoniae RM200 was generated through a combined assembly of 454 and Illumina sequencing data using Celera Assembler v6 . 1 ( Myers et al . , 2000 ) . These contigs were aligned to the S . pneumoniae D39 reference genome ( accession code: CP000410 ) ( Lanie et al . , 2007 ) , and one cut at the origin of replication , using ACT v13 . 0 . 0 ( Carver et al . , 2008 ) . The resulting annotated assembly has been deposited in the European Nucleotide Archive with the sample accession code ERS2169631 . The genomes of D39 and RM200 were then aligned using MAFFT v7 . 221 with default settings ( Katoh and Standley , 2013 ) . Base substitutions and recombinations distinguishing the two sequences were identified with Gubbins v1 . 4 . 10 ( Croucher et al . , 2015a ) . Pairwise comparisons with S . pneumoniae D39 and OXC141 ( Croucher et al . , 2013b ) ( accession code: FQ312027 ) were performed with BLASTN ( Camacho et al . , 2009 ) and ACT . The COGsoft package was used to link the proteome of individual genomes to the probes on the array ( Kristensen et al . , 2010 ) . The coding sequences of S . pneumoniae RM200 , S . pseudopneumoniae IS7493 ( accession code: CP002925 ) , S . mitis B6 ( accession code: FN568063 ) and Streptococcus mutans UA159 ( accession code: AE014133 ) were identified using the methods described previously ( Corander et al . , 2017 ) . The proteins were then aligned to those from a collection of 616 genomes from Massachusetts ( Croucher et al . , 2015b ) , used to design the proteome array ( Croucher et al . , 2017 ) , using BLASTP ( Kent , 2002 ) . These comparisons were then processed using COGcognitor ( Kristensen et al . , 2010 ) . For each of the probes on the array , every protein matching the same COG in a given genome was aligned to the protein used to design the probe using MAFFT ( Katoh and Standley , 2013 ) , and the maximal sequence identity recorded ( Supplementary file 1 ) . S . mutans proteins were sufficiently divergent from those in S . pneumoniae to suggest conservation should be defined only using the mitis group species , S . mitis and S . pseudopneumoniae . An analogous COGsoft analysis was also used to link the proteins on the array to the genome of S . pneumoniae D39 to add the functional annotation to Table 1 ( Lanie et al . , 2007 ) . Solute-binding proteins were identified by using hmmscan v3 . 1 ( Eddy , 2011 ) to search the S . pneumoniae RM200 proteome using the relevant Pfam domains , identified by keyword searches: SBP_bac_1 ( accession code PF01547 . 24 ) , SBP_bac_3 ( accession code PF00497 . 19 ) , SBP_bac_5 ( accession code PF00496 . 21 ) , SBP_bac_6 ( accession code PF13343 . 5 ) , SBP_bac_8 ( accession code PF13416 . 5 ) , SBP_bac_10 ( accession code PF07596 . 10 ) , SBP_bac_11 ( accession code PF13531 . 5 ) , Peripla_BP_1 ( accession code PF00532 . 20 ) , Peripla_BP_2 ( accession code PF01497 . 17 ) , Peripla_BP_3 ( accession code PF13377 . 5 ) , Peripla_BP_4 ( accession code PF13407 . 5 ) , Peripla_BP_5 ( accession code PF13433 . 5 ) , Peripla_BP_6 ( accession code PF13458 . 5 ) , ABC_sub_bind ( accession code PF04392 . 11 ) , Bmp ( accession code PF02608 . 13 ) , DctP ( accession code PF03480 . 12 ) and ZnuA ( accession code PF01297 . 16 ) . AliA was not identified by this analysis , but was included as an SBP as it had been identified as such in a previous analysis ( Croucher et al . , 2017 ) . For the PspA , PspC , ZmpA and ZmpB variants , sequence identity values were calculated by aligning the full-length proteins on the proteome array with the corresponding allele in RM200 using MAFFT ( Katoh and Standley , 2013 ) . For the penicillin-binding protein and PclA variants , the proteins were again aligned with MAFFT , and the phylogenies generated with FastTree2 ( Price et al . , 2010 ) . The VAC-002 phase one study ( ClinicalTrials . gov identifier: NCT01537185 ) was approved by the Western Institutional Review Board and conducted in compliance with the study protocol , international standards of Good Clinical Practice and the Declaration of Helsinki . Participants were healthy adults aged 18 to 40 years at the time of consent , and had no evidence of chronic health issues , nor any history of invasive pneumococcal disease or pneumococcal vaccination . Forty-two participants were enrolled and either received a WCV dose or a saline placebo , with sequential subject assignment performed by data management using an electronic randomization block design . Pharmacy staff responsible for preparing and administering vaccinations were unblinded . All others involved in conducting the trial , including participants , remained blinded to treatment assignment . This study is a post hoc analysis of the samples available at the end of the clinical trial , and does not present the original safety and immunogenicity tests for which the trial was designed . No specific power analysis was conducted prior to the trial , or the sample selection for this subsequent study . The Streptococcus pneumoniae Pan-Genome Microarray , produced by Antigen Discovery , Inc . ( ADI , Irvine , CA , USA ) , was designed and assayed as described previously ( Croucher et al . , 2017 ) . Briefly , the microarray included 4504 full-length or fragmented proteins from the TIGR4 reference genome ( 2106 genes ) and representatives of 2190 proteins identified in the Massachusetts pneumococcal population ( Croucher et al . , 2013b ) . These correspond to 2055 clusters of orthologous genes ( COGs ) , 36 PspA variants , 57 PspC variants , 18 ZmpA variants , 16 ZmpB variants , and individual sequences for LytA , a phage amidase , ZmpE , PblB , PsrP and a choline-binding domain oligomer . Proteome microarrays were fabricated using a library of partial or complete CDSs cloned into the T7 expression vector pXI . Proteins were expressed using an E . coli in vitro transcription and translation ( IVTT ) system ( Rapid Translation System , 5 Prime , Gaithersburg , MD , USA ) and printed onto nitrocellulose-coated glass AVID slides ( Grace Bio-Labs , Inc . , Bend , OR , USA ) using an Omni Grid Accent robotic microarray printer ( Digilabs , Inc . , Marlborough , MA , USA ) . Microarrays were probed with sera and antibody binding detected by incubation with biotin-conjugated goat anti-human IgG ( Jackson ImmunoResearch , West Grove , PA , USA ) , followed by incubation with streptavidin-conjugated SureLight P-3 ( Columbia Biosciences , Frederick , MD , USA ) . Slides were scanned on a GenePix 4300A High-Resolution Microarray Scanner ( Molecular Devices , Sunnyvale , CA , USA ) , and raw spot and local background fluorescence intensities , spot annotations and sample phenotypes were imported and merged in R ( R Core Team , 2017 ) , in which all subsequent procedures were performed . Foreground spot intensities were adjusted by local background by subtraction , and negative values were converted to one . All foreground values were transformed using the base two logarithm . The dataset was normalised to remove systematic effects by subtracting the median signal intensity of the IVTT controls for each sample . With the normalised data , a value of 0 . 0 means that the intensity is no different than the background , and a value of 1 . 0 indicates a doubling with respect to background . Values below −2 in the normalised data , corresponding to less than 0 . 25 of the IVTT control probe signals , were adjusted to −2 . This affected 152 of the 304 , 590 binding values included in the study , of which a further five were missing . Immunoreactivity was defined as achieving an IgG-binding level of one or greater in the study population at any timepoint being analysed . Change in antibody levels ( Δ0→84 ) was calculated as IgG-binding levels from post-vaccination ( day 84 ) minus pre-vaccination ( day 0 ) IgG-binding levels . The t-SNE analyses were calculated after 25 , 000 iterations with a perplexity parameter of 40 using the R package Rtsne ( Krijthe , 2015 ) . Empirical Bayes ( eBayes ) analyses were conducted with the R package limma ( Ritchie et al . , 2015 ) . The eBayes analysis was designed as pairwise contrasts of Δ0→84 values from each of the vaccinated cohorts against those from the placebo group , with no intercept in the model . Only probes that exhibited immunoreactivity , as defined previously , were used to avoid reporting low level changes in IgG binding that would likely not be of biological relevance . Only the 29 trial participants with samples at both day zero and day 84 were included in this analysis . The 1584 p values associated with all immunoreactive probes , or the 1384 non-DCL immunoreactive probes , were subject to a Benjamini-Hochberg correction ( Benjamini and Hochberg , 1995 ) using the limma function topTable , and reported as significant if they were below the false discovery rate threshold of 0 . 05 . The linear mixed effects models were fitted using the R package lme4 ( Bates et al . , 2015 ) . To test for the relationship between cohort and Δ0→84 , a model was fitted to all probes , or immunoreactive probes only , using the Δ0→84 from the 29 trial participants with samples at both day zero and day 84 . The model had the form:Δ0→84 , p=Xcc+Xpp+ZS+ε Where Δ0→84 , p is the change in IgG binding to a probe p over the duration of the trial; Xc is the fixed effect of an individual’s cohort , c; Xp is the fixed effect of measuring IgG binding using probe p; Zs is the random effect associated with each trial participant s , and ε is an error parameter . The likelihood of Xc being non-zero was calculated through fitting an identical model without the Xcc term , and comparing the likelihood ratio of the two models with an ANOVA test . The per-probe model fits used the immunoreactive probe data from 20 trial participants associated with samples at each of the four timepoints in one of the three vaccinated cohorts . The normalised IgG-binding values for a given probe were then fitted to a model of the form:ip , t=Xdd+Xtt+ZS+ε Where ip , t is the IgG binding to a probe p at time t ( measured as timepoints one to four ) ; Xt is the fixed effect of increasing timepoint on IgG binding; Xd is the fixed effect of vaccine dose on IgG binding , with d corresponding to the dose in micrograms; Zs is the random effect associated with each trial participant s , and ε is an error parameter . The likelihood of Xd being non-zero was calculated through fitting an identical model without the Xdd term , and comparing the likelihood ratio of the two models with an ANOVA test . The same approach was used to calculate the likelihood of Xt being non-zero . In each case , the 1584 likelihood values associated with all immunoreactive probes , or the 1384 non-DCL immunoreactive probes , were subject to a Benjamini-Hochberg correction using the R function p . adjust , and reported as significant if they were below the false discovery rate threshold of 0 . 05 . To identify those features of proteins that were associated with vaccine-induced IgG responses , a generalised linear model was fitted to the outcome , represented by a binary variable denoting whether a protein encoded a probe that was associated with a significant elevation in IgG binding by either the eBayes or LMM analyses ( Supplementary file 2 ) , and the explanatory variables , which described protein features . These included a continuous variable representing the mean length of the coding sequences in the corresponding cluster of orthologous genes ( Croucher et al . , 2013a ) , and binary variables corresponding to the presence of domains , identified by Pfam ( Punta et al . , 2012 ) ; a signal peptide , identified by SignalP ( Petersen et al . , 2011 ) ; transmembrane helices , identified by TMHMM ( Krogh et al . , 2001 ) ; and a lipoprotein processing motif , identified by Prosite ( Sigrist et al . , 2013 ) . The proteins included in the analysis were the 1605 for which at least 90% sequence identity between the array and RM200 proteins were observed ( Supplementary file 1 ) ; once the penicillin-binding protein and DCL variants were replaced with individual representatives of the corresponding loci , this left a final dataset of 1600 proteins . Of these , 64 were deemed to be linked to an increase in IgG binding . The features included in the analysis were limited to those present in at least five proteins remaining in this dataset . The analysis was run in R ( R Core Team , 2017 ) using the safeBinaryRegression package ( Konis , 2013 ) until convergence; the stepAIC function of the MASS package ( Venables and Ripley , 2002 ) was then used to refine the model , allowing the stepwise search to run in both directions , to produce the results shown in Supplementary file 3 . For the functional categorisation shown in Figure 4C , the annotation based on pre-vaccination data was used to describe ABTs ( Croucher et al . , 2017 ) , and the set of proteins identified by either the LMM or eBayes analyses were used for WCV antigens . For both datasets , highly similar alleles split on functional information , such as the penicillin-binding proteins , were merged into single entries , whereas divergent alleles identified as separate COGs , such as the PclA variants , were kept as individual datapoints . The four DCL were each treated as a single set of orthologues that were immunogenic in both sets of proteins . Such classification resulted in datasets of 98 pre-vaccination ABTs and 74 WCV antigens , in a manner that was consistent between these groupings .	Streptococcus pneumoniae is a bug that causes pneumonia and meningitis , killing around a million people each year . Vaccines now exist to protect young children against these diseases , but they are expensive and do not work against all the strains of the bacteria . This is because these shots train the body’s immune system to recognize and attack the bacterium’s capsule , a layer of sugars that surrounds the microbe and is often different between strains . One possible solution could be a cheap , whole cell vaccine . These injections expose the body to genetically modified S . pneumoniae that do not carry the capsule . Such treatment has now been tested in a small number of people during a clinical trial . Here , Campo et al . use a technique known as panproteome array to scan samples collected during this trial , and identify which elements the body learns to recognize when it is exposed to the genetically manipulated strain of S . pneumoniae . The results show that when volunteers receive this vaccine , their body targets proteins that the capsule normally shields from the immune system . Many of these proteins are very similar across all strains of S . pneumoniae , which means that the whole cell vaccine could potentially better protect against a broad spectrum of bacteria . However , further studies are needed to assess whether this is the case , especially in infants .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2018	Panproteome-wide analysis of antibody responses to whole cell pneumococcal vaccination
DNA is a remarkably precise medium for copying and storing biological information . This high fidelity results from the action of hundreds of genes involved in replication , proofreading , and damage repair . Evolutionary theory suggests that in such a system , selection has limited ability to remove genetic variants that change mutation rates by small amounts or in specific sequence contexts . Consistent with this , using SNV variation as a proxy for mutational input , we report here that mutational spectra differ substantially among species , human continental groups and even some closely related populations . Close examination of one signal , an increased TCC→TTC mutation rate in Europeans , indicates a burst of mutations from about 15 , 000 to 2000 years ago , perhaps due to the appearance , drift , and ultimate elimination of a genetic modifier of mutation rate . Our results suggest that mutation rates can evolve markedly over short evolutionary timescales and suggest the possibility of mapping mutational modifiers . Germline mutations not only provide the raw material for evolution but also generate genetic load and inherited disease . Indeed , the vast majority of mutations that affect fitness are deleterious , and hence biological systems have evolved elaborate mechanisms for accurate DNA replication and repair of diverse types of spontaneous damage . Due to the combined action of hundreds of genes , mutation rates are extremely low–in humans , about one point mutation per 100 MB or about 60 genome-wide per generation ( Kong et al . , 2012; Ségurel et al . , 2014 ) . While the precise roles of most of the relevant genes have not been fully elucidated , research on somatic mutations in cancer has shown that defects in particular genes can lead to increased mutation rates within very specific sequence contexts ( Alexandrov et al . , 2013; Helleday et al . , 2014 ) . For example , mutations in the proofreading exonuclease domain of DNA polymerase ϵ cause TCT→TAT and TCG→TTG mutations on the leading DNA strand ( Shinbrot et al . , 2014 ) . Mutational shifts of this kind have been referred to as ‘mutational signatures’ . Specific signatures may also be caused by nongenetic factors such as chemical mutagens , UV damage , or guanine oxidation ( Ohno et al . , 2014 ) . Together , these observations imply a high degree of specialization of individual genes involved in DNA proofreading and repair . While the repair system has evolved to be extremely accurate overall , theory suggests that in such a system , natural selection may have limited ability to fine-tune the efficacy of individual genes ( Lynch , 2011; Sung et al . , 2012 ) . If a variant in a repair gene increases or decreases the overall mutation rate by a small amount–for example , only in a very specific sequence context–then the net effect on fitness may fall below the threshold at which natural selection is effective . ( Drift tends to dominate selection when the change in fitness is less than the inverse of effective population size ) . The limits of selection on mutation rate modifiers are especially acute in recombining organisms such as humans because a variant that increases the mutation rate can recombine away from deleterious mutations it generates elsewhere in the genome . Given these theoretical predictions , we hypothesized that there may be substantial scope for modifiers of mutation rates to segregate within human populations , or between closely related species . Most triplet sequence contexts have mutation rates that vary across the evolutionary tree of mammals ( Hwang and Green , 2004 ) , but evolution of the mutation spectrum over short time scales has been less well described . Weak natural mutators have recently been observed in yeast ( Bui et al . , 2017 ) and inferred from human haplotype data ( Seoighe and Scally , 2017 ) ; if such mutators affect specific pathways of proofreading or repair , then we may expect shifts in the abundance of mutations within particular sequence contexts . Indeed , one of us has recently identified a candidate signal of this type , namely an increase in TCC→TTC transitions in Europeans , relative to other populations ( Harris , 2015 ) ; this was recently replicated ( Mathieson and Reich , 2016 ) . Here , we show that mutation spectrum change is much more widespread than these initial studies suggested: although the TCC→TTC rate increase in Europeans was unusually dramatic , smaller scale changes are so commonplace that almost every great ape species and human continental group has its own distinctive mutational spectrum . One possible concern is that batch effects or other sequencing artifacts might contribute to differences in mutational spectra . Therefore we replicated our analysis using 201 genomes from the Simons Genome Diversity Project ( Mallick et al . , 2016 ) . The SGDP genomes were sequenced at high coverage , independently from 1000 Genomes , using an almost non-overlapping panel of samples . We found extremely strong agreement between the mutational shifts in the two data sets ( Figure 2 ) . For example , all of the 43 mutation types with a significant difference between Africa and Europe ( at p<10−5 ) in 1000 Genomes also show a frequency difference in the same direction in SGDP ( comparing Africa and West Eurasia ) . In both 1000 Genomes and SGDP , the enrichment of AC→CC mutations in East Asia is larger in magnitude than any other signal aside from the previously described TCC→TTC imbalance . 10 . 7554/eLife . 24284 . 013Figure 2 . Concordance of mutational shifts in 1000 Genomes versus SGDP . Each panel shows natural-log mutation spectrum ratios between a pair of continental groups , based on 1000 Genomes ( x-axis ) and SGDP ( y-axis ) data . Data points encoded by ( + ) symbols denote mutation types that are not significantly enriched in either population in the Figure 1 1000 Genomes analysis ( p<10−5 ) . These heatmaps use the same labeling and color scale as in Figure 1 . All 1000 Genomes ratios in this figure were estimated after projecting the 1000 Genomes site frequency spectrum down to the sample size of SGDP . See Figure 2—figure supplements 1 and 2 for a complete set of SGDP heatmaps and regressions versus 1000 Genomes . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01310 . 7554/eLife . 24284 . 014Figure 2—figure supplement 1 . Heatmap comparisons between continental groups in 1000 Genomes and the SGDP . Here , each 1000 Genomes population is projected down to the sample size of the corresponding SGDP population in order to sample alleles with a similar distribution of ages and frequencies . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01410 . 7554/eLife . 24284 . 015Figure 2—figure supplement 2 . Regression of the SGDP heatmap coefficients versus the corresponding 1000 Genomes heatmap coefficients . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 015 The greatest discrepancies between 1000 Genomes and SGDP involve transversions at CpG sites , which are among the rarest mutational classes . These discrepancies might result from data processing differences or random sampling variation , but might also reflect differences in the fine-scale ethnic composition of the two panels . To investigate the timescale over which the mutation spectrum change occurred , we analyzed the allele frequency distribution of TCC→TTC mutations , which are highly enriched in Europeans ( Figure 3A; p<1×10−300 for Europe vs . Africa ) and to a lesser extent in South Asians . We calculated allele frequencies both in 1000 Genomes and in the larger UK10K genome panel ( Walter et al . , 2015 ) . As expected for a signal that is primarily European , we found particular enrichment of these mutations at low frequencies . But surprisingly , the enrichment peaks around 0 . 6% frequency in UK10K , and there is practically no enrichment among the very lowest frequency variants ( Figure 3B and Figure 3—figure supplement 1 ) . C→T mutations on other backgrounds , namely within TCT , CCC and ACC contexts , are also enriched in Europe and South Asia and show a similar enrichment around 0 . 6% frequency that declines among rarer variants ( Figure 3C ) . This suggests that these four mutation types comprise the signature of a single mutational pulse that is no longer active . No other mutation types show such a pulse-like distribution in UK10K , although several types show evidence of monotonic rate change over time ( Figure 3—figure supplements 3 , 4 and 5 ) . 10 . 7554/eLife . 24284 . 016Figure 3 . Geographic distribution and age of the TCC mutation pulse . ( A ) Observed frequencies of TCC→TTC variants in 1000 Genomes populations . ( B ) Fraction of TCC→TTC variants as a function of allele frequency in different samples indicates that these peak around 1% . See Figure 3—figure supplement 1 for distributions of TCC→TTC allele frequency within all 1000 Genomes populations , and see Figure 3—figure supplement 2 for the replication of this result in the Exome Aggregation Consortium Data . In the UK10K data , which has the largest sample size , the peak occurs at 0 . 6% allele frequency . ( C ) Other enriched C→T mutations with similar context also peak at 0 . 6% frequency in UK10K . See Figure 3—figure supplements 3 , 4 and 5 for labeled allele frequency distributions of all 96 mutation types ( most represented here as unlabeled grey lines ) . See Figure 3—figure supplement 6 for heatmap comparisons of the 1000 Genomes populations partitioned by allele frequency , which provide a different view of these patterns . ( D ) A population genetic model supports a pulse of TCC→TTC mutations from 15 , 000 to 2000 years ago . Inset shows the observed and predicted frequency distributions of this mutation under the inferred model . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01610 . 7554/eLife . 24284 . 017Figure 3—figure supplement 1 . TCC→TTC mutation fraction as a function of allele frequency in all 1000 Genomes populations . To enable better comparison with the 1000 Genomes data , the UK10K SNPs have been downsampled to 200 individuals . The age distribution of alleles of a given frequency varies as a function of the number of lineages being sampled–this is why the UK10K pulse peaks around 0 . 6% frequency when measured in a dataset of thousands of lineages , but peaks around 2% in a subsample of only 400 lineages . Some African and East Asian population names have been omitted for clarity since the TCC→TTC mutation fraction is so uniform within these continental groups . Red = European populations; Blue = South Asian; Orange = Americas; Purple = Africa; Green = East Asia . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01710 . 7554/eLife . 24284 . 018Figure 3—figure supplement 2 . Fraction of TCC→TTC mutations as a function of allele frequency in ExAC . Lek et al . compiled data from 60 , 706 exomes to create the Exome Aggregation Consortium dataset , which enables the analysis of ultra-rare human variation ( Lek et al . , 2016 ) . The overall fraction of TCC→TTC mutations is slightly higher in exome data than in whole genome data because exons contain a skewed distribution of triplet contexts , but the pulse pattern from Figure 3B reproduces unmistakably . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01810 . 7554/eLife . 24284 . 019Figure 3—figure supplement 3 . Mutation type enrichment as a function of allele frequency in UK10K ( Part I of III ) . The eleven panels in Figure 3—figure supplements 2 , 3 and 4 show the full dependence of mutation spectrum on allele frequency in the UK10K data . If we let F⁢ ( f , m ) denote the fraction of SNVs of frequency f that are of type m and let F⁢ ( m ) denote the fraction of all mutations that are of type m , the enrichment of mutation type m as a function of frequency is F⁢ ( f , m ) /F⁢ ( m ) . This function is expected to fluctuate around y=1 unless the rate of m has recently increased or decreased . All 96 mutation types are visualized in every panel , but most corresponding lines are greyed out to enhance readability . Some lines deviate from y=1 due to the effects of biased gene conversion ( BGC ) –this occurs when one of the ancestral or derived alleles is a weak base ( A or T , abbreviated W ) and the other allele is a strong base ( G or C , abbreviated S ) . W→S mutations are more abundant at high allele frequencies , while S→W mutations are more abundant at low frequencies . These effects are visible but modest in panels D , G , H , and I , but much more pronounced in panels B , C , and F , which focus on mutations in the CpG context . Transitions of the type CpA→CpG , which create CpG motifs , are extremely enriched at high frequencies , and this pattern may be an artifact of ancestral misidentification ( Hernandez et al . , 2007 ) . CpG motifs have such high mutation rates that CpG→CpT transitions often happen at the same site in humans and chimps , and these low-frequency double mutations are misclassified as high-frequency CpT→CpG mutations . Although it is not surprising to see a peak of CpT→CpG transitions at high frequencies in panel F , it is somewhat surprising to see CpG→GpG transversions peak in abundance at high frequencies in panel C . This might be a signature of recent declines in the rates of these mutations , since neither ancestral misidentification nor biased gene conversion is thought to produce such a pattern . In addition , neither of these processes can explain the strong enrichment of certain A→T mutations at high frequencies that is observed in panel K . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 01910 . 7554/eLife . 24284 . 020Figure 3—figure supplement 4 . Mutation type enrichment as a function of allele frequency in UK10K ( Part II of III ) . The eleven panels in this three-part figure show the full dependence of mutation spectrum on allele frequency in the UK10K data . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02010 . 7554/eLife . 24284 . 021Figure 3—figure supplement 5 . Mutation type enrichment as a function of allele frequency in UK10K ( Part III of III ) . The 11 panels in this three-part figure show the full dependence of mutation spectrum on allele frequency in the UK10K data . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02110 . 7554/eLife . 24284 . 022Figure 3—figure supplement 6 . Mutation spectrum comparisons partitioned by allele frequency . Each of these heatmaps shows a subset of the data used to construct Figure 1B , partitioned by allele frequency to show how rare variants are the most highly differentiated between populations . Black dots highlight mutation types that are significantly different in abundance between two populations in a particular frequency class at the p<10−5 level according to a chi-square test . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 022 We used the enrichment of TCC→TTC mutations as a function of allele frequency to estimate when this mutation pulse was active . Assuming a simple piecewise-constant model , we infer that the rate of TCC→TTC mutations increased dramatically ∼15 , 000 years ago and decreased again ∼2000 years ago . This time-range is consistent with results showing this signal in a pair of prehistoric European samples from 7000 and 8000 years ago , respectively ( Mathieson and Reich , 2016 ) . We hypothesize that this mutation pulse may have been caused by a mutator allele that drifted up in frequency starting 15 , 000 years ago , but that is now rare or absent from present day populations . Although low frequency allele calls often contain a higher proportion of base calling errors than higher frequency allele calls do , it is not plausible that base-calling errors could be responsible for the pulse we have described . In the UK10K data , a minor allele present at 0 . 6% frequency corresponds to a derived allele that is present in 23 out of 3854 sampled haplotypes and supported by 80 short reads on average ( assuming 7x coverage per individual ) . When independently generated datasets of different sizes are projected down to the same sample size , the TCC→TTC pulse spans the same range of allele frequencies in both datasets ( Figure 3—figure supplements 1 and 2 ) , which would not be the case if the shape of the curve were a function of low-frequency errors . Encouraged by these results , we sought to find other signatures of recent mutation pulses . We generated heatmaps and PCA plots of mutation spectrum variation within each continental group , looking for fine-scale differences between closely related populations ( Figure 4 and Figure 4—figure supplement 1 through 6 ) . In some cases , mutational spectra differ even between very closely related populations . For example , the AC→CC mutations with elevated rates in East Asia appear to be distributed heterogeneously within that group , with most of the load carried by a subset of the Japanese individuals . These individuals also have elevated rates of ACA→AAA and TAT→TTT mutations ( Figure 4A and Figure 4—figure supplement 4 ) . This signature appears to be present in only a handful of Chinese individuals , and no Kinh or Dai individuals . As seen for the European TCC mutation , the enrichment of these mutation types peaks at low frequencies , that is , ∼1% . Given the availability of only 200 Japanese individuals in 1000 Genomes , it is hard to say whether the true peak is at a frequency much lower than 1% . 10 . 7554/eLife . 24284 . 023Figure 4 . Mutational variation among east Asian populations . ( A ) PCA of east Asian samples from 1000 Genomes , based on the relative proportions of each of the 96 mutational types . See Figure 4—figure supplement 2 through 6 for other finescale population PCAs . ( B ) Heatmaps showing , for pairs of east Asian samples , the ratio of the proportions of SNVs in each of the 96 mutational types . Points indicate significant contrasts at p <10−5 . See Figure 4—figure supplement 1 for additional finescale heatmaps . ( C ) and ( D ) Relative enrichment of each mutational type in Japanese and Dai , respectively as a function of allele frequency . Six mutation types that are enriched in JPT are indicated . Populations: CDX=Dai , CHB=Han ( Beijing ) ; CHS=Han ( south China ) ; KHV=Kinh; JPT=Japanese . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02310 . 7554/eLife . 24284 . 024Figure 4—source data 1 . This text file shows the number of SNPs in each of the 96 mutational categories that passed all filters in each finescale 1000 Genomes population . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02410 . 7554/eLife . 24284 . 025Figure 4—figure supplement 1 . Mutation spectrum differences within Africa , Europe , East Asia , and South Asia . Figure 4B of the main text shows heat map comparisons between East Asian populations , which display fine-scale differences that are exceptionally well defined . For completeness , this figure shows finescale heatmap comparisons within all 1 kG continental groups . We can see that CAC→CCC and TAT→TTT are heterogeneously distributed within multiple continents , but to the greatest extent in East Asia . In addition , the TCC→TTC signature is somewhat heterogeneously distributed within Europe and South Asia , being depleted in Finns and enriched in the Punjabi and Gujarati . Each continental group in the 1000 Genomes data is divided into five sub-populations . These heat maps compare the mutation spectra of these fine-scale populations to each other . African populations are: MSL = Mende in Sierra Leone; LWK = Luhya in Webuye , Kenya; YRI = Yoruba in Ibadan , Nigeria; GWD = Gambian in Western Divisions; ESN = Esan in Nigeria . European populations are: IBS = Iberian Population in Spain; TSI = Toscani in Italia; GBR = British in England and Scotland; CEU = Utah Residents ( CEPH ) with Northern and Western Ancestry; FIN = Finnish in Finland . East Asian populations are: CDX = Chinese Dai in Xishuangbanna , China; JPT = Japanese in Tokyo , Japan; CHB = Han Chinese in Bejing , China; CHS = Southern Han Chinese; KHV = Kinh in Ho Chi Minh City , Vietnam . South Asian populations are: ITU = Indian Telugu from the UK; GIH = Gujarati Indian from Houston , Texas; PJL = Punjabi from Lahore , Pakistan; BEB = Bengali from Bangladesh; STU = Sri Lankan Tamil from the UK . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02510 . 7554/eLife . 24284 . 026Figure 4—figure supplement 2 . PCA of American populations . Population abbreviations are: CLM = Colombians from Medellin , Colombia; MXL = Mexican Ancestry from Los Angeles , USA; PUR = Puerto Ricans from Puerto Rico; PEL = Peruvians from Lima , Peru; ACB = African Caribbeans in Barbados; ASW = Americans of African Ancestry in SW USA . Admixed populations from the Americans show structure that mirrors the continental groups , with PC1 essentially measuring the ratio between African and non-African ancestry and PC2 measuring the ratio between European and Native American ancestry . The accompanying heat maps show the loadings of the first two principal components . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02610 . 7554/eLife . 24284 . 027Figure 4—figure supplement 3 . PCA of African populations . Population abbreviations are: MSL = Mende in Sierra Leone; LWK = Luhya in Webuye , Kenya; YRI = Yoruba in Ibadan , Nigeria; GWD = Gambian in Western Divisions; ESN = Esan in Nigeria . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02710 . 7554/eLife . 24284 . 028Figure 4—figure supplement 4 . PCA of East Asian populations . Population abbreviations are: CDX = Chinese Dai in Xishuangbanna , China; JPT = Japanese in Tokyo , Japan; CHB = Han Chinese in Bejing , China; CHS = Southern Han Chinese; KHV = Kinh in Ho Chi Minh City , Vietnam . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02810 . 7554/eLife . 24284 . 029Figure 4—figure supplement 5 . PCA of South Asian populations . Population abbreviations are: ITU = Indian Telugu from the UK; GIH = Gujarati Indian from Houston , Texas; PJL = Punjabi from Lahore , Pakistan; BEB = Bengali from Bangladesh; STU = Sri Lankan Tamil from the UK . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 02910 . 7554/eLife . 24284 . 030Figure 4—figure supplement 6 . PCA of European populations . Population abbreviations are: IBS = Iberian Population in Spain; TSI = Toscani in Italia; GBR = British in England and Scotland; CEU = Utah Residents ( CEPH ) with Northern and Western Ancestry; FIN = Finnish in Finland . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 030 PCA reveals relatively little fine-scale structure within the mutational spectra of Europeans or South Asians ( Figure 4—figure supplements 5 and 6 ) . However , Africans exhibit some substructure ( Figure 4—figure supplement 3 ) , with the Luhya exhibiting the most distinctive mutational spectrum . Unexpectedly , a closer examination of PC loadings reveals that the Luhya outliers are enriched for the same mutational signature identified in the Japanese . Even in Europeans and South Asians , the first PC is heavily weighted toward AC→CC , ACA→AAA , and TAT→TTT , although this signature explains less of the mutation spectrum variance within these more homogeneous populations . The sharing of this signature may suggest either parallel increases of a shared mutation modifier , or a shared aspect of environment or life history that affects the mutation spectrum . Finally , given our finding of extensive fine-scale variation in mutational spectra between human populations , we hypothesized that mutational variation between species is likely to be even greater . To compare the mutation spectra of the great apes in more detail , we obtained SNV data from the Great Ape Diversity Panel , which includes 78 whole genome sequences from six great ape species including human ( Prado-Martinez et al . , 2013 ) . Overall , we find dramatic variation in mutational spectra among the great ape species ( Figure 5 and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 24284 . 031Figure 5 . Mutational differences among the great apes . ( A ) Relative abundance of SNV types in 5 ape species compared to Bornean Orangutan; data from ( Prado-Martinez et al . , 2013 ) . Boxes indicate labels in ( B ) . For additional comparisons see Figure 5—figure supplement 1 . ( B ) Schematic phylogeny of the great apes highlighting notable changes in SNV abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 03110 . 7554/eLife . 24284 . 032Figure 5—figure supplement 1 . Mutation spectra of great apes . These heatmap comparisons demonstrate that closely related great apes such as Chimpanzees and Bonobos have more similar mutation spectra than more distantly related apes do . DOI: http://dx . doi . org/10 . 7554/eLife . 24284 . 032 As noted previously ( Moorjani et al . , 2016a ) , one major trend is a higher proportion of CpG mutations among the species closest to human , possibly reflecting lengthening generation time along the human lineage , consistent with previous indications that species closely related to humans have lower mutation rates than more distant species ( Goodman , 1961; Li and Tanimura , 1987; Scally and Durbin , 2012 ) . However , most other differences are not obviously related to known processes such as biased gene conversion and generation time change . The A→T mutation rate appears to have sped up in the common ancestor of humans , chimpanzees , and bonobos , a change that appears consistent with a mutator variant that was fixed instead of lost . It is unclear whether this ancient A→T speedup is related to the A→T speedup in East Asians . Other mutational signatures appear on only a single branch of the great ape tree , such as a slowdown of A→C mutations in gorillas . The widespread differences captured in Figures 1 and 2 may be footprints of allele frequency shifts affecting different mutator alleles . But in principle , other genetic and non-genetic processes may also impact the observed mutational spectrum . First , biased gene conversion ( BGC ) tends to favor C/G alleles over A/T , and BGC is potentially more efficient in populations of large effective size compared to populations of smaller effective size ( Galtier et al . , 2001 ) . However , despite the bottlenecks that are known to have affected Eurasian diversity , there is no clear trend of an increased fraction of C/G→A/T relative to A/T→C/G in non-Africans vs Africans , or with distance from Africa ( Figure 1—figure supplement 7 ) , and previous studies have also found little evidence for a strong genome-wide effect of BGC on the mutational spectrum in humans and great apes ( Do et al . , 2015; Moorjani et al . , 2016a ) . For these reasons , we think that evolution of the mutational process is a better explanation than BGC or selection for differences that have been observed between the spectra of ultra-rare singleton variants and older human genetic variation ( Carlson et al . , 2017 ) ; It is also known that shifts in generation time or other life-history traits may affect mutational spectra , particularly for CpG transitions ( Martin and Palumbi , 1993; Amster and Sella , 2016 ) . Most CpG transitions result from spontaneous methyl-cytosine deamination as opposed to errors in DNA replication . Hence the rate of CpG transitions is less affected by generation time than other mutations ( Hwang and Green , 2004; Moorjani et al . , 2016b; Gao et al . , 2016 ) . We observe that Europeans have a lower fraction of CpG variants compared to Africans , East Asians and South Asians ( Figure 1B ) , consistent with a recent report of a lower rate of de novo CCG→CTG mutations in European individuals compared to Pakistanis ( Narasimhan et al . , 2016 ) . Such a pattern may be consistent with a shorter average generation time in Europeans ( Moorjani et al . , 2016b ) , although it is unclear that a plausible shift in generation-time could produce such a large effect . Apart from this , the other patterns evident in Figure 1 do not seem explainable by known processes . In summary , we report here that , mutational spectra differ significantly among closely related human populations , and that they differ greatly among the great ape species . Our work shows that subtle , concerted shifts in the frequencies of many different mutation types are more widespread than dramatic jumps in the rate of single mutation types , although the existence of the European TCC→TTC pulse shows that both modes of evolution do occur ( Harris , 2015; Moorjani et al . , 2016b; Mathieson and Reich , 2016 ) . At this time , we cannot exclude a role for nongenetic factors such as changes in life history or mutagen exposure in driving these signals . However , given the sheer diversity of the effects reported here , it seems parsimonious to us to propose that most of this variation is driven by the appearance and drift of genetic modifiers of mutation rate . This situation is perhaps reminiscent of the earlier observation that genome-wide recombination patterns are variable among individuals ( Coop et al . , 2008 ) , and ultimate discovery of PRDM9 ( Baudat et al . , 2010 ) ; although in this case it is unlikely that a single gene is responsible for all signals seen here . As large datasets of de novo mutations become available , it should be possible to map mutator loci genome-wide . In summary , our results suggest the likelihood that mutational modifiers are an important part of the landscape of human genetic variation . All datasets analyzed here are publicly available at the following websites: Mutation spectra were computed using 1000 Genomes Phase 3 SNPs ( Auton et al . , 2015 ) that are biallelic , pass all 1000 Genomes quality filters , and are not adjacent to any N’s in the hg19 reference sequence . Ancestral states were assigned using the UCSC Genome Browser alignment of hg19 to the PanTro2 chimpanzee reference genome; SNPs were discarded if neither the reference nor alternate allele matched the chimpanzee reference . To minimize the potential impact of ancestral misidentification errors , SNPs with derived allele frequency higher than 0 . 98 were discarded . We also filtered out regions annotated as ‘conserved’ based on the 100-way PhyloP conservation score ( Pollard et al . , 2010 ) , download from http://hgdownload . cse . ucsc . edu/goldenPath/hg19/phastCons100way/ , as well as regions annotated as repeats by RepeatMasker ( Smit et al . , 2013 ) , downloaded from http://hgdownload . cse . ucsc . edu/goldenpath/hg19/database/nestedRepeats . txt . gz . To be counted as part of the mutation spectrum of population P ( which can be either a continental group or a finer-scale population from one city ) , a SNP should not be a singleton within population P–at least two copies of the ancestral and derived alleles must be present within that population . An identical approach was used to extract the mutation spectrum of the UK10K ALSPAC panel ( Walter et al . , 2015 ) , which is not subdivided into smaller populations . The data were filtered as described in Field et al . ( 2016 ) . The filtering procedure performed by Field et al . ( 2016 ) reduces the ALSPAC sample size to 1927 individuals . We also computed mutation spectra of the Simons Genome Diversity Panel ( SGDP ) populations ( Mallick et al . , 2016 ) . Four of the SGDP populations , West Eurasia , East Asia , South Asia , and Africa , were compared to their direct counterparts in the 1000 Genomes data . Three additional SGDP populations , Central Asia and Siberia , Oceania , and America , had no close 1000 Genomes counterparts and were not analyzed here ( although each project contained a panel of people from the Americans , the composition of the American panels was extremely different , with the 1000 Genomes populations being much more admixed with Europeans and Africans ) . SGDP sites with more than 20% missing data were not utilized . All other data processing was done the same way described for the 1000 Genomes data . The following table gives the same size of each population panel , as well as the total number of SNPs segregating in the panel that are used to compute mutation type ratios: Biallelic great ape SNPs were extracted from the Great Ape Diversity Panel VCF ( Prado-Martinez et al . , 2013 ) , which is aligned to the hg18 human reference sequence . Ancestral states were assigned using the Great Ape Genetic Diversity project annotation , which used the Felsenstein pruning algorithm to assign allelic states to internal nodes in the great ape tree . In the Great Ape Diversity Panel , the most recent common ancestor ( MRCA ) of the human species is labeled as node 18; the MRCAs of chimpanzees , bonobos , gorillas , and orangutans , respectively , are labeled as node 16 , node 17 , node 19 , and node 15 . We extracted the state of each MRCA at each SNP in the alignment and used it to polarize the ancestral and derived allele at that site; a SNP was discarded whenever the ancestral node was assigned an uncertain or polymorphic ancestral state . As with the human data , SNPs with derived allele frequency higher than 0 . 98 were not used , and both repeats and PhyloP-annotated conserved regions were filtered away . The mutation type of an SNP is defined in terms of its ancestral allele , its derived allele , and its two immediate 5’ and 3’ neighbors . Two mutation types are considered equivalent if they are strand-complementary to each other ( e . g . ACG→ATG is equivalent to CGT→CAT ) . This scheme classifies SNPs into 96 different mutation types , each that can be represented with an A or C ancestral allele . To compute the frequency fP⁢ ( m ) of SNP m in population P , we count up all SNPs of type m where the derived allele is present in at least one representative of population P ( which can be either a specific population such as YRI or a broader continental group such as AFR ) . After obtaining this count CP⁢ ( m ) , we define fP⁢ ( m ) to be the ratio CP⁢ ( m ) /∑m′CP⁢ ( m′ ) , where the sum in the denominator ranges over all 96 mutation types m′ . The enrichment of mutation type m in population P1 relative to population P2 is defined to be fP1⁢ ( m ) /fP2⁢ ( m ) ; these enrichments are visualized as heat maps in Figures 1B , 3B and 4A . To track changes in the mutational spectrum over time , we compute fP⁢ ( m ) in bins of restricted allele frequency . This involves counting the number of SNPs of type m that are present at frequency ϕ in population P to obtain counts CP⁢ ( m , ϕ ) and frequencies fP⁢ ( m , ϕ ) =CP⁢ ( m , ϕ ) /∑m′CP⁢ ( m′⁢ϕ ) . Deviation of the ratio fP⁢ ( m , ϕ ) /fP⁢ ( m ) from one indicates that the rate of m has fluctuated recently in the history of population P . To make the sampling noise approximately uniform across alleles of different frequencies , alleles of derived count greater than five were grouped into approximately log-spaced bins that each contained similar numbers of UK10K SNPs . More precisely , we defined a set of bin endpoints b1 , b2 , … such that the total number of SNPs ranging in derived allele count between bi and bi+1−1 is greater than or equal to the number of 5-ton SNPs , while the total number of SNPs ranging in derived allele count from bi to bi+1-2 is less than the number of 5-ton SNPs . In some cases , for example Figure 2 , Figure 2—figure supplement 1B and Figure 3—figure supplement 1 , site frequency spectra were projected down to a smaller sample size before counting SNPs in order to more accurately compare datasets of different sample sizes . A binomial sampling approach was used to project a sample of N haplotypes does to a smaller sample size n . Letting CP ( N ) ⁢ ( m , ϕ ) denote the SNP counts in the large sample of N haplotypes , effective SNP counts CP ( n ) ⁢ ( m , ϕ ) in a sample of n haplotypes are computed as follows:CP ( n ) ⁢ ( m , k/n ) = ( nk ) ⁢∑ℓ=1N-1 ( ℓ/N ) k⁢ ( 1-ℓ/N ) n-k⁢CP ( N ) ⁢ ( m , ℓ/N ) One central goal of this paper is to test whether many mutation types differ in rate between human populations or whether mutation spectrum shifts have been rare events affecting only a small proportion of mutation types . A simple statistical method for answering this question would be to perform 96 separate chi-square tests , one for each triplet-context-dependent mutation type , as follows: Let Si denote the total number of SNPs segregating in population Pi , and let Si ( m ) denote the number of SNPs of mutation type m . If mutation type m is more prevalent in population P1 than in population P2 , a chi-square test provides a natural way of assessing the significance of this difference . As described in Harris ( 2015 ) , this test is performed on the following two-by-two contingency table: It would be appealing to conclude that every mutation type ‘passing’ this chi-square test is a mutation type that has changed in rate during recent human history . However , if we were to perform the full set of 96 tests , they would not be independent . A sufficiently large increase in the rate of one mutation type m1 in population P1 after divergence from P2 could cause another mutation type m2 , whose rate has remained constant , to comprise significantly different fractions of the SNPs from P1 and P2 . To minimize this effect , we formulate the following iterative procedure of conditionally independent tests: first , compute a chi-square significance value punordered⁢ ( m ) for each mutation type m using the two-by-two chi-square table above . We then use these values to order the SNPs from lowest p value to highest and compute a set of ordered p values pordered⁢ ( m ) . For the mutation type m0 with the lowest unordered p value , punordered⁢ ( m0 ) =pordered⁢ ( m0 ) . For mutation type mi , which has the ith lowest unordered p value and i<96 , pordered⁢ ( mi ) is computed from the following contingency table: For mutation type m96 , which has the highest unordered p value , the ordered p value is computed from the contingency table This procedure is guaranteed to find fewer mutation types to differ significantly in rate between populations compared to separate chi-square tests . The python package matplotlib . mlab . PCA was used to perform PCA on the complete set of 1000 Genomes diploid genomes . First , the triplet mutational spectrum of each haplotype h was computed as a 96-element vector encoding the mutation frequencies ( fh⁢ ( m ) ) m of the non-singleton derived alleles present on that haplotype . The mutational spectrum of each diploid genome was then computed by averaging together the spectra of its two constituent haplotypes . In the same way , a separate PCA was performed on each of the five continental groups to reveal finescale components of mutation spectrum variation . We estimated the duration and intensity of TCC→T rate acceleration in Europe by fitting a simple piecewise-constant rate model to the UK10K frequency data . To specify the parameters of the model , we divide time into discrete log-spaced intervals bounded by time points t1 , … , td , assigning each interval a TCC→T mutation rate r0 , …⁢rd . In units of generations before the present , the time discretization points were chosen to be: 20 , 40 , 200 , 400 , 800 , 1200 , 1600 , 2000 , 2400 , 2800 , 3200 , 3600 , 4000 , 8000 , 12 , 000 , 16 , 000 , 20 , 000 , 24 , 000 , 28 , 000 , 32 , 000 , 36 , 000 , 40 , 000 . We assume that the total rate r of mutations other than TCC→T stays constant over time ( a first-order approximation ) . In terms of these rate variables , we can calculate the expected shape of the TCC→T pulse shown in Figure 2B of the main text . The shape of this curve depends on both the mutation rate parameters ri and the demographic history of the European population , which determines the joint distribution of allele frequency and allele age . To account for the effects of demography , we use Hudson’s ms program to simulate 10 , 000 random coalescent trees under a realistic European demographic history inferred from allele frequency data ( Tennessen et al . , 2012 ) and condition our inference upon this collection of trees as follows: Let A⁢ ( m , t ) be the function for which ∫titi+1A⁢ ( m , t ) ⁢d⁢t equals the coalescent tree branch length , averaged over the sample of simulated trees , that is ancestral to exactly m lineages and falls between time ti and ti+1 . Given this function , which can be empirically estimated from a sample of simulated trees , the expected frequency spectrum entry k/n isE⁢ ( k/n ) =∑i=1d∫ti-1tiA⁢ ( k , t ) ⁢d⁢t∑j=1n∑i=1d∫ti-1tiA⁢ ( j , t ) ⁢d⁢t and the expected fraction of TCC→T mutations in allele frequency bin k/n isE⁢ ( fTCC→T⁢ ( k/n ) ) =∑i=1dri⁢∫ti-1tiA⁢ ( k , t ) ⁢d⁢tr⁢∑i=1d∫ti-1tiA⁢ ( k , t ) ⁢d⁢t . The expected value of the TCC→T enrichment ratio being plotted in Figure 2B isE⁢ ( rTCC→T⁢ ( k/n ) ) =∑i=1dri⁢∫ti-1tiA⁢ ( k , t ) ⁢d⁢t⋅∑j=1n∑i=1d∫ti-1tiA⁢ ( j , t ) ⁢d⁢t∑i=1d∫ti-1tiA⁢ ( k , t ) ⁢d⁢t⋅∑j=1n∑i=1dri⁢∫ti-1tiA⁢ ( j , t ) ⁢d⁢t In Figure 2B , enrichment ratios are not computed for every allele frequency in isolation , but for allele frequency bins that each contain similar numbers of SNPs . Given integers 1≤km<km+1≤n , the expected TCC→T enrichment ratio averaged over all SNPs with allele frequency between km/n and km+1/n is:E⁢ ( rTCC→T⁢ ( km/n ) ) =∑i=1dri⁢∫ti-1ti∑k=kmkm+1A⁢ ( k , t ) ⁢d⁢t⋅∑j=1n∑i=1d∫ti-1tiA⁢ ( j , t ) ⁢d⁢t∑i=1d∫ti-1ti∑k=kmkm+1A⁢ ( k , t ) ⁢d⁢t⋅∑j=1n∑i=1dri⁢∫ti-1tiA⁢ ( j , t ) ⁢d⁢t We optimize the mutation rates r1 , … , rd using a log-spaced quantization of allele frequencies k1/n , … , km/n defined such that all bins contain similar numbers of SNPs . The chosen allele count endpoints k1 , … , km are: 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 20 , 30 , 40 , 50 , 60 , 70 , 80 , 90 , 100 , 200 , 300 , 400 , 500 , 600 , 700 , 800 , 900 , 1000 , 2000 , 3000 , 4000 . Given this quantization of allele frequencies , we optimize r1 , … , rd by using the BFGS algorithm to minimize the least squares distance D⁢ ( r0 , … , rd ) between E⁢ ( rTCC→T⁢ ( km/n ) ) and the empirical ratio rTCC→T⁢ ( km/n ) computed from the UK10K data . This optimization is subject to a regularization penalty that minimizes the jumps between adjacent mutation rates ri and ri+1:D⁢ ( r0 , … , rd ) =∑m=1d ( E⁢ ( rTCC→T⁢ ( km/n ) ) -rTCC→T⁢ ( km/n ) ) 2+0 . 25⁢∑i=1d ( ri-1-ri ) 2 Although the underlying model of mutation rate change assumed here is very simple , it still represents an advance over the method used in ( Harris , 2015 ) to estimate of the timing of the TCC→TTC mutation rate increase . That method relied upon explicit estimates of allele age from a dataset of less than 100 individuals , which are much noisier than integration of a joint distribution of allele age and frequency across a sample of thousands of haplotypes .	DNA is a molecule that contains the information needed to build an organism . This information is stored as a code made up of four chemicals: adenine ( A ) , guanine ( G ) , cytosine ( C ) , and thymine ( T ) . Every time a cell divides and copies its DNA , it accidentally introduces ‘typos’ into the code , known as mutations . Most mutations are harmless , but some can cause damage . All cells have ways to proofread DNA , and the more resources are devoted to proofreading , the less mutations occur . Simple organisms such as bacteria use less energy to reduce mutations , because their genomes may tolerate more damage . More complex organisms , from yeast to humans , instead need to proofread their genomes more thoroughly . Recent research has shown that humans have a lower mutation rate than chimpanzees and gorillas , their closest living relatives . Humans and other apes copy and proofread their DNA with basically the same biological machinery as yeast , which is about a billion years old . Yet , humans and apes have only existed for a small fraction of this time , a few million years . Why then do humans need to replicate and proofread their DNA differently from apes , and could it be that the way mutations arise is still evolving ? Previous research revealed that European people experience more mutations within certain DNA motifs ( specifically , the DNA sequences ‘TCC’ , ‘TCT’ , ‘CCC’ and ‘ACC’ ) than Africans or East Asians do . Now , Harris ( who conducted the previous research ) and Pritchard have compared how various human ethnic groups accumulate mutations and how these processes differ in different groups . Statistical analysis of the genomes of thousands of people from all over the world did indeed show that the mutation rates of many different three-letter DNA motifs have changed during the past 20 , 000 years of human evolution . Harris and Pritchard report that when groups of humans left Africa and settled in isolated populations across different continents , each population quickly became better at avoiding mutations in some genomic contexts , but worse in others . This suggests that the risk of passing on harmful mutations to future generations is changing and evolving at an even faster rate than was originally suspected . The results suggest that every human ethnic group carries specific variants of the genes which ensure that DNA replication and repair are accurate . These differences appear to influence which types of mutations are frequently passed down to future generations . An important next step will be to identify the genetic variants that could be controlling mutational patterns and how they affect human health .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2017	Rapid evolution of the human mutation spectrum
Chronic infection perturbs immune homeostasis . While prior studies have reported dysregulation of effector and memory cells , little is known about the effects on naïve T cell populations . We performed a cross-sectional study of chronic hepatitis C ( cHCV ) patients using tetramer-associated magnetic enrichment to study antigen-specific inexperienced CD8+ T cells ( i . e . , tumor or unrelated virus-specific populations in tumor-free and sero-negative individuals ) . cHCV showed normal precursor frequencies , but increased proportions of memory-phenotype inexperienced cells , as compared to healthy donors or cured HCV patients . These observations could be explained by low surface expression of CD5 , a negative regulator of TCR signaling . Accordingly , we demonstrated TCR hyperactivation and generation of potent CD8+ T cell responses from the altered T cell repertoire of cHCV patients . In sum , we provide the first evidence that naïve CD8+ T cells are dysregulated during cHCV infection , and establish a new mechanism of immune perturbation secondary to chronic infection . Functional impairments of CD8+ T cells have been characterized in several persistent viral infections , including human immunodeficiency virus ( HIV ) and hepatitis C virus ( HCV ) infection in humans , simian immunodeficiency virus ( SIV ) infection in macaques , and lymphocytic choriomeningitis virus ( LCMV ) infection in mice ( Ahmed and Rouse , 2006 ) . In particular , it has been shown that chronic infection skews memory/effector CD8+ T cell differentiation ( Stelekati and Wherry , 2012 ) , and drives virus-specific CD8+ T cells towards an « exhausted » phenotypic state , as marked by high expression of the programmed cell death-1 ( PD-1 ) molecule ( Kim and Ahmed , 2010 ) . Chronic infections have also been reported to impair immune responses to unrelated infectious microbes in mouse models ( Stelekati and Wherry , 2012; Richer et al . , 2013 ) , as well as in humans infected with HCV ( Park and Rehermann , 2014 ) . This phenomenon correlates with a interferon ( IFN ) stimulated gene ( ISG ) transcriptional signature , suggesting an indirect effect of systemic type I IFN secondary to innate immune activation ( Stelekati et al . , 2014 ) . Following from these observations , we hypothesized that chronic infection may alter the T cell preimmune repertoire , which plays an important role in shaping the adaptive immune responses ( Jenkins and Moon , 2012 ) . Employing a newly validated approach for the study of low-frequency ( < 10–5 ) antigen-specific T cells ( Alanio et al . , 2010 ) , we evaluated this prediction in patients with chronic viral infection of the liver . The α/β T cell preimmune repertoire is defined as the set of mature but antigen inexperienced lymphocytes that circulate in blood and secondary lymphoid organs , ready to be activated by cognate high-affinity peptide-class I MHC ( pMHC ) complexes ( Jenkins et al . , 2010 ) . They are maintained in the periphery by survival factors such as IL-7 , as well as transient contacts with low affinity non-cognate pMHC complexes ( Sprent and Surh , 2011 ) . Over the last decade , studies using newly-developed tetramer-enrichment assays - sensitive enough to detect and track antigen-specific populations prior to immunization - have provided new insights into the impact of preimmune repertoire heterogeneity ( Jenkins et al . , 2010 ) . First , the number of antigen-specific T cells ( i . e . precursor frequency ) is not equivalent across inexperienced populations , with the absolute number positively correlating with the magnitude of responses that are induced upon priming ( Obar et al . , 2008; Moon et al . , 2007; Kwok et al . , 2012; Schmidt et al . , 2011; Kotturi et al . , 2008; Tan et al . , 2011 ) . Second , antigen-inexperienced CD4+ and CD8+ T cell populations contain variable proportions of memory-phenotype ( MP ) cells ( Legoux et al . , 2010; Su et al . , 2013 ) . These cells have been explained in the literature as a result of cross-reactivity or homeostatic proliferation ( Sprent and Surh , 2011 ) . Cross-reactivity is now recognized as an essential feature of the T-cell receptor ( TCR ) / MHC interaction ( Mason et al . , 1998 ) , and a major determinant of virus-specific MP cells in the CD4+ T cell repertoire of unexposed healthy donors ( Su et al . , 2013 ) . Alternatively , homeostatic proliferation may occur in settings of lymphopenia ( Jones et al . , 2013 ) . Finally , differential CD5 expression by antigen-specific T cell populations has been shown to dictate clonal recruitment and expansion ( Fulton et al . , 2015; Tabbekh et al . , 2013 ) . To date , the impact of non-heritable influences such as human chronic viral infection on the quantitative and qualitative aspects of the preimmune repertoire remains unknown . In our study , we focused on patients with chronic hepatitis C virus infection ( cHCV ) , which show CD8+ T cell dysfunction ( Park and Rehermann , 2014; Rehermann and Nascimbeni , 2005 ) . In particular , HCV-specific responses are typically ( i ) weak – both in term of numbers and function , ( ii ) of low avidity , and ( iii ) blocked in their differentiation into central memory cells , despite the availability of cognate pMHC complexes ( Park and Rehermann , 2014 ) . cHCV is to date the only chronic viral infection that can be cured , offering the unique possibility to interrogate the reversibility of immune perturbations post-viral clearance ( Pol et al . , 2013 ) . Herein , we applied the highly sensitive tetramer-associated magnetic enrichment ( TAME ) technique for investigating at the antigen-specific level the impact of chronic viral hepatitis infection on the CD8 T cell preimmune repertoire ( Alanio et al . , 2010 ) . Although precursor frequencies were similar to healthy controls , we observed significant impairments of the preimmune repertoire in cHCV patients . Inexperienced T cell populations contained increased proportions of MP cells . This correlated with naïve-phenotype CD8+ T cells having lower surface expression of CD5 , which accounted for a lower threshold for TCR signaling and the generation of potent immune responses from cHCV patients . Importantly , the positive effect of chronic infection on naïve T cell recruitment into immune responses is transient , as cHCV patients who clear their virus following successful therapy ( referred to as Sustained Virologic Responders or SVR ) can experience a reversion towards a healthy naïve T cell repertoire within 2 years . These data provide the first evidence for chronic infection resulting in the bystander dysregulation of the antigen-specific preimmune repertoire in humans , and highlight the added benefit of early viral clearance in patients with chronic HCV infection . To test the hypothesis that chronic viral infection perturbs preimmune repertoire homeostasis , we evaluated the influence of cHCV infection on the phenotype of circulating CD8+ T cells . 29 cHCV and 37 Sustained Virologic Responders ( SVR , i . e . patients achieving clearance of the virus after therapy ) patients were included in the study ( Table 1 ) . 62% of the chronic and 100% of the SVR patients received at least one anti-HCV treatment ( of those treated , 69% received conventional IFN-ribavirin bitherapy , 31% IFN + direct antiviral agent ( DAA ) , and IFN-free DAA combination therapy alone in the case of a single SVR patient ) . Healthy donors from the blood bank were included as controls . Total lymphocyte numbers were within the normal range for all tested patients ( median 2 . 2 +/- 0 . 6 G/l ) . Within the CD3+ lymphocyte population , we observed similar percentages of circulating CD8+ T cells ( Figure 1—figure supplement 1 ) . However , absolute numbers of CD3+ were significantly increased in our cohort of cHCV ( KW p<0 . 0001 ) , translating into increased absolute numbers of CD8+ T cells in cHCV patients ( KW p=0 . 0002 ) ( Figure 1—figure supplement 2 ) . We further subsetted the CD8+ T cells according to their surface expression of CD45RA and CD27 . Based on prior studies ( Alanio et al . , 2010; De Rosa et al . , 2001 ) and our confirmatory experiments using 5 phenotypic markers for naïve or memory T cells , we determined that co-expression of high levels of CD45RA and CD27 were sufficient to classify naïve T cells in both HD and cHCV patients ( Figure 1—figure supplement 3 ) . Decreased percentages of naïve CD8+T cells have previously been reported in cHCV ( Shen et al . , 2010 ) . Here , we confirmed these findings in age- and CMV- matched chronically infected patients ( KW p=0 . 0007 , Figure 1A , B ) . Interestingly , we found that after correcting for the higher CD8+ T cell numbers in cHCV patients , the absolute numbers of naïve CD8+ T cells were within the normal range as determined by the study of healthy donors ( Figure 1C ) . We therefore interpreted the lower proportion of naïve T cells to simply be a result of an expansion of the memory cell compartment . 10 . 7554/eLife . 07916 . 003Figure 1 . Perturbed naïve CD8+ T cell repertoire during chronic HCV infection . Percentages and absolute numbers of CD3+ and CD3+CD8+ cells in Healthy Donors ( HD ) , Sustained Virologic Responder ( SVR ) , and chronic HCV ( cHCV ) patients are provided in Figure 1—figure supplement 1 and 2 . ( A ) Representative examples of CD45RA+CD27+ naïve CD8+ T cell compartment in the three donor subsets . FACS plots are gated on Live CD3+CD8+ cells . Validation of CD45RA/CD27 gating strategy for identifying naïve CD8+ T cells in cHCV patients is provided in Figure 1—figure supplement 3 . ( B ) Percentages of naïve CD8+ T cells in the three donor subsets . ( C ) Absolute numbers ( G/L ) of naïve CD8+ T cells in HD , SVR , and cHCV patients . ns ( not significant , p>0 . 05 ) , * ( p≤0 . 05 ) , ( p≤0 . 01 ) , and * ( p≤0 . 001 ) refer to Dunn’s multiple comparison test of each subset toward HD . ( D ) Normalized numbers of sjTRECs per 150 , 000 naïve CD8+ in HD and cHCV samples . Normalized numbers of sjTRECs per total CD8+ T cells are provided in Figure 1—figure supplement 4 . ( E ) Representative example of the distribution of 24 FACS-screened Vβ families in naïve CD8+ T cells from one HD and one cHCV sample . Families are ordered by increasing size in both individuals . ( F ) Lorenz curves representing the cumulative distribution of % of usage of 24 FACS-screened Vβ families from 7 HD and 7 cHCV patients . Mean Gini coefficients and standard deviations are indicated . Red line indicates an extreme example of an unequal distribution , observed in the case of a T-cell lymphoma where >90% of the TCR repertoire is explained by one particular Vβ chain . ( G ) Individual Gini coefficients of all tested samples are represented for HD and cHCV subgroups . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00310 . 7554/eLife . 07916 . 004Figure 1—figure supplement 1 . Comparable proportions of CD8+ T cells circulate in cHCV patients and HD . ( A ) Representative examples of CD3+CD8+ compartment in HD , SVR , and cHCV patients . FACS plots are gated into LiveFSCloSSClo PBMCs . ( B ) Percentages of CD3+CD8+ T cells in the three donor subsets . ( C ) Absolute numbers ( G/L ) of CD3+ T cells in HD , SVR , and cHCV patients . ( D ) Absolute numbers ( G/L ) of CD8+ T cells in HD , SVR , and cHCV patients . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00410 . 7554/eLife . 07916 . 005Figure 1—figure supplement 2 . Increased absolute numbers of CD3+ and CD8+ T cells in cHCV and SVR patients . ( A ) Absolute numbers ( G/L ) of CD3+ T cells in HD , SVR , and cHCV patients . ( B ) Absolute numbers ( G/L ) of CD3+ T cells in HD , SVR , and cHCV patients . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00510 . 7554/eLife . 07916 . 006Figure 1—figure supplement 3 . Validation of CD45RA/CD27 gating strategy for identifying naïve CD8+ T cells in cHCV patients . ( A ) Example of CD45RA/CD27 gating strategy in one cHCV patient . CD8+naïve T cells are identified as CD45RA+CD27+ . Pattern of CD127 , CD45RO and CCR7 expression of each identified population is displayed on histogram overlays ( black line represents the subpopulation of interest; grey ones the total CD8+ population ) . Of note , the minor population ( 9% ) of CD127 negative cells in the displayed CD45RA+CD27+ gate are >99% CD45RO negative and CCR7 positive . ( B ) CCR7 expression on bulk CD3+CD8+ T cells ( grey ) , and bulk ( left ) or Mart1-specific ( right ) CD3+CD8+ CD45RA+CD27+ T cells ( black ) from one HD and one cHCV . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00610 . 7554/eLife . 07916 . 007Figure 1—figure supplement 4 . Decreased number of sjTRECs in total CD3+CD8+ T cells from cHCV patients . Normalized numbers of signal-joint T cell receptor excision circle ( sjTREC ) per 150 , 000 CD3+CD8+ cells in HD and cHCV samples . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00710 . 7554/eLife . 07916 . 008Table 1 . Donors included in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 008All donorscHCVSVRHDn=29n=37n=25Male , n ( % ) 16 ( 55 ) 21 ( 57 ) 12 ( 48 ) Age , years , median ( IQR1-3 ) 48 ( 42-55 ) 48 ( 44-58 ) 38 ( 31-46 ) IgG anti-CMV positive , n ( % ) 14 ( 48 ) 24 ( 51 ) 11 ( 44 ) Cirrhosis , n ( % ) 5 ( 17 ) 8 ( 22 ) naTreatment experienced , n ( % ) 18 ( 62 ) 37 ( 100 ) naTreatment ( n per type: 0/1/2/3 ) 11/12/6/00/26/10/1naDelay post-treatment , years , median ( IQR1-3 ) 3 . 8 ( 3 . 4-4 . 2 ) 1 . 7 ( 0 . 9-3 . 2 ) na To directly test this prediction , we isolated CD8+ T cells and measured the frequency of signal joint TCR excision circles ( sjTREC ) , by-products of TCR rearrangement , and previously validated as a measure of thymic production ( Rehermann and Nascimbeni , 2005; Clave et al . , 2009 ) . Confirming previous studies , we found a significant decrease in sjTREC content of CD8+ T cells ( MW p=0 . 01 , Figure 1—figure supplement 4 ) . To address the bias due to differential naïve T cell number , we isolated CD45RA+/CD27+ naïve CD8+ T cells and assessed sjTREC frequencies . Surprisingly , we also observed within the naïve compartment a significantly lower sjTREC content in cHCV patients as compared to HD ( MW p=0 . 03 , Figure 1D ) . To further characterize this phenotype , we assessed the Vβ distribution within the naïve repertoire of cHCV patients . cHCV patients showed a biased repertoire with increased representation of selected Vβ families . A representative example of Vβ usage plotted as percentage accross the 24 tested families , and ordered by increasing size from one cHCV patient and one HD is shown ( Figure 1E ) . To compare distributions , Lorenz curves were constructed as a graphical representation of the diversity of the repertoire ( Figure 1F ) . Inequality measurements in the Vβ distribution , comparing cHCV patients to HD , indicated proportions of naïve T cells being altered in their Vβ usage . In brief , for a given percentage ( x ) of the 24 Vβ chains , Lorenz curves indicate the proportion of the T cell population that have Vβ chains among the 24 * x% least abundant ones . An equal distribution is represented as the dotted line . By contrast , an extreme , unequal distribution is shown in red , as in the case of a T-cell lymphoma where >90% of the TCR repertoire is explained by one particular Vβ chain ( red line ) . We included Gini coefficient as a numeric measure of Lorenz curve’s based observations . It corresponds to the ratio of the area between the line representing equal use of all Vβ chains ( dotted line ) and the observed Lorenz curve to the total area below the line representing equal use . The higher the coefficient , the more unequal is the distribution . In line with our observation , we found Gini coefficients increased in cHCV patients ( M-W p=0 . 03 , see Material and Methods for details of calculation ) ( Figure 1G ) . These data support an overall perturbed naïve CD8+ T cell repertoire in cHCV patients , with increased peripheral expansion of selected populations . To evaluate more precisely the impact of these perturbations on antigen-specific populations , we applied recently developed strategies to detect , quantify and phenotype rare inexperienced antigen-specific CD8+ T cells ( Klenerman and Thimme , 2012; Alanio et al . , 2013 ) . Specifically , we utilized TAME to enumerate and subdivide Mart1-specific T cell populations . While similar absolute numbers of Mart1-specific CD8+ T cells were observed in our respective study groups ( Figure 2A , B ) , SVR and cHCV patients showed a more differentiated phenotype ( Figure 2C ) , defined by fewer CD45RA+/CD27+ and increased proportions of memory-phenotype ( MP ) cells ( KW p<0 . 0001 , Figure 2D ) . Of note , these MP cells were mostly of central-memory ( CD45RA-CD27+ ) phenotype ( Figure 2—figure supplement 1 ) . Also , when considering only naïve-phenotype Mart1-specific cells , precursor frequencies were still comparable across the different study groups ( Figure 2—figure supplement 2 ) . We were able to purify sufficient numbers of Mart1-specific naïve- and memory- phenotype CD8+ T cells from one HCV patient to perform an immunoscope analysis on the Vβ chain usage ( Figure 2E ) . In line with our data in bulk T cells populations ( Figure 1 ) , we observed a restricted repertoire of Mart1-specific naïve T cells , with evidence of an expanded Vβ clonotype in memory cells . These data argue in favor of MP cells being the progeny of a perturbed naïve T cell repertoire . Although they could be expanded in response to either specific or non-specific signals , we favor the latter hypothesis based on prior knowledge of Mart1 antigen pattern of expression ( Pittet et al . , 1999 ) . 10 . 7554/eLife . 07916 . 009Figure 2 . Peripheral differentiation of Mart1-specific CD8+ T cells during cHCV infection . ( A ) Representative examples of Mart1-specificCD8+ T cellpopulations in HD , SVR , cHCV patients . FACS plots are gated on TAME-enriched LiveFSCloSSCloCD3+CD8+ PBMCs . ( B ) Precursor frequency of Mart1-specific cellsin the three donor subsets . Precursor frequency of naïve-phenotype Mart1-specific cellsis provided in Figure 2—figure supplement 1 . ( C ) Representative examples of the CD45RA/CD27 phenotype of TAME-enriched Mart1-specificpopulations in patients subsets as in A . D/ Percentages of memory-phenotype ( MP ) cells in Mart1-specific populations in the three donor subsets . Further subsetting of MP inexperienced T cells into CD45/CD27-based T cell differentiation phenotype is provided in Figure 2—figure supplement 1 . E/ Immunoscope profile of naïve and memory Mart1-specific populations FACS-sorted from one cHCV patient . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 00910 . 7554/eLife . 07916 . 010Figure 2—figure supplement 1 . CD45RA/CD27-based subsetting of Mart1-specific T cells enriched from HD , SVR , cHCV . Percentages of CD45RA+CD27+ ( naïve ) , CD45RA-CD27+ ( central memory ) , CD45RA-CD27- ( effector memory ) , CD45RA+CD27- ( late effector memory ) cells within Mart1-specific T cell population in HD , cHCV and SVR patients . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01010 . 7554/eLife . 07916 . 011Figure 2—figure supplement 2 . Precursor frequency of Mart1 naïve-phenotype cellsin the three donor subsets . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 011 We next extended our observations to other antigen specificities by using four additional multimers ( hTERT1572-580 , human CMV pp65495-503 , Ebola NP202-210 ( Sundar et al . , 2007 ) , HIV-1 Gag p1777-85 ) that are expected to detect inexperienced self- and virus-specific CD8+ T cell populations in tumor-free , CMV- , Ebola- and HIV- seronegative individuals . Here again , we found high proportions of T cells with a memory-phenotype in both self ( Mart1- and hTERT- specific ) and viral ( CMV- , Ebola- and HIV- specific ) antigen-inexperienced populations of cHCV patients as compared to healthy donors ( representative plots are shown in Figure 3A and B; and combined results from 2–6 individuals per group in Figure 3C; self-specific: KW p<0 . 001; non-self-specific: KW p=0 . 009 ) . When subsetted using CD45RA and CD27 phenotypic markers , the MP cells found in cHCV patients were preferentially of CD45RA-CD27+ central memory phenotype ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 07916 . 012Figure 3 . Memory-phenotype cells within self and non-self antigen-inexperienced populations . ( A ) Representative examples of Mart1- , hTERT- , CMV- , Ebola- and HIV- specificpopulations from HD , SVR , and cHCV patients . Enriched tetramer-specific populations are overlaid on total CD8+ T cells . ( B ) CD45RA/CD27 phenotype of tetramer-specific populations gated in A . ( C ) Percentages of memory-phenotype cells in Mart1- and hTERT- ( self ) ; CMV- , Ebola- and HIV- ( non-self ) specific populations from HD , SVR and cHCV patients . Further subsetting of MP inexperienced T cells into CD45/CD27-based T cell differentiation phenotype is provided in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01210 . 7554/eLife . 07916 . 013Figure 3—figure supplement 1 . CD45RA/CD27-based subsetting of Mart1- , Ebola- , and HIV- specific T cells enriched from cHCV patients . Percentages of CD45RA+CD27+ ( naïve ) , CD45RA-CD27+ ( central memory ) , CD45RA-CD27- ( effector memory ) , CD45RA+CD27- ( late effector memory ) cells within Mart1- , Ebola- , and HIV- specific T cell populations in HD , cHCV and SVR patients . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 013 We next compared cHCV patients to those who achieved viral clearance . Sequential samples ( available from five patients who achieved cure ) suggested that immune restoration of the naïve compartment is possible ( positive time dependency p-value p=0 . 03 , Figure 4A and B ) . These patients were all treated by IFN-RBV biotherapy ( n = 3 ) , or triple therapy that included an NS3 inhibitor ( n = 2 , patients S2 and S12 ) . Testing our observation in our cross-sectional cohort , we replicated our findings , showing a statistically significant recovery of naïve antigen-specific CD8+ T cells as a function of time ( MW p=0 . 04; Figure 4C ) . These results indicate that the differentiated cells within the perturbed repertoire of cHCV patients are a reflection of active HCV infection , and likely not a result of cross-reactivity or true memory T cell differentiation . Together the results in Figures 1–4 highlight an overall perturbation of the preimmune CD8+ T cell compartment during active cHCV infection . 10 . 7554/eLife . 07916 . 014Figure 4 . Memory phenotype of Mart1-specific CD8+ T cells during chronic infection may be reversed by viral clearance . ( A ) Example of CD45RA/CD27 phenotype of Mart1-specific cells during chronic phase , and over time after viral clearance in one HLA-A0201 SVR patients ( patient S7 ) . ( B ) Percentages of Mart1 memory-phenotype cells over time after viral clearance on 5 SVR patients with longitudinal sampling – including S7 presented in E . ( C ) Percentages of memory-phenotype cells in Mart1-specific populations vs . time elapsed since clearance of the virus in SVR patients ( time-stratified , in years ) . These data include all HLA A0201 SVR patients; first available data is incorporated for follow-up patients presented in F . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 014 To establish a mechanistic understanding of our findings , we considered the key homeostatic factors governing maintenance of the naïve CD8+ T cell compartment ( Jenkins et al . , 2010; Ho and Hsue , 2009; Hazenberg et al . , 2000 ) . We hypothesized that an altered threshold for TCR activation could explain the differentiation phenotype of inexperienced T cells . CD5 expression has been shown to correlate with the threshold of activation in mice ( Grossman and Paul , 2015 ) . It is typically high on naïve T cells , showing diminished levels as a function of T cell differentiation ( Figure 5—figure supplement 1 ) . We observed phenotypic changes ( i . e . low CD5 expression ) that were significant for the comparison between CD45RA+/CD27+ naïve CD8+ T cells in cHCV vs HD ( KW p=0 . 02 , Figure 5A and B ) . Based on its role in regulating TCR signaling , we predicted that lower CD5 expression on naïve T cells would result in their hyperactivation upon stimulation . This was tested functionally by evaluating TCR signaling in naïve CD8+ T cells , stimulating cells with low doses of plate-bound anti-CD3 and anti-CD28 Abs . While only weak induction of phosphorylated ERK ( p-ERK ) could be observed in HD during the first hour of stimulation , TCR stimulation induced a strong p-ERK signal in naïve cells from seven of sixteen cHCV patients tested ( histograms from one responding cHCV and one HD are shown in Figure 5C; MW p=0 . 03 , Figure 5D ) . Using the same stimulation protocol , we investigated expression of activation markers ( i . e . , CD25 , CD69 ) measured after 24 hr stimulation . Consistent with p-ERK data , we observed higher percentages of cells expressing CD25 on naïve CD8+ T cells from cHCV as compared to HD ( representative example from one cHCV and one HD are shown in Figure 5E; MW p=0 . 02 , Figure 5F ) . Similar results were obtained for CD69 analysis ( data not shown ) . We also observed increased percentages of naïve CD8 T cells undergoing activation-induced cell death – as assessed by active caspase 3 staining after 24 hr - in cHCV patients as compared to HD ( MW p=0 . 002; Figure 5—figure supplement 2 ) . These findings are all consistent with strong TCR engagement despite the use of low doses of cross-linking antibodies in cHCV patients . 10 . 7554/eLife . 07916 . 015Figure 5 . Decreased cell surface expression of CD5 on cHCV naïve CD8+ T cells correlates with hypersensitivity to TCR activation . ( A ) Representative histograms of CD5 on naïve CD8+ T cells from one HD and one cHCV patient . ( B ) MFI of CD5 on the surface of naïve CD8+ T cells from HD , SVR , and cHCV patients . Representative histograms and MFI of CD5 on the other T cell differentiation subsets are provided in Figure 5—figure supplement 1 . ( C ) Representative overlay of histograms of phospho-ERK ( p-ERK ) signal at different time points following TCR stimulation from one HD and one cHCV patient . Plots are gated on naïve CD8+ T cell populations . ( D ) Percentages of p-ERK positive cells in naïve CD8+ T cells from HD and cHCV patients 5 min after CD3/CD28 stimulation . ( E ) Representative overlay of histograms of CD25 expression , detected at 24 hr after TCR stimulation from one HD , and one cHCV patient . Plots are gated on naïve CD8+ T cell populations . ( F ) Percentages of CD25+ cells in naïve CD8+ T cells from HD and cHCV patients 24 hr after CD3/CD28 stimulation . Representative examples and percentages of active-caspase 3-expressing cells after similar stimulation are provided in Figure 5—figure supplement 2 . ( G and H ) Percentages of p-ERK ( 5mins ) , and CD25 ( 24 hr ) after TCR stimulation in naïve CD8+ T cells from HD , with or without prior CD5 blockade with α-CD5 antibodies . Percentages of active-caspase 3-expressing cells under similar conditions are provided in Figure 5—figure supplement 3 . Impact of CD5 blockade on TCR activation in cHCV patients is provided in Figure 5—figure supplement 4 . Similar evaluation of naïve CD8+ T cell repertoire during chronic HBV infection is provided in Figure 5—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01510 . 7554/eLife . 07916 . 016Figure 5—figure supplement 1 . Evolution of MFI of CD5 over T cell differentiation in HD and cHCV patients . ( A ) Representative histograms of CD5 in CD8+ T cell differentiation subsets in one HD and one cHCV patient . ( B ) MFI of CD5 on CD8+ T cell differentiation subsets from HD and cHCV donors . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01610 . 7554/eLife . 07916 . 017Figure 5—figure supplement 2 . Increased activation-induced cell death after TCR stimulation in cHCV patients . ( A ) Representative overlay of histograms of active-caspase 3 ( a-Casp3 ) , detected intracellularly at 24 hr after TCR stimulation from one HD , and one cHCV patient . Plots are gated on naïve CD8+ T cell populations . ( B ) Percentages of a-Casp3+ cells in naïve CD8+ T cells from HD and cHCV patients 24 hr after CD3/CD28 stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01710 . 7554/eLife . 07916 . 018Figure 5—figure supplement 3 . CD5 blockade leads to increased activation-induced cell death after TCR stimulation in HD . Percentages of a-Casp3+ ( 24 hr ) after TCR stimulation in naïve CD8+ T cells from HD , with or without prior CD5 blockade with α-CD5 antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01810 . 7554/eLife . 07916 . 019Figure 5—figure supplement 4 . Impact of CD5 blockade on TCR activation in cHCV patients . Impact of preincubation with anti-CD5 antibodies on% CD25 ( A ) and% active-caspase 3 ( B ) after CD3/CD28 stimulation in HD ( A and B ) and cHCV patients ( all patients in A and B , and paired representation for HCV patients in C and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 01910 . 7554/eLife . 07916 . 020Figure 5—figure supplement 5 . Distinct perturbation of naïve CD8+ T cell repertoire during chronic HBV infection . ( A ) Representative example of CD45RA+CD27+ naïve CD8+ compartment in one cHBV patient , and percentages in multiple HD and cHBV donors . Absolute numbers could not be calculated in this cohort where only white blood cells counts were evaluated at the time of their sampling – and not lymphocyte numbers . ( B ) Mart1 precursor frequency in multiple HD and cHBV donors . ( C ) Percentages of memory-phenotype Mart1-specific T cells in multiple HD and cHBV donors . ( D ) Representative histograms of CD5 on naïve CD8+ T cells from one HD ( white ) and one cHBV patient ( black ) , and MFI of CD5 on naïve CD8+ T cells from HD and cHBV patients . ( E ) Percentages of p-ERK and CD25-positive cells in naïve CD8+ T cells from HD and cHBV patients 24 hr after CD3/CD28 stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 020 To test the mechanistic link between CD5 expression and hyperactivation of naïve T cells , we evaluated the effect of blocking CD5 signaling . When PBMCs from HD were exposed to blocking anti-CD5 Abs ( αCD5 ) prior to TCR stimulation , we observed ( i ) increased levels of p-ERK after 5 min ( Wilcoxon p=0 . 007 , Figure 5G ) , ( ii ) increased CD25 expression after 24 hr ( Wilcoxon p=0 . 03 , Figure 5H ) , and ( iii ) increased percentages of dying naïve CD8 T cells as assessed by active caspase 3 staining after 24 hr ( Wilcoxon p=0 . 003 ) ( Figure 5—figure supplement 3 ) . When compared to cHCV patients , αCD5 partially reproduced the hyperactivation phenotype of naïve T cells from cHCV patients ( Figure 5—figure supplement 4A , B ) . By contrast , when αCD5 was applied to the cHCV patients , we observed no further increase in TCR-induced activation ( Figure 5—figure supplement 4 A–D ) . These data provide direct evidence for a negative role of CD5 on TCR-induced activation and activation-induced cell death , and support the concept that CD5 molecule is responsible , in part , for the hyperactivation phenotype observed in naive T cells of cHCV patients . Together , these data support a model where low expression of CD5 on naïve T cells in cHCV patients results in dysregulation of the homeostatic TCR threshold . We next evaluated the consequences of a low threshold for TCR activation on the ability of inexperienced T cells to expand and differentiate after stimulation with cognate peptide . After 8–11 days of in vitro priming , we observed increased percentages of Mart1-specific CD8+ T cells when expanded from PBMCs of cHCV patients as compared to those from HD ( cHCV vs . HD , Day 8 , M-W p=0 . 02 , cHCV vs HD , Day 11 , M-W p=0 . 009 , Figure 6A and B; individual FACS plots for all donors are provided in Figure 6—figure supplement 1 ) . The positive impact of chronic infection on naïve T cell expansion was titratable , with more striking differences in the proportion of MP cells after expansion observed when cells were primed with high doses of peptide ( Sprent and Surh , 2011 ) ( Day 8 , M-W p=0 . 03; Figure 6—figure supplement 2 ) . Finally , we found that the Mart1-specific CD8+ T cells generated from cHCV patients express slightly higher amounts of granzyme B ( representative example from three cHCV and three HD is shown in Figure 6C; MW p=0 . 02 , Figure 6D ) . Interestingly , a tendancy for similar differences in Granzyme B expression could be seen in freshly isolated Mart1-specific CD8+ T cell populations in cHCV patients ( M-W p=0 . 09 as compared to HD , Figure 6—figure supplement 3 ) . Hyperreactive preimmune repertoire was further supported by our observation of increased secretion of IFNγ by freshly isolated and antigen-restimulated cells – shown for Mart1 , hTERT and CMV peptides in tumor-free , CMV-seronegative cHCV donors ( 2-way Anova p=0 . 0002; Figure 6E and F ) . 10 . 7554/eLife . 07916 . 021Figure 6 . Memory phenotype cells can be expanded to generate robust CD8+ T cell responses . ( A ) Examples of Mart1-specific populations expanded from HD and cHCV patients after 8 days of in vitro priming ( IVP ) with low ( 10–8 , upper line ) and high ( 10–6 , bottom line ) doses of Mart1 peptide . FACS plots from all donor tested are provided in Figure 6—figure supplement 1 . ( B ) Percentages of Mart1-specific cells expanded after 8 and 11 days of IVP with low and high doses of Mart1 peptide as in A . Proportions of MP cells within those expanded populations are indicated in Figure 6—figure supplement 2 . ( C ) Representative histograms of intracellular granzyme-B expression by Mart1-specific T cells expanded from 3 HD and 3 cHCV after 8 days of IVP with high doses of peptide as in A . ( D ) Percentages of granzyme-B-expressing Mart1-specific T cells expanded from HD and cHCV patients after 8 days of IVP with low and high doses of Mart1 peptide . Baseline percentages are indicated in Figure 6—figure supplement 3 . ( E ) Representative examples of IFNγ detection intracellularly after in vitro restimulation with CMV or Mart1 peptides in CMV seronegative , tumor-free HD and HCV patients . IFNγ-positive populations are overlaid on total CD8+ T cells . ( F ) Percentages of cells with IFNγ-positive staining after Mart1- , hTERT- , and CMV- in vitrorestimulation in HD and cHCV patients . sn , seronegative; sp , seropositive . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 02110 . 7554/eLife . 07916 . 022Figure 6—figure supplement 1 . FACS plots of Mart1-specific populations expanded in vitro from cHCV and HD . FACS plots of Mart1-specific populations ( A ) and naïve/memory subseting of gated tetramer-positive cells ( B ) expanded from cHCV and HD patients after 8 days of IVP with low and high doses of Mart1 peptide as in Figure 6A . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 02210 . 7554/eLife . 07916 . 023Figure 6—figure supplement 2 . Increased proportions of memory-phenotype cells within Mart1 populations expanded from cHCV patients . ( A ) % of memory-phenotype Mart1-specific T cells generated from HD and cHCV after 8 days of IVP as in Figure 6A . ( B ) FACS plots of naïve/memory subseting of Mart1-specific T cell populations expanded from cHCV and HD patients after 8 days of IVP with low and high doses of Mart1 peptide as in Figure 6A . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 02310 . 7554/eLife . 07916 . 024Figure 6—figure supplement 3 . Baseline proportions of granzyme B-expressing cells in cHCV patients and HD . Percentages of cells expressing granzyme-B intracellularly at baseline within Mart1-specific CD8+ T cell population ( without peptide stimulation ) in HD and cHCV patients . DOI: http://dx . doi . org/10 . 7554/eLife . 07916 . 024 Together , our results favor a model where low levels of CD5 on naïve-phenotype cells from cHCV donors allow low-affinity interactions with non-cognate antigens to result in T cell differentiation , thereby providing an explanation for the increased frequency of MP cells in cHCV patients . Additionally , our data indicate that qualitative alterations of the CD8+ T cell preimmune repertoire in cHCV patients may result in a boosted response to cognate immune stimulation . Testing our ability to identify preimmune repertoire perturbations in other clinical conditions , we collected 18 cHBV patients using standard sampling procedures . We found normal percentages of CD3+CD8+ T cells ( data not shown ) , and decreased percentages of bulk naïve CD8+ T cells ( MW p=0 . 009 , Figure 5—figure supplement 5A ) . With the limited amount of cells available , we focused our analysis to ( i ) absolute count and phenotype of Mart1-specific T cells , ( ii ) CD5 expression on bulk naïve T cells , and ( iii ) response to TCR cross-linking . We demonstrated lower absolute numbers of Mart1-specific CD8+ T cells in HBV patients ( MW p=0 . 003 , Figure 5—figure supplement 5B ) and increased frequencies of MP Mart1-specific cells ( MW p=0 . 03 , Figure 5—figure supplement 5C ) as compared to HD , but ( ii ) similar levels of CD5 expression ( MW p=ns , Figure 5—figure supplement 5D ) , and ( iii ) a similar a activation profile of bulk naïve T cells as compared to HD ( MW p=ns , Figure 5—figure supplement 5E ) . These results indicate that different persistent viral infections of the liver can trigger distinct preimmune repertoire perturbations . Additional studies will be required to fully evaluate the heterogeneous disease pathogenesis of HBV infections as reflected by the observed immune phenotypes . Our study provides novel evidence for chronic viral infection as a cause of CD8+ T cell preimmune repertoire dysregulation . Specifically , we demonstrated that naïve CD8+ T cells are dysregulated in the context of cHCV , marked by ( i ) decreased sjTRECs levels , ( ii ) a restricted Vβ repertoire , and ( iii ) a lower threshold for TCR engagement . Prior examples suggestive of preimmune repertoire perturbations have been documented in humans . An increased threshold for TCR activation in naïve CD4+ T cells in elderly persons has been proposed as participating in the diminished response to vaccination that occurs with increasing age ( Li et al . , 2012 ) . Conversely , a decreased threshold for TCR activation , secondary to sustained cytokine production , leads to diverse autoimmune manifestations in rheumatoid arthritis patients ( Singh et al . , 2009; Deshpande et al . , 2013 ) . With respect to chronic infection , functional defects in the naïve T cell compartment have also been documented in HIV-infected individuals , with non-cognate activation of T cells correlating with disease progression ( Favre et al . , 2011 ) . One major caveat for these studies is that their analysis was limited to global dysregulation of the bulk naïve T cell repertoire . The challenge of studying perturbations of antigen-specific populations is their low precursor frequency . Taking advantage of the possibility to study viremic vs . cured patients , we chose to investigate this question in cHCV patients . Analyzing rare ( i . e . , frequency = 10–7 - 10–5 ) antigen-specific inexperienced CD8+ T cells populations , we show increased proportions of memory-phenotype cells in cHCV patients , and demonstrate that this correlates with naïve T cells being hyperreactive to TCR signaling in the context of the chronic infection . Despite these altered phenotypes , the absolute number of antigen-specific cells was comparable to healthy donors . Of note , cHCV patients are not thought to experience altered thymic output . As such , our findings provide direct evidence that MP antigen-specific T cells can arise in non-lymphopenic humans . It has been suggested that a high degree of cross-reactivity with environmental antigens is the trigger for differentiation and MP conversion ( Sprent and Surh , 2011 ) . This finding has been reported for human viral peptide / MHC restricted CD4+ T cells in unexposed donors ( Su et al . , 2013 ) . While cross-reactivity is a possible explanation for our findings , we demonstrate in cured patients that the antigen-specific inexperienced T cell populations are restored to a naïve phenotype . This result will need to be confirmed in a larger longitudinal cohort study . It favors an alternative model , where homeostatic proliferation accounts for the perturbed naïve T cell repertoire in cHCV patients . Supporting this conclusion , we note the evidence for rapid reversibility to a healthy preimmune repertoire after transient lymphopenia ( Jones et al . , 2013 ) . Consistent with our findings , Jones et al . studied multiple sclerosis patients and showed an anti-CD52 ( also known by alemtuzumab ) treatment-induced narrowing of the Vβ repertoire and the dilution of sjTREC after treatment , with a complete restoration of normal levels two years post-therapy ( Jones et al . , 2013 ) . Infection and inflammation is known to lower the threshold of TCR signaling in memory T cells , making them more sensitive to activation ( Richer et al . , 2013 ) . This effect is mediated by inflammatory cytokines ( Raué et al . , 2013 ) . Our results extend this concept to naïve T cells and introduce CD5 downregulation as a mechanism for hyperreactivity . CD5 tunes the TCR signaling threshold in peripheral T cells , with naïve cells expressing higher levels than central memory or effector T cells ( Tabbekh et al . , 2013 ) . In mice , Hawiger et al demonstrated that anti-CD5 blocking antibodies , or the use of CD5-/- transgenic MOG-specific T cells , resulted in higher sensitivity to experimental autoimmune encephalitis ( Hawiger et al . , 2004 ) . In B cells , CD5 has also been shown to regulate activation and low CD5 expression correlates with high sensitivity to activation induced cell death ( Tabbekh et al . , 2013 ) . In line with these findings , we demonstrate an increased sensitivity of CD5lo naïve CD8+ T cells to TCR ligation in cHCV patients . We further provide direct evidence that this hypersensitivity phenotype can be partially reproduced in HD by blocking CD5 . While not evaluated in our patient cohort , we propose that elevated levels of inflammatory cytokines may be responsible for the altered CD5 expression on naïve cells ( Park and Rehermann , 2014 ) . Finally , we applied our strategy for evaluating preimmune repertoire perturbations to other clinical conditions , and demonstrate in cHBV patients that a distinct persistent infection of the liver triggers a different preimmune signature . This observation may be related to the differing innate inflammation induced as a result of infection ( Duffy et al . , 2014 ) . The combination of low levels of CD5 and increased proportions of MP in inexperienced antigen-specific populations may provide a compounded effect , resulting in a highly reactive CD8+ T cell compartment . We provide evidence here that chronic HCV infection facilitates the generation of robust self-specific responses from the pool of preimmune cells . Given the important role for cellular immunity in the pathogenesis of autoimmune manifestations ( Palermo et al . , 2001 ) , we speculate that circulating self-reactive effector CD8+ T cells may contribute to the systemic immune activation observed during chronic HCV infection , and account for some of the extra-hepatic autoimmune-like manifestations ( Lee et al . , 2012 ) . If our prediction is correct , the ability to restore a physiologically normal preimmune repertoire in cured patients may thus justify early treatment as a means to limit immune-mediated manifestations of the disease . Further investigation in longitudinal cohorts is warrented to confirm these hypotheses , as well as assess the impact on the generation of non-self-specific responses ( e . g . , in the context of vaccination ) . In summary , our study demonstrates that naïve CD8+ T cells are dysregulated during cHCV , with marked perturbations of the preimmune repertoire . Specifically , low levels of CD5 at the surface of naïve T cells , and high proportions of memory-phenotype cells represent two mechanisms by which antigen-inexperienced CD8+ T cells are susceptible to stimulation and antigen-induced expansion . These findings should be considered when designing future immunotherapeutic strategies . 29 cHCV , 37 SVR , and 18 cHBV patients were included ( Table 1 ) . All subjects were followed in the Liver Unit of Hôpital Cochin ( Paris , France ) or the Department of Internal Medicine II ( Freiburg , Germany ) . French samples were obtained as part of study protocol C11-33 approved by the INSERM clinical investigation department with ethical approval from the CPP Ile-de-France II , Paris ( ClinicalTrials . gov identifier: n° NCT01534728 ) . German samples were obtained in the University Hospital Freiburg according to regulations of local ethic committee . Both study protocols conformed to the ethical guidelines of the Declaration of Helsinki , and patients provided informed consent . Patient peripheral blood mononuclear cells ( PBMCs ) were obtained from leukapheresis , or whole blood collections . Healthy donor PBMCs were obtained from buffy coat preparations or whole blood collections ( Etablissement Français du Sang , France ) . PBMCs were processed within 5 hr of their collection . They were used either fresh , or frozen and thawed when needed – and in both cases , cells were rested overnight in serum-free RPMI at 37° before performing functional studies . Absolute lymphocyte counts were determined on the day of collection at the hospital laboratories for HCV and SVR patients , and on fresh samples using AccuCheck Counting Beads ( Life Technologies , France ) for healthy donors . For all samples , PBMCs were isolated by Ficoll-Paque gradient separation ( GE Healthcare , France ) after 1:4 dilution in RPMI1640 ( Gibco , Life Technologies , France ) and controlled for viability ( >90% ) . Molecular HLA-A and –B loci typing were determined using extracted genomic DNA according to standard clinical laboratory procedures ( Hôpital St Louis , Paris , France ) . Photocleavable-HLA-A02:01 multimers were constructed using peptide exchange technology as previously described ( Jenkins et al . , 2010; Toebes et al . , 2006; Altman and Davis , 2003; Hadrup et al . , 2009 ) . Briefly , heavy chain of HLA-A0201 and β2m were produced separately in E . coli . Refolding was achieved by diluting each subunit in buffer containing the A0201 UV photocleavable peptide ( KILGFVFJV , 95% purity , PolyPeptide , France ) ( Toebes et al . , 2006; Blattman et al . , 2002 ) . After biotinylation with recombinant BirA enzyme ( Avidity , Denver , USA ) , monomers were selected by size exclusion chromatography ( Akta Purifier , GE Healthcare , France ) and stored at -80°C until use . For specific peptides , synthetic 9mer were purchased ( 75% purity , BioMatik , Toronto , Canada ) : MART126-35 ( Leu27 ) ( ELAGIGILTV ) , hCMV pp65495-503 ( NLVPMVATV ) , hTERT1572-580 ( RLFFYRKSV ) , Ebola NP202-210 ( RLMRTNFLI ) ( Sundar et al . , 2007 ) , and HIV-1 Gag p1777-85 ( SLYNTVATL ) . 200 μM peptides were exchanged on calculated amounts of monomers ( 2 μM final concentration ) for 1h under UV-lamp ( 366nm , 28W , Chromacim , France ) . Titrated amounts of PE or APC-streptavidin ( Invitrogen , France ) were added . After incubation with D-biotin ( 25 μM final , Sigma , France ) , fluorescently labeled multimers were kept in the dark at 4°C until use . Mart1 PE pentamers were purchased ( ProImmune , UK ) as quality control for our in-house production . TAME was performed as previously described ( Alanio et al . , 2010; Alanio et al . , 2013; Kyewski and Klein , 2006 ) . Briefly , purified PBMCs ( 2x107 to 4x108 ) were incubated with FcR blocking reagent ( Miltenyi , France ) , then stained with PE and/or APC pMHC-multimers at 20nM final concentration for 30 min . Samples were incubated with anti-PE-microbeads and positive selection was performed using MS MACS separation columns ( Miltenyi , France ) . Unbound cells ( “Depleted” fraction ) were collected . Bound cells ( “Enriched” fraction ) were eluted . As previously published ( Alanio et al . , 2010 ) , tetramer-positive populations were gated as LiveDump-CD8+Tetramer+ cells . To approximate the number of the epitope-specific T cells within each sample , we used a calculation previously described by Moon et al ( Arstila et al . , 1999; Moon et al . , 2009 ) . Precursor frequency is defined as the number of tetramer-positive events in the “Enriched” fraction divided by the number of total CD8+ in the sample . PBMCs were stained with titrated amounts of monoclonal Ab ( mAbs ) obtained from BD Biosciences , Biolegend , or eBiosciences ( Supplementary file 1 ) . Live/Dead Fixable Aqua reagent ( Life Technologies , France ) was included at the same incubation step ( dilution 1/200 ) in order to exclude dead cells . For PhosFlow experiments , cells were stained with surface Abs for 20 mins , then fixed with PFA 3 . 2% for 10 min at 37°C , and permeabilized by addition of 90% methanol on ice . Intracellular staining of granzyme B was performed using Transcription Factor Buffer Set ( BD Biosciences ) . Samples were acquired using an LSR Fortessa cell analyzer ( BD Biosciences , France ) . Data were analyzed using FACS DIVA 6 . 0 ( BD ) and FlowJo 8 . 8 . 7 ( Tree Star ) softwares . Where indicated , stained cells were sorted using a FACS AriaII ( BD ) in a P2+ facility . Human PBMCs were rested overnight in RPMI 1640 GlutaMAX-10% pooled human serum . Cells were plated at 5x106/mL in 24-well plates , and restimulated in vitro with MART126-35 ( Leu27 ) , hCMV pp65495-503 , or hTERT1572-580 peptides ( 10 μM final ) . After 1 hr of stimulation , GolgiPlug ( 5 μg/mL final , BD ) was added . After 7 hr , cells were stained for surface Abs , then intracellularly using standard procedures ( Cytofix/Cytoperm; BD ) . One million FACS-sorted T cells were lysed in TRIzol Reagent ( Life Technologies , France ) . Genomic DNA was extracted following manufacturer’s instructions . Quantification of thymic sjTREC was performed by RT-PCR ( ABI PRISM7700; Applied , France ) ( Obar et al . , 2008; Moon et al . , 2007; Moon et al . , 2009; Talvensaari et al . , 2002 ) . Data were expressed per 150 000 cells , after normalization for the albumin genomic copy number . After TAME , 1500 naïve and memory Mart1-specific CD8+ T cells were sorted into RLT Buffer ( Qiagen , France ) . Total RNA was extracted ( Qiagen Microkit ) . cDNA were generated using the Supercript II enzyme ( Invitrogen , France ) . RT-PCR reactions , thermal cycling conditions , calculations for relative usage of each Vβ family , and immunoscope profiles were performed as previously described ( Alanio et al . , 2013; Bouvier et al . , 2011 ) ( Supplementary file 2 ) . One million PBMCs were stained for T cell surface markers and a set of three Abs directed against TCR-Vβ families ( Supplementary file 3; IOTest Beta Kit , Beckman/Coulter , France ) . TCR-Vβ families were classified in increasing order of percentage usage . The Lorenz curve was constructed as a graphical representation of the diversity of the repertoire ( Alanio et al . , 2010; De Maio , 2007 ) . After ordering Vβ chains by abundance , from lowest to highest , the Lorenz curve shows the cumulative distribution : for a given percentage ( x ) of the 24 Vβ chains , it indicates the proportion of the T cell population which have Vβ chains that are among the 24 * x% least abundant ones . Gini coefficient was calculated as the ratio of « area between the line representing equal use of all Vβ chains ( dotted line ) and the observed Lorenz curve » to « total area below the line representing equal use » . As such , the higher the Gini coefficient , the more unequal the distribution is . 96-well plates were coated overnight with biotin anti-human CD3 and anti-human CD28 ( 1 μg/mL and 0 . 5 μg/mL final concentration , respectively ) . Unstimulated and PMA/Ionomycin conditions ( 50 ng/mL and 1 μg/mL respectively ) were used as negative and positive controls . Measurements for T cell activation included: PhosFlow , as described above; and phenotypic activation , as measured by expression of CD25 and CD69 following a 24–48 hr culture . For experiments with blocking CD5 , cells were preincubated with 5μg/mL anti-human CD5 for 1 hr before being plated for TCR stimulation . PBMCs from HLA-A0201-positive donors were primed in vitro using the ELAGIGILTV ( ELA ) peptide derived from Melan-A/MART-1 antigen ( residues 26–35 ) , using previously published method with minor adaptations ( Martinuzzi et al . , 2011 ) . Briefly , thawed PBMCs were resuspended in AIM medium ( Invitrogen ) , plated at 5x106 cells/well in a 24-well tissue culture plate , and stimulated with 10nM ( low dose , 10–8 ) or 1 µM ( high dose , 10–6 ) of Mart1 peptide ELAGIGILTV in the presence of GM-CSF ( 0 . 2 μg/ml , R&D Systems ) . After 24 hr , dendritic cells maturation was induced by the addition of a cytokine cocktail comprising TNF-α ( 1000 U/mL ) , IL-1β ( 10 ng/mL ) , IL-7 ( 0 . 5 ng/mL ) and PGE2 ( 1 μM ) ( R&D Systems ) . On day 2 , fetal calf serum ( FCS; Gibco ) was added to reach 10% by volume per well . Fresh RPMI-1640 ( Gibco ) enriched with 10% FCS was used to replace the medium every 3 days . Frequency and phenotype of ELA-specific CD8+ T-cells were determined on day 8–11 . CMV serology was determined on plasma samples from HD and HCV patients by ELISA for CMV-specific IgG Abs ( Liaison XL , Diasorin ) . Donors were defined as seropositive for CMV if specific IgG>13 U/mL , and seronegative if IgG<13 U/mL . Statistics were performed using Prism 5 , GraphPad software ( San Diego , USA ) . Single continuous variable data were analyzed by Mann-Whitney ( MW ) , or Kruskal-Wallis ( KW ) followed by Dunn’s Multiple Comparison Test . Multi-feature continuous variable data sets were analyzed by Anova and Bonferroni post-test . Paired non-parametric datasets were analysed using Wilcoxon’s statistical test . Correlation were analysed using Spearman linear regression . For all these tests , a cut-off value of p≤0 . 05 was chosen ( p≤0 . 05; p≤0 . 01; *p≤0 . 001 ) . For longitudinal data on SVR patients , after linearisation of the data by squaring , a mixed model was fitted with a fixed time effect and random patient effects for both the slope and the intercept . p-value gives significance for the fixed slope effect . The R function lme ( package nlme ) was used .	Long-lasting or “chronic” infections massively perturb the immune system as a way to favor their own growth . In particular , they can stop T cells – a subtype of immune cells that help to destroy viruses – from working well . For example , HIV and hepatitis C viruses can overwork T cells and cause them to die . This can make individuals vulnerable to other infections . In healthy people , T cells that have participated in the fight against particular infections continue to live to provide a memory of those past infections . Groups of “naïve” T cells that have not yet encountered an infected cell also patrol the body , ready to respond to infections by a new virus . There are relatively few virus-specific naïve T cells in the body , so until recently it has been hard to study them . As a result , researchers know little about how these cells are affected by long-lasting infections , and whether chronic infection affects our capacity to fight unrelated infections . Alanio et al . have now used a highly sensitive technique to compare naïve T cells found in the blood of three groups of people: those with chronic hepatitis C infections , those who have been cured of a chronic hepatitis C infection , and healthy people . This revealed that the naïve T cells are negatively affected by chronic hepatitis C infections , and become hypersensitive: they get easily overexcited , which can lead to their death . This compromises the immune defenses at the moment they are most needed . Closer inspection showed that the naïve T cells of patients with hepatitis C are hypersensitive because they have less of a protein called CD5 on their surface . This protein acts as a natural brake for the T cells , and thus having less results in the T cells mounting stronger immune responses . Although this might be beneficial when fighting certain infections , this may also account for conditions where T cells attack healthy tissues . Finally , Alanio et al . found evidence that people who have been cured of a chronic hepatitis C infection recover a healthy set of naïve T cells within two years . Treating patients as soon as an infection is diagnosed therefore has several benefits: as well as clearing the virus , this will reset the immune system balance and reduce the damage that hyperactive immune cells cause to the body . The results also have implications for vaccinations , which work by pushing naïve T cells to arm themselves against a particular virus . The discovery that naïve T cells are hypersensitive in patients with hepatitis C suggests that we may need a distinct strategy for efficiently vaccinating these patients . It is indeed possible that standard vaccines – tested in groups of healthy people – may result in unexpected and unwanted immune responses in individuals with hepatitis C . These open questions will be addressed in further studies . It will also be of interest to know if other chronic viruses have the same ability to alter the activity of naïve T cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2015	Bystander hyperactivation of preimmune CD8+ T cells in chronic HCV patients
Morphogen gradients expose cells to different signal concentrations and induce target genes with different ranges of expression . To determine how the Nodal morphogen gradient induces distinct gene expression patterns during zebrafish embryogenesis , we measured the activation dynamics of the signal transducer Smad2 and the expression kinetics of long- and short-range target genes . We found that threshold models based on ligand concentration are insufficient to predict the response of target genes . Instead , morphogen interpretation is shaped by the kinetics of target gene induction: the higher the rate of transcription and the earlier the onset of induction , the greater the spatial range of expression . Thus , the timing and magnitude of target gene expression can be used to modulate the range of expression and diversify the response to morphogen gradients . The Nodal signaling pathway plays essential roles in animal development . Nodal signaling induces and patterns mesendoderm and establishes left-right asymmetry ( Conlon et al . , 1994; Shen , 2007; Grande and Patel , 2009; Schier , 2009; Duboc et al . , 2010; Shiratori and Hamada , 2014 ) . The Nodal signaling pathway regulates dozens of genes , ranging from transcription factors to cytoskeletal components , in order to pattern embryonic tissues ( Bennett et al . , 2007; Liu et al . , 2011; Fodor et al . , 2013 ) . In embryonic stem cells , Nodal signaling is required for self-renewal as well as specification of endoderm and mesoderm ( James et al . , 2005; Vallier et al . , 2005; Schier , 2009; Oshimori and Fuchs , 2012; Chen et al . , 2013 ) . Nodal signals can form concentration gradients and can act as morphogens ( Chen and Schier , 2001; Williams et al . , 2004; Müller et al . , 2012 , 2013; Xu et al . , 2014a ) . It is unclear , however , how different Nodal concentrations induce different target genes and give rise to different cell types . The classic morphogen threshold model postulates that Nodal signals are secreted from a source and form a concentration gradient that induces different fates in the target tissue according to local ligand concentration ( Ashe and Briscoe , 2006; Barkai and Shilo , 2009; Rogers and Schier , 2011 ) . According to this model , high-threshold genes require high levels of Nodal signaling and thus are expressed close to the source ( short-range genes ) , whereas low-threshold genes require lower levels of Nodal and are expressed at a greater distance from the source ( long-range genes ) . Studies of mesendoderm patterning by Nodal in fish and frog have provided five lines of evidence that support the concentration threshold model . First , Nodal signals are produced locally starting at mid-blastula stages , and by the beginning of gastrulation , cells overlapping or close to the Nodal source express endodermal markers , while cells farther away express mesodermal genes ( Feldman et al . , 1998; Sampath et al . , 1998; Gritsman et al . , 2000; Chen and Schier , 2001; Harvey and Smith , 2009 ) . Second , a gradient of activated Smad2 , the principal transducer of the pathway , peaks at the Nodal source ( Faure et al . , 2000; Yeo and Whitman , 2001; Harvey and Smith , 2009 ) with high levels of activated Smad2 in endodermal progenitors and lower levels in mesodermal precursors . Third , reduction of Nodal signaling during blastula stages leads to the absence of endodermal fates but leaves most mesodermal fates intact ( Schier et al . , 1997; Feldman et al . , 1998 , 2000; Gritsman et al . , 2000; Dougan et al . , 2003; Hagos and Dougan , 2007 ) . Fourth , ubiquitous low concentrations of Nodal induce mesodermal markers , whereas high Nodal concentrations induce endodermal markers ( Gritsman et al . , 2000; Thisse et al . , 2000; Dougan et al . , 2003 ) . Fifth , an ectopic source of Nodal can induce short- and long-range expression of endodermal and mesodermal markers , respectively ( Thisse et al . , 2000; Chen and Schier , 2001; Williams et al . , 2004; Müller et al . , 2012; Xu et al . , 2014a ) . These observations suggest that different concentration thresholds induce different gene expression patterns . In addition to the contribution of Nodal concentration to target gene induction , the timing of signaling affects Nodal interpretation . For example , the Nodal gradient is not static as signaling activity increases in range and amplitude between the initiation of Nodal expression and the onset of zebrafish gastrulation 2 hr later ( Harvey and Smith , 2009; Müller et al . , 2012 ) . Moreover , delayed activation or premature inhibition of Nodal activity affects mesendoderm patterning ( Gritsman et al . , 2000; Dougan et al . , 2003; Hagos and Dougan , 2007 ) . Two models have addressed how duration of exposure and changes in concentration contribute to Nodal interpretation . In the snapshot model , cells rapidly adapt their output to the increasing concentration of Nodal , regardless of the duration and history of exposure ( Rogers and Schier , 2011 ) . Indeed , increases in activated Smad2 levels are accompanied by an expansion of target gene expression domains ( Harvey and Smith , 2009 ) . In this model , the only role of time is to allow the gradient to expand and reach the thresholds that trigger the expression of short- and long-range genes ( Harvey and Smith , 2009 ) . The alternative ‘cumulative dose’ or ‘integration’ model postulates that the duration of Nodal signaling plays a critical role in Nodal interpretation . Cells adopt progressively more marginal fates with increasing duration of exposure to Nodal ( Gritsman et al . , 2000; Hagos and Dougan , 2007 ) . In this model , induction of long-range genes only requires Nodal for short periods of time , whereas activation of short-range genes depends on an extended period of exposure to high Nodal levels . It has therefore been suggested that the total cumulative dose of Nodal signaling determines the cell fate but it is unclear at which level in the pathway a cumulative dose would be measured ( Hagos and Dougan , 2007 ) . Studies of TGFβ signaling in other contexts have suggested additional mechanisms for the time-dependent interpretation of Nodal signaling . In Xenopus , analysis of signaling by Activin , a TGFβ signal related to Nodal , has suggested a ratchet model: the response to the signal is maintained once the ligand has bound the receptor . Indeed , a short pulse of Activin is sufficient to induce and maintain target gene expression several hours after the pulse ( Gurdon et al . , 1995 , 1998; Dyson and Gurdon , 1998; Bourillot et al . , 2002 ) . This molecular ‘memory’ has been shown to rely on the persistence of active receptor-ligand complexes ( Jullien and Gurdon , 2005 ) and allows changes in signaling output only in response to increasing Activin concentrations but not to decreasing concentrations . Cell culture studies have suggested that time-dependent Nodal interpretation is dictated by the dynamics of the signaling pathway ( Inman et al . , 2002; Xu et al . , 2002; Nicolás et al . , 2004; Schmierer and Hill , 2005; Guzman-Ayala et al . , 2009 ) . TGFβ signaling pathways operate through distinct steps: ligand binding to its receptor , phosphorylation and nuclear accumulation of Smad2 , and induction of target gene expression ( ten Dijke and Hill , 2004; Massagué , 2012 ) . Several studies have revealed parameters that affect the levels of activated Smad2 . For example , cultured human keratinocytes take approximately 60 min of ligand exposure to generate the maximum level of activated Smad2 ( Inman et al . , 2002 ) . Other studies have shown that the rates of Smad2 phosphorylation and nucleo-cytoplasmic transport affect signaling output ( Clarke et al . , 2006; Zi and Klipp , 2007; Schmierer et al . , 2008; Vizán et al . , 2013 ) or that the speed and frequency of TGFβ ligand presentation influences target gene response ( Sorre et al . , 2014 ) . These cell culture studies highlight the potential roles of signaling dynamics in target gene induction but it is unclear how these dynamics affect the response to Nodal in vivo . To distinguish between the numerous proposed mechanisms for Nodal morphogen interpretation , we studied the temporal and spatial dynamics of Smad2 activation and target gene induction in the early zebrafish embryo . We find that not only Nodal concentration and time of exposure but also the kinetics of target gene induction are key determinants of the response to Nodal morphogens . In particular , our study indicates that a target gene's transcription rate and onset of activation are major determinants of expression range , revealing previously unrecognized layers in the interpretation of morphogen gradients . Smad2 activation has been used as a read-out for Nodal signaling , but it has been unclear whether this transcriptional regulator is the main transducer of Nodal signaling in zebrafish ( Dick et al . , 2000; Jia et al . , 2008 ) . To test the role of Smad2 in Nodal signal transduction , we used TILLING ( Wienholds et al . , 2003 ) to recover a non-sense mutation in smad2 and generated embryos lacking maternal and zygotic Smad2 ( MZsmad2; Figure 1A , B ) . Endoderm and head and trunk mesoderm are absent in MZsmad2 embryos , a phenotype very similar or identical to Nodal loss-of-function mutants ( Figure 1C–E ) ( Feldman et al . , 1998; Gritsman et al . , 1999 ) . MZsmad2 mutants could be rescued by ubiquitous expression of wild-type Smad2 and GFP-Smad2 and the larval lethality of Zsmad2 mutants could be rescued to adulthood by a GFP-Smad2 transgene ( Figure 1F–H; Table 1 ) . Moreover , neither Nodal nor Activin displayed any activity in MZsmad2 mutants ( Figure 1I , J ) . These results demonstrate that Smad2 is an essential transducer of Nodal signaling during mesendoderm specification . 10 . 7554/eLife . 05042 . 003Figure 1 . Maternal Smad2 is necessary for mesendoderm specification by Nodal signaling . ( A ) Illustration of the Smad2 protein showing the position of the ENU-induced non-sense mutation . ( B ) Western blot against Smad2/3 on 24 hpf embryos of different genotypes for smad2 . MZ , maternal-zygotic homozygotes , Z−/− , zygotic homozygotes , Z+/− , zygotic heterozygotes . The pool of maternally contributed Smad2 protein persists for at least 24 hr in zygotic homozygous embryos while it is depleted in MZsmad2 mutants . ( C–J ) Phenotypic analysis of 36 hpf zebrafish embryos . ( C ) Wild-type embryo . ( D ) MZoep embryo: maternal-zygotic mutant for one-eyed pinhead ( oep ) , a cell surface protein required for Nodal signaling ( Gritsman et al . , 1999 ) . ( E ) MZsmad2 embryo . Msmad2 mutants display a very similar phenotype ( not shown ) . ( F ) MZsmad2 embryo rescued with 20 pg of smad2 mRNA . ( G–H ) MZsmad2 embryo rescued with 50 pg of gfp-smad2 mRNA ( brightfield ( G ) , epifluorescence ( H ) ) . smad2 mRNA appears to be more effective in rescuing the prechordal plate defects in MZsmad2 mutants as compared to gfp-smad2 mRNA . ( I ) MZsmad2 embryo injected with 5 pg mRNA for the zebrafish Nodal homolog squint . ( J ) MZsmad2 embryo injected with 5 pg mRNA for activin . Note that while Activin can activate the Nodal pathway in the absence of oep ( Gritsman et al . , 1999; Cheng et al . , 2003 ) , neither Squint nor Activin can activate the pathway in the absence of Smad2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00310 . 7554/eLife . 05042 . 004Table 1 . β-actin::GFP-Smad2 transgene rescues smad2/smad2 adult lethalityDOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 004smad2/+ X smad2/+; Tg ( GFP-Smad2 ) /+Genotype+/+Tg ( gfp-smad2 ) /++/+11 ( 37% ) 4 ( 15% ) smad2/+19 ( 63% ) 17 ( 60% ) smad2/smad20 ( 0% ) 7 ( 25% ) smad2/+ fish were crossed to smad2/+; Tg ( GFP-Smad2 ) /+ fish and their progeny was raised to adulthood and genotyped for smad2 and for Tg ( GFP-Smad2 ) . The only recovered adult progeny homozygous for smad2 contains a copy of the GFP-Smad2 transgene . The nuclear accumulation of GFP-Smad2 is a well-established reporter of TGFβ signaling ( Nicolás et al . , 2004; Xu and Massagué , 2004; Harvey and Smith , 2009 ) . This approach has been applied in embryos to visualize how the activated Smad2 gradient evolves over time ( Harvey and Smith , 2009 ) , but it has not yet been determined how Smad2 activity changes in individual cells and how cell movements might influence gradient interpretation ( Xiong et al . , 2013 ) ( Figure 2A ) . We therefore generated stable transgenic lines in which both GFP-Smad2 and histone H2B-RFP were ubiquitously expressed ( Figure 2B , C ) , and tracked GFP-Smad2 nuclear accumulation at the single cell level over time and space . 10 . 7554/eLife . 05042 . 005Figure 2 . Dynamics of Nodal signaling in vivo . ( A ) Illustration of Nodal signaling input–output relationship during blastula stage . Gray = yolk , white = blastoderm . Nodal is produced at the margin , diffuses and forms a gradient along the vegetal–animal axis . Nodal signaling induces a gradient of activated Smad2 , which induces long- and short-range target gene expression . ( B and C ) Maximal intensity projection of a confocal stack of a Histone 2B-RFP ( B ) , GFP-Smad2 ( C ) double transgenic embryo at 50% epiboly ( blue box in ( A ) ) . GFP-Smad2 strongly accumulates in the nuclei of cells close to the margin , the source of Nodal signals . ( D ) Heatmap of the nucleo-cytoplasmic ( NC ) ratio of GFP intensity from the embryo in ( B and C ) . Each dot represents the position of a cell ( overlay of five consecutive frames , 3-min intervals per frame ) . Each cell is color-coded according to its GFP NC ratio ( see Figure 2—figure supplement 2 for movement of cells ) . ( E ) Examples of single cell tracks at different locations along the vegetal–animal axis , showing changes in GFP-Smad2 NC ratio over time . The position of most cells relative to the margin remains constant during blastula stage . Cells close to the margin activate Nodal signaling earlier and at higher levels than cells at a distance from the margin . The short bursts observed in some cell tracks are caused by transient nuclear accumulation of GFP-Smad2 at the onset of nuclear envelope breakdown and are observed even in the absence of Nodal signaling . ( F ) NC ratio dynamics of tracked cells along the vegetal–animal axis . ( G ) Mean NC ratio values from ( F ) in 30 min bins . Note that the range and amplitude of the Smad2 activity gradient increase over the course of 90 min . Basal NC ratio is higher in younger embryos ( see Figure 2G , 3 . 5 hpf ) . Since this phenomenon is also observed in the absence of Nodal signaling ( MZoep mutants ) , the higher NC ratio is unlikely to reflect early Smad2 activation , but a higher nuclear import/export ratio of GFP-Smad2 during early development . ( H ) Time course of ntl ( upper panel ) and gsc ( bottom panel ) expression detected by RNA in situ hybridization . ntl begins to be induced as early as 3 . 5 hpf and its domain of expression expands over time to 100–120 µm from the margin; gsc begins to be induced 30 min later than ntl and its domain of expression expands to 50 µm from the margin . Close-up views of dorsal side , animal pole to the top . Right panel , heatmap for the grayscale intensity of in situ hybridization signals along the vegetal–animal axis showing the increase in range and intensity of ntl and gsc expression over time . See Figure 2—figure supplement 3 for comparison of probes and Figure 2—figure supplement 4 for independent validation of gsc and ntl expression domains using Seurat . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00510 . 7554/eLife . 05042 . 006Figure 2—source data 1 . Individual cell tracks and NC ratio . See Supplementary file 1 for description . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00610 . 7554/eLife . 05042 . 007Figure 2—figure supplement 1 . GFP-Smad2 as a sensor of Nodal activity in vivo . ( A ) GFP-Smad2 NC ratio as a function of distance from the margin . Black dots represent individual cells and the thick red line shows a polynomial fit . ( B ) Dose response analysis of Smad2 and GFP-Smad2 phosphorylation by Western blot following increasing amounts of activin mRNA in MZoep mutant . MZoep embryos lack endogenous Nodal activity but ectopic Activin can activate the Nodal pathway in the absence of oep . Numbers indicate amounts of pGFP-Smad2 and pSmad2 relative to GFP-Smad2 and Smad2 signals , respectively . ( C ) GFP-Smad2 NC ratio distribution following increasing amounts of activin mRNA in MZoep background . Both Smad2/GFP-Smad2 phosphorylation and the GFP-Smad2 NC ratio increase as the Activin levels increase . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00710 . 7554/eLife . 05042 . 008Figure 2—figure supplement 2 . Cell movements during blastula stages . Tracks of 50 random cells from the embryo shown in Figure 2D over a period of 30 min . Because of epiboly movements , cells move towards the vegetal pole , but their relative position from the margin barely changes ( see Figure 2E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00810 . 7554/eLife . 05042 . 009Figure 2—figure supplement 3 . Detection sensitivity of ntl and gsc by in situ hybridization . ( A ) Illustration of the experiment: a ntl-gsc fusion mRNA was injected at the one cell stage at four different concentrations ( 2 , 8 , 25 and 100 pg ) . Embryos were then fixed at the 128–256 cell stage and processed for in situ hybridization with either ntl or gsc probes . ( B ) Representative images of embryos hybridized with gsc ( left ) or ntl ( right ) probes at different concentrations . ( C ) Mean signal intensity and standard deviation as a function of injected fusion mRNA concentration for gsc and ntl ( n = 5 embryos ) . The gsc probe is less sensitive at low concentrations . p = 0 . 03 at 2 pg , p = 0 . 03 at 8 pg , p = 0 . 12 at 25 pg , and p = 0 . 14 at 100 pg ( Two-sample t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 00910 . 7554/eLife . 05042 . 010Figure 2—figure supplement 4 . Comparison of ntl and gsc expression pattern from single-cell RNAseq analysis . Gene expression patterns of ntl ( left ) and gsc ( right ) computed from single-cell RNAseq data spatially assigned to a 50% epiboly zebrafish embryo using Seurat ( Satija et al . , in press ) . Lateral view , dorsal side to the right . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 010 To enable accurate quantification , we determined how Smad2 phosphorylation , GFP-Smad2 phosphorylation and GFP-Smad2 nucleo-cytoplasmic ( NC ) ratio increased with increasing Nodal signaling ( Figure 2—figure supplement 1 ) . These calibrations established that the GFP-Smad2 NC ratio could serve as a read-out for pathway activity and confirmed the graded nuclear accumulation of Smad2-GFP along the vegetal–animal axis ( Harvey and Smith , 2009 ) ( Figure 2D , Figure 2—figure supplement 1 ) . To follow the trajectory of each cell , we tracked individual blastomeres over time ( Figure 2D–F , Figure 2—figure supplement 2 , Figure 2—source data 1 ) , determined their GFP-Smad2 NC ratio , and measured their distance from the margin . The resulting spatio-temporal map of Smad2 activity revealed that ( 1 ) the position of cells relative to the margin did not change extensively until the onset of gastrulation ( Figure 2E ) ; ( 2 ) cells close to the margin tended to activate Smad2 early and reached the highest levels of activated Smad2; ( 3 ) cells located farther away from the margin tended to activate Smad2 with a delay and the levels of activated Smad2 remained low ( Figure 2F , G ) . Thus , during the 1 . 5 hr from mid- to late-blastula stage a low-amplitude short-range gradient of activated Smad2 is transformed into a high-amplitude long-range gradient . To determine how the expression range of Nodal target genes correlates with Smad2 activity , we analyzed the expression of the long-range and short-range genes ntl and gsc , respectively ( Figure 2H , Figure 2—figure supplements 3 , 4 ) . ntl was first faintly detected in a few cells on the presumptive dorsal side of the embryo at the mid-blastula stage . Subsequently , its expression domain intensified and progressively extended animally until the onset of gastrulation . By contrast , gsc expression initiated ∼30 min later and remained confined to a narrow domain on the dorsal side ( Figure 2H ) . Comparing the spatio-temporal maps of Nodal target gene expression and Smad2 activity confirmed that the long-range target gene ntl was induced at both high and low levels of activated Smad2 , whereas the expression of the short-range gene gsc correlated with high Smad2 levels and sustained Smad2 activity . The spatiotemporal maps of Smad2 activity and target gene expression are consistent with previous proposals postulating that signaling thresholds determine target gene induction ( Harvey and Smith , 2009 ) . To directly test the threshold model of Nodal signaling , we wished to determine whether high Smad2 activity fully predicts the activation of both short- and long-range Nodal target genes . Using transplantation assays , we exposed GFP-Smad2 cells to high Nodal levels for different periods of time and analyzed the relationships between activated Smad2 levels and target gene expression ( Figure 3A ) . GFP-Smad2 NC ratios were similar in cells exposed to Nodal for either 1 or 2 hr ( Figure 3B , E ) . However , while the long-range gene ntl was expressed both after one or 2 hr of exposure to Nodal ( Figure 3D , G ) , the expression of the short-range gene gsc was only detected after 2 hr of exposure ( Figure 3C , F ) . These results are inconsistent with the strictest forms of the threshold model—the level of Smad2 activity at a given time predicts target gene expression—and reveal that the duration of signaling influences morphogen interpretation primarily at the level of target gene induction . 10 . 7554/eLife . 05042 . 011Figure 3 . Testing the threshold model . ( A ) Schematic of the transplantation experiment . Animal pole cells ( black circles ) from a GFP-Smad2 transgenic embryo were transplanted into the animal pole of a host embryo that had been injected with mRNA for squint , a zebrafish Nodal gene ( red ) . Host cells were unresponsive to Nodal because they were maternal-zygotic mutants for one-eyed pinhead ( MZoep ) , a cell surface protein required for Nodal signaling . This strategy prevents feedback loops and restricts target gene expression to donor cells . The developmental age of donor cells was matched to host embryos . Black parallelograms indicate imaging plane in subsequent panels . ( B–G ) Nodal signaling response of donor cells after 1 hr ( B–D ) or 2 hr ( E–G ) of exposure to Nodal . ( B and E ) Projection of confocal stacks of transplanted embryos and associated NC ratio ( mean ± std ) . Activated Smad2 levels are similar in both cases . See Figure 3—figure supplement 1 for time course of GFP-Smad2 N/C ratio . ( C and F ) RNA in situ hybridization for gsc . ( D and G ) RNA in situ hybridization for ntl . ntl is expressed after 1 ( n = 12/12 ) or 2 hr ( n = 16/16 ) of Nodal exposure while gsc signal in transplanted cells is only detected after 2 hr of exposure ( n = 1/15 at 1 hr , n = 12/14 after 2 hr ) . Images in B–G are from different embryos . Note that the differences in the duration of Nodal exposure uncouple the activated Smad2 level from target gene expression . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01110 . 7554/eLife . 05042 . 012Figure 3—figure supplement 1 . Time course of GFP-Smad2 NC ratio . Boxplot of the NC ratio of GFP-Smad2 cells over time . GFP-Smad2 cells ( n ∼50 ) were transplanted into a MZoep host embryo injected with 30 pg of squint mRNA ( as in Figure 3A , bottom panel ) , and the NC ratio was determined at different time intervals . Note that the NC ratio is never higher at any time between 3 . 5 and 5 . 5 hpf than as compared to 5 . 5 hpf . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01210 . 7554/eLife . 05042 . 013Figure 3—figure supplement 2 . Testing the ratchet model . ( A ) Top: Illustration of transplantation experiments to move cells from a region with Nodal ( red ) to a region without Nodal . Cells located at the margin of a 30% epiboly GFP-Smad2 transgenic embryo were transplanted to the animal pole of a stage-matched wild-type embryo . The NC ratio of GFP-Smad2 was measured over time by time-lapse microscopy ( n = 3 embryos ) . Bottom: boxplot of the NC ratio distribution at different time intervals after transplantation . Smad2 activity progressively decreases and reaches basal levels after 60 min . Two-sample t-tests p-values are indicated: NS , not significant; p ≤ 0 . 05; p ≤ 0 . 01 . ( B ) Top: Illustration of the transplantation experiment . Bottom: RNA in situ hybridization for ntl 30 min ( left ) or 120 min ( right ) after marginal cells were transplanted to the animal pole . After 30 min , the majority of ectopically transplanted marginal cells express ntl ( n = 12/14 transplantations ) whereas 2 hr after transplantation , 90% of embryos are devoid of ectopic ntl expression ( n = 17/19 transplantations ) . These results indicate that Nodal pathway activity cannot be maintained for prolonged periods in the absence of Nodal . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 013 The spatiotemporal maps of Smad2 activity and target gene expression support one prediction of the ratchet model—cells respond to increases in ligand concentration . To test the other tenet of the ratchet model—cells remember the highest ligand concentration they have been exposed to—we determined whether response is refractory to decreasing Nodal levels . We transplanted GFP-Smad2 cells from the blastula margin ( where Nodal concentration is high ) to the animal pole ( where Nodal concentration is low ) . Inconsistent with the ratchet model , Smad2 activity progressively decreased and reached basal levels after ∼60 min ( Figure 3—figure supplement 2A ) . Similarly , the expression of the long-range gene ntl disappeared over time ( Figure 3—figure supplement 2B ) . Thus , pathway activity and target gene expression cannot be maintained for extended periods after transient exposure to Nodal . Since the threshold and ratchet models do not fully account for Nodal morphogen interpretation , we sought an alternative model based on the biochemistry and biophysics of signaling . The changes in Smad2 activity and gene expression suggested that the kinetics of signal transduction and gene induction might be major factors in Nodal morphogen interpretation . To determine how time and concentration might translate into pathway activity and target gene response , we developed a mathematical description of the kinetics of Nodal signaling ( Chen et al . , 2010 ) ( Source code 1 ) . To reduce the complexity of the system and the numbers of free parameters , we focused on the three main steps in the pathway ( Figure 4A ) : ( 1 ) the diffusion of Nodal from a local source , ( 2 ) the Nodal-dependent phosphorylation of Smad2 ( pSmad2 ) , and ( 3 ) the pSmad2-dependent transcription of target genes . Three coupled differential equations were formulated to implement the kinetic model . All equations were based on standard reaction-diffusion models and mass-action kinetics . ( 1 ) ∂N∂t=P ( x , t ) +DN . ∇2 . N−k1 . N . 10 . 7554/eLife . 05042 . 014Figure 4 . A kinetic model for Nodal morphogen interpretation . ( A ) Diagram of the Nodal signaling pathway used for modeling ( left ) and coupled differential equations describing the changes of Nodal , activated Smad2 and target genes over time ( right ) . The Nodal ligand is locally produced , diffuses and via kinase receptors phosphorylates Smad2 . Phosphorylated Smad2 acts as a transcription regulator and binds to target genes to induce transcription . ( B ) Spatiotemporal gene expression patterns were simulated over 3 hr using the kinetic model . Each panel depicts the expression pattern resulting from a unique combination of the four free parameters involved in mRNA production ( transcription rate α , degradation rate β , dissociation constant Kd and Hill coefficient ) while other parameters are held constant . Note how changes in these parameters change the range of target gene expression . See Figure 4—figure supplement 1 for more extensive simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01410 . 7554/eLife . 05042 . 015Figure 4—figure supplement 1 . Screening for parameters regulating range and onset of target gene expression . ( A , C , E , G , I ) Contour plots of maximum range of expression after 3 hr of signaling . ( B , D , F , H , J ) Contour plots of onset of expression . ( A and B ) Combined effect of Tx rate ( x axis ) and Kd ( y axis ) with a Hill coefficient of 1 and a degradation rate of 10−5 s−1 . Tx rate and Kd contribute equally to the range of expression , except at low Kd and low Tx rate , where the range is more sensitive to changes in Tx rate . ( C and D ) Combined effect of Tx rate ( x axis ) and Kd ( y axis ) with a Hill coefficient of 4 and a degradation rate of 10−5 s−1 . The range becomes more sensitive to changes in Kd and less sensitive to changes in Tx rate . ( E and F ) Combined effect of Tx rate ( x axis ) and Kd ( y axis ) with a Hill coefficient of 1 and a degradation rate of 5 × 10−4 s−1 ( G and H ) Combined effect of Tx rate ( x axis ) and degradation rate ( y axis ) with a Hill coefficient of 1 and Kd of 6 nM . Range and onset of expression are only sensitive to changes in degradation rates when RNA half-life is very short ( half life <30 min ) . ( I and J ) Combined effect of Kd ( x axis ) and degradation rate ( y axis ) for a Hill coefficient of 1 and Tx rate of 0 . 1 count/s . Range and onset of expression are only sensitive to changes in degradation rates when RNA half-life is very short ( half life <30 min ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 015 Equation 1 describes the change of Nodal ( N ) levels over time . Nodal is produced from a source , diffuses and is degraded . Nodal levels at a distance from the source increase with increases in Nodal production ( P ) and diffusion ( D ) and with decreases in clearance ( k1 ) . ( 2 ) dSpdt=k2 . N . S−k3 . Sp . Equation 2 describes the change in activated ( phosphorylated ) Smad2 ( Sp ) levels over time . Smad2 activation is proportional to Nodal and non-activated ( non-phosphorylated ) Smad2 ( S ) concentrations . Thus , when Nodal concentration increases , activated Smad2 levels increase . Smad2 is deactivated ( de-phosphorylated ) at rate k3 . ( 3 ) dRNAtargetdt=α . SpnKdn+Spn−β . RNAtarget . Equation 3 describes the induction of Nodal target genes ( RNAtarget ) over time . For each target gene , levels of expression and dynamics of induction are defined by its maximal transcription rate ( α ) , degradation rate of its RNA ( β ) , and the affinity of pSmad2 for its promoter/enhancer ( Kd ) . The expression of a given target gene increases as α increases , Kd decreases , or the degradation rate decreases . As the concentration of pSmad2 increases , target gene transcription increases . The Hill coefficient n defines the cooperativity that modulates the sensitivity of the response . To test the effectiveness of the kinetic model in explaining and predicting Nodal gradient interpretation , we wished to run simulations with a realistic set of parameters . The effective diffusion coefficients and clearance rates of Nodals have been experimentally determined ( Müller et al . , 2012 ) , but other parameters of the system have not been measured . Exploring the contribution of each of these parameters in regulating target gene expression revealed that multiple parameter combinations could simulate the expression patterns observed in vivo ( Figure 4B and Figure 4—figure supplement 1 ) . In particular , the range of expression is affected most dramatically by changes in transcription rate , Kd or Hill coefficient . We therefore decided to constrain the parameter space by performing a detailed quantification of pSmad2 levels and target gene expression at different Nodal concentrations and durations of exposure . To precisely control the levels and timing of ligand input , we injected different amounts of recombinant mouse Nodal protein into the extracellular space of blastula embryos that lacked endogenous Nodal ligands ( Figure 5A ) . Nodal-injected embryos were collected at different time points and pSmad2 levels were determined by Western blotting . Target gene expression levels were measured by NanoString analysis using a custom-designed codeset ( Figure 5—source data 1 ) . This technique combines fluorescently barcoded probes with microimaging to detect and count hundreds of transcripts simultaneously in a single hybridization reaction and without amplification . It thus avoids the primer-specific amplification biases of qRT-PCR experiments and allows the direct measurement and comparison of transcript levels ( Geiss et al . , 2008; Su et al . , 2009; Strobl-Mazzulla et al . , 2010 ) . 10 . 7554/eLife . 05042 . 016Figure 5 . Constraining the kinetic model through in vivo measurements . ( A ) Experimental design: Wild-type embryos were injected at the one-cell stage with squint and cyclops MOs to knock down endogenous Nodal signaling . Morphant embryos were further injected either at 3 . 5 or 4 . 5 hpf with recombinant mouse Nodal protein at different concentrations in the extracelluar space . They were then incubated for different periods of time and processed for Western blot to determine pSmad2 levels or for NanoString to assess mRNA levels . ( B ) Dose-response ( left panel ) and time course ( middle panel ) of Smad2 activation at high ( 100 nM , dark green ) and low ( 10 nM , light green ) Nodal concentrations . Dots represent experimental data points and orange lines show model simulations with k2 = 3 . 13 × 10−6 nM−1s−1 and k3 = 1 . 8 × 10−4 s−1 . Black lines represent the 95% confidence intervals of data predictions . ( Right panel ) Simulated spatial distribution of Smad2 activation in a one-dimensional column of cells from 3 . 5 to 5 hpf in response to Nodal production from a source that extends from L = 0 to 25 µm . ( C ) Dose-response ( top ) and time course ( bottom ) data of 12 direct Nodal targets ( black dots ) . Given a specific set of parameters for each gene , the model ( red line ) recapitulates the dynamics of gene expression . Black lines represent fits encompassing the 95% prediction confidence intervals . gsc and ntl dynamics are highlighted within black boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01610 . 7554/eLife . 05042 . 017Figure 5—source data 1 . NanoString Probeset . Listed are all the genes and target sequences included in the probeset . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01710 . 7554/eLife . 05042 . 018Figure 5—source data 2 . Smad2 associated peaks after Nodal injection . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01810 . 7554/eLife . 05042 . 019Figure 5—source data 3 . Smad2 associated peaks after Nodal signaling inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 01910 . 7554/eLife . 05042 . 020Figure 5—source data 4 . FoxH1 associated peaks after Nodal injection . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02010 . 7554/eLife . 05042 . 021Figure 5—source data 5 . FoxH1 associated peaks after Nodal signaling inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02110 . 7554/eLife . 05042 . 022Figure 5—source data 6 . NanoString counts of Nodal target genes . Nanostring counts of the 61 direct and indirect Nodal target genes . See Source code 2 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02210 . 7554/eLife . 05042 . 023Figure 5—source data 7 . Nodal target genes identified in the NanoString codeset and their associated characteristics . ( A ) Simulated transcription rate associated with the best fit . ( B ) Range of transcription rates encompassing the 95% confidence intervals . ( C ) Simulated degradation rate associated with the best fit . ( D ) Range of degradation rates encompassing the 95% confidence intervals . ( E ) Simulated Kd associated with the best fit . ( F ) Range of Kd encompassing the 95% confidence intervals . ( G ) Simulated Hill coefficient associated with the best fit . ( H ) Range of Hill coefficient values encompassing the 95% confidence intervals . ( I ) Coefficient of determination value of the best fit . ( J ) Maximum mRNA levels obtained in NanoString experiments ( counts ) . ( K ) Maximum range of simulated spatial expression . The predicted spatial gene expression ranges only build on the simulated response to Nodal signaling kinetics . Regulation by other pathways is not taken into account and might change the in vivo expression patterns of these genes . ( L ) Degree of insensitivity to cycloheximide treatment: Strong ( + ) , mild ( ± ) or no ( − ) induction by Nodal after cycloheximide . ( M and N ) Amplitude of FoxH1 and Smad2 associated peaks: ++high , +medium , ±low , -background levels . See text and methods for details . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02310 . 7554/eLife . 05042 . 024Figure 5—figure supplement 1 . Characterization of direct Nodal target genes . Left panel: binding peaks of Smad2 and its associated transcription factor FoxH1 at dome stage after injection of zebrafish Nodal Squint mRNA ( red ) or after treatment with the Nodal signaling inhibitor SB505124 ( blue ) . Peaks called by the MACS algorithm are indicated ( gray blocks ) . Middle panel: NanoString count levels after injection of recombinant mouse Nodal protein ( red ) or after SB505124 treatment ( blue ) in the absence ( −cycloH ) or in the presence ( +cycloH ) of the translation inhibitor cycloheximide . Right panel: time course induction of target genes after Nodal injection at 3 . 5 hpf ( green ) or 4 . 5 hpf ( orange ) . In each case , ( 1 ) a specific Smad2 peak associated with a FoxH1 peak appears in the vicinity of the TSS upon Squint injection , ( 2 ) the Nodal-induced expression is not abolished by cycloheximide treatment , and ( 3 ) induction kinetics have similar trajectories independently of the stage of Nodal exposure , suggesting that these genes are direct targets of Nodal signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 024 To test whether the kinetic model can recapitulate and predict the temporal and spatial pattern of Nodal target gene expression , we ran simulations in a one-dimensional column of cells spanning the vegetal–animal axis . We let Nodal be produced and diffuse from a point source and used the parameters identified in the previous section to simulate the spatial Smad2 activation and transcriptional response over time . Using these conditions , the spatiotemporal pattern of activated Smad2 correlated well with the endogenous pattern of Smad2 activation ( Figure 5B ) : a short-range low-amplitude gradient was transformed over time into a long-range high-amplitude gradient , as observed in vivo for the GFP-Smad2 activity gradient ( Figure 2F , G ) ( Harvey and Smith , 2009 ) . The simulated spatiotemporal patterns of gene expression also fit well with the in vivo data . For example , in our simulations , ntl expression began in cells close to the margin 45 min after ligand production started , and the range of ntl continuously increased and reached cells located more than 100 µm away from the margin after 3 hr ( Figures 2H , 6A ) . By contrast , gsc expression was delayed and its range of expression was confined to cells close to the source ( Figures 2H , 6B ) . 10 . 7554/eLife . 05042 . 025Figure 6 . The kinetic model predicts gene expression patterns . Comparison of kinetic model simulations and RNA fluorescent in situ hybridization for ntl ( A ) , gsc ( B ) , foxa3 ( C ) , efnb2a ( D ) . Left panels: simulations of spatiotemporal expression patterns over 3 hr along a 200 µm-high column of cells using gene-specific parameters identified in the parameter screen . Right panels: RNA fluorescent in situ hybridization at 3 , 4 . 5 and 6 hpf . The size of the embryonic field is 100 µm wide and 200 µm high . Animal pole to the top . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 025 To test the predictive power of the kinetic model , we determined the expression patterns of genes that had not been analyzed in detail with respect to their range . As predicted by the simulations , foxa3 mRNA rapidly accumulated at high levels up to four cell tiers ( ∼60 µm ) from the source and then extended up to 80–100 µm by the onset of gastrulation ( Figure 6C ) . efnb2a mRNA also readily accumulated but was expressed in a narrower domain , as predicted from the simulations ( Figure 6D ) . These results reveal the power of the kinetic model in recapitulating and predicting the response of target genes to Nodal morphogen signaling . Since the kinetic model predicted target gene expression , we wished to determine which parameters were the major contributors to the range of gene expression . In the simulations described above ( Figure 4B and Figure 4—figure supplement 1 ) , genes whose Kd is low and maximal transcription rate is high are expressed at high levels and at long range . In contrast , the degradation rate influences the range of expression only when mRNA half-lives are very short ( Figure 4—figure supplement 1 ) . In agreement with the simulations , we found that genes that are highly induced by Nodal generally display a long range of expression ( Figure 7A , B ) . Strikingly , the maximal transcription rate , not the Kd or the degradation rate , was the best predictor of gene expression range ( Figure 7C , D ) . For example , while the degradation rate , Kd and Hill coefficient for the long-range gene foxa3 and the short-range gene gsc are very similar ( Figure 5—source data 7 ) , their maximal transcription rate , and therefore their maximal level of expression , differ by a factor of 20 . These results raise the possibility that the maximal transcription rate is a major contributor to target gene expression range: the higher the maximal rate of transcription , the longer the range . Moreover , multiple hypotheses analysis indicates that a model in which the maximal transcription rate is gene-specific and Kd is identical for all the genes performs better than a model where Kd is gene-specific and the maximal transcription rate is constant ( see Nodal signaling modeling section in ‘Materials and methods’ ) . Although the Kd may affect target gene response , these analyses indicate that the maximal transcription rate is a key parameter in determining the range of expression . 10 . 7554/eLife . 05042 . 026Figure 7 . Range of expression correlates with maximal transcription rate . ( A ) Bar graph showing the number of counts detected 90 min after injection of 100 nM of recombinant Nodal protein for the 61 Nodal-responsive genes ( direct and indirect ) identified in the NanoString codeset . Some of the genes used in this study are highlighted . ( B ) Scatter plot comparing maximal expression and simulated range . Highly expressed genes tend to have a longer range of expression . ( C and D ) Scatter plots comparing fitted Kd and maximal transcription rate ( Tx rate ) in relation to maximal expression ( C ) and in relation to simulated spatial range of expression ( D ) . Most Kd values remain in a narrow range while transcription rates spread over several orders of magnitude . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 026 To analyze additional predictions generated by the kinetic model , we asked whether a delay in gene induction might affect target gene response . To simulate this scenario , we extended the kinetic model with a co-factor that is produced later than and independently of Nodal and acts together with Smad2 to activate gene transcription . Simulations revealed not only the expected delay but also a reduced range of target gene induction: a long-range gene could be transformed into a short-range gene by introducing a delay in gene induction ( Figure 8A ) . 10 . 7554/eLife . 05042 . 027Figure 8 . Delayed onset of transcription restricts expression range . ( A ) Simulation of efnb2b expression using the kinetic model without ( left ) or with ( right ) a co-transcriptional activator Y . The dependence on Y delays the onset of efnb2b expression and reduces its range . ( B ) Top: Experimental design . Bottom: Time-course induction of efnb2b after injecting recombinant Nodal protein at 3 . 5 hpf ( green ) and 4 . 5 hpf ( orange ) . The induction kinetics of this gene are very slow , but the later Nodal is injected , the faster its induction . Note that counts for the expression of late target genes are higher after early injection compared to later injections . This effect might be due to the fact that after early injections phospho-Smad2 levels are high for a longer period before a gene becomes competent to respond as compared to late injections , when there is a shorter time window of high phospho-Smad2 levels . There might be a priming mechanism in which longer exposure to activated Smad2 increases gene expression when competence is reached . ( C ) RNA fluorescent in situ hybridization for efnb2b at 3 , 4 . 5 and 6 hpf . Expression of efnb2b is only detected at 6 hpf , although Nodal signaling and the expression of most other Nodal targets commences much earlier . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02710 . 7554/eLife . 05042 . 028Figure 8—figure supplement 1 . Characterization of co-regulated Nodal target genes . Left panel: binding peaks of Smad2 and the associated transcription factor FoxH1 at dome stage after injection of Squint mRNA ( red ) or after treatment with the Nodal signaling inhibitor SB505124 ( blue ) . Peaks called by the MACS algorithm are indicated ( gray blocks ) . Middle panel: NanoString count levels after mNodal injection ( red ) or after SB treatment ( blue ) in the absence ( −cycloH ) or the presence ( +cycloH ) of the translation inhibitor cycloheximide . Right panel: time course induction of target genes after Nodal injection at 3 . 5 hpf ( green ) or 4 . 5 hpf ( orange ) . Although these genes have Smad2/FoxH1 binding sites , their induction is abolished in the presence of cycloheximide , suggesting that an additional transcriptional co-regulator controls their transcription . Moreover , these genes are delayed in their onset of expression . The length of the delay depends on the stage at which Nodal is applied . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 02810 . 7554/eLife . 05042 . 029Figure 8—figure supplement 2 . Transcriptional competence regulates the onset and range of bra expression . ( A ) Concentration-dependent induction of flh ( left ) , bra ( middle ) and gsc ( right ) . bra and flh can be induced at low Nodal concentrations . ( B ) Induction dynamics of bra after injecting Nodal at 3 . 25 , 4 . 25 and 5 . 25 hpf as a function of time after injection ( left ) or as a function of absolute embryonic time ( right ) . Arrowheads in the right panel indicate the time of Nodal injection . bra can only be induced when the embryo has reached a specific embryonic stage . bra expression is detected by RT-qPCR . ( C ) Fluorescent RNA in situ hybridization with bra probe at 3 , 4 . 5 , 6 hpf . bra is only detected at 6 hpf in 5 cell tiers . DOI: http://dx . doi . org/10 . 7554/eLife . 05042 . 029 To determine whether such delayed genes might exist in vivo , we screened our NanoString data for Nodal targets whose induction upon Nodal exposure was delayed ( Figure 8B , Figure 8—figure supplement 1 ) . We discovered a small set of genes that were induced slowly after Nodal exposure at 3 . 5 hpf but more rapidly after exposure at 4 . 5 hpf ( Figure 8B , Figure 8—figure supplement 1 ) . For example , when Nodal was injected at 3 . 5 hpf , efnb2b was only induced after approximately 2 hr . By contrast , when Nodal was injected at 4 . 5 hpf , the delay in efnb2b induction was reduced by more than 30 min ( Figure 8B ) . In contrast , most other genes responded rapidly to Nodal exposure at either time point ( Figure 5—figure supplement 1 ) . This result revealed that the delay was gene-specific and did not reflect a general lack of competence to respond to Nodal signaling or activate gene expression . Similar to the canonical target genes , genes with delayed induction contained Smad2/FoxH1 binding sites and showed a clear response upon injection of Nodal ( Figure 8—figure supplement 1 ) but their induction was abolished in the presence of cycloheximide ( Figure 8—figure supplement 1 ) . These Nodal target genes are therefore likely to be regulated not only by Nodal signaling but additional factors . Strikingly , and as predicted by the delayed induction model , efnb2b expression in the embryo was detected only late ( 6 hpf ) and at a short range ( 5 cell tiers ) ( Figure 8C ) . Similarly , the Nodal target gene bra ( Martin and Kimelman , 2008 ) could only be induced shortly before gastrulation and , as predicted by the model , was expressed at low levels and at a short range ( Figure 8—figure supplement 2 ) . These results reveal that a delay in transcriptional response can be used to limit the range of morphogen-induced gene expression . Numerous models have been proposed to explain how morphogen gradients are interpreted to generate diverse gene expression patterns . To interrogate these models , we have taken a quantitative approach to measure the parameters that underlie gradient formation ( Müller et al . , 2012 ) and interpretation ( this study ) . This approach reveals that the kinetics of target gene induction is a major determinant of morphogen interpretation and suggests that a kinetic model of morphogen interpretation is better suited for the Nodal morphogen system than the prominent threshold and ratchet models . The kinetic model recapitulates the dynamics of Smad2 activation and reveals how distinct gene expression patterns can be generated: ( 1 ) the Nodal morphogen gradient forms and extends through diffusion; ( 2 ) rapid phosphorylation generates a corresponding gradient of activated Smad2; target genes are induced based on ( 3 ) their affinity for activated Smad2 , ( 4 ) their maximal transcription rate , and ( 5 ) their competence to respond to activated Smad2 . Thus , a target gene can be induced rapidly and at a long range by high transcription rate , high Smad2 affinity and early onset of induction . Conversely , low affinity for Smad2 , low transcription rate or late onset of induction generate short-range gene expression patterns . Our analysis identifies transcription rates and induction delays as two novel strategies to modulate morphogen interpretation . Previous models of morphogen interpretation have emphasized the importance of differential DNA ( or chromatin ) affinity: the higher the affinity for the transcription regulator , the longer the range of target gene expression . Our results do not contradict such models but reveal that in a rapidly developing system , the intrinsic rate of transcription of a target gene can be a major determinant of gene expression range: high affinity binding sites cannot overcome the limits imposed on gene expression range by low levels of intrinsic transcription . Similarly , delays in transcriptional onset can turn high affinity target genes into short-range genes . The Nodal morphogen system stands in contrast to two other well-studied morphogen systems , Sonic Hedgehog ( Shh ) and Bicoid . The Shh gradient patterns the dorsoventral axis of the mammalian neural tube over several days ( Briscoe and Ericson , 1999; Nishi et al . , 2009; Cohen et al . , 2013 ) . Although different concentrations of Shh elicit different transcriptional responses , feedback and cross-regulatory interactions are key determinants of patterning . For example , cells are progressively desensitized to Shh activity by upregulating patched1 ( Dessaud et al . , 2007 ) , and downstream targets regulate each other to generate discrete domains of expression ( Balaskas et al . , 2012 ) . Thus , in contrast to the Nodal system which rapidly establishes target gene expression patterns , the Shh system makes extensive use of feedback inhibition and cross-regulation . At the other extreme , the Bicoid morphogen has already formed a quasi steady-state gradient before its target genes can be activated during zygotic genome activation ( Driever and Nüsslein-Volhard , 1988; Gregor et al . , 2007; Porcher et al . , 2010 ) . Bicoid concentration and affinity to regulatory chromatin elements are important ( but not the sole [Ochoa-Espinosa et al . , 2009; Chen et al . , 2012] ) determinants of target gene expression along the anterior-posterior axis of the Drosophila embryo ( Driever et al . , 1989; Burz et al . , 1998 ) . Thus , in contrast to the Nodal morphogen gradient , which evolves and is interpreted continuously under pre-steady state conditions , the Bicoid morphogen system makes only limited use of temporal strategies to modulate target gene response . The influence of transcription rates and delays in morphogen interpretation raises the question how these processes might be regulated at the molecular level . Transcriptional delay might be achieved by a co-activator for target gene induction . Alternatively , a repressor might have to be eliminated for a target gene to become competent to respond . Transcription rates might be influenced by local chromatin structure , promoter strength , and by co-activators that boost or repressors that dampen the levels of target gene expression ( Li et al . , 2007; Lupien et al . , 2008; Hager et al . , 2009; Kanodia et al . , 2012; Peterson et al . , 2012; Coulon et al . , 2013; Oosterveen et al . , 2013; Foo et al . , 2014; Xu et al . , 2014b ) . In either case , our study suggests that the intensity and onset of target gene transcription can be major determinants in shaping morphogen gradient interpretation . Similar mechanisms might modulate other rapid and dynamic pattern formation processes ( Bolouri and Davidson , 2003; Lewis , 2003; de-Leon and Davidson , 2010; Oates et al . , 2012 ) . Fish were raised and maintained under standard conditions . Wild-type embryos were collected from TLAB in-crosses . MZoeptz57 embryos were obtained as previously described ( Zhang et al . , 1998; Gritsman et al . , 1999 ) . Mutations in the smad2 gene ( ENSDARG0000006389 , zv9 ) were screened for in the sperm of ENU-treated males by TILLING with primers encompassing exons 9 and 10 . Live embryos were embedded in 0 . 8% low melting point agarose on a glass bottom culture dish ( MatTek , Ashland , MA ) , with the marginal region facing the objective . The dish was filled with fish water ( Instant Ocean sea salt [0 . 6 g/l] in RO water , 0 . 01 mg/l methylene blue ) to prevent dehydration . Images were acquired on a PASCAL confocal microscope ( Zeiss , Germany ) using a 25× objective ( LCI Plan-Neofluar/0 . 8 ) equipped with a heated stage set at 28°C . Samples were simultaneously excited with an argon laser at 488 nm and a Helium laser at 546 nm . Four confocal planes were imaged at 3 µm intervals ( 512 × 512 size , 12-bit depth , line averaging eight times ) every 3 min for a period of 3 hr . Embryonic position of the recorded field was assessed morphologically at the end of the imaging session . Image stacks were processed using custom-made Matlab scripts to measure centroid localization of nucleus , nucleo-cytoplasmic ratio of GFP-Smad2 intensity , distance from the margin , and cell tracks ( see Supplementary file 1 ) . Channels of stacked confocal images were split in ImageJ and saved as grayscale TIFF image sequences ( 8-bit for H2B-RFP , 16-bit for GFP-Smad2 ) . H2B-RFP images were further converted to binary images , by applying a threshold using Otsu's method . Objects smaller than 20 pixels were then removed , and the resulting images were segmented using the Moore-Neighbor tracing algorithm modified by Jacob's stopping criteria . The centroid location and area of each nucleus were then extracted . The binary image of the H2B-RFP was used as a mask on the corresponding GFP-Smad2 image to extract nuclear only- and cytoplasmic only GFP-Smad2 signals . The ratio between the mean nuclear GFP intensity and the mean cytoplasmic GFP intensity was used to define Smad2 activity at the single cell level . In MZoep mutants ( Nodal insensitive ) , the mean NC ratio value is 1 . 19 ± 0 . 07 . Cell tracking was perfomed using the nearest-neighbor strategy based on the centroid position of each nucleus at different time frames . The nearest centroid of the next frame was selected as being part of the cell track if it was less than 10 pixels apart . This process was reiterated through all the frames to generate cell tracks . Based on visual checks of the resulting tracks , ∼90% of the tracks are estimated to be accurate . The distance between each centroid and the margin was measured at each time point . The position of the margin was defined using a user interface: the maximal projection of the H2B-RFP channel was displayed and six reference points were manually selected along the yolk-blastoderm boundary . The whole margin position was then extrapolated by fitting a polynomial curve . The fitted function was used to determine the distance of each centroid from the margin . Embryos for Smad2 and FoxH1 ChIP were collected at dome stage after 5 pg squint mRNA injection or after treatment with the Nodal signaling inhibitor SB505124 ( Sigma S4696 ) at 20 µM final . For FoxH1 ChIP , embryos were injected with 5 pg of FoxH1-flag mRNA at 1-cell stage , and anti-flag antibody was used for the pull down . For each ChIP , 800 embryos were collected and fixed in 1 . 85% formaldehyde for 15 min at 20°C . Formaldehyde was quenched by adding glycine to a final concentration of 0 . 125 M . Embryos were rinsed three times in ice-cold PBS , and resuspended in cell lysis buffer ( 10 mM Tris-HCl pH7 . 5/10 mM NaCl/0 . 5% NP40 ) and lysed for 15 min on ice . Nuclei were collected by centrifugation , resuspended in nuclei lysis buffer ( 50 mM Tris-HCl pH 7 . 5/10 mM EDTA/1% SDS ) and lysed for 10 min on ice . Samples were diluted three times in IP dilution buffer ( 16 . 7 mM Tris-HCl pH 7 . 5/167 mM NaCl/1 . 2 mM EDTA/0 . 01% SDS ) and sonicated to obtain fragments of ∼500 bp . Triton X-100 was added to a final concentration of 0 . 75% and the lysate was incubated overnight while rotating at 4°C with 25 µl of protein G magnetic Dynabeads ( Invitrogen ) pre-bound to an excess amount of antibody . Antibodies used were anti-FLAG M1 ( Sigma F3165 ) , anti-Smad2/3 ( Invitrogen , Grand Island , NY 51–1300 ) . Bound complexes were washed six times with RIPA ( 50 mM HEPES pH7 . 6/1 mM EDTA/0 . 7% DOC/1% Igepal/0 . 5 M LiCl ) and TBS and then eluted from the beads with elution buffer ( 50 mM NaHCO3/1% SDS ) . Crosslinks were reversed overnight at 65°C and DNA purified by the QIAquick PCR purification kit ( Qiagen ) . Libraries were prepared according to the Illumina sequencing library preparation protocol and sequenced on an Illumina HiSeq 2000 . ChIP-seq reads were mapped to the zebrafish genome ( UCSC Zv9 assembly ) and peaks were called using MACS ( Zhang et al . , 2008 ) . The goal of modeling Nodal signaling is to predict the range of expression of Nodal target genes in the embryonic blastula and to analyze the key parameters regulating gene response . Equations 1–3 were used to model the kinetics of Nodal signaling . ∂N∂t=P ( x , t ) +DN . ∇2 . N−k1 . N , dSpdt=k2 . N . S−k3 . Sp , dRNAtargetdt=α . SpnKdn+Spn−β . RNAtarget . P: Production rate of Nodal from the source , where P=γ . t1+t when x ≤ 25 µm and P = 0 when x > 25 µm . DN: Diffusion coefficient of Nodal . k1: clearance rate of Nodal . k2: activation ( phosphorylation ) rate of Smad2 . k3: de-activation ( de-phosphorylation ) rate of Smad2 . α: maximal transcription rate of Nodal target gene . β: degradation rate of Nodal target gene . Kd: effective dissociation constant of activated Smad2 for target gene enhancer . n: Hill coefficient . We assume that the pool of total Smad2 remains constant such that Stotal = S + Sp . We used two different scenarios to reflect the experimental set up: the ‘homogenous’ scenario , where ectopic Nodal ligand is injected uniformly into a Nodal depleted embryo , and the ‘spatial gradient’ scenario , where Nodal is produced locally on one side of a one-dimension column of cells . Most models of morphogen signaling and interpretation use large numbers of parameters and can thus suffer from overfitting . We thus considered six different models with different numbers of parameters to describe Smad2-dependent transcription of Nodal target genes and compared the probability that these models can generate the data . The same rate of Smad2 activation is shared among these models:dSpdt=k1 . N . S−k2 . Sp , where Sp , N and S are phosphorylated Smad2 , Nodal and non-phosphorylated Smad2 concentrations , respectively , with k1 = 3 . 1 × 10−6 nM−1s−1 and k2 = 1 . 8 × 10−4 s−1 . All models for RNA production have two terms: a pSmad2-dependent mRNA transcription rate , and a linear mRNA degradation rate . In Model 1 , we assume that the effective transcription rate is linearly proportional to pSmad2 concentration:M1: dRNAdt=α . Sp−β . RNA . In Model 2 , we assume that the transcription rate is regulated by a dissociation constant for pSmad2:M2: dRNAdt=SpKd+Sp−β . RNA . Model 3 is similar to Model 2 , with the addition of a maximum transcription rate coefficient:M3: dRNAdt=α . SpKd+Sp−β . RNA . Model 4 is similar to Model 2 , with the addition of a Hill coefficient and a fixed transcription rate coefficient . M4: dRNAdt=A . SpnKdn+Spn−β . RNA , A = 0 . 1 count/s , corresponding to the mode value of the maximal transcription rate coefficient distribution of our fully developed model ( see Table 1 ) . Model 5 is similar to Model 4 , except that in this case , the maximal transcription rate coefficient is let free while the dissociation constant is fixed:M5: dRNAdt=α . SpnCn+Spn−β . RNA , C = 6 . 7 nM , corresponding to the mode value of the dissociation constant distribution of our fully developed model ( see Table 1 ) . Finally , Model 6 is the fully developed model:M6: dRNAdt=α . SpnKdn+Spn−β . RNA . Assuming that our NanoString measurements {yi} are noisy with a standard deviation of {σi} , we can consider yi as a Gaussian random variable with a mean value f ( ti;θ ) of the underlying model containing a vector of parameters θ and a variance σi2 ( Bialek , 2012 ) . We thus have , P ( yi\|ti , θ ) =12πσi2exp[− ( yi−f ( ti;θ ) ) 22σi2] , andP ( {yi}\|{ti} , θ ) =∏i=1NP ( yi\|ti , θ ) . The probability of the data given the underlying model isP ( {ti , yi}\|θ ) =[∏i=1NP ( yi\|ti , θ ) ][∏iP ( ti ) ] . Given χ2=∑i\| ( yi−f ( ti;θ ) σi\|2 , P ( {ti , yi}\|θ ) =exp[∑i=1NlnP ( ti ) −12∑i=1Nln ( 2πσi2 ) −12χ2] . Therefore minimizing χ2 by fitting the parameters θ increases the probability that the model could have produced the data . However , different classes of models with different numbers of parameters whose values are unknown have to be considered . To determine the probability of the data given a class of models with unknown K parameters , an integration over all the possible values of the parameters , weighted by some prior knowledge , has to be computedP ( {ti , yi}\|class ) =∫ dKθP ( θ ) P ( {ti , yi}\|θ ) =∫ dKθP ( θ ) exp[−12∑i=1Nln ( 2πσi2 ) −12χ2 ( θ;{ti , yi} ) ][∏nP ( tn ) ] , where P ( θ ) is the probability of the a priori distribution of the parameters . χ2 is proportional to N , and we can writeP ( {ti , yi}\|class ) =exp[−12∑i=1Nln ( 2πσi2 ) ][∏nP ( tn ) ]∫ dKθe−Nf ( θ ) , wheref ( θ ) =12Nχ2 ( θ;{ti;yi} ) −1Nln P ( θ ) . We use a saddle point approximation such that∫ dKθe−Nf ( θ ) ≈e−Nf ( θ∗ ) ( 2π ) K2exp[−12ln det ( NH ) ] , where θ is the value at which f ( θ ) is minimized , and H is the Hessian matrix of the second derivatives at this point . Taking the negative log probability of the data given the model class , we have−ln P ( {ti , yi}\|class ) ≈∑i=1Nln ( 2πσi2 ) −∑i=1Nln P ( ti ) +12χmin2+ln P ( θ∗ ) −K2ln 2 π+12ln det ( NH ) . Since H is a K × K matrix , det ( NH ) =NKdet ( H ) , and we finally have−ln P ( {ti , yi}\|class ) ≈∑i=1Nln ( 2πσi2 ) −∑i=1Nln P ( ti ) +12χmin2+K2ln N+12ln det ( H ) +ln P ( θ∗ ) −K2ln 2π . The negative log probability measures the length of the shortest code for the data being generated given the class of models . This length depends on the sample size of the data , the number of parameters , the quality of the fit , and some prior on the parameters that we consider flat in our case . Therefore , the model giving the smallest value of the code is to be considered the best model explaining the data given the sample size . The NanoString data and associated noise ( which is ∼10% of the count value based on the analysis of the positive spikes across a cartridge ) are identical in all our models , so we are left to compareCMX=12χmin2+K2ln N+12lndet ( H ) −K2ln 2 π . Calculating the mean value across all genes , we found CM1 = 601 . 1 CM2 = 767 . 6 CM3 = 681 . 0 CM4 = 264 . 3 CM5 = 134 . 5 CM6 = 204 . 1 CM5 < CM6 < CM4 < CM1 < CM3 < CM2 . Thus , among all the six different models we considered , model M5 and M6 are the most probable models given the data , highlighting the importance of the maximal transcription rate .	How a cell can tell where it is in a developing embryo has fascinated scientists for decades . The pioneering computer scientist and mathematical biologist Alan Turing was the first person to coin the term ‘morphogen’ to describe a protein that provides information about locations in the body . A morphogen is released from a group of cells ( called the ‘source’ ) and as it moves away its activity ( called the ‘signal’ ) declines gradually . Cells sense this signal gradient and use it to detect their position with respect to the source . Nodal is an important morphogen and is required to establish the correct identity of cells in the embryo; for example , it helps determine which cells should become a brain or heart or gut cell and so on . The zebrafish is a widely used model to study animal development , in part because its embryos are transparent; this allows cells and proteins to be easily observed under a microscope . When Nodal acts on cells , another protein called Smad2 becomes activated , moves into the cell's nucleus , and then binds to specific genes . This triggers the expression of these genes , which are first copied into mRNA molecules via a process known as transcription and are then translated into proteins . The protein products of these targeted genes control cell identity and movement . Several models have been proposed to explain how different concentrations of Nodal switch on the expression of different target genes; that is to say , to explain how a cell interprets the Nodal gradient . Dubrulle et al . have now measured factors that underlie how this gradient is interpreted . Individual cells in zebrafish embryos were tracked under a microscope , and Smad2 activation and gene expression were assessed . Dubrulle et al . found that , in contradiction to previous models , the amount of Nodal present on its own was insufficient to predict the target gene response . Instead , their analysis suggests that the size of each target gene's response depends on its rate of transcription and how quickly it is first expressed in response to Nodal . These findings of Dubrulle et al . suggest that timing and transcription rate are important in determining the appropriate response to Nodal . Further work will be now needed to find out whether similar mechanisms regulate other processes that rely on the activity of morphogens .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Response to Nodal morphogen gradient is determined by the kinetics of target gene induction
Chemokines are secreted proteins that regulate a range of processes in eukaryotic organisms . Interestingly , different chemokine receptors control distinct biological processes , and the same receptor can direct different cellular responses , but the basis for this phenomenon is not known . To understand this property of chemokine signaling , we examined the function of the chemokine receptors Cxcr4a , Cxcr4b , Ccr7 , Ccr9 in the context of diverse processes in embryonic development in zebrafish . Our results reveal that the specific response to chemokine signaling is dictated by cell-type-specific chemokine receptor signal interpretation modules ( CRIM ) rather than by chemokine-receptor-specific signals . Thus , a generic signal provided by different receptors leads to discrete responses that depend on the specific identity of the cell that receives the signal . We present the implications of employing generic signals in different contexts such as gastrulation , axis specification and single-cell migration . Chemokines are small proteins that signal upon binding seven-pass-transmembrane G protein-coupled receptors ( GPCRs ) ( Zlotnik and Yoshie , 2000 ) . Chemokine receptors are classified into four categories namely , CXCR , CCR , XCR and CX3CR ( Nomiyama et al . , 2011 ) . Chemokines were originally shown to function in the context of immune response , but were thereafter implicated in a range of developmental processes such as angiogenesis ( Strieter et al . , 1995 ) , neural development ( Zou et al . , 1998 ) and migration of non-immune cells . Following binding to their ligands , chemokine receptors activate a wide range of effectors , including adenylyl cyclases , phospholipase isoforms , protein tyrosine kinases , ion channels , and mitogen-activated protein kinases ( MAPKs ) . These responses can result from the activation of G proteins , as well as from other second messengers to initiate G-protein-independent signaling ( Steen et al . , 2014 ) . For example , in addition to signaling through the G protein Gαi and Gβγ , chemokine receptors can initiate JAK/STAT signaling and signal through β-arrestin , in the context of chemotaxis of hematopoietic progenitor cells and in the context of activation and release of granules in neutrophils , respectively ( Barlic et al . , 2000; Zhang et al . , 2001 ) . Thus , chemokine receptor signaling through a range of second messenger molecules potentially expands an individual chemokine receptor’s ability to control qualitatively different cellular responses . In many cases , the same chemokine receptor is expressed in different cell types , where it initiates very different types of biological responses . For example , Cxcr4 expression in hematopoietic progenitors is important for these cells’ mobilization ( Möhle et al . , 1998 ) , yet , the same receptor , when expressed in neuronal progenitor cells inhibits their proliferation ( Krathwohl and Kaiser , 2004 ) . Similarly , CCR7 is expressed by T lymphocytes to facilitate their homing to secondary lymphoid organs ( Sallusto et al . , 1999 ) , but it is also expressed in and important for the development of the human placenta ( Drake et al . , 2004 ) . On the other hand , in early zebrafish embryos Ccr7 is expressed broadly and is involved in proper dorsoventral axis formation ( Wu et al . , 2012 ) . These observations raise the question of how different chemokine receptors control such a wide range of different processes . Several models have been suggested to explain this phenomenon and these are collectively referred to as ‘signaling bias’ ( Steen et al . , 2014 ) . Related to that , it has been recently demonstrated that the extracellular and membrane-spanning domains of chemokine receptors are not responsible for signaling specificity ( Xu et al . , 2014 ) . In this work , the extracellular and transmembrane domains of rhodopsin were combined with the intracellular domains of Cxcr4 . In this case , in response to light ( the ligand of rhodopsin ) , the chimeric protein elicited Cxcr4 signaling , such that it could direct the migration of T-cells . A receptor could show signaling bias by preferentially activating different signaling cascades depending on the specific ligand or agonist it binds . For example , when chemokine receptor Ccr7 is bound by Ccl19 , it induces β-arrestin recruitment more potently than when it is bound by Ccl21 ( Kohout et al . , 2004 ) . Another type of signaling bias , which could increase the range of processes that chemokine receptors control involves the initiation of different signaling cascades upon binding of the same ligand . For example , Cxcr4 interaction with its ligand Cxcl12 triggers both Gαi and β-arrestin signaling ( Dumstrei et al . , 2004; Sun et al . , 2002; Thomsen et al . , 2016 ) , while Cxcr7 receptor interaction with Cxcl12 was reported to activate β-arrestin but not G-protein signaling ( Rajagopal et al . , 2010 ) . Overall , the above-mentioned studies help explain how chemokine receptor signaling bias occurs , as these receptors and their ligands can differentially activate certain second messengers . However , previous studies have not thoroughly examined the relevance of specific differences in second messenger activation with respect to the resulting biological responses in vivo . Here , we demonstrate that distinct responses to chemokine receptor signaling depend on the responding cell type rather than on the specific receptor activated by its ligand . We demonstrate this principle via chemokine receptor signaling in zebrafish embryos by examining the function of four different chemokine receptors: Cxcr4a , Cxcr4b , Ccr7 , and Ccr9 . We gathered several lines of evidence to show that these chemokine receptors initiate specific biological processes in a way that depends on the cell types they are expressed in . These results present chemokine receptor signal interpretation module ( CRIM ) as a new mechanism for biasing the biological response resulting from chemokine receptor signaling . Specifically , we showed that each of those receptors is capable of controlling biological processes that it normally does not regulate , indicating that different chemokine receptors provide the same signal when activated . We thus suggest that different cell types express specific response modules or CRIM that interpret generic signals produced by different types of chemokine receptors . As a result of an additional genome duplication in teleosts relative to other vertebrates ( Lu et al . , 2012; Meyer and Schartl , 1999 ) , the zebrafish genome encodes two cxcr4 genes , specifically cxcr4a and cxcr4b ( Chong et al . , 2001 ) . These two genes were shown to regulate very different biological events and respond to different ligands; Cxcr4a is activated by the chemokine Cxcl12b and Cxcr4b is activated by the chemokine Cxcl12a ( Boldajipour et al . , 2011 ) . Cxcr4a plays a central role in vascular system patterning by guiding multicellular vessel growth ( Siekmann et al . , 2009 ) . Additionally , Cxcr4a controls endodermal cell-matrix adhesion , thereby ensuring proper gastrulation movements ( Nair and Schilling , 2008 ) . At the same time , Cxcr4b is involved in different processes such as the guided migration of primordial germ cells ( PGCs ) ( Doitsidou et al . , 2002; Knaut et al . , 2003 ) . To examine whether qualitative differences between Cxcr4a and Cxcr4b signaling exist , we investigated this issue in the contexts of gastrulation ( Cxcr4a-controlled endodermal cell adhesion ) and directional migration ( Cxcr4b-controlled single-cell migration ) . First , we expressed Cxcr4a instead of the Cxcr4b receptor in PGCs and examined whether the foreign receptor ( Cxcr4a ) could function in the context of guided single-cell migration , which is normally directed by Cxcr4b . We assayed the function of the receptor by monitoring the position of the PGCs in 12 hours post-fertilization ( hpf ) old embryos , a stage when cxcl12b ( encoding for the Cxcr4a ligand ) and cxcl12a ( encoding for the Cxcr4b ligand ) exhibit distinct expression patterns ( Boldajipour et al . , 2011 ) , allowing us to determine the response of the cells towards each of the ligands . In this experimental setup , embryos homozygous for the cxcr4b odysseus nonsense mutation , inactivating the gene ( Knaut et al . , 2003 ) were used . In odysseus mutant embryos injected with control RNA encoding for the human CD14 the PGCs were randomly distributed , whereas injecting RNA encoding for Cxcr4b in the PGCs reversed the phenotype , such that the cells clustered at regions where the ligand Cxcl12a was expressed ( Figure 1A ) . Intriguingly , PGCs expressing Cxcr4a were located closer to the midline at the region where the RNA encoding for the Cxcr4a ligand Cxcl12b is normally expressed ( see green label in Figure 1A right panel , n = 60 embryos in three experimental repeats ) . Thus , the mis-expressed chemokine receptor Cxcr4a is capable of directing PGC migration toward sites where its ligand Cxcl12b is expressed , despite the fact that it is normally not involved in this process . This result is consistent with the idea that the signals provided by the two receptors are qualitatively similar , allowing the cells to respond in a similar way to the signals generated by either of the receptors . To further investigate if the signals the two receptors elicited were indeed equivalent , we examined the ability of Cxcr4b to support a process normally controlled by Cxcr4a and its ligand Cxcl12b , namely the proper adhesion and positioning of endodermal cells during gastrulation . In this experiment , we made use of a transgenic fish line ( sox17::GFP , [Mizoguchi et al . , 2008] ) in which all the endodermal cells and the dorsal forerunner cells are labeled with GFP . Inhibiting the translation of cxcr4a along with cxcl12b RNAs using antisense morpholino oligonucleotides elicited the previously described abnormal displacement of endoderm from the dorsal forerunner cells , as seen in Figure 1B ( Nair and Schilling , 2008 ) . Unlike the case of chemokine-guided migration , in the context of controlling the interaction of the endoderm with the mesoderm , the distribution of the ligand is not critical . Accordingly , global expression of the chemokine acting in a paracrine or autocrine manner is expected to effectively control the process . Indeed , the cxcr4a/cxcl12b morpholino–induced phenotype was effectively rescued by co-expressing the morpholino-resistant cxcr4a and cxcl12b mRNAs in the embryos ( Figure 1B ) . Interestingly , consistent with the idea that the intracellular signals generated by the two receptors are equivalent , the expression of cxcr4b and cxcl12a RNAs in embryos knocked down for cxcr4a and cxcl12b reversed the phenotype as well . Thus , Cxcr4b signaling in endodermal cells could effectively replace that of Cxcr4a as determined by the reduction in the displacement between endodermal cells and forerunner cells . The results presented above show that distinct CXC receptors can control processes they are not normally involved in . Nevertheless , Cxcr4a and Cxcr4b show relatively high similarity in their protein sequence ( Figure 2—figure supplement 1 ) . Therefore , to examine the equivalence of chemokine receptor signaling more rigorously , we performed analogous experiments where we exchanged CC and CXC receptors in different processes . Here , Ccr9 and Ccr7 ( and their ligands Ccl25 and Ccl19 , respectively ) , which do not share high-sequence similarity with Cxcr4a , Cxcr4b ( Figure 2—figure supplement 2 ) were tested in the context of Cxcr4-controlled PGC directional migration and endoderm cell adhesion . To examine the general nature of chemokine receptor signals , we tested the potency of Ccr7 and Ccr9 in regulating endodermal cell movement . We expressed these receptors with their cognate ligands in early embryos and observed their effects on endodermal cell positioning in the embryos knocked down for Cxcr4a , the receptor that normally regulates this process . The ubiquitous expression of chemokine receptors and their ligand in the early embryos by way of injecting the RNA results in a uniform expression pattern , which in the context of this process is similar to that of the endogenous receptor-ligand pair ( Cxcr4a and Cxcl12b ) . Remarkably , both Ccr9 and Ccr7 reversed the Cxcr4a phenotype concerning the displacement between endoderm cells and dorsal forerunner cells , effectively controlling endodermal cell positioning ( Figure 2A , B ) To further test the capability of receptors to direct cell migration , we expressed receptors in PGCs and their cognate ligands in one half of the embryo in the absence of the regular endogenous signals guiding the cells ( i . e . the Cxcl12a in the environment and Cxcr4b in the PGCs ) . In this experimental setup , we thus generated spatially restricted source of chemokine , simulating the uneven distribution of the endogenous guidance cue within the embryo ( see Figure 3 for a schematic representation of the experimental setup ) . If the receptor can direct cell migration , it would lead to PGC accumulation within the part of the embryo expressing the ligand , as compared with a random distribution in control ( Doitsidou et al . , 2002 ) . Interestingly , in contrast with PGCs expressing control RNA that were randomly distributed throughout the embryo , under conditions where the endogenous Cxcl12a signals were knocked down , PGCs expressing Ccr9 were preferentially present on the part of the embryo engineered to express Ccl25 ( Figure 3A , C ) . Similar results were observed when testing the activity of Ccr7 and its ligand Ccl19 ( Figure 3A , D ) and Cxcr4a and its ligand Cxcl12b ( Figure 3A , B ) , demonstrating that receptors from the same and from different families can control directional PGC migration . Since the biological contexts studied above are based on cell migration , we further tested the equivalency of chemokine receptor signals in the context of dorsoventral fate specification in the early zebrafish embryo . Ccr7 was previously shown to be important for dorsoventral axis specification in early zebrafish embryos ( Wu et al . , 2012 ) . Here , the activated Ccr7 limits β-catenin-induced dorsalization throughout the early embryo , thereby controlling the relative size of dorsal and ventral domains . Embryos lacking Ccr7 function ( maternal zygotic ccr7stl7/stl7 , MZccr7 mutants ) do not appear morphologically dorsalized as do ccr7 morpholino-treated embryos ( Wu et al . , 2012 ) , a finding that could be attributed to off-target effects of the morpholinos on genes in addition to ccr7 . Nevertheless , MZccr7 mutant embryos do exhibit increased sensitivity to β-catenin-induced dorsalization . Consequently , a very low dose of RNA encoding for Δβ-catenin ( 2 . 5 pg ) caused expansion of the dorsal region in MZccr7 homozygous mutants , whereas in wild-type embryos the same dose of Δβ-catenin has no effect ( Figure 4—figure supplement 1 ) . To quantify the effect of chemokine signaling on the size of the dorsal tissue induced by Δβ-Catenin , we used MZccr7 mutant embryos expressing GFP under the control of the goosecoid promoter ( Doitsidou et al . , 2002 ) . We counted the number of pixels showing GFP expression above the auto threshold in those embryos following different experimental manipulations . MZccr7 mutants Δβ-catenin RNA-sensitized embryos co-injected with control RNA showed high level of GFP expression at 5 hpf as compared with non-injected embryos ( Figure 4A–D ) . Interestingly , goosecoid promoter-driven GFP expression was reduced in Δβ-catenin RNA-sensitized MZccr7 embryos in which different chemokine receptors were activated by expression of cxcr4a , cxcr4b and ccr9 RNAs along with their cognate ligands ( Figure 4A–D ) . These results show that Ccr7’s effect on the extent of dorsalization ( Wu et al . , 2012 ) can be directed by other chemokine receptors from different families . In contrast to the notion that different receptors initiate different signaling cascades to mediate various biological processes , our results suggest that different chemokine receptors initiate similar signaling . Accordingly , the specific response to receptor signaling might depend on its interpretation by the cell type within which the receptor was activated . To test this idea , we examined the signaling downstream of chemokine receptors in the context of directional cell migration . Cxcr4 was shown to signal through Gαi in response to Cxcl12 binding ( Moepps et al . , 1997 ) . Accordingly , Gαi was shown to be important for Cxcr4b-mediated directed PGC migration in zebrafish ( Dumstrei et al . , 2004 ) . To examine if the guidance signals that receptors other than Cxcr4b transmitted are Gαi dependent as well , we expressed pertussis toxin ( PTX ) in the PGCs . PTX catalyzes ADP ribosylation of Gαi impairing its interaction with the receptor thereby inhibiting G-protein-dependent signaling ( Casey et al . , 1989; Mangmool and Kurose , 2011 ) . We examined if this treatment that inhibits Gαi signaling affected the activity of different chemokine receptors in steering PGCs toward ligand-expressing domains within the embryo ( Figure 5A ) . Indeed , the guidance of PGCs mediated by the four receptor-ligand pairs ( Cxcr4b-Cxcl12a , Cxcr4a-Cxcl12b , Ccr9-Ccl25 and Ccr7-Ccl19 ) was abrogated by inhibiting Gαi function ( Figure 5A–E ) . These results are consistent with the idea that Gαi is essential for the directional cues the four chemokine receptors provide to the motile PGCs . The finding that different types of chemokine receptors depend on the same signaling cascade to control the same process highlights the importance of tight regulation over their expression . This would ensure that distinct processes are regulated by the specific ligands that are expressed at the correct time and location . To demonstrate this point , we ectopically expressed the Cxcr4a receptor in the PGCs rendering them responsive to its cognate ligand Cxcl12b ( in addition to the endogenous Cxcr4b ligand Cxcl12a ) . Interestingly , despite the expression of Cxcl12a within regions toward which the PGCs migrate , making the cells responsive to Cxcl12b affected their migration . Specifically , we found that PGCs co-expressing the two chemokine receptors were more dispersed within the embryo , consistent with the idea that they responded to spatially distinct conflicting signals encoded by the two different ligands . Indeed , PGCs could be found in locations ( e . g . in the head region , Figure 5—figure supplement 1 ) where Cxcl12b is expressed ( Thisse and Thisse , 2005 ) . The results provided above support the notion that chemokine receptors from different groups can initiate the same signaling pathways . These findings raise the possibility that chemokine receptors in a particular cell type may act redundantly among themselves or with receptors belonging to other GPCRs classes to control specific processes , thereby conferring genetic robustness ( Krakauer and Plotkin , 2002 ) . According to this proposition , receptors that are not considered to play a role in certain processes since their function appears dispensable for them , are actually important for those events , but are redundant . To examine this proposition , we studied the role of two classes of GPCRs expressed during early stages of embryogenesis in a process where they were not known to function before . We analyzed the involvement of the chemokine receptor Cxcr4b and phospholipid receptors ( S1p and LPA receptors ) in the process of gastrulation . To this end , we overexpressed LPPs ( lipid phosphate phosphatases ) , which dephosphorylate active lipids such as S1p and LPA , a treatment that should reduce signaling by lipid receptors Lpar and S1pr ( Brindley and Pilquil , 2009 ) . Conducting this treatment in embryos lacking Cxcr4 function allows for studying the effect of simultaneous inhibition of two seemingly unrelated receptors . Overexpression of LPPs in WT embryos had no visible effect on gastrulation and development ( Figure 6A ) . Interestingly , however , overexpression of LPPs in cxcr4b mutant embryos led to a strong delay in epiboly movements ( Figure 6A and B ) and somitogenesis as compared with a similar manipulation in wild-type embryos ( Figure 6C and D ) . These results are consistent with the idea that the two unrelated receptors , despite belonging to different groups of GPCRs , cooperate in ensuring proper progression of early processes in early embryonic development . According to our findings , chemokine-induced signaling elicits a qualitatively similar cascade that is interpreted differently by different types of cells . At the same time , a specific cell type can interpret chemokine signals in a distinct way that is dictated by the specific chemokine receptor signal interpretation module ( CRIM ) . For example , if a chemokine receptor-induced signaling cascade leads to directional migration toward a ligand , the same signaling cannot induce migration away from the source of ligand in the same cell type ( Poznansky et al . , 2000 ) . However , our model appears to be incompatible with cell behavior during fugetaxis ( cell movement away from the chemoattractant [Vianello et al . , 2005] ) , and retrotaxis ( cell migration down chemoattractant gradients [Hamza et al . , 2014] ) . Relevant for the guided migration of PGCs , T-cells were shown to actively migrate away from high concentrations of Cxcl12 ( Poznansky et al . , 2000 ) . To examine the behavior of PGCs upon exposure to a high concentration of a ligand , simulating the conditions the cells would experience upon arrival at their target , we expressed the ligand in the forming endoderm of the embryo . This was achieved by co-injecting RNA encoding for the activated version of the TARAM-A receptor ( TARAM-A* , [Peyriéras et al . , 1998] ) with RNA encoding for the Cxcl12a into one blastomere at the 16-cell stage embryo . In this experiment , we expressed low ( 25 pg ) and high ( 400 pg ) amounts of cxcl12a in endoderm and observed the behavior of Cxcr4-expressing PGCs in a 10 hpf embryo . As expected , the PGCs were found to be located on the Cxcl12a-expressing area when the level of the ligand was low ( Figure 7A , Figure 7—Video 1 ) , they continued to migrate within this region and only very rarely would leave the Cxcl12a-expressing domain ( Figure 7B and C ) . Surprisingly , unlike the behavior PGCs exhibited with respect to low-Cxcl12a-expressing domains , when high levels of the ligand ( 400 pg ) were expressed , the PGCs were not localized within the area where the ligand was expressed ( Figure 7A ) . Instead , the cells initially actively migrated toward ligand-expressing area , but often turned away from the region where the ligand was expressed , a behavior resembling reverse migration , as observed , for example , for neutrophils at a resolution phase of inflammation ( de Oliveira et al . , 2016; Mathias et al . , 2006 ) Figure 7B , Figure 7—Video 2 . Since PGCs performed reverse migration only when exposed to a high concentration of the ligand , we reasoned that rapid receptor internalization due to exposure to high levels of the ligand could lead to this behavior . According to this model , when cells reach the location of high ligand concentration , they can move away as they lost the ability to respond to the chemokine signal and migrate randomly . To examine this possibility , we compared the effect of ligand concentration on the levels of the Cxcr4b receptor on the membrane of PGCs . To this end , we labeled the PGC membrane with mCherry , and compared this signal with that of an EGFP-tagged Cxcr4b . The ratio between the mCherry and the EGFP signals reported the relative amount of functional receptor present on the membrane of the cell . Indeed , PGCs exposed to a high concentration of the ligand retained significantly fewer receptors on their membranes as compared with PGCs exposed to a low concentration of the ligand ( Figure 7D and E ) . Thus , the level of receptor internalization could be correlated with reverse migration and could constitute the basis for this behavior . In this work , we show that the same chemokine receptor can direct distinct responses in different cell types , while different receptors elicit the same biological response in a specific type of cells . These findings are consistent with the idea , that the biological consequences of chemokine receptor signaling depend on the cell type rather than on qualitative differences in the signal produced by specific receptors . We demonstrate that the identity of the activated receptor is immaterial for the actual interpretation of the signal that results in distinct biological responses in different cells . Our findings suggest that based upon their specific differentiation state , different cell types contain specific chemokine receptor signal interpretation modules ( CRIM ) that interpret the generic signals produced by chemokine receptors . The suggestion that the same receptor can elicit different cellular responses is presented graphically in Figure 8A . While it is possible that different receptors induce the response more efficiently or less , the qualitative features of the signaling , at least for the receptors and processes we examined , appear to be generic . Consistently , the cell-specific biological response to the signal appears to be robust as it can be observed when different levels of the receptor were expressed in the cells ( Figure 3—figure supplement 1 ) . This situation is analogous to heterozygosity for mutated chemokine and chemokine receptor alleles that has no phenotypic consequences ( Knaut et al . , 2003; Kupperman et al . , 2000 ) , as well as to a situation where the level of the Cxcr4b and Cxcl12a is altered by alleviating the miRNA regulation , a manipulation that has no phenotypic consequences ( Goudarzi et al . , 2013 ) . In support of our model , it was shown that in the context of the immune system the same cell type can respond in a similar way to signaling of different chemokine receptors . For example , CCR1 , CXCR1 and CXCR2 were shown to trigger arrest of rolling monocytes ( Ley , 2003; Luscinskas et al . , 2000; Weber et al . , 2001 ) . An interesting feature of the model we propose is that it allows positioning of cells by different ligands expressed in spatially distinct locations . Such a scenario was described in the case of neutrophil mobilization from the bone marrow to the blood stream and is schematically presented in Figure 8—figure supplement 1A . In this case , the positioning of the cells is dictated by a tug-of-war situation in which CXCL12 expressed by osteoblasts functions as a retention signal that maintains the neutrophils within the bone and CXCL2 emanating from the endothelium that attracts the cells toward the blood vessels ( Eash et al . , 2010 ) . Another aspect of the model is presented schematically in Figure 8B . In this case , concurrent trafficking of different cell types to distinct locations can be achieved using different chemokines . For example , in humans two subsets of memory T-cells , central memory T cells ( TCM , CCR7+ ) and effector memory T-cells ( TEM , CCR7– ) exhibit different behavior and localization . TCM tend to home to secondary lymphoid organs , where ligand for CCR7 is expressed , whereas TEM , that express CCR1 , CCR3 , CCR5 migrate toward the inflamed tissue , where the corresponding ligands for those receptors are expressed ( Sallusto et al . , 1999 ) . Thus , while several cell types share CRIMs , their actions are compartmentalized by dynamic and distinct spatiotemporal expression patterns of the receptors and their cognate ligands . In the context of embryonic development , the differentiation of cells and tissues dictates the presence of specific downstream signaling molecules in different cell types , which facilitates differential interpretation of a generic signal generated by chemokine receptors . This differential cell competence allows concomitant processes to be controlled by chemokines by regulating the expression pattern of receptors and ligands in specific cell populations . For example , during zebrafish gastrulation Cxcr4a regulates endodermal cell movement by controlling adhesion levels in response to uniform chemokine signaling . At the same time , the migration of germ cells is directed by specific patterns of a different ligand that interact with the chemokine receptor Cxcr4b . These two different events can take place at the same time within the same region of the embryo despite the generic signal the receptors produce as the different cell types interpret it differently . A particular feature of the generic downstream signaling is to increase the robustness of physiologically important processes through cooperation between different receptors in supporting specific processes . For example , by inhibiting the function of lipid-activated GPCRs , we revealed a novel function of Cxcr4 in promoting epiboly during gastrulation . This function was not described before , as Cxcr4 function is redundant to that of the LPA and S1P receptors . Similarly , T cells utilize Ccr9 , Cxcr4 and Ccr7 for homing to the thymus in early mouse embryos , but eliminating the function of one or two of these receptors is not sufficient to abrogate thymus homing . Only in the absence of all three receptors is thymic homing of T-cells completely abolished ( Calderón and Boehm , 2011 ) . The implication of these findings is that to determine the role of chemokine signaling in a certain process , one should mutate all the chemokine receptors/GPCRs in a tissue . Our findings suggest that even if cells express multiple chemokine receptors , at a specific differentiation or physiological state they can respond to receptor activation in only one way , such as directed migration ( Doitsidou et al . , 2002 ) , interaction with other cell types ( Nair and Schilling , 2008 ) or embryonic patterning ( Wu et al . , 2012 ) . As S1P/LPA receptors and Cxcr4b were able to elicit a similar biological response ( Figure 6 ) , this principle could be relevant for other G-protein-coupled receptors . Indeed , G-protein-coupled receptors of different families were shown to result in similar responses in other contexts . For example , neutrophils were shown to migrate towards fMLP and Cxcl8 , ligands that bind distinct receptors ( Gallin et al . , 1983; Ludwig et al . , 1997 ) . Interestingly , despite the fact that the receptors for these ligands ( fMLP receptor and Cxcr2 , respectively ) differ in some different downstream effectors they activate , the response to ligand binding is qualitatively identical ( Heit et al . , 2008 ) . Thus , despite differences in the biochemical response to such signals , the specific biological response of a specific cell type is identical . In light of our findings , chemokine-mediated attraction , fugetaxis or retrotaxis do not represent the activation of distinct signaling pathways . We suggest that reverse migration does not involve a qualitatively different signaling pathway , but represents a lack of receptor signaling due to internalization and desensitization processes , as proposed by Holmes and colleagues based on mathematical modeling ( Holmes et al . , 2012 ) . In contrast , cells that encounter regions where lower levels of the ligand are expressed maintain the receptor on their membranes and are therefore retained within those regions by positively responding to the chemokine distribution in the tissue . It would be interesting to determine if in other cases where retrotaxis was described for example in the case of LXA4 ( Hamza et al . , 2014 ) , which has also been shown to act as a potent chemoattractant ( Maddox and Serhan , 1996 ) , G-protein-coupled receptor internalization or desensitization of signaling provides the basis for reverse migration . While we present here a role for a chemokine receptor signal interpretation modules ( CRIM ) in a range of biological process in the context of different receptors function , some previous findings report on qualitatively different responses in the same cell type ( Gerszten et al . , 1999 ) . Such cases do not necessarily contradict our model if the cells investigated undergo a maturation process , which alter their interpretation of the signal . We suggest that since , among other differences , specific cell types differ from one another by the specific response network they harbor , the outcome of chemokine signaling can differ between cells of different lineages and sublineages ( see Figure 8—figure supplement 1B for a graphical explanation of the concept ) . Apparently differential response to chemokine signaling was demonstrated in the case of leukocyte arrest in response to CXCL1 activation of CXCR2 , while CCL2 binding to CCR2 could not lead to a similar biological response in the same cell type ( Huo et al . , 2001 ) . These findings are , however , contradictory to another study according to which CCL2 could actually induce leukocyte arrest ( Gerszten et al . , 1999 ) . It would thus be interesting to critically examine such contrasting results in light of the model we suggest . Similar statements suggesting different biological responses elicited by the action of distinct ligands ( e . g . [Zohar et al . , 2014] ) should be carefully assessed for equal experimental starting conditions and cell states between the different treatments . While the principle of generic signals and cell-specific interpretation presented here was tested in the context of chemokine receptors , it could also be relevant for other receptor families such as receptor tyrosine kinases . For example , for the receptor tyrosine kinase EGFR , the same ligand was shown to control different biological processes upon activating a specific receptor within different cell types ( Freeman and Gurdon , 2002; Queenan et al . , 1999 ) . Similarly , activation of the G-protein-coupled receptor for Acetylcholine was shown to induce different biological responses in different cell types ( Caulfield , 1993 ) . Such findings were interpreted as an indication that the same signaling pathway can lead to multiple cellular and developmental consequences , depending on the context and time ( Freeman and Gurdon , 2002 ) . While based on our results chemokine receptors elicit qualitatively similar signals , which appear to be equivalent to those LPA and S1P receptors produce , the signal may not be universal for all GPCRs . Indeed , the endoderm migration defect resulting from loss of Apela , a GPCR ligand ( Pauli et al . , 2014 ) could not be suppressed by expression of Cxcr4b or CCR9 with their corresponding ligands ( data not shown ) . Further experiments should be conducted to determine if this finding reflects a qualitative difference between the signal the chemokine receptors produce and the receptor for Apela . Alternatively , the dynamics of the expression of Apela and its receptor , which we could not mimic by uniformly expressing receptors and their ligands is responsible for this result . Our findings suggest that in addition to the expression of classic cell-specific differentiation markers , an important aspect of cell specification that dictates its fate and behavior is the expression of signaling interpretation modules . While the specific components of these modules are likely to be expressed in many cell types , the relative level of second messenger molecules as well as of molecules further downstream could provide specificity to the response . Understanding this relatively less explored yet important layer of cell differentiation is likely to shed more light on how cells regulate a range of processes and respond to different signals . We suggest that based on the differentiation state of a cell , these second messengers and molecules further downstream of them could be present in different proportions . In this way , the interpretation of GPCR signaling would vary in each cell type , leading to different outcomes . It would thus be interesting to determine those parameters for different responses to chemokine receptors and determine if one can modulate the consequence of chemokine signaling in a predicted way . Zebrafish ( Danio rerio ) of AB background were used as wild-type fish . Embryos from transgenic fish carrying sox17:EGFP ( Mizoguchi et al . , 2008 ) were used to investigate the capability of different chemokine receptors to regulate endoderm positioning . ccr7stl7/stl7 homozygous mutant embryos ( see below ) carrying gsc:GFP transgene and AB fish carrying gsc:GFP transgene were used to assess the competence of different chemokine receptors in regulating dorsoventral axis maintenance . The odysseus ( ody; Knaut et al . , 2003 ) fish line , homozygous for the mutation in cxcr4b gene , was used to assess the capability of different chemokine receptors to induce directed migration in PGCs , to demonstrate the ability of PGCs to undergo reverse migration upon exposure to high concentration of ligand and to test the effect of LPP overexpression on gastrulation . The ccr7 mutant was generated by TALEN system . The sequences of ccr7 TALEN targets are 5’TCCAACATGACTGAACAC and 5’TCATACTCTGTTGTAG , designed by ZiFit ( http://zifit . partners . org/ZiFiT/ ) ( Sander et al . , 2011 ) and TALEN plasmids were constructed using REAL Assembly TALEN Kit ( Addgene TALEN kit 1000000017 ) as described previously ( Sander et al . , 2011; Shin et al . , 2014 ) . The TALEN RNAs were synthesized using SP6 mMessageMachine Kit ( Ambion ) and injected into the one-cell stage eggs with 20–50 pg of each RNA . The mutagenic activity of the TALENs and mutant screen were assessed by restriction fragment length polymorphism ( RFLP ) analysis . Briefly , we first isolated the genomic DNAs from the TALEN RNAs injected embryos , amplified the TALEN target region of ccr7 locus using the primer set ( 5’TCCAACATGACTGAACACCAAATG and 5’AGGTCAGGATGACCAGAAAGTTCC ) , and checked whether HphI recognition sequences are mutated in the TALEN target region . Fragment size were: 162 , 37 bp for WT allele and 199 , 162 , 37 bp for ccr7 heterozygous mutant allele . Sequence alignment of wild-type ccr7 and mutated ccr7 can be found in Supplementary file 1; ccr7stl7 allele is 8 bp deletion in Exon three that is predicted to cause a frameshift and premature stop codon . A list of all the constructs used in the study and amounts used for injection are provided in Supplementary file 1 . A list of all the primers and morpholinos used in the study is provided in Supplementary file 1 . To express proteins preferentially in germ cells , the corresponding ORF was cloned upstream of the 3’UTR of nanos3 gene ( Köprunner et al . , 2001 ) . To express proteins globally , the corresponding ORFs were cloned upstream of the 3’UTR of the Xenopus globin gene . To direct the expression of Cxcl12a to endoderm its mRNA was co-injected with mRNA encoding for the constitutively active form of the TARAM-A protein into one of the 16 blastomeres ( Peyriéras et al . , 1998 ) . Capped mRNA used for injection was synthesized using mMessageMachine kit from Ambion . To determine relative position of PGCs expressing control , Cxcr4a or Cxcr4b mRNA with respect to expression pattern of Cxcl12a and cxcl12b , RNAscope in situ hybridization procedure was performed as previously described ( Gross-Thebing et al . , 2014 ) . For dorsoventral axis specification experiments , eggs were harvested immediately after they were laid and ramped in 1 . 5% agarose ramps . 2 nl of receptor and ligand or control RNA were injected into the newly forming cell . Δβ-catenin RNA was then injected into the cell as well . The same needle was used to inject Δβ-catenin in control and experimental embryos , to maintain identical volume of injection Still images of live zebrafish embryos were acquired using Zeiss Axioplan2 and a Zeis Axiolmager . Z1 microscopes controlled by the Visiview software , or on a Zeiss stereo microscope controlled by the Zeiss software . Still images of 5 hpf , 8 hpf and 10 hpf embryos were acquired at 10X magnification . Time-lapse movies showing reverse migration of PGCs were acquired at 10X magnification as well . For the time-lapse movies , an image was acquired every 2 min over a period of 180 min with exposure of 80 ms over 15 different focal planes . Still images showing internalization of receptors were acquired at 63X Magnification , using 488 nm and 561 nm laser sources . In the endoderm positioning experiment , the displacement of this germ layer was evaluated by measuring the vertical distance between the lowest positioned endodermal cell and forerunner cells , as illustrated in the Figure 1—figure supplement 1 . In the dorsoventral axis specification experiment , background was subtracted in the FIJI software using the rolling ball algorithm with 130 size . Median filter with size two was applied following Autothreshold . Either Yen or Default autothreshold algorithm was used in the study to determine the pixels showing GFP expression above threshold . In the reverse migration experiment , PGCs were tracked using the Imaris software . Movement of the endoderm tissue was averaged and subtracted , followed by analysis of PGCs movement with respect to the endoderm . Kruskal-Wallis test was performed , correcting for multiple testing . Error bars represent S . E . M . ns = nonsignificant . p≤0 . 05 . p≤0 . 01 , p≤0 . 001 , **p≤0 . 0001 .	Every process in the body is regulated by a complex network of interactions between different molecules and cells . Chemokines , for example , are tiny molecules produced by a cell that are involved in a range of processes , from development to immune responses and cancer . When chemokines bind to a specific protein on another cell , called the chemokine receptor , it stimulates different signaling pathways inside the cell . Consequently , chemokine receptors are equally important for regulating processes as diverse as the movement of cells during development and growth , or activating immune responses . Mammals have over 20 different chemokine receptors , and the same receptor can have different roles depending in which cell type it is found in . For example , in one cell type it may stimulate an action such as cell growth , but in another , it may block this process . Until now , it was unclear how chemokine receptors can achieve such different effects . One theory was that chemokine receptors initiate a distinct signaling cascade , a phenomenon termed ‘signaling bias’ , depending on the type of chemokine or receptor . Here , Malhotra et al . used zebrafish embryos to investigate how four specific chemokine receptors regulate different events during early development . They found that the same chemokine receptor could direct different reactions in distinct cell types , while different receptors could also cause the same response in a specific cell type . In other words , the effect of a chemokine receptor depends on the cell type rather than the type of receptor . Since each of these receptors was able to control processes that it normally does not regulate in other cells , Malhotra et al . suggest that different chemokine receptors provide the same generic signal when activated , which the specific cell types then interpret accordingly . A next step will be to test how other chemokine receptors behave in different contexts , for example during an immune response . If the receptors work on the same principle regardless of the process , it could help to explain why faulty expression of chemokine receptors play such an important role during development and in disease . It could further highlight why blocking one receptor may not have any consequences , as they are dispensable and can be replaced by other receptors in the cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "immunology", "and", "inflammation" ]	2018	Spatio-temporal regulation of concurrent developmental processes by generic signaling downstream of chemokine receptors
We have developed an open-source software called bi-channel image registration and deep-learning segmentation ( BIRDS ) for the mapping and analysis of 3D microscopy data and applied this to the mouse brain . The BIRDS pipeline includes image preprocessing , bi-channel registration , automatic annotation , creation of a 3D digital frame , high-resolution visualization , and expandable quantitative analysis . This new bi-channel registration algorithm is adaptive to various types of whole-brain data from different microscopy platforms and shows dramatically improved registration accuracy . Additionally , as this platform combines registration with neural networks , its improved function relative to the other platforms lies in the fact that the registration procedure can readily provide training data for network construction , while the trained neural network can efficiently segment-incomplete/defective brain data that is otherwise difficult to register . Our software is thus optimized to enable either minute-timescale registration-based segmentation of cross-modality , whole-brain datasets or real-time inference-based image segmentation of various brain regions of interest . Jobs can be easily submitted and implemented via a Fiji plugin that can be adapted to most computing environments . The mapping of the brain and neural circuits is currently a major endeavor in neuroscience and has great potential for facilitating an understanding of fundamental and pathological brain processes ( Alivisatos et al . , 2012; Kandel et al . , 2013; Zuo et al . , 2014 ) . Large projects , including the Mouse Brain Architecture project ( Bohland et al . , 2009 ) , the Allen Mouse Brain Connectivity Atlas ( Oh et al . , 2014 ) , and the Mouse Connectome project , have mapped the mouse brain ( Zingg et al . , 2014 ) in terms of cell types , long-range connectivity patterns , and microcircuit connectivity . In addition to these large-scale collaborative efforts , an increasing number of laboratories are also developing independent , automated , or semi-automated frameworks for processing brain data obtained for specific projects ( Fürth et al . , 2018; Ni et al . , 2020; Niedworok et al . , 2016; Renier et al . , 2016; Wang et al . , 2020a; Iqbal et al . , 2019 ) . With the improvement of experimental methods for dissection of brain connectivity and function , development of a standardized and automated computational pipeline to map , analyze , visualize , and share brain data has become a major challenge to all brain connectivity mapping efforts ( Alivisatos et al . , 2012; Fürth et al . , 2018 ) . Thus , the implementation of an efficient and reliable method is fundamentally required for defining the accurate anatomical boundaries of brain structures , by which the anatomical positions of cells or neuronal connections can be determined to enable interpretation and comparison across experiments ( Renier et al . , 2016 ) . The commonly used approach for automatic anatomical segmentation is to register an experimental image dataset within a standardized , fully segmented reference space , thus obtaining the anatomical segmentation for this set of experimental images ( Oh et al . , 2014; Ni et al . , 2020; Renier et al . , 2016; Kim et al . , 2015; Lein et al . , 2007 ) . There are currently several registration-based high-throughput image frameworks for analyzing large-scale brain datasets ( Fürth et al . , 2018; Ni et al . , 2020; Niedworok et al . , 2016; Renier et al . , 2016 ) . Most of these frameworks require the user to set a few parameters based on the image intensity or graphics outlines or to completely convert the dataset into a framework-readable format to ensure the quality of the resulting segmentation . However , with the rapid development of sample labeling technology ( Lee et al . , 2016; Richardson and Lichtman , 2015; Schwarz et al . , 2015 ) and high-resolution whole-brain microscopic imaging ( Economo et al . , 2016; Gong et al . , 2013; Nie et al . , 2020; Liu et al . , 2017; Li et al . , 2010 ) , the heterogeneous and non-uniform characteristics of brain structures make it difficult to use traditional registration methods for registering datasets from different imaging platforms to a standard brain space with high accuracy . In this case , laborious visual inspection , followed by manual correction , is often required , which significantly reduces the productivity of these techniques . Therefore , the research community urgently needs a robust , comprehensive registration method that can extract a significant number of unique features from image data and provide accurate registration between different types of individual datasets . Moreover , though registration-based methods can achieve full anatomical annotation in reference to a standard atlas for whole-brain datasets , their region-based 3D registration to a whole-brain atlas lacks the flexibility to analyze incomplete brain datasets or those focused on a certain volume of interest ( Song and Song , 2018 ) , which is often the case in neuroscience research . Though some frameworks can register certain types of brain slabs that contain complete coronal outlines slice by slice ( Fürth et al . , 2018; Song and Song , 2018; Ferrante and Paragios , 2017 ) , it remains very difficult to register a small brain block without obvious anatomical outlines . As neural networks have emerged as a technique of choice for image processing ( Long et al . , 2015; Chen et al . , 2018a; He et al . , 2019; Zhang et al . , 2019 ) , deep-learning-based brain mapping methods have also recently been reported to directly provide segmentation/annotation of primary regions for 3D brain datasets ( Iqbal et al . , 2019; Akkus et al . , 2017; Chen et al . , 2018b; Milletari et al . , 2017; de Brebisson and Montana , 2015 ) . Such deep-learning-based segmentation networks are efficient in extracting pixel-level features and thus are not dependent on the presence of global features such as complete anatomical outlines , making them better suited for processing of incomplete brain data , as compared to registration-based methods . On the other hand , the establishment of these networks still relies on a sufficiently large training dataset , which is often laboriously registered , segmented , and annotated . Therefore , a combination of image registration and a neural network can possibly provide a synergistic improved analysis method and lead to more efficient and versatile brain mapping techniques . Here , we provide an open-source software as a Fiji ( Schindelin et al . , 2012 ) plugin , termed bi-channel image registration and deep-learning segmentation ( BIRDS ) , to support brain mapping efforts and to make it feasible to analyze , visualize , and share brain datasets . We developed BIRDS to allow investigators to quantify and spatially map 3D brain data in its own 3D digital space with reference to Allen CCFv3 ( Wang et al . , 2020b ) . This facilitates analysis in its native status at cellular level . The pipeline features: ( 1 ) A bi-channel registration algorithm integrating a feature map with raw image data for co-registration with significantly improved accuracy and ( 2 ) a mutually beneficial strategy in which the registration procedure can readily provide training data for a neural network , while this network can efficiently segment incomplete brain data that is otherwise difficult to register with a standardized atlas . The whole computational framework is designed to be robust and flexible , allowing its application to a wide variety of imaging systems ( e . g . , epifluorescent microscopy or light-sheet microscopy ) and labeling approaches ( e . g . , fluorescent proteins , immunohistochemistry , and in situ hybridization ) . The BIRDS pipeline offers a complete set of tools , including image preprocessing , feature-based registration and annotation , visualization of digital maps and quantitative analysis via a link with Imaris , and a neural network segmentation algorithm that allows efficient processing of incomplete brain data . We further demonstrate how BIRDS can be employed for fully automatic mapping of various brain structures and integration of multidimensional anatomical neuronal labeling datasets . The whole pipeline has been packaged into a Fiji plugin , with step-by-step tutorials that permit rapid implementation of this plugin in a standard laboratory computing environment . Figure 1 shows our bi-channel registration procedure , which registers experimental whole-brain images using a standardized Allen Institute mouse brain average template , and then provides segmentations and annotations from CCFv3 for experimental data . The raw high-resolution 3D images ( 1 × 1 × 10 μm3 per voxel ) , obtained by serial two-photon tomography ( STPT , see Materials and methods ) , were first down-sampled into isotropic low-resolution data with a 20 μm voxel size identical to an averaged Allen template image ( Figure 1a ) . The re-sampling ratios along the x ( lateral-medial axis ) , y ( dorsal-ventral axis ) , and z ( anterior-posterior , AP axis ) axes were thus 0 . 05 , 0 . 05 and 0 . 5 , respectively . It should be noted that , in addition to the individual differences , the preparation/mounting steps can also cause non-uniform deformation of samples , thereby posing extra challenges to the precise registration of experimental image to an averaged template ( Figure 1b , original dataset ) . To mitigate this non-uniform deformation issue before registration , we applied a dynamic re-sampling ratio rather than using a fixed value of 0 . 5 to the z reslicing . We first subdivided the entire image stack into multiple sub-stacks ( n = 6 in our demonstration , Figure 1a ) according to seven selected landmark planes ( Figure 1a , Figure 1—figure supplement 1 ) . Then we applied a dynamic z re-sampling ratio calculated corresponding to the positions of the landmark planes in the Allen template and sample data ( varying from ~0 . 35 to 0 . 55 ) to each sub-stack , to finely compress ( <0 . 5 ) or stretch ( >0 . 5 ) the z depth of the sub-stacks , thereby better matching the depth of each sub-stack to the Allen template brain and rectifying the deformation along the AP axis ( Figure 1a , Materials and methods ) . The rectified whole-brain stack assembled by these dynamically re-sampled sub-stacks showed higher original similarity to the Allen template brain as compared to a raw experimental image stack ( Figure 1b ) . The implementation of such a preprocessing step was beneficial for the better alignment of non-uniformly morphed brain data to a standardized template ( Figure 1—figure supplement 2 ) . After data preprocessing , we applied a feature-based iterative registration using the Allen reference images to the preprocessed experimental images . We note that previous registration methods were vulnerable to inadequate alignment accuracy ( Niedworok et al . , 2016; Renier et al . , 2016; Goubran et al . , 2019 ) , which was associated with inadequate registration information provided by merely using the raw background image data . To address this issue , in addition to the primary channel containing the background images of each sample and template brains , we further generated an assistant channel to augment the image registration and enhance the accuracy . First , we used a phase congruency ( PC ) algorithm ( Kovesi , 2019 ) to extract the high-contrast edge and texture information from both the experimental and template brain images based on their relatively fixed anatomy features ( Figure 1c , Materials and methods ) . Then , we obtained the geometry features of both brains along their lateral–medial , dorsal–ventral , and anterior–posterior axes with enhanced axial mutual information ( MI ) extracted using a grayscale reversal processing ( Maes et al . , 1997; Thévenaz and Unser , 2000 ) ( Figure 1c , Figure 1—figure supplement 3 , Materials and methods ) . Finally , the primary channel containing raw brain images , in conjunction with the assistant channel containing the texture and geometry maps of brains , were included in the registration procedure to fulfill an information-augmented bi-channel registration requirement ( Figure 1—figure supplement 4 ) , which was verified to show notably better registration accuracy as compared to conventional single-channel registration methods ( aMAP [Niedworok et al . , 2016] , ClearMap [Renier et al . , 2016] , and MIRACL [Goubran et al . , 2019] ) . During registration , through an iterative optimization of the transformation from an averaged Allen brain template to the experimental data , the MI gradually reached its maximum when the inverse grayscale images , PC images , and the raw images were finally geometrically aligned ( Figure 1d ) . The displacement was presented in a grid form to illustrate the non-linear deformation effects . The geometry wrapping parameters obtained from the registration process were then applied to the Allen annotation file to generate a transformed version specifically for experimental data ( Figure 1—figure supplement 4 ) . Our dual-channel registration achieved fully automated registration/annotation at sufficiently high accuracy when processing STPT experimental data of an intact brain ( Han et al . , 2018 ) . As for low-quality or highly deformed brain data ( e . g . , clarified brain with obvious shrinkage ) , though the registration accuracy of our method was accordingly reduced , our method still quite obviously surpassed other methods ( Figure 2 ) . For such challenging data types , we also developed an interactive graphic user interface ( GUI ) to readily permit manual correction of the visible inaccuracies in the annotation file , through finely tuning the selected corresponding points ( Figure 1e ) . Finally , an accurate 3D annotation could be generated and applied to experimental data , either fully automatically ( STPT data ) or after mild manual correction ( light-sheet ﬂuorescence microscopy [LSFM] data of clarified brain ) , as shown in Figure 1f . Next , we merged our experimental brain image with a registered annotation file to generate a 3D annotated image and quantitatively compared its registration accuracy with aMAP , ClearMap , and MIRACL results . We made comparisons of both STPT data from intact brains that contained only minor deformations ( Figure 2a ) and LSFM data from clarified brains ( u-DISCO ) that showed obvious shrinkage ( Figure 2b ) . It should be noted here that the annotated results of either previous single-channel methods or our bi-channel method were all using automatic registration without any manual correction applied , and the averaged manual annotations by our experienced researchers served as a ground truth for quantitative comparisons . It was visually obvious that , as compared to the other three methods ( green: aMAP; red: ClearMap; and blue: MIRACL in Figure 2a , b ) , the Allen annotation files transformed and registered by our BIRDS method ( yellow in Figure 2a , b ) were far better aligned with both STPT ( as shown in VISC , CENT , AL , and PAL regions , Figure 2a ) and LSFM ( as shown in HPF , CB , VIS , and COA regions , Figure 2b ) images . Furthermore , we manually labeled 10 3D fiducial points of interest ( POIs ) across the registered Allen template images together with their corresponding experimental images ( Figure 2c ) and then measured the error distances between the paired anatomical landmarks in the two datasets , so that the registration accuracy by each registration method could be quantitatively evaluated ( Figure 2—figure supplement 1 ) . As shown in Figure 2d , the error distance distributions of POIs in five brains ( two STPT + three LSFM ) registered by the abovementioned four methods were then quantified , showing the smallest median error distance ( MED ) was obtained using our method for all five brains ( Supplementary file 3 ) . In two different sets of STPT data , only our BIRDS method could provide an MED below 100 μm ( ~80 μm , n = 2 ) , and this value slightly increased to ~120 μm for LSFM data ( n = 3 ) , but was still smaller than all the results obtained using the other three methods ( aMAP , ~342 μm , n = 3; ClearMap , ~258 μm , n = 3; and MIRACL , ~175 μm , n = 3 ) . Moreover , the Dice scores ( Dice , 1945 ) , defined as a similarity scale function used to calculate the similarity of two samples , for each method were also calculated at the nucleus precision level based on nine functional regions in the five brains . The comparative results were then grouped by brain and region , as shown in Figure 2e , f , respectively . The highest Dice scores with an average median value of >0 . 89 ( Supplementary file 3 , calculated for five brains , 0 . 75 , 0 . 81 , and 0 . 81 for aMAP , ClearMap , and MIRACL ) or >0 . 88 ( Supplementary file 3 , calculated using nine regions , 0 . 74 , 0 . 77 , and 0 . 84 for aMAP , ClearMap , and MIRACL , respectively ) were obtained by BIRDS , further confirming the superior registration accuracy of our method . Through a comparative Wilcoxon test , our results were demonstrated to be superior to the other three methods ( providing larger Dice scores ) with a p value < 0 . 05 calculated either by brain or by region . More detailed comparisons of registration accuracies can be found in Figure 2—figure supplements 2–4 . A 3D digital map ( CCFv3 ) based on the abovementioned bi-channel registration was generated to support automatic annotation , analysis , and visualization of neurons in a whole mouse brain ( see Materials and methods ) . The framework thus enabled large-scale mapping of neuronal connectivity and activity to reveal the architecture and function of brain circuits . Here , we demonstrated how the BIRDS pipeline visualizes and quantifies single-neuron projection patterns obtained by STPT imaging . A mouse brain containing six GFP-labeled layer-2/3 neurons in the right visual cortex was imaged with STPT at 1 × 1 × 10 μm3 resolution ( Han et al . , 2018 ) . After applying the BIRDS procedure to this STPT image stack , we generated a 3D map of this brain ( Figure 3a ) . An interactive hierarchal tree of brain regions in the software interface allowed navigation through the corresponding selected-and-highlighted brain regions with its annotation information ( Figure 3b , Video 1 ) . Through linking with Imaris , we visualized and traced each fluorescently labeled neuronal cell ( n = 5 ) using the filament module of Imaris across the 3D space of the entire brain ( Figure 3c , Materials and methods , Video 2 ) . The BIRD software can also apply reverse transformation to a raw image stack to generate a standard template-like rendered 3D map , including both traced axonal projections and selected whole-brain structures , which faithfully captures true 3D axonal arborization patterns and anatomical locations , as shown in Figure 3d . This software can also quantify the lengths and arborizations of traced axons according to the segmentation of the 3D digital map generated using the BIRDS pipeline ( Figure 3e ) . BIRDS can be linked to Imaris to perform automated cell counting with higher efficiency and accuracy ( Materials and methods ) . Here , we demonstrate it with an example brain where neurons were retrogradely labeled by CAV-mCherry injected to the right striatum and imaged by STPT at 1×1×10 μm3 resolution ( Han et al . , 2018 ) . The whole-brain image stacks were first processed by BIRDS to generate a 3D annotation map . Two of the example segregated brain areas ( STR and BS ) are outlined in the left panel of Figure 4a . The annotation map and the raw image stack were then transferred to Imaris , which processed the images within each segregated area independently . Imaris calculated the local image statistics for cell recognition only using the image stack within each segregated area; therefore , it fit the dynamic range of the local images to achieve better results , as shown in the middle column in the right panel of Figure 4a . In contrast , the conventional Imaris automated cell counting program processed the whole-brain image stack at once to calculate the global cell recognition parameters for every brain area , which easily resulted in false positive or false negative counts in brain areas where the labeling signal was too strong or too weak compared to the global signal , as demonstrated in the STR and SB in the right column of the right panels of Figure 4a , respectively . The BIRDS–Imaris program could perform automated cell counting for each brain area and reconstructed them over the entire brain . The 3D model of the brain-wise distribution of labeled striatum-projecting neurons was visualized using the BIRDS–Imaris program as a 3D rendered brain image and projection views from three axes in Figure 4b . The BIRDS program could calculate the volume of each segregated region according to the 3D segregation map and the density of labeled cells across the brain as shown in Figure 4c . Meanwhile , manual cell counting was also performed with every one out of four sections using an ImageJ plugin ( Figure 4d ) . Compared to conventional Imaris results , our BIRDS–Imaris results were more consistent with a manual one , especially for brain regions where the fluorescent signal was at the high or low end of the dynamic range ( BS and STR , Figure 4e ) . Thanks to the 3D digital map generated by the BIRDS pipeline , BIRDS–Imaris can process each segmented brain area separately , namely calculating the parameters for the cell recognition algorithm using local image statistics instead of processing the whole-brain image stack at once . Such a segmented cell counting strategy is much less demanding on computation resources , and moreover , it is optimized for each brain area to solve the problem that the same globe cell recognition parameter works poorly in certain brain regions with signal intensity at either of the two extreme ends of the dynamic range of the entire brain . In practice , acquired brain datasets are often incomplete , due to researcher’s particular interest in specific brain regions , or limited imaging conditions . The registration of such incomplete brain datasets to an Allen template is often difficult due to the lack of sufficient morphology information for comparison of both datasets . To overcome this limitation , we further introduced a deep neural network ( DNN ) -based method for efficient segmentation/annotation of incomplete brain sections with minimal human supervision . Herein , we optimized a Deeplab V3+ network , which was based on an encoding-decoding structure , for our deep-learning implementation ( Figure 5a ) . The input images passed through a series of feature processing stages in the network , with pixels being allocated , classified , and segmented into brain regions . It should be noted that the training of a neural network fundamentally requires a sufficiently large dataset containing various incomplete brain blocks which have been well segmented . Benefiting from our efficient BIRDS method , we could readily obtain a large number of such labeled datasets through cropping processed whole brains and without experiencing time-consuming manual annotation . Various types of incomplete brains , as shown in Figure 5b , were hereby generated and sent to our DNN for iterative training , after which the skilled network could directly infer the segmentations/annotations for new modes of incomplete brain images ( Materials and methods ) . Next , we validated the network performance on three different modes of input brain images cropped from the registered whole-brain dataset ( STPT ) . The DNN successfully inferred annotation results for a cropped hemisphere , irregular cut of hemisphere , and a randomly cropped volume , as shown in Figure 5c–e , respectively . The inferred annotations ( red lines ) were found to be highly similar to the registered annotation results ( green lines ) in all three types of incomplete data . To further quantify the inference accuracy , the Dice scores of the network-segmented regions were also calculated by comparing the network outputs to ground truth , which was the registration results after visual inspection and correction ( Figure 5—figure supplement 1 ) . The averaged median Dice scores for the individual sub-regions in the hemisphere , irregular cut of hemisphere , and random volumes were 0 . 86 , 0 . 87 , and 0 . 87 , respectively , showing a sufficiently high inference accuracy in most of brain regions , such as the isocortex , HPF , OLF , or STR . It is worth noting that the performance of our network for segmentation using PAL , MBsta , P-sen regions remained sub-optimal ( Dice score 0 . 78–0 . 8 ) , due to their lack of obvious borders , and large structural variations across planes ( Figure 5—figure supplement 1 ) . Finally , we applied our network inferences to generate 3D atlases for these three incomplete brains , while segmenting the hemisphere into 18 regions such as the Isocortex , HPF , OLF , CTXsp , STR , PAL , CB , DORpm , DORsm , HY , MBsen , MBmot , MBsta , P-sen , P-mot , P-sat , MY-sen , and MY-mot , while we processed an irregular cut of half the telencephalon into 10 regions as Isocortex , HPF , OLF , CTXsp , STR , PAL , DORpm , DORsm , and HY , MY-mot , and the random volume into seven regions , defined as the Isocortex , HPF , STR , PAL , DORpm , DORsm , and HY ( Figure 5f , g , h ) . Therefore , our DNN performed reasonably well even if the brain was highly incomplete . Furthermore , it could achieve second-level fine segmentation within a small brain region of interest . For example , we successfully segmented the hippocampus ( CA1 , CA2 , CA3 , and DG ) , as shown in Figure 5—figure supplement 2 . Such a unique capability of our DNN was possibly derived from the detection of pixel-level features rather than regions , and thereby substantially strengthened the robustness of our hybrid BIRDS method over conventional brain registration techniques when the data is highly incomplete/defective . More detailed performance comparisons between our DNN-based inference and other methods are shown in Figure 5—figure supplements 1–5 . In summary , we demonstrate a bi-channel image registration method , in conjunction with a deep-learning framework , to readily provide accuracy-improved anatomical segmentation for whole mouse brain in reference to an Allen average template , and direct segmentation inference for incomplete brain datasets , which were otherwise not easily registered to standardized whole-brain space . The addition of a brain feature channel to the registration process greatly improved the accuracy of automatically registering individual whole-brain data with a standardized Allen average template . It should be noted that the registration was based on two-photon template images provided by Allen CCF , so it is currently limited to using on like-imaged brains , for example , brains imaged using wide-field , confocal , or light-sheet microscopes , etc . For processing various incomplete brain datasets , which were challenging for registration-based methods while remaining very common in neuroscience research , we applied our deep neural network to rapidly infer segmentations . The sufficiently accurate results shown using different types of incomplete data verify the advances of network segmentation . Though a full annotation using a neural network is currently too computationally demanding as compared to registration-based segmentation , it is undoubtedly a good complement to registration-based segmentation . Therefore , in our hybrid BIRDS pipeline , the DNN inference greatly reinforced the inefficient side of registration , while the registration also readily provided high-quality training data for our DNN . We believe such a synergistic effect in our method could provide a paradigm shift for enabling robust and efficient 3D image segmentation/annotation for biology research . With the unceasing development of deep learning , we envision that network-based segmentation will play an increasingly important role in new pipelines . A variety of applications , such as tracing of long-distance neuronal projections and parallel counting of cell populations in different brain regions , was also enabled as a result of our efficient brain mapping . The BIRDS pipeline is now fully open source and also has been packaged into a Fiji plugin to facilitate biological researchers . We sincerely expect that the BIRDS method can immediately allow new insights using current brain mapping techniques , and thus further push the resolution and scale limits in future explorations of brain space . Brains 1 and 2 were obtained with STPT , and each dataset encompassed ~180 Gigavoxels , for example , 11 , 980 × 7540 × 1075 in Dataset 1 , with a voxel size of 1 × 1 × 10 μm3 . The procedure of sample preparation and imaging acquisition were described in Han et al . , 2018 . Briefly , the adult C57BL/6 mouse ( RRID:IMSR_JAX:000664 ) was anesthetized , and craniotomy was performed on top of the right visual cortex . Individuals neuronal axons were labeled with plasmid DNA ( pCAG-eGFP [Addgene accession 11150] ) by two-photon microscopy-guided single-cell electroporation , and the brain was fixed by cardioperfusion of 4% paraformaldehyde 8 days later . Striatum-projecting neurons were labeled by stereotactically injecting PRV-cre into the right striatum of tdTomato reporter mice ( Ai14 , JAX ) , and the brain was fixed cardioperfusion 30 days later . The brains were embedded in 5% oxidized agarose and imaged with a commercial STPT ( TissueVision , USA ) excited at 940 nm . Coronally , the brain was optically scanned every 10 μm at 1 μm/pixel without averaging and physically sectioned every 50 μm . The power of excitation laser was adjusted to compensate the depth of optical sections . Brains 3 , 4 , and 5 were obtained with LSFM and each dataset encompassed ~700 Gigavoxels ( ~10 , 000 × 8000 × 5000 ) , with an isotropic voxel size of 1 μm3 . Brain tissues of eight-week-old Thy-GFP-M mice ( RRID:IMSR_JAX:007788 ) were first clarified with u-DISCO protocol ( Pan et al . , 2016 ) before imaging . Brains 3 and 4 were acquired using a custom-built Bessel plane illumination microscope , a type of LSFM modality employing non-diffraction thin Bessel light-sheet . Brain 5 was whole-brain 3D image of a Thy-GFP-M mice acquired using a lab-built selective plane illumination microscope ( Nie et al . , 2020 ) , another LSFM modality combining Gaussian light-sheet with multi-view image acquisition/fusion . We run BIRDS on a Windows 10 workstation equipped with dual Xeon E5-2630 V3 CPUs , 1 TB RAM , and two GeForce GTX 1080 Ti graphic cards . In Supplementary file 1 , we showed the memory and time consumptions of our BIRDS plugin for processing 180 GB and 320 GB brain datasets . The image preprocessing time at stage 1 is approximately proportional to the size of data . In contrast , since the datasets for registration are down-sampled to the same size to match the Allen CCF template , the time and memory consumption at stages 2 and 3 are nearly the same for two datasets . The time and memory consumption for generating the 3D digital framework at stage 4 is proportional to the data size . It should be noted that memory with capacity at least 1 . 5 times of the data size is required at this step . Therefore , when applying BIRDS to larger datasets , such as rat brain , we will recommend a powerful workstation with at least one XEON CPU , 1 TB memory , and one state-of-the-art graphic card ( Geforce RTX 3090 ) , to guarantee a smooth running of whole pipeline . We have made our pipeline open access for the community . We have provided source code of the BIRDS for Windows 10 at https://github . com/bleach1by1/birds_reg ( Wang , 2021a; copy archived at swh:1:rev:22cf3d792c3887708a65ddae43d6dde7ed8b7836 ) . FIJI plugin and ancillary installation packages are provided at https://github . com/bleach1by1/BIRDS_plugin ( Wang , 2021b; copy archived at swh:1:rev:41e90d4518321d6ca8e806ccadb2809bfa6bd475 ) . BIRDS contains five core modules , image preprocessing , bi-channel registration , manual correction , link with Imaris software , and deep-learning segmentation , all of which can be executed on a GUI . We also provided a step-by-step tutorial and test data to facilitate the program implementation by other researchers .	Mapping all the cells and nerve connections in the mouse brain is a major goal of the neuroscience community , as this will provide new insights into how the brain works and what happens during disease . To achieve this , researchers must first capture three-dimensional images of the brain . These images are then processed using computational tools that can identify distinct anatomical features and cell types within the brain . Various microscopy techniques are used to capture three-dimensional images of the brain . This has led to an increasing number of computational programs that can extract data from these images . However , these tools have been specifically designed for certain microscopy techniques . For example , some work on whole-brain datasets while others are built to analyze specific brain regions . Developing a more flexible , standardized method for annotating microscopy images of the brain would therefore enable researchers to analyze data more efficiently and compare results across experiments . To this end , Wang , Zeng , Yang et al . have designed an open-source software program for extracting features from three-dimensional brain images which have been captured using different microscopes . Similar to other tools , the program uses an ‘image registration’ method that is able to recognize and annotate features in the brain . These tools , however , are limited to whole-brain datasets in which the complete anatomy of each feature must be present in order to be recognized by the software . To overcome this , Wang et al . combined the image registration method with a deep-learning algorithm which uses pixels in the image to identify features in isolated regions of the brain . Although these neural networks do not require whole-brain images , they do need large datasets to ‘learn’ from . Therefore , the image registration method also benefits the neural network by providing a dataset of annotated features that the algorithm can train on . Wang et al . showed that their software program , named BIRDS , could accurately recognize pixel-level brain features within imaging datasets of brain regions , as well as whole-brain images . The deep-learning algorithm could also adapt to analyze various types of imaging data from different microscopy platforms . This open-source software should make it easier for researchers to share , analyze and compare brain imaging datasets from different experiments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2021	Bi-channel image registration and deep-learning segmentation (BIRDS) for efficient, versatile 3D mapping of mouse brain
Hippocampal firing is organized in theta sequences controlled by internal memory processes and by external sensory cues , but how these computations are coordinated is not fully understood . Although theta activity is commonly studied as a unique coherent oscillation , it is the result of complex interactions between different rhythm generators . Here , by separating hippocampal theta activity in three different current generators , we found epochs with variable theta frequency and phase coupling , suggesting flexible interactions between theta generators . We found that epochs of highly synchronized theta rhythmicity preferentially occurred during behavioral tasks requiring coordination between internal memory representations and incoming sensory information . In addition , we found that gamma oscillations were associated with specific theta generators and the strength of theta-gamma coupling predicted the synchronization between theta generators . We propose a mechanism for segregating or integrating hippocampal computations based on the flexible coordination of different theta frameworks to accommodate the cognitive needs . The hippocampal formation flexibly combines computations subserving spatial navigation , driven by external environmental cue stimuli ( McNaughton et al . , 1983; O’Keefe and Nadel , 1978 ) , but also memory processing , dependent on internally generated firing sequences ( Pastalkova et al . , 2008; Wang et al . , 2015 ) . The characteristic oscillatory activity patterns in brain networks have been proposed as a mechanism to organize different computations and , depending on the cognitive needs , integrate or segregated them in oscillatory cycles ( Buzsáki and Draguhn , 2004; Engel et al . , 2001; Lisman and Jensen , 2013 ) . In the hippocampus , theta and gamma oscillations are the most prominent rhythms recorded in freely moving animals ( Buzsáki , 2002; Colgin , 2016; Colgin , 2013; Vanderwolf , 1969 ) . Theoretical and experimental work in the hippocampus have associated the processing of environmental cues and the encoding of memories with the input from the entorhinal cortex ( EC ) arriving at CA1 at a particular phase of the theta cycle , and the retrieval of memories with the CA3 output to CA1 in a different phase of the cycle ( Douchamps et al . , 2013; Hasselmo et al . , 2002; Siegle and Wilson , 2014 ) . Information transmission between these regions is proposed to occur in gamma oscillations of different frequencies organized in the phase of the slower CA1 theta rhythm ( Colgin et al . , 2009; Lisman and Idiart , 1995; Lisman and Jensen , 2013 ) . These theta-gamma associations , known as cross-frequency coupling ( CFC ) , are modulated during exploration and memory-guided behaviors ( Cabral et al . , 2014; Canolty et al . , 2006; Colgin , 2015; Fernández-Ruiz et al . , 2017; Lasztóczi and Klausberger , 2016; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014; Tort et al . , 2008 ) . Recent studies have further shown that theta-gamma interaction may vary in a cycle-by-cycle manner within a global hippocampal theta rhythmicity ( Dvorak et al . , 2018; Lopes-Dos-Santos et al . , 2018; Zhang et al . , 2019 ) . However , theta oscillations originating in different anatomical layers of the hippocampus are known to coexist ( Alonso and García-Austt , 1987; Bland and Whishaw , 1976; Buzsáki , 2002; Charpak et al . , 1995; Green and Rawlins , 1979; Vanderwolf C et al . , 1973; Kramis et al . , 1975; Montgomery et al . , 2009; Vanderwolf , 1969; Winson , 1974 ) and , therefore , theta-gamma interactions need to be interpreted in the context of multiple rhythm generators . In addition to the classical medial septum/diagonal band of Broca input imposing a global rhythmicity to the hippocampus and EC , important rhythm generators are located in EC layers II ( EC2 ) and III ( EC3 ) , whose activity reach the dentate gyrus ( DG ) and hippocampus proper through the perforant and temporoammonic pathways , respectively , and from CA3 activity reaching CA1 stratum radiatum through the Schafer collaterals ( Buzsáki , 2002 ) . Importantly , although theta oscillations in the hippocampus are most commonly studied as a unique coherent oscillation across hippocampal layers , exhibiting a characteristic amplitude/phase vs . depth variation ( Buzsáki , 2002 ) , the frequency and phase of the CA3 theta rhythm generator was shown to change relatively independently from the EC theta inputs ( Kocsis et al . , 1999; Montgomery et al . , 2009 ) . How these multiple theta rhythm generators and pathway-specific gamma oscillations interact in the hippocampus is not well understood . One appealing possibility is that different theta oscillations may represent different theta-gamma coding frameworks , providing the substrate to segregate , but also to integrate computations . Here we investigated the function of pathway-specific synchronization of oscillatory activity in the hippocampus of rats freely exploring known and novel environments and resolving a T-maze . Using high density electrophysiological recordings aided by source separation techniques we characterized the dynamic properties of three different theta and three different gamma dipoles in the hippocampus with origins in the CA3 Schaffer collateral layer , the EC3 projection to the stratum lacunosum-moleculare and the EC2 projection to the mid-molecular layer of the DG , respectively , and found strong support for the existence of different theta-gamma frameworks . Using optogenetic tools targeted to CA3 parvalbumin interneurons , we show the specific modulation of the CA3-associated vs . the EC-associated theta generators , demonstrating independent theta oscillations in the hippocampus . Nevertheless , phase shifts between the identified theta frameworks served to coordinate them in pairs or triads , in a sub-second timescale . We then characterized theta-gamma interactions between the different pathways and established an association with the synchronization state in the hippocampal network . Theta-gamma CFC was stronger during higher theta synchronization and we found that pathway-specific gamma oscillations consistently precede theta phase shifts . Finally , we investigated the functional role of these theta-gamma frameworks for contextual learning requiring the update of an existent memory with the changes found in the environment . We found that both , theta-gamma CFC and the coordination between theta oscillations , were consistently higher during mismatch novelty and memory guided decisions , in situations in which the representation of the context from memory is compared against the incoming sensory information . We performed electrophysiological recordings using linear array electrodes across the dorsal hippocampus in five rats ( see Materials and methods , Figure 1 ) . Recordings were carried out while the animal freely explored a familiar open field ( Figures 1 , 2 , 3 , 4 , 5 , 6 ) , a novel open field ( Figure 6 ) or a T-maze ( Figure 7 ) . Using spatial discrimination techniques to separate LFP sources contributed by different synaptic pathways , based on independent component analysis ( ICA , Figure 1—figure supplement 1; Benito et al . , 2014; Fernández-Ruiz and Herreras , 2013; Herreras , 2016; Herreras et al . , 2015; Lęski et al . , 2010; Makarov et al . , 2010; Makarova et al . , 2011; Ortuño et al . , 2019; Schomburg et al . , 2014 ) , we dissected three robust components in all subjects ( Figure 1; Materials and methods ) . The maximum voltages ( Figure 1b ) and dipoles in the current source density ( CSD ) depth profiles ( Figure 1c ) of the three components matched the stratified distribution of known terminal fields in the hippocampus and the currents resulting from stimulation of the corresponding pathways , as previously shown ( Benito et al . , 2014; Fernández-Ruiz and Herreras , 2013; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014 ) . The first component was located in the stratum radiatum , where the CA3 Schaffer collateral/commissural pathway targets the CA1 region ( labelled as Schaffer component or Sch-IC ) . The second matched the EC3 projection in the stratum lacunosum-moleculare ( lm-IC ) , and the third one the perforant pathway from EC2 to the mid-molecular layer of the DG ( PP-IC ) . These three components , referred to as pathway-specific LFPs or IC-LFPs , represent the synaptic contributions with distinct anatomical origins recorded in the LFP ( Herreras , 2016 ) . The power spectra of these signals exhibited a clear peak at theta frequency ( 6–10 Hz ) and broadband gamma activity ( Figure 1e ) . CA3 and EC3 neurons have been shown to fire phase locked to discrete gamma band oscillations in the downstream Sch-IC and lm-IC , respectively ( Fernández-Ruiz et al . , 2012a; Schomburg et al . , 2014 ) , with gamma oscillations segregated in the phases of the theta wave recorded in CA1 ( Colgin et al . , 2009; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014 ) . In good agreement , pathway-specific gamma activities were distributed in the theta cycle , with lm-IC close to the theta peak ( π radians ) and followed by Sch-IC ( Figure 1f ) , showing consistency with the firing properties of principal neurons in their respective upstream afferent layers . Similarly , entorhinal principal cells in EC2 and EC3 were shown to fire in anti-phase , relative to the theta oscillation ( Mizuseki et al . , 2009 ) and , accordingly , large amplitude gamma oscillations in PP-IC and lm-IC in our recordings were found shifted 180° ( Figure 1f ) . These results support the use of multichannel recordings and source separation tools to investigate interactions between theta and gamma current generators in multiple layers of the hippocampal formation . The relative phase and coherence of these components was first compared with the theta oscillation recorded at the pyramidal layer of CA1 , commonly used as the reference for temporal interactions in the hippocampus . We measured coherence and the inter-cycle phase clustering index ( ICPC ) which , in addition to a measure of coherence , computes the phase differences between signals in a cycle by cycle basis ( see Materials and methods ) . All IC-LFPs exhibited prominent coherence with the LFP signal mainly in the theta range ( and its first harmonic , Figure 1g ) . Similarly , the coherence at theta frequency was high between IC-LFP pairs ( 0 . 41 , 0 . 31 , 0 . 61 , for Sch-lm , Sch-PP and lm-PP , respectively , Figure 1g ) , being larger between EC-associated generators ( p<0 . 0001 , ANOVA with degrees of freedom corrected by Greenhouse-Geisser , F ( 1 . 091 , 4 . 363 ) =89 . 33 ) . To illustrate the contribution of source separation analysis ( ICA ) to these results , we compared the coherence between IC-LFPs against that of LFP signals recorded at the site with the maximum contribution to each IC-LFP ( i . e . str . radiatum , lacunosum-moleculare and hilus for Sch-IC , lm-IC and PP-IC , respectively; Figure 1g , dashed lines ) . This comparison revealed higher coherence between raw LFPs at different frequency bands , likely due to volume conduction between channels , not present in the IC-LFPs . For the theta band , the differences were most evident in the stratum radiatum , where theta coherence was significantly lower between IC-LFPs ( Figure 1g ) . Differences were also notable in the gamma bands for all regions , which will be relevant in further analysis below . Note at this point that the extraction of highly coherent signals is perfectly compatible with ICA . This methodology finds components that are spatially distributed , and only requires small differences in their temporal co-variation ( i . e . temporal jitter and/or amplitude variation ) . Therefore , ICA allows the separation of sources , even if there is a high coherence between them ( Makarova et al . , 2011; see Materials and methods ) . The coherence results were confirmed by the ICPC analysis , demonstrating a significant coupling in the theta range with the reference LFP signal ( ICPC = 0 . 50/0 . 73/0 . 61 for Sch-IC/lm-IC/PP-IC vs . CA1 LFP , respectively , p<0 . 0001 , surrogate test ) , which also showed the characteristic phase shift across layers ( π/2 , 0 . 8π and 1 . 1π radians for Sch-IC , lm-IC and PP-IC , respectively; Figure 1h ) . Interestingly , the lack of coherence closer to the unit between theta oscillations in the IC-LFPs was already suggesting the coexistence of different theta current generators with certain degree of independency , rather than the artificial breakdown of a unique theta rhythm into spatially segregated components . In the latter , the coherence between the oscillations should be maximum , as they would be three components of one single wave . To provide direct evidence of the independency between theta current generators , we next used an optogenetic approach ( Figure 2 ) . Using a transgenic rat cre line ( LE-TG[Pvalb-iCre]2Ottc , NIDA , USA ) and adeno-associated virus ( AAV1-EF1a-DIO-hChR2 ( H134R ) -eYFP-WPRE-hGH , Penn Vector Core , USA ) injected in the dorsal CA3 ( Figure 2a ) , we expressed the excitatory Channelrhodopsin-2 ( ChR2 ) in parvalbumin positive ( PV+ ) interneurons ( Figure 2b , Materials and methods ) . Animals ( n = 3 ) implanted bilaterally with optic fibers targeting dorsal CA3 and multichannel electrophysiological recordings as before ( Figure 1 ) were used to test the hypothesis that hippocampal theta generators can be modulated independently by activating CA3 PV-interneurons and decreasing the Schaffer collateral output . As shown in Figure 2c–d , blue light illumination ( 460 nm ) in animals freely-exploring an open field significantly and specifically decreased the power of theta in the Sch-IC and the corresponding pathway-specific low gamma oscillations . In contrast , the oscillatory activities ( power and peak frequency ) in lm-IC and PP-IC were preserved ( Figure 2c ) . This finding was highly robust across animals ( Figure 2e ) . The modulation of theta power specifically in the Sch-IC with the preservation of peak theta frequencies in the three generators , conclusively demonstrated the existence of independent theta oscillators in CA3 and EC . We next investigated in more detail the functional interactions between the three IC-LFP theta frequencies . To reduce the error of the theta phase estimation to less than 1% of the theta cycle , we selected for further analysis only those epochs in which theta power was four times higher than delta ( 1–4 Hz ) activity ( Materials and methods and Figure 1—figure supplement 2 for a mathematical validation of this threshold ) . As expected from the results in Figure 1g and h , theta interactions between pathways were not constant in time , but presented periods of high and low synchronization ( Figure 2f ) . To get insight into these states , we computed a dynamic ICPC for each theta cycle , measuring the variation of the phase relationship between theta oscillations with respect to the previous and consecutive cycles ( Materials and Methods , Figure 1—figure supplement 3 ) . The dynamic ICPC was measured for all pairs of IC-LFPs and for the three signals simultaneously ( Figure 2g , Figure 2—figure supplement 1 ) . The distribution of the ICPC over all animals showed a peak close to perfect phase locking , but with an important tail of low-synchronized epochs , with approximately the 20% of the cycles presenting an ICPC lower than 0 . 8 . Again , the highest synchronization was found between PP-IC and lm-IC , in agreement with the coherence analyses ( Figure 1g ) and consistent with the likely origin of these generators in two sublayers of the same cortical regions ( EC ) . As we did for the coherence analysis , we also computed the ICPC from the raw LFP signals recorded at str . radiatum , lacunosum-moleculare and hilus . This analysis showed significantly higher estimates of phase coupling based on LFPs than for IC-LFPs ( averaged ICPC = 0 . 88/0 . 93 for the three IC-LFPs/raw LFPs , p<0 . 01 , paired t-test , t = 5 . 35 , N = 5 ) . Perfect phase locking ( ICPC = 1 ) was strongly reduced in IC-LFP signals , unveiling the spurious coupling measured on the raw LFPs likely due to the mixture of sources by volume conduction . Therefore , by separating the sources , ICA allowed us to clearly identify shifts in theta synchronization across hippocampal layers . In the previous analysis , we used three consecutive theta cycles to compute the ICPC . We took this value as a trade-off between time resolution and a robust estimation of the metric . However , recent works have demonstrated that theta dynamics in the hippocampus may rapidly change between single cycles ( Dvorak et al . , 2018; Lopes-Dos-Santos et al . , 2018; Zhang et al . , 2019 ) . To overcome this limitation and understand better the temporal dynamics of the theta couplings , we analyzed the variability of the ICPC across time by comparing the value of a given cycle to that of the previous ones ( Figure 2i ) . These results provided a monotonously rising curve up to 0 . 75 s; from there on , the curve hardly increased . This indicates that the coupling strength between consecutive cycles spreads on a time scale in the order of one second , thus expecting dynamical changes in the ICPC in this time scale . Overall , this methodology allowed the characterization of the temporal synchronization between theta generators with a time resolution of one theta cycle , highlighting dynamical changes in the coupling strength between hippocampal pathways in the theta range . The above results supported the coexistence of different temporal frames in the theta range to organize hippocampal activity . Thus , since gamma activity is nested to the theta cycle , it opened the possibility to multiple theta-gamma interactions ( Figure 3a ) . For comparison , we first followed a conventional approach to the analysis of theta-gamma phase-amplitude CFC , taking as a phase reference the theta in the LFP recorded in the pyramidal layer , as is usually done ( Colgin , 2015; Colgin et al . , 2009; Csicsvari et al . , 1999; Fernández-Ruiz et al . , 2017; Lasztóczi and Klausberger , 2016; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014; Tort et al . , 2009; Tort et al . , 2008 ) , and using the modulation index ( MI ) introduced by Tort et al . , 2008 ( Figure 3b ) . This analysis identified the typical coupling between CA1 theta and a slow gamma band of CA3 origin ( Sch-IC; maximal modulation at 37 . 5 ± 5 Hz; Colgin et al . , 2009; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014 ) and a medium gamma band of EC3 origin ( lm-IC; 82 . 5 ± 4 Hz medium gamma; Colgin et al . , 2009; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014 ) . The analysis also revealed an additional theta-nested fast gamma band ( 130 ± 10 Hz , ) in the mid-molecular layer of the DG overlapping the terminal field of EC2 inputs , compatible with the previously found theta-gamma CFC in the DG ( Bragin et al . , 1995 ) . We then computed the CFC using as references the different theta oscillations separated in the IC-LFPs . The key new finding was the systematic observation of stronger theta-gamma CFC in IC-LFPs vs . LFPs ( Figure 3b ) . We further tested the robustness of these phase-amplitude CFCs by using the alternative methodology proposed in Canolty et al . , 2006 ( Figure 3—figure supplement 1 ) , obtaining similar results . The result was not totally unexpected , since we had found that the theta oscillation recorded in the LFP and typically used as a reference in CFC analysis was indeed a mixture of different theta generators of variable coherence ( see Figure 2d–e above ) . It has been argued that , in the case of low gamma frequencies in the hippocampus , the measured theta-gamma CFC could be a spurious effect due to the asymmetry in the theta wave ( Belluscio et al . , 2012; Cole and Voytek , 2019; Cole and Voytek , 2017 ) and/or theta harmonics ( Juhan et al . , 2015; Lozano-Soldevilla et al . , 2016 ) . This limitation , however , can be mitigated by a definition of the theta oscillation that takes into account its asymmetry instead of just applying a band-pass filter at theta frequency ( Belluscio et al . , 2012; Cole and Voytek , 2019; see Materials and methods ) . We checked the effect of theta asymmetry in our dataset with a multiple linear regression analysis , where the power at each gamma band was determined by theta power and asymmetry ( Colgin , 2016; Zheng et al . , 2015; Figure 3—figure supplement 2 ) . We also included running speed as it has been shown to co-vary with the power and frequency of hippocampal gamma oscillations ( Ahmed and Mehta , 2012; Zheng et al . , 2015 ) . We considered two factors to measure theta asymmetry: the ratio between the duration of rise and decay phases in each cycle and the ratio between the duration of the peak and the trough ( Cole and Voytek , 2018; see Materials and methods ) . The analysis confirmed the influence of theta power and speed on gamma power ( Zheng et al . , 2015; Figure 3—figure supplement 2 ) , and a negligible contribution of theta asymmetry . This result supports the existence of a genuine low-gamma activity band in CA1 ( Colgin et al . , 2009; Dvorak et al . , 2018; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014; Tort et al . , 2009; Zhang et al . , 2019 ) and the physiological value of its coupling with the theta oscillation . We finally asked whether pathway specific gamma activities were preferentially coupled to the theta oscillation in their same afferent pathway , likely reflecting local computations , or in different pathways , thus reflecting inter-pathway interactions , or both . Results in Figure 3c–d demonstrated a dominant CFC between oscillations recorded in the same IC-LFP . Therefore , theta-gamma CFC mainly reflects pathway-specific interactions , rather than a unique carrier theta wave to which the gamma activity from different origins is multiplexed in segregated theta-gamma channels . The higher theta-gamma CFC found by using pathway-specific theta references , provided yet another indication of the coexistence and relevance of distinct temporal theta frameworks in the hippocampus . Having shown that theta generators can be modulated independently and present variable synchrony ( Figure 2 ) and gamma nesting is pathway-specific ( Figure 3 ) , we next explored the theta and gamma features accounting for the different synchronization states . We found that theta power in all IC-LFPs correlated with the ICPC , with larger theta power associated with states of higher synchronization ( Figure 4a , b and c ) . Interestingly , the frequency of the theta oscillation was constant across synchronization states in the Sch-IC , but varied in the two EC-associated generators ( Figure 4a ) . Theta frequencies in lm-IC and PP-IC increased with ICPC ( Figure 4b and c ) . Regarding gamma activity , broadband power did not correlate with theta synchronization ( not shown ) , in contrast to narrowband power ( slow/medium/fast gamma for Sch-IC/lm-IC/PP-IC , respectively ) , which correlated with the ICPC in lm-IC , but not in Sch-IC nor PP-IC ( Figure 4b ) . Because running speed also correlates with hippocampal theta power and frequency ( Vanderwolf , 1969 ) , we performed a multiple linear regression analysis including running speed , theta power and theta frequency as explanatory variables to predict the ICPC ( Figure 4c ) . This analysis allowed us to estimate the contribution of each variable to the ICPC that cannot be accounted by any other variable in the model . We used for the analysis all theta cycles recorded while animals were exploring a familiar open-field . The MI was not included in the multiple linear regression since its value for individual theta cycles is not reliable . The result demonstrated the main effect of theta power on the ICPC value , with a lower , but significant , contribution of theta frequency and running speed ( Figure 4c , p<0 . 05 , t-test against zero between beta values of each factor , Bonferroni corrected , Materials and methods ) . Finally , this analysis unveiled a striking correlation between the CFC and theta synchronization ( Figure 4b and d ) . Strong theta-gamma modulation was associated with high ICPC values , while weak or nearly absent CFC was found in periods of low synchronization . Note that , as mentioned above , only cycles with high theta power activity were selected in this analysis ( Figure 4d ) , so that signal’s power could not affect the estimation of its phase ( Figure 1—figure supplement 2 ) , thus preventing the introduction of any bias in the synchronization measurement ( Figure 1—figure supplement 2 , Materials and methods ) . This result indicated that within-pathway CFC was associated to the synchronization between pathways . Theta and gamma oscillations reflect the extracellularly added excitatory and inhibitory synaptic and active dendritic currents of two processes occurring at different timescales ( Herreras , 2016 ) . We hypothesize that CFC may reflect a mechanism through which fast excitation-inhibition interactions organize the activity of principal cells in different theta frameworks found in our analysis . We then looked for an indication of directionality in the interaction between the two frequencies , and computed the cross-frequency directionality index ( CFD; Jiang et al . , 2015 ) , based on the phase-slope index to compute the phase difference between two signals . This methodology was specially developed to estimate the directionality between signals with large differences in signal to noise ratio , as theta and gamma frequencies , demonstrating in these conditions to be more efficient than classical approaches such as Granger Causality ( Granger , 1969; Jiang et al . , 2015; Nolte et al . , 2010 ) . In CFD , an increase of the phase difference between the theta phase and the gamma amplitude with frequency gives rise to a positive slope of the phase spectrum ( i . e . a positive CFD value ) when the phase of the slow oscillation consistently precedes the amplitude of the fast , this is , when the time difference between a theta phase and the next burst of gamma activity is constant . The slope is negative when the amplitude of the fast oscillation consistently precedes the phase of the slow or , in our analysis , when the delay from the gamma activity to the next theta cycle is constant . As shown in Figure 5a for the group data , and Figure 5—figure supplement 1 for individual animals , CFD resulted in negative values ( amplitude-phase coupling , APC ) for the specific gamma bands nested to the theta oscillations in the corresponding IC-LFPs . This gamma amplitude to theta phase directionality is given by the consistent anticipation of the gamma activity to the theta phase . In Figure 5b , there is a representative example of the gamma-to-theta directionality in PP-IC . The delay from gamma to theta is almost fixed , while in the opposite direction ( theta to gamma ) is highly variable . It should be noted that the CFD is not exempt of limitations and , as the CFC , the presence of harmonics and theta asymmetries may result in spurious measurements of directionality ( Lozano-Soldevilla et al . , 2016 ) . To validate the finding , we also computed the CFD directly in the LFP signals recorded in the different hippocampal layers . To compare with the IC-LFPs , we chose the LFP signals from the channels matching the site of maximum contribution to each IC-LFP ( Figure 5c and d ) . Negative values of CFD were found in the LFPs in str . lacunosum-moleculare and in the DG , supporting the driving role of gamma oscillations over the phase of the theta waves . We could not find a significant directionality in the case of the str . radiatum LFP and , in all cases , the absolute value of the CFD was higher using IC-LFPs than LFPs ( Figure 5d ) , demonstrating that source-separation tools outperform the use of raw LFPs to investigate pathway interactions . Overall , our CFD analysis suggests that the neuronal circuits supporting gamma oscillations in the hippocampus set the timing of principal cells activity in the theta range , as reflected in the phase of the recorded theta oscillations . Previous studies have shown that both CFC and inter-regional coherence , independently , correlate with learning ( Canolty et al . , 2006; Engel et al . , 2001; Fries , 2015; Fries , 2005; Palva et al . , 2005; Tort et al . , 2009; Tort et al . , 2008 ) . Our analysis ( Figure 4 ) now showed that both phenomena seem to be linked . Therefore , in our final set of experiments , we looked for behavioral evidence in support of the hypothesis that they are part of a common mechanism to flexibly integrate or segregate neuronal computations . More specifically , we hypothesized that layer-specific interactions would phase-lock theta oscillations between layers to facilitate the integration of CA3-mediated and EC-mediated information streams in CA1; for instance , in learning conditions requiring the comparison of context representations from memory ( Sch-IC pathway ) and from the environment ( lm-IC and PP-IC pathways; Buzsáki and Moser , 2013; Dudai and Morris , 2013; Wang and Morris , 2010 ) . One such learning conditions is mismatch novelty ( Lever et al . , 2006 ) , in which the subject is re-exposed to a previously visited context which has been modified . The ‘mismatch’ occurs when comparing the expected representation from memory and the found representation in the environment . To test whether ICPC and CFC increase in parallel during mismatch novelty , we trained the animals in a task in which , after habituation to an open field ( 8 min session one per day during 8–10 days , Figure 6a control ) , we introduced a novel tactile stimulus in the floor of the otherwise unchanged field ( Figure 6a novelty , see Materials and methods ) . We computed and compared theta synchrony and CFC between the novelty session and the habituation session the day before . When the animal entered the arena , the ICPC between theta oscillations was high and comparable in both conditions during the first two minutes of exploration ( Figure 6b , t1 ) . As the animal explored the context , synchronization remained high during novelty , but rapidly decayed in the known environment ( Figure 6b , t2 ) . Consistent with the notion of information transmission to update an existing memory , by the end of the exploration time both conditions decreased to the same level of theta synchronization ( Figure 6b , t3 ) , when the introduced tactile stimulus had lost its novelty . As a control , we tested locomotor activity comparing movement velocity between novelty and habituation sessions ( Figure 6c ) , without finding differences between sessions ( p>0 . 3 , t-test , Figure 6c ) . Therefore , differences in the ICPC cannot be solely explained by changes in locomotor activity . We used a multiple linear regression analysis as before , to investigate now the independent contribution of theta power and frequency , experimental condition ( control vs . novelty sessions ) , running speed and time in the task to the measured ICPC ( Figure 6d ) . We found that theta power and the experimental condition are the main factors that contribute to the ICPC value , with other variables such us running speed and time marginally contributing . We then computed the CFC index ( MI ) in the same recordings and found that it paralleled the changes in theta synchronization during the complete session in both conditions , as shown in Figure 6b and e . The CFC strength was higher in the three IC-LFPs during the novelty sessions associated with the higher theta synchronization , and decreased towards the end of the session in parallel with the ICPC ( Figure 6e ) . Previous studies have shown that CFC in EC pathways preferentially occurs when the animal is rearing on its hind legs , an exploratory response to novelty ( Lever et al . , 2006 ) , which is also associated with increased theta frequency ( Barth et al . , 2018 ) . To investigate the potential contribution of rearing behavior to our findings in the mismatch novelty task , we removed from our recordings the epochs in which animals were rearing on their hind legs and then reanalysed ICPC and CFC . As shown in Figure 6—figure supplement 1 , the increased theta synchronization between the three IC-LFPs was maintained during novelty in the absence of rearing epochs . Similarly , the MI was higher during novelty , although more variable , likely reflecting the decrease in the number of data samples after rearing removal . We concluded that the changes found in mismatch novelty cannot be solely explained by the rearing behavior . Furthermore , we measured theta frequency in the complete time series and compared it between control and novelty conditions , and found a significant decrease for Sch-IC and PP-IC theta frequency in t2 ( Figure 6—figure supplement 2; Wells et al . , 2013 ) . In a second behavioral experiment , a hippocampus-dependent delayed spatial alternation task was used in which the animal needed to remember the arm visited in the previous trial and to update the memory with the choice made in the current trial ( Ainge et al . , 2007; Montgomery and Buzsáki , 2007; Wood et al . , 2000 ) , again relying on the interaction between context representations from memory and from external sensory cues . Rats learned in an 8-shaped T-maze to alternate between the left or right arms on successive trials for water reward ( Figure 7a ) until they reached performances above 80% . In this task , the central arm is associated with memory recall , decision making and encoding of the current decision ( DeCoteau et al . , 2007; Montgomery and Buzsáki , 2007; Tort et al . , 2008; Wood et al . , 2000 ) , while neuronal recordings in the side arm are thought to convey little information to predict behavioral outcomes in the following trial ( Pastalkova et al . , 2008; Schomburg et al . , 2014 ) . Using this task , previous independent studies showed a phase shift between theta oscillations recorded in CA1 and CA3 pyramidal layers ( Montgomery et al . , 2009 ) and increased CFC in the CA1 radiatum and lacunosum-moleculare IC-LFPs ( Schomburg et al . , 2014 ) associated to the central arm . Therefore , in this analysis we wanted to validate our hypothesis finding concomitant increases in theta-gamma CFC and theta ICPC in the central arm , and extend previous findings by incorporating the PP-IC into the analysis . We computed and compared theta-gamma CFC and theta synchronization in recordings obtained from the central and side arms in correct trials , selecting only those epochs were the movement velocity was comparable in both arms ( Figure 7b ) . We found significantly increased CFC in the central arm for the three IC-LFPs ( Figure 7c ) . Importantly , concomitant with CFC , we also found an increase in theta ICPC in the central arm during the same epochs ( Figure 7d and e ) . The results of the two behavioral tasks , thus , demonstrate that theta-gamma CFC and theta ICPC across generators are linked and preferentially occur during memory-guided exploration and mismatch novelty detection , two conditions in which internally generated memory representations need to be integrated with the incoming sensory information about external cues . They support the idea that different theta-gamma frameworks may flexibly coordinate information transmission in the hippocampus . In this work , we have separated the LFP sources contributed by different synaptic pathways using spatial discrimination techniques based on independent component analysis ( Benito et al . , 2014; Fernández-Ruiz and Herreras , 2013; Herreras , 2016; Makarov et al . , 2010; Makarova et al . , 2011 ) . This processing step allowed us to work with a more reliable representation of the local electrophysiological dynamics , as compared to raw LFPs or CSDs ( Martín-Vázquez et al . , 2013 ) . The main drawback of LFPs is the multisource origin of the signals , a blend of dipolar ( or quadrupolar ) field potentials . While the CSD of multisource raw LFPs avoids the problem of volume conduction , it does not separate the co-activating current sources in the recorded region , hence the CSD of pathways targeting the same cells ( e . g . CA1 pyramidal cells ) overlap and add/subtract , cancelling each other . In these conditions , the time course of the CSD is a composite one ( as it is that of native LFPs ) and cannot be unambiguously assigned to any of the co-activating sources . Therefore , source separation techniques are necessary to obtain the correct time course of each individual synaptic contribution . Extensive previous research has demonstrated the existence of multiple theta rhythms and current generators in the hippocampus and EC ( Buzsáki , 2002 ) . While septal activity is required for theta rhythmicity , and lesions targeting the medial septum eliminate theta oscillation in both structures , intrinsic hippocampal activity from CA3 and extrinsic EC inputs do also contribute to the recorded theta oscillations ( Buzsáki , 2002 ) . Surgical removal of the EC unveils a theta oscillation that depends on the integrity of CA3 and is highly coherent across hippocampal layers ( Bragin et al . , 1995 ) . In the presence of an intact EC , however , the coherence between theta signals in the stratum radiatum and lacunosum-moleculare is reduced ( Kocsis et al . , 1999 ) , consistent with an input competition between CA3 and EC3 . Now , using optogenetic tools targeted to CA3 PV+ interneurons ( Figure 2 ) , we provide new results that conclusively support the coexistence of independent theta oscillations in the hippocampus , by showing the specific modulation of the CA3-associated Sch-IC vs . the EC-associated theta generators ( lm-IC and PP-IC ) . Furthermore , variations in theta power and frequency in each generator occurred dynamically and independently ( Figure 4 ) . Nevertheless , phase shifts between the identified theta frameworks served to coordinate them in pairs or triads , in a sub-second time-scale ( Figure 2 and Figure 2—figure supplement 1 ) . In turn , we speculate , these hippocampal theta coupling states would be associated with distinct brain-wide network states . In support of this view , we found selective behavioral/cognitive functions associated with different states of between-framework theta synchronization ( Figures 6 and 7 ) . Finally , the dynamic change in theta frequency observed in individual pathway-specific LFPs ( Figure 4 and Figure 6—figure supplement 2 ) also argues against the view of the identified theta frameworks as monolithic oscillations driven by external pacemakers , and rather suggests the cooperation of weakly-coupled local oscillators and global rhythm generators for the fine-tuning of theta oscillations . Thus , theta activity in the hippocampus is neither unique , nor monolithic . Overall , taking brain oscillations as rhythmic changes in neuronal excitability that can define sequential information packages ( Canolty and Knight , 2010; Fries , 2005 ) , the dynamic variation in theta synchrony found in the hippocampus likely reflects multiple theta-coordinated time-frames , with phase differences between oscillations having a large impact on the timing of neuronal firing in the respective layers . Synchronization of theta frameworks would , in turn , coordinate , though not necessarily synchronize ( Mizuseki and Buzsaki , 2014 ) , firing sequences in consecutive hippocampal stations . The processing streams thus generated could transmit independent information , e . g . driven by memory retrieval or external environmental cues , or the result of integrating/comparing both information sources , depending on the cognitive needs . Interactions between the phase of the theta oscillation and the amplitude of the gamma activity have been extensively documented and proposed as an effective mechanism to integrate activity across different spatial and temporal scales ( Bragin et al . , 1995; Bruns and Eckhorn , 2004; Buzsáki and Draguhn , 2004; Canolty et al . , 2006; Canolty and Knight , 2010; Colgin , 2015; Colgin et al . , 2009; Engel et al . , 2001; Fell and Axmacher , 2011; Lakatos et al . , 2008; Lakatos et al . , 2005; Lisman and Idiart , 1995; Lisman and Jensen , 2013; Mormann et al . , 2005; Palva et al . , 2005; Saleh et al . , 2010; Soltesz and Deschênes , 1993; Tort et al . , 2009; Tort et al . , 2008; Zheng et al . , 2016 ) . Our analysis demonstrates that phase-amplitude CFC between theta and gamma oscillations in the hippocampus is selective for theta-frameworks ( Figure 3 ) . Interactions between theta and gamma were higher when IC-LFP theta oscillations were used as the temporal reference for the corresponding layer-specific gamma activities , instead of a single LFP recording , as commonly done . This observation , together with previous and important evidence demonstrating that firing of principal cells in CA3 and EC3 is phase-locked to downstream theta-nested gamma oscillations recorded in the CA1 stratum radiatum and lacunosum-moleculare , respectively ( Colgin et al . , 2009; Lasztóczi and Klausberger , 2014; Schomburg et al . , 2014 ) , suggests that local layer-specific circuits interact with upstream afferent pathways to organize hippocampal cell assemblies in multiple theta-gamma frameworks . The significant positive correlation between CFC strength and theta synchronization found in our study ( Figure 4 ) further suggests that layer-specific CFC might reflect the mechanism for theta framework coordination . Co-modulation of gamma amplitude and theta phase can be the result of a theta-driven process increasing gamma activity , a gamma-driven modulation of theta phase , or due to the presence of a common external drive for both components , fast and slow , simultaneously . The CFD index has been previously shown to reveal both fast-to-slow and slow-to-fast frequency interactions in modelled data and real electrophysiological recordings ( Helfrich et al . , 2019; Helfrich et al . , 2018; Jiang et al . , 2015; Zheng et al . , 2017 ) . We applied it here , for the first time to hippocampal IC-LFPs , and unveiled , in contrast to a generalized assumption in the field , a predominant gamma-to-theta interaction ( Figure 5 ) . This directionality was confirmed directly on the LFP signals ( Figure 5 ) , although the contribution of a third input controlling simultaneously both rhythms cannot be fully discarded . We do not take this result as an indication of theta oscillations in the hippocampus being generated by gamma activity . On the contrary , we suggest that gamma activities , reflecting the interplay of inhibitory-excitatory networks ( Cardin et al . , 2009; Neymotin et al . , 2011; Orbán et al . , 2006; Rotstein et al . , 2005; Stark et al . , 2013; Tort et al . , 2007 ) , impose phase shifts on the on-going theta oscillations in their corresponding layers . Therefore , local gamma-generating circuits , driven by afferents from their respective upstream layers , might not be activated at a particular theta phase , but rather be coordinating principal cells activity and setting the phase of the local theta oscillation . While dissecting the precise circuit mechanisms supporting the above gamma-to-theta interaction is out of the scope of the present work , several possibilities exist . Computational works have demonstrated that theta-gamma CFC emerges from the interactions between functionally distinct interneuron populations interconnected in a network of principal cells receiving an external theta rhythm generator , such as the septal input ( Neymotin et al . , 2011; Orbán et al . , 2006; Rotstein et al . , 2005; Tort et al . , 2007 ) . Subsets of interneurons can phase-lock to different hippocampal rhythms ( Klausberger et al . , 2003; Klausberger and Somogyi , 2008 ) and , interestingly , recent findings have shown in the CA1 region that some interneurons can specifically phase-lock to slow-gamma and others to medium-gamma , supporting the idea that different classes of interneurons drive slow and medium gamma oscillations ( Colgin , 2015; Fernández-Ruiz et al . , 2017; Lasztóczi and Klausberger , 2014 ) . Thus , an appealing mechanism for the gamma-modulation of theta phase would be the control of different interneuron classes by pathway-specific inputs , which would entrain specific gamma networks modulating principal cell excitability and firing in response to on-going theta inputs , advancing or delaying theta phases . In support of this hypothesis , recent analyses of theta-gamma associations on a theta cycle-by-cycle basis have demonstrated a significantly higher spike-field phase synchrony for interneurons than pyramidal cells in the theta band ( Zhang et al . , 2019 ) . Finally , spiking resonance in principal cells may contribute to this mechanism too , since optogenetic activation of basket interneurons ( PV-cells ) in the hippocampus and neocortex pace pyramidal cell firing in the theta range , by virtue of postinhibitory rebound of Ih activity ( Stark et al . , 2013 ) . In that experiment , theta-band firing of excitatory neurons required rhythmic activation of inhibitory basket cells , as white noise activation effectively modulated their activity but did not entrained pyramidal theta-band firing ( Stark et al . , 2013 ) . Feed-forward activation of interneurons from upstream layers or an external rhythmic input ( i . e . cholinergic or GABAergic inputs form the septum ) are thus required for resonance amplification . Intrinsic cellular properties and network mechanisms may thus interact to support gamma-dependent coordination of theta phases across hippocampal layers . The above interpretation would also explain phase-phase coupling between CA1 theta and CA1 slow- and medium-gamma ( Belluscio et al . , 2012 ) , as the consequence of theta phase driven by pathway-specific gamma activity entrained by upstream inputs in CA3 and EC3 , respectively . Recent studies , however , highlighted the importance of frequency harmonics and waveform asymmetry when measuring phase-phase coupling ( Scheffer-Teixeira and Tort , 2016 ) and also amplitude-phase CFC ( Cole and Voytek , 2017 ) . Waveform asymmetry in oscillatory activity introduces spectral content that cannot be defined solely by sinusoidal components ( Amzica and Steriade , 1998 ) and , therefore , may result in spurious CFC and CFD ( Juhan et al . , 2015; Cole and Voytek , 2017; Kramer et al . , 2008; Lozano-Soldevilla et al . , 2016; Scheffer-Teixeira and Tort , 2016 ) . Several approaches have been developed to overcome these limitations ( Cole and Voytek , 2019; Kramer et al . , 2008 ) , improving the estimation of the theta phase and minimizing the effect of sharp edges . We applied these methods in our analysis ( Materials and methods ) . The specific waveform of an oscillation should not be seen as a problem but as a source of physiological information when appropriately considered ( Cole and Voytek , 2017 ) . The proposed scenario provides a mechanism to coordinate distributed computations organized in theta waves by synchronizing theta oscillations through theta-gamma CFC . We reasoned that layer-specific interactions would phase-lock theta oscillations between layers when the integration of CA3- and EC-associated information streams is required ( Buzsáki and Moser , 2013; Dudai and Morris , 2013 ) . We selected two well-known behavioral tasks to test this hypothesis . We first used a mismatch novelty task ( Lever et al . , 2006 ) in which the memory representation of the context , involving the CA3-associated pathway , is compared against the novel ( mismatch ) sensory input , conveyed by the EC-associated pathways ( to CA1 and DG ) . Our hypothesis predicted that , during the novelty condition , a concomitant increase in CFC and theta synchronization in the three theta-gamma frameworks should occur , something that we found experimentally ( Figure 6 ) . This result could not be explained solely by the animal’s speed , which was indistinguishable in our experiments between known and novelty conditions , nor by rearing behavior , sometimes associated with novelty exploration ( Barth et al . , 2018 ) . A link between CFC and theta synchrony was also found in the central arm of the 8-shaped T-maze after correcting for running speed ( Figure 7 ) , in the location where the interaction between context representations from memory and from external sensory cues take place for decision making and encoding ( DeCoteau et al . , 2007; Montgomery and Buzsáki , 2007; Tort et al . , 2008; Wood et al . , 2000 ) . These results bring together previous independent findings showing phase shifts between theta oscillations recorded in CA1 and CA3 pyramidal layers ( Montgomery et al . , 2009 ) and increased CFC in the CA1 radiatum and lacunosum-moleculare IC-LFPs ( Schomburg et al . , 2014 ) associated to the central arm . A recent study using an uncharted novelty test showed increased theta-gamma CFC exclusively in EC pathways , but not in the CA3 pathway ( Barth et al . , 2018 ) . Importantly , in contrast to mismatch novelty , uncharted novelty involves the exposure to a previously unvisited context and therefore it lacks a memory representation . Thus , in the absence of a memory representation , only EC-pathways conveying information about the environmental cues demonstrate enhanced theta-gamma coupling , lending support to our hypothesis . Finally , we found that theta frequency decreased in the mismatch novelty condition in the Sch-IC and PP-IC ( Figure 6—figure supplement 2; Wells et al . , 2013 ) , while it was reported to increase in the EC-pathways in uncharted novelty ( Barth et al . , 2018 ) , suggesting that theta frequency modulation was required to couple the three theta frameworks during mismatch novelty . Important recent studies have investigated theta oscillations in a cycle-by-cycle manner ( Dvorak et al . , 2018; Lopes-Dos-Santos et al . , 2018; Zhang et al . , 2019 ) , demonstrating highly dynamic changes in spectral components and theta-gamma interactions associated to different behaviors . These studies support the notion that individual theta cycles represent flexible temporal units to transiently organize CA1 computations . We found that layer-specific theta oscillations coexisting in the hippocampus couple and decouple dynamically , and we propose that it reflects a mechanism to integrate or segregate computations , respectively . This possibility is fundamentally different from previous ones based on the segregation of computations in the phase of the theta wave ( Colgin et al . , 2009; Lisman and Idiart , 1995; Lisman and Jensen , 2013 ) , or in single theta cycles as indicated above ( Dvorak et al . , 2018; Lopes-Dos-Santos et al . , 2018; Zhang et al . , 2019 ) , in that those were based on the rapid alternation of computational modes between phases or cycles , respectively , but always of a unique theta framework . In contrast , our new proposal contemplates parallel processing in cell assemblies receiving information from different theta frameworks . A decrease in the coherence between the theta oscillations would decouple the processing streams , segregating the underlying cognitive processes ( i . e . retrieval from encoding ) . An increase in the coherence would rather couple them , facilitating the integration in CA1 neurons and downstream regions of both information streams ( i . e . when stored and ongoing contextual information need to be compared ) . Interestingly , however , the two models complement each other , since computations in each theta framework would likely vary in a cycle-by-cycle manner , representing an even more versatile coding framework . Interactions between slow and fast brain oscillations have been measured in multiple brain regions during perception , attention , learning and memory formation ( Buzsáki and Draguhn , 2004; Engel et al . , 2001; Lisman and Jensen , 2013 ) . Despite its ubiquitous presence in fundamental cognitive processes , its function is largely unknown . Our results provide a mechanism for parallel processing in the hippocampus based on the coexistence of multiple theta frameworks that support both , segregated or integrated computations , depending on their synchronization level . Important questions remain to be answered . How theta synchronization in the hippocampus relates to hippocampal-neocortical interactions ( Siapas et al . , 2005; Sirota et al . , 2008 ) known to be favoured at theta and beta frequencies ( Igarashi et al . , 2014; Moreno et al . , 2016 ) and modulated by synaptic plasticity in the hippocampus ( Álvarez-Salvado et al . , 2014; Canals et al . , 2009 ) ? The conditions triggering the coordination between theta-gamma frameworks are not well understood , but given that theta-gamma uncoupling seems to represent an early electrophysiological signature of hippocampal network dysfunction in Alzheimer’s disease ( Goutagny et al . , 2013; Iaccarino et al . , 2016; Palop and Mucke , 2009; Verret et al . , 2012 ) as well as for schizophrenia and other psychiatric disorders ( Olypher et al . , 2006; Phillips and Silverstein , 2003; Uhlhaas and Singer , 2006 ) , further and detailed mechanistic investigations are granted . Five male Long-Evans rats , with a weight of 250–300 g . were trained in different behavioral tasks , with a multichannel electrode recording the electrophysiological activity in the hippocampus ( data are available at http://dx . doi . org/10 . 20350/digitalCSIC/12537 ) . The sample size was selected based on previous reports with analysis of hippocampal theta and/or gamma in a T-Maze task ( Montgomery and Buzsáki , 2007; Schomburg et al . , 2014; Tort et al . , 2008 ) . All of them were implanted with a 32 channels silicon probe ( Neuronexus Technologies , Michigan , USA ) connected in turn to a jumper consisting of two corresponding connectors joined by 5 cm of flexible cable . An Ag/AgCl wire ( World Precision Instruments , Florida , USA ) electrode was placed in contact with the skin on the sides of the surgery area , and used as ground . The data were acquired at 5 kHz , with an analog high-pass filter at 0 . 5 Hz . After digitalization , we initially low-pass filtered them at 300 Hz , removed the net noise with Notch filters at 50 Hz and 100 Hz and down-sampled the signals at 2 . 5 kHz . We adjusted the final position of both electrodes using as a reference the typical evoked potentials at the dentate gyrus ( Andersen et al . , 1966 ) , so that a maximal population spike in the dentate gyrus was recorded . After the surgery , the rats were left for at least 10 days until they recovered completely . During the first 72 hr , they were injected subcutaneously with analgesic twice per day ( Buprenorphine , dose 2–5 μg/kg , RB Pharmaceutical Ltd . , Berkshire , UK ) . During 1 week , they had as well antibiotic dissolved in the water ( Enrofloxacin , dose 10 mg/kg , Syva , León , Spain ) . The behavioral tasks were not started until the animals showed no signs of discomfort with the manipulation of the implants . All subjects were trained before the surgery following the next protocol . The first three days consisted on a habituation process with two 10 min sessions per day in an open field , with freedom of movement . The environment was a methacrylate sandbox of 50 × 50 cm , opened at the top and with three visual cues in three of the walls . After that , they carried out two new sessions per day for 8 days , first repeating the habituation and then performing a modified T-maze task , that has been described previously ( Wood et al . , 2000 ) . It consisted in several tracks in 8-like shape ( 132/102/80 cm long/wide/high with track wide 8 cm , Figure 6e ) . The starting point was located at the beginning of the central rail ( Figure 6e , start ) and the rat was forced to run across that arm ( Figure 6e , center ) , blocking other pathways with black panels . At the end of the track , it must choose one of the two directions of the T-junction and a small drop of water was delivered at the corner ( Figure 6e , reward ) in successfully trials . Each repetition is considered successful if the rat chooses the opposite direction with respect to the previous trial , finding always a reward at the corner after the T-junction . Then , another panel located after the water prevented the rat from retracing its route , forcing it to go to the starting point across the corresponding side arm ( Figure 6e , side ) , for a new trial . Each session had a duration of 20 min with around 30 trials , and all the subjects reached a performance greater than 80% in the last session . Only correct trials were considered for further analysis . After the surgery and recovery , we repeated the same protocol for 8 days . For further electrophysiological analysis , we considered only those sessions were the subjects kept a high level of performance ( 80% ) without any interference . There were in total between 2 and 5 sessions of 4 subjects . During the 9th day , we carried out a ‘novelty’ test . The rats were exposed to a novelty by introducing them in a ‘novelty chamber’ located inside the familiar open field; such chamber was a transparent methacrylate box with a square base 35 cm wide , and 40 cm high , opened at the top , with sand paper on the floor to provide a noticeable tactile stimulus . After this time , the novelty chamber was removed , and the animals were left another 10 min in the open field , considering this session as the control condition for the analysis . Except for the results in Figure 6 , all the analyses were carried out during the control session ( last session ) , with freedom of movement in a well-known environment . Four male Long-Evans transgenic rats , expressing Cre recombinase under the rat parvalbumin promoter ( LE-TG[Pvalb-iCre]2Ottc , NIDA , USA ) , were bred in our facilities , housed in pairs with food and water available ad libitum and maintained on a 12/12 hr light-dark cycle . For the surgery , rats were anesthetized with isoflurane ( 4 . 5% induction , 1–2% maintenance , in 0 . 8 l/min O2 ) and locally anesthetized by subcutaneous injection of bupivacaine ( 0 . 2 ml ) . All rats weighted 300–330 gr at the time of the first surgery . The Cre-dependent viral vector AAV1-EF1a-DIO-hChR2 ( H134R ) -eYFP-WPRE-hGH ( Penn Vector Core ) was bilaterally injected in dorsal CA3 ( AP −3 . 5 mm , LM ± 3 . 6 mm from bregma and DV –2 . 8 mm from the brain surface ) using a Hamilton syringe attached to an infusion pump ( 1 µl per hemisphere at 1 µl/min ) . Thus , ChR2 is specifically expressed in PV+ cells ( PV-ChR2 ) . Two weeks after the virus injection , rats underwent a second surgery for two fiber-optic cannulas and one recording electrode implantation . First , five screws were attached to the skull to strengthen the fixation of the implant . As in the previous group of rats , a 32 channels silicon probe connected to a jumper ( Neuronexus Technologies , Michigan , USA ) was placed in the left hippocampus covering dorsal CA1 and DG . Reference wires were attached to one of the screws . The coordinates for the electrode implantation were AP −3 . 5 mm , LM ± 2 . 5 mm from bregma and DV –3 . 0 mm from the brain surface , although its final position was adjusted based on the electrophysiological potentials evoked by stimulating the perforant pathway . Then , a fiber-optic cannula ( 200 µm diameter , 0 . 66 NA , 10 mm length; Doric Lenses , Quebec , Canada ) was placed in the dorsal CA3 of both hemispheres with an angle of 20° in the coronal plane at the coordinates AP −3 . 5 mm , LM ± 5 . 2 mm from bregma and DV –3 . 2 mm from the brain surface ( Figure 2a ) . Once the electrode and the two fibers were positioned , the stimulation electrode was removed and several layers of dental cement ( SuperBond or Palacos ) were applied to ensure enough fixation of all the components . The post-operative care was the same as in the previous group of rats ( see above ) . One of the subjects was excluded from the experiment after the surgery and prior to any analysis due to the reduced quality of its electrophysiological recordings . All behavioral procedures were conducted during the dark cycle . After the complete recovery following the implantation surgery and before starting the experiments , we handled the animals during 5 days in order to habituate them to the experimenter as well as to the manipulation of the implant ( connection of the headstage and the fiber-optic patch cords ) . Rats performed an 8 min daily session during 5 consecutive days in an already known open field ( 50 × 50×40 cm box of black methacrylate ) with bedding covering the floor . Animals were allowed to freely explore the arena in each session while receiving ON/OFF periods of light stimulation . The light for excitation of ChR2 was delivered at 50 mW/mm2 by a blue LED source ( Prizmatix , Canada ) at a wavelength of 460 nm . Stimulation protocol consisted on 5 s 40 Hz trains with 1 ms light pulses each 30 s during the entire session . For the bilateral stimulation , we used a branching fiber-optic patch cord ( 500 µm diameter , 0 . 63 NA; Doric Lenses ) connected to a rotatory joint ( Prizmatix ) which in turn connects to the LED source by a fiber-optic patch cord ( 1 mm diameter , 0 . 63 NA; Doric Lenses ) . The power density of the delivered light was measured prior to each session using a powermeter ( Thorlabs ) to ensure the same power density in all sessions ( 50 mW/mm2 ) . Light pulses were triggered by a stimulus generator ( STG2004 , Multichannel Systems , Reutlingen , Germany ) controlled by MC_Stimulus software ( Multichannel Systems ) . Electrophysiological data were recorded at 5 kHz sampling rate with an open-source acquisition system ( Open Ephys ) and synchronized with light stimulation and video recording by using an I/O board ( Open Ephys ) . After the performance of the experiments , the rats were perfused intracardially with PFA 4% . Brains were kept in post-fixation for 3 hr at RT and then stored in PBS at 4°C o/n . Then , brains were cut in 50µm-slices to corroborate the viral infection as well as the correct position of the recording electrode and the fiber-optic cannulas . Slices were incubated with monoclonal PV antibody developed in mouse ( 1:2000 , Swant , Switzerland ) and afterwards with an anti-mouse secondary antibody developed in goat ( 1:500 , Alexa Fluor 594 dye , Life Technologies , USA ) . After completion of histological treatments , brain sections were imaged using a fluorescence microscope ( DM4000B , Leica ) coupled to a Neurolucida software ( MicroBrightField , Inc ) and images were processed with Image J software . The animals were monitored during all the tasks , and their behavior was recorded using a standard camera located at the top of the room . Using those videos , the location of the subjects was tracked with the software tracker ( physlets; https://www . physlets . org/tracker/ ) , taking their centroid as the refence point . The synchronization of the video and the electrophysiological recordings was made triggering a red LED and matching the temporal mark that it left in the recordings with the first frame with light . The first approach to achieve the information of the sources contributing to the LFPs was the use of CSD analysis ( Freeman and Nicholson , 1975; Herreras , 1990; Holsheimer , 1987; Mitzdorf , 1985 ) . It measures the transmembrane currents , providing a spatiotemporal distribution of the local sinks and sources ( inward and outward currents , respectively ) . Contrary to the LFPs , these currents represent spatially localized phenomena , increasing the spatial resolution . The membrane currents can be achieved following the Laplace equation and using the measured field potentials and the conductivity of the medium . As the hippocampus is a layered structure , the one-dimensional approach in the direction parallel to the recording electrode was used: ( 1 ) CSDmt=-σh2um-1t-2umt+um+1t , where umt is the LFP recorded at the m-th site , h is the distance between channels and σ is the conductivity of the extracellular space . We assumed the whole structure as an isotropic and homogeneous medium . Though hippocampal strata present different resistivities , they do not affect much to the temporal dynamics of specific locations ( Herreras , 1990; Holsheimer , 1987; López-Aguado et al . , 2001 ) . Therefore , the distance and conductivity are constants , and they merely act as a scale factor . In this work , the distance between contacts was h = 100 μm and we assumed a constant conductivity σ = 350 Ω−1cm−1 ( López-Aguado et al . , 2001 ) . Though the CSD presents higher spatial resolution than the LFPs , it does not discriminate contributions from different pathways . Multiple membrane currents with different origins may overlap spatiotemporally , and local currents are also affected by the activity in nearby domains ( Herreras , 2016; Korovaichuk et al . , 2010; Martín-Vázquez et al . , 2013 ) . To overcome these limitations , we applied an independent component analysis ( ICA ) . Methods such as CSD analysis make it possible to isolate the local transmembrane currents by eliminating propagated field potentials ( see above ) . Nevertheless , different pathways are contributing to these currents , and their activities may overlap spatiotemporally . To disentangle the specific sources that generate the LFPs , we applied an ICA . The effectiveness of this approach has been well studied and established in the hippocampus ( Fernández-Ruiz et al . , 2012b; Herreras et al . , 2015; Korovaichuk et al . , 2010; Makarov et al . , 2010; Makarova et al . , 2011; Schomburg et al . , 2014 ) . It aims to solve the problem of separating N statistically independent sources that have been mixed in M output channels . To do that , it performs a blind separation of patterns , because the different distributions of the sources are unknown . Moreover , it assumes spatial immobility of the sources or , in other terms , a fixed location of the axon terminals . The contribution of their synaptic currents to the LFP conforms the different independent components ( ICs ) or generators to unravel . Each recorded time-series um ( t ) is modeled as the sum of N neuronal sources multiplied by a constant factor: ( 2 ) umt=∑n=1NVmnsnt , m=1 , 2 , … , M , where Vmn is the mixing matrix with the voltage loadings of N LFP generators on M electrodes and sn ( t ) is the time-series associated to the n-th LFP generator . As the number of ICs with significant variance is usually low ( 4–7 out of 32 ) ( Benito et al . , 2014; Korovaichuk et al . , 2010 ) , we applied a dimension reduction of the loading matrix by prior use of principal component analysis , keeping 99% of the original LFP variance . For each structure and electrographic state , the number of optimal components is determined by stepwise increase of the number of principal components until the new ICs are only noise ( Makarov et al . , 2010 ) . Since noisy components contribute negligible variance ( in absence of artefacts in the signal ) we always choose this number plus two . There are several algorithms to compute the mixing matrix that transform LFP data into ICs , nevertheless , all of them share a common theory framework . In this work , we have used the information-maximization approach RUNICA ( Bell and Sejnowski , 1995 ) , implemented in the matlab toolbox ‘ICAofLFPs’ ( http://www . mat . ucm . es/~vmakarov/downloads . php ) . For comparison purposes , the kernel density ICA algorithm KDICA ( Chen , 2006 ) was also computed , obtaining similar results . By definition , the ICA may extract as many generators as the number of LFP signals . To correctly identify the presynaptic specificity of an IC , several conditions must be taken into account . First , each IC contributes differently to the total variance ( power ) of the LFP . Only those with a significant contribution ( >1% in this work ) were considered for further analysis . Second , the anatomic structure of each generator is fixed in each subject ( Castellanos and Makarov , 2006; Korovaichuk et al . , 2010; Makarov et al . , 2010 ) . In other words , the spatial profile of each ICs must be stable along the time . This was assessed by applying the ICA in different short-term epochs ( Korovaichuk et al . , 2010 ) . Only those components present in all conditions with a stable spatial loading may represent true current generators . Moreover , a certain degree of similarity is expected between subjects and a specific pathway should have a comparable profile for different animals . Third , the synaptic specificity of each generator was determined by stimulating their respective excitatory pathways with subthreshold evoked activity ( Makarova et al . , 2011 ) . Fourth , not every synaptic input leaves a footprint in the LFP . The geometry of the region and the distribution of axons and dendrites determine the real contribution of each pathway to the field potential ( Buzsáki et al . , 2012; Herreras , 2016; Herreras et al . , 2015 ) . This requires specific realistic models to test the multiple origins of the measured currents . Applying ICA in our data recordings and considering all the conditions mentioned above , we were able to extract three common and stable generators in all subjects ( Figure 1 and Figure 1—figure supplement 1 ) . They correspond to pathway-specific inputs to the hippocampus . Two of them were in CA1: one in str . radiatum , which corresponded to the synaptic terminals of Schaffer collaterals from CA3 to the pyramidal cells in CA1 ( Benito et al . , 2014; Fernández-Ruiz et al . , 2012a; Korovaichuk et al . , 2010; Makarova et al . , 2011; Martín-Vázquez et al . , 2016; Schomburg et al . , 2014 ) ( Sch-IC ) ; the other component had a current sink in str . lacunosum-moleculare ( lm-IC ) , where are located the inputs from EC3 to the pyramidal cells in CA1 ( Benito et al . , 2014; Martín-Vázquez et al . , 2016; Schomburg et al . , 2014 ) . A third component was identified in the DG , which corresponded to the axons projected from the EC layer II ( EC2 ) to the dendrites of the granular cells through the perforant-pathway ( Benito et al . , 2014; Korovaichuk et al . , 2010; Makarova et al . , 2011 ) ( PP-IC ) . Note that the active synaptic domain of PP-IC was in the molecular layer of the DG , but its field potential was dominant in the hilar region ( Figure 1 ) . This was generated by the volume conduction of the cell membranes into the molecular layer . The field potentials of common currents , above and below the hilus , are overlapped in this region , thus increasing their electric field ( Benito et al . , 2014; Herreras , 2016; Herreras et al . , 2015 ) . The extracted ICs represent the current sources of specific pathways to the hippocampus . Therefore , the temporal dynamics and rhythms of each generator reflects the activity generated in different nearby regions ( CA3 , EC3 and EC2 for Sch-IC , lm-IC and PP-IC , respectively ) . One limitation of this approach is that it cannot separate distinct temporal patterns within the same origin ( i . e . theta and gamma oscillations ) . This is the consequence of two main effects . First , the same neuron could fire in multiple modes ( Vinogradova , 2001 ) . Moreover , the currents generated by synaptic terminals from the same region are fully overlapped in the space and their combination made up a single generator . It should be noted at this point that ICA requires independence in space and time . Therefore , two spatially separated sources with exactly the same temporal dynamics would converge in a single component . Nevertheless , small differences in the signals’ co-variation ( i . e . temporal jitter and/or imperfect amplitude co-variation ) would allow a correct separation of the two sources , even if there is a high coherence between them ( Makarova et al . , 2011 ) . Another consideration is that the strongest generators could introduce contamination in other components ( Korovaichuk et al . , 2010; Schomburg et al . , 2014 ) . To ensure that the huge theta power is not affecting the discrimination of ICs , we separated the LFP in slow and fast rhythms by filtering the raw LFPs at 30 Hz ( low-pass at <30 Hz and a high-pass at >30 Hz , respectively ) . The ICA was applied to each filtered data-set separately . We compared the resultant ICs with those using the unfiltered data , confirming that the same generators were found in all conditions with quite similar time-series ( Figure 1—figure supplement 1 ) . ICA does not ensure the correct polarity and amplitude of each generator . However , as the ICA algorithm is invertible , the LFPs generated by each component can be retrieved separately . The CSD can be applied to these reconstructed signals , obtaining the sinks and sources of each specific pathway . Such partial signals do have the correct polarity and amplitude ( Korovaichuk et al . , 2010; Martín-Vázquez et al . , 2013 ) . After computing the ICA algorithm , the dataset corresponding to each subject was composed by three time-series . These signals were downsampled at 625 Hz to improve the speed of computational analysis . They were also normalized , imposing to each dataset an averaged mean value of zero and a standard deviation of one to each signal separately . This way , we increase the similarities inter-subject and facilitate their comparison . Power spectra were estimated using the multitaper method ( Thomson , 1982 ) . For power analysis at specific frequency bands we used an approach based on filtering and Hilbert transform ( Jackson et al . , 2006; Ólafsdóttir et al . , 2017 ) . First , the signal is bandpass filtered with a FIR filter between the frequencies of interest . Then , we computed the Hilbert transform and the instantaneous power was estimated as the squared complex modulus of the signal at each time point . The mean value was obtained as the averaged power in a certain time window . We defined the following frequency bands which are used along the text ( unless otherwise indicated ) : delta ( 1–4 Hz ) , theta ( 6–10 Hz ) , slow gamma ( 30–60 Hz ) , medium gamma ( 60–100 Hz ) and fast gamma ( 100–150 Hz ) . The linear interaction between IC-LFPs at each specific frequency was assessed using a coherence analysis . It measures the ratio between the cross power spectral density and their individual power spectral densities and was computed using the mscohere . m function in Matlab . The statistical significance was determined by a surrogate analysis ( 1000 surrogates in this work ) . With this methodology , the temporal relationship between signals was broken by randomly displacing one signal respect to the other . Then , the coherence surrogated results at each frequency were approximated to a Gaussian distribution and the significance threshold was the value for which the previous cumulative distribution was 0 . 95 ( p=0 . 05 ) . To evaluate the distribution of the gamma activity along the phase of theta , the signals were first filtered at the frequencies of interest ( gamma and theta ) and the amplitude and phase were extracted using the Hilbert transform . Then , for each theta cycle , the envelope of the gamma activity was divided into N equidistant bins; an average along all cycles was then taken . Similar to the coherence analysis , the statistical significance was assessed by a surrogate analysis ( 1000 surrogates ) , randomly shifting the gamma signal with respect to the theta phase . The surrogate distribution was estimated by averaging the results of all simulations . From the whole recordings , only those epochs with a real theta rhythm in all components where considered for further analysis , that is , with high power at that band . Moreover , as oscillations with low amplitude could result in a less accurate estimation of their phase , we selected only those cycles with a minimum value of theta power to avoid this issue . To find such threshold we have modeled theta rhythm data as the combination of theta oscillations ( Xθt ) and pink noise ( Xnt ) to simulate the neural noise of the recordings: ( 3 ) Xθnt=Xnt+Xθt The noise was computed using the function pinknoise from Matlab , while Xθt was composed by d segments or cycles defined as: ( 4 ) Sik ( ti ) =A ( sin⁡ ( 2πfiti+1 . 5π ) ) Where k=1 , 2 , … , d and fiti where selected in order to that cycle had a duration of Ti∈0 . 1 , 0 . 145 seconds , randomly chosen from a normal distribution with mean 0 . 125 and standard deviation 0 . 02 . Knowing each Sik ( ti ) , we could estimate perfectly the phase of the theta rhythm , that is this was the ground truth . Briefly , for each simulation we changed the relative power between both components by varying the A parameter and we estimated the phase of the theta oscillation ( see below ) . Doing so , we were able to measure the error between our estimation and the ground truth as a function of the relative theta power and then find the value that minimizes that error . A detailed description of these steps follows . For each dataset , we bandpass filtered each signal at delta ( 1-4 Hz ) and theta ( 6-10 Hz ) frequencies . Then , we computed the Hilbert transform and the instantaneous power was estimated as the squared complex modulus of the signal at each time point . The relative theta power was obtained as the ratio between the averaged theta by the delta power ( Ólafsdóttir et al . , 2017; Jackson et al . , 2006 ) . The phase of the oscillation was estimated for the theta filtered signal through Hilbert , being zero and π radians the values corresponding to the trough and the peak of the cycle , respectively . Finally , the error was measured as the averaged distance ( in milliseconds ) between each trough of the real phase and the estimated one . Additionally , we computed the minimum error as that obtained when Xnt is set to zero . This value corresponds to the noise introduced by the method used to estimate the phase and does not depend on the power . The relationship between the ratio and the error is shown in Figure 1—figure supplement 3 . Using the simulated data , we considered that the theta power was not influencing the estimation of the theta phase when the error due to the amplitude ( i . e . not considering the one introduced by the filtering ) was lower than 1 ms . This corresponded to a ratio of theta power 3 . 78 times ( we took four for simplicity ) higher than delta ( Figure 1—figure supplement 3 ) . In the real recordings we expect not only neural noise but also activity at the delta band . Thus , the theoretical threshold obtained by this procedure represent a conservative measurement . For those cases with high delta activity , the threshold would be more restrictive as the ratio decreases , but the theta power would be always high enough to guarantee a correct estimation of the phase . To detect the theta rhythm in all the ICs recorded , we used a sliding window of one theta cycle ( approximately 125 ms ) and selected only those epochs where the ratio between their theta power by delta ( computed as described above ) was higher than 4 . Moreover , analysis taken thresholds of 6 and 8 were also done for comparison purposes , showing no significant differences . In the Cre-expressing transgenic subjects we analyzed whether the light stimulation of CA3 PV-interneurons had an effect exclusively in the Schaffer collateral theta output . We compared time windows of two seconds immediately after the stimulation with the first two seconds of the stimulus . This period was chosen to minimize the influence of different locomotor activities between windows . For all trials , only those with theta oscillation ( see above ) were further considered . Then , we computed the power spectrum of each IC-LFP using a multitaper approach ( Thomson , 1982 ) . To estimate the relative theta phase between ICs , we used a modified inter-trial phase clustering approach ( Cohen , 2014 ) to account for differences between cycles instead of trials ( ICPC ) . In this methodology , each trial is defined as a vector with modulus one and the angle corresponding to the phase of the oscillation measured at a specific time point ( φt ) . Then , the ICPC is computed as the modulus of the averaged vector among all trials: ( 5 ) ICPC=1N∑t=1Neiφt If the distribution of angles is uniform along the polar axis , then the ICPC value is zero . On the contrary , values near one indicate a preferred phase in the distribution , being one when all trials have the same phase . For the analysis in Figure 1h , the phases of each signal were extracted following Cole and Voytek , 2019 to have a better characterization of the shape of the theta rhythm . The LFP measured in the pyramidal layer of CA1 ( pyr . CA1 ) was considered as reference , where its trough and peak coincide with 0 and π radians , respectively . We calculated the ICPC for the different components separately , where each trial was the phase of the IC measured at each trough of pyr . CA1 . Thus , the number of trials corresponded to the number of theta cycles in pyr . CA1 and the angle and value of the ICPC can be interpreted as the phase difference and the stability of the IC with respect to pyr . CA1 , respectively . The statistical significance was assessed by a surrogate analysis ( 1000 surrogates in this work ) , randomly shifting the phase of the IC and keeping pyr . CA1 constant . For each simulated dataset , the ICPC was computed fitting all the surrogated results into a normal distribution . Then , the p-value associated to the ICPC of the IC was obtained as one minus the previous normal cumulative distribution evaluated at the ICPC value . Using the ICPC approach , the degree of coherence between theta rhythms was estimated for each cycle separately . Doing this process , a dynamic measurement of synchronization can be done identifying time epochs of high and low coherence . Considering two ICs , one acting as the reference , the ICPC value of each theta cycle was computed using only three cycles , which corresponded to the relative phase at that cycle and the previous and consecutive ones ( Figure 2—figure supplement 1 ) . If the waves were highly coherent , then their phase would be similar along theta cycles , resulting in an ICPC value close to 1; for those with different phases , the ICPC would be lower . In this work , we considered the theta oscillation in lm-IC as the reference to compute the ICPC , as it had the highest amplitude at that frequency . Moreover , the ICPC between pairs of signals ( i . e . lm-IC vs . Sch-IC and lm-IC vs . PP-IC ) was averaged as an approximation of the global synchronization of the network at each time epoch . The instantaneous ICPC can be compared to other metrics as frequency or power , analyzing the correlation between synchronized state and the features of the signals . To compute each correlation , the data were classified into 10 groups as a function of the ICPC , with ten equidistant bins from 0 . 75 to 1 . We chose these values as they contain at least the 90% of the cycles in all subjects and 80 cycles per bin ( around 10 s ) . Then , the averaged value of the different metrics was computed for each group , analyzing the correlation between the means and the ICPC value . To identify reliably relationships , we compared if the resultant correlation value ( ρ ) was higher than the obtained by a surrogate analysis . Each simulated dataset ( 100 surrogates in this work ) was built by randomly shifting the IC components , breaking any temporal relationship between them . Then , the correlation between the ICPC and other features of interest were computed , fitting the results to a gaussian distribution . We considered that a correlation was significant if its value was higher than the 95th percentile of the surrogatedistribution . The joint analysis of several features as predictors of the ICPC was done using a multiple linear regression . Firstly , we ranked all the values associated to each feature ( function tiedrank . m in matlab ) to minimize the effect of outlayers in the dataset . Then , a single model is fitted using each predictor multiplied by a beta factor: ( 6 ) ICPC=β0+β1power+β2frequency+β3speed+…+ε Where β0 is a constant value and ε are the residuals . The contribution of one specific predictor to the total explained variance of the model is estimated by fitting a reduced multiple linear regression without that predictor and computing the difference between variances in the full model minus the reduced one . To estimate if each variable is significantly contributing to the ICPC across subjects , we followed Montgomery and Buzsáki , 2007 and tested their associated beta values for a statistical difference from zero ( t-test , Bonferroni corrected ) . Interactions between the phase of a low frequency oscillation and the amplitude of a faster one were measured using an approach based on the modulation index ( MI ) ( Canolty et al . , 2006; Tort et al . , 2008 ) . The original method computed the phases and amplitudes through filtering and Hilbert transform . Nevertheless , theta rhythms in the hippocampus are non-sinusoidal oscillations and filtering the signal results in errors at estimating the waveform shape and could introduce spurious coupling ( Cole and Voytek , 2017; Kramer et al . , 2008 ) . A new methodology has been proposed to overcome this issue and estimate the instantaneous phase of this kind of oscillations ( Cole and Voytek , 2019 ) . Briefly , we used a combination of a narrowband filter to detect the zero-crossing points ( which correspond to the ascendant and descendant slope of the oscillation , or the phases π/2 and 3π/3 , respectively ) and a broadband filter to find the trough and the peak ( phases 0 and π , respectively ) . This way , the phase of the signal does not vary monotonically ( as in the case of a sinusoidal one ) but could follow fast changes in the cycle as an abrupt ascendant slope and a soft descent . We named this signal as xθφ ( t ) . The amplitude of the faster oscillation is computed as the envelope of the signal filtered at the specific frequency which we want to analyze . First , we used a filter centered at that frequency and with a bandwidth at least two times the frequency of the phase signal where the coupling is expected to be maximum ( Juhan et al . , 2015 ) . Considering that the main rhythm in the hippocampus is around 8 Hz , the bandwidth should have at least 16 Hz ( we used 20 Hz in this work ) . Then , the envelope is computed using the Hilbert transform . The resultant signal is labelled as xγAt . To compute the MI , we divide each cycle of xθφ ( t ) into N bins . To avoid the issues introduced in this method due to the previously mentioned non-uniform theta oscillation ( van Driel et al . , 2015; Cole and Voytek , 2017 ) , these bins were not of the same size , but were equalized along the phase of the theta cycle . Instead of using N bins , we divided each cycle into 4 segments which correspond to the epochs between the peak , the trough and both ascendant and descendant pendants ( Cole and Voytek , 2019 ) , and then divided that segment into N/4 bins of the same size . Therefore , N/4 bins were used from the trough to the middle of the ascendant phase , N/4 bins from this point to the peak and so on . After that , we computed the mean amplitude of xγAt at each bin , calling <xγA>φ ( j ) the amplitude at the phase bin j . From them , we can calculate the entropy H , defined by: ( 7 ) H=−∑j=1Npjlog⁡pjwhere N was set to 20 , and pj is given by ( 8 ) pj=<xyA>ϕ ( j ) ∑j=1N<xyA>ϕ ( j ) The value of MI is defined as the entropy H normalized by the maximal entropy ( Hmax ) , given by the uniform distribution pj=1/N ( i . e . Hmax=log⁡N ) : ( 9 ) MI= Hmax-HHmax A value of MI near 0 indicates lack of phase-to-amplitude modulation , while larger MI values reflect higher coupling between both signals . The values obtained using the MI given by Equation 9 were compared to those provided by the phase-amplitude coupling index proposed by Canolty and colleagues ( mean vector length; Canolty et al . , 2006 ) . The latter is estimated as the degree of asymmetry of the probability density function of the gamma amplitude across the phase of theta . Both methods yielded similar results . The statistical significance has been assessed following the steps proposed by Canolty and colleagues ( Canolty et al . , 2006 ) , by a surrogate analysis ( n = 100 surrogates ) in which each surrogate is built by cutting the phase signal at a random point and exchanging the resultant segments . This breaks the temporal relationship between both time-series minimizing the distortion of their dynamics ( Juhan et al . , 2015 ) . The MI estimated among surrogates represents the coupling due to the oscillatory nature of the signals but not by a real temporal relationship . Therefore , the MI values of all surrogates are approximated to a gaussian distribution , whose 95th percentile is considered as a significance threshold . The MI is a measurement of the degree of interaction between the phase and the amplitude of two frequencies , but it has no information regarding the directionality in this coupling . On the one hand , the theta phase would modulate the amplitude at gamma frequencies while , on the other hand , the gamma activity could be leading the phase . To identify leader and follower in this interaction , we have used the CFD index ( Jiang et al . , 2015 ) . The main idea is that , supposing that the phase component precedes the amplitude with a fixed time delay of k ms , the distance from the peak of the phase to the next peak of the amplitude should be k ms ( i . e . there is an increase of gamma activity k ms after the peak of theta phase ) . Thus , as not all phase cycles have the same duration , the distance from the peak of the gamma amplitude to the next peak of theta may vary . Nevertheless , in the case of gamma activity preceding theta , the result would be the opposite . The peak of gamma activity should appear k ms before the peak of the phase , while the timing from phase to amplitude would be different for each cycle . Note that the CFD is an estimation of the temporal relations between signals , but not a measurement of causality per se as , for example , a third source could be interacting with both signals . The CFD identifies this relationship using the phase-slope index ( PSI; Nolte et al . , 2008 ) , a measurement of directionality between time series . Briefly , if the oscillation of one signal x ( t ) at a certain frequency is driving a second one y ( t ) with a time delay , then the phase difference between them at that specific delay will change consistently with the frequency of the signals . The slope of the phase is obtained in function of the frequency , and its sign will indicate who is the driver . If the slope is positive ( higher the frequency , higher the phase difference between signals ) then x ( t ) leads y ( t ) in time , while negative values would indicate that y ( t ) precedes x ( t ) . The CFD is a variant of the PSI , where one signal is the theta component , and the other the envelope of the gamma activity . Calling x ( t ) to the original signal , xγAv ( t ) to the power envelope of the signal at a v gamma frequency , and being X and XγAv their Fourier transform , respectively , the CFD is defined as the PSI by: ( 10 ) ψv , fj=Im∑fj-β2fj+β2Cv , fjCv , fj+∆fwhere ( 11 ) Cv , fj=∑s=1SXs ( XγAv , s ) ∑s=1SXs2∑s=1SXγAv , s2is the complex coherency , Im is the imaginary part , ‘*’ denotes the complex conjugate , fj is the theta frequency under study , S is the number of segments in which the signal has been divided and β is the bandwidth for which the phase slope is measured , and it has been fixed at 2 Hz , 4 times the resolution ( ∆f=0 . 5 Hz ) . This methodology has been proposed specifically for those frameworks with phase-amplitude coupling as more classical approaches like Granger Causality ( Granger , 1969 ) have some unavoidable limitations due to the use of filters ( Barnett and Seth , 2011 ) and the differences of signal-to-noise ratio in both components ( Nolte et al . , 2010 ) , which provoke that they cannot identify the correct directionality in CFC models ( Jiang et al . , 2015 ) . To provide statistical significance , a new surrogate test ( n = 100 surrogates ) has been developed following the same steps than in the MI analysis . Note that in this case , two thresholds are imposed , considering both tails of the gaussian distribution ( i . e . positive and negative values ) . In this work , we have used the Matlab toolbox HERMES ( Niso et al . , 2013 ) and the implementation of PSI in Matlab code ( http://doc . ml . tu-berlin . de/causality/ ) . All the Matlab code developed to compute cross-frequency analysis is freely available at https://canalslab . com/ . To emphasize the directionality in the region of higher CFC , the MI comodulogram has been redefined as a mask , with values from 0 to 1 ( minimum and maximum value MI ) . Applying this mask to the CFD comodulogram , areas without phase-amplitude coupling are attenuated , while the main cluster remains constant .	In the brain , a vast number of neurons coordinate their activity to support complex cognitive processes . One of the best places to see this in action is the hippocampus , a brain structure with a key role in memory and navigation . The hippocampus shows waves of electrical activity , which represent the synchronized firing of large numbers of neurons . The hippocampus can generate multiple rhythms at once . The two main rhythms are theta and gamma . Theta waves are slow , with a frequency of about 8 Hertz . Gamma waves are faster with a frequency of up to 120 Hertz or even more . Theta waves are always present in the brains of freely moving animals , whereas gamma waves occur in brief bursts . These bursts usually correspond to a particular point on the theta wave . One burst may occur just before each peak of the theta wave , for example , whereas another burst may occur just after the peak . This separation enables individual bursts of gamma to carry different messages without them becoming mixed up . This is similar to how radio stations broadcast their signals at different carrier frequencies to avoid interference . By recording hippocampal activity in rats exploring a maze , Lopez-Madrona et al . now show that the hippocampus has not one , but three generators of theta waves . Having three sources of theta , each of which can be synchronized with gamma , provides a more versatile system for encoding and sending information . It also means that the three theta generators can vary the degree to which they coordinate their firing . This helps the brain combine or separate streams of information as required . By working together to create a single theta rhythm , for example , the three theta generators can help animals combine information stored in memory with incoming sensory input . How the coordination of theta rhythms in the hippocampus influences the activity of other brain regions involved in learning and memory remains unclear . However , uncoupling of theta and gamma waves seems to be an early sign of Alzheimer’s disease and can also be seen in the brains of people with schizophrenia and other psychiatric disorders . Understanding how this process occurs could provide clues to the origin of these disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Different theta frameworks coexist in the rat hippocampus and are coordinated during memory-guided and novelty tasks
Gastrulation generates three layers of cells ( ectoderm , mesoderm , endoderm ) from a single sheet , while large scale cell movements occur across the entire embryo . In amniote ( reptiles , birds , mammals ) embryos , the deep layers arise by epithelial-to-mesenchymal transition ( EMT ) at a morphologically stable midline structure , the primitive streak ( PS ) . We know very little about how these events are controlled or how the PS is maintained despite its continuously changing cellular composition . Using the chick , we show that isolated EMT events and ingression of individual cells start well before gastrulation . A Nodal-dependent ‘community effect’ then concentrates and amplifies EMT by positive feedback to form the PS as a zone of massive cell ingression . Computer simulations show that a combination of local cell interactions ( EMT and cell intercalation ) is sufficient to explain PS formation and the associated complex movements globally across a large epithelial sheet , without the need to invoke long-range signalling . Before gastrulation , the embryo of reptiles , birds and most mammals is a large flat disc of epithelial cells ( epiblast ) ( Pasteels , 1940 ) . In the chick , the 50 , 000 or so cells that comprise the embryonic epiblast ( area pellucida , 3–5 mm in diameter ) move as two bilaterally symmetrical whorls , known as the ‘Polonaise’ pattern ( Gräper , 1929; Wetzel , 1929 ) ( Figure 1 , stage EGK XI-XIV ) . The movements continue for 8–10 hr , culminating in the formation of a stable morphological structure in the posterior midline , the primitive streak ( PS ) ( Figure 1 , stage HH2 ) . Stage HH2 is very brief , as the PS then quickly narrows and elongates along the midline of the embryo , reaching about 2/3 of the diameter of the area pellucida in a further 8–10 hr ( Figure 1 , stages HH3 and 3+ ) . Once the PS forms , cells in the epiblast lateral to the PS start moving directly into it along trajectories perpendicular to its axis ( for reviews see Spratt , 1946; Nicolet , 1971; Bellairs , 1986; Stern , 2004b ) ( Figure 1 , stages HH3-3+ ) . The PS acts as a gateway for gastrulation as epiblast cells internalize via epithelial-to-mesenchymal transition ( EMT ) ( Nieto , 2011 ) and generate mesoderm and endoderm beneath the epiblast layer . At present we do not understand the cellular or molecular mechanisms of any of these events , nor do we know whether they are controlled separately or represent the manifestation of a single underlying process . 10 . 7554/eLife . 01817 . 003Figure 1 . Diagrams depicting the early stages of chick development . The upper row of diagrams shows embryos at stages XI-XIV ( pre-primitive streak ) , 2 ( early streak ) , 3 ( mid-streak ) and 3+ ( mid- to late streak ) , viewed from the dorsal ( epiblast ) side . The arrows denote the main morphogenetic movements ( ‘Polonaise’ ) occurring within the plane of the epiblast . After stage 4 ( end of gastrulation ) , convergence of cells towards and ingression through the anterior part of the streak slows down or ceases ( although these movements continue through the middle and posterior parts of the streak ) , while the epiblast anterior to the streak ( prospective neural plate ) elongates ( Sheng et al . , 2003 ) ; later , the streak starts to regress , further lengthening the neural plate posteriorly ( Spratt , 1947 ) . The lower row of diagrams shows an exploded view of the embryos at each of the above stages , with the top row of diagrams representing the upper layer ( epiblast , shades of yellow ) , the bottom row showing the lower layer ( shades of blue/green: hypoblast in dark green , endoblast in light green , definitive or gut endoderm in blue ) and the centre row showing the middle ( mesodermal ) layer ( primitive streak , in red ) . Within the epiblast , the central ( yellow ) region is the area pellucida and the outer ( mustard ) region the extraembryonic , area opaca . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 003 Many theories have been proposed to account for the early movements of the epiblast ( Table 1 in Chuai and Weijer , 2009 ) . One type of model invokes long-range , diffusible chemotactic attractants or repellents emanating from various parts of the embryo , to which epiblast cells respond as individuals . For example , ( Vasiev et al . , 2010 ) suggested that the tip of the PS produces repellents for cells in the rest of the epiblast , while ( Sandersius et al . , 2011 ) proposed that the PS acts as a ‘chemotactic dipole’ , secreting repellents at the tip and attractors at the base , to which epiblast cells respond . Differential adhesion between cells destined to ingress and the rest of the epiblast is also invoked by some models ( Vasiev et al . , 2010 ) . All of these models are complicated by the fact that the extracellular matrix , presumed to be the substrate over which the epiblast moves , is secreted by both epiblast and underlying hypoblast and actually moves along with the cells ( Harrisson et al . , 1985a; Harrisson et al . , 1985b; Zamir et al . , 2006; Zamir et al . , 2008 ) . Some models do not envision the extracellular matrix as a substrate for cell movements . One of these ( Wei and Mikawa , 2000 ) focused on streak elongation , proposing that oriented cell division could drive this process . Another class of mechanism involves epithelial intercalation of epiblast cells at right angles to the future midline in the presumptive domain of the PS , which is initially located along the posterior edge of the epiblast: this could drive the elongation of this domain and may also contribute to the Polonaise movements ( Voiculescu et al . , 2007 ) . However none of these models is sufficient to account for all four major movements of chick gastrulation: the Polonaise of the early epiblast , the elongation of the PS , the movement of epiblast cells towards the streak and their ingression through the streak ( Table 1 ) . To date , only a very complex combination of various unrelated mechanisms , involving oriented cell division in the streak , secretion of signals by the streak that repel its tip ( ‘mechanism M3’ of Vasiev et al . ( 2010 ) ) , induced cell polarization of the epiblast and differential adhesion of the prospective mesendoderm to neighbouring cells ( ‘mechanism 11’ of Table 1 in Vasiev et al . ( 2010 ) ) , has come close to delivering the full repetoire of key movement patterns . 10 . 7554/eLife . 01817 . 004Table 1 . Summary of the four main classes of model ( with an example of each ) proposed to explain aspects of chick gastrulationDOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 004ModelMechanism ( s ) PS elongationEarly epiblast movements ( Polonaise ) Late epiblast movementsIngressionWei and Mikawa , 2000oriented cell division in PSxBodenstein and Stern , 2005movement and incorporation of lateral cells into PS +/− active movement within PSxVoiculescu , et al . , 2007intercalation in PS regionxxVasiev , et al . 2010repulsion by tip of PSxxSandersius , et al . , 2011repulsion and attraction by PSxxThe last four columns summarise the specific cell movements that are explained ( x ) or not explained ( blank ) by each model . None of the existing classes of model is sufficient by itself to account for all the movements observed . Several other problems are not addressed by any current model . One of these is how the PS is maintained as a morphologically stable structure despite the fact that cells are continuously moving into and out of it . We know very little about the dynamics of EMT for individual cells , how collective EMT arises and how the PS is maintained as a stable structure despite its constantly changing cellular composition . The PS acts like the blastopore ( a canal connecting the outer and inner layers of the embryo ) of lower vertebrates , but the PS does not have an obvious opening , raising the question of how cells are internalized through an apparently solid structure . It is also unclear how the epiblast preserves its integrity and characteristic columnar epithelial organisation of cells with apical-basal polarity during this process . Here we address these questions and provide evidence that the epiblast is highly dynamic and that local cell interactions are sufficient to explain global morphogenetic movements across a large epithelial sheet without the need for long-range signalling . Conventional time-lapse video microscopy reveals that the PS appears abruptly , forming a triangular structure within 10–30 min ( Figure 2A–E , Video 1 ) . This event defines the transition between stages XIV ( Eyal-Giladi and Kochav , 1976 ) and 2 ( Hamburger and Hamilton , 1951 ) ( Figure 1 ) . Scanning Electron Microscopy ( SEM ) of embryos at successive stages , fractured perpendicular to the forming streak , reveals a growing population of middle layer cells ( prospective mesoderm and endoderm ) underlying an uninterrupted flat sheet of epiblast ( Figure 2F–K ) . Close to the PS , the epiblast displays cells with various degrees of apical narrowing and baso-lateral expansion ( designated 1–5 in Figure 2L–P ) , indicative of bottle-like cells undergoing EMT . The PS only develops a marked midline groove many hours later , by which time it contains many deep cells and its length has extended to about 2/3 of the diameter of the area pellucida ( stage 3+ , Hamburger and Hamilton , 1951; Figure 1 ) ; even then , it does not contain a blastopore-like opening that could act as a portal for gastrulation ( Bancroft and Bellairs , 1974; Vakaet , 1984; Figure 2K ) . 10 . 7554/eLife . 01817 . 005Figure 2 . EMT in the formation of the primitive streak ( PS ) . ( A–E ) Images from a time-lapse sequence of entire embryos ( Video 1 ) , showing the uniform epiblast 6 hr ( A , stage EG&K XII ) and just before primitive streak formation ( B , stage EG&K XIV ) , the first appearance of the primitive streak ( C , stage HH2 ) , accumulation of mesoderm beneath the flat streak ( D , stage HH3 ) , appearance of a groove in the PS and emigration of mesoderm ( E , stage HH3+ ) . ( F–K ) SEM of fractured embryos before ( F–H ) and after ( I–K ) streak formation . White arrows indicate possible EMT before PS formation . ( L–P ) SEM of fractured PS , showing EMT cells with various degrees of apical constriction and basolateral expansion ( classified as ‘ingression stages 1–5’ ) . ( Q ) This embryo was cultured for 1 hr after electroporation of a control , fluorescent morpholino into the entire epiblast at stage XI , then sectioned sagitally and viewed under fluorescence . Labelled cells in the epiblast show similar morphologies to those in SEMs ( panels L–P , ‘ingression stages 1–5’ ) . ( R ) This embryo was cultured for 4 hr after electroporation of a control , fluorescent morpholino into the entire epiblast at stage XI , then fixed ( at stage XII ) , sectioned sagitally and stained with anti-fluorescein antibody ( brown ) . The section shows several cells that have left the epiblast and are now in the underlying space throughout the anterior-posterior extent of the embryo ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 00510 . 7554/eLife . 01817 . 006Video 1 . The primitive streak forms abruptly . Chick embryo development from late blastula to full primitive-streak stages . Time is indicated in hh:mm . The video shows the entire embryo ( about 3 mm in diameter ) and was made using a 2 . 5x objective and a conventional upright compound microscope with bright field optics . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 006 To examine when EMT begins , we labelled cells in the epiblast before PS formation ( stages X-XIII , Eyal-Giladi and Kochav , 1976 ) by widespread electroporation of a fluorescein-labelled control Morpholino . Within 1 hr after labelling and at all embryonic stages examined , all regions of the epiblast contain some cells at different stages of EMT: Figure 2Q shows an example , where cells with different morphologies have been classified into five ‘ingression stages’ ( 1–5 ) equivalent to those seen by SEM ( see above ) . 4 hr after labelling ( Figure 2R ) , some cells have left the epiblast but can be distinguished from hypoblast cells because the latter are much larger . Multi-photon time-lapse sequences reveal individual ingression events widely distributed in the epiblast , as early as stages X-XII , 6–15 hr before streak formation ( Figure 3A; Video 2 ) . 10 . 7554/eLife . 01817 . 007Figure 3 . Clustering of seemingly stochastic EMT underpins the formation of PS . ( A and B ) Uniform distribution of EMT in the epiblast before PS formation ( A ) and acceleration of EMT as the PS appears ( B ) . Locations are plotted from 6 hr time-lapse sequences ( see Videos 3 and 4 , respectively ) and the time of ingression is colour-coded ( numbers represent minutes ) . Each field of view is 600 × 600 μm , in the central posterior epiblast ( where the primitive streak arises ) . ( C and D ) Apical surface of the epiblast seen in SEM at PS formation stages . ( E ) An individual epiblast cell followed in time-lapse before ( see Video 5 ) undergoing repeated attempts at full EMT . ( F ) Multi-photon time-lapse sequence of EMT at PS stages . The top left-hand panel shows a diagram of the embryo with the area imaged enclosed in a square . The other panels represent views in the x-z ( top right ) , y-z ( bottom-left ) and x-y ( bottom-right ) . The positions of selected , colour-coded cells at successive time points ( 10 min intervals ) are connected with lines . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 00710 . 7554/eLife . 01817 . 008Video 2 . Isolated EMT in the epiblast before primitive streak formation . Multi-photon time-lapse sequence of the posterior epiblast at pre-primitive streak stages . The embryo was labelled by electroporation of a fluorescein-coupled control morpholino at stage EG&K XI , imaged every 10 min until stage EG&K XII ( time indicated in hh:mm ) . Top view perpendicular to the epiblast ( maximum intensity projection , scan depth 100 μm , z-spacing of 3 μm; scanned area 600 μm × 600 μm ) . Prospective ingressing cells in this sequence are marked by a blue dot; a red dot marks each ingression event . Relates to Figure 2A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 008 What underlies the transition from isolated EMT scattered across the epiblast to massive internalization at the PS ? Using SEM , we observed that the apical surface of the epiblast displays depressions 3–4 cells wide ( Bancroft and Bellairs , 1974 ) at stages XIII-2 ( just before and as the PS appears ) ; these increase in width and depth by stage 3 ( Figure 3C , D ) , suggesting coordinated apical constriction . In multi-photon time-lapse sequences , ingression events can be seen to accelerate as the streak forms ( Figure 3B , Figure 4; Videos 3 , 4 ) . Individual cells scattered throughout the epiblast undergo repeated cycles of incomplete delamination as they move towards the streak; this continues even at later PS stages ( Video 5; Figure 3E ) . Few cells far from the streak complete their delamination ( Video 6 ) . Ingression increases as cells approach the PS so that most of them ingress within 1 hr; however , cells reaching it at the same time do not necessarily ingress synchronously ( Video 7; Figure 3B , F ) and some ingressions occur away from the PS midline . For example in Figure 3F , a cluster of cells highlighted with different colours at the far left of the lower right hand panel ( dorsal view of the epiblast ) shows that cells that are close to each other ingress at different times and different positions along their trajectory towards the PS ( middle of the panel ) : one of the red cells ( higher in the panel ) ingresses furthest from the streak and earliest , whereas the other red cell ( the lowest in the group ) ingresses only when it reaches the streak . Together , these observations show that ingression of epiblast cells occurs throughout the epiblast at a low rate , but this rate increases markedly in the region of the forming streak . 10 . 7554/eLife . 01817 . 009Figure 4 . Quantification of ingression from the epiblast with time . Cell ingression accelerates as the PS forms and cells approach its midline . ( A ) the first and last frames of Video 4 ( left and right panels , respectively ) , highlighting the triangles used for measuring . ( B ) relative change in surface area of each triangle over time ( min ) . ( C ) relative change in surface area of each triangle as a function of distance to the midline ( in μm ) . ( D ) correlation coefficient ( r2 ) of the size reduction plotted against initial distance to the midline ( μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 00910 . 7554/eLife . 01817 . 010Video 3 . EMTs accelerate and cluster at the time of primitive streak formation . Multi-photon time-lapse sequence of the posterior epiblast around the time of primitive streak formation . The embryo was labelled by electroporation of a plasmid driving the expression of H2B-EGFP , imaged every 10 min between stages EG&K XIII and HH 3 ( time indicated in hh:mm ) . The upper panel is a top view of the epiblast ( maximum intensity projection , scan depth 100 μm , z-spacing of 3 μm; scanned area 600 μm × 600 μm ) . The lower panel is a side view ( YZ projection , along the forming primitive streak ) . Prospective ingressing cells are marked by a blue dot in this sequence and a red dot marks each ingression event . Relates to Figure 2A , B and Video 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01010 . 7554/eLife . 01817 . 011Video 4 . Changes in surface area in regions close to the site of primitive streak formation . The coloured dots mark some cells which do not ingress and which could be followed throughout the sequence in Video 3 . 15 triangles were drawn to connect sets of three cells . The surface area of each of these triangles was measured at each time point ( every 10 min ) , and the relative changes used to assess the net rate of ingression in each region; the results are plotted in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01110 . 7554/eLife . 01817 . 012Video 5 . Cells attempt EMT several times before full ingression . Multi-photon time-lapse sequence of an embryo at stage HH 3+ , whose epiblast was electroporated with a plasmid driving DsRed-Express , imaged every 10 min ( time indicated in hh:mm ) . 3D-reconstruction with the basal side of the epiblast towards the viewer and the axis of the primitive streak running from top ( anterior ) to bottom ( posterior ) . Cells of the wire-frame grid cells are squares 30 μm × 30 μm . One cell attempting EMT is highlighted in green and shown magnified in the insert to the right . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01210 . 7554/eLife . 01817 . 013Video 6 . Isolated EMTs outside the primitive streak . Multi-photon imaging of another embryo at stage HH 3+ , which had been electroporated with DsRed-Express plasmid around the primitive streak and imaged at 10 min intervals ( time indicated in hh:mm ) . 3D-reconstruction with the basal side of the epiblast towards the viewer ( similar to the one in Video 5 ) , with the primitive streak running from upper right ( posterior ) to lower left ( anterior ) . The arrow in the first frame points to a cell which will ingress outside the primitive streak . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01310 . 7554/eLife . 01817 . 014Video 7 . EMTs seem stochastic even at full primitive streak stages . Tracking of cell nuclei in an embryo at stage HH 3+ , which had been electroporated with H2B-EGFP plasmid and imaged at 10 min intervals . Time indicated in hh:mm; the tracks are colour-coded as indicated in the time bar ( lower-right; time indicated in hh:mm ) . The green balls show the positions of chosen nuclei at each time point . To allow visualization of all tracks , two views are shown from slightly different angles in the left and right main panels . In both , the apical side of the epiblast is towards the viewer and its basal side away; the primitive streak runs along the middle ( anterior towards the top , posterior towards the bottom ) . The black insets ( top right corner of each main panel ) show an overview of the entire volume scanned . Cells of the grid box are squares , 50 mm × 50 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 014 The increase in ingression rate in the proximity of the future PS territory suggests that cells that have already ingressed may favour ingression of their neighbours . To test this , we grafted small groups of ingressed cells from the posterior early PS ( stage 2–3 ) of quail embryos , under the epiblast of pre-PS stage chick embryos ( stage XII–XIII ) ( Figure 5A ) . We chose these cells because they only contribute to lateral mesoderm and not to axial tissue ( Hatada and Stern , 1994; Psychoyos and Stern , 1996 ) ; this differs from grafts of organizer or Koller's sickle ( Izpisua-Belmonte et al . , 1993; Bachvarova et al . , 1998 ) , both of which contribute to and induce an organizer ( Izpisua-Belmonte et al . , 1993; Bachvarova et al . , 1998; Streit et al . , 2000 ) . The grafted cells induce PS markers in the adjacent epiblast within 4 hr ( cBra , 8/8 embryos , cSnail2 , 10/10 ) and massive ingression ensues ( Figure 5B ) . After 14 hr , a second , host-derived streak develops from the graft site ( 8/11; Figure 5C ) , whereas the grafted cells ( prospective lateral mesoderm [Hatada and Stern , 1994; Psychoyos and Stern , 1996] ) migrate away . When a similar graft is made using mesoderm from more lateral cells that have emerged from the PS , no such induction occurs ( see below ) . This is consistent with a previous study ( Vakaet , 1973 ) using full-thickness grafts of posterior mature PS ( ‘nodus posterior’ ) . Our results implicate the mesoderm as the source of the inductive signals . Early ingressed cells induce mesendodermal identity and increase the probability of other epiblast cells undergoing EMT , suggesting that the PS forms and maintains itself by positive feedback mediated by a community effect ( Gurdon , 1988 ) . 10 . 7554/eLife . 01817 . 015Figure 5 . Cells in EMT trigger a chain reaction of EMT in a Nodal-dependent manner . ( A ) EMT cells from the early PS of a quail embryo ( left ) are grafted under the epiblast of a pre-PS chick embryo ( right ) . ( B ) Grafted cells ( brown stain , thin black arrow ) upregulate EMT markers ( cSnail2 , purple ) and trigger EMT from the epiblast above , after 4 hr . ( C ) Grafted embryo after 15 hr . The grafted quail cells ( brown ) have migrated away , and the new PS they triggered ( ‘2o PS’ ) is composed of host cells . The PS developing along the original orientation is labelled ‘1o PS’ . In grafts combined with COS cells secreting Cerberus ( E ) or Cer-S ( F ) , or beads soaked in SB431542 ( H ) or SB505124 ( I ) , EMT ( thickening ) from the epiblast and induction of cSnail2 ( purple ) in the epiblast ( red arrows ) are abolished . Control COS cells ( D ) or beads soaked in DMSO ( G ) do not abolish the induction by the grafted mesoderm ( black arrows ) . Mesoderm from a region lateral to the PS cannot induce EMT or cSnail2 either alone ( not shown ) or in the presence of GFP-transfected COS cells ( J ) or beads soaked in solvent alone ( K ) ( red arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 015 What is the molecular basis of this community effect ? Candidates include pathways implicated in mesendoderm induction ( FGF , TGFβ/Nodal ) and/or patterning ( canonical Wnt , BMP ) ( Carnac and Gurdon , 1997; Standley et al . , 2001; Stern , 2004a ) . We co-transplanted recently ingressed cells with COS cells secreting specific inhibitors ( Figure 5E–F ) or beads soaked in chemical modulators of each pathway ( Figure 5H–I ) . SU5402 ( FGF-inhibitor ) , Crescent , Dkk and alsterpaullone ( canonical-Wnt-modulators ) , chordin and noggin ( BMP-inhibitors ) did not inhibit induction ( n = 9 each except Dkk , n = 7 ) . However , Cerberus ( BMP- and Nodal-inhibitor , 9/9; Figure 5E ) and Cerberus-Short ( Nodal-inhibitor , 8/9; Figure 5F ) , as well as SB4315412 ( 10/12; Figure 5H ) and SB505124 ( 11/12; Figure 5I ) ( inhibitors of TGFβ superfamily receptors ALK4/ALK7 ) all prevented both the induction of PS markers ( cBra , cSnail2 ) and ingression of epiblast cells adjacent to the graft . Importantly , they can do this without loss of the markers in the graft cells themselves ( Figure 5E–F ) . Control COS cells ( Figure 5D ) and beads ( Figure 5G ) do not prevent induction by the grafted mesoderm . Grafts of mesoderm from outside the PS do not induce the markers either in the presence ( Figure 5J , K ) or absence ( not shown ) of beads or COS cells . These results suggest that TGFβ-related factors , and most likely Nodal , are required for the community effect by newly-ingressed mesendoderm . Nodal is expressed before streak formation in a posterior domain of the epiblast ( Bertocchini and Stern , 2002; Skromne and Stern , 2002 ) , but its activity is initially blocked by Cerberus ( Bertocchini and Stern , 2002 ) , an antagonist produced by the hypoblast . This expression domain seems to be identical to the region in which we previously found cells to undergo intercalation parallel to the marginal zone , driven by the Wnt-PCP pathway ( Voiculescu et al . , 2007 ) . The domain of Nodal expression and intercalation adopts the shape of the forming streak . Thus , two separable local cell interactions ( intercalation and EMT amplified by a community effect ) are necessary for PS formation . Are they sufficient to explain PS shape and appearance as well as the complex pattern of tissue movements before and during gastrulation ? To address this question we used an agent-based model where these cell behaviours are explicitly added to a simple representation of a bounded epithelial sheet ( ‘Materials and methods–Description of the model’ ) . The model assigns various states ( e . g . , Wnt-PCP , Nodal ) to cells ( Figure 6; Table 2 ) ; cells modify their states and execute behaviours based upon their current internal state and interactions with their neighbours ( e . g . , oriented intercalation , self-amplifying EMT; see Table 3 for a summary of the model rules ) . 10 . 7554/eLife . 01817 . 016Figure 6 . Different views of a simulation of normal development . These diagrams provide an explanatory key for the simulation videos and illustrate the principal signals , cell behaviours and the major tissues involved in gastrulation . Three time points are shown: stage XI , stage 2 and stage 3+ . The upper 7 rows are dorsal views onto the epiblast; the lower 3 rows are oblique views . Colours are additive when a cell is positive for more than one displayed state ( see e . g . , the row labelled ‘combined’ , which symbolises the sum of all features in the rows above it for the forming primitive streak ) . Nodal ( + ) cells are shown in red ( top row ) , Wnt-PCP ( + ) cells in yellow ( second row ) . Cells positive for both Nodal and Wnt-PCP appear orange ( third row ) . At Stage XI all cells in the future streak-forming region are Nodal and Wnt-PCP positive . Later , most continue to have both activities but some cells are only positive for Nodal ( red ) . Cells undergoing EMT are shown in blue and ‘mesendodermal’ cells in aquamarine ( fourth row ) . For combinations of Nodal , Wnt-PCP , EMT and mesendoderm , note that Nodal ( + ) -EMT cells appear purple ( red + blue ) ; if also Wnt-PCP ( + ) then approximately violet ( red + yellow + blue ) ( ‘combined’ ) . The hypoblast is shown chocolate-coloured and the endoblast greenish-slate ( rows 6 and 8 ) . Hypoblast displacement by the endoblast ( at stage XIV; between stages XI and 2 in the Figure ) disinhibits Nodal in the overlying epiblast ( see text ) . Sequential cell positions are integrated by remembering all previous time points to form ‘trails’ , as shown in row 7 . For clarity , trails made from 15% of the cells are shown . The last three rows depict the embryo viewed from an oblique angle . In row 8 ( ‘hypoblast and endoblast’ ) , the position of the lower layer can be seen ( also see above , lower layer ) . Initially this consists only of hypoblast ( chocolate ) . At later stages , endoblast ( greenish-slate ) partially displaces the hypoblast . The epiblast is also seen from below ( ‘epiblast ventral view’ , row 9 ) , allowing clear visualization of EMT ( blue/purple/violet ) and emerging and emerging middle layer ( aquamarine ) cells . The final row , ‘epiblast dorsal view’ ( row 10 ) , displays the epiblast from above with a pseudo-surface applied , simulating indentations caused by ingressing cells . These indentations sum as cells approach the posterior midline , generating a midline groove at the PS . The pseudo-surface is created by tessellating points representing the top of each epithelial cell ( using the cell body for cells undergoing EMT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01610 . 7554/eLife . 01817 . 017Figure 6—figure Supplement 1 . Modelling: hierarchical time implementation . Time is represented as ‘ticks’ . Each simulation tick executes activities that include a set of actions for the entire embryo ( ‘organism tick’ ) . The organism tick in turn executes activities including a cell tick for each cell in the organism . Cell ticks calculate and execute activities for each cell . Note that many of these calculations and activities are themselves iterative . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01710 . 7554/eLife . 01817 . 018Figure 6—figure Supplement 2 . Modelling: MZ displacement vectors . For each MZ cell a displacement vector ( white arrowhead ) is calculated as the vector sum of ‘curvature’ ( orange ) , ‘density’ ( green ) and ‘area correction’ ( blue ) vectors . A mark ( red dot ) identifies the common origin of each . Vectors are shown magnified 50x for illustration . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01810 . 7554/eLife . 01817 . 019Figure 6—figure Supplement 3 . Modelling: schematic representation of EMT and Nodal expression . ( A ) Shown is a schematic representation of EMT in the model . Right side: Nodal ( − ) epithelial cells ( grey ) may convert to emt cells ( blue ) which at first are tethered to the epithelium ( t-emt ) but then become untethered ( u-emt ) as they descend into the middle layer . They complete the transition as mesenchymal ( meso ) cells ( aquamarine ) . While still tethered and with cell body above the basement membrane ( BM ) , some will revert and rejoin the epithelium ( double-headed arrow ) . Left side: Nodal ( − ) epithelial cells convert to Nodal ( + ) ( red ) in the region of the PS . The rate of EMT increases with increasing Nodal activity from the cell and its neighbours; Nodal-active emt cells ( red + blue = purple ) lose the ability to rejoin the epithelium ( thicker , single-headed arrow ) . Conversion to Nodal-positivity and the enhanced rate of EMT is inhibited by the hypoblast and disinhibited when the endoblast displaces the hypoblast . ( B ) Shown are cell interactions leading to Nodal expression . Nodal ( − ) epithelial cells ( grey ) are converted to Nodal ( + ) cells by near-neighbour Nodal ( + ) epithelial cells ( red ) and local neighbour Nodal ( + ) emt cells ( purple = blue[emt] + red[Nodal] ) . For local neighbours the effect falls off with distance but is particularly enhanced for near-neighbour epithelial cells ( arrow widths ) . A similar scheme ( not shown ) applies to Wnt-PCP conversion and to EMT recruitment . Numbers of cells , distances and proportions not to scale . BM: basement membrane , t-emt: tethered emt cell , u-emt: untethered emt cell , meso: mesenchymal cell . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 01910 . 7554/eLife . 01817 . 020Figure 6—figure Supplement 4 . Modelling: cell movements within the plane of the epiblast . ( A ) Schematic diagram of the equilibration algorithm . Cells ( solid circles ) are distributed in a hexagonal array . Voronoi regions ( VR ) are transiently generated around these cells ( solid lined hexagons ) . Cell centroids and VR centroids correspond ( dot ) and the tissue is at equilibrium . When cell ‘A’ shifts to a new position ( dashed circle ) , new VR's are generated ( dotted hexagons ) , making the centroid of the new VR for neighbouring cell ‘B’ move to a new position ( cross ) . Cell ‘B’ then shifts towards this new position to reestablish equilibrium ( arrow ) . The vector from the original centroid of cell ‘B’ to the centroid of the new VR of cell B is the equilibrium displacement vector ( vequil ) . ( B ) Propagation of oriented intercalation orientation vectors . MZ-cells maintain a reference OI-orientation vector state perpendicular to the MZ ( determined by the local curvature ) . Epiblast cells calculate their individual OI-orientation vector states by averaging their current vector with the consensus of their near-neighbours , including MZ-cells ( see text ) . Since the MZ-cell vectors are fixed , epiblast cells abutting the MZ will tend to align their vectors to those of the MZ-cells . Note that although this state is stored as a vector in the model , it has angular but not heads vs tails orientation . ( C ) Diagram of the oriented intercalation algorithm . A cell and its near-neighbour ( NN ) both possess OI-orientation information ( double-headed arrows ) . A sequential displacement vector is calculated , oriented from the cell to its target and with a magnitude equal to \|sinθ\| . This is applied iteratively for all cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02010 . 7554/eLife . 01817 . 021Table 2 . List of model cell states which include anatomic cell types as well as signalling mediators and behavioural descriptorsDOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 021StateTypeVisualizationDescriptionInitiation , maintenance and propagationCell TypesMZbooleangreenMZ pseudo-cells surround the epiblast and form a boundaryinitial ring of cells with shape altered to minimize local curvature and epiblast cell density , then adjusted to control epiblast areaepiblastbooleangreyepiblast epithelial cellsinitial disc of cells; increased by cell division; decreased by ingressionemtbooleanmagentaepiblast cells undergoing EMTcells attempting to ingress from the epiblast in a Nodal-dependent processtetheredbooleanby shapeflags whether or not EMT cell remains tethered to epiblast epitheliumtethered emt cells may re-incorporate into the epithelium whereas untethered cells are committed to progress to mesenchymemesobooleanbluecells which have completed transition to mesenchymeend result of EMT; a terminal , inactive state in these simulationsOther StatesNodalbooleanredcells expressing/ secreting Nodalinitially present in PS forming region; cells may be converted to positive by neighboursWnt-PCPbooleanyellowWnt-PCP ( + ) cells capable of oriented intercalationinitially present in the PS forming region; cells may be converted to positive by neighboursOI-vectorvectorline segmentorients intercalation of Wnt-PCP ( + ) cellscalculated by consensus among Wnt-PCP ( + ) cellsColour codes can be matched to simulation images in Figures 6 and 7 and Supplementary Videos . 10 . 7554/eLife . 01817 . 022Table 3 . Description of the rules used in the modelDOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 022Tissue Structure cells in the epiblast are arranged in a flat epithelial layer; cells undergoing EMT descend beneath this layer the epiblast is surrounded by a ring of marginal zone ( MZ ) cells that acts as a malleable boundaryCell State Activities an initial cohort of cells in the posterior epiblast is positive for Nodal and the Wnt-PCP system Nodal-negative cells may become Nodal-positive if they receive a Nodal signal from neighbours Wnt-PCP-negative , Nodal-positive cells may be converted to Wnt-PCP-positive if surrounded by Wnt-PCP-positive neighbours epithelial cells are more likely to undergo EMT if neighbouring cells are undergoing EMT ( a community effect mediated in the model by Nodal ) the appearance of the endoblast at Stage XIV displaces the hypoblast ( which secretes the Nodal antagonist Cerberus ) resulting in Nodal disinhibition in the posterior part of the embryoCell Physical Activities epithelial cells undergo a cell cycle and divide in the plane of the epiblast epithelial cells maintain spatial equilibrium by centring themselves amongst their near-neighbours epithelial cells may convert to EMT cells which are initially tethered to the epithelium some EMT cells may become untethered , exit the epithelium and ingress to become mesenchyme cells not experiencing Nodal activity ( either by being far from the primitive streak ( PS ) where Nodal is expressed , or by having Nodal inhibited by Cerberus from the hypoblast ) undergo EMT at a low rate and may revert back to epithelium Nodal-active epiblast cells undergo EMT at an enhanced rate and do not back-convert Wnt-PCP-positive cells undergo oriented intercalation with an orientation based on a consensus of the contiguous cohort of intercalating cells , oriented relative to the MZ ( intercalation occurs at approximately right angles to the tissue radius ) For details , including mathematical formulations , see ‘Material and methods—Description of the Model’ . In the model , the localized intercalation behaviour , first appearing in the pre-PS epiblast , can recreate movements similar to the early Polonaise seen in real embryos ( Figure 7A–E , F–H , K–M; Videos 8 , 9 ) ; the isolated , uniform EMT occurring at these stages has minimal effect . When cooperativity of EMT is triggered in the intercalation domain ( by disinhibition of Nodal activity [Bertocchini and Stern , 2002] , because of the displacement of the hypoblast away from the posterior Nodal-expressing zone ) , massive ingression occurs . In line with experimental observations , this causes the movement pattern to be altered , with cells now entering the PS along direct lateral-to-medial trajectories . The simulations faithfully recreate the large-scale Polonaise movements as well as PS formation and its role as a gateway for gastrulation via cell ingression . Importantly , the global Polonaise movements follow passively from active events localized to the posterior PS-forming region and then the PS itself . 10 . 7554/eLife . 01817 . 023Figure 7 . A model based on local cell behaviours explains the global movements in the epiblast and experimental conditions . ( A–E ) Epithelial intercalation in a posterior domain ( orange ) and EMT ( blue , isolated events , cooperative in the pink domain ) are sufficient to explain the formation of the PS . ( A–C ) sequence in time , vertical view; ( D ) ventral view of the epiblast; ( E ) apical view of the epiblast . ( F–H ) Sequence from a time-lapse experiment , with cells in the intercalation domain electroporated with control morpholino ( green ) and other locations in the epiblast labelled with DiI ( red ) . ( F ) initial condition , 6 hr before streak formation; ( G ) movements prior to streak formation; ( H ) movements over 6 hr after PS forms . ( I and J ) Movements observed in the same time-frame as in F–H , when intercalation is blocked by electroporating morpholinos ( green ) against the Wnt-PCP pathway . ( K–O ) The computer model correctly simulates the observed movements both in normal embryos ( K–M ) and in intercalation-compromised condition ( N and O ) . ( P–R ) Hypoblast rotation at pre-PS stages leads to bending of the PS . ( P ) Experimental embryo , with the PS marked by Bra expression; the model accounts for this result ( red in Q ) by the induction of a new intercalation domain ( yellow in R ) which deforms the original one and the field of cooperative ingression ( orange in R ) . ( S and T ) EMT cells can trigger a chain reaction of EMT and initiate a new PS in both experimental embryos ( S ) and in the model ( T ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02310 . 7554/eLife . 01817 . 024Figure 7—figure Supplement 1 . Effect of key parameters on the behaviour of the computer simulation model . The figure shows the composite effects of changing the value of mN and σd on PS morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02410 . 7554/eLife . 01817 . 025Figure 7—figure Supplement 2 . Model predictions compared with the results of Spratt ( 1946 ) . Spratt: global epiblast movements as described by Spratt from carbon-particle marking experiments ( Spratt , 1946 ) ( adapted from Spratt 1946 ) . The diagrams combine a movement schematic and a representation of the PS . Model: a series of stages in a normal simulation showing the fates of horizontal bands of marked cells ( upper row ) and formation of the PS ( lower row ) . The simulated pattern is also consistent with more recent analysis of epiblast cell movements and the Polonaise ( Foley et al . , 2000; Wei and Mikawa , 2000; Voiculescu et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02510 . 7554/eLife . 01817 . 026Video 8 . Movements of the epiblast cells before and during gastrulation . Cells in a posterior crescent of the epiblast were electroporated with control morpholino ( green ) , and various locations in the rest of the epiblast labelled with DiI ( red ) at stage EG&K XII and the embryo filmed in a conventional fluorescence microscope in time-lapse . Time indicated as hh:mm before ( negative values ) and after primitive streak formation . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02610 . 7554/eLife . 01817 . 027Video 9 . Simulation of normal chick gastrulation . Different views of videos showing simulations of normal embryo development ( the videos are synchronised with each other ) . Left column: all cells in the embryonic epiblast are shown in white , confined by the marginal zone ( green ) . In the upper panel , the lower layers are displayed in the background , the hypoblast in pale brown and the endoblast in pale green; in the lower panel , only the epiblast cells are shown . The epiblast cells performing oriented intercalation in the posterior crescent are shown in orange and the early ingressing cells in blue . Cells ingressing by a community effect are displayed in pink . See also Figures 6 , 7 , Tables 2 and 3 , and ‘Materials and methods—Description of the Model for details and colour codes . Middle column: cell movements in the epiblast . In the upper panel , horizontal bands of cells are coloured differently , to allow comparisons with the results in Gräper 1929; in the lower panel , cells in the posterior domain were coloured green and groups of cells in other epiblast locations in red , allowing comparisons with the experimental observations presented here ( Video 8 ) . Lower right panel: global movements in the epiblast . A uniform grid of individual objects were tracked over time , and their trajectories are time-coded in rainbow colours . Top right: pattern of ingression in the epiblast . The apical aspect of the epiblast is shown at the top , and its basal side in the lower side of the panel . Cells engaged in EMT are shown in blue , and the locations of completed EMTs in turquoise . Time indicated as hh:mm before ( negative values ) and after primitive streak formation , as for the experimental observations in Video 8 . See also Figure 7F–H , K–J . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 027 We tested the effect of changing several parameters . As expected , increasing the strength of intercalation causes the streak to converge and elongate more rapidly , and vice versa . We also examined the kinetics of ingression by changing the mean of the logistic curve defining the probability of ingression ( mN in Equation 5 . 1 , ‘Materials and methods–Description of the model’ ) which effectively alters the community effect threshold and by changing the spatial range of effector cells ( σd in Equation 5 . 2; dmax was co-modified to remain 4 × σd ) . Figure 7—figure supplement 1 shows images from simulations in which mN is varied from 6 . 0 to 10 . 0 ( bracketing 8 . 0 used in the study ) and images from simulations where σd is varied from 5 . 0 to 20 . 0 μm ( bracketing the 10 . 0 μm used in this study ) . Parameter changes that increase the strength of the ingression community effect ( smaller mN or larger σd ) tend to produce more diminutive streaks since incorporation of new cells into the streak ( Equations 6 . 1 , 6 . 2 , 7 . 1 , 7 . 2 ) is less able to compensate for increased loss by ingression . In the model , events in the PS ( intercalation and ingression ) coupled with the cells in the non-PS epiblast spatially equilibrating amongst themselves results in the global epiblast movement pattern . As the PS converges , cells just lateral and anterior shift posteriorly and medially to ‘fill-in’ the area being evacuated . This shifting is propagated to successively more distant cells . The anterior and lateral epiblast thus sweeps posteriorly and medially . At the same time , convergence and extension in the PS generates movement along the midline; this is mainly directed anteriorly , as the marginal zone limits posterior extension . Together these processes yield both the circumferential ( lateral epiblast ) and anterior ( posterior midline epiblast ) components of the Polonaise movement pattern as seen in normal embryos ( Spratt , 1946; Figure 7—figure supplement 2 ) . To distinguish the roles of intercalation and ingression ( EMT ) in the global Polonaise cell movements , we blocked medio-lateral intercalation by electroporation of a mixture of morpholinos against components of the Wnt-PCP pathway ( Voiculescu et al . , 2007 ) and followed the movement of cells elsewhere in the epiblast by using the carbocyanine dye DiI to label a lattice of cells throughout the epiblast ( Figure 7I–J ) . At pre-streak stages , we observe a complete arrest of the Polonaise movements ( Gräper , 1929; Wetzel , 1929 ) across the entire epiblast ( Videos 8 , 10; in Figure 7 compare G , normal embryo , with I , where Wnt-PCP was blocked in the intercalation domain ) . Massive ingression is still triggered at the appropriate time , but in a posterior domain close to the margin of the embryo rather than at the midline . In these experiments , cells move posteriorly towards this zone of concentrated ingression along abnormal , straight-line trajectories ( in Figure 7 compare H , normal embryo , with J , blocked intercalation ) . These findings suggest that shaping of the early PS in the normal embryo is mainly driven by intercalation , which results in coalescence and extension in the midline and the Polonaise movement pattern . After its formation , the PS is maintained as a zone of massive ingression that generates a more direct pattern of cell convergence , with cell trajectories perpendicular to the axis of the PS ( Video 8; see Figure 1 , stage 3+ ) . We used the model to simulate this experiment . With standard parameters and without changing any other conditions , the model faithfully reproduces the altered movement pattern associated with the abrogation of intercalation using Wnt-PCP-Morpholinos in posterior cells ( Voiculescu et al . , 2007; Video 11; Figure 7N–O ) . 10 . 7554/eLife . 01817 . 028Video 10 . The Polonaise movements are driven by intercalation in the posterior domain . Cells in the posterior crescent of the epiblast electroporated with a combination of morpholinos blocking the Wnt-PCP pathway ( green ) , and various locations in the rest of the epiblast labelled with DiI ( red ) at stage EG&K XII . Embryo filmed by conventional epifluorescence in time-lapse . Time indicated as hh:mm before ( negative values ) and after primitive streak formation . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 02810 . 7554/eLife . 01817 . 029Video 11 . Simulation of experimental abrogation of intercalation in the posterior epiblast . In this simulation , oriented intercalation was abolished in 50% of the cells in the domain where this normally occurs ( to simulate the Morpholino experiment , Video 10 and Figure 7 ) . Cell states , domains and trajectories are represented in the same way as for the normal embryos , lower row in Video 10 . The middle panel allows direct comparison with the experimental blockage of Wnt-PCP pathway in the posterior crescent ( Video 10 ) with electroporated cells depicted in green and dots at random locations labelled in red ( to simulate the DiI labelled cells ) . See also Figure 7I , J , N , O . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 029 All above experiments and simulations suggest that just two cell behaviours , cell intercalation and EMT , the latter subsequently amplified locally by a community effect , are sufficient to account for the movements of gastrulation . To test whether they can also account for other reported experimental manipulations , we simulated the effects of hypoblast rotation . In real embryos , rotation of the hypoblast by 90° bends the PS ( Waddington , 1932 ) because of altered cell movements ( Foley et al . , 2000 ) . Signals from the hypoblast can induce a new domain of PCP activity which was proposed to account for these events ( Voiculescu et al . , 2007 ) . We used the model to test whether this is a sufficient explanation for bending of the PS; simulations suggest that it is ( Video 12; Figure 7P–R ) . Finally , we tested whether the model can also account for our present finding that an ectopic streak can be induced by a graft of ingressed cells . Again , the model can simulate this result without changing any parameters ( see above , Video 13; Figure 7S–T ) . 10 . 7554/eLife . 01817 . 030Video 12 . Simulation of hypoblast rotation experiment . Based on experimental findings ( Voiculescu et al . , 2007 ) , hypoblast rotation by 90° induces a supplementary domain of Wnt-PCP gene expression and oriented intercalation ( shown here in yellow ) was added to the simulation , to mimic the rotation of hypoblast . As in experimental cases ( Waddington , 1932; Foley et al . , 2000; Voiculescu et al . , 2007 ) , hypoblast rotation leads to bending of the primitive streak . The colour coding follows the scheme employed in Video 9 , lower left panel . See also Figure 7P–R . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 03010 . 7554/eLife . 01817 . 031Video 13 . Simulation of EMT induced by a mesoderm implant . A group of ingressing cells ( pink , as in the colour scheme in Video 9 , lower left panel ) was ‘grafted’ at pre-primitive streak stages to a lateral region of a simulated normal embryo . As in the experiments presented in Figure 5 ( see also Figure 7S , T ) , this results in the induction of an ectopic primitive streak from host cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01817 . 031 In conclusion , our experimental observations and computer model suggest that just two cell behaviours , cell intercalation localized to a posterior domain of the area pellucida epiblast ( future streak forming region ) together with EMT events , amplified by a community effect mediated by Nodal , are sufficient to explain all four major movements of chick gastrulation: the Polonaise of the early epiblast , the elongation of the PS , the movement of epiblast cells towards the streak and their ingression through the streak . Several theories have been put forward to account for the origin of the PS and for the associated cell movements . In the chick , some argue that all precursors of the streak arise from and multiply in a very small region of the posterior epiblast ( Wei and Mikawa , 2000 ) , whereas others propose that the precursors of the early mesendoderm are present in all regions of the early epiblast , based on the observation that HNK1+/Acetylcholinesterase+ cells are scattered randomly throughout the pre-PS epiblast ( Drews , 1975; Canning and Stern , 1988 ) , contribute to mesendoderm and are required for PS formation ( Stern and Canning , 1990 ) . Here we show that ingression of cells starts well before gastrulation , by individual cells sparsely scattered across the epiblast , in a pattern similar to the HNK1+ cells . Our results strongly suggest that the early ingressing cells correspond to , or are a subset of the HNK1+ population , which also explains why early ingressed mesoderm cells can rescue PS formation in embryos from which HNK1+ cells have been ablated ( Stern and Canning , 1990 ) . There are interesting parallels with the sea urchin , where gastrulation is initiated by maternally specified ‘pioneer’ cells , the Primary Mesenchyme cells ( PMCs ) ( Sherwood and McClay , 1999; McClay et al . , 2000; Sweet et al . , 2002 ) , also characterised by their expression of Acetylcholinesterase ( Drews , 1975 ) , which carries the HNK1 epitope ( Bon et al . , 1987; Canning and Stern , 1988 ) . At present , it is not known when and how HNK1+ cells are specified in the chick epiblast , but they share two key characteristics with their sea urchin counterparts: they are required for proper gastrulation ( mesendoderm formation ) and have inductive abilities . We propose that as in the sea urchin , amniote mesendoderm formation is initiated before the PS forms by ‘pioneer’ cells that ingress as individuals at relatively low frequency throughout the epiblast . Our results indicate that , in the presence of Nodal activity , these early ingressing cells can trigger a chain reaction of EMT , induce the expression of mesendodermal markers and PS formation . In zebrafish , when the Nodal pathway is compromised , only about 60 cells still ingress instead of the ∼2500 that normally do ( Keller et al . , 2008 ) , consistent with a community-effect mediated by Nodal being conserved in vertebrates . In sea urchin , however , the inductive effect of PMCs on gastrulation is mediated by Notch signalling ( Sherwood and McClay , 1999; McClay et al . , 2000; Sweet et al . , 2002 ) . The amniote hypoblast ( a transient layer of cells ) plays a crucial role in coordinating the timing of PS formation with other cell movements . Medio-lateral intercalation in the epiblast prior to the start of gastrulation acts to displace and re-shape the Nodal domain and the prospective mesendodermal territory to the midline . Later in development , cell intercalation seems to continue to play a role in axial elongation by driving convergence/extension movements in the midline mesoderm and the overlying neuroectoderm ( prospective floor plate ) , after the initial appearance of notochord cells . These later movements in the mesoderm and prospective floor plate are also found in anamniotes , as has been demonstrated in Xenopus and zebrafish ( Yeo et al . , 2001; Ezin et al . , 2006 ) , whereas the early ( pre-gastrulation ) movements are unique to amniotes . Our results provide a mechanistic explanation for how the displacement of the chick hypoblast ( Bertocchini and Stern , 2002 ) ( expressing Cerberus ) by the endoblast , or of the mouse anterior visceral endoderm ( AVE ( Perea-Gomez et al . , 2002 ) , expressing Cerberus and Lefty1 ) leads to extensive EMT and PS formation . We propose that apart from a role in nutrition of the embryo , the hypoblast/AVE acquired the function of delaying PS formation while repositioning the streak precursor cells to the midline ( Stern and Downs , 2012 ) . This occurs because the Nodal-expressing domain also expresses components of the Wnt-PCP pathway and undergoes intercalation , independently of ingression ( Voiculescu et al . , 2007 ) . Distinct molecular pathways mediate the dual role of the hypoblast: FGF , which controls the Wnt-PCP pathway and positions the PS ( Voiculescu et al . , 2007 ) , and Nodal antagonism ( perhaps together with Wnt antagonism ) , which regulates the timing of PS formation ( Bertocchini and Stern , 2002 ) . We suggest the following model of amniote gastrulation . ( I ) In the stages leading up to gastrulation , cells in a crescent-shaped posterior region of the epiblast express Nodal and the Wnt-PCP system . The Wnt-PCP system drives oriented intercalation of epithelial cells , parallel to the marginal zone ( perpendicular to the future body axis ) . Nodal is a potent enhancer of EMT activity ( ingression ) and sensitizes cells to activity in neighbouring cells ( community effect ) . However , underlying the epiblast is the hypoblast , a suppressor of Nodal activity ( through Cerberus ) . Thus cells outside the posterior region ( Nodal-negative ) and cells within the region ( Nodal-positive but hypoblast suppressed ) only attempt EMT at a low rate , and most of these attempts are unsuccessful: they do not result in cell ingression . The Wnt-PCP system causes this posterior cohort of Nodal-positive cells to converge to , and extend along , the midline . Movement of these cells towards the midline draws in neighbouring cells , the displacements are propagated outwards and , through the geometry of the roughly circular epiblast confined within the marginal zone , the Polonaise movement pattern ensues . ( II ) Endoblast derived from the posterior germ wall ( the deep , yolky cells of the area opaca , Stern , 1990 ) displaces the hypoblast away from the posterior part of the embryo , unleashing Nodal activity in the Nodal-positive population , now localized at the posterior midline . Under the influence of Nodal , ingression accelerates and becomes self-reinforcing , generating the PS . The now massive loss of cells through ingression within the midline PS pulls in lateral neighbours and the displacements are propagated laterally , resulting in a transverse movement pattern . As lateral cells enter the PS , they become Nodal and Wnt-PCP positive and fuel the process . The PS continues to elongate through incorporation of lateral cells and oriented intercalation . Together , our results suggest that amniote gastrulation is a population event . The PS is not a fixed gateway for cell internalization but rather a dynamic , self-reinforcing concentration of individually ingressing cells . These results provide a mechanism for the self-maintenance of stable morphological structures as their cell composition changes ( Joubin and Stern , 1999 ) . They also demonstrate that large-scale movements and morphogenesis of entire epithelial sheets can be driven by local cell interactions , without the need for signalling over long distances . Nudge++TM ( a product of Olana Technologies , Inc . – info@olanatech . com ) is an agent-based modelling system designed to study multi-cellular morphogenesis . The current model builds on previously described versions ( Bodenstein and Stern , 2005; Fisher and Bodenstein , 2006 ) . Simulated biological cells are the model agents . A two- or three-dimensional simulated tissue is constructed as a cohort of these model cells . Cells execute individual cellular programs leading to actions . The cellular programs reference internal states and external cues , the latter of which may include the states and actions of neighbouring cells . The pooled behaviour of the entire cohort of cells leads to tissue morphogenesis ( Bodenstein , 1986 ) .	A key process during the development of an embryo involves a single layer of cells reorganizing into three ‘germ layers’: the ectoderm , which becomes the skin and nervous system; the mesoderm , which gives rise to the skeleton , muscles and the circulatory and urinogenital systems , and the endoderm , which gives rise to the lining of the gut and associated organs . The process of forming these three layers is known as gastrulation . To date most experiments on gastrulation in vertebrates have been performed on frog embryos . However , the embryos of amniotes , the group of ‘higher’ vertebrates that comprises reptiles , birds and mammals , differ from those of frogs in a number of ways . Now Voiculescu et al . have used a combination of experimental and computational techniques to shed new light on gastrulation in chick embryos . Just prior to gastrulation , the cells of the amniote embryo are arranged in a flat disk , one cell thick , called the epiblast . The cells of the epiblast then move to form the mesoderm and endoderm ( in a process called epithelial-to-mesenchymal transition ) . These cell movements also lead to the formation of a structure called the primitive streak that establishes the left-right symmetry of the organism , and also defines the midline of the body . Now Voiculescu et al . have shown that the epithelial-to-mesenchymal transition starts before the primitive streak appears , and that two main processes drive gastrulation . One involves cells inserting themselves between other cells at the midline of the epiblast , which causes a double whorl-like movement within the plane of the epiblast . At the same time small numbers of cells leave the epiblast , and as these cells accumulate under the epiblast , they initiate a positive feedback effect by which they encourage more cells to leave the epiblast . Voiculescu et al . found that this ‘community effect’ involves signalling by a protein called Nodal . This protein effectively amplifies the epithelial-to-mesenchymal transition and leads to the appearance of the primitive streak at the midline . Using computational modelling , Voiculescu et al . argue that the movements of gastrulation can be explained entirely based on local interactions between cells , without the need for cells to send signals over long distances to guide cell movements , as had been generally believed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	Local cell interactions and self-amplifying individual cell ingression drive amniote gastrulation
Signaling pathways often transmit multiple signals through a single shared transcription factor ( TF ) and encode signal information by differentially regulating TF dynamics . However , signal information will be lost unless it can be reliably decoded by downstream genes . To understand the limits on dynamic information transduction , we apply information theory to quantify how much gene expression information the yeast TF Msn2 can transduce to target genes in the amplitude or frequency of its activation dynamics . We find that although the amount of information transmitted by Msn2 to single target genes is limited , information transduction can be increased by modulating promoter cis-elements or by integrating information from multiple genes . By correcting for extrinsic noise , we estimate an upper bound on information transduction . Overall , we find that information transduction through amplitude and frequency regulation of Msn2 is limited to error-free transduction of signal identity , but not signal intensity information . Cellular signaling pathways often exhibit a bowtie topology ( Csete and Doyle , 2004 ) : multiple distinct signal inputs converge on a single master regulator , typically a transcription factor ( TF ) , which then controls the expression of partially overlapping sets of downstream target genes . This raises two general questions: first , how can the cell encode information about different signals in the activity of a single master TF ? Second , can this information be decoded by target genes to elicit a specific output for each input ? One way the cell can encode signal information is by regulating the activation dynamics of a single master TF ( Figure 1A ) . For example , p53 , a tumor suppressor TF , exhibits an intensity-dependent number of nuclear pulses in response to γ-radiation , but a sustained pulse of nuclear localization with intensity-dependent amplitude during UV-radiation ( Lahav et al . , 2004; Batchelor et al . , 2011 ) . Akin to p53 , the yeast multi-stress response TF Msn2 exhibits short pulses of nuclear localization with intensity-dependent frequency under glucose limitation , but sustained nuclear localization with intensity-dependent amplitude under oxidative stress ( Hao et al . , 2013; Hao and O'Shea , 2012; Jacquet et al . , 2003; Petrenko et al . , 2013 ) . Thus , p53 and Msn2 dynamics encode both signal identity and signal intensity . Beyond p53 and Msn2 , amplitude- or frequency encoding of signal identity and intensity information is conserved throughout eukaryotic signaling pathways ( see also Berridge et al . , 2000; Werner et al . , 2005; Cai et al . , 2008; Warmflash et al . , 2012; Albeck et al . , 2013; Aoki et al . , 2013; Imayoshi et al . , 2013; Dalal et al . , 2014; Harima et al . , 2014 ) . Such encoding of signal identity and intensity information in TF activation dynamics has led to the hypothesis that TF target genes can reliably decode this dynamical information to elicit distinct gene expression programs with fine-tuned expression levels ( Figure 1A ) ( Behar et al . , 2007; Behar and Hoffmann , 2010; de Ronde and ten Wolde , 2014; Hansen and O'Shea , 2013; Levine et al . , 2013; Purvis and Lahav , 2013; Yosef and Regev , 2011 ) . 10 . 7554/eLife . 06559 . 003Figure 1 . Encoding and transmitting signal identity and intensity information in the dynamics of a single transcription factor ( TF ) . ( A ) Different signals ( e . g . , stress or ligand exposure ) can be encoded in the dynamics of a single TF . Signal identity is encoded in the type of TF dynamics: a sustained pulse ( signal A ) or nuclear pulsing ( signal B ) . Signal intensity ( e . g . , ligand concentration ) is encoded in the amplitude for signal A , but in the frequency for signal B . Different dynamical patterns of TF activation can activate distinct , but specific , downstream gene expression programs . ( B ) Applying an information theoretic framework to cell signaling , a gene promoter can be considered a channel . A graded population-level dose–response belies the complexity of the single-cell response: it shows the mean expression at points a , b , c , and d , but not the width or variance of their distributions . ( C ) Two extreme models . In the ‘rheostat model’ , signal intensity information encoded in the frequency or amplitude of a TF leads to non-overlapping gene expression distributions ( a , b , c , and d ) . Thus , by reading the gene expression output the cell can accurately determine the input signal intensity and high information transmission is achieved . Conversely , in the ‘noisy switch model’ , as a consequence of overlapping gene expression distributions ( a , b , c , and d ) information about signal intensity is permanently lost: the cell can distinguish ON/OFF ( signal identity ) , but the expression of a target gene cannot be fine-tuned to the stress intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 003 However , non-genetic cell-to-cell variability ( noise ) in gene expression limits the fidelity with which information can be decoded by TF target genes ( Coulon et al . , 2013; Sanchez and Golding , 2013 ) . This is important because the capacity of any signaling pathway for information transduction is limited by the capacity of its weakest node or bottleneck ( Cover and Thomas , 2006 ) . Thus , even though information can reliably be encoded in TF activation dynamics ( Selimkhanov et al . , 2014 ) , this information will be lost unless genes can reliably decode it . We therefore focus on the response of single genes and ask: can cells reliably transmit both signal identity and intensity information in the amplitude and frequency of TFs to target genes in the presence of biochemical noise ? In other words , what are the limits on amplitude- and frequency-mediated information transduction ? We investigate this by applying tools from information theory to quantify how much of the information ( in bits ) encoded in the amplitude and frequency of a TF can be transmitted through gene promoters to fine-tune the gene expression level . Originally developed by Claude Shannon for communication systems ( Shannon , 1948 ) , information theory has recently been applied to cell signaling ( reviewed in Tkacik and Walczak , 2011; Waltermann and Klipp , 2011; Nemenman , 2012; Rhee et al . , 2012; Bowsher and Swain , 2014; Levchenko and Nemenman , 2014; Mc Mahon et al . , 2014 ) . Mutual information quantifies how much information an output can carry about an input across a noisy channel ( Figure 1B ) . Mathematically , information is quantified in bits: to resolve two different signal intensities without error requires at least 1 bit of information , to resolve four different signal intensities without error requires at least 2 bits of information and so forth . However , 1 bit of information does not guarantee that two intensities can be distinguished without error . Similarly , 1 bit may allow multiple intensities to be distinguished , albeit with some associated error ( Bowsher and Swain , 2014 ) . As an example of how information theory can be applied , consider a dose–response relationship ( Figure 1B ) . A graded population-level dose–response can belie the complexity of the single-cell response ( Ferrell and Machleder , 1998 ) . For example , if different TF amplitudes or frequencies lead to distinguishable gene expression outputs ( points a , b , c and d ) , signal intensity information is accurately transmitted and the cell can fine-tune the expression of stress genes to the stress intensity like a ‘rheostat’ ( Figure 1C , rheostat model ) . However , biochemical noise can degrade signal information: if gene expression outputs are no longer resolvable , the cell can no longer fine-tune the expression level of stress genes to stress intensity ( Figure 1C , noisy switch model ) . In the noisy switch model , the cell can distinguish no stimulus ( point a , OFF ) from maximal stimulus ( point d , ON ) —but intermediate stimuli ( points b and c ) cannot reliably be distinguished based on the gene expression output and signal intensity information has been lost ( Figure 1C ) . Information theory provides a framework for capturing and quantifying these differences . Thus , we can distinguish these two models by measuring information transduction by promoters: the noisy switch model requires ∼1 bit , whereas the rheostat model requires substantially higher mutual information . Previous applications of information theory have been theoretical ( Ziv et al . , 2007; Tostevin and ten Wolde , 2009; Lestas et al . , 2010; de Ronde et al . , 2011; Bowsher and Swain , 2012; Rieckh and Tkacik , 2014 ) or have focused on upstream signaling and development ( Gregor et al . , 2007; Tostevin et al . , 2007; Skerker et al . , 2008; Tkacik et al . , 2008 , 2009; Mehta et al . , 2009; Cheong et al . , 2011; Dubuis et al . , 2013; Uda et al . , 2013; Selimkhanov et al . , 2014; Voliotis et al . , 2014 ) . However , despite gene expression being the final bottleneck in cell signaling , gene expression has received little attention ( Uda et al . , 2013 ) . Estimating an upper limit on the information transduction capacity of a gene has not previously been possible due to extrinsic noise: even when studying genetically identical single cells , the cells can exhibit non-genetic differences , that is , in cell cycle phase or variability in TF concentration , which means the measured mutual information will be an underestimate ( Elowitz et al . , 2002; Toettcher et al . , 2013 ) . Here , we overcome this limitation through a combined experimental and theoretical approach that corrects for extrinsic noise and allows us to estimate an upper limit on the information transduction capacity of individual Msn2 target genes . We combine high-throughput microfluidics to control the amplitude and frequency of Msn2 nuclear translocation with information theory to determine the information transduction capacity of Msn2 target genes . We find that Msn2 target genes can transduce just over 1 bit of information , consistent with the ‘noisy switch model’ . Although individual Msn2 target genes can only transduce little information , we illustrate how the cell can improve information transduction capacity by modulating promoter cis-elements , by integrating the response of more than one gene , or by having multiple copies of the same gene . We show that more information can be transduced through amplitude than through frequency modulation ( FM ) of Msn2 activation dynamics . Nevertheless , while previous studies have shown that significant amounts of information can be encoded in TF activation dynamics ( Selimkhanov et al . , 2014 ) , we find that noise in the decoding step severely limits information transduction . Specifically , our results indicate that information about signal identity , but not signal intensity , can be transmitted nearly without error in the amplitude and frequency of Msn2 and decoded by Msn2-responsive promoters . Information theory quantifies information transduction across a channel between a signal and a response ( Shannon , 1948; Cover and Thomas , 2006 ) . If a channel is noisy , a given signal input will give rise to a distribution of response outputs . This represents a loss of information since the signal input can no longer reliably be learned from observing the response output ( Figure 1B–C ) . A ‘black-box’-framework , information theory was originally developed for telecommunication channels , but it can also be applied to other ‘channels’ such as gene promoters or cell signaling pathways provided that the signal input ( here amplitude or frequency of Msn2 activation ) can be precisely controlled and the response output distribution precisely measured ( here single-cell gene expression ) . Mutual information , MI ( R;S ) , measured in bits , quantifies the amount of information about the signal input ( S ) that can be obtained by observing the response output ( R ) and , given discretized data , is defined as: ( 1 ) MI ( R;S ) =∑i , jp ( Ri , Sj ) log2 ( p ( Ri , Sj ) p ( Ri ) p ( Sj ) ) . The response distribution , p ( R ) , is the experimentally measured distribution of gene expression output . The signal distribution , p ( S ) , is the relative probability of each Msn2 amplitude or frequency . Since MI ( R;S ) depends on p ( S ) and since p ( S ) , that is , how often a cell might be exposed to a particular intensity of oxidative stress , is unknowable , hereafter we consider the maximal mutual information , I ( I ( R;S ) =maxp ( S ) [MI ( R;S ) ] ) which is the maximal amount of information that can be transduced through a ‘promoter channel’ . I can be thought of as a channel capacity , though a gene promoter is effectively a ‘single-use’ channel and I therefore has units of bits , whereas messages are sent repeatedly through a Shannon channel and , accordingly , the channel capacity has units of bits/s ( [Bowsher and Swain , 2014]; a detailed discussion is given in Supplementary file 2 ) . To measure how much information Msn2 target genes can transduce , we took advantage of a pharmacological method for controlling Msn2 nuclear localization using a small molecule , 1-NM-PP1 , ( Bishop et al . , 2000; Hao and O'Shea , 2012; Zaman et al . , 2009 ) and high-throughput microfluidics coupled to quantitative time-lapse microscopy ( Hansen et al . , 2015; Hansen and O'Shea , 2013 ) . With this setup ( Figure 2—figure supplement 1A; Video 1 ) , we can control and measure the amplitude and frequency of activation of an Msn2-mCherry fusion protein over time and generate single-cell traces that mimic the natural Msn2 dynamics under oxidative stress ( a sustained nuclear pulse with signal intensity–dependent amplitude; Figure 2A ) and glucose limitation ( short pulses with signal intensity–dependent frequency; Figure 2B ) ( Hao and O'Shea , 2012; Petrenko et al . , 2013 ) . To measure stress-relevant gene expression , we use dual cyan and yellow fluorescent protein ( CFP/YFP ) reporters and focus on two specific Msn2 target genes: HXK1 , which is induced under glucose limitation ( Herrero et al . , 1995 ) and SIP18 , which is induced in response to oxidative stress ( Rodriguez-Porrata et al . , 2012 ) . Using this setup , we have previously shown that , at the population level , individual genes differentially decode Msn2 dynamics ( Hansen and O'Shea , 2013; Hao and O'Shea , 2012 ) : oscillatory Msn2 activation induces gene class B ( e . g . , HXK1 ) without inducing gene class A ( e . g . , SIP18 ) , whereas sustained Msn2 activation preferentially induces gene class A ( Figure 1A ) . Thus , this represents an ideal setup for studying promoter decoding of Msn2 dynamics in single cells , which enables us to quantify information transduction . 10 . 7554/eLife . 06559 . 004Video 1 . A typical experiment . Mut B cells were grown in a microfluidic device and exposed to six 5-min Msn2 pulses separated by 10 min and phase contrast ( top left ) , Msn2-mCherry ( top right ) , CFP ( bottom left ) , and YFP ( bottom right ) reporter expression monitored . Video 1 consists of 64 frames at 2 . 5 min resolution and images have been compressed , cropped , and contrast adjusted , but not corrected for photobleaching . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 00410 . 7554/eLife . 06559 . 005Figure 2 . Information transduction by promoters with respect to amplitude and frequency modulation . ( A ) Cells containing either the hxk1::YFP or sip18::YFP reporter were exposed to either no activation or a 70-min pulse of seven increasing amplitudes from ca . 25% ( 100 nM 1-NM-PP1 ) to 100% ( 3 μM 1-NM-PP1 ) of maximal Msn2-mCherry nuclear localization and single-cell gene expression monitored . For each single-cell time-trace , YFP concentration is converted to a scalar by taking the maximal YFP value after smoothing . For each Msn2-mCherry input ( a fit to the raw data is shown on the left ( AM: Msn2 input ) ) , the gene expression distribution is plotted as a histogram of the same color on the right for HXK1 and SIP18 . The population-averaged dose–response ( top ) is obtained by calculating the YFP histogram mean for each Msn2 input condition . ( B ) Cells containing either the hxk1::YFP or sip18::YFP reporter were exposed to either no activation or from one to nine 5-min pulses of Msn2-mCherry nuclear localization ( ca . 75% of maximal nuclear Msn2-mCherry , 690 nM 1-NM-PP1 ) at increasing frequency . All calculations were performed as in ( A ) . ( C ) Cells containing either the pSIP18 mut A::YFP reporter or the pSIP18 mut B::YFP reporter were exposed to amplitude modulation ( AM ) as in ( A ) . ( D ) Cells containing either the pSIP18 mut A::YFP reporter or the pSIP18 mut B::YFP reporter were exposed to frequency modulation ( FM ) as in ( B ) . Maximal mutual information , I , and its error are calculated as described in Supplementary file 2 . Full details on data processing are given in ‘Materials and Methods’ . Each plot of an Msn2 input pulse and YFP expression is based on data from ca . 1000 cells from at least three replicates . All raw single-cell time-lapse microscopy source data for HXK1 ( 15 , 259 cells ) , SIP18 ( 21 , 242 cells ) , pSIP18 mut A ( 18 , 203 cells ) , and pSIP18 mut B ( 17 , 655 cells ) for this Figure are available online as Supplementary file 1 and in ( Hansen and O'Shea , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 00510 . 7554/eLife . 06559 . 006Figure 2—Figure supplement 1 . How time-lapse data are converted to histograms and promoter maps and noise data . ( A ) Overview of strains . To visualize and quantify the subcellular localization of Msn2 it was C-terminally tagged with the red fluorescent protein mCherry . A nuclear protein , NHP6a , was C-terminally tagged with iRFP , an infrared fluorescent protein ( Filonov et al . , 2011; Hansen and O'Shea , 2013 ) , to visualize the nucleus for segmentation purposes . All three catalytic subunits of Protein Kinase A ( PKA ) were mutated to contain an analogue-sensitive M→G mutation ( TPK1M164G TPK2M147G TPK3M165G ) . These mutations render all three PKA subunits sensitive to the small molecule 1-NM-PP1 ( Bishop et al . , 2000; Hao and O'Shea , 2012; Zaman et al . , 2009 ) . Thus , when 1-NM-PP1 is added , PKA is inhibited , Msn2-mCherry is no longer phosphorylated by PKA , Msn2-mCherry therefore gets dephosphorylated , and translocates into the nucleus where it can bind to and activate target genes . To visualize gene expression , the ORFs of target genes were replaced with YFP ( mCitrineV163A ) and CFP ( SCFP3A ) on homologues chromosomes in diploid cells , as it has been described previously ( Hansen et al . , 2015; Hansen and O'Shea , 2013; Kremers et al . , 2006 ) . The inhibitor , 1-NM-PP1 is shown on the right and its synthesis has been described previously ( Hansen and O'Shea , 2013 ) . ( B ) An illustration of how the YFP histograms are obtained for each condition . For a specific amplitude or frequency , the response of ∼1000 cells is measured ( only ∼300 cells shown here for ease of visualization ) . For each single cell time-trace , a moving average smoothing filter is applied to remove any technical noise and the maximal YFP value is determined after the trace has reached a plateau . This is repeated for all single cells and a YFP histogram is generated by binning . The procedure is then repeated for all the Msn2 conditions ( e . g . , the no input and all the AM conditions ) to generate a full single-cell dose–response ( right ) . These data are then used to calculate the maximal mutual information with respect to amplitude modulation , IAM . ( C ) Promoter nucleosome occupancy maps . The upstream promoter region ( −800 to 0 bp from ATG site ) is shown for each promoter . Msn2 binding sites ( STRE 5′-CCCCT-3′ ) are shown in red triangles and nucleosome occupancy data ( grey ) are from ( Hansen and O'Shea , 2013 ) . The SIP18 promoter has three Msn2 binding sites . The most upstream site is seemingly non-functional—removing it does not affect gene induction . The two sites ( close to −400 bp ) are required—removing these two sites abolishes gene induction . pSIP18 mut A and mut B have three and four new binding sites , respectively , in between the two nucleosomes close to the transcription start site . ( D ) Dynamic range and noise . Removing the two WT Msn2 binding sites and replacing them with three or four binding sites , respectively , substantially increases the dynamic range ( defined as the response to a 70 min pulse at 3 μM 1-NM-PP1 ) . How the total ( red ) , intrinsic ( blue ) and extrinsic ( green ) noise scales with the Msn2 amplitude ( for a 70 min pulse; top ) or the frequency ( at 690 nM 1-NM-PP1; bottom ) for all four promoters is shown . The y-axis scale is different in each case . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 00610 . 7554/eLife . 06559 . 007Figure 2—Figure supplement 2 . Data processing and control of measurement noise . ( A ) Data processing illustration . Controlling measurement noise is important , because high measurement noise will cause measurements of mutual information to be underestimates . To minimize effects of measurement noise coming from , for example , improper focusing by the microscope , autofluorescence and camera noise , slight errors in cell segmentation and other sources , multiple YFP measurements are made . For each single cell , the YFP level is measured 64 times at 2 . 5 min time resolution . In general , measurement noise is modest at very low YFP expression—in part due to cellular autofluorescence—but negligible at high YFP expression . As an example of very low YFP expression , a single cell time-trace is shown on the left ( SIP18 , 70 min , 100 nM 1-NM-PP1 ) . By smoothing the raw YFP data ( black circles ) , an accurate estimation of the YFP level can be obtained ( red line ) . As an example of very high YFP expression , a raw and smoothed single cell time-trace is shown on the right ( mut B , 70 min , 3 μM 1-NM-PP1 ) . ( B ) Example of raw data at very low YFP expression ( SIP18 , 70 min , 100 nM 1-NM-PP1 ) . Raw YFP time-traces of 100 randomly chosen single cells are shown on the left and the same YFP time-traces , after smoothing as illustrated in ( A ) , are shown on the right . Although the raw YFP data suffer from modest measurement noise , the actual YFP level can be accurately estimated by smoothing . ( C ) Example of raw data at low YFP expression ( mut A , 70 min , 100 nM 1-NM-PP1 ) . Raw YFP time-traces of 100 randomly chosen single cells are shown on the left and the same YFP time-traces , after smoothing as illustrated in ( A ) , are shown on the right . ( D ) Example of raw data at high YFP expression ( mut B , 70 min , 3 μM 1-NM-PP1 ) . Raw YFP time-traces of 100 randomly chosen single cells are shown on the left and the same YFP time-traces , after smoothing as illustrated in ( A ) , are shown on the right . Furthermore , all raw single-cell time-trace data for HXK1 ( 15 , 259 cells ) , SIP18 ( 21 , 242 cells ) , pSIP18 mut A ( 18 , 203 cells ) , pSIP18 mut B ( 17 , 655 cells ) , 1× reporter diploid ( 21 , 236 cells ) , and 2× reporter diploid ( 19 , 222 cells ) are available as Supplementary file 1 and in ( Hansen and O'Shea , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 007 To measure information transduction through the HXK1 and SIP18 promoters with respect to amplitude modulation ( IAM ) , we exposed thousands of cells to increasing amplitudes of a 70 min Msn2 pulse to mimic oxidative stress , measured the single-cell distribution of responses for each amplitude with minimal measurement noise ( Figure 2—figure supplement 2 ) , and determined the population-averaged dose–response ( Figure 2A; all raw single-cell data are available online as Supplementary file 1 and in ( Hansen and O'Shea , 2015 ) ; see also Figure 2—figure supplement 1B ) . We quantify gene expression as the maximal YFP concentration after the YFP time-trace has reached a plateau ( ‘Materials and Methods’ ) . Surprisingly , for both HXK1 and SIP18 , IAM was 1 . 2–1 . 3 bits—enough to distinguish ON from OFF without error ( the ‘no Msn2 input’ and the ‘brown’ distributions are clearly distinguishable; Figure 2A ) , but with limited ability to distinguish signal intensities . One way to think about this result is to ask , given the HXK1 YFP expression output , how much information does that provide about the input amplitude ? For example , considering the HXK1 AM histograms in Figure 2A , for most YFP outputs the cell can exclude the ‘no Msn2 input’ condition , but appears to be unable to discern which of the other amplitudes it was exposed to without a high error rate . Consequently , HXK1 and SIP18 can distinguish no stress from high oxidative stress ( high Msn2 amplitude ) without error , but cannot accurately transmit information about stress intensity . Next , we measured information transduction of HXK1 and SIP18 with respect to frequency modulation ( IFM ) using 5-min Msn2 pulses at frequencies similar to those observed under glucose limitation ( Figure 2B ) . Even though HXK1 is physiologically induced during Msn2 pulsing , IFM was only 1 . 11 bits—again enough for distinguishing ON from OFF essentially without error like a ‘noisy switch’ , but insufficient to accurately fine-tune the HXK1 expression level to each Msn2 frequency like a ‘rheostat’ . SIP18 , required only under oxidative stress , largely filters out Msn2 pulsing and therefore has a negligible IFM . It is generally assumed that gene expression levels are fine-tuned ( de Nadal et al . , 2011 ) , but the very low IAM and IFM of HXK1 and SIP18 are incompatible with this idea . One possibility is that mutual information for promoters is biophysically constrained to ∼1 . 0–1 . 3 bit , but another possibility is that HXK1 and SIP18 are not optimized for AM- and FM-mediated information transduction . To investigate this and explore the relationship between promoter cis-elements and information transduction , we focused on SIP18 , which has the lowest I and suffers from high gene expression noise ( Figure 2—figure supplement 1D ) , and asked if altering promoter architecture could improve information transduction . We removed the two functional Msn2 binding sites in the SIP18 promoter and added three and four new binding sites in the nucleosome-free region closer to the transcription start site ( promoter architecture maps are shown in Figure 2—figure supplement 1C ) to generate pSIP18 mut A and pSIP18 mut B , which differ from the wild-type SIP18 promoter by 14 and 18 nucleotides , respectively . We then repeated the experiments for mut A and mut B to measure their IAM and IFM . With respect to AM , both mutants had significantly higher IAM of 1 . 42 bits ( mut A ) and 1 . 55 bits ( mut B ) ( Figure 2C ) . We attribute this increase to a combination of three factors: a more linear dose–response , a higher dynamic range , and significantly lower gene expression noise ( Figure 2—figure supplement 1D ) . The wild-type SIP18 promoter filters out oscillatory input and therefore has a negligible IFM . In contrast , with respect to FM mut A shows a slightly higher IFM of 0 . 88 bits and mut B a significantly higher IFM of 1 . 39 bits ( Figure 2D ) . Notably , although HXK1 presumably evolved to decode Msn2 pulsing , as is observed under glucose limitation , mut B now shows a higher IFM than even HXK1 . Although I could be different for natural Msn2 dynamics ( Hao and O'Shea , 2012 ) , these results show that for the AM and FM signals studied here , natural Msn2 target genes are not optimized for information transduction and do not have their maximal I even though promoters with higher IAM/FM are only a few mutations away . Furthermore , IAM exceeds IFM for all four promoters , which shows that , at least in these four cases , transmitting gene expression information in the amplitude of TF activation dynamics is more reliable than transmitting it in the frequency . Thus , the promoter information transduction capacity is tunable in cis: by modulating Msn2 binding sites , we can control both how a promoter decodes Msn2 dynamics and how much information it can transmit . Natural Msn2 target promoters appear to have I ≤ 1 . 3 bits . Thus , we observe high information loss during gene expression . Information loss comes from two sources: gene-intrinsic and gene-extrinsic noise ( Elowitz et al . , 2002 ) . Intrinsic noise originates from the inherently stochastic nature of biochemical reactions , such as stochastic binding of Msn2 at individual promoters . Information loss due to intrinsic noise is therefore unavoidable for the cell . Extrinsic noise comes from the intracellular environment , which may differ between cells in a population . Even though we consider genetically identical cells grown in a microfluidic chemostat , the cell population could exhibit non-genetic differences in cell-cycle phase and Msn2 abundance or dynamics , etc . This could cause the dose–response to be different between single cells ( Figure 3A ) , as was observed in a recent study on Ras/ERK signaling ( Toettcher et al . , 2013 ) . For example , a cell with a higher-than-average Msn2 abundance might show higher gene expression . When we carefully quantify Msn2-mCherry dynamics , we observe loss of information between the microfluidic 1-NM-PP1 input and nuclear Msn2 due to variability in Msn2 abundance between cells ( Figure 3—figure supplement 1 ) . Likewise , the cell cycle is a major source of extrinsic gene expression noise ( Zopf et al . , 2013 ) . Therefore , measuring mutual information in a cell population subject to extrinsic noise , as we did in Figure 2 , underestimates the intrinsic information transduction capacity of a promoter . 10 . 7554/eLife . 06559 . 008Figure 3 . An algorithm for estimating intrinsic mutual information . ( A ) Genetically identical cells can have shifted single-cell dose-responses due to gene-extrinsic effects , such as variation in Msn2 abundance and cell cycle phase . Measuring the response of a single reporter ( YFP ) therefore underestimates mutual information . By introducing an additional reporter ( CFP ) , we can distinguish extrinsic noise such as a shifted dose–response since this affects both CFP and YFP equally , from true intrinsic stochasticity . ( B ) Overview of algorithm . By fitting a gamma distribution to the raw YFP data , calculating the CFP/YFP covariance and filtering this component out of the total variance , an intrinsic YFP distribution can be estimated ( left ) . By repeating this for each dose–response distribution , intrinsic mutual information can be estimated ( right ) . Full details on the algorithm are given in Supplementary file 2 . ( C ) By applying the algorithm to the data from Figure 2 ( solid bars ) , we can estimate intrinsic mutual information ( hatched bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 00810 . 7554/eLife . 06559 . 009Figure 3—Figure supplement 1 . Input noise and variability in Msn2 abundance . ( A ) Variability in Msn2 abundance . One source of noise in our system is non-genetic cell-to-cell variability in Msn2 abundance . Msn2 is a low-abundance protein: there are only a few hundred molecules in each cell ( Ghaemmaghami et al . , 2003 ) . Therefore , precisely measuring Msn2 abundance is challenging . Furthermore , the nucleus moves in and out of focus during time-lapse acquisition . To estimate the variation in Msn2 abundance , cells ( pSIP18 mut B ) were grown in the microfluidic device and exposed to a 70-min pulse of either 0 , 100 nM , 175 nM , 275 nM , 413 nM , 690 nM , 1117 nM or 3 μM 1-NM-PP1 . Msn2-mCherry nuclear localization was measured using a 5-frame z-stack series of 0 , ±1 . 2 μm , ± 2 . 4 μm above and below the focal plane using a 500 ms exposure time and imaging every 10 min . Msn2-mCherry fluorescence was corrected for photobleaching . We collected two frames before and after 1-NM-PP1 exposure to calculate the baseline level of Msn2 before 1-NM-PP1 treatment . In ( A ) , we show the mean and standard deviation for each timepoint for each concentration . ( B ) To calculate mutual information between 1-NM-PP1 input and Msn2-mCherry dynamics , we use the data from ( A ) and calculate IAM ( 1-NM-PP1; Msn2 ) = 2 . 06 ± 0 . 03 bits . We quantify Msn2-mCherry localization in absolute units as the mean nuclear Msn2 level across the seven measurements while Msn2 is nuclear—this also corresponds to the total time-integrated nuclear level of Msn2 ( Msn2 ‘Area Under the Curve’ or AUC ) . In total , we measured 2996 single cells . Using Msn2 variability in response to 3 μM 1-NM-PP1 , we estimate the cell-to-cell variability of Msn2 to be CV∼15% . However , given measurement noise we stress that CV∼15% and I∼2 . 06 bits are likely over- and underestimates , respectively . Note that Msn2 is a low abundance protein ( Ghaemmaghami et al . , 2003 ) . Previous proteomic studies showed that essentially no yeast proteins have CV<10% ( Newman et al . , 2006 ) . Therefore , Msn2 is among the least variable low abundance proteins in yeast . ( C ) This figure is plotted using data from Figure 2 for HXK1 ( Hansen and O'Shea , 2015 ) . In red is shown the input and in black is shown traces from 10 representative single cells . We did not do a finely spaced z-stack series for this experiment , which is necessary to accurately quantify the concentration of Msn2 in the nucleus—this causes too much photobleaching to be compatible with imaging at reasonable temporal resolution ( 2 . 5 min here ) . Nonetheless , as can be seen , the black traces faithfully track the input with limited noise . For each cell plotted above , we also measured hxk1::CFP and hxk1::YFP gene expression . ( D ) To accurately quantify Msn2-mCherry dynamics during FM input , we acquired a finely spaced z-stack series at high time-resolution ( 1 min ) . This causes too high photobleaching to be compatible with sustained time-lapse imaging . Therefore , we are only able to collect data at this resolution for a single 5-min pulse . The mean ( black dots ) and standard deviation ( error bars ) for 132 single cells ( pSIP18 mut B ) are shown . As can be seen , Msn2-mCherry accurately tracks the microfluidic 1-NM-PP1 input with limited noise also during FM input . In a population of cells , Msn2 translocates to the nucleus in every single cell during 1-NM-PP1 exposure . ( F ) To calculate mutual information between 1-NM-PP1 input and Msn2-mCherry dynamics , we use the data from ( D ) and estimate IFM ( 1-NM-PP1; Msn2 ) = 2 . 23 ± 0 . 03 bits . Given measurement noise , this is likely an underestimate . We quantify Msn2-mCherry localization in absolute units as the total nuclear Msn2 level across the ten measurements while Msn2 is nuclear ( five during the pulse , five after the pulse ) —this also corresponds to the total time-integrated nuclear level of Msn2 ( Msn2 ‘Area Under the Curve’ or AUC ) . With data from ( D ) , we measure the distribution of cell-to-cell variability for a single 5-min pulse . To calculate IFM , we then extrapolate by multiplying the AUC probability distribution by the pulse number of each experiment since Msn2 tracks the 1-NM-PP1 input as faithfully for the first pulse as for the subsequent pulses . Ideally , one would measure the Msn2 AUC at 1-min time resolution and with finely spaced z-stacks throughout the entire time-lapse experiment , but this is not technically possible due to photobleaching . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 009 Although it is in principle possible to correct for cell cycle phase , Msn2 abundance and other gene-extrinsic factors individually , it is impossible to correct for all factors . To overcome this limitation and estimate the intrinsic I ( Iint ) , we developed a method based on the dual-reporter approach ( Elowitz et al . , 2002; Swain et al . , 2002; Hilfinger and Paulsson , 2011 ) . By having two gene expression reporters in diploid cells on homologous chromosomes that differ only by their color ( CFP and YFP ) but share the same intracellular environment , the extent to which they co-vary in the same cell allows us to infer how much gene-extrinsic factors , such as cell-cycle phase and Msn2 variability , and so on . contribute altogether ( extrinsic noise ) , without having to specify each factor . Or phrased differently , if the dose–response is shifted in a cell , both the CFP and YFP reporter will be affected in a correlated manner and their covariance allows us to quantify this ( Figure 3A ) . Therefore , we developed an algorithm that uses the CFP/YFP covariance to estimate what the intrinsic I ( Iint ) would have been in the absence of extrinsic noise . Briefly , our algorithm takes the following steps ( Figure 3B ) : First , the raw YFP histogram is fitted to a gamma distribution ( YFP∼Γ ( a , b ) ) . Second , the extrinsic component ( covariance ) of the total variance is determined ( σext2=〈CFP·YFP〉−〈CFP〉〈YFP〉 ) . Third , keeping the mean constant , a new gamma distribution without the extrinsic component is inferred ( YFPint∼Γ ( aint , bint ) ) . Fourth , this is repeated for each Msn2 input ( e . g . , amplitude or frequency ) . Finally , this inferred data set is discretized and then used to estimate Iint ( Figure 3B; see Supplementary file 2 for a detailed discussion of the algorithm ) . We verified our algorithm in silico by systematically simulating five linear and five non-linear gene expression models with and without extrinsic noise and compared the true Iint to the algorithm-inferred Iint . The algorithm tended to slightly underestimate the true Iint , but the mean error was less than 2% and the error was always less than 5% ( Supplementary file 2 ) . Therefore , by using dual-reporter strains we can determine how much of the information loss is extrinsic , apply the algorithm and estimate Iint in each case ( IAM , int and IFM , int ) . We find that filtering out extrinsic noise significantly increases I ( hatched bars , Figure 3C ) . Since the cell most likely incorporates some gene-extrinsic factors into a decision , but most likely does not incorporate all gene-extrinsic factors , we interpret Iraw and Iint as a lower and upper bound , respectively , on the true I . Thus , our approach allows us to estimate an upper bound , Iint , on a promoter's information transduction capacity . Even after correcting for extrinsic noise , IAM , int for HXK1 and SIP18 only reach ∼1 . 5–1 . 6 bits ( Figure 3C ) . And IFM , int for HXK1 is just 1 . 36 bits—that is , three ranges of inputs can only be distinguished with some associated error . Thus , even when considering Iint , which is the upper limit on the maximal mutual information , neither natural Msn2 target gene can transmit information about stress intensity without some error . That is , consistent with the ‘noisy switch model’ , expression of HXK1 and SIP18 is not reliably fine-tuned to stress intensity . In contrast , for mut B , IFM , int is 1 . 55 bits and IAM , int is ∼2 bits ( Figure 3C ) . Thus , mut B almost approaches a range where information about both signal identity and intensity could conceivably be transduced nearly without error like a ‘rheostat’ , though the natural Msn2 target genes , HXK1 and SIP18 , do not . Filtering out extrinsic noise substantially increases I ( Figure 3C ) . Next , we considered how reducing intrinsic noise might increase I . In principle , as the number of gene copies increases , information loss due to intrinsic noise decreases due to simple ensemble averaging and mutual information increases—in the limit of infinite copies , intrinsic noise is zero and all information loss is due to extrinsic noise ( Cheong et al . , 2011 ) . To test this we generated diploid strains with either one ( 1× ) or two ( 2× ) copies of the hxk1::CFP and sip18::YFP reporters in the same cell . We repeated the AM and FM experiments for the 1× and 2× diploids ( Figure 4—figure supplement 1 and Figure 4—figure supplement 2 ) . Comparing the 1× and 2× diploids ( Figure 4A ) , we see that having two copies of a gene generally improves I by ∼0 . 05–0 . 20 bits . For example , on going from haploid ( 1× ) to diploid ( 2× ) , HXK1 IAM increases from 1 . 30 to 1 . 47 bits . Therefore , in terms of information transduction , being diploid confers a small but robust advantage . 10 . 7554/eLife . 06559 . 010Figure 4 . Integrating the response of more than one gene improves information transmission . ( A ) The AM and FM experiments ( Figure 2 ) were repeated for diploid strains containing either one copy ( 1× ) of the hxk1::CFP and sip18::YFP reporters or two copies ( 2× ) of the hxk1::CFP and sip18::YFP reporters and individual and joint mutual information determined ( full details on calculations are given in Supplementary file 2 ) . ( B ) 2× sip18::YFP vs 2× hxk1::CFP scatterplot showing expression for three experiments: no input ( light purple ) , five 5-min pulses of 690 nM 1-NM-PP1 separated by 13-min intervals ( orange ) or one 70-min pulse of 3 μM 1-NM-PP1 ( green ) . For each condition , 600 cells are shown . The YFP/CFP expression is the maximal value after each time-trace has reached a plateau . The inset shows a zoom-in highlighting the ‘no input’ condition . All raw single-cell time-lapse microscopy source data for the 1× reporter diploid ( 21 , 236 cells ) and 2× reporter diploid ( 19 , 222 cells ) for this Figure are available online as Supplementary file 1 and in ( Hansen and O'Shea , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 01010 . 7554/eLife . 06559 . 011Figure 4—Figure supplement 1 . Summary of results for 1× reporter diploid . This figure shows single and joint distribution histograms for the 1× reporter diploid ( sip18::YFP hxk1::CFP ) . Top panel , left: Cells containing both the sip18::YFP and hxk1::CFP reporters were exposed to either no activation or a 70-min pulse of seven increasing amplitudes from ca . 25% ( 100 nM 1-NM-PP1 ) to 100% ( 3 μM 1-NM-PP1 ) of maximal Msn2-mCherry nuclear localization and single-cell gene expression was monitored . For each single-cell time-trace , YFP expression is converted to a scalar by taking the maximal YFP value after smoothing . For each Msn2-mCherry input ( a fit to the raw data is shown on the left [AM: Msn2 input] ) , the gene expression distribution is plotted as a histogram of the same color on the right for HXK1 and SIP18 . The population-averaged dose–response ( top ) is obtained by calculating the YFP histogram mean for each Msn2 input condition . Top panel , right: Cells containing both the sip18::YFP and hxk1::CFP reporters were exposed to either no activation or from one to nine 5-min pulses of Msn2-mCherry nuclear localization ( ca . 75% of maximal nuclear Msn2-mCherry , 690 nM 1-NM-PP1 ) at increasing frequency . All calculations were performed as described above . Middle panel: The discretized joint AM distribution is shown with sip18::YFP on the y-axis and hxk1::CFP on the x-axis . The color of each bin corresponds to the probability—dark blue means unoccupied and red corresponds to the highest probability . The single-cell time-traces were converted to scalars as illustrated in Figure 2—figure supplement 1B . Each individual subplot corresponds to a different condition ( Msn2 amplitude ) and the data have been binned such that the low expression bins are much smaller and therefore harder to see on the plot . Bottom panel: Same as for the joint AM distribution in the middle panel except for the joint FM distribution . Each subplot now corresponds to a specific frequency ( and thus number of pulses ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 01110 . 7554/eLife . 06559 . 012Figure 4—Figure supplement 2 . Summary of results for 2× reporter diploid . This figure shows single and joint distribution histograms for the 2× reporter diploid ( 2× sip18::YFP 2× hxk1::CFP ) . Top panel , left: Cells containing both the 2× sip18::YFP and 2× hxk1::CFP reporters were exposed to either no activation or a 70-min pulse of seven increasing amplitudes from ca . 25% ( 100 nM 1-NM-PP1 ) to 100% ( 3 μM 1-NM-PP1 ) of maximal Msn2-mCherry nuclear localization and single-cell gene expression monitored . For each single-cell time-trace , YFP expression is converted to a scalar by taking the maximal YFP value after smoothing . For each Msn2-mCherry input ( a fit to the raw data is shown on the left [AM: Msn2 input] ) , the gene expression distribution is plotted as a histogram of the same color on the right for HXK1 and SIP18 . The population-averaged dose–response ( top ) is obtained by calculating the YFP histogram mean for each Msn2 input condition . Top panel , right: Cells containing both the 2× sip18::YFP and 2× hxk1::CFP reporters were exposed to either no activation or from one to nine 5-min pulses of Msn2-mCherry nuclear localization ( ca . 75% of maximal nuclear Msn2-mCherry , 690 nM 1-NM-PP1 ) at increasing frequency . All calculations were performed as described above . Middle panel: The discretized joint AM distribution is shown with 2× sip18::YFP on the y-axis and 2× hxk1::CFP on the x-axis . The color of each bin corresponds to the probability—dark blue means unoccupied and red corresponds to the highest probability . The single-cell time-traces were converted to scalars as illustrated in Figure 2—figure supplement 1B . Each individual subplot corresponds to a different condition ( Msn2 amplitude ) and the data have been binned such that the low expression bins are much smaller and therefore harder to see on the plot . Bottom panel: Same as for the joint AM distribution except for the joint FM distribution . Each subplot now corresponds to a specific frequency ( and thus number of pulses ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 012 So far we have considered information transduction from Msn2 to a single gene . Yet , Msn2 controls the expression of hundreds of genes in response to different stresses ( Elfving et al . , 2014; Hao and O'Shea , 2012; Huebert et al . , 2012 ) , we therefore extend our approach to information transduction from Msn2 to multiple genes . We next asked whether one way the cell might overcome the low I of individual genes would be to integrate the response of two or more different genes . To simulate and test this , we used diploid strains with both hxk1::CFP and sip18::YFP in the same cell , which allows us to measure the joint mutual information , I ( R1 , R2;S ) . We find that the AM joint mutual information ( IAM , joint ) is significantly higher in both the 1× and 2× cases than the individual IAM of HXK1 and SIP18 ( Figure 4A ) . For example , the total joint mutual information ( IAM+FM , joint; combining both the AM and FM responses ) is 1 . 67 bits and 1 . 83 bits for the 1× and 2× diploids , respectively ( Figure 4A ) . Therefore , although HXK1 and SIP18 individually can only distinguish ON from OFF without error ( Figure 2A ) , their joint response can distinguish three inputs ( no input , FM , or AM ) nearly without error ( Figure 4B ) . Thus , these results show that although the information transduction capacities of individual genes may be low , by integrating the response of two different genes the cell can improve information transduction . Therefore , by integrating the response of even more than two genes , the cell could potentially substantially improve the information transduction capacity of a pathway . Here , we use information theory to investigate the hypothesis that cells can transduce both signal identity and signal intensity information in the amplitude and frequency of TF activation dynamics to control gene expression . As a conceptual framework , we introduce two extreme models of information transmission ( Figure 1C ) : in the ‘noisy switch model’ , the cell only transmits information sufficient to turn ON or OFF particular genes or pathways in response to external signals or stresses , whereas in the ‘rheostat model’ the cell is accurately fine-tuning the expression levels of relevant genes to the intensity of a signal or stress . For a TF responding to multiple stresses , we can extend this framework beyond a single gene . Extending the noisy switch model to two genes , the stress-relevant gene HXK1 is reliably induced during FM pulsing of Msn2 ( as seen under glucose limitation ) , whereas both HXK1 and the stress-relevant SIP18 gene are reliably induced during AM activation of Msn2 ( as seen under oxidative stress ) ( Figure 4B ) . Therefore , three inputs ( no input , FM , or AM ) can be distinguished essentially without error ( Figure 4B ) . However , given the modest joint information transduction capacities with respect to AM and FM combined ( IAM+FM , joint; Figure 4A ) , the cell cannot fine-tune HXK1 and SIP18 expression levels without significant error to the stress intensity . Thus , signal identity information for two distinct stresses can be transduced in the amplitude and frequency of Msn2 essentially without error , but intensity information can only be transduced with high error . A central result in information theory is that the information transduction capacity of a signaling pathway is limited by and equal to the capacity of its weakest node or bottleneck ( see also Supplementary file 2 for a discussion ) . In other words , once information has been lost , no amount of post-processing can recover it , as is seen in the game of ‘broken telephone’ . Therefore , by measuring information transduction of individual Msn2 target genes to be ∼1 . 0–1 . 3 bits , we can establish that the expression of Msn2 target genes cannot transduce stress signal intensity information without significant error at least for the AM and FM signals studied here—we can draw this conclusion without knowing all the relevant upstream components of the signaling pathway , how they mechanistically interact and how much information they can transmit . Thus , this approach can provide insight into the purpose of a pathway ( e . g . , noisy switch vs rheostat ) and can readily be applied to other signaling pathways . Why does information transduction by Msn2 resemble a ‘noisy switch’ rather than a ‘rheostat’ ? Or phrased differently , why should the cell not fine-tune the expression level of stress genes to the stress intensity ? One possibility is that the stochasticity inherent in the biophysical process of transcription fundamentally constrains information transduction by a promoter to ∼1 . 0–1 . 3 bit . However , since the information transduction capacity of SIP18 can be substantially increased by modulating promoter cis-elements ( Figures 2 and 3 ) , the low I of natural Msn2 target genes is not solely due to inherent biophysical constraints . Another speculative possibility is that variability is selected for: since evolutionary selection works at the population-level , variability in gene expression can create phenotypic diversity within an isogenic population ( Balaban et al . , 2004; Blake et al . , 2006 ) . It is also important to note that under natural stress a network of factors could be activated , whereas here we study the limits on amplitude- and frequency-mediated transduction of gene expression information in the dynamics of a single master TF . Many biological signaling pathways transmit information through the amplitude or frequency of a shared signaling molecule ( Figure 1A ) and this has raised the long-standing question: can more information be transmitted through the amplitude or the frequency of a signaling molecule ( Rapp et al . , 1981; Li and Goldbeter , 1989 ) ? This question has not previously been experimentally addressed for TFs responding to multiple signals in an amplitude- or frequency-dependent manner . We show that more gene expression information can be transduced through the amplitude than through the frequency of Msn2 activation dynamics for all genes studied here ( Figures 2 and 3 ) . Although the FM dose-responses tend to be more linear , the AM dose-responses have higher dynamic range and lower noise ( Figure 2 and Figure 2—figure supplement 1D ) . While we show that gene promoters have higher information transduction capacities for amplitude- than frequency-encoded information ( Figures 2 and 3 ) , maximal information transduction can be achieved for TFs that exhibit both amplitude- and frequency-encoding ( Figure 4 ) . The amount of information promoters measured in this study can transmit is limited ( Figures 2–4 ) ; yet we stress that for many ‘house-keeping’ genes or genes expressed at steady-state information transduction is likely significantly higher , in part due to time-averaging . Indeed , the gene expression response to a transient signal is noisier than a response at steady-state ( Hansen and O'Shea , 2013 ) and inducible genes tend to show higher expression noise ( Bar-Even et al . , 2006; Newman et al . , 2006 ) . One way the cell can improve information transduction is by integrating the response of more than one gene or by having multiple copies of a gene ( Figure 4 ) . An example of this is ribosome biogenesis where , by having multiple copies of each gene encoding a subunit and employing elaborate feedback control , the cell can fine-tune its translational capability to its growth and energy status ( Lempiainen and Shore , 2009 ) . Another example is morphogen or cytokine secretion: although the amount produced by each single cell might be noisy , the average amount produced by a large number of cells can be highly precise ( Gregor et al . , 2007; Cheong et al . , 2011 ) . Hence , a number of strategies for increasing information transmission exist . In conclusion , we have investigated the reliability of transmitting gene expression information in the amplitude and frequency of a TF . We show that the information transduction capacity of a gene can be tuned in cis and the amount of information transmitted increased by integrating the response of multiple genes . Nonetheless , for individual genes our results are consistent with the Msn2 pathway transmitting essentially error-free signal identity information , but unreliable signal intensity information , and therefore functioning more like a ‘noisy switch’ than a ‘rheostat’ . Since many similar master regulators , such as p53 , NF-κB , ERK , and Hes1 , also transduce information through the regulation of signaling dynamics , it will be interesting to investigate whether dynamic cell signaling is generally limited to error-free transduction of only signal identity information . Microscopy experiments were performed essentially as described previously ( Hansen et al . , 2015; Hansen and O'Shea , 2013 ) . Briefly , yeast cells were grown overnight at 30°C with shaking at 180 rpm to an OD600 nm of ca . 0 . 1 in low fluorescence medium , quickly collected by suction filtration , loaded into the five channels of a microfluidic device pretreated with concanavalin A and the setup mounted on a Zeiss Axio Observer Z1 inverted fluorescence microscope ( Carl Zeiss , Jena , Germany ) equipped with an Evolve EM-CCD camera ( Photometrics , Tuscon , AZ ) , 63× oil-immersion objective ( NA 1 . 4 , Plan-Apochromat ) , Zeiss Colibri LEDs for excitation and an incubation chamber kept at 30°C . Solenoid valves programmed using custom-written software ( MATLAB ) control whether medium with or without 1-NM-PP1 is delivered to each microfluidic channel and the flow ( ca . 1 μL/s ) is driven by gravity . Control of 1-NM-PP1 delivery enables the control of Msn2 pulsing ( Figure 2 ) and a unique pulse sequence can be delivered to each of the five microfluidic channels . The microscope maintains focus and moves between each channel to acquire phase-contrast , YFP , CFP , RFP ( mCherry ) , and iRFP images for 64 frames with a 2 . 5 min time resolution . For the AM experiments , 1-NM-PP1 was added to each microfluidic channel for 70 min at the following concentrations: 100 nM , 175 nM , 275 nM , 413 nM , 690 nM , 1117 nM , 3 μM . For the FM experiments , a concentration of 690 nM 1-NM-PP1 was used together with the following pulse sequences: one 5-min pulse; two 5-min pulses separated by a 40-min interval; three 5-min pulses separated by 25-min intervals; four 5-min pulses separated by 17 . 5-min intervals; five 5-min pulses separated by 13-min intervals; six 5-min pulses separated by 10-min intervals; seven 5-min pulses separated by 7 . 86-min intervals; eight 5-min pulses separated by 6 . 25-min intervals; nine 5-min pulses separated by 5-min intervals . Control software for the microfluidic device and a full protocol are provided elsewhere ( Hansen et al . , 2015 ) . Image analysis was performed using custom-written software ( MATLAB ) that segments , tracks and quantifies single-cell time-traces and has been described previously ( Hansen et al . , 2015; Hansen and O'Shea , 2013 ) . All raw single-cell data are available online as Supplementary file 1 and in ( Hansen and O'Shea , 2015 ) . The mutual information for a single reporter is defined in Equation 1 and the maximal mutual information given by:I ( R;S ) =maxp ( S ) [MI ( R;S ) ] for ∑ip ( Si ) =1; p ( Si ) ≥0 . The p ( S ) that maximizes the mutual information is determined using the iterative Blahut-Arimoto algorithm . An unbiased I was estimated using jackknife sampling to correct for undersampling as it has previously been described ( Strong et al . , 1998; Slonim et al . , 2005; Cheong et al . , 2011 ) . The data were discretized by binning as shown in Figure 2 . Maximal mutual information , I , and its error are reported as the mean and standard deviation , respectively , from calculating the unbiased I using 15 to 35 bins , inclusive . To determine the maximal joint mutual information , I ( Figure 4A ) , first consider the joint mutual information between the signal S and two responses R1 ( e . g . , YFP ) and R2 ( e . g . , CFP ) :MI ( R1 , R2;S ) =MI ( R1;S ) +MI ( R2;S\|R1 ) , Where MI ( R1;S ) is known from Equation 1 and MI ( R2;S\|R1 ) is given by:MI ( R2;S\|R1 ) =∑i , j , kp ( R1 ( i ) , R2 ( j ) , S ( k ) ) log2 ( p ( R1 ( i ) ) p ( R1 ( i ) , R2 ( j ) , S ( k ) ) p ( R1 ( i ) , R2 ( j ) ) p ( R1 ( i ) , S ( k ) ) ) . The maximal joint mutual information is then given by:I ( R1 , R2;S ) =maxp ( S ) [MI ( R1 , R2;S ) ] for ∑ip ( Si ) =1; p ( Si ) ≥0 . As before , p ( S ) is obtained using the Blahut-Arimoto algorithm , and the mean and error of I are obtained as for a single reporter , except using 8 to 20 bins , inclusive . Full details are given in Supplementary file 2 . Briefly , the total , intrinsic and extrinsic noise for each condition is calculated using dual-reporters ( CFP/YFP ) ( Elowitz et al . , 2002; Swain et al . , 2002 ) . The expression distributions in the absence of extrinsic noise are required to determine Iint . This is an intractable problem ( Hilfinger and Paulsson , 2011 ) . To estimate it , the raw , empirical YFP distribution is fitted to a gamma distribution ( YFP∼Γ ( a , b ) ) . Keeping the mean fixed , a new gamma distribution representing the YFP response in the absence of extrinsic noise is then inferred by filtering out the extrinsic contribution to the variance . This is repeated for each condition , each distribution is then discretized and the maximal mutual information , I , determined as above . The accuracy of the algorithm was tested by simulating five linear and five non-linear stochastic gene expression models for both a fast and a slow promoter using the Gillespie algorithm under AM ( 10 conditions ) . Extrinsic noise is added by picking the translation rate and TF concentration for each iteration from a gamma distribution . The algorithm was then applied to each data set with extrinsic noise and compared to simulation results with only intrinsic noise and the error calculated . In all 80 cases ( 10 models , 2 promoters , 4 levels of extrinsic noise ) , the error was less than 5% ( in bits ) and the mean error was less than 2% . Full details are given in Supplementary file 2 . Measurement noise is a major concern for information theoretical calculations and can lead to underestimates of mutual information . To control and minimize effects of noise , the following data processing pipeline was employed . For each single-cell , a time-trace of 64 YFP measurements is made ( 2 . 5 min interval ) . The fluorescence ( in AU ) is the mean pixel-intensity per cell corresponding to the YFP concentration . As can be seen in Figure 2—figure supplement 1B and Figure 2—figure supplement 2 from the single-cell YFP traces , YFP concentration generally reaches a plateau around or after the 100 min time-point ( element 43 in the YFP vector ) . So the maximal YFP level in the cell is measured approximately 20 times before the experiment ends ( element 64 in the YFP vector ) . Although there is slight noise in each measurement of the YFP concentration as shown in Figure 2—figure supplement 2A ( black circles ) , because YFP is independently measured ∼20 times after it has reached a plateau , the actual YFP level can accurately be determined by smoothing ( Figure 2—figure supplement 2A , red line ) . The YFP trace is smoothed using an 11-point moving average filter and the vector is subsequently converted to a scalar by taking the maximal YFP value in the ( 33;64 ) range of elements . The scalar YFP concentration ( Figure 2—figure supplement 1B ) is used for all information theoretical calculations . We believe that the protein concentration is the most biologically relevant measure of gene expression . For example , the activity of a stress response enzyme is generally determined by its concentration . But we note that had a different measure been used , that is , had the dynamics of the YFP time-trace been included , different estimates of I would be obtained ( see also Supplementary file 2 for a further discussion ) . The following factors , among others , contribute to measurement noise: slight variations in microscope focusing; fluctuations in cellular autofluorescence; instrumentation variability ( e . g . , camera noise ) ; day-to-day experimental variability; slight errors from automated image analysis . Nonetheless , as is also evident from Figure 2—figure supplement 2 measurement noise is small . For HXK1 and SIP18 IAM and IFM were independently measured twice in different strains: the SIP18 dual-reporter strain ( EY2813/ASH94 ) , the HXK1 dual-reporter strain ( EY2810/ASH91 ) , and the 1× reporter diploid ( EY2972/ASH194 ) . The results are shown in the table below:IGene::YFP / gene::CFP strain1x sip18::YFP / hxk1::CFP strainIAM ( sip18::YFP ) 1 . 21 ± 0 . 03 bits1 . 17 ± 0 . 02 bitsIFM ( sip18::YFP ) 0 . 52 ± 0 . 06 bits0 . 50 ± 0 . 05 bitsIAM ( hxk1::CFP/YFP ) 1 . 30 ± 0 . 01 bits1 . 30 ± 0 . 01 bitsIFM ( hxk1::CFP/YFP ) 1 . 11 ± 0 . 01 bits1 . 14 ± 0 . 01 bits As is clear from the table above , the measurements of IAM and IFM between different strains ( with slightly different genetic backgrounds ) are highly similar and within error . This provides high confidence in the measurements and shows that the measurements are robust between different clones . Nonetheless , a constant noise source would cause all measurements to be underestimates by similar amounts . Therefore , the consistency of the measurements does not exclude the presence of a constant noise source . However , it is also important to note that most noise sources are ‘extrinsic’ to the gene and will therefore partially be filtered out by the algorithm during the correction for extrinsic noise . All strains used in this study are listed in Table 1 . The diploid strains containing fluorescent reporters for the SIP18 ( ASH94/EY2813 ) and HXK1 ( ASH91/EY2810 ) promoters have been described previously ( Hansen and O'Shea , 2013 ) . These and all other Saccharomyces cerevisiae strains used in this study are from an ADE+ strain in the W303 background ( MATa [EY0690] and MATα ( EY0691 ) trp1 leu2 ura3 his3 can1 GAL+ psi+ ) . Standard methods for growing and genetically manipulating yeast were used throughout this study and all manipulations were performed in the same manner in both haploid mating types unless otherwise stated . Mating was performed by mixing haploids and selecting for diploids on SD–TRP–LEU plates . All genetic manipulations were verified by polymerase chain reaction ( PCR ) . 10 . 7554/eLife . 06559 . 013Table 1 . List of strains . DOI: http://dx . doi . org/10 . 7554/eLife . 06559 . 013StrainTypeStrain detailsEY0690MATaW303 ( trp1 leu2 ura3 his3 can1 GAL+ psi+ ) ( not generated in this study ) EY0691MATαW303 ( trp1 leu2 ura3 his3 can1 GAL+ psi+ ) ( not generated in this study ) EY2808/ASH89MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX hxk1::mCitrine_V163A-spHIS5 ( not generated in this study ) EY2809/ASH90MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX hxk1::SCFP3A-spHIS5 ( not generated in this study ) EY2810/ASH91DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1/LEU2 MSN2-mCherry NHP6a-iRFP::kanMX hxk1::mCitrineV163A/SCFP3A-spHIS5 ( not generated in this study ) EY2811/ASH92MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-spHIS5 ( not generated in this study ) EY2812/ASH93MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::SCFP3A-spHIS5 ( not generated in this study ) EY2813/ASH94DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1/LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrineV163A/SCFP3A-spHIS5 ( not generated in this study ) EY2964/ASH139MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-HIS3 pSIP18 Mut A 3 STREsEY2965/ASH140MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-HIS3 pSIP18 Mut B 4 STREsEY2966/ASH188MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::SCFP3A-HIS3 pSIP18 Mut B 4 STREsEY2967/ASH189DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1/LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A/SCFP3A-HIS3 pSIP18 Mut B 4 STREsEY2968/ASH190MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::SCFP3A-HIS3 pSIP18 Mut A 3 STREsEY2969/ASH191DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1/LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A/SCFP3A-HIS3 pSIP18 Mut A 3 STREsEY2970/ASH192MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-HIS3 hxk1::URA3EY2971/ASH193MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX hxk1::SCFP3A-HIS3 sip18::URA3EY2972/ASH194DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1/LEU2 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-HIS3 hxk1::URA3 / hxk1::SCFP3A-HIS3 sip18::URA3 ( 1x reporter diploid ) EY2973/ASH195MATaTPK1M164G TPK2M147G TPK3M165G msn4Δ::TRP1 MSN2-mCherry NHP6a-iRFP::kanMX sip18::mCitrine_V163A-HIS3 hxk1::SCFP3A-HIS3EY2974/ASH196MATαTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX hxk1::SCFP3A-HIS3 sip18::mCitrine_V163A-HIS3EY2975/ASH197DiploidTPK1M164G TPK2M147G TPK3M165G msn4Δ::LEU2 MSN2-mCherry NHP6a-iRFP::kanMX 2x hxk1::SCFP3A_JCat-HIS3 2x sip18::mCitrine_V163A-HIS3 ( 2x reporter diploid ) To generate the pSIP18 promoter mutants , the relevant segment of the promoter was replaced by URA3 and followed by replacing the URA3 fragment with a PCR generated fragment containing the relevant mutations and counterselection against URA3 . The full sequence of the wild-type SIP18 promoter and the mutant promoters is listed below .	The way that a cell responds to an external stimulus is governed by a sequence of events called a signalling pathway . While cells are exposed to a wide range of external stimuli—such as different types of chemicals and different forms of radiation—the last stage of the signalling pathway usually involves a gene being expressed as a protein or some other gene product . The amount of protein that is produced depends on the intensity of the signal that reaches the end of the pathway . Proteins called transcription factors have an important role in this gene expression stage , and it is quite common for several signalling pathways to pass through the same transcription factor . How does the cell ensure that the information travelling along a particular pathway reaches the relevant gene and that the correct level of gene expression takes place ? Biologists have been using information theory—a set of ideas developed by computer scientists and engineers—to understand signalling pathways at a fundamental level . It turns out that just as radio stations can broadcast on FM ( which is short for frequency modulation ) or AM ( amplitude modulation ) , cells can do something similar . Msn2 is a transcription factor that is found in yeast: when the supply of glucose to the yeast cells is limited , Msn2 becomes active in short bursts , with the frequency of the bursts depending on the severity of the glucose shortage ( which is similar to FM radio ) . However , when the yeast cells are exposed to chemicals that cause oxidative stress , Msn2 becomes active for prolonged periods , with the amplitude of this activity depending on level of oxidative stress ( similar to AM radio ) . In the language of information theory , the behaviour of Msn2 encodes two types of information: information about identity ( short bursts of activity , or FM , mean that there is a shortage of glucose; sustained bursts , or AM , mean that the cell is experiencing oxidative stress ) , and information about intensity ( that is , information about the severity of the glucose shortage or the level of oxidative stress ) . But how much of this information is transmitted to the relevant genes ? Hansen and O'Shea have used a combination of experiment and information theory to explore this question . For both the AM and FM cases , it is found that the cell can transmit the identity information but not the intensity information . However , the amount of information transmitted can be increased by having multiple copies of the same gene , by combining information from more than one gene , or by modifying a region of DNA called a promoter that is involved in the regulation of genes . Finally , unlike radio broadcasting , where FM is generally favoured over AM , Hansen and O'Shea find that AM signalling is more reliable than FM signalling in cells . In the future , it will be a priority to investigate whether these results apply more generally beyond the Msn2 system in yeast .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2015	Limits on information transduction through amplitude and frequency regulation of transcription factor activity
Adenosine triphosphate ( ATP ) , the chemical energy currency of biology , is synthesized in eukaryotic cells primarily by the mitochondrial ATP synthase . ATP synthases operate by a rotary catalytic mechanism where proton translocation through the membrane-inserted FO region is coupled to ATP synthesis in the catalytic F1 region via rotation of a central rotor subcomplex . We report here single particle electron cryomicroscopy ( cryo-EM ) analysis of the bovine mitochondrial ATP synthase . Combining cryo-EM data with bioinformatic analysis allowed us to determine the fold of the a subunit , suggesting a proton translocation path through the FO region that involves both the a and b subunits . 3D classification of images revealed seven distinct states of the enzyme that show different modes of bending and twisting in the intact ATP synthase . Rotational fluctuations of the c8-ring within the FO region support a Brownian ratchet mechanism for proton-translocation-driven rotation in ATP synthases . In the mitochondria of eukaryotes , adenosine triphosphate ( ATP ) is produced by the ATP synthase , a ∼600 kDa membrane protein complex composed of a soluble catalytic F1 region and a membrane-inserted FO region . The ATP synthase is found in the inner membranes of mitochondria , with the F1 region in the mitochondrial matrix and the FO region accessible from the inter-membrane space between the mitochondrial outer and inner membranes . In the mammalian enzyme , the subunit composition is α3β3γδε for the F1 region with subunits a , e , f , g , A6L , DAPIT , a 6 . 8 kDa proteolipid , two membrane-inserted α-helices of subunit b , and the c8-ring forming the FO region ( Walker , 2013 ) . The rotor subcomplex consists of subunits γ , δ , ε , and the c8-ring . In addition to the rotor , the F1 and FO regions are linked by a peripheral stalk composed of subunits OSCP , d , F6 , and the hydrophilic portion of subunit b . Approximately 85% of the structure of the complex is known at high resolution from X-ray crystallography of constituent proteins , which have been assembled into a mosaic structure within the constraints of a cryo-EM map at 18 Å resolution ( Walker , 2013; Baker et al . , 2012 ) . The proton motive force , established by the electron transport chain during cellular respiration , drives protons across the FO region through the interface between the a subunit and the c8-ring , inducing rotation of the rotor ( Boyer , 1997; Walker , 1998 ) . While the mechanism by which ATP synthesis and hydrolysis are coupled to rotation of the γ subunit is now understood well ( Walker , 2013 ) , it is still unresolved how rotation of the central rotor is coupled to proton translocation through the FO region . The most popular model suggests that proton translocation occurs through two offset half channels near the a subunit/c subunit interface ( Junge et al . , 1997 , Junge , 2005 ) . In this model , one half channel allows protons to move half-way across the lipid bilayer in order to protonate the conserved Glu58 residue of one of the c subunits . The other half channel allows deprotonation of the adjacent c subunit ( Lau and Rubinstein , 2012 ) , setting up the necessary condition for a net rotation of the entire c-ring . Rotation does not occur directly from the protonating half channel to the deprotonating half channel , but in the opposite direction so that the protonated , and therefore uncharged , Glu residues traverse through the lipid environment before reaching the deprotonating half channel . The deprotonated Glu residue prevents the ring from rotating in the opposite direction , which would place the charged residue in the hydrophobic environment of the lipid bilayer . Rotation of the c-ring occurs due to Brownian motion , making the enzyme a Brownian ratchet . A recent cryo-EM map of the Polytomella sp . ATP synthase dimer showed two long and tilted α-helices from the a subunit in contact with the c10-ring of that species ( Allegretti et al . , 2015 ) . This arrangement of α-helices from the a and c subunits was also seen in the Saccharomyces cerevisiae V-type ATPase ( Zhao et al . , 2015a ) . Cryo-EM of the S . cerevisiae V-ATPase demonstrated that images of rotary ATPases could be separated by 3D classification to reveal conformations of the complex that exist simultaneously in solution . In the work described here , we obtained and analyzed cryo-EM images of the bovine mitochondrial ATP synthase . 3D classification of the images resulted in seven distinct maps of the enzyme , each showing the complex in a different conformation . By averaging the density for the proton-translocating a subunit from the seven maps , we generated a map segment that shows α-helices clearly . Analysis of evolutionary covariance in the sequence of the a subunit ( Göbel et al . , 1994 ) allowed the entire a subunit polypeptide to be traced through the density map . The resulting atomic model for the a subunit was fitted into the maps for the different rotational states , suggesting a path for protons through the enzyme and supporting the Brownian ratchet mechanism for the generation of rotation ( Junge et al . , 1997; 2005 ) , and thereby ATP synthesis , in ATP synthases . The FO regions of all seven maps also revealed a remarkable feature not resolved previously in cryo-EM maps of ATP synthases ( Baker et al . , 2012; Allegretti et al . , 2015; Lau et al . , 2008; Rubinstein et al . , 2003 ) . The feature appears to consist of an elongated membrane-embedded density , possibly an α-helix , that extends from the rotor-distal portion of FO to the c8-ring . The orientation of this density would cause it to pass through the inter-membrane space of the mitochondrion ( Figure 1B and C , orange arrow ) . While not identified in the previous cryo-EM map of the enzyme at 18 Å resolution , the structure is consistent with a poorly-resolved ridge along the surface of the FO region seen in the earlier map ( Baker et al . , 2012 ) . Because it extends from the bent end of the FO region , this feature may correspond to the soluble part of the e subunit . Indeed , a similar structure was observed in single particle EM of negatively stained ATP synthase dimers from bovine heart mitochondria , and was proposed to be interacting e subunits ( Minauro-Sanmiguel et al . , 2005 ) . However , in the present structure the feature is not positioned to interact between dimers of the enzyme and its role in the complex remains unclear . In order to improve the signal-to-noise ratio for the FO region of the complex , the membrane regions from the seven different maps were aligned and averaged . Averaging maps increases the signal-to-noise ratio where the structures are similar , but blurs regions where the maps differ . In principle , this method could also be applied to other map regions of the ATP synthase or other heterogeneous protein complexes by applying an appropriate transform before averaging . Averaging the FO region provides a clear view of the portion of the FO region adjacent to the rotor , allowing the trans-membrane α-helices from the a , b , and A6L subunits to be identified reliably ( Figure 1C and D , green arrows , and Figure 2 ) . The c-ring has a lower density in the averaged membrane region than in the original maps , suggesting that its position differs between maps ( Figure 1C and D , purple arrows ) . The averaged density for the FO region revealed the a subunit to have five membrane-inserted α-helices and an additional α-helix along the plane of the membrane surface ( Figure 2 ) . Three additional trans-membrane α-helices are also apparent , presumably two from the b subunit ( Walker et al . , 1987 ) and one from the A6L subunit ( Fearnley and Walker , 1986 ) . The mammalian mitochondrial a subunit possesses the two highly tilted α-helices in contact with the c-ring that were seen previously for the Polytomella sp . F-type ATP synthase ( Allegretti et al . , 2015 ) and S . cerevisiae V-ATPase ( Zhao et al . , 2015a ) ( Figure 2A ) . 10 . 7554/eLife . 10180 . 008Figure 2 . 3D structure of the FO region . ( A ) In the FO region of the complex , density was segmented for the a subunit ( green ) , the b subunit ( red ) , the A6L subunit ( blue ) , and the structure thought to arise from subunits e and g ( orange ) . ( B ) The a subunit sequence could be placed unambiguously into the cryo-EM density ( green ) by including constraints for residues predicted to be near to each other due to evolutionary covariance ( red lines ) . ( C ) The a subunit ( coloured with a gradient from blue to red to denote directionality from the N to C terminus ) possesses six α-helices , numbered 1–6 . Trans-membrane α-helices from subunits b and A6L are shown as volumes ( red and blue , respectively ) . Five of the α-helices of subunit a are membrane-inserted while helix #2 runs along the matrix surface of the FO region . The N terminus of the a subunit is on the inter-membrane space side of the subunit while the C terminus is on the matrix side . The highly conserved residue Arg159 is on the elongated and highly tilted α-helix #5 . Scale bar , 25 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 00810 . 7554/eLife . 10180 . 009Figure 2—figure supplement 1 . Analysis of evolutionary covariance of residues . ( A ) The top 90 predicted couplings between residues of the a subunit are indicated , along with trans-membrane helices predicted by the MEMSAT-SVM algorithm ( 52 ) , shown in green , and highly conserved residues , shown in red . Residues modeled as membrane-inserted α-helices based on the cryo-EM density are indicated with dark blue rectangles outside of the sequence and residues modeled as a soluble α-helix based on the cryo-EM density is indicated with a light blue rectangle . ( B ) The top six predicted couplings between residues of the a subunit and residues on the outer surface of the c-ring are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 009 A model for the a subunit was built into the cryo-EM density map using constraints from analysis of evolutionary covariance in sequences of the a subunit from different species . Analysis of covariance in evolutionarily related protein sequences can identify pairs of residues in a protein structure that are likely to interact physically with each other ( Göbel et al . , 1994; Cronet et al . , 1993; Hopf et al . , 2012; Ovchinnikov et al . , 2014 ) . Spatial constraints from covariance analysis were sufficient not only to identify tentatively trans-membrane α-helices of the a subunit that are adjacent to each other , but also suggest which face each α-helix presents to the other α-helices ( Figure 2B and Video 2 , red lines ) . The constraints show patterns of interaction consistent with the predicted α-helical structure of the a subunit ( Figure 2—figure supplement 1A ) , as well as interactions between the a subunit and the outer α-helix of a c subunit in the c8-ring ( Figure 2—figure supplement 1B ) . As a result , we were able to trace the path of the a subunit polypeptide through the cryo-EM density map . The fit of the α-helices in the a subunit density was improved by molecular dynamics flexible fitting ( MDFF ) ( Trabuco et al . , 2008 ) and the 34 residue long connecting loop from residues 115 to 148 was built with Rosetta ( Rohl et al . , 2004 ) ( Figure 2B and Video 2 ) . This connecting loop was built to be physically reasonable , but because its structure is not derived from experimental data it is not included in the discussion below . The final model places the α carbons of the residues in the co-varying pairs within 15 Å of each other in 94% of the top 90 identified pairs , with an average Cα to Cα distance of 10 . 3 Å . The 6% of constraints that are violated by the model are consistent with the false positive rate observed when testing covariance analysis approaches with proteins of known structure ( Marks et al . , 2011 ) . 10 . 7554/eLife . 10180 . 010Video 2 . Fold of the a subunit . The density corresponding to the a subunit ( green ) , membrane-inserted portion of the b subunit ( red ) , and A6L subunit ( blue ) are shown , in addition to a ribbon diagram for the a subunit ( green ) and the top 90 constraints from analysis of covarying residues in the a subunit sequence ( red lines ) . 6% of the constraints could not be satisfied , which is consistent with the false positive rate from known structures ( Marks et al . , 2011 ) . Scale bar , 25 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 010 The mammalian a subunit appears to consist of six α-helices , with five α-helices that penetrate into the membrane ( Figure 2C ) . The N terminus of the subunit is in the inter-membrane space of the mitochondrion . The first α-helix extends vertically across the FO region distal to the contact of the a subunit and c8-ring . The two trans-membrane α-helices of the b subunit are packed against one surface of helix #1 while the single trans-membrane α-helix from the A6L subunit is packed against its opposite surface . The second density region , interpreted as an α-helix of the a subunit , is not membrane-inserted and extends along the matrix surface of the FO region connecting the membrane-inserted α-helix #1 with a membrane-inserted helical-hairpin composed of α-helices #3 and #4 . This hairpin of the third and fourth α-helices does not appear to cross the FO region fully , as seen previously in the Polytomella sp . ATP synthase ( Allegretti et al . , 2015 ) . The final two trans-membrane helices are the two highly tilted α-helices seen previously with the Polytomella sp . ATP synthase and S . cerevisiae V-ATPase , with the C terminus of the a subunit on the matrix side of the FO region . Within this structure , Arg159 , which is essential and completely conserved , is found near the middle of the long tilted α-helix #5 nearer the inter-membrane space side of the FO region , different from its predicted position in the Polytomella sp . enzyme ( Allegretti et al . , 2015 ) . To analyze the different enzyme conformations detected by 3D classification , the maps were segmented and available crystal structures for the F1:IF1 complex ( Gledhill et al . , 2007 ) , F1 peripheral stalk complex ( Rees et al . , 2009 ) , peripheral stalk alone ( Dickson et al . , 2006 ) , and F1-c8 complex ( Watt et al . , 2010 ) were combined into each of the maps by MDFF ( Trabuco et al . , 2008 ) . Residues for the b subunit were extended from the N terminus of the b subunit crystal structure into the membrane region based on trans-membrane α-helix prediction . While MDFF with maps in this resolution range cannot be used to determine the locations or conformations of amino acid side chains , loops , or random-coil segments of models , it can show the positioning of α-helices in the structures . Figures 3A and B compare the fitting for state 1a ( Figure 3A ) and state 1b ( Figure 3B ) , illustrating the accuracy with which α-helical segments could be resolved in the maps of different sub-states . The atomic model alone for state 1a is shown in Figure 3C , with the c8-ring removed for clarity in Figure 3D . Transitions between the different states were illustrated by linear interpolation ( Video 3 ) . As seen previously for the S . cerevisiae V-ATPase , almost all of the subunits in the enzyme undergo conformational changes on transition between states ( Zhao , et al . , 2015a ) . Because there were two sub-states identified for states 1 and 3 there is only a single sub-state to sub-state transition for these two states . In comparison , three different sub-states were identified for state 2 and consequently there are three sub-state to sub-state transitions that are possible for this state . All of the sub-state to sub-state transitions include a slight rotation of the c8-ring against the a subunit . It is possible that this movement is due to partial disruption of the subunit a/c8-ring interface . However , the structural differences within the FO regions of different classes are significantly smaller than the structural differences seen elsewhere in the enzyme , suggesting that these changes do not originate from disruption within the membrane region of the complex and instead reflect flexibility in the enzyme . 10 . 7554/eLife . 10180 . 011Figure 3 . Docking of atomic models into the cryo-EM maps . Fitting of all available atomic models into the density map is shown for state 1a ( A ) and state 1b ( B ) . State 1a is also shown in a different orientation and without the density map ( C ) and with the c8-ring removed for clarity ( D ) . The apparent gap between the c8-ring and γ and δ subunits is filled with amino acid side chains and is the same as was seen in the crystal structure of the F1-c8 complex ( Watt et al . , 2010 ) . Scale bar , 25 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 01110 . 7554/eLife . 10180 . 012Figure 3—figure supplement 1 . Differences between sub-states . The differences between sub-states can be seen by overlaying maps and models for state 1a ( red ) and 1b ( green ) ( A ) , 2a ( red ) and 2b ( green ) ( B ) , 2b ( red ) and 2c ( green ) ( C ) , and 3a ( red ) and 3b ( green ) ( D ) . These differences can be represented approximately as a rigid body rotation of the 33 hexamer by 10 ( state 1a to 1b ) , 11 ( state 2a to 2b ) , 12 ( state 2b to 1c ) , and 16 ( state 3a to 3b ) . The axes of these rotations are shown as black rods . This movement is most easily seen in Videos 4 and 5 . Scale bar , 25 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 01210 . 7554/eLife . 10180 . 013Video 3 . Conformation changes during the rotary cycle . Linear interpolation is shown between one of each of the main states identified by 3D classification ( state 1a , 2a , and 3a ) showing the large conformation changes that occur during rotation . Scale bar , 25 Å . [please view movie as loop]DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 013 The largest change between the sub-states of each state can be approximated by rotations of the α3β3 hexamer relative to the rest of the complex by angles ranging from 10 to 16° . A comparison of the maps for the different sub-states and the axes of these rotations are shown in Figure 3—figure supplement 1 . The resulting conformational changes can be seen most clearly in Videos 4 and 5 . The state 1a to 1b transition reveals a bending of the peripheral stalk towards the top of the F1 region near the OSCP and F6 subunits ( Videos 4 and 5 , panel A ) . In comparison , the state 3a to 3b transition reveals bending of the peripheral stalk near where the b subunit enters the membrane ( Videos 4 and 5 , panel D ) . Transitions between the three sub-states of state 2 show both motions: the transition between 2a and 2b shows mostly bending of the peripheral stalk near OSCP and F6 subunits , while the 2b and 2c transition shows mostly bending near the membrane-inserted portion of the peripheral stalk ( Videos 4 and 5 , panels B and C , respectively ) . The transition from state 2a to 2c shows a combined bending at both of these positions . It is most likely that the different modes of bending exist in all of the states and further classification of larger datasets would be expected to reveal these complex motions . The different sub-states do not appear to have a specific sequence or represent specific intermediates in the catalytic rotation sequence . Instead , the differences in conformation between sub-states when taken together illustrate the flexibility of the enzyme , a property that has been linked to its rapid rate of enzymatic activity ( Zhao , et al . , 2015a; Zhou et al . , 2014 ) . The functional significance of the sub-states may also be determined by the orientation of the c8-ring with respect to the a subunit , as discussed below . 10 . 7554/eLife . 10180 . 014Video 4 . Conformational differences between the different ATP synthase maps . Differences between conformations detected by 3D classification are illustrated by linear interpolation between state 1a and 1b , showing bending of the peripheral stalk near the OSCP and F6 subunits ( A ) , state 2a and 2b showing a similar bending of the peripheral stalk near the OSCP and F6 subunits ( B ) , state 2b and 2c showing bending of the peripheral stalk near the membrane region ( C ) , and state 3a and 3b showing a similar bending of the peripheral stalk near the membrane region ( D ) . Scale bar , 25 Å . [please view movie as loop]DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 01410 . 7554/eLife . 10180 . 015Video 5 . Conformational differences between the different ATP synthase maps . The same interpolations as shown in Video 4 , except viewed from the F1 region towards the FO region . Scale bar , 25 Å . [please view movie as loop]DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 015 Predicting the path of protons through membrane protein complexes has proven difficult , even in cases where high-resolution atomic models including bound water molecules are available from X-ray crystallography ( Hosler et al . , 2006 ) . Nonetheless , features in the structure of the bovine mitochondrial FO region suggest a possible path for proton translocation similar to a model put forward based on the structure of the Polytomella sp . ATP synthase ( Allegretti et al . , 2015 ) . The arrangement of α-helices in the FO region is remarkably similar to the arrangement of α-helices in the VO region of the yeast V-ATPase ( Zhao , et al . , 2015a ) , even though the V-ATPase a subunit has eight α-helices and little detectable sequence similarity with the F-type ATP synthase a subunit . The conserved general architecture of the membrane-inserted regions in F-type ATP synthases and V-type ATPases suggests that the observed arrangement of α-helices is functionally important and likely involved in proton translocation ( Figure 4A and B ) . The matrix half channel of the ATP synthase is likely to be formed by the cavity between the c8-ring and the matrix ends of tilted α-helices #5 and #6 of the a subunit . The lumenal half channel in the V-ATPase is probably formed entirely from α-helices from the a subunit , whereas the corresponding inter-membrane space half channel in the ATP synthase is likely composed of the inter-membrane space ends of α-helices #5 and #6 and one or both of the two trans-membrane α-helices of the b subunit . Defining the exact placement of half channels will likely require higher-resolution maps from cryo-EM or X-ray crystallography that reveal amino acid side chain density and bound water molecules . In addition to bending and twisting of the peripheral stalk and central rotor of the enzyme , the differences between the sub-states of each state show variability in the rotational position of the c8-ring in relation to the a subunit ( Figure 4C and D ) , even in the nucleotide-depleted conditions in which cryo-EM grids were frozen for this analysis . This lack of a rigid interaction between the c8-ring and a subunit is consistent with the Brownian ratchet model of proton translocation ( Junge et al . , 1997 ) . In the Brownian ratchet model , the rotational position of the ring fluctuates due to Brownian motion , but the ring cannot turn to place the Glu58 residue of a c subunit into the hydrophobic environment of the lipid bilayer until the Glu58 is protonated at the inter-membrane space half channel . Therefore , with this model , the different sub-states would correspond to energetically equivalent or nearly-equivalent conformations that occur due to Brownian motion . Video 6 illustrates the extent of rotational oscillation predicted from the transition between states 2a and 2c . It is most likely that this oscillation occurs as each c subunit passes the interface with the a subunit , with 8/3 c subunits on average contributing to the synthesis of one ATP molecule . The rotational flexibility of the c8-ring that exists even when the γ subunit is locked within the α3β3 hexamer suggests that flexing and bending of the components of the ATP synthase smooths the coupling of the 8-step rotation of the c8-ring with the 3-step rotation of the F1 region . This model suggests that the observed flexibility in the enzyme , which apparently complicates determination of atomic resolution structures directly from cryo-EM data , is also essential to the mechanism of ATP synthesis . 10 . 7554/eLife . 10180 . 016Figure 4 . Model for proton translocation . ( A and B ) The a subunit ( green ) , along with the membrane-intrinsic α-helices of the b subunit ( red ) , form two clusters that could be the half channels needed for trans-membrane proton translocation . ( C and D ) The map segment corresponding to the c8-ring is shown for state 2a ( pink ) and state 2c ( purple ) . The difference in rotational position of the c-ring is consistent with the Brownian fluctuations predicted for the generation of a net rotation . Scale bar , 25 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 10180 . 016 Bovine mitochondrial ATP synthase was purified as described previously ( Runswick et al . , 2013 ) and cryo-EM specimen grids were prepared as described previously , except that glycerol was removed from specimens prior to grid freezing with a 7 kDa molecular weight cutoff Zeba Spin centrifuged desalting column ( Thermo Scientific ) and nano-fabricated grids with 500 nm holes were used ( Marr et al . , 2014 ) . After optimization of grid freezing conditions , micrographs were recorded from three grids on a Titan Krios microscope ( FEI ) operated at 300 kV with parallel illumination of a 2 . 5 μm diameter area of the specimen and an electron fluency of 3 el-/Å2/s . A 70 μm objective aperture was employed with a nominal magnification of 18 , 000 × onto a K2 Summit direct detector device ( Gatan Inc . ) operated in super-resolution mode with a 1 . 64 Å physical pixel and 0 . 82 Å super-resolution pixel . With no specimen present , the rate of exposure of the detector was 8 el-/pixel/s . Exposure-fractionated movies of 20 . 1 s were recorded as stacks of 67 frames , so that selected specimen areas were exposed with a total of 60 . 3 el-/Å2 . Data collection was automated with SerialEM ( Mastronarde , 2005 ) . Magnification anisotropy ( Zhao et al . , 2015b ) under the conditions described above was measured previously from images of a standard cross-grating specimen with the program mag_distortion_estimate ( Grant and Grigorieff , 2015a ) . The linear scaling parameters were 0 . 986 and 1 . 013 , the azimuth of the distortion was 134 . 0° , and the program mag_distortion_correct was used to correct for this distortion in each dose-fractionated frame . The frames were then down-sampled to a pixel size of 1 . 64 Å by Fourier-space cropping and aligned with each other with the program Unblur ( Grant and Grigorieff , 2015a ) . Defocus parameters were estimated from aligned sums of frames using CTFFIND4 ( Rohou and Grigorieff , 2015 ) . Particle images were selected in Relion and subjected to 2D classification ( Scheres , 2015; Scheres , 2012 ) , yielding a set of 195 , 233 single particle coordinates selected from 5 , 825 movies . Local beam-induced motion was corrected for each particle with the program alignparts_lmbfgs ( Rubinstein and Brubaker , 2015 ) . Aligned dose-fractionated particle images were filtered and summed to optimize the signal-to-noise ratio at all spatial frequencies ( Grant and Grigorieff , 2015b; Rubinstein and Brubaker , 2015; Baker et al . , 2010 ) , giving a set of particle images that were 256 × 256 pixels . These images were down-sampled to 128 × 128 pixels ( pixel size of 3 . 28 Å ) for determining particle orientations . Initial single-particle alignment parameter values were obtained by 5 rounds of iterative grid search and reconstruction in FREALIGN's mode 3 ( Grigorieff , 2007 ) , using the earlier published map of the enzyme as an initial reference ( Baker et al . , 2012 ) . FREALIGN's likelihood-based classification algorithm ( Lyumkis et al . , 2013 ) was then used to classify particles images into several maps , alternating between refinement of orientation parameters every 3rd or 4th iteration and class occupancy during other iterations . The final classification yielded 12 classes , of which 7 gave interpretable 3D maps . Only spatial frequencies up to 1/10 Å-1 were used during refinement to avoid fitting noise to high-resolution features in maps . All seven 3D maps had Fourier shell correlation values greater than 0 . 8 at this frequency . Segmentation of 3D maps was performed with UCSF Chimera ( Goddard et al . , 2007; Pintilie et al . , 2010 ) and atomic structures were fit flexibly into 3D maps using NAMD with Molecular Dynamics Flexible Fitting ( MDFF ) ( Trabuco et al . , 2008 ) . The FO regions from the seven different 3D maps were aligned and averaged in real space with UCSF Chimera and Situs ( Wriggers et al . , 1999 ) . Co-varying pairs of residues were detected in the full bovine mitochondrial ATP synthase a subunit sequence ( NCBI reference YP_209210 . 1 ) with the program EVcouplings ( Hopf et al . , 2012 ) using a pseudo-likelihood maximization approach and the top 90 connections were considered in the analysis . The protein was not assumed to have trans-membrane α-helices and the job was run as a quick launch with all other parameters at default settings . Evolutionary couplings between the a subunit and ATP synthase c subunit were detected with GREMLIN ( Ovchinnikov et al . , 2014 ) with an E-value threshold for multiple sequence alignments ( MSAs ) of 1 × 10-10 and Jackhmmer was used to produce MSAs over 8 iterations . To build a model of the a subunit , six straight α-helices ( φ = -57° and ψ = -47° ) were built in UCSF Chimera . These α-helices were fit manually in the map of the average FO region in the only orientations that satisfied constraints from evolutionary covariance analysis . For illustration , but not interpretation , loops connecting these helices were also included in the model . Randomly structured connecting loops between the α-helices were built with Modeller ( Eswar et al . , 2006 ) within UCSF Chimera and fitted into the density with MDFF with a low density scaling factor ( gscale = 0 . 3 ) over 200 , 000 steps ( 200 ps ) . Bond lengths and angles were then idealized with Rosetta ( idealize_jd2 command ) and the loop between residues 115 and 148 rebuilt in Rosetta ( loopmodel command ) using the quick_ccd method of remodelling ( Dimaio et al . , 2009 ) . Each output structure included an all-atom relaxation in the density map with a score weight of 0 . 1 . The lowest-energy model of 100 models was selected and angles were idealized and the structure energy-minimized with UCSF Chimera . Loops beside the one from residues 115 and 148 were too short for this process to be useful . The b subunit crystal structure was extended into the FO region of the map based on trans-membrane α-helix prediction from MEMSAT-SVM ( Nugent and Jones , 2009 ) .	A molecule called adenosine triphosphate ( ATP ) is the energy currency in cells . Most of the ATP used by cells is made by the membrane-embedded enzyme ATP synthase . This enzyme is found in membranes inside specialized compartments known as mitochondria . ATP synthase is made up of many protein subunits that work together as a molecular machine . Hydrogen ions flow across the membrane through the ATP synthase , turning a rotor structure within the enzyme , which leads to the production of ATP . It is not known how the transport of hydrogen ions causes rotation of the rotor . Some researchers have proposed that the enzyme works as a ratchet that is driven by the random Brownian motion of the rotor . That is , the rotational position of the rotor fluctuates randomly , but a ratchet mechanism ensures that there is a net rotation in one direction . However , there is currently little experimental evidence to back up this theory , which is known as the Brownian ratchet model . Zhou , Rohou et al . used a technique called electron cryomicroscopy ( or cryo-EM ) to study ATP synthase from cows . The cryo-EM data made it possible to use computer software to construct a three-dimensional model of the enzyme that is more detailed than previous attempts . Zhou , Rohou et al . show that the structure of ATP synthase is flexible , with the different protein subunits bending , flexing , and rotating relative to each other . This variability in the position of the rotor is consistent with the Brownian ratchet model . Together , these findings reveal important new details about the structure of ATP synthase and provide some of the first experimental evidence for the Brownian ratchet model . The new three-dimensional structure of ATP synthase will open the door to testing hypotheses of how the ATP synthase works .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Structure and conformational states of the bovine mitochondrial ATP synthase by cryo-EM
Novel regenerative therapies may stem from deeper understanding of the mechanisms governing cardiovascular lineage diversification . Using enhancer mapping and live imaging in avian embryos , and genetic lineage tracing in mice , we investigated the spatio-temporal dynamics of cardiovascular progenitor populations . We show that expression of the cardiac transcription factor Nkx2 . 5 marks a mesodermal population outside of the cardiac crescent in the extraembryonic and lateral plate mesoderm , with characteristics of hemogenic angioblasts . Extra-cardiac Nkx2 . 5 lineage progenitors migrate into the embryo and contribute to clusters of CD41+/CD45+ and RUNX1+ cells in the endocardium , the aorta-gonad-mesonephros region of the dorsal aorta and liver . We also demonstrated that ectopic expression of Nkx2 . 5 in chick embryos activates the hemoangiogenic gene expression program . Taken together , we identified a hemogenic angioblast cell lineage characterized by transient Nkx2 . 5 expression that contributes to hemogenic endothelium and endocardium , suggesting a novel role for Nkx2 . 5 in hemoangiogenic lineage specification and diversification . Development of the cardiovascular system takes place during the early stages of embryogenesis . Cardiac progenitors residing in the cardiac crescent are formed from the first heart field ( FHF ) located in the anterior lateral plate mesoderm ( LPM ) . As the embryo develops , FHF progenitors fuse at the midline to form the primitive heart tube , which begins to beat and , as a consequence , blood begins to circulate ( DeRuiter et al . , 1992; Stalsberg and DeHaan , 1969 ) . Second heart field ( SHF ) progenitors residing within the pharyngeal mesoderm ( Diogo et al . , 2015 ) contribute to subsequent growth and elongation of the heart tube ( Kelly et al . , 2001; Mjaatvedt et al . , 2001; Waldo et al . , 2001 ) . In both chick and mouse embryos , the FHF gives rise to myocytes of the left ventricle and parts of the atria , whereas the SHF contributes to myocardium of the outflow tract , right ventricle , and atria ( Buckingham et al . , 2005 ) . Recent studies suggest that these heart fields contain both unipotent and multipotent mesodermal progenitors that give rise to the diverse lineage types within the heart ( Kattman et al . , 2006; Lescroart et al . , 2014; Meilhac et al . , 2004; Moretti et al . , 2006; Wu et al . , 2006 ) . For example , bipotent SHF progenitors generate endocardium or smooth muscle cells as well as cardiomyocytes ( Lescroart et al . , 2014; Moretti et al . , 2006 ) . Cardiovascular progenitors sequentially express the cardiac transcription factors Mesp1/2 , Islet1 ( Isl1 ) and Nkx2 . 5 , and , in response to cues from the microenvironment , undergo lineage diversification and differentiation ( Laugwitz et al . , 2008; Prall et al . , 2007; Saga et al . , 1999 ) . The formation of blood vessels begins with the appearance of blood islands in the extraembryonic region . In the chick embryo , this occurs in the area vasculosa around St . 3–5 . Sabin first proposed that some blood cells differentiate directly from endothelial cells ( Sabin , 1920 ) . Indeed , endothelial and blood cells that form the rudimentary circulatory system have long been thought to originate from bipotent mesoderm progenitors termed ‘hemangioblasts’ ( Choi et al . , 1998 ) . Recent experimental advances revealed the existence of a specialized cell , hemogenic endothelium , that harbours the potential to generate hematopoietic progenitors ( Boisset et al . , 2010; Jaffredo et al . , 1998 ) . These cells arise early in embryonic development and migrate from extra- to intra-embryonic locations ( Tanaka et al . , 2014 ) . However , while blood cell formation from hemogenic endothelium has been visualised directly in multiple animal models ( Bertrand et al . , 2010; Boisset et al . , 2010; Kissa and Herbomel , 2010; Lam et al . , 2010 ) , and in vitro during ES cells differentiation ( Eilken et al . , 2009; Lancrin et al . , 2009 ) , the existence in vivo of a bipotential hemangioblast that contributes to both blood and endothelial cells remains controversial ( Hirschi , 2012 ) . Recent lineage tracing and live imaging studies addressing the ontogenic origins of hemoangiogenic cells in mouse embryos suggest that all hemogenic cells derive from a Flk1+/Runx1+/Gata1- ‘hemogenic angioblast’ population located in extraembryonic mesoderm bordering the region forming blood islands ( Tanaka et al . , 2014 ) . As for angioblasts that contribute to the embryonic vasculature , these cells actively migrate to embryonic sites of hemopoiesis before the establishment of circulation . Runx1 activation in the aorta-gonad-mesonephros ( AGM ) region is essential at this early stage for the specification of hemogenic endothelial cell fate and for definitive hemopoiesis ( Tanaka et al . , 2012 ) . The origins of the myocardium have been studied intensively , although that of the endocardium remains largely obscure ( Harris and Black , 2010 ) . Early lineage tracing studies in chick have shown that myocardial and endocardial progenitors segregate at primitive streak stages ( Wei and Mikawa , 2000 ) . In line with this view , we previously demonstrated that the endocardium originates , in part , from committed vascular endothelial progenitors that segregate medially from myocardial progenitors within the cardiac crescent ( Milgrom-Hoffman et al . , 2011 ) . Numerous studies have demonstrated the close ontological relationship between the blood and vascular lineages , and the myocardium and endocardium of the forming heart . Heart progenitor cells in the embryo , or those derived from differentiating embryonic stem cells , co-express markers of blood ( Tal1 , Gata1 ) and endothelial ( Flk1 ) lineages; conversely , vascular progenitors in the yolk sac express at some point in their lineage histories multiple cardiac transcription factors including Nkx2 . 5 , Isl1 and Tbx20 ( Keenan et al . , 2012; Stanley et al . , 2002; Stennard et al . , 2005 ) . Studies in both fish and mouse show that the blood and heart-forming territories develop in close proximity , and are mutually antagonistic ( Bussmann et al . , 2007; Schoenebeck et al . , 2007; Van Handel et al . , 2012 ) . These territories appear to represent distinct spatio-temporal domains of the nascent mesoderm regulated in common by the transcription factor Mesp1 ( Chan et al . , 2013 ) . Nkx2 . 5 has long been known as a key transcription factor for cardiac development , specifically in the specification and differentiation of the myocardial lineage ( Evans et al . , 2010; Lyons et al . , 1995; Turbay et al . , 1996 ) . However , the mis-patterning of the heart and arrest of cardiac development in Nkx2-5 null mutant mouse embryos has precluded a deep analysis of the role of this transcription factor in other mesodermal lineages . Cre recombinase-based lineage tracing with Nkx2-5irescre driver mice revealed a contribution of labelled cells to endothelial and blood cells in the yolk sac blood islands ( Stanley et al . , 2002 ) , consistent with severe defects reported in remodelling of the yolk sac vasculature in Nkx2-5 null embryos ( Tanaka et al . , 1999 ) . However , it was not determined whether this was a primary affect of loss of Nkx2-5 in yolk sac , or secondary to hemodynamic effects arising from arrested heart development . Recent studies have begun to shed new light on the role of Nkx2-5 in hemoangiogenic lineage specification . Nkx2-5 was found to directly bind a cis-regulatory element of the Etv2 gene , encoding an ETS domain family transcription factor essential for formation of the embryonic and extra-embryonic endothelium , endocardium and blood lineages , and Nkx2-5 induces its expression in vitro ( Ferdous et al . , 2009 ) . In zebrafish , clusters of nkx2 . 5+ cells segregate from the main nkx2 . 5+ myogenic fields prior to heart formation , giving rise to the pharyngeal arch arteries ( Paffett-Lugassy et al . , 2013 ) . Here , nkx2 . 5 is essential for endothelial cell fate specification and expression of etsrp/etv2 and scl/tal1 . Consistent with these results , genetic lineage tracing in the mouse using the Nkx2-5irescre driver showed a contribution of Nkx2-5+ cells to the pharyngeal arch arteries and aortic sac , and analysis of Nkx2-5 null embryos demonstrated that Nkx2-5 is required for pharyngeal arch artery formation ( Paffett-Lugassy et al . , 2013 ) . Nkx2-5 and Isl1 are expressed in and required for a distinct subset of endocardial cells defined as the hemogenic endocardium , proposed to contribute to primitive and transient definitive hematopoiesis in the embryo ( Nakano et al . , 2013 ) . Finally , expression of the human NKX2-5 gene occurs in a subset of paediatric T and B cell acute lymphoblastic leukaemias defined by a t ( 5;14 ) translocation ( Nagel et al . , 2003; Su et al . , 2008 ) . Despite this knowledge , the origins of hemogenic endothelium and endocardium , and the role of Nkx2 . 5 in their lineage networks in vertebrates , are poorly understood . In this study , we used chick and mouse embryonic models to investigate the origins and plasticity of early cardiovascular progenitors . Using defined early Nkx2-5 and Isl1 embryonic enhancers driving expression of GFP and RFP as surrogate lineage tracing tools in the chick , we revealed the origin of the cardiac hemogenic endocardium to be hemogenic angioblasts in the extraembryonic/lateral plate mesoderm . These mesodermal cells express Nkx2 . 5 already in the primitive streak before migrating to the posterior LPM and extraembryonic regions , and , subsequently , into the endocardium and DA , where they form hemogenic endothelial clusters . We further demonstrated in chick , through gain-of-function experiments , that Nkx2 . 5 induces the expression of hemoangiogenic markers in nascent mesoderm . In mouse , Nkx2-5irescre lineage tracing demonstrated activation of Nkx2-5 in extraembryonic tissue proximal to the cardiac crescent , and Nkx2-5+ cells make a substantial contribution to Runx1+ and CD41+ hemogenic endothelium in the DA and to a lesser extent endocardium . Our data provide the first in vivo evidence that Nkx2 . 5 has a role in initiating the hemoangiogenic program and that Nkx2 . 5+ progenitors contribute via the LPM/extraembryonic region to hemogenic endothelium of the endocardium and AGM region , and subsequently blood . Furthermore , we identify a novel population of endocardial progenitors outside the classical FHF and SHF domains . The chick embryo is highly suitable for the study of early development due to its accessibility for in vivo manipulations such as tissue grafting and cell fate mapping . In order to investigate the spatiotemporal dynamics of early cardiogenesis , we combined bioinformatic enhancer identification with DNA electroporation and time-lapse imaging of enhancer reporters in live chick embryos cultured ex ovo ( Uchikawa et al . , 2004 ) . We first analyzed the expression in chick embryos of a previously characterized 513 bp early mouse Nkx2-5 cardiac enhancer ( Lien et al . , 1999 ) ( Nkx2-5-en; Figure 1A–E; n = 30 ) . Despite apparent lack of sequence conservation with the chick Nkx2 . 5 locus , Nkx2-5-en was able to drive robust reporter gene expression in cardiac progenitors in live chick embryos ( Figure 1B–C ) . Nkx2-5-en expression corresponded well with the mesoderm expression domain of endogenous chick Nkx2 . 5 at embryonic St . 8 ( Figure 1E–E”’ ) . We then identified a conserved 830 bp regulatory element flanking the mouse genomic locus of the Isl1 gene , which acts as an early cardiac enhancer in chick embryos ( Isl1-en; Figure 1F–J’; n = 30 , Video 1; n = 3 ) . The expression of Isl1-en recapitulated the endogenous mesoderm and endoderm ( RNA and protein ) expression domains of chick Isl1 ( Figure 1I–J ) , but was absent from ectoderm ( Figure 1J’ and Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 20994 . 003Figure 1 . Identification of novel Nkx2-5 and Isl1 cardiac enhancers . ( A ) Location of the Nkx2-5 enhancer in the mouse genome . ( B–C ) Expression patterns of the enhancer-driven GFP/RFP formed in the cardiac crescent at St . 8 , compared to corresponding control vectors . ( D ) In situ hybridization of endogenous Nkx2 . 5 compared to ( E ) the mouse Nkx2-5 enhancer . ( E'–E’’’ ) Cross-section of the electroporated embryo after immunostaining for Nkx2 . 5 ( magenta ) , GFP ( green ) , and DAPI ( blue ) . Separated channels for the Nkx2-5-en ( E” ) and Nkx2 . 5 protein ( E’’’ ) . ( F ) . Location of the Isl1 enhancer in chick and mouse genomes . ( G–H ) Expression patterns of the enhancer-driven GFP formed in the cardiac crescent at St . eight for both chick and mouse elements . The control RFP vector is expressed in all cells . ( I ) In situ hybridization of endogenous Isl1 , compared to ( J ) stage-matched mouse enhancer-electroporated embryo . ( J'–J’’’ ) Cross-section of the electroporated embryo after immunostaining for Isl1 ( magenta ) , GFP ( green ) , and DAPI ( blue ) . Separated channels for the mouse Isl1-en ( J” ) and the Isl1 protein ( J’’’ ) . B-E , n = 30; F-J , n = 30 . FHF: first heart field; SHF: second heart field; CPM: cranial paraxial mesoderm; NT: neural tube; CC: cardiac crescent . See also Figure 1—figure supplement 1 , Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 00310 . 7554/eLife . 20994 . 004Figure 1—figure supplement 1 . Characterization of the novel cardiac Isl1 enhancer . ( A–D ) Expression pattern of the Isl1-en at different developmental stages . ( E ) The GFP pattern formed in the cardiac crescent at St . eight overlapped with the Isl1 mRNA expression pattern and its protein expression by staining for Isl1 ( red ) GFP ( green ) and Dapi ( blue ) . ( F–J ) Cross sections through different stages of embryonic development corresponding to the embryos shown in A-D . The mouse Isl1-en ( green , GFP ) and control pCAAG-RFP ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 00410 . 7554/eLife . 20994 . 005Video 1 . Time lapse movie of a developing embryo expressing the Isl1 enhancer . Relates to Figure 1 . An embryo electroporated with the Isl1-en ( GFP ) plasmid at St . 3 . The time-lapse video follows from St . 5to St . 11 when the heart tube has already formed and the embryo has begun to turn . oft: outflow; inf: inflow . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 005 To study the dynamics of cardiac progenitor specification and migration in vivo , we co-electroporated Isl1 and Nkx2-5 enhancers into primitive streak-stage chick embryos ( Figure 2A ) . Embryos were incubated overnight and monitored from St . 7 onwards . Live-cell imaging revealed the presence of both common and distinct cell populations within the cardiac crescent and adjacent developing area vasculosa ( Figure 2B; n = 30 , Video 2; n = 3 ) . At St . 7–8 , Nkx2-5-en-GFP and Isl1-en-RFP double-positive cells ( yellow ) migrated to the anterior and lateral regions of the embryo and contributed to the cardiac crescent ( Figure 2B , C ) . However , a small proportion of electroporated cells expressed either GFP or RFP alone ( Figure 2B–E ) . The anterior part of the heart tube ( outflow region ) preferentially contained GFP+ ( Nkx2 . 5+ ) cardiac progenitors , while the posterior part ( inflow ) was mostly RFP+ ( Isl1+ ) ( Figure 2B–E ) . This is consistent with studies showing that the myocardium of the inflow pole of the mouse heart can form from Isl1+ progenitors in the absence of Nkx2 . 5 expression ( Christoffels et al . , 2006 ) . 10 . 7554/eLife . 20994 . 006Figure 2 . Nkx2-5 and Isl1 enhancers mark distinct cardiac progenitor populations . ( A ) Experimental design of the ex ovo electroporation technique . An electric field is used to introduce different plasmids into the embryo at gastrulation ( St . 3 ) , followed by EC culture and imaging . ( B–E ) Images taken from a 24 hr time-lapse video of an embryo co-electroporated with the Nkx2-5 ( GFP ) and Isl1 ( RFP ) enhancers from St . 7 to St . 11 . Higher magnification panels ( B’’’–E’’’ ) of the dotted square boxes highlight the differential Nkx2-5/Isl1 enhancers expression . Green arrow: Nkx2-5-en+ cells; Red arrow: Isl1-en+ cells; Yellow arrow: Nkx2-5-en+/Isl1-en+ cells . CC: cardiac crescent; LPM: lateral plate mesoderm; HT: heart tube; smt: somites . Scale bar: 100 µm . See also Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 00610 . 7554/eLife . 20994 . 007Figure 2—figure supplement 1 . Migration of Nkx2-5-en+ cells from outside the cardiac crescent towards the heart tube . ( A–H ) Time lapse analysis of a live chick embryo expressing GFP under the control of Nkx2-5-en analyzed at different time points . A single cell ( white arrow and circle ) migrating from the border of the extraembryonic/LPM and the cardiac heart field into the developing heart . Higher magnification images of the red boxes are included in the upper left corner of each panel and bring to view the traced cell . The GFP ( Nkx2-5-en+ cells ) and BF channels are separated . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 00710 . 7554/eLife . 20994 . 008Video 2 . Time lapse movie of a developing embryo co-expressing the Nkx2-5 and Isl1 enhancers . Relates to Figure 2 . A St . 3 embryo was co-electroporated with the Isl1-en ( RFP ) and Nkx2-5-en ( GFP ) plasmids . The embryo was cultured overnight and time-lapse analysis started at St . 6 . The embryo was allowed to grow to St . 11–12 when the heart tube is visible and begins to loop . The BF channel was omitted from the movie to enhance the visibility of distinct cardio-vascular progenitor populations . cc: cardiac crescent; lpm: lateral plate mesoderm; ee: extraembryonic; da: dorsal aorta; shf: second heart field . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 008 By St . 9–10 , FHF/cardiac crescent cells have already fused into the primitive heart tube , which continues to grow with the addition of SHF cells at its inflow and outflow poles ( Figure 2C–E ) . In the chick and mouse , Nkx2 . 5 is expressed throughout cardiac development , while Isl1 expression is transient and progressively downregulated as heart progenitors differentiate into specialised cardiomyocytes ( Nathan et al . , 2008 ) . Consistently , the middle part of the primary heart tube exhibited stronger GFP expression ( Figure 2E ) . Surprisingly , we could detect a trail of GFP+/RFP- ( Nkx2-5-en+Isl1-en- ) cells extending from the LPM and extraembryonic regions towards the heart ( Figure 2D , green arrowheads ) . This expression was not observed for Isl1-en-RFP , which was restricted to the primitive heart tube including both outflow and inflow tracts . Further time-lapse live imaging suggested that Nkx2-5-en+ ( Isl1-en- ) progenitors migrate into the forming heart via the inflow tract ( Figure 2D , Figure 2—figure supplement 1 , Video 2 ) . Taken together , our findings reveal the existence of cardiac progenitors within and outside the cardiac crescent that differ in their Isl1 and Nkx2 . 5 spatiotemporal expression patterns . Our dynamic enhancer imaging suggests that Nkx2-5-en-GFP marks a broad mesoderm population with both cardiac and angioblastic characteristics . The GFP+ angioblast progenitors form a continuum between the cardiac crescent defined by Isl1-en-RFP , and LPM and extraembryonic mesoderm , and are incorporated into the forming heart tube . To explore this idea further , we used a hemangioblast enhancer ( Hb-en ) located in the cis-regulatory region of the chick Cerberus gene , which marks blood islands and migratory angioblasts in the yolk sac ( Teixeira et al . , 2011 ) . Co-electroporation of Nkx2-5-en-RFP and Hb-en-GFP enhancers at St . 3 with analysis at St . 7 revealed Nkx2-5-en-RFP+ progenitors within the cardiac crescent in a pattern similar to that of Nkx2 . 5 mRNA , as well as in LPM and extraembryonic mesoderm ( Figure 3A–B ) . Hb-en-GFP was expressed only in posterior LPM/extraembryonic mesoderm bordering the area vasculosa of the yolk sac ( Figure 3A–C; n = 17/20 ) . Many cells co-expressed Nkx2-5-en-RFP and Hb-en-GFP within the LPM ( Figure 3A’’’ , yellow arrowheads ) . The Nkx2-5-en expression domain in the LPM/extraembryonic region overlapped with that of the endothelial/blood marker Tal1 ( Figure 3C ) . 10 . 7554/eLife . 20994 . 009Figure 3 . The Nkx2-5 enhancer is expressed in a hemangiogenic cell population . ( A–A''' ) A St . 7 embryo co-electroporated with the Nkx2-5-en ( RFP ) and Hb-en ( GFP ) . The dashed line marks the boundaries of the cardiac crescent . Red arrows represent Nkx2-5-en+ cells in the cardiac crescent; yellow arrows represent Nkx2-5-en+/Hb-en+ cells; green arrows represent Hb-en+ cells ( B ) In situ hybridization for the endogenous Nkx2 . 5 mRNA of a St . 7 embryo . ( C ) In situ hybridization for the endogenous Tal1 mRNA of a St . 7 embryo . Tal1 expression pattern is compared to a representative Nkx2-5-en expression pattern from different embryo . CC: cardiac crescent; AP: area pellucida; LPM: lateral plate mesoderm; YS: yolk sac . Scale bar: 100 µm . n = 17/20 . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 00910 . 7554/eLife . 20994 . 010Figure 3—figure supplement 1 . The Hb enhancer is active in endothelial and blood progenitors . ( A ) In situ hybridization of Tal1 mRNA at St . 8 ( B–B" ) The Hb-en is expressed in the extra-embryonic region with a similar pattern to Tal1 endogenous mRNA . Expression can also be detected in the cardiac crescent where endothelial cells give rise to the endocardium . ( C ) Example of a posterior LPM explant expressing the Hb-en , cryo-sectioned and ( D–E"' ) immune-stained for early hematopoietic and endothelial markers . ( F ) A posterior LPM explant expressing the Nkx2-5-en ( red ) cryo-sectioned and stained for early hematopoietic marker Runx1 . ( F’–F’’’ ) is a higher magnification image of the square box in F . Co-expressing cells are marked by white arrow heads . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 010 We next explored the ability of Hb-en+ and Nkx2-5-en cells to generate vascular endothelial and hemogenic cells in vitro by culturing posterior LPM explants following the electroporation of the two reporters ( Figure 3—figure supplement 1 ) . GFP+ and RFP+ cells in LPM explants expressed endothelial ( Cdh5 ) and hemogenic ( Runx1 and CD45 ) mesodermal markers ( Figure 3—figure supplement 1 ) , suggesting that Hb-en+/Nkx2-5-en+ double positive enhancer labels hemogenic angioblasts located in the yolk sac and LPM . Based on the anatomical location , we hypothesize that Nkx2-5-en/Hb-en double positive cells represent hemogenic angioblasts ( Tanaka et al . , 2014; Teixeira et al . , 2011 ) and that Nkx2-5-en marks distinct cardiomyocyte and hemogenic angioblast populations in the chick embryo . We next assessed the contribution of Hb-en+ hemogenic angioblasts to the developing heart tube by electroporating Hb-en-GFP at St . 3 , with analysis at St . 6–11 ( Figure 4A–G ) . Live-imaging analysis revealed GFP+ cells outside the cardiac crescent at St . 6 that were eventually incorporated into the heart at St . 11 ( Video 3; n = 3 ) . These cells or their daughter cells ( Figure 4A–G , orange and blue arrowheads ) migrated towards and incorporated into the venous pole of the heart ( Figure 4B–E ) , while some daughters remained in the vitelline vasculature . 10 . 7554/eLife . 20994 . 011Figure 4 . Hemangiogenic progenitors migrate to the heart through the inflow tract , and contribute to the endocardium . ( A–G ) Time-lapse images taken from an embryo electroporated with Hb-en ( GFP ) . Images span a 24 hr time frame , from St . 6 to St . 11 of the same embryo . The orange and cyan circles distinguish between two distinct Hb-en+ cells . Cells were manually marked and tracked throughout the 24 hr window . ( H ) A St . 11 embryo co-electroporated with Nkx2-5-en and Hb-en . The embryo was sectioned at the level of the heart to show the distribution of cells within the heart tube . Arrows mark Hb-en+ ( green ) , Hb-en+/Nkx2-5-en+ ( yellow ) and Nkx2-5-en+ ( red ) cells . ( H’–H’’’ ) include higher magnification images of single channels from the square box in H . ( I ) A St . 11 embryo electroporated with Hb-en . The section was immunostained for Nkx2 . 5 ( magenta ) , and its expression pattern compared to that of Hb-en . ( J ) An embryo electroporated with the Hb-en-GFP and control vector pCAGG-RFP at St . 12 . ( J’ ) Cryo-section through the outflow tract region ( OFT ) of the heart shown in J . ( J” ) Cryo-section through the inflow tract region ( IFT ) of the heart shown in J . Yellow arrows correspond to Hb-en+ /pCAGG+ cells in the endocardium . ( K ) Quantification of Hb-en+ cell contribution to the myocardium compared to the contribution of pCAGG-RFP+ cells . ( L ) Quantification of Hb-en+ cell contribution to the endocardium relative to the contribution of pCAGG-RFP+ cells . The quantifications represent two independent experiments with n = 3 . For both quantifications , double positive cells ( RFP+/GFP+ ) were manually counted and compared to the total number of RFP+ cells . ( M ) Schematic representation of relative Hb-en+ cell contribution to different parts of the endocardium . Hb-en+ cells numbers are increased towards the venous ( IFT ) pole of the heart . CC: cardiac crescent; IFT: inflow tract; OFT: outflow tract; vent: ventricle; myo: myocardium; end: endocardium . Scale bars: 100 µm . n = 9/12 . See also Video 3 and Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 01110 . 7554/eLife . 20994 . 012Figure 4—source data 1 . The distribution of Hb-en+ cells in the chick embryo . Analysis is based on six embryos from two independent experiments . The first column represents the tissues checked for Hb-en+ cells . The second column represents the percentage of embryos in which Hb-en+ cells were detected . The third column represents the percentage of double positive Hb-en+/CD45+ cells out of the total CD45+ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 01210 . 7554/eLife . 20994 . 013Figure 4—figure supplement 1 . Double positive Nkx2-5-en+/Hb-en+ cells migrate towards the heart and yolk sac . Time-lapse images taken from an embryo electroporated with Hb-en ( GFP ) and Nkx2-5-en ( RFP ) . Images span an 18 hr time frame , from St . six to St . 11 of the same embryo . Images were taken in a way that the cardiac crescent , LPM and extraembryonic tissues are visible simultaneously . ( A–E’ ) Tracking an Hb-en+/Nkx2-5-en+ cell at the border between the cardiac crescent and the LPM ( white arrow ) . ( D–J’ ) Tracking an Hb-en+/Nkx2-5-en+ cell that is located outside of the cardiac crescent and eventually incorporated into the heart tube ( cyan arrow ) . Note the large number of double positive cells extending from the LPM towards the extraembryonic tissue that are migrating to the yolk sac in a circulation independent matter . that cc - cardiac crescent; ee - extraembryonic tissue; lpm - lateral plate mesoderm; nt - neural tube . Scale bar: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 01310 . 7554/eLife . 20994 . 014Video 3 . Time lapse movie of a developing embryo expressing the Hb enhancer . Relates to Figure 4 . A St . 3 embryo electroporated with the Hb-en plasmid ( GFP ) and cultured overnight . The movie covers the development of the embryo between St . 6 to St . 11–12 . The BF and GFP channels were merged to better assess the migration of cells towards the inflow of the heart . Hb: hemangioblast; ee: extraembryonic; cc: cardiac crescent; nt: neural tube; hn; hensen’s node; ap: area pellucida; da: dorsal aorta; oft: outflow; inf: inflow . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 014 We next co-electroporated embryos with both Nkx2-5-en and Hb-en plasmids and followed the fate of double-labelled cells . Migration of Hb-en+/Nkx2-5-en+ cells towards the heart and yolk sac could be tracked ( Figure 4—figure supplement 1 , Video 4 ) . Histological sections showed that expression of Hb-en ( GFP+ , green arrow ) was restricted to the endocardium ( Figure 4H–I ) , while Nkx2-5-en expression ( RFP+ , red arrow ) was detected in both myocardium and endocardium ( Figure 4H; n = 12 ) . Importantly , Nkx2-5-en+/Hb-en+ ( yellow ) cells were found only in the endocardium ( Figure 4H , yellow arrows; Figure 4—source data 1 ) . Nkx2 . 5 protein was expressed only in the myocardium at this stage ( Figure 4I ) , indicating that Nkx2-5-en marks cells expressing Nkx2 . 5 protein at an earlier progenitor stage . 10 . 7554/eLife . 20994 . 015Video 4 . Time lapse movie of a developing embryo co-expressing the Hb and Nkx2-5 enhancers . Relates to Figure 4 . A St . 3 embryo electroporated with both the Hb-en ( GFP ) and Nkx2-5-en ( RFP ) plasmids subsequently cultured overnight . The movie covers the development of the embryo between St . 6 to St . 11 . The GFP and RFP channels were merged to visualize the migration of Hb-en+/Nkx2-5-en+ cells towards the inflow and the yolk sac . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 015 To quantify the contribution of hemogenic angioblasts to the endocardium , Hb-en-GFP was co-electroporated with a control pCAGG-RFP vector as a reference for electroporation efficiency . At St . 12 , a large number of cells across the heart were pCAGG-RFP+ , while Hb-en-GFP+ cells could be detected in a salt and pepper pattern ( Figure 4J ) . Embryos were subsequently cryo-sectioned at the level of the heart tube and cells positive for both GFP and RFP were quantified separately in the myocardium and endocardium ( Figure 4J’–J” ) . As expected , the contribution of Hb-en+ cells to the myocardium was <0 . 1% ( Figure 4K ) . In the endocardium , ~15% of electroporated pCAGG-RFP+ cells expressed Hb-en-GFP ( Figure 4J–J’’ , L ) . Further analysis revealed that the Hb-en+ contribution was graded slightly but significantly between the two poles of the heart - from 13% at the arterial pole ( outflow tract ) to 18% at the venous pole ( inflow tract ) where the heart connects with the vitelline veins ( Figure 4M ) . Taken together , our results reveal a novel population of cardiovascular progenitors located in the LPM/extraembryonic region outside the cardiac crescent , which contributes specifically to the endocardium . This progenitor population is distinct from previously characterized SHF endocardial progenitors located medial to myocardial progenitors in the cardiac crescent ( Milgrom-Hoffman et al . , 2011 ) . Formation of the DA involves the integration of early endothelial cells from the LPM ( Sato , 2013 ) . We next determined whether Hb-en+/Nkx2-5-en+ cells contribute to the endothelial layer of the DA . To visualize endothelium as well as hemogenic endothelium in the DA , we used our enhancers in conjunction with a set of markers including VE-cadherin ( Cdh5 ) , CD45 and Runx1 ( Figure 5 ) . While Cdh5 labelled the entire DA endothelium , CD45 expression was restricted to hemogenic clusters in the floor of the DA at St . 17 ( Figure 5A ) . To investigate the contribution of Hb-en+ cells to the hemogenic endothelium , embryos were electroporated with Hb-en and cultured for 3 days until St . 14–15 . Hb-en expression was detected in the endothelial layer of the floor of the DA ( n = 4/6 ) , as well as in the endocardium ( n = 6/6 ) but not in the cardinal vein ( Figure 5B , Figure 4—source data 1 ) . Co-expression of the Hb-en-GFP with CD45 and Runx1 revealed hemogenic clusters in the DA , highlighting the hemogenic angioblast origin of the hemogenic endothelium ( Figure 5B ) , in accordance with previous findings ( Lancrin et al . , 2009; Tanaka et al . , 2014 ) . We propose that hemogenic cells are incorporated into the floor of the DA during its formation from LPM . 10 . 7554/eLife . 20994 . 016Figure 5 . Hb-en+ and Nkx2-5-en+ cells contribute to the hemogenic endothelium of the dorsal aorta . ( A–A” ) A St . 17 embryo sectioned at the level of the AGM and stained with novel anti Cdh5 ( VE-cadherin ) and CD45 antibodies delineating the endothelium and hematopoietic cells , respectively . ( B–B''' ) St . 14–15 embryo electroporated with the Hb-en ( GFP ) and sectioned at the level of the early fused dorsal aorta stained for CD45 , Runx1 and GFP . The arrow marks the initiation of hemogenic clusters indicated by CD45 expression . ( C ) Section of a St . 14 embryo expressing the Nkx2-5-en in the right aspect of the newly fused dorsal aorta ( arrowheads ) . While no CD45 expression is detected the Nkx2-5-en expression co-localized with Cdh5 . ( D–D" ) Section of an early St . 14 embryo expressing the Nkx2-5-en . Hemogenic endothelium cells are marked by Runx1 which precedes the expression of CD45 . DA–dorsal aorta . Scale bars: 100 μm ( A ) , 20 μm ( B–D ) . Scale bar: 100 µm . n = 4/6 . See also Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 016 We also found Nkx2-5-en+ cells in the endothelial layer of the DA that co-expressed Cdh5 ( Figure 5C , white arrows; n = 5/7 ) . Earlier , at St . 13–14 , these cells were localized at the lateral edge of the newly fused DA and failed to express CD45 , although , Runx1 , an early hematopoietic stem cell marker known to be activated in hemogenic cells of the DA ( North et al . , 2002 ) , was co-expressed with Nkx2-5-en+ cells at this stage ( Figure 5D ) . These results suggest that Nkx2 . 5+ hemogenic angioblasts contribute to the endothelial layer and hemogenic endothelium of the DA during its formation . We next sought to investigate the contribution of Hb-en+ and Nkx2-5-en+ cells to hemogenic endocardium , recently reported to derive from the Nkx2-5+ lineage in the mouse ( Nakano et al . , 2013 ) . Currently , the origin of these cells is unknown . Endocardial cells co expressing Nkx2-5-en+ and Cdh5+ could be detected in St . 13–14 chick embryos ( Figure 6A ) and we found that a subset of these cells had the same marker profile as in the DA ( Figure 6B ) . At the level of the developing ventricle , we observed Hb-en-GFP+ cells co-expressing either CD45 or Runx1 , or both ( Figure 6C–D ) . Taken together , our data suggest that the hemogenic endothelium within the primitive heart tube of the chick embryo is derived from hemogenic angioblasts within the LPM/extraembryonic mesoderm . 10 . 7554/eLife . 20994 . 017Figure 6 . Hb-en+ and Nkx2-5-en+ cells give rise to hemogenic endocardium . ( A ) Section through a heart of a St . 12–13 embryo expressing the Nkx2-5-en . Staining for Cdh5 delineates the endocardium ( white arrow ) which expresses the Nkx2-5-en . ( B ) High magnification of a section through the heart of a St13-14 embryo expressing Nkx2-5-en . A hemogenic cluster is attached to the endocardium delineated by Cdh5 and CD45 staining . ( C ) St . 14–15 embryo electroporated with the Hb-en sectioned at the level of the heart . ( C'–C''' ) Higher magnification of the square box area marked in C . The section was stained for CD45 and shows a large hemogenic cluster in the endocardium . The arrow indicates Hb-en+ cells that express CD45 . The endocardium and myocardium are delineated by grey and pink dashed lines , respectively . The dashed red line delineates the hemogenic cell cluster . ( D ) St . 14–15 embryo electroporated with the Hb-en and sectioned at the level of the heart tube . ( D'–D''' ) Higher magnification of the square box area marked in D . The section was stained for CD45 , Runx1 and GFP . White arrow highlights an Hb-en+ cell co-expressing CD45 and Runx1 . For each dotted red line a higher magnification image is shown at the upper right side of the panel . end: endocardium; myo: myocardium . Scale bars: 100 µm , B – 20 µm . n = 4/6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 017 To examine the potential of Nkx2 . 5 to induce cardiovascular/blood lineage commitment in chick embryos , we first compared the expression profile of key cardiovascular and blood transcription factors at St . 7 ( cardiac crescent stage; Figure 7A ) . Regional expression of these factors can be seen along the medial-lateral and anterior-posterior axes in zones generally defined as ‘cardiac’ ( Tbx5 , Isl1 , Nkx2 . 5 ) , ‘cardiovascular’ ( Isl1 , Nkx2 . 5 , Gata4 , Tal1 ) , and ‘blood’ ( Tal1 , Runx1 ) -forming territories . In chick , Tbx5 is expressed in FHF cells as they begin to differentiate into cardiomyocytes ( Nathan et al . , 2008; Plageman and Yutzey , 2004; Tirosh-Finkel et al . , 2006 ) , while Nkx2 . 5 and Isl1 are expressed throughout the anterior LPM in both FHF and SHF progenitors and pharyngeal endoderm ( Nathan et al . , 2008; Tirosh-Finkel et al . , 2006 ) . Gata4 is expressed more lateral and posterior to these two factors . The endothelial/blood marker Tal1 is highly expressed in the posterior LPM and extraembryonic tissues , whereas the blood/hematopoietic lineage marker Runx1 marks only extraembryonic blood islands , more lateral to cells expressing Tal1 . This ISH analysis reveals the existence of both overlapping and distinct developmental fields ( illustrated in Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 20994 . 018Figure 7 . Nkx2 . 5 is expressed in the nascent mesoderm in the chick primitive streak . ( A ) The cardiovascular network gradient . In situ hybridization of key genes involved in blood , vascular and cardiac development . ( B ) Semi qRT-PCR analysis of cardiovascular genes in early St . 3 ( red box ) and St . 4 ( blue box ) embryos . ( C–C’ ) A St . 3–4 embryo electroporated with the Nkx2-5-en . ( C’ ) shows a higher magnification of the square box in ( C ) , displaying the GFP channel and the merge with the BF channel . ( D ) Wholemount immunofluorescence staining for Nkx2 . 5 on late St . 4 embryo . ( E ) Wholemount immunofluorescence of St . 8 embryo . Nkx2 . 5 expression is restricted to the cardiac crescent . ( F ) Semi qRT-PCR time course analysis of Nkx2 . 5 mRNA expression at different embryonic stages and tissues . Nkx2 . 5 expression is highlighted in yellow boxes . GAPDH is used as control . CC: cardiac crescent; NT: neural tube; PS: primitive streak . Scale Bars – C: 100 µm; C’: 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 01810 . 7554/eLife . 20994 . 019Figure 7—figure supplement 1 . Key cardiovascular gene expression gradient in a chick embryo . ( A ) Super-position of the different cardiovascular factors on a single embryo based on in situ hybridization and semi qRT-PCR analysis . ( B ) Semi qRT-PCR analysis of cardiovascular genes in a St . 8- embryo . The embryo is divided to four sections that are manually dissected and subjected to RNA analysis . Representative results of three different biological repeats . lpm – lateral plate mesoderm; np – neural plate; cc – cardiac crescent; ee – extraembryonic tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 019 Nkx2 . 5 mRNA expression at St . 7 is restricted to the cardiac crescent , while the enhancer is also expressed in the posterior LPM ( Figures 2 , 3 and 7 ) . Enhancer activity in the LPM could be explained by the lack of negative cis regulatory elements within this fragment . Alternatively , it could reflect the stability and perdurance of the fluorescent reporters ( GFP or RFP ) , which would serve as lineage tracing tools marking all cells that have expressed Nkx2 . 5 in their recent history . To address this issue , we performed RT-PCR analysis to ask whether Nkx2 . 5 is expressed in eartly gastrulating St . 3–4 embryos , when early cardiovascular progenitors emerge ( López-Sánchez and García-Martínez , 2011 ) . At these early stages cardiac progenitors were localized at the anterior primitive streak , while blood progenitors and Nkx2-5-en+ were restricted to the mid-posterior streak ( Schoenwolf et al . , 1992; Schultheiss et al . , 1995 ) . Analysis of mid-primitive streak tissue detected Nkx2 . 5 , but not Isl1 , RNA expression , along with the earliest markers of mesoderm , Mesp2 ( homolog of mouse Mesp1 ) ( Saga et al . , 1999 ) and Brachyury ( Bry ) ( Figure 7B; red and blue boxes , n = 18 ) . Hemoangiogenic markers Ets1 , Flk1 , Gata4 , CD31 , Runx1 and Tal1 , were also detected at these stages . Consistent with these finding , we detected expression of Nkx2-5-en in the mid-streak at these same stages , marking nascent mesodermal progenitors . Wholemount immunofluorescence staining of St . 4 embryos revealed multiple Nkx2 . 5+ cells in and outside the primitive streak region ( Figure 7D ) , while at St . 8 Nkx2 . 5 was expressed exclusively in the cardiac crescent ( Figure 7E ) . Nkx2 . 5 mRNA was also transiently expressed in the posterior LPM at St . 5–6 ( Figure 7F , yellow box ) in addition to strong and sustained expression within the anterior LPM ( cardiac crescent ) . Lower expression was observed in extraembryonic tissue at St . 6 , which was downregulated when embryos reached St . 7 , in-line with inhibition of the cardiac developmental program in the posterior LPM and extraembryonic tissue ( Schoenebeck et al . , 2007; Van Handel et al . , 2012 ) . In-depth RT-PCR analysis at St . 7 embryos was performed for the extraembryonic mesoderm , LPM , cardiac crescent and neural plate tissues ( Figure 7—figure supplement 1 , n = 20 ) . A mixture of endothelial and blood gene expression signatures ( Flk1 , Tal1 , Ets1 , CD31 , Gata1 , Runx1 ) was observed in both extraembryonic and LPM tissues , while the cardiac crescent expressed Nkx2 . 5 , Isl1 and Tbx5 ( Figure 7—figure supplement 1 ) . Taken together , our molecular analyses suggest that in the chick transient expression of Nkx2 . 5 marks hemoangiogenic progenitors in extra-cardiac regions from early primitive streak stages through St 6 , just prior to formation of the cardiac crescent . These results are consistent with Nkx2-5-en serving as a lineage-tracing tool for Nkx2 . 5-expressing mesodermal cells in gastrula and immediate post-gastrula stages . Our findings suggest that Nkx2 . 5 marks hemoangiogenic progenitors in chick nascent mesoderm . To examine the competence of nacent mesoderm to respond to Nkx2 . 5 expression , we performed Nkx2 . 5 gain-of-function experiments by electroporating the pCIG-Nkx2 . 5 construct at St . 3 , with analysis at St . 7 ( cardiac crescent stage , Figure 8 ) . Strikingly , enforced expression of Nkx2 . 5 at St . 3 induced robust expression of hemoangiogenic genes , including Tal1 , Flk1 , Ets1 and Lmo2 ( red arrows , n = 6/8; n = 8/11; n = 4/5; n = 5/7 , respectively , Figure 8A’–D’ ) . Ectopic expression of these genes was observed at sites of high Nkx2 . 5 expression ( lower panels ) , indicating a cell autonomous effect . Furthermore , Nkx2 . 5 overexpression inhibited somite formation and Pax3 expression , indicating strong lateralization of the paraxial mesoderm ( Figure 8E–E”; n = 5/6 ) . 10 . 7554/eLife . 20994 . 020Figure 8 . Ectopic expression of Nkx2 . 5 induces angioblast gene expression . ( A–E ) In situ hybridization for Tal1 , Flk1 , Ets1 and Lmo2 and Pax3 in control embryos . ( A’–E’ ) Embryos electroporated at St . 3 with a vector over-expressing Nkx2 . 5 ( pCIG-Nkx2 . 5-GFP ) . Embryos were subsequently incubated to St . 7 and then subjected to in situ hybridization for the respected genes . Red arrow indicates where the ectopic expression of a gene was detected compared to its control embryo . ( A”–E” ) The same embryos as in A'–E' merged with the GFP channel visualizing the Nkx2 . 5 over-expressing tissues . The over-expressing phenotype corresponds to the area where the vector expression was the highest . Scale bar: 100 μm . See also Figure 9—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02010 . 7554/eLife . 20994 . 021Figure 8—figure supplement 1 . Tal1 over-expression inhibits cardiogenesis . The coding sequence of chick Tal1 was cloned into an expression vector and electroporated into St . 3 embryos . ( A , C , E , G ) Expression profile in WT control embryos . ( B , D , F , H ) Expression profile in Tal1 gain-of-function embryos . ( B' , D' , F' , H' ) GFP expression marks the cells where Tal1 is expressed . ( B" , D" , F" , H" ) Overlay of the GFP signal and the in-situ hybridization assay . The yellow dotted line represents the normal borders of the cardiac crescent . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02110 . 7554/eLife . 20994 . 022Figure 8—figure supplement 2 . Tal1 over-expression impairs normal heart development . ( A ) Control St . 13 WT embryo . ( B ) Tal1 over-expressing embryo with edema . ( C ) Tal1 over-expressing embryo with defected cardiac looping and size . ( D ) Tal1 over-expressing embryo with ectopic intra-embryonic blood islands . ( E–I ) Higher magnification of the square box in D . Tal1 – GFP; Nkx2-5-en – RFP . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 022 In a similar fashion , we performed overexpression experiments with Tal1 ( Figure 8—figure supplements 1 and 2 ) . Tal1 overexpression suppressed cardiac genes , mostly within the caudal part of the cardiac crescent ( Figure 8—figure supplement 1 , n = 21/27 ) . Furthermore , the cardiac crescent appeared smaller in Tal1 electroporated embryos . At later stages ( St . 12–13 ) , Tal1 expression resulted in cardiac looping defects and the appearance of edemas ( Figure 8—figure supplement 2A–C , n = 14/17 ) . Furthermore , we observed ectopic blood island formation in the posterior LPM , where Nkx2-5-en was also expressed ( Figure 8—figure supplement 2D–I , n = 10/17 ) . Collectively , our data indicate that in the chick , nascent mid-streak mesoderm transiently expresses Nkx2 . 5 mRNA . Furthermore , it is competent to respond to Nkx2 . 5 over-expression through engagement of a hemoangiogenic program . Overexpression of Tal1 , a definitive hemogenic transcription factor , overides the endogenous cardiac program and redirects it to blood . Our data suggest that Nkx2-5-en provides a robust readout of Nkx2 . 5+ lineage expansion and migration in the chick , and that the first role for Nkx2 . 5 in the embryo is as a hemoangiogenic transcription factor , before its latter involvement in cardiac regulatory networks . Previous CRE/loxP genetic lineage tracing studies in mouse embryos have demonstrated a contribution of Nkx2-5 lineage+ cells to yolk sac blood island endothelium at E10 . 5 ( Stanley et al . , 2002 ) and hemogenic endocardium at E9 . 5 ( Nakano et al . , 2013 ) . However , the origins of these cells and their contribution to the AGM region of the DA have not been explored . We crossed the Nkx2-5irescre mice to Rosa-LacZ , Rosa-YFP or Z/EG reporter lines , and examined embryos for marked tissues at E7 . 5-E10 . 5 . Whole mount staining for β-gal in Nkx2-5irescre/+;RosalacZ/+ embryos revealed the presence of isolated labelled cells in extra-embryonic yolk sac mesoderm at ~E70 . 5-E8 . 0 , coincident with the earliest first appearance of labelling in the cardiac crescent , although not in pre-crescent embryos ( Figure 9A; Figure 9—figure supplement 1A , A’ ) . More labelled yolk sac cells were evident during heart tube formation ( Figure 9B ) . Labelled cells tended to be proximal to the crescent and restricted to the anterior half of the yolk sac ( Figure 9A , B ) . As development progressed further , the number of labelled cells increased dramatically , such that by E9 . 0 there were many positive cells associated with the yolk sac vasculature ( Figure 9D , Figure 9—figure supplement 1D , D’ ) . At this stage , isolated lineage-tagged cells were also seen within the cardio-pharyngeal region of the embryo extending into the head , as well as more caudally in the trunk and tail ( Figure 9C ) . Histological sections of yolk sac showed that from E8 . 0 , lineage tagged cells contributed only to the mesodermal layer including the endothelial lining of forming vessels ( Figure 9—figure supplement 1B–D’’ ) . Moreover , at E9 . 0 , the majority of Nkx2-5 lineage-traced cells in the yolk sac vasculature also expressed endothelial markers Vegfr2/Flk1 and VE-cadherin/Cdh5 ( Figure 9—figure supplements 2A–E’’’ ) , suggesting their endothelial nature . However , the yolk sac vasculature was a mosaic of lineage labelled and unlabelled cells ( Figure 9—figure supplement 1D’ ) . Labelled blood cells were also evident within vessels ( Figure 9—figure supplement 1D , D’’ ) . We note that Nakano and colleagues did not see significant labelling in yolk sac at E9 . 5 using identical ( Nkx2-5irescre/+;Rosa26lacZ/+ ) or similar ( Nkx2-5irescre/+;Rosa26YFP/+ ) lineage tracing crosses ( Nakano et al . , 2013 ) , which we attribute to the lower sensitivities of CRE reporters on different genetic backgrounds . 10 . 7554/eLife . 20994 . 023Figure 9 . An Nkx2-5+ lineage contributes to hemogenic endothelium of the dorsal aorta in the mouse . ( A–D ) Lineage tracing β-gal staining of Nkx2-5irescre;R26R embryos . ( A ) E7 . 5 showing β-gal staining beginning in the cardiac crescent ( cc ) and yolk sac ( ys ) . ( B ) E8 . 0 showing β-gal staining in the forming heart ( fh ) and yolk sac . Red arrows in A and B indicate β-gal positive cells in the anterior half of yolk sac . ( C ) E9 . 5 showing β-gal staining in the looped heart ( ht ) and associated pharyngeal region extending into the head ( potentially angiogenic cells or blood cells ) . Isolated cells are also found within the trunk and tail regions of the embryo ( red arrows ) . ( D ) E9 . 5 showing β-gal staining in the yolk sac marking angiogenic and potentially hemoangiogenic cells within the vasculature ( red arrows; see sections in Figure 9—figure supplement 1B–D’ ) . ( E–H ) Nkx2-5 expression β-gal staining of Nkx2-5lacZ/+ embryos . ( E–F ) E7 . 5 showing Nkx2-5-LacZ expression in the cardiac crescent and proximal yolk sac . Red arrows show single β-gal positive cells in yolk sac in close proximity to the cardiac crescent . ( G–H ) E8 . 0 ( G ) and E9 . 0 ( H ) embryos showing Nkx2-5-LacZ expression in the forming heart and looped heart respectively . Red arrows show caudal parts of the cardiac zone that contain positive cells in yolk sac . ( I–I’’’ ) E10 . 5 Nkx2-5irescre;ROSAYFP embryos sectioned at the level of the aorta gonad mesonephros ( AGM ) region and immunostained for Runx1 ( red ) and YFP ( green ) . Co-labelled Runx1+ YFP+ cells are present mostly in the floor of the dorsal aorta ( da ) . ( J–J’’’ ) E10 . 5 Nkx2-5irescre;ROSAYFP embryo depicting Nkx2-5 lineage+ ( YFP+ ) hemogenic cells ( Runx1+ ) protruding from the endothelial layer of the dorsal aorta in the AGM region . ( K–L’’’ ) E10 . 5 embryos showing co-staining for Nkx2-5-lineage+ ( YFP+ ) and Runx1+ ( K–K’’’ ) or CD41+ ( L–L’’’ ) cells in the developing liver . Yellow arrows in I-K’’’ mark YFP+ Runx1+ cells and in L-L’’’ mark YFP+ CD41+ co-labelled cells . Green arrows in J and J’’ depict YFP+ Runx1- single positive cells . Scale bar: A-H - 100 µm; I-J’’ - 50 µm; K-L’’ - 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02310 . 7554/eLife . 20994 . 024Figure 9—figure supplement 1 . Nkx2-5 is transiently expressed in yolk sac mesoderm from cardiac crescent stages . ( A ) Lateral view of whole mount β-gal staining in an E7 . 0 mid-gastrula-stage Nkx2-5+/+;R26R ( control ) embryo . Weak background is detected sporadically across the yolk sac ( ys ) . ( A’ ) Lateral view of an E7 . 0 Nkx2-5irescre/+;R26R embryo . Note lack of β-gal-positive cells in the embryo and yolk sac . ( B–D’’ ) Cryo-sections of Nkx2-5irescre/+;R26R embryos through the forming heart ( ft ) and yolk sac at E8 . 0 , E8 . 5 , and E9 . 0 , respectively , showing isolated β-gal-positive cells in the yolk sac vessel mesothelium . Labelling is also seen in the foregut ( fg ) and pericardium ( p ) . Vessel staining is mosaic - in some vessels most or all mesothelial cells are labelled ( C’ , D ) while in others few or none are labelled ( D’ ) . A positive blood cell within a yolk sac vessel is also shown ( D; arrowhead ) . ( D’’ ) Additional section showing labelled blood cells within a yolk sac vessel at E9 . 0 ( arrowhead ) . ( E–E’ ) early and mid-primitive streak stage Nkx2-5lacZ/+embryos at ~E70 . 0 . No β-gal-positive cells are seen in the embryo or yolk sac . ( F ) E7 . 5 Nkx2-5lacZ/+embryo showing the earliest detectable labelling of in the cardiac crescent ( cc ) and yolk sac ( arrows ) . ( G–G’ ) Cryo-section of E7 . 5 embryos showing clusters of labelled cells in the cardiac crescent , and isolated cells embedded within the mesodermal layer of the yolk sac ( arrows ) . ( H ) Cryosection through the forming heart region of E8 . 0 embryo showing β-gal-positive cells in yolk sac mesoderm and endoderm ( arrows ) only in close association with the heart . Scale bars: 100 μm . ( I ) Graph showing quantification of Runx1+ and YFP+ cells in the floor of the dorsal aorta in the AGM region of E10 . 5 Nkx2-5irescre/+;ROSAYFP/+ embryos ( n = 3 embryos; average number of cells/section with standard deviation ) . ( J–J’’’ ) Transverse section of E10 . 5 Nkx2-5irescre/+;ROSAYFP/+ embryo through the region of AGM immunostained for Runx1 ( red ) and Nkx2-5+ lineage traced YFP+ cells . Nuclei indicated by Hoechst . Yellow arrows mark YFP+ Runx1+ cells; red arrows mark YFP- Runx1+ cells; green arrows mark YFP+ Runx1- cells in the endothelium . Note , the majority of Runx1+ cell are YFP- both in the endothelium and sub-endothelial regions of the AGM . These stainings complement those shown in Figure 9I–J’’’ and indicate a section in which there are several Nkx2-5-lineage traced YFP+ cells that are Runx1- ( quantified in panel I ) . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02410 . 7554/eLife . 20994 . 025Figure 9—figure supplement 2 . Nkx2-5+ lineage contributes to hemogenic endothelium of yolk sac vasculature . Co-immunostaining of E9 . 0 Nkx2-5irescre;ROSAYFP ( A–C’’’ ) and Nkx2-5irescre;Z/EG ( D–E’’’ ) embryos . ( A-B’’’; D–D’’’ ) . Cryosections of yolk sac immunostained for YFP ( green ) and endothelial marker Vegfr2/Flk1 ( Red ) . Co-localization of Nkx2-5+ lineage traced YFP+ cells and Vegfr2 was observed in the mesothelium of vessels as marked by arrows . ( C-C’’’; E–E’’’ ) . Immunostaining of yolk sac cryosections for Nkx2-5-lineage traced YFP+ cells ( green ) and VE-cadherin ( red ) . Arrows depict co-localization of YFP+ cells with VE-cadherin . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02510 . 7554/eLife . 20994 . 026Figure 9—figure supplement 3 . Nkx2-5 expression in single cells of mouse E7 . 0–7 . 75 embryos . ( A ) t-SNE plots were used to visualize single cells transiting through stages of mouse mesodermal development in vivo , with cell clustering and identification of subtypes based on categorizations determined previously ( left plot ) ( Scialdone et al . , 2016 ) . Nkx2-5 expression is shown overlaid across these populations ( right plot ) . ( B–C ) The fraction of cells expressing Nkx2-5 ( B ) and violin plots depicting the distribution of Nkx2-5 expression levels ( C ) are shown for each of the nine different cell populations identified including mesodermal populations as well as epiblast and extra-embryonic ectoderm . Dashed line in C represents chosen threshold of 1 normalised read count/cell . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02610 . 7554/eLife . 20994 . 027Figure 9—figure supplement 4 . Spatial RNA-seq analysis of Nkx2-5 expression in mouse gastrula-stage embryos . ( A–C ) ‘Corn plots’ showing examples of the spatial domains of Nkx2-5 expression in the epiblast , ectoderm and mesoderm of mouse embryo at different stages of gastrulation ( taken from published RNA-seq data at E7 . 0 ( n = 3 ) ( Peng et al . , 2016 ) and unpublished data at E6 . 5 ( n = 4 ) and E7 . 5 ( n = 3 ) ) . ( A ) E6 . 5: anterior ( A ) and posterior ( P ) halves at seven proximo-distal levels . In the example shown , only segment A3 showed expression . ( B ) E7 . 0: A , P , left ( L ) and right ( R ) quadrants at 11 proximo-distal levels ( A and P only from level 1 ) ( upper panel ) ; and mesoderm pooled from both sides ( M ) and posterior epiblast quadrant containing primitive streak ( P ) at 12 proximo-distal levels ( lower panel ) . ( C ) E7 . 5: A and P quadrants , left and right anterior lateral quadrants ( L1 , R1 ) and left and right posterior lateral half quadrants ( L2 , R2 ) at nine proximo-distal levels ( A and P only in level 1 ) ( upper panel ) ; and separate left and right mesodermal wings ( M1 and M2 ) compared to posterior ectoderm quadrant at nine proximo-distal levels ( lower panel ) . Scales represent levels of Nkx2-5 expression as log10 of fragments per kilobase million ( FPKM +1 ) ( D ) Nkx2-5 expression level ( mean log10 of FPKM ± standard deviation of replicates ) in the positive segments describe in ( A–C ) above ( Peng et al . , 2016 ) . ( E ) Table of Nkx2-5 spatial expression values ( log10 of FPKM ) used to construct corn plots shown in ( A–C ) . Segments in which there were no detectable transcripts are not shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 02710 . 7554/eLife . 20994 . 028Figure 9—figure supplement 5 . Nkx2-5 lineage+ cells in the cardio-pharyngeal region . ( A–E” ) Co-immunostaining of cryo-sections from E10 . 5 Nkx2-5irescre; ROSAYFP ( A–D ) and Nkx2-5irescre;Z/EG ( E–E” ) embryos . ( A ) Section showing Nkx2-5 lineage traced cells ( YFP+ ) in the endothelium of the pharyngeal arch arteries ( paa ) and endoderm of foregut ( fg ) . In pharyngeal arch arteries , YFP+ cells do not express Runx1 ( white arrows ) , however note co-expression of YFP and Runx1 in the foregut floor and ectoderm lateral to one pharyngeal arch artery . ( B ) Nkx2-5 lineage traced YFP+ cells contribute extensively to the myocardium of outflow tract ( oft ) , atrium ( a ) , ventricle ( v ) and atrio-ventricular cushion tissue ( c ) , as well as liver ( li ) . Few Runx1+YFP- cells were observed in cushion tissues ( arrows ) which are potentially blood cells . Other Runx1+YFP- blood cells can be seen within the atrium and pericardium ( haemorrhaged during preparation ) . ( C–D ) CD41+YFP+ endothelial ( and potentially hemogenic ) cells observed in the atrium , ventricles and paired anterior dorsal aorta ( da ) ( yellow arrows ) . One double positive cell in the atrium may be in the process of extrusion into the lumen . One CD41+YFP- cell is also shown ( white arrow ) . ( E–E” ) Co-immunostaining demonstrates Nkx2-5 lineage traced GFP+ cells embedded within Pecam1+ endothelial cells of anterior dorsal aorta ( white arrows ) . Scale bar: A-D; 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 028 To explore further the lineage identities of those cells first expressing Nkx2-5 in mouse embryos at E7 . 5 , we interrogated recently published transcriptome data of single cells derived from the epiblast and mesoderm of E7 . 0-E7 . 75 embryos , whose lineage identities had been determined based on cell surface marker status ( Flk1; CD41 ) and transcriptome signature ( Scialdone et al . , 2016 ) . No cells classified as epiblast or extra-embryonic ectoderm expressed Nkx2-5 ( Figure 9—figure supplement 3A–C ) . However , Nkx2-5 was expressed in ~7% of cells classified as nascent mesoderm and ~30% of cells classified as posterior and pharyngeal mesoderm ( likely cardiac progenitors ) . Importantly , ~30% of endothelial cells and an equivalent proportion of cells classified as allantois ( one extra-embryonic site of hemopoiesis ) were Nkx2-5+ , as were ~10% of early blood progenitors . These data confirm the hemoangiogenic nature of Nkx2-5 lineage-traced cells in early mouse embryos . It is noteworthy that Nkx2-5-Cre mediated lineage-tagged cells were not found in late gastrula/pre-crescent stage embryos at E7 . 0 ( Figure 9—figure supplement 1A , A’ ) . Lack of significant Nkx2-5 mRNA expression in early to mid-gastrula stage embryos was supported by interrogation of a previously published high-resolution spatial RNA-seq map of E7 . 0 mid-gastrula stage embryos ( Peng et al . , 2016 ) , as well as similar maps from E6 . 5 early gastrula and E7 . 5 late gastrula/pre-cardiac crescent stage embryos ( unpublished data from S . S . , J-D . J . H . , G . P . , N . J . and P . P . L . T . ; summarized in Figure 9—figure supplement 4A–E ) . No consistent spatial pattern of Nkx2-5 expression was obtained in proximo-distal and anterior-posterior embryonic segments of epiblast at early gastrulation ( n = 4 , Figure 9—figure supplement 4A ) . Where expression did register , it occurred in sporadic segments , and the mean and maximum normalized read counts ( log10 of fragments per kilobase million [FPKM] ) were very low ( mean 0 . 18 ± 0 . 08 , range 0 . 09–0 . 25; Figure 9—figure supplement 4D , E ) . Similarly , at mid-gastrulation ( E7 . 0; n = 3 ) , only very low read counts registered in epiblast ( mostly in a single replicate; mean 0 . 07 ± 0 . 07 , range 0 . 032–0 . 32 ) , and nascent mesoderm ( mean 0 . 58 ± 0 . 67 , range 0 . 09–1 . 35 ) ( Figure 9—figure supplement 4B , D , E ) . In contrast , in E7 . 5 late gastrula/pre-cardiac crescent stage embryos , significantly higher read counts registered in coherent clusters of segments in nascent mesoderm in 2 of the three replicates ( mean 5 . 6 ± 5 . 9 , range 0 . 07–20 . 1 ) while only low counts were found in ectoderm in sporadic segments ( mean 0 . 32 ± 0 . 36 , range 0 . 03–1 . 1 ) ( Figure 9—figure supplement 4D , E ) . The higher number of counts in clusters of segments in E7 . 5 nascent mesoderm likely reflects the upward slope of the Nkx2-5 expression profile during specification of cardiomyocytes in the cardiac crescent . The RNA-seq profiles in mouse embryos are therefore consistent with our lineage tracing data detecting the onset of Nkx2-5 expression at the pre-crescent/crescent transition . Taking a complementary approach , we explored the timing and location of Nkx2-5 expression using the sensitive Nkx2-5LacZ knockin allele ( Prall et al . , 2007 ) ( Figure 9E–H ) . Similar to our findings with lineage tracing and RNA-seq , the earliest expression of β-gal in extra-embryonic tissue of Nkx2-5LacZ/+ embryos was at E7 . 5 , when the cardiac crescent was established ( Figure 9E , F; Figure 9—figure supplement 1E–F ) . Even when only few cells were β-gal positive in the crescent , a comparable number of positive cells were evident in the yolk sac ( Figure 9—figure supplement 1F ) . Labelling was in the mesodermal layer proximal to the cardiac crescent ( Figure 9E , F; Figure 9—figure supplement 1F–G’ ) . However , by forming heart tube and looping heart stages , positive cells in the yolk sac were only rarely evident ( Figure 9G , H ) , although visual inspection of whole mount stained E8 . 0 embryos and histological sectioning , showed positive cells in yolk sac at the caudal territories of the cardiac zone ( Figure 9G , arrows ) and in yolk sac mesoderm and endoderm immediately overlying the heart region ( Figure 9—figure supplement 1H , H’ ) . Interestingly , both lineage tracing and Nkx2-5-LacZ expression showed that labelled endodermal cells were not seen in the yolk sac layer distal to the cardiac zone , and those located proximally likely remain associated with the foregut . We conclude that in mouse , Nkx2-5 is activated transiently in extra-embryonic mesoderm in close association with the cardiac crescent at E7 . 5-E8 . 0 . The combination of expression analysis , lineage tracing and RNA-seq suggests that these cells expand as they move deeper into the yolk sac and populate vascular structures . We anticipate that the Nkx2-5+ cells present at E7 . 5-E8 . 0 are the founders of the Nkx2-5 extra-cardiac lineage and are analogous to those detected in extra-embryonic regions in chick embryos . We hypothesize that some Nkx2-5+ cells correspond to hemoangiogenic precursors that later populate the endothelium of yolk sac blood islands and hemogenic endocardium ( our study and Nakano et al . , 2013; Stanley et al . , 2002 ) . We next explored whether the mouse Nkx2-5 extra-embryonic lineage+ cells described above also contribute to the hemogenic endothelium of the AGM region in the DA . Nakano and colleagues did not detect Nkx2 . 5 lineage+ cells in the AGM at E9 . 5 using the crosses mentioned above ( Nakano et al . , 2013 ) . However , in Nkx2-5irescre/+;RosaYFP/+ embryos at E10 . 5 , we found abundant YFP+ cells in the AGM region , including in endothelium and sub-endothelium ( Figure 9I–J”’; Figure 9—figure supplement 1J–J’’’ ) , as seen previously for cells in the AGM expressing the earliest hemogenic marker Runx1 ( North et al . , 2002 ) . We also saw cells with the same expression profile within the liver ( Figure 9K–K’’’ ) , the first site of definitive mammalian embryonic hemopoiesis . Most YFP+ cells in the DA co-expressed Runx1 , although there were also many Runx1+ YFP- cells in the region ( Figure 9I–J’’’; Figure 9—figure supplement 1I ) . Quantifications showed that ~ 15% of Runx1+ cells in the endothelium and ~16% of Runx1+ cells in the sub-endothelium were also YFP+ , while overall >77% of YFP+ cells were also Runx1+ ( Figure 9—figure supplement 1I ) . Consistently , we calculated a similar ~15% ratio of Runx1+/Nkx2-5-en+ in the chick DA ( data not shown ) . Clusters of Runx1+ YFP+ cells appeared to be budding from the endothelial layer of the DA towards the lumen ( Figure 9J–J''' , Figure 9—figure supplement 1J–J''' ) , behaviour consistent with that of hemogenic endothelium ( Jaffredo et al . , 1998 ) . With a single exception , none of the YFP+ cells in the AGM expressed the hemogenic marker CD41 ( integrin alpha2b ) that is known to act downstream of Runx1 ( Mikkola et al . , 2003 ) . However , there were significant numbers of YFP+ CD41+ as well as YFP+ Runx1+ cells in the liver ( Figure Figure 9K-L''' ) . In line with previous mouse data ( Nakano et al . , 2013; Paffett-Lugassy et al . , 2013; Stanley et al . , 2002 ) and chick lineage tracing studies presented above , mouse Nkx2-5 lineage+ cells were detected within the endothelium of the pharyngeal arch arteries and paired dorsal aorta , and endocardium of the atria and ventricles , at E9 . 5 ( data not shown ) and E10 . 5 ( Figure 9—figure supplement 5A–E’’ ) . Using Nkx2-5irescre/+;Z/EG embryos , we confirmed that Nkx2-5 lineage+ ( GFP+ ) cells were embedded within Pecam1+ endothelium of the DA ( Figure 9—figure supplement 5E–E” ) . Some Nkx2-5 lineage+ endothelial cells in the cardiac region at E10 . 5 were also CD41+ , although these were very rare ( Figure 9—figure supplement 5C–D ) . The present study provides novel insights into early cardiogenesis and hemangiogenesis in vertebrate embryos . First , we revealed an early contribution of mesodermal cells residing outside the classical cardiac crescent region , to the developing heart ( Figure 10 ) . Cellular and molecular analyses indicate that a subset of extra-embryonic/LPM-derived hemogenic angioblasts travels via the inflow tract to populate the heart endocardium . Moreover , our results in both chick and mouse models demonstrate that the hemogenic endothelium in the DA and endocardium derives , at least in part , from Nkx2 . 5 lineage+ angioblast progenitors that may also contribute to hematopoietic stem cells ( HSCs ) in the fetal liver ( Figure 10 ) . Based on these studies and gain of function experiments , we propose a broader role for Nkx2 . 5 in the cardiovascular and blood lineage diversification programs than currently appreciated . 10 . 7554/eLife . 20994 . 029Figure 10 . Nkx2 . 5 marks hemogenic angioblasts that contribute to the formation of the endocardium and dorsal aorta . Cardiovascular progenitors begin to form at the onset of gastrulaion . These populations segregate as the cells begin to migrate in a lateral fashion towards the extraembryonic tissue . During early gastrulation in the chick ( St . 4 ) or late gastrula stages in the mouse ( cardiac crescent , E7 . 5 ) Nkx2 . 5+ hemogenic angioblasts are specified . While Nkx2 . 5 expression is maintained at the cardiac crescent its expression in the hemogenic angioblasts is downregulated . Nkx2 . 5 lineage-derived cells populate the cardiac crescent , extraembryonic tissue and lateral plate mesoderm in both species ( St . 8 and E7 . 5 ) . As the embryo develops Nkx2 . 5 derived hemogenic angioblasts migrate to the endocardium and dorsal aorta , there generating blood cells via the hemogenic endothelium ( St . 18 and E10 . 5 ) . In both chick and mouse Nkx2 . 5 lineage derived cells contribute massively to the yolk sac vasculature and to HSCs in the fetal liver ( mouse ) . CC – cardiac crescent; DA – dorsal aorta; EE – extraembryonic; FHF – first heart field; hrt – heart; LB – liver bud; LPM – lateral plate mesoderm; PS – primitive streak . DOI: http://dx . doi . org/10 . 7554/eLife . 20994 . 029 The cardiovascular and hematopoietic lineage specification programs begin at gastrulation , when nascent mesodermal cells ingress through the primitive streak . In the chick , we demonstrated that Nkx2 . 5 is transiently expressed in the mid-primitive streak and the posterior LPM/extra-embryonic tissues at St . 4–6 . Moreover , overexpression of Nkx2 . 5 induced endothelial/hematopoietic gene expression . In the mouse , Nkx2-5 activation in extra-embryonic tissue occurred later in gastrulation than in the chick , at the time of cardiac crescent formation , but positive cells nonetheless contributed extensively to the yolk sac vasculature and blood . Expression was also seen in hemogenic endothelium of the AGM , although we have not yet shown definitively in mouse whether positive cells migrate to the AGM from extra-embryonic tissue or whether Nkx2-5 is expressed de novo in hemogenic angioblasts arising in embryonic LPM . Confirming findings of Nakano and colleagues ( Nakano et al . , 2013 ) , we found Nkx2-5 lineage+ blood cells in the yolk sac and embryonic vessels . Nakano et al . found that Nkx2-5 lineage+ blood cells expressed haemoglobins typical of primitive then definitive erythropoiesis , although no positive blood cells were present in adults . These results resonate with those of Yoder et al . , who demonstrated that the yolk sac contains HSCs capable of sustaining long-term multi-lineage hemopoiesis when transferred to conditioned hemopoietic organs of neonates although not adults ( Yoder et al . , 1997 ) . Taken together our findings suggest that Nkx2 . 5 plays a conserved role in the establishment of the cardiovascular and early blood cell lineages , in apparent agreement with the severe defects observed in the yolk sac vasculature and hematopoiesis seen in Nkx2-5 mutant mouse embryos ( Tanaka et al . , 1999 ) . The electroporation technique in early chick embryos enabled us to follow nascent mesoderm progenitors with hemogenic angioblastic characteristics . What makes these cells hemangioblast-like ? First , these extra-embryonic/LPM mesoderm progenitors were labelled by both Nkx2-5-en and hemangioblast enhancers . Morphologically , we observed the expression of these two enhancers within blood islands , and could track their migration into the DA ( and other vessels ) , as well as to the endocardium . These cells broadly expressed hematopoietic and endothelial lineage markers such as Runx1 , Flk1 , Tal1 , Ets1 , Gata1 , Lmo2 and Gata4 . When cultured in vitro , these cells differentiated into endothelial and blood progenitors , although we have not yet shown a bipotent differentiation potential in a clonal assay . Finally , we showed that in vivo , these cells contributed to the hemogenic endothelium in the DA and heart endocardium . Tight molecular and cellular relationships exist between myocardial , endocardial , endothelial , and hematopoietic lineages in different organisms . A number of transcription factors control these cell fate choices , primarily through cross-inhibition mechanisms ( Caprioli et al . , 2011; Schoenebeck et al . , 2007; Van Handel et al . , 2012 ) . Heart , vessels and blood likely represent an ancient lineage triad controlled by Nkx2-5 that has contributed to cardiovascular and hemogenic developmental plasticity in vertebrates . This plasticity is highlighted by activation of the hematopoietic program within hemogenic endothelium at various embryonic sites , including the yolk sac ( Palis et al . , 1999 ) , DA ( Jaffredo et al . , 1998 ) , extra-embryonic arteries ( Gordon-Keylock et al . , 2013 ) , and endocardium ( Nakano et al . , 2013 ) . The mouse enhancer for Nkx2-5 labels a broad cardiovascular mesoderm population in the chick , including FHF cardiac progenitors , endothelial and hematopoietic progenitors . Because of the demonstrated early expression of Nkx2 . 5 protein and mRNA in the nascent mesoderm at St . 4–6 , we suggest that the reporter expression provides a valid readout of the transient expression of Nkx2 . 5 in these cells , which is later downregulated in extra-cardiac lineages by endothelial and hematopoietic transcription factors . In the mouse , we showed that Nkx2-5 lineage+ cells arise in extra-embryonic tissues later in gastrulation associated with cardiac crescent formation . Linage+ cells also contributed substantially to the hemogenic endothelium within the AGM , liver , and , to a lesser degree , the endocardium . The extra-cardiac Nkx2-5 lineage+ cells that contribute to mouse hemopoiesis in the embryo from E8 . 5-E15 . 5 ( Nakano et al . , 2013 ) , likely derive from the yolk sac or AGM region via the liver , rather than the endocardium as proposed ( Nakano et al . , 2013 ) . However , whether there is a contribution of Nkx2-5 lineage+ cells to primitive hemopoiesis in the mouse yolk sac , and how Nkx2-5 lineage+ cell arrive in the AGM region , remain to be explored . Moreover , how Nkx2-5 lineage+ cells relate to the Runx1+ Gata1- Etv2+ VE-cadherin+ hemogenic angioblasts which are already present in the yolk sac at E7 . 5 ( Eliades et al . , 2016; Lie-A-Ling et al . , 2014; Tanaka et al . , 2014 ) , needs detailed analysis . Notwithstanding the differences in the timing of activation of Nkx2-5 in extraembryonic tissue in chick and mouse , our findings on the Nkx2-5 lineage in these two models seem highly concordant . Bioinformatics and in vitro analyses demonstrated that several Gata and Ets sites are present and active within the Nkx2-5 enhancer ( data not shown and Lien et al . , 1999 ) . Our ISH and RT-PCR analysis revealed the expression of Gata4 in both LPM and cardiac crescent ( and also nascent mesoderm ) corresponding to the expression of the Nkx2-5-en . Hence , Gata factors could drive the early expression of Nkx2 . 5 . Scl/Tal1 was previously shown to be a key regulator of hemangioblasts ( Bussmann et al . , 2007; Schoenebeck et al . , 2007; Van Handel et al . , 2012 ) . Our findings demonstrating the ectopic activation of Tal1 by Nkx2 . 5 suggest that the latter acts upstream of Tal1 in the hemangioblast hierarchy and may have a role in priming hemopoiesis . Activation of Tal1 in the chick inhibited cardiac gene expression , in line with the repressive role of Scl/Tal1 on cardiogenesis , as demonstrated in Scl/Tal1 mutant mouse embryos in which ectopic cardiomyogenesis was observed in the yolk sac vasculature and endocardium ( Van Handel et al . , 2012 ) . Similar to Scl/Tal1 , loss of etsrp/etv2 induced cardiac gene expression in endocardial progenitors ( Palencia-Desai et al . , 2011; Rasmussen et al . , 2011 ) . Taken together , these studies and our data highlight the dynamic developmental interplay between embryonic mesoderm progenitors for cardiac , endothelial and blood lineages . Endocardial cells are known to be heterogeneous in origins ( Harris and Black , 2010; Milgrom-Hoffman et al . , 2011; Vincent and Buckingham , 2010 ) . In vitro studies in chick and mouse provided evidence for bipotential progenitors for myocardial and endocardial cells that exist in the SHF ( Hutson et al . , 2010; Kattman et al . , 2006; Lescroart et al . , 2014; Moretti et al . , 2006 ) , but probably not in the FHF in which myocardial progenitors are unipotent ( Lescroart et al . , 2014; Später et al . , 2013 ) . Our findings are consistent with the idea that endocardial and myocardial lineage separation occurs very early during embryogenesis , in line with other studies ( Li et al . , 2015; Milgrom-Hoffman et al . , 2011; Paffett-Lugassy et al . , 2013 ) . Understanding the origin and regulation of definitive HSCs is a subject of great interest and contention . It is clear that HSCs are mesoderm-derived and their genesis involves differentiation from endothelial cells . In the chick , zebrafish and mouse embryos , the splanchnic mesoderm has been considered to be the source of hemogenic endothelium-derived HSCs ( Childs et al . , 2002; Lancrin et al . , 2009; Zovein et al . , 2010 ) . More recent data suggest that hemogenic endothelium at embryonic sites have their origins in the yolk sac ( Tanaka et al . , 2014 ) . It is likely that multiple sources of endothelial cells contribute to the formation and establishment of HSCs including extra-embryonic hemogenic angioblasts ( Tanaka et al . , 2014 ) and intra-embryonic angioblasts that are induced to form HSCs by local signalling ( Nguyen et al . , 2014; Richard et al . , 2013 ) . Consistent with these findings we suggest that part of the hemogenic endothelium in both DA and endocardium derives from extra-embryonic/LPM hemogenic angioblast progenitors previously expressing Nkx2 . 5 . The detection of Nkx2 . 5 lineage+ hematopoietic progenitors in DA , liver and endocardium , suggests that these progenitors derive from a common embryonic origin , although this remains to be formally tested . Potentially relevant to this work , a lineage tracing study with Nfatc1-Cre mice suggests that endocardial cells , which are in contact with liver epithelium , contribute to the liver vasculature ( Zhang et al . , 2016 ) . As in the DA , endocardial cells have multiple origins , some of which derive from specialized angioblasts with hemogenic properties . Indeed , the existence of specialized angioblasts has now been documented in the floor of the cardinal vein that generates lymph , arterial and venous cell fates ( Nicenboim et al . , 2015 ) . Thus , analysis of heart development has revealed on the one hand major contributions from the well-characterised cardiac cell populations ( e . g . FHF , SHF , cardiac neural crest , epicardium ) , along with minor contributions from other mesodermal progenitors that provide further developmental plasticity and functionality . Our studies and those of others ( see Introduction ) suggest a conserved role for Nkx2 . 5 in both heart and hemoangiogenic lineage development . In addition to its role in the specification of cardiac mesoderm , the Drosophila Nkx2 . 5 homologue , tinman , is essential for formation of the fly larval lymph gland , a hemopoietic organ proposed to be similar to the hemogenic endothelium of the mammalian AGM region ( Mandal et al . , 2004 ) . Thus , cardio-hemoangiogenic lineage development may be one of the ancestral functions of Nkx2 . 5 . In summary , the role of Nkx2 . 5 in endothelial and blood specification programs is evolutionarily conserved from fly to human . In vertebrates , Nkx2 . 5 expression is mostly restricted to cardiac progenitors , but in a narrow window of time may prime vascular and blood development , as shown already in pharyngeal arch arteries and endocardium ( Nakano et al . , 2013; Paffett-Lugassy et al . , 2013 ) . The definitive vasculature and blood networks ultimately utilise other transcription factors for lineage specialisation , which also prevent ectopic heart formation in these highly plastic mesodermal cells ( Van Handel et al . , 2012 ) . Unravelling the dynamic roles of Nkx2 . 5 in cardiac , endothelial , and blood lineage specification will lead to a deeper understanding of the cardiovascular/blood regulatory networks , and may point the way toward novel regenerative therapies for cardiovascular disease . Fertilized white eggs from commercial sources were incubated to the desired stage at 38 . 0°C in a humidified incubator , according to Hamburger and Hamilton ( Hamburger and Hamilton , 1951 ) . Mice used in this study were bred and maintained in the Victor Chang Cardiac Research Institute BioCore facility according to the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes . For the Nkx2-5-lineage study , the Nkx2-5irescre strain ( Stanley et al . , 2002 ) was crossed with RosalacZ ( Soriano , 1999 ) , RosaYFP ( Srinivas et al . , 2001 ) or Z/EG ( Novak et al . , 2000 ) reporter lines . Litters of required embryonic stages were dissected in 1X ice cold PBS , fixed and processed accordingly . For whole mount β-gal staining , embryos were fixed in 0 . 1 M sodium phosphate buffer containing 0 . 2% glutaraldehyde , 1 . 48% formaldehyde , 5 mM EGTA and 2 mM magnesium chloride , pH7 . 3 from 30 min to 60 min at room temperature depending on the size of the embryo ( for examples , E9 . 0 embryos were fixed for 1 hr ) . Further , embryos were washed in 0 . 1M sodium phosphate buffer containing 2 mM of magnesium chloride , 0 . 01% sodium deoxycholate and 0 . 02% of Nonidet-P40 detergent followed by β-gal staining in 0 . 1M sodium phosphate buffer containing 1X potassium ferrocynide , 1X potassium ferricynide and 1 mg/ml of X-gal ( Promega , V3941 ) . Imaging of whole mount beta-gal stained embryos was done with Leica M125 microscope fitted with a Leica DFC295 camera . Embryos were fixed in 4% PFA at 4° C overnight followed by PBS washes . Subsequently , embryos were transferred through a 15% and 30% sucrose gradation followed by embedding in Tissue-Tek O . C . T . compound ( Thermo scientific , Waltham , MA USA ) . 10 μM sections were prepared using Leica cryostat . For immune-staining sections were fixed in 4% PFA for 5 min on ice , washed in a 1X PBS , blocked in serum containing 3% goat serum , 3% bovine serum albumin and 0 . 1% Triton X-100 followed by incubation with primary antibodies overnight at 4° C . Primary antibodies used were GFP ( 1:200 , ab13970 ) , Runx1 ( 1:200 , ab92336 ) , CD41 ( 1:300 , ab33661 ) , CD31 ( 1:100 , 553370 , BD Pharmingen ) , Vegfr2/Flk1 ( 1:100 , sc48161 ) and VE-cadherin ( 1:100 , sc-6458 ) . Secondary antibodies , anti-rabbit biotin , anti-rat biotin , anti-goat biotin and anti-chicken Alexa Fluor-488 ( Life Technology ) were used at 1:200 concentrations . For biotinylated antibodies signal was amplified using ABC ( Vectastatin ) and cy3-Tyramide amplification kit ( Perkin Elmer ) . Images were taken using a Zeiss LSM 700 Upright confocal microscope . Runx1+ YFP- , Runx1- YFP+ and Runx1+ YFP+ cells were quantified in endothelium or sub-endothelium of the DA in the region of AGM from three different embryos . Confocal Z-stack images from six to eight 10 µm sections per embryo were used for counting . The average number of single or double positive cells per section were graphed with standard deviation . Whole mount in situ hybridization was performed , using digoxigenin ( dig ) -labelled antisense riboprobes synthesized from total cDNA . Briefly riboprobes were generated using T7 polymerase in the presence of dig-labelled UTP ( Roche , Penzberg , Germany ) . Embryos were incubated to the desired stage , and fixed in 4% paraformaldehyde ( PFA ) overnight . Dehydration was performed with methanol series , after which the embryos were stored in absolute methanol overnight at −20°C . Rehydration was performed in methanol series , in PBT . Embryos were treated with proteinase K and fixed with 4% PFA/0 . 2% glutaraldehyde . Prehybridization ( 50% formamide , 5XSSC pH 4 . 5 , 2% SDS , 250 μg/ml yeast tRNA , 50 μg/ml heparin ) was performed at 68°C , prior to hybridization overnight at that temperature . To remove unbound probe , a series of washes ( 50% formamide , 5XSSC pH 4 . 5 , 1% SDS ) was performed . After a second series of washes in 0 . 1M maleic acid ( Ph 7 . 4 ) , 0 . 15M NaCl , and 1% Tween-20 ( MABT ) , the embryos were incubated in blocking solution ( 20% whole goat serum in MABT ) for 2 hr; an anti-dig alkaline phosphatase-conjugated antibody ( Roche ) was then added to the blocking solution , and embryos were incubated at 4°C overnight . A colour reaction was performed using NBT/BCIP substrates ( Roche ) . To stop the reaction , embryos were fixed in 4% PFA . For wholemount immunostaining St . 3–4 embryos were dissected with their yolk sac , fixed in 4% PFA for 1 hr and then subjected to standard immunostaining protocol . For cryo-sections , embryos were fixed in 4% PFA , incubated overnight in 30% sucrose in PBS at 4°C and then embedded in a Peel-A-Way plastic mold ( VWR , Radnor , PA USA ) in OCT compound ( Sakura , VWR ) . The embryos were sectioned between 7–15 μm , using a Leica cryostat . Sections were blocked with 5% whole goat serum in 1% bovine serum albumin in PBS , prior to incubation with primary antibodies . We used the following primary antibodies: Nkx2 . 5 ( 1:300 , Santa-Cruz SC-8697 ) was amplified using the Tyramid amplification kit ( Perkin-Elmer Waltham , MA USA and Life Technologies , Carlsbad , CA USA ) . Isl1 ( DSHB , 40 . 3A4 , 1:5 ) , CD45 ( 1:100 , Prionics HIS-C7 ) , Runx1 ( 1:100 , Abcam ab92336 ) , EGFP ( 1:100 , Santa Cruz SC-101536 ) , and a novel in-house Cdh5 ( 1:100 , VE-cadherin ) . Secondary antibodies: Cy2/488 , Cy3/594 and Cy5/647-conjugated anti-mouse , anti-rabbit , anti-rat IgG ( Jackson ImmunoResearch laboratories , West Grove , PA USA and Thermo Fischer ) were diluted 1:100 . Images are obtained using a Leica MZ 16FA stereomicroscope attached to a Leica digital camera ( DC300F , Leica Microsystems , Wetzlar , Germany ) and an assembled upright fluorescent microscope ( Nikon ECLIPSE 90i ) with a CCD camera . For time lapse filming a temperature controlled chamber ( semi-automated ) is used under the software advanced acquisition system ( Image-Pro AMS version 6 . 0 , Media Cybernetics , Bethesda , MD , USA ) . The system allows multi channel ( bright field and GFP ) Z-stacks ( for constructing a focused image ) photography at multi time points for generating a time lapse movie . Cell tracking in live embryos was performed by manual analysis of 7 focal planes at each time point during in vivo imaging . Time lapse imaging was also performed using the DeltaVision Elite system ( Applied Precision , USA ) , on an Olympus IX71 inverted microscope , running softWoRx 6 . 0 by a CoolSnap HQ2 CCD camera ( Roper Scientific , USA ) . Still images were also obtained with the Zeiss Cell Observer Spinning Disk Confocal microscope ( Oberkochen , Germany ) . Captured images were analyzed by ZEN , Photoshop , Image-Pro AMS and ImageJ softwares . To introduce DNA vectors into the embryonic mesoderm , a method was used involving electroporation of mesodermal cells while they are still epithelial , prior to their ingression during gastrulation . Chick embryos were cultured in ‘EC-culture’ ( Chapman et al . , 2001 ) . Embryos were then placed ventral side up ( the vitelline membrane on the bottom ) above the cathode , made of a 2 × 2 mm platinum plate located in a concavity made in the plastic platform . A solution of 1 μg/μl DNA and Fast Green ( 10 μg/ml ) was injected between the blastoderm and the vitelline membrane , using a glass capillary . An anodal electrode , with an inter-electrode distance of 4 mm , was quickly placed on the hypoblast side of the embryo , and electroporation was performed using an ECM830 electroporator ( BTX Co . Ltd , Holliston , MA USA ) . St . 3 embryos were electroporated with 3 pulses of 9V each , for 35 ms at intervals of 300 ms . After electroporation was complete , embryos were transferred back to the EC culture plates , and incubated to the desired stage . A novel Isl1 cardiac enhancer was identified , using comparative genomic analysis ( genomes of chick , rat , zebrafish and human were used ) . The enhancer was cloned into previously described pTK vectors that contain a minimal promoter from the HSV thymidine kinase gene upstream of GFP/RFP , and were shown to be permissive for enhancer activity in chick embryos ( Uchikawa et al . , 2004 ) . Control vectors for broad expression within the embryo , the reporter plasmids pCAGG-GFP/RFP , which were previously described ( Nathan et al . , 2008 ) , were injected . The mouse Nkx2-5 cardiac enhancer was obtained from Prof . Eric Olson ( Lien et al . , 1999 ) and subcloned into the pTK vector for chick expression . To study the relationship between the hemangioblast and the cardiac mesoderm , we obtained a hemangioblast enhancer based on the chick Cerberus gene , that drives expression of GFP ( Teixeira et al . , 2011 ) . For ectopic expression of Nkx2 . 5 and Tal1 , chick coding sequences were cloned into pCAGG-IRES-GFP , which drives ubiquitous expression in every avian cell ( Megason and McMahon , 2002 ) . Dissection of different explants from chick embryos was carried out using a tungsten needle . Explants were isolated together with all germ layers at selected embryonic stages and locations . Posterior LPM explants were cultured for 4 . 5 days on a collagen drop covered with 700 μl of dissection medium ( 10% Fetal Calf Serum , chick embryo extract 2 . 5% and pen/strep 1% in αMEM medium , Biological Industries , Israel ) in a four-well plate . Total RNA was extracted using Qiagene RNeasy Micro Kit ( Qiagen , Germany ) , followed by reverse transcription using the cDNA Reverse Transcription kit ( Applied Biosystems , Foster City , CA USA ) . The cDNA product was then amplified using different sets of primers , via semi-quantitative RT-PCR . The RNA analysis was performed using semi-quantitative RT-PCR with Green Master Mix ( Promega , Madison , WI USA ) . Gene-level read counts data from single cell RNA-seq of single cells from E7-0-7 . 75 mouse embryos ( Scialdone et al . , 2016 ) were downloaded from http://gastrulation . stemcells . cam . ac . uk . Read counts were normalized for sequencing depth using size factors calculated with DESeq ( Anders and Huber , 2010 ) . Only samples from E7 . 0-E7 . 75 stage embryos were used . t-SNE plots were generated with highly variable genes as described ( Scialdone et al . , 2016 ) . Single cell RNA-seq samples were categorized as expressing Nkx2-5 if they had normalized read counts > 1/cell .	As an animal embryo develops , it establishes a circulatory system that includes the heart , vessels and blood . Vessels and blood initially form in the yolk sac , a membrane that surrounds the embryo . These yolk sac vessels act as a rudimentary circulatory system , connecting to the heart and blood vessels within the embryo itself . In older embryos , cells in the inner layer of the largest blood vessel ( known as the dorsal aorta ) generate blood stem cells that give rise to the different types of blood cells . A gene called Nkx2 . 5 encodes a protein that controls the activity of a number of complex genetic programs and has been long studied as a key player in the development of the heart . Nkx2 . 5 is essential for forming normal heart muscle cells and for shaping the primitive heart and its surrounding vessels into a working organ . Interfering with the normal activity of the Nkx2 . 5 gene results in severe defects in blood vessels and the heart . However , many details are missing on the role played by Nkx2 . 5 in specifying the different cellular components of the circulatory system and heart . Zamir et al . genetically engineered chick and mouse embryos to produce fluorescent markers that could be used to trace the cells that become part of blood vessels and heart . The experiments found that some of the cells that form the blood and vessels in the yolk sac originate from within the membranes surrounding the embryo , outside of the areas previously reported to give rise to the heart . The Nkx2 . 5 gene is active in these cells for only a short period of time as they migrate toward the heart and dorsal aorta , where they give rise to blood stem cells These findings suggest that Nkx2 . 5 plays an important role in triggering developmental processes that eventually give rise to blood vessels and blood cells . The next step following on from this work will be to find out what genes the protein encoded by Nkx2 . 5 regulates to drive these processes . Mapping the genes that control the early origins of blood and blood-forming vessels will help biologists understand this complex and vital tissue system , and develop new treatments for patients with conditions that affect their circulatory system . In the future , this knowledge may also help to engineer synthetic blood and blood products for use in trauma and genetic diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	Nkx2.5 marks angioblasts that contribute to hemogenic endothelium of the endocardium and dorsal aorta
Centrioles are microtubule-based organelles crucial for cell division , sensing and motility . In Caenorhabditis elegans , the onset of centriole formation requires notably the proteins SAS-5 and SAS-6 , which have functional equivalents across eukaryotic evolution . Whereas the molecular architecture of SAS-6 and its role in initiating centriole formation are well understood , the mechanisms by which SAS-5 and its relatives function is unclear . Here , we combine biophysical and structural analysis to uncover the architecture of SAS-5 and examine its functional implications in vivo . Our work reveals that two distinct self-associating domains are necessary to form higher-order oligomers of SAS-5: a trimeric coiled coil and a novel globular dimeric Implico domain . Disruption of either domain leads to centriole duplication failure in worm embryos , indicating that large SAS-5 assemblies are necessary for function in vivo . Most eukaryotes harbor microtubule-based cylindrical organelles called centrioles that exhibit a striking ninefold radial symmetry , and which are crucial for a wide range of cellular functions ( reviewed in Gönczy , 2012; Agircan et al . , 2014 ) . In resting cells , centrioles are usually found near the plasma membrane where they organize the formation of flagella and cilia , whereas in proliferating cells centrioles typically reside adjacent to the nucleus , where they recruit pericentriolar material to form the centrosome , the major microtubule organizing center of animal cells . Centrosomes play a major role in directing cellular architecture during interphase and bipolar spindle assembly during mitosis . Centriole numbers are tightly regulated , with centriole duplication occurring only once per cell cycle , in concert with replication of the genetic material ( reviewed in Firat-Karalar and Stearns , 2014 ) . Abnormalities in centriole formation can impair cell signaling and motility owing to defective cilia or flagella , as well as cause spindle positioning defects and genome instability due to aberrations in centrosome numbers and/or sizes . Thus , it is not surprising that centriolar defects are at the root of multiple medical conditions , including primary microcephaly , male sterility and possibly cancer ( reviewed in Nigg and Raff , 2009; Arquint et al . , 2014; Chavali et al . , 2014; Godinho and Pellman , 2014; and Nachury , 2014 ) . Five proteins required for centriole assembly were originally identified in Caenorhabditis elegans through genetic analysis and functional genomics ( reviewed in Gönczy , 2012 ) ; these include the recruiting factor SPD-2 ( Kemp et al . , 2004; Pelletier et al . , 2004 ) , the kinase ZYG-1 ( O'connell et al . , 2001 ) , and the coiled-coil domain containing proteins SAS-5 , SAS-6 and SAS-4 ( Kirkham et al . , 2003; Leidel and Gönczy , 2003; Dammermann et al . , 2004; Delattre et al . , 2004; Leidel et al . , 2005 ) . Following localization of these five proteins to the site of new centriole formation , recruitment of microtubules completes the assembly process , giving rise to a ninefold- symmetric centriole ∼100 nm in diameter ( Pelletier et al . , 2006 ) . Functionally equivalent proteins have now been identified throughout eukaryotes ( Carvalho-Santos et al . , 2010; Hodges et al . , 2010 ) , indicating an evolutionary shared assembly pathway for centriole formation . Whereas SAS-6 is critical for establishing the ninefold radial symmetry of centrioles ( reviewed in Gönczy , 2012; and Hirono , 2014 ) , the underlying structural mechanism differs between C . elegans and other species . Crystallographic and/or electron microscopic analysis supports the view that recombinant SAS-6 proteins from Chlamydomonas reinhardtii , Danio rerio and Leishmania major form ninefold-symmetric rings ( Kitagawa et al . , 2011b; Van Breugel et al . , 2011; Van Breugel et al . , 2014 ) . Such SAS-6 rings are thought to dictate the ninefold- symmetrical assembly of the entire centriole . In contrast , similar analysis of C . elegans SAS-6 suggests formation of a spiral oligomer with 4 . 5-fold symmetry per turn , thus generating ninefold symmetry upon two turns of the spiral ( Hilbert et al . , 2013 ) . C . elegans SAS-6 physically interacts with SAS-5 ( Leidel et al . , 2005; Qiao et al . , 2012; Hilbert et al . , 2013; Lettman et al . , 2013 ) , a protein that shuttles rapidly between the cytoplasm and centrioles throughout the cell cycle ( Delattre et al . , 2004 ) . The presence of SAS-6 and SAS-5 at centrioles is essential for formation of the central tube , a cylindrical structure at the core of the emerging centriole ( Pelletier et al . , 2006 ) . Depletion of SAS-5 ( Dammermann et al . , 2004; Delattre et al . , 2004 ) or SAS-5 mutants that are unable to bind SAS-6 ( Delattre et al . , 2004; Qiao et al . , 2012; Lettman et al . , 2013 ) prevent central tube formation , and therefore centriole assembly . Although SAS-5 has been proposed to assist SAS-6 organization ( Qiao et al . , 2012; Lettman et al . , 2013 ) , the mechanisms by which this may be achieved are not known , in part because the architecture of SAS-5 has not yet been resolved . Here , we employ biophysical methods and X-ray crystallography , together with functional assays in C . elegans embryos , to demonstrate that large assemblies of SAS-5 are necessary for centriole formation . Our results lead us to propose a working model in which SAS-5 oligomers may assist function by providing a multivalent framework for the assembly of C . elegans SAS-6 oligomers . Previous attempts at recombinant expression of full-length C . elegans SAS-5 ( SAS-5FL , amino acids 1–404 , Figure 1A ) yielded insoluble or marginally soluble material ( Qiao et al . , 2012; Lettman et al . , 2013 ) . To tackle this problem , we constructed a bacterial expression vector system harboring 13 different solubility-tags , which allowed us to obtain soluble SAS-5FL fused to MsyB ( Zou et al . , 2008 ) in quantities sufficient for biophysical analysis ( Figure 1—figure supplement 1 , which shows all recombinant proteins in this study ) . Characterization of such purified SAS-5FL using circular dichroism ( CD ) revealed the presence of protein aggregates ( Figure 1—figure supplement 2A , B ) , a conclusion also supported by size-exclusion chromatography multi-angle light scattering ( SEC-MALS , Figure 1—figure supplement 2C ) and negative-stain electron microscopy ( Figure 1—figure supplement 2D ) . We sought to locate the motif in SAS-5 that causes aggregation by expressing a series of MsyB-tagged C-terminal truncation constructs , and observed a significant increase in solubility when comparing a construct encompassing residues 2–279 with one containing residues 2–296 ( Figure 1—figure supplement 3 ) . Secondary structure prediction suggested a β-strand between residues 282 and 295 ( Figure 1—figure supplement 4 ) . Excising this region from untagged SAS-5FL ( SAS-5Δ282–295 ) or replacing it with a Gly-Ser-Ala-rich flexible linker of equal length ( SAS-5FLEX ) resulted in proteins with CD spectra characteristic of mainly α-helical proteins ( Figure 1B ) . These results suggest that the predicted β-strand between residues 282–295 of SAS-5 drives formation of large protein aggregates . 10 . 7554/eLife . 07410 . 003Figure 1 . SAS-5 comprises two independently folded domains . ( A ) Schematic representation of SAS-5 architecture showing the relative locations and residue boundaries of the coiled-coil and Implico domains , the putative SAS-4 binding site ( Hatzopoulos et al . , 2013 ) and the SAS-6 binding site ( Qiao et al . , 2012; Hilbert et al . , 2013 ) . ( B ) Overlaid CD spectra of SAS-5Δ282–295 and SAS-5FLEX samples recorded at 10°C , shown as per residue molar elipticity vs wavelength . The semi-quantitative contribution of secondary structure elements in each spectrum is deconvoluted in the bar charts on the left . Grey color corresponds to random coil , green to β-strand and red to α-helical segments . ( C ) Thermal unfolding profiles of the same samples monitored by recording molar elipticity at 222 nm as a function of temperature , and graphical representation of the melting transition temperatures observed in each sample . ( D–G ) Similar CD spectra and thermal unfolding profiles of ( D , E ) SAS-5CC and SAS-5Imp , and ( F , G ) SAS-5125–265 samples . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 00310 . 7554/eLife . 07410 . 004Figure 1—figure supplement 1 . Recombinant proteins sample quality . ( A , B ) Sections of SDS-PAGE showing Coomassie-stained samples of recombinant proteins used in this study . Panel A groups proteins with molecular weight over 30 kDa; panel B shows smaller proteins . Individual gel sections have been compressed or expanded along the vertical axis to match the molecular weight marker , and their contrast levels have been equalized . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 00410 . 7554/eLife . 07410 . 005Figure 1—figure supplement 2 . SAS-5 forms protein aggregates in vitro . ( A ) CD spectrum of SAS-5FL at 10°C shown as per residue molar elipticity against wavelength , and ( B ) thermal unfolding profile of the same protein monitored by recording molar elipticity at 222 nm as a function of temperature . The CD spectrum is highly similar to that of protein aggregates characterized elsewhere ( Digambaranath et al . , 2010 ) . ( C ) Overlay of SEC-MALS chromatograms of SAS-5FL samples at multiple concentrations , showing scaled light scattering intensity vs elution volume . ( D ) Representative negative stain electron micrograph of SAS-5FL and high-magnification view of the boxed section . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 00510 . 7554/eLife . 07410 . 006Figure 1—figure supplement 3 . A short SAS-5 segment promotes protein aggregation . ( A , B ) Coomassie-stained SDS-PAGE of metal-affinity purified SAS-5 variants , corresponding to MsyB-tagged full-length protein and C-terminal truncations ( A ) . The expected SAS-5-MsyB protein bands are indicated by black arrows . Soluble protein yields increased significantly upon truncation of residues 280–296 ( compare lanes 3 and 4 ) . Excision of this protein segment ( SAS-5Δ282–295-MsyB ) or substitution by a flexible linker ( SAS-5FLEX-MsyB ) improved soluble yields of the full-length protein ( B , compare to right-most lane of panel A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 00610 . 7554/eLife . 07410 . 007Figure 1—figure supplement 4 . SAS-5 secondary structure and disorder predictions . ( A ) Schematic representation and relative locations of secondary structure elements predicted from the SAS-5 amino acid sequence by PSIPRED ( Jones , 1999 ) . ( B ) Disorder probability per amino acid residue predicted from the SAS-5 sequence by DISOPRED3 ( Jones and Cozzetto , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 00710 . 7554/eLife . 07410 . 008Figure 1—figure supplement 5 . The SAS-5 N- and C-terminal segments are unstructured in isolation . ( A ) Overlaid CD spectra of SAS-5 N-terminal ( residues 2–122 ) and C-terminal ( residues 269–404 ) fragments recorded at 10°C . The semi-quantitative contribution of secondary structure elements in each spectrum is deconvoluted in the bar chart left . Grey colour corresponds to random coil , green to β-strand and red to α-helical segments . ( B ) Thermal unfolding profiles of the same samples based on their CD signal at 222 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 008 Interestingly , thermal unfolding of the soluble SAS-5Δ282–295 and SAS-5FLEX variants showed a two-step cooperative melting process , thus revealing the presence of two independently folded domains ( Figure 1C ) . Sequence-based prediction suggested the presence of a coiled coil spanning residues 125–180 , as well as of three tightly-spaced α-helices ( residues 210–265 ) ; these two elements are separated from each other by a presumed disordered linker of ∼30 residues ( Figure 1—figure supplement 4 ) . We expressed both the coiled coil ( SAS-5CC ) and the predicted helical region ( SAS-5Imp for Implico , see below ) and performed CD analysis . SAS-5CC displayed a double-minimum spectrum ( Figure 1D ) characteristic of α-helical coiled coils , and exhibited moderate thermal stability in isolation ( apparent melting transition temperature , Tm ∼ 37°C , Figure 1E ) . In contrast , CD of SAS-5Imp revealed a very stable ( Tm ∼ 72°C ) α-helical domain ( Figure 1D , E ) . A SAS-5 construct that combined both the coiled coil and the α-helical domain ( SAS-5125–265 , residues 125–265 ) showed CD spectra and a two-step thermal unfolding profile highly similar to that of SAS-5Δ282–295 and SAS-5FLEX ( compare Figure 1F , G with Figure 1B , C ) . We attribute the first melting transition of SAS-5125–265 ( Tm of ∼50°C ) to unfolding of the coiled-coil domain and the second ( Tm of ∼70°C ) to unfolding of the Implico domain . Moreover , CD spectra of isolated SAS-5 N-terminal ( residues 2–122 ) or C-terminal ( 269–404 ) fragments showed no persistent secondary structure or cooperative thermal unfolding ( Figure 1—figure supplement 5 ) , consistent with similar previous analysis of the SAS-5 N-terminus ( Shimanovskaya et al . , 2013 ) . Overall , these findings led us to conclude that the segment between residues 125–265 contains two independently folded domains and encompasses all structured elements of SAS-5 . We next set out to determine the X-ray crystallographic structures of the two independently folded domains . Native crystals of SAS-5CC diffracted to 1 . 8 Å resolution , whilst phases for structure solution were determined by single anomalous dispersion ( SAD ) using trimethyl lead acetate derivatized crystals ( Tables 1 , 2 ) . The structure of SAS-5CC suggested a parallel trimeric coiled coil , although this arrangement is distorted by crystal packing that intercalates the ends of successive triple helical bundles ( Figure 2A ) . To correct for these distortions , we performed molecular dynamics ( MD ) simulations starting from the crystallographic structure of SAS-5CC , which rapidly converged to a trimeric coiled-coil structure ( Figure 2B–D ) . SEC-MALS analysis of a slightly longer coiled-coil construct ( SAS-5CC-L , residues 101–206 ) demonstrated formation of a trimer in solution in a concentration-dependent manner , with apparent Kd of ∼3 μM ( Figure 2G , H ) . Both the crystallographic model and MD simulations show a series of hydrophobic residues ( W133 , M137 , L141 , I144 , I148 , L159 , M167 and M171 ) forming the core of the coiled coil and enabling trimer formation ( Figure 2E , F ) . Interestingly , we found that all these residues are conserved or conservatively substituted in SAS-5 homologs in other Caenorhabditis species , suggesting functional relevance ( Figure 2—figure supplement 1 ) . To investigate the stability of the coiled-coil domain across nematode evolution , we expressed constructs corresponding to SAS-5CC-L from Caenorhabditis brenneri , Caenorhabditis briggsae , Caenorhabditis remanei , Caenorhabditis sinica and Caenorhabditis tropicalis . In all cases , SEC-MALS analysis showed similar oligomerization properties and apparent affinities as for C . elegans SAS-5CC-L ( Figure 2—figure supplement 2 ) . Taken together , these data suggest that the trimeric coiled coil is an evolutionary conserved feature of nematode SAS-5 proteins . 10 . 7554/eLife . 07410 . 009Table 1 . Crystallographic data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 009ProteinSAS-5CCSAS-5CC ( Pb-derivative ) SAS-5ImpSAS-5Imp ( Hg-derivative ) PDB code4YV4–4YNH–Space groupP 21 21 21P 21 21 21P 1 21 1P 1 21 1Unit cell ( Å , ° ) a = 46 . 52;a = 46 . 32;a = 27 . 18;a = 27 . 85;b = 55 . 10;b = 52 . 99;b = 36 . 29;b = 36 . 65;c = 191 . 50c = 191 . 55c = 42 . 60;c = 42 . 13;β = 97 . 50β = 97 . 74BeamlineDLS/I04DLS/I03ESRF ID14-4DLS/I02Wavelength ( Å ) 0 . 97950 . 9470 . 9001 . 007Resolution range ( Å ) 41 . 90-1 . 8041 . 70-1 . 9018 . 49-1 . 0041 . 75-1 . 70High resolution shell ( Å ) 1 . 84-1 . 801 . 95-1 . 90†1 . 05-1 . 001 . 74-1 . 70Rmerge‡0 . 047 ( 1 . 092 ) 0 . 089 ( 1 . 088 ) 0 . 108 ( 0 . 763 ) 0 . 065 ( 0 . 541 ) Rpim‡0 . 034 ( 0 . 988 ) 0 . 014 ( 0 . 680 ) 0 . 044 ( 0 . 314 ) 0 . 017 ( 0 . 151 ) Completeness‡ ( % ) 98 . 1 ( 97 . 8 ) 80 . 1 ( 33 . 6 ) §99 . 4 ( 98 . 3 ) 85 . 6 ( 39 . 4 ) #Multiplicity‡2 . 9 ( 2 . 9 ) 9 . 3 ( 3 . 6 ) 6 . 9 ( 6 . 8 ) 18 . 4 ( 13 . 6 ) Mean I/σ ( I ) ‡7 . 4 ( 0 . 7 ) 13 ( 0 . 9 ) 12 . 0 ( 4 . 2 ) 26 . 5 ( 4 . 5 ) Phasing No . of heavy atom sites–9–2 Resolution–41 . 70-1 . 90–41 . 75-1 . 70 FOM initial¶–0 . 39–0 . 39 FOM DM–0 . 61–0 . 75Refinement statistics Rwork ( reflections ) 20 . 9% ( 30 , 118 ) –12 . 8% ( 42 , 016 ) – Rfree ( reflections ) 23 . 8% ( 1600 ) –14 . 9% ( 2232 ) –Number of atoms Protein atoms3647–2040 ( including H ) – Ligands25––– Water261–124–Average B factors ( Å2 ) Protein atoms50 . 9–10 . 9– Water46 . 5–23 . 5–RMSD from ideal values Bonds / angles ( Å/° ) 0 . 01 / 0 . 95–0 . 007 / 1 . 152–MolProbity statistics†† Ramachandran favored ( % ) 99 . 5%–99 . 2%– Ramachandran disallowed ( % ) 0%–0%– Clashscore ( percentile ) 2 . 49 ( 99th ) –0 . 98 ( 98th ) – MolProbity score ( percentile ) 1 . 11 ( 100th ) –0 . 79 ( 99th ) –Anisotropic diffraction of 2 . 17 Å , 2 . 59 Å and 1 . 73 Å highest resolution along the a* , b* and c* axes , respectively , based on a mean I/σ ( I ) > 2 . 0 criterion . †Anisotropic diffraction of 2 . 18 Å , 3 . 03 Å and 2 . 17 Å highest resolution along the a* , b* and c* axes , respectively , based on a mean I/σ ( I ) > 2 . 0 criterion . ‡By Aimless ( Evans and Murshudov , 2013 ) , values in parentheses correspond to the high resolution shell . §98 . 0% complete to 2 . 26 Å#98 . 8% complete to 2 . 03 Å¶From PHASER ( Mccoy et al . , 2007 ) . *From RESOLVE ( Terwilliger , 2000 ) . ††From MolProbity ( Chen et al . , 2010 ) . 10 . 7554/eLife . 07410 . 010Table 2 . Anisotropy correction statistics for native SAS-5CC crystallographic dataDOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 010Resolution ( Å ) Number of observed reflectionsRedundancyCompleteness ( % ) Rmerge ( % ) Mean I/σ ( I ) BeforeAfterBeforeAfterBeforeAfterBeforeAfterBeforeAfter8 . 05142814282 . 32 . 392 . 392 . 32 . 92 . 922 . 321 . 95 . 69261725982 . 52 . 592 . 192 . 23 . 33 . 322 . 321 . 94 . 65332233172 . 62 . 694 . 494 . 44 . 04 . 022 . 4224 . 03394939232 . 62 . 693 . 393 . 23 . 63 . 622 . 722 . 33 . 6458145272 . 72 . 795 . 895 . 83 . 73 . 821 . 721 . 33 . 29492549572 . 62 . 695 . 195 . 34 . 14 . 119 . 819 . 43 . 04576057032 . 82 . 896 . 996 . 75 . 05 . 017 . 116 . 82 . 85602660032 . 82 . 896 . 396 . 45 . 15 . 114 . 213 . 92 . 68669666492 . 92 . 997 . 897 . 86 . 56 . 412 . 111 . 92 . 55709870982 . 9398 . 698 . 57 . 57 . 510 . 19 . 92 . 43717871362 . 82 . 898 . 298 . 19 . 49 . 38 . 28 . 12 . 32782477813398 . 999 . 111 . 811 . 76 . 86 . 72 . 23784472332 . 82 . 697 . 491 . 115 . 914 . 95 . 25 . 42 . 158551660932 . 399 . 079 . 620 . 415 . 94 . 45 . 32 . 08875757263299 . 066 . 929 . 320 . 03 . 14 . 42 . 01870745422 . 91 . 598 . 751 . 841 . 822 . 92 . 34 . 11 . 959329335531 . 199 . 133 . 959 . 124 . 01 . 74 . 21 . 9915023412 . 90 . 798 . 120 . 593 . 330 . 21 . 13 . 81 . 859824153230 . 599 . 211 . 6114 . 028 . 60 . 94 . 51 . 897485022 . 80 . 197 . 93 . 9143 . 131 . 80 . 83 . 9total13331492 , 9602 . 9297 . 668 . 24 . 94 . 48 . 211 . 3Derived from the UCLA Diffraction Anisotropy Server ( Strong et al . , 2006 ) . 10 . 7554/eLife . 07410 . 011Figure 2 . The SAS-5 coiled coil forms a parallel trimer . ( A ) Schematic representation of the SAS-5CC crystallographic structure in purple . The protein N- and C-termini are indicated , as is the area where successive molecules intercalate in the crystal , leading to distortions of the coiled-coil structure; an intercalating molecule is shown here in white . ( B , C ) Average distance between the centers of the three α-helices at the N-terminus ( B ) or the C-terminus ( C ) of the coiled coil as a function of time in MD simulations . The distances at the initially frayed ends of the coiled coil decreased rapidly from ∼16 nm to ∼11 nm as the structure converged to a canonical trimeric arrangement . ( D ) Schematic representation of the coiled-coil structure at the start ( left , purple ) and the end ( right , grey ) of MD simulations . The helix centers whose distances are plotted in panels B and C are shown as blue and red spheres , respectively . ( E , F ) Magnified view of the coiled-coil hydrophobic core from the crystallographic structure . Residues at the C-terminus ( E ) and N-terminus ( F ) of the coiled coil are shown as sticks; charged substitutions at L141 and M167 disrupted coiled-coil oligomerization . ( G ) Overlay of SEC-MALS chromatograms of SAS-5CC-L at multiple concentrations , showing scaled light scattering intensity vs elution volume . The calculated molecular weight for each trace is shown as continuous line over the chromatogram peak , and corresponds to the right–hand axis . Chromatograms of samples at the highest two concentrations show evidence of protein overloading on the size-exclusion column . ( H ) Plot of average molecular weight from SEC-MALS analysis as a function of on-column protein concentration . The apparent Kd , and molecular sizes of monomers , dimers and trimers are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 01110 . 7554/eLife . 07410 . 012Figure 2—figure supplement 1 . Alignment of SAS-5 coiled coils from the Caenorhabditis genus . Alignment of nematode SAS-5 coiled-coil sequences from the Caenorhabditis genus by MergeAlign ( Collingridge and Kelly , 2012 ) and plotted by ESPript ( Robert and Gouet , 2014 ) . The protein secondary structure inferred from the crystallographic structure of SAS-5CC is shown schematically on top . Universally conserved residues are highlighted in red; conservatively substituted residues are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 01210 . 7554/eLife . 07410 . 013Figure 2—figure supplement 2 . The trimeric SAS-5 coiled coil is an evolutionary conserved feature . ( A ) Overlay of SEC-MALS chromatograms of a C . briggsae SAS-5 construct spanning residues 96–199 , equivalent to C . elegans SAS-5CC-L , in multiple concentrations , and ( B ) plot of average molecular weight calculated from SEC-MALS analysis as a function of on-column protein concentration . The apparent Kd and molecular sizes of monomers , dimers and trimers are indicated . Chromatograms of samples at the highest protein concentration show evidence of overloading the size exclusion column . ( C–J ) Similar chromatograms and plots of C . brenneri SAS-5 residues 1–82 ( C , D ) , C . remanei SAS-5 residues 98–206 ( E , F ) , C . sinica SAS-5 residues 94–204 ( G , H ) and C . tropicalis SAS-5 residues 94–199 ( I , J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 013 We next turned our attention to the second independently folded domain , SAS-5Imp . We found that this newly recognized domain is highly conserved amongst Caenorhabditis species ( Figure 3—figure supplement 1 ) . Importantly , SEC-MALS experiments revealed that SAS-5Imp forms a stable dimer ( Figure 3A ) . Crystals of SAS-5Imp diffracted to 1 . 0 Å resolution and experimental phases were determined by SAD from mercury acetate derivatized samples ( Table 1 ) . The SAS-5Imp structure revealed a dimer composed of interlocked chains arranged in an antiparallel fashion ( Figure 3B ) . Each chain features three α-helices ( A1 to A3 and B1 to B3 ) that interact in a pairwise manner ( A1–B3 , A2–B2 and A3–B1 ) . Proline residues at the linkers between helices allow for tight 90° turns ( Figure 3C ) , resulting in a very compact dimer of just ∼4 nm in diameter . The dimer interface ( Figure 3D ) involves the majority of hydrophobic core residues ( I219 , I225 , A229 , L230 , I232 , I233 , L237 , F243 , I247 , V250 , L251 ) ; thus , breaking the dimer is expected to unfold the SAS-5Imp domain . Given the high thermal stability of SAS-5Imp ( Figure 1E ) , this suggests formation of stable dimers even at very low protein concentrations . This interlocked tight dimer is a novel structural feature that we named Implico , from Latin to entangle . 10 . 7554/eLife . 07410 . 014Figure 3 . The SAS-5 Implico domain comprises a novel type of protein dimer . ( A ) Overlay of SEC-MALS chromatograms of SAS-5Imp in multiple concentrations showing formation of a stable dimer . ( B ) Two orthogonal views of the SAS-5Imp crystallographic structure in schematic representation . The α-helices are denoted A1–A3 and B1–B3 for the two protein chains ( red and grey , respectively ) . The N- and C-termini are indicated . ( C ) Magnified view of a single SAS-5Imp protein chain in schematic representation , showing the proline residues ( space-filling representation ) at the tight turns between α-helices . ( D ) Two views of the SAS-5Imp dimeric interface , with one chain represented as surface and the other as ribbon . Hydrophobic residues at the dimer interface are shown as sticks on the ribbon chain , and their location is indicated on the surface . Surface colors are a gradient from blue to orange representing residue hydrophobicity in the Kyte-Doolittle scale ( Kyte and Doolittle , 1982 ) . Blue corresponds to −4 . 5 in this scale ( most hydrophilic ) , white to 0 and orange to 4 . 5 ( most hydrophobic ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 01410 . 7554/eLife . 07410 . 015Figure 3—figure supplement 1 . Alignment of SAS-5 Implico domains from the Caenorhabditis genus . Alignment of nematode SAS-5 Implico domain sequences from the Caenorhabditis genus by MergeAlign ( Collingridge and Kelly , 2012 ) . The protein secondary structure inferred from the crystallographic structure of SAS-5Imp is shown schematically on top . Universally conserved residues are indicated by red highlight; conservatively substituted residues are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 015 Together , the two oligomerization domains of SAS-5 have the potential to drive formation of protein assemblies of higher-order than merely dimers or trimers . We envision that SAS-5 would readily dimerize even at low concentrations via the Implico domain , as oligomerization of this domain is tighter compared to that of the SAS-5 coiled coil . Such SAS-5 dimers could then come together to form higher-order configurations . To test this possibility , we performed SEC-MALS experiments on MsyB-tagged SAS-5FLEX , which revealed concentration-dependent oligomerization of SAS-5 from dimers to an approximately equimolar population of tetramers and hexamers ( Figure 4A , B ) . Analogous results were obtained with a MsyB-tagged SAS-5 construct that lacks the disordered protein C-terminus ( SAS-52–265 ) ( Figure 4C , D ) . Moreover , native mass-spectrometry of SAS-52–265 also supported the presence of higher-order protein assemblies up to hexamers in solution ( Figure 4E ) . Taken together , these data support the notion that the SAS-5 coiled coil promotes the association of protein dimers towards higher-order assemblies . Furthermore , our results are consistent with an earlier SEC-MALS analysis of MBP-tagged SAS-5 constructs of residues 1–260 , which showed formation of tetramers at an unspecified protein concentration ( Shimanovskaya et al . , 2013 ) . 10 . 7554/eLife . 07410 . 016Figure 4 . SAS-5 forms higher-order assemblies in solution . ( A ) Overlay of SEC-MALS chromatograms of MsyB-tagged SAS-5FLEX in multiple concentrations and ( B ) plot of average molecular weight from SEC-MALS analysis as a function of on-column protein concentration . The apparent Kd and molecular sizes of dimers , tetramers and hexamers are indicated . SAS-5FLEX-MsyB reaches an equimolar hexamer to tetramer ratio at the highest concentration we could assess . Fitting the experimental data under the assumption of ultimate hexamer formation yielded Kd values comparable to those of the SAS-5 coiled coil in isolation ( compared B with Figure 2H ) . ( C , D ) Similar SEC-MALS analysis of MsyB-tagged SAS-52–265 . SAS-52–265-MsyB reaches a 3:1 hexamer to tetramer ratio at the highest concentration point . ( E ) Native mass-spectrometry electrospray ionization spectrum of a 20 μM sample of SAS-52–265 , showing relative abundance of protein species as a function of mass to charge ratio . The charged states and protein oligomeric forms corresponding to specific peaks are indicated . Odd-numbered oligomeric forms ( monomers , trimers etc ) likely correspond to in-flight breakdown of higher- order protein assemblies . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 016 To further probe the architectural model of higher-order SAS-5 assemblies , we proceeded to disrupt each oligomerization interface by engineering charged residues in place of core hydrophobic amino acids . The CD spectra of the resulting SAS-5CC L141E mutant showed significant reduction of α-helical content in the coiled coil and a striking reduction in stability ( Tm < 10°C , Figure 5—figure supplement 1A , B ) , whereas that of the SAS-5Imp I247E mutant showed absence of α-helical structure and complete loss of cooperativity upon thermal unfolding ( Figure 5—figure supplement 1C , D ) . We next engineered the same substitutions in the SAS-52–265 construct to selectively disrupt each oligomerization interface whilst leaving the other intact . We found that the thermal unfolding profile of a SAS-52–265 L141E mutant showed a reduction in stability of the coiled coil by approximately 25°C , whereas stability of the Implico domain was not affected ( Figure 5A , B ) . Substituting an additional core hydrophobic residue of the coiled coil ( SAS-52–265 L141E/M167E ) effectively abrogated coiled coil formation , again with no effect on the Implico domain ( Figure 5A , B ) . As anticipated , coiled coil disruptions in either SAS-52–265 L141E or L141E/M167E mutants yielded a stable dimer in SEC-MALS experiments ( Figure 5C , D ) , mediated by the Implico domain . Furthermore , we found that a SAS-52–265 I247E mutant , where the Implico domain is disrupted , displayed a single-step thermal unfolding profile as measured by CD with a Tm similar to that of the SAS-5 coiled coil in isolation ( compare Figure 5A , B to Figure 1E ) . SEC-MALS analysis of the same construct shows a concentration-dependent trimerization , again consistent with formation of just the SAS-5 coiled coil ( Figure 5E , F ) . Finally , a SAS-5 mutant where both oligomerization interfaces are disrupted , SAS-52–265 L141E/I247E , showed no cooperativity upon thermal unfolding ( Figure 5A , B ) and remained monomeric in solution ( Figure 5G ) . Together , our results support the notion that the two SAS-5 oligomerization domains self-associate independently from one another , and that together they allow formation of higher-order SAS-5 assemblies . 10 . 7554/eLife . 07410 . 017Figure 5 . Sequential disruption of SAS-5 oligomeric interfaces . ( A ) Thermal unfolding profiles monitored by CD of SAS-52–265 wild-type and mutants that disrupt the coiled coil ( L141E , L141E/M167E ) or the Implico ( I247E ) oligomerization interface . ( B ) Graphical representation of the melting transition temperatures observed in each SAS-52–265 variant . ( C , D ) Overlay of SEC-MALS chromatograms of SAS-52–265 L141E ( C ) or L141E/M167E ( D ) variants in multiple concentrations . ( E ) Similar SEC-MALS analysis of SAS-52–265 I247E and ( F ) plot of average molecular weight from SEC-MALS analysis of this variant as a function of on-column protein concentration . The apparent Kd and molecular sizes of monomers , dimers and trimers are indicated . ( G , H ) Overlay of SEC-MALS chromatograms from the SAS-52–265 L141E/I247E variant ( G ) , and the MsyB-tagged SAS-52–265 wild-type as reference ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 01710 . 7554/eLife . 07410 . 018Figure 5—figure supplement 1 . Single amino acid mutations disrupt the SAS-5CC and SAS-5Imp domains . ( A ) Overlaid CD spectra of SAS-5CC wild-type and L141E mutant samples recorded at 10°C . The semi-quantitative contribution of secondary structure elements in each spectrum is deconvoluted in the bar charts on the left . Grey color corresponds to random coil , green to β-strand and red to α-helical segments . ( B ) Thermal unfolding profiles of the same samples from their CD signal at 222 nm , and graphical representation of the melting transition temperatures observed in each sample . ( C , D ) Similar CD spectra and thermal unfolding profiles of SAS-5Imp wild-type and I247E mutant samples . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 018 We next sought to assess the role of higher-order SAS-5 assemblies in vivo . To this end , we generated C . elegans transgenic animals expressing either GFP-SAS-5 ( wild-type ) , or GFP-SAS-5 L141E or GFP-SAS-5 I247E mutant versions . All exons of the sas-5 sequence were recoded in the transgenic constructs to confer resistance to RNAi directed against endogenous sas-5 . Worms were subjected to sas-5 RNAi and the resulting embryos analyzed by time-lapse differential interference contrast ( DIC ) microscopy as a means to assay centriole formation . In wild-type embryos ( Figure 6A and Video 1 ) , the sperm contributes the sole pair of centrioles to the newly fertilized embryo . Following the first round of centriole duplication in the zygote , the two centrosomes , now each with a pair of centrioles , direct assembly of a bipolar spindle during the first mitotic division . At the two-cell stage , centrioles separate and duplicate , leading to bipolar spindle assembly in each blastomere and resulting in a signature 4-cell stage configuration following mitotic exit ( Figure 6A and Video 1 ) . In sas-5 ( RNAi ) embryos , the sperm , which is not affected by RNAi under these experimental conditions , contributes two normal centrioles to the zygote; these paternally contributed centrioles each recruit pericentriolar material and direct bipolar spindle assembly during the first mitotic division ( Delattre et al . , 2004 ) . However , due to depletion of the maternal pool of SAS-5 , centriole duplication does not occur , leading to monopolar spindle assembly in each blastomere at the second cell cycle ( Figure 6B , Video 2 ) . As a result , a four-cell configuration is never observed in sas-5 ( RNAi ) embryos ( Figure 6B ) . 10 . 7554/eLife . 07410 . 019Figure 6 . Both SAS-5 oligomerization interfaces are essential for centriole duplication in C . elegans embryos . Wild type ( A , B ) , gfp::sas-5 ( C ) , gfp::sas-5[L141E] ( D ) , and gfp::sas-5[I247E] ( E ) adult worms were subjected to sas-5 ( RNAi ) or left untreated ( A ) . The resulting embryos were imaged by time-lapse DIC microscopy ( A–E , see corresponding Videos 1–5 ) or fixed and analyzed by immunofluorescence ( F–H ) . ( A–E ) Embryos at the onset of cleavage furrow ingression ( top ) and at the end of the second cell cycle ( bottom ) ; time stamp in mm:ss from the beginning of cleavage furrow ingression at the first mitosis . The percentage of embryos that reach the four-cell stage is shown below each image , along with the number of embryos ( n ) analyzed per condition . ( F–H ) Embryos stained with antibodies against GFP ( green ) and the centriolar marker IFA ( magenta ) ; DNA is visualized in blue . Scale bar 10 μm . Insets are fivefold enlargements of boxed regions . ( I ) Centrosomal GFP localization was scored in one-cell stage embryos as shown in panels F–H as absent , weak focus or strong focus . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 01910 . 7554/eLife . 07410 . 020Figure 6—figure supplement 1 . GFP-SAS-5 and mutants are expressed , and resistant to sas-5 RNAi . Wild type ( A , E , F ) , gfp::sas-5 ( B ) , gfp::sas-5 L141E ( C ) , and gfp::sas-5 I247E ( D ) worms were treated with sas-5 RNAi ( A–D ) or simultaneous gfp and sas-5 RNAi ( E , F ) , and the resulting embryos imaged by dual DIC and fluorescence time-lapse microscopy . Top panels: one-cell stage embryos , bottom panels: same embryo ∼25 min later . GFP images are maximum intensity projections of 8 z-sections taken at 0 . 5 μm intervals . Time from the first image is shown in mm:ss . Insets in panels B , C , E and F show 2 . 5 fold enlargements of boxed regions . ( G ) Quantification of GFP fluorescence intensity in live one-cell stage embryos , shown as a percentage of the average value for wild-type GFP-SAS-5 . Rescue of endogenous sas-5 depletion upon dual RNAi treatment depended on the level of exogenous GFP-SAS-5 expression . Importantly , note that levels of GFP-SAS-5 below those of either of the GFP-SAS-5 mutants , reduced using gfp RNAi , were able to rescue the sas-5 ( RNAi ) phenotype , indicating that the lack of rescue in the latter two strains is not due to insufficient expression levels . To rule out the possibility of sas-5 ( RNAi ) rescue being due to dilution of the sas-5 RNAi with gfp RNAi , embryos expressing GFP-SAS-5 L141E were scored under the same conditions , with all embryos showing the sas-5 ( RNAi ) phenotype ( data not shown ) Furthermore , GFP-SAS-5 levels at only ∼5% of the levels in the sas-5 ( RNAi ) condition , failed to rescue , further indicating that sas-5 ( RNAi ) is functional under these conditions . ( H ) Western blots of 50 adult worms from each indicated strain probed with antibodies against SAS-5 or GFP , as indicated , and against α-tubulin as loading control . Molecular weight markers are indicated as black dashes . Endogenous SAS-5 exists as two isoforms: the long 404 amino acid isoform and a shorter 288 amino acid isoform , which is likely the result of trans-splicing ( Wormbase , F35B12 . 5c ) . Note that exogenous gfp::sas-5 results in the expression of the longer isoform only . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 02010 . 7554/eLife . 07410 . 021Video 1 . Time-lapse DIC microscopy of C . elegans embryo over the first two cell cycles . Untreated wild type embryo . Frames were captured every 5 s and the movie is played at 12 frames/s . Embryo is oriented with anterior to the left and posterior to the right; elapsed time is shown in minutes and seconds . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 02110 . 7554/eLife . 07410 . 022Video 2 . Time-lapse DIC microscopy of C . elegans embryo over the first two cell cycles . Wild type embryo subjected to sas-5 ( RNAi ) . Frames were captured every 5 s and the movie is played at 12 frames/s . Embryo is oriented with anterior to the left and posterior to the right; elapsed time is shown in minutes and seconds . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 022 We confirmed the expression of the GFP fusion proteins by Western blot analysis and live imaging after sas-5 RNAi ( Figure 6—figure supplement 1B–D , G ) . Importantly , using time-lapse DIC microscopy to analyze the three transgenic lines generated in this study , we found that expression of GFP-SAS-5 fully rescued the sas-5 ( RNAi ) phenotype , allowing all analyzed embryos to progress to the 4-cell stage ( Figure 6C and Video 3 ) , as in untreated wild-type embryos . In stark contrast , GFP-SAS-5 mutants that disrupt either the coiled coil ( L141E ) or the Implico oligomerization domain ( I247E ) were unable to rescue the sas-5 ( RNAi ) phenotype , such that all embryos underwent monopolar spindle assembly in each blastomere at the second cell cycle and thus failed to result in a 4-cell configuration ( Figure 6D , E , and Videos 4 , 5 ) . We conclude that both SAS-5 oligomerization interfaces , and hence formation of SAS-5 assemblies larger than merely dimers or trimers , are essential for SAS-5 function in vivo . 10 . 7554/eLife . 07410 . 023Video 3 . Time-lapse DIC microscopy of C . elegans embryo over the first two cell cycles . Embryo expressing GFP-SAS-5 treated with sas-5 ( RNAi ) . Frames were captured every 5 s and the movie is played at 12 frames/s . Embryo is oriented with anterior to the left and posterior to the right; elapsed time is shown in minutes and seconds . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 02310 . 7554/eLife . 07410 . 024Video 4 . Time-lapse DIC microscopy of C . elegans embryo over the first two cell cycles . Embryo expressing GFP-SAS-5 L141E treated with sas-5 ( RNAi ) . Frames were captured every 5 s and the movie is played at 12 frames/s . Embryo is oriented with anterior to the left and posterior to the right; elapsed time is shown in minutes and seconds . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 02410 . 7554/eLife . 07410 . 025Video 5 . Time-lapse DIC microscopy of C . elegans embryo over the first two cell cycles . Embryo expressing GFP-SAS-5 I247E treated with sas-5 ( RNAi ) . Frames were captured every 5 s and the movie is played at 12 frames/s . Embryo is oriented with anterior to the left and posterior to the right; elapsed time is shown in minutes and seconds . DOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 025 We next investigated the distribution of the fusion proteins in one-cell stage embryos treated with sas-5 ( RNAi ) by live imaging . Whereas GFP-SAS-5 was observed in foci at centrosomes ( Figure 6—figure supplement 1B ) , foci were not seen for either of the oligomerization mutant versions ( Figure 6—figure supplement 1C , D ) . In order to investigate the localization of the fusion proteins using a more sensitive approach , we analyzed fixed one-cell stage embryos by immunofluorescence with antibodies against GFP and the centriolar marker IFA ( Figure 6F–H ) . We found that whereas GFP-SAS-5 is enriched at centrioles ( Figure 6F , I and Delattre et al . , 2004 ) , as was also observed by live imaging , the situation differed for both oligomerization interface mutants . In the case of GFP-SAS-5 L141E , only ∼30% of centrosomes exhibited a clear GFP focus , with the remaining centrosomes exhibiting either no GFP focus , or only a weak focus ( Figure 6G , I ) . In the case of GFP-SAS-5 I247E , the vast majority of embryos exhibited no detectable GFP focus at centrioles ( Figure 6H , I ) . We conclude that SAS-5 oligomerization is important for centriole localization , with the Implico domain playing a particularly critical role in this respect . Moreover , the presence of GFP-SAS-5 L141E at centrioles in one third of embryos , combined with the fact that this mutant completely failed to rescue the phenotype , raises the possibility that formation of higher-order SAS-5 assemblies is essential for function even if centriolar localization has been achieved . Cottee , Muchalik et al . ( Cottee et al . , 2015 ) have now shown that Drosophila Ana2 and vertebrate STIL , the functional equivalents of SAS-5 , can also form higher-order oligomers that are essential for function . Both Ana2 and STIL tetramerize through their coiled-coil domains . It remains to be determined whether the different SAS-5 and Ana2/STIL architectures reflect a divergence in centriole architecture between these organisms . We have constructed a line of entero-bacterial expression vectors , called pFloat , derived from pET30a ( Novagen , Madison , WI ) and engineered to include N-terminal His6- and solubility tags , and a human rhinovirus 3C protease cleavage sequence . To accelerate construct screening we further inserted a ccdB promoter and gene cassette , amplified from pDONR211 ( Invitrogen , Grand Island , NY ) , inside the multi-cloning site . Various solubility tags included thioredoxin ( TRX ) , glutathione S-transferase ( GST ) , S-tag , colicin E9 immunity protein ( Im9 ) , maltose-binding protein ( MBP ) , small ubiquiting-like modifier ( SUMO ) , haloalkane dehalogenase ( HALO ) , trigger factor ( TF ) , N utilization substance protein A ( NusA ) , acidic protein MsyB , yellow fluorescent protein ( YFP ) and enhanced green fluorescent protein ( eGFP ) ; the solubility tag sequences were amplified from the pOPIN ( Bird , 2011 ) and pGEX ( GE Healthcare , Little Chalfont , UK ) vector systems , or directly from the Escherichia coli strain BL21 genome . All amplifications were performed using high fidelity Phusion polymerase ( New England Biolabs , Ipswich , MA ) , and vectors were constructed using seamless cloning methods ( Gibson Assembly , New England Biolabs ) . SAS-5 gene fragments from C . brenneri , C . remanei , C . sinica and C . tropicalis were synthesised by IDT , Coralville , IA . C . briggsae SAS-5 was amplified from genomic DNA ( gifted by Marie-Anne Félix , Ecole Normale Supérieure ) . Full-length or gene fragments encoding for SAS-5 constructs were cloned in pFloat vectors , transformed to E . coli strain BL21 ( DE3 ) Rosetta2 ( Novagen ) and grown at 37°C in lysogeny broth ( LB ) media . Protein expression was induced at OD600 0 . 6 by isopropyl-β-D-1-thiogalactopyranoside ( IPTG , 0 . 5 mM final concentration ) and lowering the temperature to 18°C . After ∼16 hr incubation cells were harvested and resuspended in PBS solution ( 10 mM Na2HPO4 , 1 . 8 mM KH2PO4 , 137 mM NaCl , 2 . 7 mM KCl , pH 7 . 4 ) supplemented with 350 mM NaCl , 0 . 1 mM phenylmethanesulfonyl fluoride ( PMSF ) , a protease inhibitor cocktail ( cOmplete , Roche , Basel , Switzerland ) , 0 . 25 mg/ml lysozyme and 10 μg/ml DNAse I . Resuspended cells were disrupted by sonication while chilled on ice , and cell lysate was cleared by centrifugation at 40 , 000×g and 0 . 22 μM filtration . Imidazole at 20 mM final concentration was added to clarified lysates prior to incubation with TALON ( GE Healthcare ) metal affinity resin pre-equilibrated with PBS supplemented with 350 mM NaCl and 20 mM imidazole . Resin beads were washed with PBS supplemented with 350 mM NaCl , 1 mM β-mercaptoethanol and 20 mM imidazole , and protein was eluted in PBS supplemented with 350 mM NaCl , 1 mM β-mercaptoethanol and 500 mM imidazole . Further purification of proteins was performed at 4°C by anion- ( HiTrap Q HP , GE Healthcare ) or cation-exchange ( HiTrap SP HP , GE Healthcare ) chromatography in 10 mM Tris–HCl or Na2HPO4 , 50 mM NaCl , 3 mM β-mercaptoethanol , pH 8 . 0 ( for anion-exchange ) or pH 6 . 5 ( for cation-exchange ) running buffer , and proteins were eluted using a NaCl gradient to 2 M . His6- and solubility-tags were cleaved , whenever necessary , by overnight incubation with Prescission protease ( GE Healthcare ) ; proteins were further purified by size-exclusion chromatography in HiLoad Superdex 75 , Superdex 200 or Superose 6 columns ( GE Healthcare ) equilibrated in 10 mM Tris–HCl , 300 mM NaCl , 2 mM β-mercaptoethanol , pH 7 . 5 buffer . Pure protein fractions were buffer-exchanged by dialysis , and concentrated by centrifugal ultrafiltration . Protein concentration was estimated by UV absorption at 280 nm , and protein identity confirmed by electrospray ionization mass-spectrometry . Far-UV CD spectra ( 260–180 nm ) were measured using a Jasco ( Easton , MD ) J-815 Spectropolarimeter at 10°C . Protein samples were exchanged to a 10 mM NaH2PO4 , 100 mM NaF , 1 mM β-mercaptoethanol , pH 7 . 5 buffer . CD sample concentrations are shown in Table 3 . Data were truncated at high tension voltage exceeding 800 V . After buffer background subtraction , data were smoothed using a Savitzky–Golay filter ( Savitzky and Golay , 1964 ) . CD spectra were deconvoluted using the DICHROWEB server ( Whitmore and Wallace , 2008 ) . Measurements of thermal stability monitored CD signal at 222 nm while temperature was increased by 1°C/min between 4°C and 90°C . 10 . 7554/eLife . 07410 . 028Table 3 . CD and SEC-MALS experimental detailsDOI: http://dx . doi . org/10 . 7554/eLife . 07410 . 028CDProtein nameConcentration ( μM ) SAS-5FL1 . 5SAS-5Δ282–2953SAS-5FLEX3SAS-52–2656SAS-52–265 L141E6SAS-52–265 L141E/M167E6SAS-5–265 I247E6SAS-52–265 L141E/I247E6SAS-5125–26510SAS-5CC20SAS-5CC L141E20SAS-5Imp20SAS-5Imp I247E20SAS-5 N-terminus20SAS-5 C-terminus5SEC-MALSProtein nameMonomeric MW ( kDa ) Size exclusion columnMinimum concentration ( μM ) Maximum concentration ( μM ) SAS-5FL46Superose60 . 100 . 57SAS-5FLEX-MsyB62Superose60 . 039 . 5SAS-52–265-MsyB ( Figure 4 ) 46Superose60 . 0241SAS-52–265-MsyB ( Figure 5 ) 46Superdex2000 . 0541SAS-52–265 L141E46Superdex2001 . 3263SAS-52–265 L141E/M167E46Superdex2001 . 1663SAS-52–265 I247E46Superdex2000 . 0563SAS-52–265 L141E/I247E46Superdex2000 . 8263SAS-5Imp6 . 5Superdex752 . 9292SAS-5CC-L12 . 5Superdex750 . 18152C . briggsae SAS-5 ( 96–199 ) 12 . 7Superdex750 . 10149C . brenneri SAS-5 ( 1–82 ) 10 . 2Superdex750 . 21186C . remanei SAS-5 ( 98–206 ) 13Superdex750 . 21146C . sinica SAS-5 ( 94–204 ) 13Superdex750 . 16145C . tropicalis SAS-5 ( 94–199 ) 12 . 5Superdex750 . 12152Refers to on-column concentration calculated from the protein differential refractive index . SEC-MALS analysis was performed at 20°C using analytical Superdex 75 , Superdex 200 or Superose 6 columns ( GE Healthcare ) and a Shimadzu ( Kyoto , Japan ) chromatography system , connected in-line to a Heleos8+ multi-angle light scattering detector and an Optilab T-rEX refractive index ( RI ) detector ( Wyatt Technologies , Goleta , CA ) . A 1:1 dilution series of protein samples in 10 mM Tris–HCl , 300 mM NaCl , 2 mM β-mercaptoethanol , pH 7 . 5 buffer were injected in this system , and the resulting MALS , RI and UV traces processed in ASTRA 6 ( Wyatt Technologies ) . On-column protein concentration was calculated from the differential RI , assuming dn/dc of 0 . 1850 ml/g . SEC-MALS data were fit with a two-state model to derive association parameters , and the resulting Kd estimated from the inflection point of the fit . Details of SEC-MALS sample concentrations , monomeric molecular weight and column used are shown in Table 3 . Native mass-spectrometry was performed on protein samples in 250 mM NH4Acetate , pH 7 . 5 buffer using a modified Synapt G1 High Definition Mass Spectrometry Quadrupole Time of Flight instrument ( Waters , Milford , MA ) ( Bush et al . , 2010 ) calibrated using 10 mg/ml CsI . 2 μl aliquots of sample were delivered by nano-electrospray ionization via gold-coated capillaries , prepared in house ( Hernandez and Robinson , 2007 ) . Instrumental parameters were as follows: source pressure 6 . 0 mbar , capillary voltage 1 . 20 kV , cone voltage 50 V , trap energy 10 V , bias voltage 5 V and trap pressure 0 . 0163 mbar . For negative stain electron microscopy SAS-5FL protein samples ( 5 μl , 0 . 1 mg/ml ) in 10 mM Tris–HCl , 300 mM NaCl , 2 mM β-mercaptoethanol , pH 7 . 5 buffer were applied to glow-discharged homemade carbon-coated copper mesh grids and allowed to adsorb for 1 min . After blotting off any excess liquid , negative stain was applied by adding 3 μl of 1% wt/vol uranyl acetate to the grid . Blotting and staining was repeated 2 times , and the grids were then allowed to air dry for 10 min . The grids were imaged in a JEOL ( Tokyo , Japan ) JEM-2010 electron microscope equipped with a high brightness LaB6 filament and an ORIUS SC1000 ( Model 832 ) camera ( Gatan , Abingdon , UK ) . Images were acquired using an accelerating voltage of 200 kV , at a nominal magnification of 16 , 000× to 52 , 000× . Crystals were obtained using the sitting drop vapor diffusion technique . A Mosquito robot ( TTP LabTech , Melbourn , UK ) was used to setup 200 nL-size drops with 1:1 and 1 . 3:0 . 7 ratios of protein to mother liquor . For SAS-5CC , protein in 10 mM Tris–HCl pH 7 . 5 , 200 mM NaCl buffer and at concentration of 20 mg/ml was mixed with 0 . 1 M MOPS/HEPES-Na pH 7 . 5 , 12 . 5% wt/vol PEG 1000 , 12 . 5% wt/vol PEG 3350 , 12 . 5% wt/vol MPD , 0 . 03 M NaNO3 , 0 . 03 M Na2HPO4 , 0 . 03 M ( NH4 ) 2SO4 . Crystals developed at 4°C in 30 days , were flash-cooled in liquid nitrogen and diffracted to a maximum resolution of 1 . 73 Å at the Diamond Light Source ( DLS , Harwell , UK ) beamline I04 . The space group was determined as P212121 with eight molecules per asymmetric unit . Crystallographic data ( statistics in Table 1 ) were integrated in MOSFLM ( Leslie and Powell , 2007 ) or XDS ( Kabsch , 2010 ) and scaled by Aimless ( Evans and Murshudov , 2013 ) or XSCALE ( Kabsch , 2010 ) . Scaling reveals substantial diffraction anisotropy , which was corrected using the Diffraction Anisotropy Server ( UCLA Molecular Biology Institute , Strong et al . , 2006 ) whilst applying a high resolution cutoff of 1 . 8 Å . Statistics of the anisotropically corrected data are shown in Table 2 . Phase information for SAS-5CC was obtained from a highly redundant dataset with maximum resolution of 1 . 9 Å collected on Pb-derivatized crystals at a wavelength of 0 . 947 Å at the DLS beamline I03 . For Pb-derivatization the native SAS-5CC crystals were incubated with 1 mM trimethyl lead ( IV ) acetate for 16 hr prior to cooling . Phasing by SAD was performed using PHENIX . autosol ( Adams et al . , 2002 ) which located and refined 9 Pb sites to produce a density map with initial figure of merit of 0 . 36 . Initial model building was done with PHENIX . autobuild ( 291 residues built , 222 identified out of 427 residues in the asymmetric unit ) . Iterative model building with COOT ( Emsley and Cowtan , 2004 ) and refinement against the anisotropically corrected native data in Buster 2 . 10 ( Bricogne et al . , 2011 ) using automatic NCS and TLS restraints ( Smart et al . , 2012 ) yielded the final SAS-5CC model . For SAS-5Imp , protein in 10 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM DTT buffer and at concentration of 10 mg/ml was mixed with 0 . 1 M MES/imidazole pH 6 . 4 , 10% wt/vol PEG 20 , 000 , 20% vol/vol PEG MME 550 , 0 . 1 M carboxylic acids mixture ( Gorrec , 2009 ) . Crystals developed at 18°C in 3 days , were cryo-protected by brief immersion to mother liquor supplemented with 20% vol/vol glycerol , flash-cooled in liquid nitrogen and diffracted to 1 . 0 Å at the European Synchrotron Radiation Facility ( ESRF , Grenoble , France ) beamline ID14-4 . The space group was determined as P1211 with two molecules per asymmetric unit . Data integration and scaling was performed as in SAS-5CC ( Table 1 ) , however no anisotropy correction was required . Phase information was obtained from a highly redundant dataset with maximum resolution of 1 . 7 Å collected on Hg-derivatized crystals at a wavelength of 1 . 007 Å at the DLS beamline I02 . For Hg-derivatization native SAS-5Imp crystals were incubated with 2 mM mercury ( II ) acetate for 16 hr prior to cooling . Phasing by SAD was performed using PHENIX . autosol ( Adams et al . , 2002 ) which located and refined 2 Hg sites to produce a density map with initial figure of merit of 0 . 39 . Initial model building was done with PHENIX . autosol ( 86 residues built and 51 identified out of 116 residues in the asymmetric unit ) . Iterative model building with COOT ( Emsley and Cowtan , 2004 ) and refinement against the native data in PHENIX . refine ( Adams et al . , 2002 ) using explicit hydrogens and atom-level anisotropic restraints . Model quality was assessed by MolProbity ( Chen et al . , 2010 ) . For graphical representation , we used UCSF Chimera ( Pettersen et al . , 2004 ) . Initial crystal inspection and data collection took place at the DLS beamline I04-1 . All MD simulations were performed using GROMACS 4 ( Pronk et al . , 2013 ) and the GROMOS96 53a6 force field ( Oostenbrink et al . , 2004 ) . The crystal structure of SAS5CC was placed in a cubic simulation box of 1452 nm3 volume with periodic boundary conditions . The protein was solvated by adding 10 , 637 explicit SPC water molecules . Na+ and Cl− ions were added to the final concentration of 0 . 1 M . The system was first minimized with the steepest descend algorithm , equilibrated in constant temperature ( NVT , 300 K ) for 100 ps , and followed by equilibration in constant temperature and pressure ( NPT , 1 bar ) for 500 ps . The pressure was controlled with an isotropic Parrinello-Rahman barostat ( Parrinello and Rahman , 1981 ) applied to the entire system with a time constant of 2 . 0 ps and compressibility of 4 . 5 × 10−5 bar−1 . The temperature was controlled with two velocity rescaling thermostats ( Berendsen et al . , 1984 ) applied to the protein and solvent , with a time constant of 0 . 1 ps . Electrostatic interactions were calculated using the particle-mesh Ewald ( Darden et al . , 1993 ) summation method . Ten replicate simulations of 45 ns duration were performed in the Oxford Advanced Research Computing facility . The genomic sequence of sas-5 ( Wormbase , F35B12 . 5a ) was recoded for resistance to RNAi directed against the endogenous gene , and synthesized by Genscript ( Piscataway , NJ ) . Intronic sequences were kept intact , as well as the bases neighboring the splice sites: 3 bases at 3′ ends and 6 bases at 5′ ends of exons . The codon adaptation index was preserved at the level of the wild type ( WT: 0 . 777 , recoded: 0 . 712 ) . The recoded gene was amplified and cloned in pIC26 using the SpeI restriction site . pIC26 contains a pie-1 promoter and 3′UTR , as well as a GFP coding sequence fused upstream of and in frame with sas-5 ( Cheeseman and Desai , 2005 ) . C . elegans culture was performed according to standard procedures ( Brenner , 1974 ) . GFP–SAS-5 transgenic animals were generated by bombardment ( Praitis et al . , 2001 ) . Integrated lines were recovered after bombardment with gfp::sas-5 L141E ( strain GZ1301; isIs51{pie-1::gfp::sas-5[recL141E]} ) and I247E ( strain GZ1302; isIs52{pie-1::gfp::sas-5[recI247E]} ) , but GFP-SAS-5 bombardment led to the recovery of non-integrated lines . For these experiments a line with <50% transmission of the extra chromosomal array was used ( strain GZ1300; isEx5{pie-1::gfp::sas-5[rec]} ) . RNAi was carried out by selecting L3–L4 wild-type hermaphrodites and feeding them for 26 hr at 24°C using the sas-5 ( RNAi ) feeding strain , which targets nucleotides 301–1170 of the genomic sequence ( Delattre et al . , 2004 ) . Embryos were prepared for immunofluorescence essentially as described ( Gönczy et al . , 1999 ) , with a fixation time of 2 min in −20°C methanol and an additional blocking step in PBS , 2% wt/vol BSA before antibody incubation . Primary and secondary antibodies were diluted in PBS as follows: rabbit anti-GFP ( gift from Viesturs Simanis , EPFL , 1/500 ) , mouse anti-IFA ( Leung et al . , 1999 , 1/50 ) , goat anti-GFP ( Abcam , Cambridge , UK , ab6673 , 1/250 ) , goat anti-rabbit Alexa Fluor 488 , goat anti-mouse Alexa Fluor 568 , donkey anti-goat Alexa Fluor 488 , donkey anti-mouse Alexa Fluor 568 ( Life Technologies , Grand Island , NY , all 1/500 ) . Primary antibodies were incubated overnight at 4°C and secondary incubations were at room temperature for 1 hr . 1 μg/ml Hoechst 33 , 258 ( Sigma-Aldrich , St . Louis , MO ) was used to visualize DNA . Time-lapse DIC microscopy of early embryos was carried out as described ( Gönczy et al . , 1999 ) , recording one image every 5 s at 24°C . A Zeiss ( Oberkochen , Germany ) LSM 700 confocal microscope was used for fluorescence microscopy . Z-sections were imaged at an interval of ∼0 . 3 μm . Fluorescent images in Figure 6F–H are maximum intensity projections . Dual time-lapse and fluorescence DIC imaging was performed on a Zeiss Axioplan 2 ( Bellanger and Gönczy , 2003 ) . The motorised filter wheel , two external shutters , and the 1392 × 1040 pixel , 12-bit Photometrics CoolSNAP ES2 ( Photometrics , Tucson , AZ ) were controlled by μManager ( Edelstein et al . , 2010 ) . Images were acquired with an exposure time of 100 ms for the DIC and 200 ms for the GFP channel using the Zeiss Filter Set 10 ( GFP ) . The GFP images presented in Figure 6—figure supplement 1 are maximum intensity projections of 8 z-sections acquired with an interval of 0 . 5 μm . For quantification of fluorescence intensities , the integrated intensity was measured from three areas of fixed size within each embryo and then averaged . The average background fluorescence of wild-type ( GFP-negative ) embryos was subtracted , then values were normalized to the average value of the GFP-SAS-5 wild type intensity . All images were processed and fluorescence intensities measured using ImageJ ( Schneider et al . , 2012 ) . Embryos were synchronized by bleaching adult worms , hatched and matured at 24°C; 50 adult worms were then collected manually , boiled in Laemmli buffer for 5 min , sonicated , and loaded on a 4–12% wt/vol SDS-PAGE ( Biorad , Hercules , CA ) . Proteins were transferred to a 0 . 45 μm nitrocellulose membrane , blocked with 5% wt/vol milk powder in PBS , 0 . 1% vol/vol Tween-20 , and blotted with rabbit anti-SAS-5 ( full length , 1/1000 , ( Delattre et al . , 2004 ) ) or rabbit anti-GFP ( 1/1000 , gift from Viesturs Simanis , EPFL ) , then re-blotted with mouse anti-α-tubulin ( 1/10 , 000 , DM1α , Sigma-Aldrich ) antibodies . Immunocomplexes were visualized using anti-rabbit or anti-mouse HRP conjugated secondary antibodies at 1/5000 dilution ( Promega , Madison , WI ) , chemiluminescence kit ( Roche ) , and X-ray films ( Fujifilm , Tokyo , Japan ) . SAS-5 protein sequences in this study correspond to UniProt database entries Q20010 ( C . elegans ) , Q60M48 ( C . briggsae ) , G0NH40 ( C . brenneri ) and E3LPN2 ( C . remanei ) . C . sinica and C . tropicalis SAS-5 sequences were identified from the respective genomes by blast searches starting from the C . elegans variant . For C . sinica the University of Edinburgh sp . 5 genome project ( Csp5_scaffold_00353 ) was used , while for C . tropicalis the Genome Institute at Washington University Caenorhabditis sp . 11 genome project ( Csp11 . Scaffold629 ) served as reference . The flexible linker used in SAS-5FLEX to replace the ‘sticky’ amino acid segment has the sequence SGAAGSSGAAGSSG .	Most animal cells contain structures known as centrioles . Typically , a cell that is not dividing contains a pair of centrioles . But when a cell prepares to divide , the centrioles are duplicated . The two pairs of centrioles then organize the scaffolding that shares the genetic material equally between the newly formed cells at cell division . Centriole assembly is tightly regulated and abnormalities in this process can lead to developmental defects and cancer . Centrioles likely contain several hundred proteins , but only a few of these are strictly needed for centriole assembly . New centrioles usually assemble from a cartwheel-like arrangement of proteins , which includes a protein called SAS-6 . In the worm Caenorhabditis elegans , SAS-6 associates with another protein called SAS-5 . This interaction is essential for centrioles to form , but the reason behind this is not clearly understood . Now , Rogala et al . have used a range of techniques including X-ray crystallography , biophysics and studies of worm embryos to investigate the role of SAS-5 in C . elegans . These experiments revealed that SAS-5 proteins can interact with each other , via two regions of each protein termed a ‘coiled-coil’ and a previously unrecognized ‘Implico domain’ . These regions drive the formation of assemblies that contain multiple SAS-5 proteins . Next , Rogala et al . asked whether SAS-5 assemblies are important for centriole duplication . Mutant worm embryos , in which SAS-5 proteins could not interact with one another , failed to form new centrioles . This resulted in defects with cell division . An independent study by Cottee , Muschalik et al . obtained similar results and found that the fruit fly equivalent of SAS-5 , called Ana2 , can also self-associate and this activity is required for centriole duplication . Further work is now needed to understand how SAS-5 and SAS-6 work with each other to form the initial framework at the core of centrioles .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	The Caenorhabditis elegans protein SAS-5 forms large oligomeric assemblies critical for centriole formation
CaV1 . 3 channels regulate excitability in many neurons . As is the case for all voltage-gated channels , it is widely assumed that individual CaV1 . 3 channels behave independently with respect to voltage-activation , open probability , and facilitation . Here , we report the results of super-resolution imaging , optogenetic , and electrophysiological measurements that refute this long-held view . We found that the short channel isoform ( CaV1 . 3S ) , but not the long ( CaV1 . 3L ) , associates in functional clusters of two or more channels that open cooperatively , facilitating Ca2+ influx . CaV1 . 3S channels are coupled via a C-terminus-to-C-terminus interaction that requires binding of the incoming Ca2+ to calmodulin ( CaM ) and subsequent binding of CaM to the pre-IQ domain of the channels . Physically-coupled channels facilitate Ca2+ currents as a consequence of their higher open probabilities , leading to increased firing rates in rat hippocampal neurons . We propose that cooperative gating of CaV1 . 3S channels represents a mechanism for the regulation of Ca2+ signaling and electrical activity . CaV1 . 3 channels are widely expressed in neurons throughout the brain and spinal cord ( Tan et al . , 2011 ) , where they serve a number of critical functions including the modulation of resting potentials , the amplification of synaptic currents and the generation and shaping of repetitive firing ( Guzman et al . , 2009; Olson et al . , 2005; Striessnig et al . , 2006 ) . These channels are dihydropyridine-sensitive L-type Ca2+ channels composed of a pore-forming CaV1 . 3α1 subunit and accessory β and α2-δ subunits . The carboxy-terminus ( C-terminus ) of the α1D subunit is structurally complex , containing an EF hand domain as well as pre-IQ and IQ domains to which the Ca2+-binding protein calmodulin ( CaM ) binds ( Ben-Johny and Yue , 2014 ) . Alternative splicing results in the expression of 'long' and 'short' CaV1 . 3 channel isoforms that differ in the length of the C-terminus ( Singh et al . , 2008 ) . The splice variant 42A ( CaV1 . 3S ) has a short C-terminus of 183 amino acids long compared to the 695 amino acids of the long isoform ( CaV1 . 3L ) . The CaV1 . 3S channels lack the so-called C-terminal modulatory domain ( CTM ) , comprised of proximal ( PCRD ) and distal ( DCRD ) regulatory domains that block CaM binding to the IQ domain ( Figures 1A and B , left ) . As a consequence , CaV1 . 3S channels activate at lower voltages , have a higher open probability , and inactivate faster than CaV1 . 3L channels ( Bock et al . , 2011; Singh et al . , 2008; Tan et al . , 2011 ) . 10 . 7554/eLife . 15744 . 003Figure 1 . Ca2+ enhances the activity of CaV1 . 3S , but not CaV1 . 3L , channels . ( A ) Left: Schematic of CaV1 . 3L channel splice variant , depicting the domains important for Ca2+-mediated regulation: pre-IQ ( green ) , IQ ( blue ) , proximal and distal C-terminal regulatory domains ( PCRD , DCRD , gray ) . Middle: Representative ICa and IBa of CaV1 . 3L channels expressed in tsA-201 cells . Currents were evoked by a 300-ms depolarization from holding potential of -80 mV to a test potential of -10 mV , with 2 mM Ba2+ ( black ) or 2 mM Ca2+ ( red ) as the charge carrier in the same cell . Right: Voltage dependence of CaV1 . 3L channel CDI . r300 is the fraction of ICa or IBa remaining after 300 ms . f300 is the difference between ICa and IBa r300 values at 0 mV . ( B ) Left: Schematic of CaV1 . 3S channel splice variant . Middle: Representative ICa and IBa of CaV1 . 3S channels . Right: Voltage dependence of CaV1 . 3S channel CDI , format as in ( A ) . ICa is presented as normalized to IBa , currents analyzed for these experiments were in a range between 200 and 900 pA . ( C and D ) Representative iCa single channel recordings from CaV1 . 3L ( C ) and CaV1 . 3S channels ( D ) expressed in tsA-201 cells during step depolarizations from -80 to -30 mV . ( E and F ) all-points iCa amplitude histograms for CaV1 . 3L ( E ) and CaV1 . 3S channels ( F ) , the black line is the best fit to the data with a multi-Gaussian function with a quantal unit value of -0 . 48 ± 0 . 07 for CaV1 . 3L and -0 . 49 ± 0 . 01 pA for CaV1 . 3S channels , respectively ( constructed from n = 6 cells each ) . Single channel recordings were also performed using Ba2+ as the charge carrier for both channels ( see Figure 1—figure supplement 1 ) . ( G ) Ensemble average single-channel currents from multiple sweeps . ( H ) Current-voltage relationships of CaV1 . 3L currents ( left ) and CaV1 . 3S currents ( right ) in the presence of 2 mM Ca2+ ( red ) or Ba2+ ( black ) as the charge carrier . Data were normalized to the maximum current in the presence of Ba2+ . Symbols are averages of 7 cells ± SEM . ( I ) Scatter plots of NPo ( at -10 mV ) of CaV1 . 3L ( left ) and CaV1 . 3S ( right ) channels in the presence of Ba2+ and Ca2+ . ( J ) Change in NPo for CaV1 . 3L and CaV1 . 3S channels for currents recorded in the presence of Ca2+ and then Ba2+ . The horizontal bar shows the mean value ( p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 00310 . 7554/eLife . 15744 . 004Figure 1—figure supplement 1 . Single-channel recordings of iBa for CaV1 . 3S and CaV1 . 3L channels . ( A and B ) Representative iBa single channel recordings from CaV1 . 3S ( A ) and CaV1 . 3L channels ( B ) expressed in tsA-201 cells during step depolarizations from -80 to -30 mV . ( C ) Average single-channel current-voltage relationship ± SEM . Data from 5 cells for CaV1 . 3S and 7 cells for CaV1 . 3L . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 004 CaV1 . 3 channels carry about 20% of the L-type calcium current in hippocampal neurons ( Moosmang et al . , 2005 ) . As CaV1 . 3S channels activate at low voltages , they are a particular good candidate for underlying the sustained the neuronal firing and the persistent low- voltage-activated current observed in CA3 neurons ( Avery and Johnston , 1996 ) . In fact , CaV1 . 3 channels support spontaneous firing of dopaminergic neurons in the substantia nigra and mid-spiny striatal neurons ( Guzman et al . , 2009; Olson et al . , 2005 ) . The function of CaV1 . 3 channels is tightly regulated by changes in intracellular Ca2+ ( [Ca2+]i ) . The opening of CaV1 . 3 channels causes a local increase in [Ca2+]i that induces two opposing regulatory mechanisms: Ca2+-dependent inactivation ( CDI ) and Ca2+-dependent facilitation ( CDF ) ( Ben-Johny and Yue , 2014 ) . CDF manifests as an increase in the magnitude of the CaV1 . 3 current ( ICa ) with repetitive activation . In neurons , CDF can induce a persistent ICa that increases firing rate and may even lead to self-sustained firing ( Fransen et al . , 2006; Major and Tank , 2004; Sheffield et al . , 2013 ) . It has been proposed that CDF of the CaV1 . 3 channel depends on Ca2+/CaM-dependent kinase II ( CaMKII ) -mediated phosphorylation , as has also been proposed for the closely related CaV1 . 2 channel ( Hudmon et al . , 2005; Xiao et al . , 1994; Yuan and Bers , 1994 ) . This phosphorylation requires the presence of a second protein , densin , which binds to the PDZ domain located in the most distal part of the C-terminus of the channel ( Jenkins et al . , 2010 ) . Because CaV1 . 3S lacks that PDZ domain , CaMKII-mediated phosphorylation is unlikely to be responsible for CDF in CaV1 . 3S channels . Thus , the mechanisms underlying the CDF of the widely expressed CaV1 . 3S channel have not yet been resolved . Two recent studies by Dixon et al . ( Dixon et al . , 2012; 2015 ) have suggested the tantalizing hypothesis that Ca2+-induced interactions between the C-termini of neighboring CaV1 . 2 channels facilitates Ca2+ influx by increasing the activity of adjoined channels in cardiac muscle . At present , however , whether this physical and functional coupling of CaV1 . 2 channels is a common mechanism for the control of Ca2+ influx via voltage-gated Ca2+ channel function , including CaV1 . 3 channels is unknown . Furthermore , the possibility that cooperative CaV1 . 3 channel gating regulates neuronal excitability is also unclear . In the present study , using electrophysiological , optogenetic , and super-resolution imaging approaches , we discovered that CaV1 . 3S channels form functional clusters of two or more channels along the surface membrane of hippocampal neurons . Clustered CaV1 . 3S channels undergo Ca2+-induced physical interactions that increase the activity of adjoined channels , facilitate Ca2+ currents and thereby increase firing rates in hippocampal neurons . We propose that cooperative gating of CaV1 . 3S channels is a new general mechanism for the amplification of Ca2+ signals in excitable cells . Because CaV1 . 3 channels are alternatively spliced , we first sought to determine whether CaV1 . 3S and CaV1 . 3L channels are differentially regulated by [Ca2+]i . Macroscopic currents were recorded from tsA-201 cells expressing either CaV1 . 3S or CaV1 . 3L channels in the presence of 2 mM Ba2+ or 2 mM Ca2+ . Currents were activated by a depolarizing pulse ( 300 ms ) from a holding potential of -80 mV to -10 mV . With Ba2+ in the external solution , membrane depolarization induced large CaV1 . 3L currents that inactivated slowly ( Figure 1A , center ) . Switching to a perfusion solution containing Ca2+ decreased the amplitude of CaV1 . 3L currents by nearly 40% and increased the rate of inactivation . Like CaV1 . 3L currents , CaV1 . 3S currents inactivated faster when Ca2+ was used as a charge carrier however , in agreement with previous reports ( Bock et al . , 2011; Singh et al . , 2008 ) , we observed more pronounced CDI ( defined as the difference between inactivation of IBa and ICa ) in CaV1 . 3S channels compared to the CaV1 . 3L variant ( Figure 1B , right versus Figure 1A , right ) . As discussed above , this difference in the magnitude of CDI has been attributed to the lack of the CTM domain in CaV1 . 3S channels . Curiously , the amplitude of CaV1 . 3S currents decreased to a lesser extent ( only about 15% at -10 mV ) upon changing the external solution from Ba2+ to Ca2+ ( Figure 1B , center ) . We investigated whether differences in the amplitude of elementary CaV1 . 3L and CaV1 . 3S channel currents could , at least in part , account for these disparities in macroscopic Ca2+ and Ba2+ currents . Single CaV1 . 3L and CaV1 . 3S channel currents were recorded from cell-attached patches with pipettes containing 20 mM Ca2+ or Ba2+ . With Ca2+ as the charge carrier , the amplitudes of elementary CaV1 . 3S and CaV1 . 3L channel currents were similar . For example , at -30 mV they were -0 . 48 ± 0 . 07 and -0 . 49 ± 0 . 01 pA for CaV1 . 3L and CaV1 . 3S channels , respectively ( Figures 1C–F ) . These values are in accordance with the unitary currents reported by Guia et al . for cardiac L-type channels using Ca2+ as charge carrier ( Guia et al . , 2001 ) . Ensemble averages revealed currents that activated quickly and then inactivated likely due to Ca2+ and voltage-dependent mechanisms ( Figure 1G ) . Furthermore , as is the case for other CaV1 channels , the single channel currents produced during the opening of CaV1 . 3L and CaV1 . 3S channels were both larger with Ba2+ as the charge carrier than with Ca2+ ( -1 . 14 ± 0 . 02 pA and -1 . 10 ± 0 . 02 pA for CaV1 . 3L and CaV1 . 3S at -30 mV , respectively ) , but not significantly different from each other ( Figure 1—figure supplement 1 ) . The amplitude of the unitary currents with Ba2+ is also in accordance with previously reported values for these channels ( Bock et al . , 2011 ) . These data suggest that the difference observed in the macroscopic currents between the CaV1 . 3L and CaV1 . 3S channels is not due to differences in unitary currents . We then tested the hypothesis that Ca2+ enhances the activation of CaV1 . 3S , but not CaV1 . 3L channels by increasing the channel activity ( NPo ) . We performed a quantitative analysis of NPo . The whole-cell current ( I ) is given by the equation I = iNPo , where i is the amplitude of the elementary current , N is the number of functional channels , and Po is the channel open probability . Since elementary CaV1 . 3L and CaV1 . 3S currents are larger when Ba2+ rather than Ca2+ is the charge carrier , for ICa to be similar to IBa , the NPo of CaV1 . 3S must be higher in the presence of Ca2+ than Ba2+ . We estimated NPo with Ca2+ and Ba2+ as charge carriers by dividing the amplitude of whole-cell CaV1 . 3S and CaV1 . 3L currents at -10 mV , by the values of unitary current . Our analysis relies on the assumption that the relatively larger peak of CaV1 . 3S currents in the presence of Ca2+ was not due to faster activation kinetic of this channel than that of CaV1 . 3L channels . This assumption is reasonable , as we found no significant difference in the activation time constants for CaV1 . 3L and CaV1 . 3S currents that were 1 . 60 ± 0 . 22 and 1 . 17 ± 0 . 051 ms ( -10 mV , n = 6 for each channel , p = 0 . 112 ) , respectively . We used elementary currents values recorded with 2 mM Ca2+ ( -0 . 16 pA ) and Ba2+ ( -0 . 24 pA ) at -10 mV ( Guia et al . , 2001 ) . We found that Ca2+ ions entering the cell through the channels increased the NPo nearly 1 . 5-fold for CaV1 . 3S channels , but not at all for CaV1 . 3L channels ( Figures 1I and J ) . Thus , assuming that the number of functional CaV1 . 3 channels ( N ) in the membrane remained constant , a reasonable assumption given the short time lapse ( ~2 min ) between recording IBa and ICa from the same cell , these data suggest that a Ca2+-dependent mechanism enhances inward Ca2+ currents by increasing the Po of CaV1 . 3S channels , but not that of CaV1 . 3L channels . One possible explanation for the Ca2+-influx–dependent increase in the Po of CaV1 . 3S channels is cooperative gating among channels in small clusters , as we have previously reported for CaV1 . 2 channels in cardiomyocytes and smooth muscle cells ( Navedo et al . , 2010 ) . To test this possibility , we made optical recordings of individual CaV1 . 3S-mediated Ca2+ influx events ( called 'sparklets' ) in CaV1 . 3L and CaV1 . 3S-expressing tsA-201 cells loaded with 200 µM Rhod-2 using total internal reflection fluorescence ( TIRF ) microscopy . CaV1 . 3S sparklets were recorded at a membrane potential of -80 mV in the presence of 20 mM external Ca2+; 10 mM EGTA was included in the patch pipette to confine the [Ca2+]i signal to within ~1 µm of the point of Ca2+ entry ( Zenisek et al . , 2003 ) . A quantal analysis of CaV1 . 3S and CaV1 . 3L sparklets revealed the presence of single-level ( elementary ) events with a mean amplitude of ~40 nM for both channels , in agreement with our previous study ( Navedo et al . , 2007 ) . Interestingly , consistent with our single channel data , we found that multi-quantal sparklets , which presumably result from the simultaneous opening of several CaV1 . 3 channels , were more commonly observed in cells expressing CaV1 . 3S than CaV1 . 3L channels ( Figure 2A ) . To determine whether channels in a cluster opened cooperatively or independently , we calculated the coupling coefficient ( κ ) among channels within a CaV1 . 3S and CaV1 . 3L sparklet site by applying a coupled Markov-chain model ( Chung and Kennedy , 1996 ) . Channels with κ > 0 . 1 were considered coupled ( Navedo et al . , 2010 ) . Using this approach , we found that the average κ value for CaV1 . 3S and CaV1 . 3L channels was 0 . 21 ± 0 . 05 ( n = 15 ) and 0 . 08 ± 0 . 04 ( n = 12 ) , respectively ( Figure 2B ) . These results support the hypothesis that CaV1 . 3S channels are more likely to undergo cooperative gating , generating persistent and greater Ca2+ influx than CaV1 . 3L channels . 10 . 7554/eLife . 15744 . 005Figure 2 . CaV1 . 3S but not CaV1 . 3L channels gate cooperatively to increase Ca2+ influx . ( A ) TIRF images of spontaneous CaV1 . 3S ( top ) and CaV1 . 3L sparklets ( bottom ) at a holding potential of -80 mV in tsA-201 cells expressing the respective channels . Traces at the right show the time course of [Ca2+]i in the sites marked by the green circles . The dotted red lines show the amplitudes of 1 to 3 quantal levels . The coupling coefficient ( κ ) is shown above each trace . ( B ) Bar chart showing the coupling coefficient for the CaV1 . 3S and CaV1 . 3L sparklets sites . Bars are averages of 5 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 005 If the signal for cooperative CaV1 . 3S channel gating is a local increase in [Ca2+]i , these channels must be in close proximity to one another . To test this hypothesis , we examined the spatial organization of endogenous CaV1 . 3 channels in hippocampal neurons using super-resolution localization microscopy ( Figure 3A–D ) . Hippocampal neurons were immunostained against CaV1 . 3 using an antibody kindly provided by Dr . William Catterall and Dr . Ruth Westenbroek . This analysis showed that CaV1 . 3 channels form clusters occupying an average area of 3660 ± 80 nm2 ( n = 5 ) . The antibody used in this study has been shown to not recognize the corresponding sequence of the closely related CaV1 . 2 α subunit in both , transfected cells and hippocampal tissue ( Hell et al . , 1993; 2013 ) . However , we were not able to test the specificity of the antibody on CaV1 . 3-KO neurons and thus , pursued our analyses of channel clustering using a heterologous expression of CaV1 . 3 channels in tsA-201 cells . 10 . 7554/eLife . 15744 . 006Figure 3 . CaV1 . 3 channels assemble into clusters in the plasma membrane of cultured hippocampal neurons . ( A ) Wide-field image of a representative cultured hippocampal neuron immunostained for CaV1 . 3 channels ( red ) and the neuronal marker microtubule-associated protein 2 ( MAP2; green ) . ( B ) Super-resolution ( GSD ) image of CaV1 . 3 channels in the outlined region in ( A ) . ( C ) Comparison of conventional ( TIRF , left ) and super-resolution ( GSD , right ) images of CaV1 . 3 clusters in zones i and ii outlined in ( B ) . ( D ) Frequency distribution of the area of CaV1 . 3 channel clusters ( n = 5320 clusters from 5 cells ) . ( E ) TIRF image of CaV1 . 3S-mGFP channels expressed in cultured hippocampal neurons ( left ) . Examples of sequential photobleaching steps for three different clusters ( right ) . ( F ) Frequency distribution of CaV1 . 3S cluster bleaching steps ( n = 1105 clusters from 18 cells ) . Clustering of CaV1 . 3S and CaV1 . 3L channels was tested in tsA-201 cells expressing the respective isoform ( See also Figure 3—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 00610 . 7554/eLife . 15744 . 007Figure 3—figure supplement 1 . CaV1 . 3S and CaV1 . 3L channels form clusters in tsA-201 cells . ( A ) Untransfected tsA-201 cells immunostained against CaV1 . 3 channels . Bright field ( left ) , TIRF ( center ) and super-resolution ( GSD , right ) images , outline depicts the periphery of the cell . ( B ) TIRF ( top ) and GSD ( bottom ) of the immunostaining of a representative tsA-201 cell expressing CaV1 . 3S channels . ( C ) Zoom in of the CaV1 . 3S clusters in zones i and ii outlined in ( B ) . ( D ) TIRF ( top ) and GSD images ( bottom ) of the immunostaining of a representative tsA-201 cell expressing CaV1 . 3L ( E ) Zoom in of CaV1 . 3L clusters in zones i and ii in ( D ) . ( F ) Frequency distributions of the area of CaV1 . 3S and CaV1 . 3L channel clusters ( n = 19 , 143 clusters from 7 cells for CaV1 . 3S and 15 , 580 cluster from 5 cells for CaV1 . 3L ) . ( G ) Bar plot of the average cluster area for CaV1 . 3S and CaV1 . 3L channel clusters ( n = 7 and 5 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 00710 . 7554/eLife . 15744 . 008Figure 3—figure supplement 2 . CaV1 . 3S organize in clusters of ~5 channels in tsA-201 cells . ( A ) TIRF image of CaV1 . 3S-mGFP channels expressed in tsA-201 cells ( left ) . Sequential photobleaching steps for three different clusters ( right ) . ( B ) Frequency distribution of CaV1 . 3S cluster bleaching steps ( n = 585 clusters from 11 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 008 The specificity of the antibody in tsA-201 cells was tested by immunostaining untransfected and CaV1 . 3S-transfected cells . No evidence of staining was observed in the untransfected cells ( Figure 3—figure supplement 1A , n = 4 ) . As our antibody cannot distinguish between CaV1 . 3 channels isoforms , we expressed CaV1 . 3L or CaV1 . 3S channels separately and found that both channel subtypes form clusters of similar size ( Figure 3—figure supplement 1B–F ) . The mean areas of CaV1 . 3L and CaV1 . 3S channel clusters were 2543 ± 50 nm2 and 2119 ± 73 nm2 , respectively ( Figure 3—figure supplement 1G , n = 7 ) . GSD images were acquired in the TIRF focal plane with a penetration depth of 130 nm . To determine the number of channels within CaV1 . 3 channel clusters , we used step-photobleaching ( Ulbrich and Isacoff , 2007 ) , of expressed green fluorescent protein ( GFP ) -fused CaV1 . 3S channels in hippocampal neurons ( Figure 3E and F ) and tsA-201 cells ( Figure 3—figure supplement 1G and H ) . The rationale for using only CaV1 . 3S-GFP channels in these studies is twofold . First , CaV1 . 3S and not CaV1 . 3L channels have the ability to undergo coupled activation . Second , the size of CaV1 . 3L and CaV1 . 3S puncta ( at least in tsA-201 cells ) is similar . Thus , CaV1 . 3L and CaV1 . 3S clusters are likely composed of the same number of channels . We identified and excited single CaV1 . 3S channel clusters in hippocampal neurons and tsA-201 cells using TIRF microscopy with a penetration depth of 130 nm . Although some intracellular signal could be collected , the use of TIRF restricts the signal mainly to the plasmalemmal fraction of the channels . After continuous photobleaching we counted the number of step-wise decreases in fluorescence intensity of CaV1 . 3S-GFP clusters . The majority ( 65% ) of CaV1 . 3S clusters underwent at least five stepwise decreases in fluorescence , with the remaining clusters showing fewer steps . A single photobleaching step was observed in only 4% of the CaV1 . 3S clusters . The mean number of CaV1 . 3S channels per cluster determined through sequential photobleaching was 8 ± 1 ( n = 1105 clusters , 18 cells ) in hippocampal neurons , this number is consistent with the average cluster area calculated from our super-resolution data and is similar to that reported in two different studies on the L-type channel distribution in hippocampal neurons using immunogold-labeling with electron microscopy and high-resolution immunofluorescence techniques ( Leitch et al . , 2009; Obermair et al . , 2004 ) . Step-photobleaching in tsA-201 cells revealed an average of 5 ± 1 CaV1 . 3S channels per cluster ( n = 585 clusters , 11 cells ) , consistent also with our super-resolution cluster area measurements ( Figure 3—figure supplement 2 ) . Collectively , these results support our working hypothesis that multi-quantal events detected by imaging Ca2+ influx represent the simultaneous activity of multiple CaV1 . 3S channels in a membrane microdomain . Furthermore , our data suggest that although channel clustering may be necessary for functional coupling of adjacent CaV1 . 3S channels , physical clustering alone is not sufficient to induce functional coupling in CaV1 . 3L channels . To determine whether a physical interaction between CaV1 . 3S channels induces ICafacilitation , we fused the C-terminus of CaV1 . 3S and CaV1 . 3L channels to an optogenetic light-induced dimerization system based on CIBN and CRY2 proteins ( Kennedy et al . , 2010 ) . Blue-light illumination ( 488 nm ) , promotes CIBN and CRY2 fusion , which forces the C-termini of the attached channels to interact ( Figure 4A ) . 10 . 7554/eLife . 15744 . 009Figure 4 . Light-induced fusion increases ICa amplitude in CaV1 . 3S but not in CaV1 . 3L channels . ( A ) Schematic of the blue light-induced dimerization system ( CIBN-CRY2 ) fused to the C-terminal of CaV1 . 3S channels . The same proteins were fused to the C-terminal of CaV1 . 3L channels ( not shown in schematic ) . ( B ) Representative current records from tsA-201 cells expressing CaV1 . 3S-CIBN/CaV1 . 3S-CRY2 ( left ) or CaV1 . 3L-CIBN/CaV1 . 3L-CRY2 ( right ) , before ( black traces ) and after ( red traces ) induction of channel coupling by excitation with a 30 s pulse of 488 nm light . ( C ) Bar plot of the averaged fold-change in ICa following 488 nm excitation in cells expressing CaV1 . 3S-CIBN/CaV1 . 3S-CRY2 ( black ) or CaV1 . 3L-CIBN/CaV1 . 3L-CRY2 . Bars are averages of 5 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 009 We transfected tsA-201 cells with CaV1 . 3S-CIBN and CaV1 . 3S-CRY2 or CaV1 . 3L-CIBN and CaV1 . 3L-CRY2 channels and measured ICa in response to a series of depolarizing voltage steps before and after a 30-s exposure to blue light ( 488 nm ) ( Figure 4B ) . As shown in Figure 4C , in cells expressing CaV1 . 3S-CIBN and CaV1 . 3S-CRY2 channels , ICa amplitude increased by 35% ( n = 6 , p<0 . 001 ) after illumination , whereas in cells expressing CaV1 . 3L-CIBN and CaV1 . 3L-CRY2 channels , there was no change in current amplitude ( 0 . 95 ± 0 . 03 n = 6 ) . These results suggest that fusing adjacent channels at the tip of their C-tail increase the probability of functional coupling between CaV1 . 3S adjoined channels but not between CaV1 . 3L channels . We used a second optogenetic approach that entailed fusing CaV1 . 3S and CaV1 . 3L channels with either the N- ( VN ) or C-terminus ( VC ) of the Venus fluorescent protein ( Kodama and Hu , 2010; Shyu et al . , 2006 ) ( Figure 5A ) . Individual VN and VC fragments are non-fluorescent , but when they come into close proximity , they can reconstitute a full Venus protein , resulting in fluorescence . Venus reconstitution is irreversible and thus , the intensity of the fluorescence signal increases with time , proportionate to the number of CaV1 . 3S/L-VN and CaV1 . 3S/L-VC channels that physically associate . 10 . 7554/eLife . 15744 . 010Figure 5 . Coupling of CaV1 . 3S channels is Ca2+-dependent and increases channel activity . ( A ) Schematic of CaV1 . 3S fused to VN and VC fragments of the Split Venus bimolecular fluorescence complementation system . ( B ) TIRF images of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ in tsA-201 cells expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( top ) . Fluorescence reconstitution was measured in response to 9-s depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ICa currents evoked at the different voltage steps ( bottom ) . ( C ) Voltage-dependence of the normalized conductance ( G/Gmax ) of the ICashown in ( B ) . Dashed curve is the fit to a Boltzmann function . ( D ) Voltage-dependence of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ . The Boltzmann function calculated in ( C ) is superimposed to compare voltage-dependence . ( E ) Bar plot of averaged Venus fluorescence in the presence of 20 mM Ca2+ at -60 mV and +20 mV . Bars are averages ± SEM ( p<0 . 05 , n = 5 cells ) . ( F ) TIRF images of Venus fluorescence reconstitution in the presence of 2 mM Ba2+ in tsA-201 cells expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( top ) . Format and protocol are as in ( B ) IBa currents evoked at the different voltage steps ( bottom ) . ( G ) Voltage dependence of normalized conductance ( G/Gmax ) of the IBashown in ( F ) . Dashed curve is the fit to a Boltzmann function . ( H ) Voltage dependence of Venus fluorescence reconstitution in the presence of 2 mM Ba2+ . The Boltzmann function calculated in ( G ) was superimposed to compare voltage-dependence . ( I ) Bar plot of averaged Venus fluorescence in the presence of 20 mM Ca2+ at -60 mV and +20 mV . Bars are averages of 5 cells ± SEM ( p<0 . 05 ) . Venus reconstitution was also tested in the presence 2 mM Ca2+ ( See Figure 5—figure supplement 1 ) ( J ) Top: TIRF images of CaV1 . 3S sparklets recorded at -80 mV in 20 mM Ca2+ , before depolarization ( left ) , after the same depolarization protocol used in ( B , F ) in the presence of 2 mM Ba2+ ( center ) , and after depolarization in the presence of 20 mM Ca2+ ( right ) . Green circles indicate sparklet sites . Bottom: Traces of the time course of [Ca2+]i in sites 1 and 2 under the three conditions . ( K ) Bar plot of the averaged CaV1 . 3S sparklet activity ( nPs ) before depolarization ( black; average is ~0 ) , after depolarization in Ba2+ ( gray ) , and after depolarization in Ca2+ ( red ) . Bars are averages ± SEM ( p<0 . 05 , n = 5 cells ) . ( L ) Bar plot of sparklet density . Format as in ( K ) . ( M ) Event amplitude histograms of CaV1 . 3S sparklets recorded after depolarization in the presence of Ba2+ ( gray ) or Ca2+ ( red ) . The amplitude of elementary CaV1 . 3 sparklets was calculated by fitting histograms to a multicomponent Gaussian function . The experiments in this figure were performed using the CaV1 . 3S channel encoded by the Addgene plasmid 26576 , similar results for split Venus reconstitution and sparklet activity were observed for the plasmid 49 , 333 ( Figure 5—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 01010 . 7554/eLife . 15744 . 011Figure 5—figure supplement 1 . Depolarization in the presence of physiological Ca2+ concentrations induces coupling in CaV1 . 3S . ( A ) TIRF images of Venus fluorescence reconstitution in the presence of 2 mM Ba2+ in tsA-201 cells expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( top ) . TIRF images of Venus fluorescence reconstitution in the presence of 2 mM Ca2+ in the same cell ( bottom ) . Fluorescence reconstitution was measured as described in Figure 5 . ( B ) Voltage-dependence of CaV1 . 3S-VN and CaV1 . 3S-VC Venus fluorescence reconstitution in the presence of 2 mM Ba2+ followed by 2 mM Ca2+ . ( C ) Bar plot of averaged Venus fluorescence in the presence of 2 mM Ba2+ and 2 mM Ca2+ at -60 mV and +20 mV . Bars are averages ± SEM ( p<0 . 05 , n = 6 cells ) . Bars are averages ± SEM ( p<0 . 05 , n = 6 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 01110 . 7554/eLife . 15744 . 012Figure 5—figure supplement 2 . CaV1 . 3S and CaV1 . 3S ( G244S ) channels exhibit Ca2+-dependent coupling and increased sparklet activity after depolarization . ( A ) Voltage-dependence of the normalized conductance ( G/Gmax ) of the ICa from tsA-201 cells expressing CaV1 . 3S ( Addgene 49333 ) or CaV1 . 3S ( G244S/A1104V ) ( Addgene 26576 ) . ( B ) Plot of the time constants of activation against the depolarization voltage . Current traces were fitted by a single exponential and the corresponding time constants were averaged and plotted in the graph . Data from 6 cells per group . ( C ) TIRF images of Venus fluorescence reconstitution in the presence of 2 mM Ca2+ in tsA-201 cells expressing CaV1 . 3S ( top ) . Fluorescence reconstitution was measured in response to 9-s depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ICa currents evoked at the different voltage steps ( bottom ) . ( D ) Voltage-dependence of Venus fluorescence reconstitution in the presence of 2 mM Ca2+ . ( E ) ( top ) TIRF images of CaV1 . 3S sparklets recorded at -80 mV in 20 mM Ca2+ , before depolarization ( left ) , after the same depolarization protocol used in ( C ) in the presence of 20 mM Ca2+ ( right ) . Green circles indicate sparklet sites . ( Bottom ) Traces of the time course of [Ca2+]i in sites 1 and 2 . ( F ) Bar plot of the averaged CaV1 . 3S sparklet activity ( nPs ) before and after . Bars are averages of 4 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 012 We simultaneously recorded ICa and obtained TIRF microscopic images from tsA-201 cells transfected with CaV1 . 3S -VN and CaV1 . 3S -VC . The first set of experiments was performed in the presence of 20 mM Ca2+ to mimic the experimental conditions used to record Ca2+ sparklets below . With 20 mM Ca2+ in the bathing solution , ICa and Venus fluorescence for CaV1 . 3S channels increased in parallel in response to depolarizing pulses from a holding potential of -80 mV ( Figure 5B ) . The normalized conductance and Venus fluorescence exhibited similar sigmoidal voltage dependencies ( Figures 5C–D ) , which can be attributed to the irreversible nature of the Venus reconstitution , resulting in an increased number of fluorescent proteins with each successive depolarization as the open probability of CaV1 . 3S channels increases . Substituting 2 mM Ba2+ for 20 mM Ca2+ in the external solution led to a robust IBa during membrane depolarization , but caused no accompanying change in Venus fluorescence ( Figures 5F–I ) . Importantly , Venus reconstitution was never observed in cells expressing CaV1 . 3L-VN and CaV1 . 3L-VC in ( see Figure 8 ) . These data indicate that CaV1 . 3S channels , but not CaV1 . 3L channels , can physically interact via their C-termini , that this association occurs in response to membrane depolarization , and that it is promoted by intracellular Ca2+ . Because Venus reconstitution is irreversible , once CaV1 . 3S-VN and CaV1 . 3S-VC channels fuse they must remain adjoined . Thus , we tested the hypothesis that Ca2+-induced fusion of CaV1 . 3S-VN and CaV1 . 3S-VC increases the activity of adjoined channels by recording CaV1 . 3S sparklets at a holding potential of -80 mV , in the presence of Ba2+ or Ca2+ , before and after applying the same depolarization protocol described above . Before depolarization , CaV1 . 3S sparklet activity was very low ( Figure 5J ) ; after depolarization , CaV1 . 3S sparklet activity ( nPs; Figure 5K ) and sparklet density ( Figure 5L ) markedly increased in the presence of Ca2+ , but not Ba2+ , without a change in the amplitude of the elementary Ca2+ influx ( Figure 5M ) . To investigate the mechanism underlying the Ca2+-dependency of CaV1 . 3S channel coupling further , we used the Venus system in conjunction with a mutagenesis approach , focusing on CaM , which is required for CDF and has been shown to bind Ca2+ and associate with the C-terminus of L-type Ca2+ channels ( Zuhlke et al . , 1999 ) . tsA-201 cells were transfected with CaV1 . 3S-VN/CaV1 . 3S-VC and divided into three groups . Cells in the first group ( controls ) were dialyzed with a standard Cs+-based intracellular solution . Cells in the second group were dialyzed with a CaM-inhibitory peptide corresponding to a 15-aa fragment of the wild-type CaM-binding domain of myosin light chain kinase ( MLCKp; 1 μM ) , which binds to CaM with high affinity ( apparent dissociation constant , ~6 pM ) in the presence of Ca2+ ( Torok and Trentham , 1994 ) and has been used by others as a competitive inhibitor of CaM ( Ciampa et al . , 2011; Mercado et al . , 2010; Piper and Large , 2004 ) . The third group consisted of cells co-expressing a dominant-negative mutant form of CaM ( CaM1234 ) that does not bind Ca2+ through its N- or C-terminal lobes . Dialysis of MLCKp or co-expression of CaM1234 prevented CaV1 . 3S-VN and CaV1 . 3S-VC fusion upon membrane depolarization ( Figure 6A–D ) . Although MLCKp and CaM1234 were equally effective in preventing Venus reconstitution , they had differential effects on ICa inactivation ( Figure 6E ) . Whereas the fraction of peak ICa remaining at 300 ms ( r300 ) in MLCKp-dialyzed cells ( 0 . 11 ± 0 . 03 , n = 5 ) was similar to that of controls ( 0 . 14 ± 0 . 03 , n = 5 ) , the rate of inactivation of ICa was slower in cells expressing CaM1234 , as reflected in a much higher r300 value ( 0 . 75 ± 0 . 05 , n = 5 ) ( Figure 6F ) . Our interpretation of these findings is that the CaM molecules involved in CDI are distinct from those involved in functional coupling of CaV1 . 3S channels . The results suggest that CaM molecules that mediate coupling could be both attached to the channels or recruited to the C-terminus during depolarization . This could explain why they are accessible to MLCKp blockade , unlike the CaM molecules that mediate CDI , which are tethered to the IQ domain of the channels ( Pitt et al . , 2001 ) . 10 . 7554/eLife . 15744 . 013Figure 6 . CaV1 . 3S coupling requires Ca2+-CaM . ( A–C ) TIRF images of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ in tsA-201 cells expressing ( A ) CaV1 . 3S-VN and CaV1 . 3S-VC , ( B ) CaV1 . 3S-VN and CaV1 . 3S-VC and dialyzed with the MLCK peptide ( MLCKp ) , ( C ) CaV1 . 3S-VN , CaV1 . 3S-VC and CaM1234 . Fluorescence reconstitution was measured in response to depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ( D ) Voltage-dependence of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ for control ( black ) , MLCKp ( blue ) , and CaM1234 ( red ) cells shown in ( A–C ) ( left ) . Bar plot of averaged Venus fluorescence in the presence of 20 mM Ca2+ at -60 mV and +20 mV ( right ) . Bars are averages ± SEM ( p<0 . 05 , n = 5 cells ) . ( E ) Normalized ICa currents evoked by a 300-ms depolarizing pulse from a holding potential of -80 mV to a test potential of 0 mV in control ( black ) , MLCKp ( blue ) , and CaM1234 ( red ) cells . Currents analyzed for these experiments were in a range between 0 . 3 and 1 . 2 nA ( F ) Bar plot of the mean fraction of r300 at 0 mV . Bars are averages ± SEM ( p<0 . 05 , n = 5 cells ) . ( G ) Top: TIRF images of CaV1 . 3S sparklets in tsA-201 cells expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( Control ) . Sparklets were recorded at -80 mV in 20 mM Ca2+ before depolarization ( left ) and after the same depolarization protocol used in ( A–C ) ( right ) . Green circles indicate sparklet sites . Bottom: Traces of the time course of [Ca2+]i in the corresponding sparklet sites 1 and 2 . ( H ) TIRF images and time course of [Ca2+]i of CaV1 . 3S sparklets in tsA-201 cells expressing CaV1 . 3S-VN/CaV1 . 3S-VC and CaM1234 . Format and protocol are as in ( G ) . ( I ) Bar plot of the averaged CaV1 . 3S sparklet activity ( nPs ) before ( gray ) and after ( black ) depolarization . Bars are averages 5 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 013 To extend this analysis , we recorded CaV1 . 3S sparklets in control and CaM1234 cells before and after depolarization to +20 mV ( Figure 6G and H ) . Although multi-quantal CaV1 . 3S sparklets were observed in control cells at rest ( e . g . , Figure 6G , trace 2 left ) , sparklets in CaM1234 cells prior to depolarization were long , single-quantal events ( Figure 6H , traces 1 and 2 , left ) . These long CaV1 . 3S sparklets were likely due to decreased CDI of CaV1 . 3S channels in cells expressing CaM1234 ( Yang et al . , 2014 ) . Importantly , the overall CaV1 . 3S sparklet density was lower in cells expressing CaM1234 than WT , suggesting that Apo-CaM does not increase CaV1 . 3S channel activity . The coupling coefficient for CaV1 . 3S sparklet sites in WT cells was 0 . 07 ± 06 , whereas that in CaM1234 cells was 0 . 02 ± 0 . 02 ( n = 5 ) . Membrane depolarization increased the coupling coefficient of CaV1 . 3S channels within multi-quantal sparklet sites in control cells ( 0 . 18 ± 0 . 03 , n = 5 ) , but not in CaM1234 cells ( 0 . 08 ± 0 . 05 , n = 5 ) . The opposing effects of CaM1234 on CaV1 . 3S sparklets — longer , but decoupled — resulted in CaV1 . 3S sparklet activity before depolarization that was similar in control ( nPS = 0 . 07 ± 0 . 05; p>0 . 05 ) and CaM1234 ( nPS = 0 . 12 ± 0 . 09 ) cells ( Figure 6I ) . However , depolarization increased CaV1 . 3 sparklet activity nearly 7-fold in control cells , but had no effect on sparklet activity in CaM1234 cells . Taken together with the ICa and Venus reconstitution results described above , these data strongly suggest that Ca2+ binding to CaM is required for physical and functional CaV1 . 3S-to-CaV1 . 3S channel coupling . Finally , to establish the molecular mechanism by which CaM might mediate these effects , we mutated different CaM-binding domains of the CaV1 . 3S . L-type channels have two binding sites for CaM in their C-terminus: the IQ domain ( aa K1601-Q1621 ) and the pre-IQ domain ( aa T1545-Q1587 ) ( Fallon et al . , 2009 ) . These sites have different affinities for CaM; whereas CaM is “pre-associated” and binds tightly to the IQ domain , the association of CaM with the pre-IQ domain is seemingly weaker and likely transient . To determine which of these sites is necessary for CaM-mediated CaV1 . 3S coupling , we generated two mutants of the CaV1 . 3S-VN and CaV1 . 3S-VC channels . The first contained a single point mutation I1608E ( CaV1 . 3-I1608E ) that disrupts CaM binding to the IQ domain ( Zuhlke et al . , 1999 ) , and the second contained a triple mutation ( L1569A , V1572A , and W1577E; CaV1 . 3S-AAE ) that prevents CaM binding to the pre-IQ domain and anti-parallel coiled-coil arrangement of the pre-IQ domains ( Fallon et al . , 2009 ) ( Figure 7A ) . CaV1 . 3S-I1608E channels showed a slower rate of inactivation than control and CaV1 . 3S-AAE channels ( Figure 7B ) , consistent with the lack of CDI . Interestingly , CaV1 . 3S-I1608E-VN/VC , but not CaV1 . 3S-AAE VN/VC , which were capable of reconstituting Venus during membrane depolarization ( Figure 7C–G ) . These data suggest that binding of CaM to the pre-IQ domain is critically involved in CaV1 . 3S channel coupling during membrane depolarization and further supports our previous assertion that the CaM pool involved in CDI is distinct from that involved in channel coupling . 10 . 7554/eLife . 15744 . 014Figure 7 . The pre-IQ domain is required for Ca2+-CaM-mediated CaV1 . 3S coupling . ( A ) Schematic of CaV1 . 3S mutations introduced to disrupt CaM binding to the IQ ( I1608E ) or the pre-IQ ( AAE ) domain; the position of the mutated amino acid is shown in the sequence below . ( B ) Normalized ICa currents evoked by a 30-ms depolarizing pulse from a holding potential of -80 mV to a test potential of 0 mV in tsA-201 cells expressing CaV1 . 3S ( Control , black ) , CaV1 . 3S ( I1608E ) ( blue ) , or CaV1 . 3S ( AAE ) ( red ) . Currents analyzed for these experiments were in a range between 100 and 600 pA ( C–E ) TIRF images of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ in tsA-201 cells expressing ( C ) CaV1 . 3S-VN and CaV1 . 3S-VC , ( D ) CaV1 . 3S ( I1608E ) -VN and CaV1 . 3S ( I1608E ) -VC , or ( E ) CaV1 . 3S ( AAE ) -VN and CaV1 . 3S ( AAE ) -VC . Fluorescence reconstitution was measured in response to depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ( F ) Voltage-dependence of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ for control ( black ) , I1608E mutant ( blue ) , and AAE mutant ( red ) from the cells shown in ( C–E ) . ( G ) Bar plot of averaged Venus fluorescence in the presence of 20 mM Ca2+ at -60 mV and +20 mV . Bars are averages of 5 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 014 We investigated one of the possible molecular mechanisms that prevent functional coupling in CaV1 . 3L channels . Singh et al ( 2008 ) showed that a CaV1 . 3L mutant lacking the last 116 amino acids of the C-terminus ( CaV 1 . 3L∆116 ) had voltage-dependencies and kinetics similar to those of CaV1 . 3S channels . The CaV1 . 3L∆116 channels lack the distal regulatory domain ( DCRD ) that binds the proximal regulatory domain ( PCRD ) located downstream to the pre-IQ and IQ domains of the channel ( Figure 8A ) . Interaction of these two regulatory domains interferes with CaM binding to the IQ domain and results in a reduction in CDI ( Bock et al . , 2011; Singh et al . , 2008 ) . If the DCRD interferes also with the binding of CaM to the pre-IQ domain , which we propose is important for channel-to-channel coupling , we would expect that removing the DCRD would allow coupling between CaV1 . 3L channels . Thus , we investigated whether or not CaV1 . 3L∆116 channels are capable of undergoing Ca2+-driven physical interactions . For these experiments , we created CaV1 . 3L∆116 channels fused to the split Venus system . As expected , CDI of ICa was faster in cells expressing the CaV1 . 3L∆116 channels compared to the full length CaV1 . 3L channels ( p<0 . 05; Figure 8B and C ) . 10 . 7554/eLife . 15744 . 015Figure 8 . Distal auto-regulatory domain ( DCRD ) is not responsible for the lack of coupling of CaV1 . 3L channels . ( A ) Schematic of CaV1 . 3L channel splice variant ( left ) , depicting the domains important for Ca2+-mediated regulation: pre-IQ ( green ) , IQ ( blue ) , proximal and distal C-terminal regulatory domains ( PCRD , DCRD , gray . Schematic of the CaV1 . 3LΔC116 channel where the last 116 aa in the C-terminal were removed ( right ) . ( B ) Representative currents of CaV1 . 3L ( black ) and CaV1 . 3L ΔC116 channels ( red ) expressed in tsA-201 cells . Currents were evoked by a 300-ms depolarization from holding potential of -80 mV to a test potential of 0 mV , with 2 mM Ca2+ as the charge carrier . Currents analyzed for these experiments were in a range between 0 . 3 and 1 nA ( C ) Bar plot of the% inactivation after 25 or 300 ms at 0 mV . Bars are averages of 5 cells ± SEM ( p < 0 . 001 ) ( D–F ) TIRF images of Venus fluorescence reconstitution in the presence of 2 mM Ca2+ in tsA-201 cells expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( D ) CaV1 . 3L-VN and CaV1 . 3L-VC ( E ) or CaV1 . 3L ΔC116-VN and CaV1 . 3L ΔC116-VC ( E ) . Fluorescence reconstitution was measured in response to depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ( G ) Bar plot of averaged Venus fluorescence at -60 mV and +20 mV for each of the aforementioned construct pairs . Bars are averages of 5 cells ± SEM ( p<0 . 05 ) . Data for CaV1 . 3L ΔC116-VN and CaV1 . 3L ΔC116-VC Venus reconstitution with 20 mM Ca2+ is presented in Figure 8—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 01510 . 7554/eLife . 15744 . 016Figure 8—figure supplement 1 . High Ca2+ concentration is not enough to induce coupling in CaV1 . 3LΔC116 channels . ( A ) TIRF images of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ in tsA-201 cells expressing CaV1 . 3L ΔC116-VN and CaV1 . 3L ΔC116-VC . Fluorescence reconstitution was measured in response to depolarizing voltage steps from a holding potential of -80 mV to test potentials of -60 mV to +60 mV . ( B ) Voltage-dependence of Venus fluorescence reconstitution in the presence of 20 mM Ca2+ . Points are averages of 5 cells ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 016 We found that , with 2 mM Ca2+ in the bathing solution , cells expressing CaV1 . 3L∆116 channels failed to reconstitute Venus fluorescence , similar to what we observed for the full length CaV1 . 3L channel ( Figure 8D–G ) . Even increasing the extracellular Ca2+ concentration to 20 mM was not enough to induce coupling in the CaV1 . 3L∆116 channels ( Figure 8—figure supplement 1 , n = 5 , p=0 . 245 -60 mV vs 20 mV ) . This result suggests that deletion of DCRD is not sufficient to allow coupling of CaV1 . 3L channels . As CaV1 . 3L∆116 channels still have a C-terminus ( 396 aa ) that is considerably longer than that of CaV1 . 3S channels , it is possible that another domain inside this region might be responsible for occluding the binding of CaM to the pre-IQ domain , and preventing CaV1 . 3L∆116 channel coupling . Differential folding between the short and long C-terminal could be another explanation for the ability of CaV1 . 3S channel-to-channel coupling during depolarization-induced Ca2+ entry . We extended our investigation of the functional consequences of CaV1 . 3S channel coupling to cultured rat hippocampal neurons . We began by recording spontaneous CaV1 . 3 sparklets from these cells . As in tsA-201 cells expressing CaV1 . 3S channels , Ca2+ sparklets in neurons were restricted to specific sites and had multi-quantal amplitudes resulting from the simultaneous opening and closing of multiple channels . The quantal unit of Ca2+ influx was about 40 nM . The coupling coefficient ( κ ) of sparklet sites ranged from 0 . 14 to 0 . 33 . The average κ value was 0 . 23 ± 0 . 04 ( n = 6 ) . Importantly , sparklet site activity was decreased by application of 300 nM of the dihydropyridine antagonist isradipine and completely eliminated when the concentration of the drug was increased to 10 µM ( Figure 9A and B ) . This is consistent with the hypothesis that sparklets in hippocampal neurons were produced by L-type calcium channels . Both CaV1 . 2 and CaV1 . 3 channels are expressed in hippocampal neurons ( Hell et al . , 1993 ) and although it is impossible to distinguish between these two L-type Ca2+ channels using either electrophysiological or pharmacological methods , it has been shown that CaV1 . 3 channels have a reduced sensitivity to dihydropyridines compared to CaV1 . 2 channels ( Lipscombe et al . , 2004; Xu and Lipscombe , 2001 ) . A previous study by Koschak et al found that 100% of CaV1 . 2 channels but only ~60% of CaV1 . 3 channels are inhibited by 300 nM isradipine ( Koschak et al . , 2001 ) , given this , it is reasonable to assume that any sparklet remaining after application of 300 nM isradipine is more likely to be generated by CaV1 . 3channels than CaV1 . 2 . In addition , near by 25% of the L-type current in hippocampal neurons is carried by CaV1 . 3 channels ( Moosmang et al . , 2005 ) , this proportion is in agreement with the remaining sparklet site density we observed after the treatment with 300 nM isradipine ( Figure 9C ) . 10 . 7554/eLife . 15744 . 017Figure 9 . Hippocampal neurons exhibit dihydropyridine-sensitive spontaneous persistent sparklet activity that is increased after depolarization . ( A ) Top: TIRF images of Ca2+sparklets recorded at -80 mV in cultured hippocampal neurons ( 4 div ) under control conditions ( 20 mM [Ca2+]o; left ) , and after exposure to low ( 1 µM; middle ) and high ( 10 µM; right ) concentrations of dihydropyridine ( DHP ) . Green circles indicate sparklet sites . Bottom: Traces of the time course of [Ca2+]i at the site indicated by the white arrow for each condition are shown below the relevant image . Dotted red lines show the amplitudes of 1 to 7 quantal levels . ( B ) Bar chart showing the mean Ca2+ sparklet activity ( nPs ) in control conditions and after exposure to 300 nM , 1 µM or 10 µM concentrations of DHP . ( C ) Bar chart showing sparklet density for each condition described in ( B ) ( p<0 . 05 , n = 6 cells ) . ( D ) Top: Ca2+ sparklets recorded at -80 mV in cultured hippocampal neurons ( 4 div ) , before depolarization ( left ) and after depolarization ( right ) . Green circles indicate sparklet sites . Bottom: Traces of the time course of [Ca2+]i in sites 1 and 2 before and after depolarization . ( E ) Bar plot of the averaged Ca2+ sparklet activity ( nPs ) before ( white ) and after depolarization ( black ) . Bars are averages of 4 cells ± SEM ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 017 As was the case for tsA-201 cells expressing CaV1 . 3S channels , we found that conditioning membrane depolarization to 0 mV increased sparklet activity nearly 3-fold in the hippocampal neurons and induced persistent sparklet activity ( Figure 9D and E ) . The coupling coefficient increased from 0 . 15 ± 0 . 04 to 0 . 43 ± 0 . 12 after membrane depolarization ( n = 8 ) . These results support the hypothesis that L-type channels undergo cooperative gating , generating persistent Ca2+ influx in hippocampal neurons . The persistent L-type channels sparklet activity evoked by membrane depolarization in the presence of Ca2+ bears a striking resemblance to the persistent cationic currents observed in several types of neurons ( Fransen et al . , 2006; Major and Tank , 2004; Moritz et al . , 2007; Powers and Binder , 2001 ) . We found that the somatic and dendritic membranes of neurons expressing both CaV1 . 3S-VN and CaV1 . 3S-VC channels displayed prominent Venus fluorescence ( Figure 10A ) , indicating fusion of CaV1 . 3S-VN and CaV1 . 3S-VC channel pairs and functional Venus reconstitution . Since Venus reconstitution is irreversible and neurons have spontaneous electrical activity , this observation suggests that the C-termini of CaV1 . 3S channels make contact during normal neuronal firing . Although spontaneous self-assembly between Venus subunits might conceivably drive an interaction that would not otherwise occur , the improved Venus system used here has a mutation ( I152L ) that minimizes non-specific interactions ( Kodama and Hu , 2010 ) . 10 . 7554/eLife . 15744 . 018Figure 10 . CaV1 . 3S coupling increases the firing rate of hippocampal neurons . ( A ) Confocal images of two representative cultured hippocampal neurons expressing tRFP as a transfection marker ( red ) and CaV1 . 3S-VC ( left , negative control ) or CaV1 . 3S-VC/CaV1 . 3S-VN ( right ) . Fluorescence of the spontaneously reconstituted Venus is shown in green . The insets show expanded views of the soma and dendritic regions marked by the dashed boxes . Overexpression of these channels does not change the cluster size observed with super-resolution microscopy ( see also Figure 10—figure supplemental 1 ) . ( B ) Representative traces of spontaneous action potentials recorded from neurons expressing CaV1 . 3S-CRY2 and CaV1 . 3S-CIBN before ( left ) and after ( middle ) the induction of fusion with 488 nm light and after subsequent treatment with 10 µM nifedipine ( right ) . ( C ) Bar plot showing the AP frequency ( normalized to the peak frequency ) for each condition . Bars are averages of 4 cells ± SEM ( p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 01810 . 7554/eLife . 15744 . 019Figure 10—figure supplement 1 . CaV1 . 3S overexpression in hippocampal neurons increased the number of CaV1 . 3S channels , but not the cluster size . ( A ) Super-resolution ( GSD ) image of CaV1 . 3 channels in representative hippocampal neurons non-transfected ( left ) or expressing CaV1 . 3S-VN and CaV1 . 3S-VC ( right ) , immunostained against CaV1 . 3 . Insets at the left corner are magnifications of the outlined regions . ( B ) Average cluster area for CaV1 . 3 channels in non-transfected ( white ) and CaV1 . 3S-VN and CaV1 . 3S-VC transfected neurons . Bars are averages ± SEM ( n = 5 cells ) . ( C ) CaV1 . 3 channels cluster density in non-transfected ( white ) and CaV1 . 3S-VN and CaV1 . 3S-VC transfected neurons . Bars are averages of 5 cells ± SEM ( *p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 019 We analyzed super resolution images of hippocampal neurons transfected with CaV1 . 3S-VN and CaV1 . 3S-VC and found that expression of these channels increased the number of CaV1 . 3 channel clusters in the cells , but not the size of the clusters ( Figure 10—figure supplemental 1 ) . These results indicate that the spontaneous fusion of CaV1 . 3S channels was not a consequence of their being over-expressed in the plasma membranes of the hippocampal neurons . A testable prediction of these observations is that CaV1 . 3S channel fusion—forced coupling—should augment the inward ICa and consequently increase neural excitability and firing rate . Figure 10B shows the spontaneous action potentials recorded from a neuron transfected with CaV1 . 3S-CIBN and CaV1 . 3S-CRY2 , before ( right trace ) and after ( center trace ) exposure to 488 nm light . Forced coupling of CaV1 . 3S-CIBN and CaV1 . 3S-CRY2 channel pairs increased the average firing rate by about 40% ( p<0 . 01 , Figure 10C ) . 488 nm illumination produced no effect on firing rate in control hippocampal neurons transfected only with CaV1 . 3S-CIBN channels ( 0 . 90 ± 0 . 04 , N = 4 ) . Adding the L-type calcium channel blocker nifedipine 10 μM to the bathing solution abolished all AP activity after light-induced fusion of CaV1 . 3S-CIBN and CaV1 . 3S-CRY2 ( Figure 10B , left trace and 10C ) . Together , these results highlight the crucial role that CaV1 . 3S channels play in regulating the action potential firing in hippocampal neurons . We have found that CaV1 . 3 channels in the plasma membrane of hippocampal neurons are arranged in clusters containing multiple ( ~8 ) channels . This clustering has important physiological consequences for the short-splice variant of the channel ( CaV1 . 3S ) , enabling proximal channels to engage in cooperative gating , generating more persistent and greater Ca2+ influx . Functional coupling of CaV1 . 3S channels is promoted by intracellular Ca2+ and involves physical interactions via channel C-termini mediated by physical interactions between Ca2+-CaM and the pre-IQ domain . We propose that Ca2+-driven CaV1 . 3S channel coupling constitutes a novel feed-forward mechanism for the activation of these channels during membrane depolarization and provides an apparatus for Ca2+-dependent facilitation of inward L-type Ca2+ currents . Further , these findings challenge a fundamental tenet of the classic Hodgkin-Huxley model of ion channel gating: that voltage-gated channels open and close independently . Our data suggest that CaV1 . 3S channel coupling is critically dependent on intracellular Ca2+ and CaM . During membrane depolarization , CaV1 . 3S channels open . Ca2+ flowing through these channels creates a local increase in [Ca2+]i — a CaV1 . 3 sparklet — near the mouth and C-terminus of the channel where the Ca2+-binding protein CaM resides . Upon binding Ca2+ , CaM associates with the pre-IQ domain of the channels , enabling the formation of CaV1 . 3S-CaV1 . 3S ‘couplets’ . The observation of pre-IQ dimers of the structurally similar CaV1 . 2 channel undergoing coiled-coil interactions in vitro ( Fallon et al . , 2009 ) and functionally couples to neighboring channels ( Dixon et al . , 2015 ) gives credence to this model . An interesting question suggested by this model is which CaM pool is involved in CaV1 . 3S channel coupling ? The observation that CaM1234 and MLCKp prevent CaV1 . 3S coupling would suggest that soluble as well as apo-CaM pre-associated with the channel could be involved in inducing physical channel-to-channel interactions . Because physically coupled CaV1 . 3S channels exhibit higher open probabilities , the overall activity of CaV1 . 3S channels within a cluster would then depend on the number of channels forming dimers or the probability of the formation of higher order oligomers . A schematic summary of our model is presented in Figure 11 . 10 . 7554/eLife . 15744 . 020Figure 11 . Proposed mechanism for Ca2+-induced functional CaV1 . 3S coupling in hippocampal neurons . CaV1 . 3S channels are organized in clusters in the plasma membrane ( SM ) of hippocampal neurons . At hyperpolarized potentials ( e . g . , -80 mV ) , where [Ca2+]i and Po of CaV1 . 3S channels is very low , the number of coupled channels is very low . Membrane depolarization increases the probability of stochastic ( i . e . , uncoupled ) openings of CaV1 . 3S channels . Ca2+ flow through these channels creates a local increase in [Ca2+]i ( yellow gradient ) . This Ca2+ binds to apoCaM , which can be tethered to the pre-IQ domain or soluble in the cytoplasm . Once CaM is activated it promotes channel-channel interaction at the pre-IQ domain . Upon association , the activity of adjoined channels increases , entering into a cooperative gating mode that facilitates Ca2+ influx and underlies the ‘depolarizing drive’ that sustain repetitive firing . DOI: http://dx . doi . org/10 . 7554/eLife . 15744 . 020 It is important to note that Minor and colleagues reported the formation of symmetric dimers of pre-IQ domains bridged by two CaM molecules ( Kim et al . , 2010 ) . However , they failed to obtain any clustering when the channels were expressed in Xenopus oocytes . The cause for this apparent lack of clustering of CaV1 . 2 channels in the frog egg are unclear , but suggest that clustering and functional coupling may be features of CaV1 channels expressed only in mammalian cells , as we have shown here for CaV1 . 3S channels . Our results suggest that while close proximity is necessary , it is certainly not sufficient to allow channel interactions . Super-resolution imaging shows that CaV1 . 3L and CaV1 . 3S channels form clusters of similar size along the surface membrane , but CaV1 . 3L channels , unlike CaV1 . 3S channels , do not undergo physical and functional coupling . Our experimental results with CaV1 . 3L∆116 channels open the question as to whether there is another regulatory domain inside the long C-terminus , capable of blocking the interaction of CaM with the pre-IQ domain . These results also suggest that the mechanism allowing neighboring CaV1 . 3S channels to interact seems to be structural rather than organizational . Although CaV1 . 3 channels are generally classified as high-voltage activated channels , their lower activation range allows them to generate subthreshold depolarizations that support repetitive firing ( Olson et al . , 2005 ) . This is in particular true for the short variants of CaV1 . 3 , which are more voltage-sensitive than CaV1 . 3L channels ( Bock et al . , 2011; Tan et al . , 2011 ) . Our results reveal a new and striking characteristic of CaV1 . 3S channels: they can coordinate their openings through a physical interaction to facilitate Ca2+ entry at low membrane potentials . In addition , the fact that CaV1 . 3 channels are clustered in dendritic spines in these neurons ( Gao et al . , 2006; Jenkins et al . , 2010 ) raises the possibility that their coordinated gating and boosting of Ca2+ entry may have a significant effect on synaptic plasticity . Subthreshold activation of CaV1 . 3 channels is also a key component of pacemaking and oscillatory behavior in neurons in the brain , such as dopaminergic neurons in the substantia nigra , the principal neurons affected in Parkinson’s disease ( Guzman et al . , 2009; Puopolo et al . , 2007 ) . In this context , it is tempting to speculate that alterations in the cooperative gating of CaV1 . 3S might contribute to the Ca2+ excitotoxicity observed in multiple pathological conditions , including Parkinson’s neurodegeneration . Future studies should examine the extent to which the cooperative gating of CaV1 . 3S channels can affect the Ca2+ load in neurons . In summary , our data indicate that Ca2+-driven physical interactions among clustered CaV1 . 3S channels lead to cooperative gating of these channels and the enhanced Ca2+ influx that underlies the self-sustained firing of hippocampal neurons . It is anticipated that future studies may reveal that cooperative CaV1 . 3S channel gating plays an important role in pathological conditions such as Parkinson’s disease , spasticity , and memory loss , as well as in physiological functions as diverse as hearing and the modulation of heart rate , where CaV1 . 3 channels play a key role . Finally , our findings point to a novel , general mechanism for the dynamic modulation of ionic currents through L-type CaV1 . 3S and CaV1 . 2 channels . For example , in SA node cells , Ca2+ entry through CaV1 . 3 and CaV1 . 2 channels is a key feature in the generation of pacemaker activity ( Mangoni et al . , 2003; Platzer et al . , 2000; Striessnig et al . , 2014 ) . In addition , it has been shown in a recent paper that CaV1 . 3 channels play a critical role in controlling pacemaker activity in SAN cells through the activation of RyR and the triggering of local Ca2+ release from the sarcoplasmic reticulum ( SR ) ( Torrente et al . , 2016 ) . The present results suggest a potential new mechanism for Ca2+-dependent control of pacing in these cells . Accordingly , spontaneous junctional SR Ca2+ release events could induce multimerization of nearby CaV1 . 3S , which , once fused , could produce persistent inward Ca2+ currents that increase the number of RyR activated , thereby driving SA node cells closer to the threshold for action potential generation . The pcDNA clones of the rat CaV1 . 3 isoforms were obtained from Addgene . We used two Addgene CaV1 . 3S plasmids#26576 and #49333 ( Xu and Lipscombe , 2001 ) , the first one containing two single point substitutions , a glycine by a serine at position 244 and an alanine by a valine at position 1104 . We found that they encoded channels with similar voltage-dependencies of activation and rate of activation ( Figure 5—figure supplement 2A and B ) . The CaV1 . 3S channels encoded by these plasmids were also capable of undergoing Ca2+-dependent dimerization and functional coupling ( Figure 5—figure supplement 2C–F ) . On the basis of these data , we concluded that the G244S and A1104V substitutions in the plasmid #26576 are functionally silent with respect to voltage dependence , rate of activation and have no effect on the capacity of adjacent CaV1 . 3S channels to undergo allosteric interactions . A prior study by Lieb et al . also reported no contribution of these mutations to the functional properties of CaV1 . 3S channels ( Lieb et al . , 2012 ) , . Both plasmids #26576 and #49333 were used to express and design the functional CaV1 . 3S constructs used in this study . CaV1 . 3L was also obtained from Addgene ( plasmid #49332 , ( Xu and Lipscombe , 2001 ) ) . Auxiliary subunits CaVβ3 and CaVα2δ1 were gift of Dr . Diane Lipscombe’s laboratory; Brown University , Providence , RI ) . The C-terminus of CaV1 . 3S andCaV1 . 3L channels were fused to different proteins depending on the experimental approach . For bimolecular fluorescence complementation , they were fused to either VN or VC fragments of the Venus protein ( Kodama and Hu , 2010 ) ( Dr . Chang-Deng Hu , Addgene plasmids 27097 , 22011 ) ; for photobleaching experiments , they were fused to monomeric GFP ( mGFPA206K ) , amplified from the pCGFP-EU vector ( Kawate and Gouaux , 2006 ) , kindly provided by Dr . Eric Goaux ( Oregon Health and Science University , Portland , OR ) ; and for the light induced cryptochrome system , they were fused with either CRY2 or CIBN ( generous gifts from Dr . Pietro Di Camilli , Yale University , New Haven , CT ) . The CaM1234 plasmid was a generous gift from Dr . Johannes Hell ( University of California , Davis , CA ) . The tsA-201 cell line , used for heterologous expression of the constructs listed above , was maintained in Dulbecco’s modified Eagle medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin antibiotic solution . Cells were transiently transfected using jetPEI transfection reagent ( Polyplus Transfection , New York , NY ) and plated onto 25-mm coverslips ( 0 . 13–0 . 17-mm thick ) . Successfully transfected cells were identified on the basis of turbo red fluorescent protein ( tRFP ) fluorescence . Imaging and electrophysiology experiments were performed within 48 hr of transfection . Hippocampal neurons were prepared from newborn ( P1 ) Sprague-Dawley rats in accordance with University of Washington ( UW ) guidelines . Animals were decapitated and their tissue harvested according to a protocol approved by the UW Institutional Animal Care and Use Committee ( IACUC ) . The hippocampi of six rat pups were dissected and cut into small pieces in cold dissection medium consisting of 12 mM MgSO4 and 0 . 3% bovine serum albumen ( BSA ) in Hank’s balanced salt solution ( HBSS ) . The pieces were incubated for 30 min at 37ºC in dissection medium containing 25 U/ml papain . The digested tissue was washed with warm Neuronal medium consisting of Minimal Essential Medium ( MEM ) supplemented with 10% horse serum , 2% B27 , 25 mM HEPES , 20 mM glucose , 2 mM GlutaMAX , 1 mM sodium pyruvate , and 1% penicillin/streptomycin antibiotic solution . The tissue was then gently homogenized in fresh Neuronal medium using a long Pasteur pipette . Neurons were plated on poly-D-lysine-coated coverslips ( 0 . 2 mg/ml for 2 hr ) at a density of 2 x 105 cells/coverslip . After incubating neurons at 35ºC for 24 hr , unattached cells were removed by replacing the medium with fresh Neuronal medium . Every 5 day , one-third of the medium was replaced with fresh Neuronal medium supplemented with the anti-mitotic agents fluorodeoxyuridine ( 20 μM ) and uridine ( 50 μM ) . After 14 d in culture , rat hippocampal neurons were transfected using 2 . 4 μg DNA , 4 . 8 μl of Lipofectamine LTX and 4 . 8 μl PLUS reagent ( Life Technologies , Grand Island , NY ) in a final volume of 1 ml of a 1:1 mixture of Neuronal medium and Opti-MEM . After 1 hr of incubation , the medium was replaced with fresh Neuronal medium . Experiments were performed within 48 hr of transfection . Successful transfection was corroborated by detection of tRFP . Ca2+ currents were recorded using the whole-cell configuration of the patch-clamp technique in voltage-clamp mode . Currents were sampled at a frequency of 10 kHz and low-pass filtered at 2 kHz using an Axopatch 200B amplifier . During the experiments , tsA-201 cells were superfused with a solution containing 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 140 mM N-methyl-D-glucamine , 1 mM MgCl2 and 2 mM CaCl2 or 2 mM BaCl2 , depending on the experiment . pH was adjusted to 7 . 4 with HCl . For the experiments using 20 mM CaCl2 , the osmolarity was adjusted by decreasing the concentration of NMDG to 113 mM . Borosilicate patch pipettes with resistances of 3–6 MΩ were filled with an internal solution containing 87 mM cesium aspartate ( CsAsp ) , 20 mM CsCl , 1 mM MgCl2 , 10 mM HEPES , 10 mM EGTA and 5 mM MgATP , adjusted to pH 7 . 2 with CsOH . A voltage offset of 10 mV , attributable to the liquid junction potential of these solutions , was corrected offline . Current–voltage relationships were obtained by subjecting cells to a series of 300-ms depolarizing pulses from a holding potential of -80 mV to test potentials ranging from -70 to +50 mV . The voltage dependence of channel activation ( G/Gmax ) was obtained from the resultant currents by converting them to conductances using the equation , G = ICa/ ( test pulse potential – reversal potential of ICa ) ; normalized G/Gmax was plotted as a function of test potential . We found that the reversal potential for CaV1 . 3L ( +51 ± 4 mV ) and CaV1 . 3S ( +54 ± 3 mV ) currents in the presence of 2 mM Ca2+ were not significantly different ( p=0 . 49 ) . The same was true with 2 mM Ba2+ in the bath ( CaV1 . 3L = +60 ± 5 mV; CaV1 . 3S = +65 ± 2 mV; p=0 . 31 ) . Thus , while Erev is similar in CaV1 . 3L and CaV1 . 3S channels with the same permeating ion , the reversal potential of Ca2+ and Ba2+ currents through these channels differ . In our patch clamp experiments in which the external solution was switched from Ba2+ to Ca2+ , 2-min intervals were inserted between the onset of the whole cell configuration to the first pulse and again after switching the external solution , to rule out any effect of the run-up of the ICa ( Tiaho et al . , 1993 ) . Single-channel currents ( iCa ) were recorded from tsA-201 using the cell-attached configuration . Cells were superfused with a high K+ solution to fix the membrane potential at ∼0 mV . The bathing solution had the following composition: 145 mM KCl , 2 mM MgCl2 , 0 . 1 mM CaCl2 , 10 mM HEPES and 10 mM glucose; pH was adjusted to 7 . 3 with KOH . Pipettes were filled with a solution containing 10 mM HEPES and either 20 mM CaCl2 or 20 mM BaCl2; pH was adjusted to 7 . 2 with CsOH . The dihydropyridine agonist BayK-8644 ( 500 nM ) was included in the pipette solution to promote longer channel open times . A voltage-step protocol from a holding potential of −80 mV to a depolarized potential of −30 mV was used to elicit currents . The single-channel event-detection algorithm of pClamp 10 . 2 was used to measure single-channel opening amplitudes . We generated all-points histograms from our cell-attached patch-clamp recordings . These histograms were fit using Prism 5 . 0a software ( GraphPad software Inc . La Jolla , CA ) with a multi-Gaussian function that included a quantal ( q; i . e . , elementary current ) parameter using the following equation:N=∑j=1n aj×e[−iCa−jq22jb] , where N is the number of events , a and b are constants , iCa is the amplitude of the current measured and q is the quantal elementary current of the channel . Spontaneous discharge of cultured hippocampal neurons was recorded in current-clamp mode using the perforated-patch configuration . Neurons were superfused with a solution containing 140 mM NaCl , 5 mM KCl , 10 mM HEPES , 10 mM glucose , 1 mM MgCl2 , 2 mM CaCl2 and 1 mM Na-pyruvate , adjusted to pH 7 . 4 with NaOH . Borosilicate patch pipettes with resistances of 3–6 MΩ were filled with an internal solution containing 5 mM NaCl , 140 mM KCl , 15 mM HEPES and 7 mM MgATP , adjusted to pH 7 . 2 with KOH; 60 μM amphotericin B was added to the solution before starting the recording . Series resistances lower than 30 MΩ were obtained within 5 min of seal formation . The sampling frequency was 10 kHz filtered at 2 kHz . For immunostaining CaV1 . 3 in tsA-201 cells or hippocampal neurons , cells were fixed by incubating in phosphate-buffered saline ( PBS ) containing 3% paraformaldehyde and 0 . 1% glutaraldehyde for 15 min . After washing with PBS , cells were incubated with 50 mM glycine at 4ºC for 10 min ( aldehyde reduction ) , washed again with PBS , and blocked by incubating with 20% SEA BLOCK ( Thermo Scientific ) and 0 . 25% v/v Triton X-100 in PBS ( blocking buffer ) for 1 hr . The cells were incubated overnight at 4ºC with primary antibodies recognizing the residues 809 to 825 located at the intracellular II-III loop of the CaV1 . 3 channel ( DNKVTIDDYQEEAEDKD , rabbit; provided by Drs . William Catterall and Ruth Westenbroek ) and the neuronal marker MAP2 ( mouse; Abcam ) , diluted in blocking buffer to a concentration of 10 μg/ml . Cells were then washed with PBS , incubated for 1 hr with Alexa Fluor 647-conjugated donkey anti-rabbit ( 2 µg/ml; Molecular Probes ) and Alexa Fluor 488-conjugated chicken anti-mouse ( 2 µg/ml; Molecular Probes ) secondary antibodies , and washed again with PBS . Our antibody was designed to bind to the intracellular loop linking the 2nd and 3rd membrane domains of CaV1 . 3 . It cannot distinguish between CaV1 . 3S and CaV1 . 3L . For super-resolution microscopy , coverslips were mounted on microscope slides with a round cavity using MEA-GLOX imaging buffer ( NeoLab Migge Laborbedarf-Vertriebs GmbH , Germany ) and sealed with Twinsil ( Picodent , Germany ) . The imaging buffer contained 10 mM MEA , 0 . 56 mg/ml glucose oxidase , 34 μg/ml catalase , and 10% w/v glucose in TN buffer ( 50 mM Tris-HCl pH 8 , 10 mM NaCl ) . A super resolution ground-state depletion system ( SR-GSD , Leica ) based on stochastic single-molecule localization was used to generate super-resolution images of CaV1 . 3 in hippocampal neurons and tsA-201 cells . The Leica SR-GSD system was equipped with high-power lasers ( 488 nm , 1 . 4 kW/cm2; 532 nm , 2 . 1 kW/cm2; 642 nm , 2 . 1 kW/cm2 ) and an additional 30 mW , 405 nm laser . Images were obtained using a 160× HCX Plan-Apochromat ( NA 1 . 43 ) oil-immersion lens and an EMCCD camera ( iXon3 897; Andor Technology ) . For all experiments , the camera was running in frame-transfer mode at a frame rate of 100 Hz ( 10 ms exposure time ) . Fluorescence was detected through Leica high-power TIRF filter cubes ( 488 HP-T , 532 HP-T , 642 HP-T ) with emission band-pass filters of 505–605 nm , 550–650 nm , and 660–760 nm . Super-resolution localization images of CaV1 . 3 channel distribution were reconstructed using the coordinates of centroids obtained by fitting single-molecule fluorescence signals with a 2D Gaussian function using LASAF software ( Leica ) . A total of 50 , 000–100 , 000 images were used to construct the images . The localization accuracy of the system is limited by the statistical noise of photon counting . Thus , assuming the point-spread functions are Gaussian , the precision of localization is proportional to DLR/√N , where DLR is the diffraction-limited resolution of a fluorophore and N is the average number of detected photons per switching event ( Dempsey et al . , 2011; Folling et al . , 2008 ) . Accordingly , we estimated a lateral localization accuracy of 16 nm for Alexa 647 ( ~1900 detected photons per switching cycle ) . CaV1 . 3 cluster size was determined using binary masks of the images in ImageJ software ( NIH ) . The number of CaV1 . 3S channels in clusters along the surface membrane was estimated using a single-molecule bleaching approach similar to that described by Ulbrich and Isacoff ( 2007 ) . Briefly , TIRF images of tsA-201 or hippocampal neurons expressing CaV1 . 3S channels fused to the monomeric GFP were acquired using our Leica GSD microscope in TIRF mode . Cells were fixed with 4% paraformaldehyde for 10 min prior to the acquisition . Images were acquired using an oil immersion 160x objective ( NA 1 . 43 ) and an Andor iXON EMCCD camera . GFP was excited with a 488 nm laser and image stacks of 2000 frames were acquired at 30 Hz . During analysis , the first 5 images of the stack were averaged and a rolling-ball background subtraction was applied using ImageJ ( NIH ) . This image was then low-pass filtered with a 2 pixel cut-off and high-pass filtered with a 5 pixel cut-off . Thresholding was then applied to identify connected regions of pixels that were above threshold . The ImageJ plugin ‘Time Series Analyzer v2 . 0’ was used to select 4x4 pixel ROIs , centered on the peak pixel in each spot . These ROIs were used to plot Z-axis intensity profiles ( where z is time ) of the entire image stack to manually detect the bleaching steps . The spontaneous interaction between the C-terminus of CaV1 . 3S channels was assessed using the split Venus system ( Shyu et al . , 2006 ) . In this approach , CaV1 . 3S channels were fused to either the VN fragments ( N1–154 ) or the VC fragment ( C155–238 ) of the Venus fluorescent protein . The Venus protein emits fluorescence only when the two fragments are close enough to interact and reconstitute the whole protein , providing a measure of the proximity between the C-terminus of the CaV1 . 3S channels . CaV1 . 3S -VN and/or CaV1 . 3S -VC constructs were expressed at a 1:1 ratio in hippocampal neurons; tRFP fluorescence was used as an indicator of successful transfection . Confocal images were acquired with a Fluoview FV1000 microscope ( Olympus , Center Valley , PA ) equipped with a UPlanS-Apochromat 60× ( NA 1 . 2 ) water-immersion objective . The Venus protein was excited using a 488 nm laser line . The calcium dependence of CaV1 . 3 spontaneous interactions was studied in tsA-201 cells expressing CaV1 . 3S -VC alone or CaV1 . 3S -VN and CaV1 . 3S -VC in a 1:1 ratio . Using the whole-cell configuration of the patch-clamp technique , cells were depolarized from a holding voltage of -80 mV to test potentials ranging from -60 to +60 mV , administered as 9-s pulses . Maturation of newly reconstituted Venus protein takes some time , hence the long depolarizing pulse ( Nagai et al . , 2002 ) . Images were acquired at a frequency of 100 Hz during each depolarizing pulse using a through-the-lens TIRF microscope built around an inverted microscope ( IX-70; Olympus ) equipped with a Plan-Apochromat ( 60×; NA 1 . 49 ) oil-immersion lens ( Olympus ) and an electron-multiplying charge-coupled device ( EMCCD ) camera ( iXON; Andor Technology , UK ) . The last 10 images of each stack were averaged , and total fluorescence was quantified using ImageJ software ( NIH ) . The images were pseudo-colored using the ‘red hot’ lookup table in ImageJ . F0 was calculated by dividing the total fluorescence for each voltage by the initial fluorescence at -80 mV . The change in F/F0 was plotted against voltage and compared to G/Gmax curves constructed as described in the electrophysiology section . The Ca2+ dependence of Venus reconstitution was tested in cells bathed in an external solution containing 20 mM Ca2+ or 2 mM Ba2+ . CaV1 . 3S sparklets were recorded using the TIRF microscopy system described above . [Ca2+]i was monitored by adding the Ca2+ indicator Rhod-2 ( 200 µM ) to the pipette solution and exciting with a 568 nm laser . The much slower Ca2+ buffer EGTA ( 10 mM ) was included with the relatively fast Ca2+ indicator , Rhod-2 , to restrict Ca2+ signals to the vicinity of the Ca2+ entry source . Sparklets were detected in tsA-201 cells expressing CaV1 . 3S channels . The driving force for Ca2+ entry was increased by holding the membrane potential at -80 mV using the whole-cell configuration of the patch-clamp technique . TIRF images were acquired at a frequency of 100 Hz using TILL Image software . Sparklets were detected and analyzed using custom software written in MATLAB ( Source code 1 ) . Fluorescence intensity values were converted to nanomolar units as described previously ( Navedo et al . , 2007 ) . Event amplitude histograms were generated from [Ca2+]i records and fitted with a multicomponent Gaussian function . We determined the activity of sparklets by calculating the nPs of each sparklet site , where n is the number of quantal levels and Ps is the probability that a quantal sparklet event is active . A detailed description of this analysis can be found in Navedo et al . ( Navedo et al . , 2006 ) . In split Venus experiments , sparklets were acquired at -80 mV before and after a depolarizing protocol from -60 to +60 mV with 9-s pulses . Sparklet images were always acquired in a solution containing 20 mM Ca2+ , whereas depolarization protocols were run in 20 mM Ca2+ or 2 mM Ba2+ to compare the effect of Ca2+-dependent CaV1 . 3S dimerization on sparklet activity . The degree of coupling between CaV1 . 3 Ca2+ sparklet sites was assessed by further analyzing sparklet recordings using a binary coupled Markov chain model , as first described by Chung and Kennedy ( 1996 ) and previously employed by our group ( Navedo et al . , 2010; Cheng et al . , 2011; Dixon et al . , 2012 ) . The custom program ( Source code 2 ) , written in the MATLAB language , assigns a coupling-coefficient ( κ ) to each record , where κ can range from 0 ( purely independently gating channels ) to 1 ( fully coupled channels ) . Elementary event amplitudes were set at 38 nM . Light-induced dimerization of CaV1 . 3 channels was accomplished by fusing CaV1 . 3 channels with the optogenetic light-induced dimerization system based on CRY2 and CIB1 proteins of Arabidopsis thaliana ( Kennedy et al . , 2010 ) . Upon blue-light illumination ( 488 nm ) , CRY2 absorbs a photon , causing a conformational change in one of its domains that promotes binding to the N-terminal region of CIB1 ( CIBN ) . CaV1 . 3-CIBN and/or CaV1 . 3-CRY2 constructs were expressed in a 1:1 ratio in hippocampal neurons . Forty-eight hours after transfection , spontaneous action potential firing was recorded in current-clamp mode . One minute after initiating recordings , neurons were exposed to a 30-s blue light pulse to induce dimerization , and the changes in the firing pattern were measured . In tsA-201 cells , Ca2+ currents were recorded before and after light illumination in response to a 20-ms depolarizing pulse at +10 mV from a holding potential of -80 mV . Experiments were performed on a Nikon ( Eclipse TE2000-S ) Swept Field confocal system controlled with Elements software and equipped with a 488 nm laser line and a Plan Apo 60× 1 . 45 N . A . oil-immersion objective . Data were collected from at least five independent experiments in each series . The data included in this paper were normally distributed . Accordingly , parametric statistics were performed and mean ± SEM are used to provide a description of the data set . Student’s t-test was used to test for statistical significance using Prism 5 . 0 a software ( GraphPad software Inc . La Jolla , CA ) . We decided , a priori , that p values <0 . 05 were indicative a statistical significance difference between or among data groups . The number of cells used for each experiment and p values are detailed in each figure legend . Paired t- tests were used to test for statistical significance of paired observations . Comparisons between three or more conditions were made by one-way ANOVA test using Prism software .	The electrical charge inside a cell is different from that outside of the cell . Neurons rely on this difference to send signals via electrical impulses . This process involves ions moving across the neuron’s membrane through proteins called ion channels . Cav1 . 3 channels are ion channels that open when the membrane’s electrical charge changes to allow positively charged calcium ions into the cell . This generates an electrical current that enables neurons in the brain to produce repetitive impulses . Calcium ions entering through a Cav1 . 3 channel can encourage the channel to allow in even more calcium ions . A closely related channel called Cav1 . 2 , which is essential to the activity of heart muscle , behaves in a similar way . Researchers have recently found that Cav1 . 2 channels are arranged in clusters in the membrane and that adjacent channels interact to allow more calcium ions through the channels . This was an unexpected finding because it had long been thought that all ion channels acted independently . Moreno et al . have now used a range of different approaches to investigate the behavior of one form of the Cav1 . 3 channel , called CaV1 . 3S , in human cells and in neurons from rat brains . Initial experiments confirmed that calcium ions stimulated these channels to open in a coordinated way and to allow in more calcium . High-resolution microscopy then revealed that the CaV1 . 3S channels do form clusters in the cell membrane . Moreno et al . went on to demonstrate that this simultaneous opening of CaV1 . 3S channels first requires a protein called calmodulin to bind calcium inside the cell . Next , the calcium-calmodulin complex associates with the parts of the channels that are also inside the cell . Further experiments showed that coupling the Cav1 . 3 channels together allows them to cooperate , and makes them more likely to be open and generate bigger calcium flows , and allowed neurons to send electrical signals more frequently . Future challenges include investigating how the clusters of Cav1 . 3 channels are established and maintained , and determining how the channels’ cooperation plays a role in both healthy and diseased states .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2016	Ca2+ entry into neurons is facilitated by cooperative gating of clustered CaV1.3 channels
Aneuploidy causes severe developmental defects and is a near universal feature of tumor cells . Despite its profound effects , the cellular processes affected by aneuploidy are not well characterized . Here , we examined the consequences of aneuploidy on the proteome of aneuploid budding yeast strains . We show that although protein levels largely scale with gene copy number , subunits of multi-protein complexes are notable exceptions . Posttranslational mechanisms attenuate their expression when their encoding genes are in excess . Our proteomic analyses further revealed a novel aneuploidy-associated protein expression signature characteristic of altered metabolism and redox homeostasis . Indeed aneuploid cells harbor increased levels of reactive oxygen species ( ROS ) . Interestingly , increased protein turnover attenuates ROS levels and this novel aneuploidy-associated signature and improves the fitness of most aneuploid strains . Our results show that aneuploidy causes alterations in metabolism and redox homeostasis . Cells respond to these alterations through both transcriptional and posttranscriptional mechanisms . Aneuploidy , a condition of having a chromosome number that is not an exact multiple of the haploid complement , has detrimental effects on the development of all eukaryotic organisms where it has been studied ( Torres et al . , 2008 ) . In humans , aneuploidy is the major cause of spontaneous abortions and mental retardation , and it is found in most solid tumors and leukemias ( Weaver and Cleveland , 2006; Nagaoka et al . , 2012 ) . To gain insight into the consequences of aneuploidy at the cellular level and its role in tumorigenesis , we studied the effects of gaining an extra chromosome in haploid yeast cells ( henceforth disomes ) . We showed that yeast cells harboring an extra chromosome share a number of phenotypes including impaired proliferation , increased genomic instability , traits indicative of proteotoxic stress and a gene expression signature known as the environmental stress response ( ESR ) , which is associated with slow growth and stress ( Gasch et al . , 2000; Torres et al . , 2007; Sheltzer et al . , 2012 ) . Importantly , these aneuploidy-associated stresses are also present in aneuploid mammalian cells ( Williams et al . , 2008; Stingele et al . , 2012 ) . Based on these findings , we proposed that the aneuploid state has general consequences beyond those conferred by the increased copy number of specific genes . A key feature of the aneuploid condition is its impact on protein homeostasis . Aneuploid yeast cells are prone to aggregation of both endogenous proteins and ectopically expressed hard-to-fold proteins ( Oromendia et al . , 2012 ) . Furthermore , they exhibit increased sensitivity to inhibitors of protein translation , degradation , or folding ( Torres et al . , 2007 ) . Aneuploid mammalian cells are also sensitive to compounds that interfere with protein quality control mechanisms such as chaperone activity or autophagy ( Tang et al . , 2011 ) . These observations suggest that the proteomic imbalances caused by an aneuploid karyotype disrupt protein homeostasis . How do aneuploid cancer cells overcome the detrimental effects of aneuploidy ? We hypothesized that they may harbor mutations that suppress the adverse effects of aneuploidy . We showed that such mutations indeed exist . In a selection , we identified mutations that improve the fitness of aneuploid yeast strains . Among them was a loss-of-function mutation in the gene encoding the deubiquitinating enzyme Ubp6 , which results in enhanced proteasomal degradation ( Hanna et al . , 2006; Torres et al . , 2010 ) . Deletion of UBP6 improved the fitness of several disomic yeast strains under standard growth conditions and attenuated the proteomic changes caused by aneuploidy . Whether deletion of UBP6 improves the fitness of aneuploid yeast strains in general or whether it is restricted to specific aneuploid karyotypes is not known , nor is the mechanism whereby deletion of the UBP6 gene suppresses the proliferation defect associated with aneuploidy . Here we investigate the impact of aneuploidy on the cell's proteome and how Ubp6 dampens the impact of the aneuploid condition . Our studies show that protein abundances largely scale with gene copy number but that ∼20% of proteins encoded by genes present on additional chromosomes are attenuated . The majority of the attenuated proteins are components of multi-subunit complexes . This finding has implications not only for understanding how cells respond to aneuploidy , but also for how protein complexes are formed and maintained in euploid cells . Importantly , our analysis revealed the existence of both transcriptionally and post-transcriptionally mediated protein expression changes indicative of slow growth as well as oxidative and metabolic stress . Deleting UBP6 attenuates the impact of aneuploidy on the proteome and fitness of all aneuploid yeast strains analyzed , highlighting the importance of proteasomal degradation for aneuploidy tolerance . To understand the global consequences of aneuploidy on the proteome , we used stable isotope labeling of amino acids in cell culture ( SILAC ) ( Ong et al . , 2002 ) and liquid chromatography—tandem mass spectrometry ( LC-MS/MS ) to profile protein abundances in 12 different disomic strains ( Disomes I , II , V , VIII , IX , X , XI , XII , XIII , XIV , XV and XVI ) ( Figure 1A , B , ‘Materials and methods’ ) . These experiments revealed quantitative information for ∼70–80% of all verified open reading frames ( ORFs ) in the disomic strains relative to wild-type cells ( Figure 1B , Figure 1—source data 1 ) . A comparison of wild-type/wild-type cells showed a standard deviation ( SD ) of the log2 ratios equal to 0 . 35 ( Figure 1—figure supplement 1 ) . Analysis of the protein abundances encoded by genes on the duplicated chromosomes of all 12 disomic strains demonstrates that on average protein levels increased approximately twofold ( Figure 1B ) . This correlation is apparent when log2 ratios of protein levels of disomic strains relative to wild-type cells are sorted by the chromosomal position of their encoding genes ( Figure 1B ) . 10 . 7554/eLife . 03023 . 003Figure 1 . Proteome quantification of aneuploid yeast strains . ( A ) Schematic of the approach utilized in stable isotope labeling of amino acids in cell culture ( SILAC ) and liquid chromatography—mass spectrometry . ( B ) The plots show the log2 ratio of the relative protein abundance of disomes compared to wild-type cells grown in synthetic medium . Protein levels are shown in the order of the chromosomal location of their encoding genes . Protein levels of duplicated chromosomes are shown in red . ( C ) Schematic of the approach utilized in isobaric tandem mass tag ( TMT ) -based quantitative mass spectrometry . ( D ) The plots show the log2 ratio of the relative protein abundance of disomes compared to wild-type cells grown in rich medium ( YEPD ) . Protein levels are shown in the order of the chromosomal location of their encoding genes . Protein levels of duplicated chromosomes are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00310 . 7554/eLife . 03023 . 004Figure 1—source data 1 . TMT and SILAC data . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00410 . 7554/eLife . 03023 . 005Figure 1—figure supplement 1 . SILAC and TMT mass spectrometry of wild-type vs wild-type cells . ( A ) The plots show the log2 ratio of the relative protein abundance of wild-type/wild-type cells grown in synthetic medium ( upper panel ) . Protein levels are shown in the order of the chromosomal location of their encoding genes . Histogram of the log2 ratios of the protein levels of wild-type/wild-type cells ( lower panel ) . Fit to a normal distribution is shown ( black line ) . ( B ) The plots show the log2 ratio of the relative protein abundance of wild-type/wild-type cells grown in YEPD medium ( upper panel ) . Protein levels are shown in the order of the chromosomal location of their encoding genes . Histogram of the log2 ratios of the protein levels of wild-type/wild-type cells ( lower panel ) . Fit to a normal distribution is shown ( black line ) . ( C ) Chart of conversion of log2 ratios to fold changes . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00510 . 7554/eLife . 03023 . 006Figure 1—figure supplement 2 . Transcriptome and proteome quantification of aneuploid yeast strains . ( A ) Gene expression and protein levels of wild-type and aneuploid strains grown in synthetic medium , ordered by chromosome position . Experiments ( columns ) are ordered by the number of the chromosome that is present in two copies . ( B ) Comparison of the mRNA vs protein levels in aneuploid strains grown synthetic medium . Pairwise comparison show a Pearson correlation coefficient ( r ) = 0 . 48 . ( C ) Gene expression and protein levels of aneuploid strains relative to wild-type cells grown in YEPD medium , ordered by chromosome position . Experiments ( columns ) are ordered by the number of the chromosome that is present in two copies . ( D ) Comparison of the mRNA vs protein levels in aneuploid strains grown YEPD medium . Pairwise comparison show a Pearson correlation coefficient ( r ) = 0 . 49 . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 006 Growth conditions can significantly influence gene expression . Because aneuploidy increases genomic instability including higher rates of chromosome loss , disomic strains are grown in medium that selects for the presence of the duplicated chromosome ( ‘Materials and methods’ ) . In addition , SILAC relies on the use of synthetic medium supplemented with ‘heavy’ or ‘light’ amino acids . To determine whether growth conditions affect the proteome composition of disomic strains , we grew cells in rich medium ( YEPD ) for a small number of generations and utilized isobaric tandem mass tag ( TMT ) -based quantitative mass spectrometry ( ‘Materials and methods’ , Figure 1C ) to assess the proteome of the 12 disomic yeast strains . In total , we obtained quantitative information for ∼65–74% of all verified ORFs in the disomes relative to wild-type-cells ( Figure 1D , Figure 1—source data 1 ) . As seen in the SILAC-based quantifications , analysis of protein levels of genes encoded on the duplicated chromosomes of 12 disomes showed an average increase of ∼twofold ( Figure 1D ) . The log2 ratios of control wild-type/wild-type cells showed low noise and high reproducibility in the data ( SD of log2 ratios = 0 . 2 , Figure 1—figure supplement 1B ) . Importantly , comparison of the changes in gene expression and protein abundances of disomes compared to wild-type cells grown under similar conditions revealed significant correlations between mRNA and protein levels ( Figure 1—figure supplement 2 , Figure 2—source data 1 ) . These results indicate that on average , increases in gene copy number lead to proportional increases in mRNA and protein levels independent of growth conditions . Dosage compensation , where a change in gene dosage does not lead to a corresponding change in protein levels , is common for genes located on sex chromosomes ( Lee and Bartolomei , 2013 ) . Whether dosage compensation also occurs on autosomes and if so , which genes are affected and how it is brought about are critical questions not only to understand the effects of aneuploidy but also to understand how protein homeostasis is maintained in normal cells . Our set of disomic yeast strains , which comprises duplications of 12 of the 16 chromosomes ( corresponding to 73% of the yeast genome ) , allowed us to address this question . We grew the 12 disomic strains in rich medium , split the cultures and analyzed mRNA and protein levels . In total , we obtained quantitative information for both mRNA and protein , reported as log2 ratios , for 2 , 581 genes located on duplicated chromosomes ( Figure 2A , B ) and 39 , 011 paired measurements for genes on non-duplicated chromosomes ( Figure 2C , D ) . The ratios of mRNA levels of duplicated genes fit a normal distribution with a mean increase of 1 . 9-fold ( SD = 0 . 3 and R2 = 0 . 99 , Figure 2A ) . Parallel analysis of the corresponding protein changes did not fit as well to a normal distribution ( R2 = 0 . 96 , Pearson's mode skewness = −0 . 12 ) . Nonlinear regression analysis of the protein data best fit a sum of two normal distributions; one with a mean increase of twofold , the other with a significantly reduced mean increase of ∼1 . 6-fold ( R2 = 1 . 00 , Figure 2B ) . In contrast , analysis of both mRNA and protein changes of non-duplicated genes showed nearly perfect normal distributions ( R2 = 0 . 99 , Figure 2C , D ) . These analyses indicate that although acquisition of an extra chromosome leads on average to twofold increases in mRNA levels of the duplicated genes , a large and statistically significant number of proteins do not increase proportionally with copy number . Importantly , neither the growth conditions nor the quantitative approach affected the degree of attenuation , as analysis of mRNA and protein levels from cells grown in selective medium and analyzed by SILAC showed similar results ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 03023 . 007Figure 2 . Attenuation of proteins encoded on duplicated chromosomes . ( A ) Histogram of the log2 ratios of the relative mRNA levels of duplicated genes of 12 disomes relative to wild-type grown in YEPD medium . Fit to a normal distribution is shown ( black line ) . ( B ) Histogram of the log2 ratios of the relative protein levels of duplicated genes of 12 disomes relative to wild-type . Fit to the sum of two normal distributions is shown ( black line ) . Fit of individual distributions are shown in dashed-line . ( C ) Histogram of the log2 ratios of the relative mRNA levels of non-duplicated genes of 12 disomes relative to wild-type grown in YEPD medium . Fit to a normal distribution is shown ( black line ) . ( D ) Histogram of the log2 ratios of the relative protein levels of non-duplicated genes of 12 disomes relative to wild-type grown in YEPD medium . Fit to a normal distribution is shown ( black line ) . ( E ) Gene Ontology enrichment analysis of 550 proteins encoded on duplicated genes that are significantly attenuated ( log2 ratios ≤ 0 . 4 ) . ( F ) Pie chart representation of the relative number of all proteins predicted to form part of complexes in the yeast genome is shown in gray ( 33% ) . Pie chart representation of the relative number of proteins that are significantly attenuated and are part of macromolecular complexes in every disome are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00710 . 7554/eLife . 03023 . 008Figure 2—source data 1 . Gene expression data . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00810 . 7554/eLife . 03023 . 009Figure 2—source data 2 . GO enrichment analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 00910 . 7554/eLife . 03023 . 010Figure 2—figure supplement 1 . Attenuation of proteins encoded on duplicated chromosomes . ( A ) Histogram of the log2 ratios of the relative mRNA levels of duplicated genes of 12 disomes relative to wild-type grown in synthetic medium . Fit to a normal distribution is shown ( black line ) . ( B ) Histogram of the log2 ratios of the relative protein levels of duplicated genes of 12 disomes relative to wild-type grown in synthetic medium . Fit to the sum of two normal distributions is shown ( black line ) . Fit of individual distributions are shown in dashed-line . ( C ) Histogram of the log2 ratios of the relative mRNA levels of non-duplicated genes of 12 disomes relative to wild-type grown in grown in synthetic medium . Fit to a normal distribution is shown ( black line ) . ( D ) Histogram of the log2 ratios of the relative protein levels of non-duplicated genes of 12 disomes relative to wild-type grown in synthetic medium . Fit to a normal distribution is shown ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01010 . 7554/eLife . 03023 . 011Figure 2—figure supplement 2 . Gene ontology analysis of attenuated proteins . ( A ) Gene Ontology enrichment analysis of 486 proteins encoded on duplicated genes that are significantly attenuated ( log2 ratios ≤ 0 . 4 ) . ( B ) Overlap between attenuated protein in disomes grown in YEPD vs synthetic medium . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 011 To characterize which genes are subject to dosage compensation , we performed a gene ontology ( GO ) analysis . Using a stringent cutoff of log2 ratio of 0 . 6 ( 3SD ) lower than the expected value of 1 . 0 , we identified a total of 550 proteins encoded by the duplicated chromosomes that were significantly attenuated in disomic strains grown in rich medium ( ∼21% of detected ORFs ) . Gene ontology analysis revealed that components of macromolecular complexes were significantly enriched ( 369 of 550 , p value = 5 . 1 E−32 ) , including all ribosomal subunits detected ( 113 of 550 , p value = 6 . 2 E−43 ) ( Figure 2E , Figure 2—source data 2 ) . Furthermore , enrichment of subunits of macromolecular complexes among the attenuated proteins was observed for every disome analyzed ( Figure 2F ) . Analysis of the identity of the attenuated proteins in disomes grown in selective medium and analyzed by SILAC showed similar results; a significant number of proteins encoded by genes located on the duplicated chromosomes were attenuated in every disomic strain and components of macromolecular complexes were significantly enriched ( 295 of 486 , p value = 2 . 7 E−17 , Figure 2—figure supplement 2A , Figure 2—source data 2 ) . Importantly , there was a significant overlap between the two experiments; 287 proteins ( 57% ) were significantly attenuated in both experiments ( Figure 2—figure supplement 2B ) . Among these proteins , 76% are subunits of macromolecular complexes ( 218 of 287 ) . We previously quantified protein abundances in two disomic strains , disomes V and XIII , relative to wild-type cells ( Torres et al . , 2010 ) . Consistent with our findings presented here , preliminary analyses indicated that subunits of macromolecular complexes were enriched among the dosage compensated genes ( Torres et al . , 2010 ) . A subsequent quantitative proteomic study of five aneuploid yeasts obtained as progeny from triploid or pentaploid meioses found no evidence for the attenuation of proteins that form multi-subunit complexes ( Pavelka et al . , 2010 ) . To better understand this discrepancy , we analyzed the protein measurements generated by Pavelka et al . ( 2010 ) ( Figure 3—figure supplement 1A ) . In two strains , the ratios of protein levels of duplicated genes fit a sum of two normal distributions; one with a mean increase close to twofold , the other with significantly reduced mean close to zero ( R2’s = 0 . 94 and 0 . 92 , Figure 3—figure supplement 1B ) . In the other three strains , the ratios of protein levels of duplicated genes fit normal distributions and show average increases significantly lower than the predicted twofold change ( mean log2 ratios equal to 0 . 79 , 0 . 77 and 0 . 65 ) ( Figure 3—figure supplement 1B ) . Pavelka et al . showed that attenuation of protein levels of subunits of macromolecular complexes were small but not significant compared to proteins not found in complexes . Using the same list of complexes in Pavelka et al . ( 2010 ) ( Gavin et al . , 2006 ) , we obtained similar results ( Figure 3—figure supplement 1C ) . However , when we used a more up to date and manually curated list of subunits of macromolecular complexes ( Pu et al . , 2009 ) , we found that statistically significant attenuation in proteins that form part of complexes takes places in all the strains ( Figure 3—figure supplement 1C ) . Next , we focused on the identity of the most attenuated proteins and asked whether subunits of complexes were enriched among them . We used a stringent cutoff of log2 ratio of 0 . 6 lower than the mean increase in protein levels of the duplicate genes in each of the five aneuploid strains and found that between 23 and 38% of duplicated proteins were significantly attenuated ( Figure 3—figure supplement 1D ) . Importantly , the attenuated proteins are enriched for subunits of macromolecular complexes in all five strains ( Figure 3—figure supplement 1D ) . We conclude that significant attenuation of subunits of macromolecular complexes is a general feature of aneuploid yeast strains . To determine the mechanisms that prevent an increase in protein levels despite increased gene dosage , we compared mRNA and protein levels of the attenuated genes in disomes grown in rich medium . Transcript levels of the attenuated proteins showed increases close to twofold and , unlike their protein products , showed no signs of compensation ( Figure 3A ) . Strikingly , ribosomal genes encoded on duplicated chromosomes showed mean increases close to twofold in their mRNA levels while every ribosomal protein exhibited attenuation ( Figure 3B ) . Similar results were obtained with cells grown in selective medium and analyzed by SILAC ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 03023 . 012Figure 3 . Attenuation takes place posttranslationally . ( A ) Histograms of the log2 ratios of the relative mRNA ( blue ) and protein levels ( red ) of the 550 attenuated proteins from disomic cells grown in YEPD medium compared to wild-type . Fits to a normal distribution are shown ( black lines ) . ( B ) Histograms of the log2 ratios of the relative mRNA ( blue ) and protein levels ( red ) of 88 ribosomal protein genes . Fits to a normal distribution are shown ( black lines ) . ( C ) The plots show the log2 ratio of the relative mRNA levels , mRNA footprints and protein abundance of disomes V and XVI compared to wild-type cells . mRNA levels , mRNA footprints and protein levels are shown in the order of the chromosomal location of their encoding genes . Log2 ratios of the duplicated chromosomes are shown in red . ( D ) Histograms of the log2 ratios of the relative mRNA footprints ( blue ) and protein levels ( red ) of attenuated genes of disomes V and XVI compared to wild-type cells ( top ) . Histograms of the log2 ratios of the relative mRNA footprints ( blue ) and protein levels ( red ) of non-attenuated genes of disomes V and XVI compared to wild-type cells ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01210 . 7554/eLife . 03023 . 013Figure 3—source data 1 . RNA-Seq and ribosome footprints of disome V . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01310 . 7554/eLife . 03023 . 014Figure 3—figure supplement 1 . Analysis of proteome changes of meiotically generated aneuploid strains . ( A ) The plots show the log2 ratio of the relative protein abundance of five meiotically generated aneuploid strains compared to wild-type cells as reported in Pavelka et al . ( 2010 ) . The duplicated chromosomes are indicated . Protein levels are shown in the order of the chromosomal location of their encoding genes . ( B ) Histogram of the log2 ratios of the relative protein levels of duplicated genes relative to wild-type genes from Pavelka et al . ( 2010 ) . In each plot , a fit to a normal distribution or the sum of two normal distributions is shown ( black line ) . Fit of individual distributions are shown in dashed-line . ( C ) Attenuation of proteins encoded on duplicated chromosomes of meiotically generated aneuploid strains . For each of the five meiotically generated aneuploid strains , the mean log2 ratio to wild-type cells for all duplicated genes is plotted separately for proteins annotated as protein complex members and those that are not . The left plot uses the set of Core Complex proteins defined in Gavin et al . ( 2006 ) . The right plot uses a set of manually curated protein complexes from Pu et al . ( 2009 ) . p-values were calculated using Welch's t test . ( D ) Attenuation of duplicated proteins in the meiotically generated set of aneuploid yeast strains . p-values were calculated using Welch's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01410 . 7554/eLife . 03023 . 015Figure 3—figure supplement 2 . Attenuation takes place posttranslationally in cells grown in selective medium . ( A ) Histograms of the log2 ratios of the relative mRNA ( blue ) and protein levels ( red ) of the 486 attenuated proteins from disomic cells grown in synthetic medium compared to wild-type . Fits to a normal distribution are shown ( black lines ) . ( B ) Histograms of the log2 ratios of the relative mRNA ( blue ) and protein levels ( red ) of 83 ribosomal protein genes . Fits to a normal distribution are shown ( black lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 015 To investigate whether translational control mechanisms participate in the attenuation of protein levels , we performed ribosomal footprinting and SILAC-based proteome analysis of disome V and wild-type cells ( Figure 3C , Figure 3—source data 1 ) . In addition , we compared mRNA footprints of disome XVI previously published ( Thorburn et al . , 2013 ) to the proteome quantification of disome XVI grown in similar conditions . In both disomic strains , we found that duplicated genes , both attenuated and not attenuated at the protein level , show similar increases in ribosomal footprints ( Figure 3C , D ) . While we cannot exclude the possibility that translational control may play a role in the attenuation of a small subset of genes , these results indicate that most of the duplicated genes are transcribed and translated . Our results show that dosage compensation is predominantly mediated by posttranslational mechanisms . To test whether protein turnover pathways mediate the attenuation of duplicated genes , we performed TMT-based quantitative proteomics on wild-type cells and two aneuploid strains , disomes II and V , following inhibition of the proteasome and vacuolar degradation by addition of 100 µM MG132 and 10 mM chloroquine , respectively ( note that the strains harbor a deletion in the gene encoding the efflux pump Pdr5 to increase the efficacy of MG132 ) . We hypothesized that after very short times , 90 and 300 s , of protein turnover inhibition , only proteins with increased translation and that are rapidly degraded could show significant increases in abundance . These experiments revealed quantitative information for ∼75% of all verified ORFs ( Figure 4A , Figure 4—source data 1 ) . As expected , very small changes in protein levels were detected in wild-type cells and the two disomes upon protein turnover inhibition ( Figure 4A , Figure 4—figure supplement 1A ) . However , analysis of the average increase in protein levels per chromosome revealed that duplicated genes increased more than the rest of the genome ( Figure 4—figure supplement 1B ) . Analysis of the identity of the duplicated genes revealed that attenuated proteins , which are enriched for subunits of complexes , account for most of the increases in protein levels ( Figure 4B , C ) . Strikingly , individual proteins show increases between 4 to 20% in their levels upon inhibition of protein turnover in such short times ( Figure 4C ) . These results provide direct evidence for protein degradation as being a mechanism for dosage compensation in aneuploid cells . 10 . 7554/eLife . 03023 . 016Figure 4 . Inhibition of protein degradation leads to increases in levels of attenuated proteins . ( A ) The plots show the log2 ratio of the relative protein abundance of disomes compared to wild-type cells and disomes II and V harboring the PDR5 deletion compared to wild-type cells at 0 , 90 and 300 s with 100 µM MG132 and 10 mM chloroquine . Protein levels are shown in the order of the chromosomal location of their encoding genes . Protein levels of duplicated chromosomes are shown in red . ( B ) Average log2 ratios of attenuated and not attenuated products of duplicated genes in disome II ( left ) and disome V ( right ) upon inhibition of protein turnover for 0 , 90 and 300 s ( * denotes p values < 1E-3 ) . ( C ) Examples of duplicated genes that are attenuated in disome II ( left ) and disome V ( right ) that show significant increases upon inhibition of protein degradation . Percent increases are shown below . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01610 . 7554/eLife . 03023 . 017Figure 4—source data 1 . TMT proteome of WT , disome II and disome V after inhibition of protein turnover . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01710 . 7554/eLife . 03023 . 018Figure 4—figure supplement 1 . Inhibition of protein degradation leads to increases in protein levels of duplicated genes . ( A ) Scatter plots of the log2 ratios of protein levels at time 0 vs 300 s after inhibition of protein turnover in disome II ( left ) and disome V ( right ) . ( B ) Average change calculated from the slope of the log2 ratios at 0 , 90 and 300 averaged per chromosome are shown . Pair-wise t test was performed . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 018 To determine which multi-subunit complexes were subject to subunit dosage compensation , we used a manually curated set of yeast protein complexes to assign complex status to all duplicated gene products ( Pu et al . , 2009 ) . We found that more than half of the proteins designated as members of macromolecular complexes were significantly attenuated in the disomes grown in rich medium ( 469 of 923 showed log2 ratios ≤ 0 . 6 ) . Nonlinear regression analysis of the distribution of the log2 ratios of their levels relative to wild-type cells showed two populations , one of which encompassed the majority with a mean ratio of 0 . 46 ( 1 . 4-fold ) , a value significantly lower than the predicted increase of twofold that would be expected if protein levels scaled with gene copy number ( Figure 5A ) . Similar results were obtained with cells analyzed by SILAC ( Figure 5—figure supplement 1A ) . In contrast , proteins encoded by duplicated genes that are not found in complexes showed little attenuation ( Figure 5B , Figure 5—figure supplement 1B ) . Nonetheless , levels of a small number of uncomplexed proteins were attenuated . To assess how these proteins contribute to the total attenuation , we analyzed their identity and the reproducibility of their attenuation in selective and rich media . We identified 88 proteins not known to function in complexes that were attenuated in both growth conditions . Gene ontology analysis did not reveal any significant enrichment for a particular function , cellular process or component . In fact , the biological function of 15 of 88 proteins is unknown ( Figure 5—figure supplement 2A ) . Our analysis not only indicates that most of the proteome attenuation observed in aneuploid cells is caused by the attenuation of components of protein complexes but also leads us to estimate that about half of all cellular proteins found in complexes ( 469 of 923 detected proteins ) are unstable and rapidly degraded unless they find their binding partners . 10 . 7554/eLife . 03023 . 019Figure 5 . Subunits of macromolecular complexes accounts for most of the attenuation . ( A ) Histograms of the log2 ratios of the relative protein levels of 923 duplicated genes found in complexes from disomic cells grown in YEPD medium compared to wild-type . Fits to a sum of two normal distributions are shown ( black lines ) . ( B ) Histograms of the log2 ratios of the relative protein levels of 1 , 658 duplicated genes not part of complexes from disomic cell grown in YEPD medium compared to wild-type . Fits to a sum of two normal distributions are shown ( black lines ) . ( C ) Log2 ratios of subunits of complexes when encoded in a duplicated chromosome relative to wild-type . Complexes that show significant attenuation mean of their subunits < 0 . 6 ( dashed red line ) are shown in red . ( D ) Comparison of the protein levels of subunits of complexes when present in a duplicated chromosome in disomic cells grown in YEPD vs synthetic medium . Pairwise comparison show a Pearson correlation coefficient ( r ) = 0 . 62 . ( E ) Protein levels in wild-type cells or cells harboring a CEN plasmid containing a single copy of RPL1B , RPL3 , RPL30 , ARP5 or CDC28 . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 01910 . 7554/eLife . 03023 . 020Figure 5—source data 1 . List of complexes analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02010 . 7554/eLife . 03023 . 021Figure 5—figure supplement 1 . Subunits of macromolecular complexes accounts for most of the attenuation . ( A ) Histograms of the log2 ratios of the relative protein levels of 977 duplicated genes found in complexes from disomic cells grown in synthetic medium compared to wild-type . Fits to a sum of two normal distributions are shown ( black lines ) . ( B ) Histograms of the log2 ratios of the relative protein levels of 1 , 950 duplicated genes not part of complexes from disomic cell grown in synthetic medium compared to wild-type . Fits to a sum of two normal distributions are shown ( black lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02110 . 7554/eLife . 03023 . 022Figure 5—figure supplement 2 . GO term analysis of attenuated proteins not found in complexes . ( A ) Comparison of the protein levels of attenuated proteins , not known to be part of complexes , when present in a duplicated chromosome in disomic cells grown in YEPD vs synthetic medium . Pairwise comparison show a Pearson correlation coefficient ( r ) = 0 . 45 . ( B ) Log2 ratios of subunits of complexes when encoded in a duplicated chromosome relative to wild-type . Complexes that show significant attenuation ( mean of their subunits < 0 . 6 , dashed red line ) are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 022 To identify which subunits of protein complexes are unstable when present in excess , we pooled quantitative information for subunits for individual complexes and calculated their average increase upon gene duplication . We limited our analysis to complexes for which quantitative information for three or more of their subunits was obtained in both TMT and SILAC datasets ( Figure 5—source data 1 ) . Analysis of 84 complexes is presented here ( Figure 5C , Figure 5—figure supplement 2B , Figure 5—source data 1 ) . Considering subunits that increase in levels by 1 . 5-fold or lower instead of the predicted to twofold ( log2 ratio of 0 . 6 and lower ) , we found that 42 complexes showed attenuation in almost every subunit ( average log2 ratios ≤ 0 . 6 , Figure 5C , Figure 5—figure supplement 2B ) . Not surprisingly , subunits of the ribosome and the nucleosome were among the most attenuated proteins ( Figure 5C , Figure 5—figure supplement 2B ) . The other 42 complexes showed average increases in their subunits higher than log2 ratios of 0 . 6 in one or both proteome datasets ( Figure 5C , Figure 5—figure supplement 2B ) . Strikingly , with the exception of the trehalose-6-phosphate synthase/phosphatase complex ( Reinders et al . , 1997 ) , every complex analyzed showed significant attenuation in at least one of its subunits ( 83 of 84 ) . Interestingly , nearly every complex also contains one or more subunits that are not attenuated , the ribosome and nucleosome being notable exceptions . A few examples include Arp3 of the heptameric Arp2/3 complex ( Robinson et al . , 2001 ) , Mtw1 of the tetrameric MIND complex ( Maskell et al . , 2010 ) , Gim5 of the hexameric prefoldin complex ( Vainberg et al . , 1998 ) , Rpp1 of the nonameric ribonuclease P complex ( Chamberlain et al . , 1998 ) , Prp19 of the octameric Prp19-associated complex ( Chen et al . , 2002 ) , Orc5 of the hexameric origin recognition complex ( Bell and Stillman , 1992 ) , and , Ost1 of the nonameric oligosaccharyltransferase complex ( Spirig et al . , 1997 ) . Examples of complexes with two stable subunits include Ccr4 and Not5 of the CCR4/NOT core complex which contains seven other subunits ( Chen et al . , 2001 ) , Sec21 and Glo3 of the COPI complex which contains six other subunits ( Hosobuchi et al . , 1992 ) , and Vma2 and Vma13 of the proton-transporting ATPase which contains 11 other subunits ( Kawasaki-Nishi et al . , 2001 ) . Other stable subunits include proteins that can be found in more than one complex such as Rpb5 which is part of all three RNA polymerase I , II and III complexes ( Woychik et al . , 1990 ) . While most cellular protein complexes appear to contain subunits that are highly unstable when present in excess , they may also require one or more stable subunits that serve as scaffolds for complex assembly . To assess the reproducibility in attenuation of individual complex subunits , we compared their log2 ratios between cells grown in rich and selective medium , excluding the ribosome and nucleosome . Figure 5D shows the high correlation and reproducibility of the degree of attenuation of such proteins ( Pearson r = 0 . 62 ) indicating that the effects described here are independent of growth conditions and quantification technique . Attenuation of protein levels of duplicated genes could be a result of inherent instability of individual subunits . Alternatively , specific cellular responses to the presence of an extra chromosome could be a contributing factor . For example , the attenuation of ribosome subunit levels could be due to down-regulation of mRNA levels , which occurs as part of the environmental stress response ( ESR ) ( Gasch et al . , 2000 ) . To distinguish between these mechanisms , we analyzed the effects of expressing ribosomal genes from centromeric plasmids on their protein levels . Western blot analysis of cells harboring plasmids with an extra copy of the ribosomal subunits RPL1B , RPL3 or RPL30 showed that protein levels of these genes did not increase with copy number ( Figure 5E ) . In contrast , cells harboring plasmids with an extra copy of ARP5 or CDC28 , which encode proteins that are not attenuated , showed increased levels . Our results indicate that the protein attenuation of ribosomal subunits is at least in part driven by protein instability rather than aneuploidy-induced cellular responses . We previously identified a pattern of transcriptional changes in aneuploid yeast with similarity to the ESR ( Torres et al . , 2007 ) . This change in gene expression leads to a corresponding change in protein levels ( Figure 6—figure supplement 1 ) . The ESR signature was also present in disomic yeast strains grown in rich medium , although with reduced intensity ( Figure 6—figure supplement 1 ) . We hypothesize that this reduced ESR is in part due to smaller differences in proliferation rates between disomic and wild-type cells grown in rich medium compared to selective medium ( Torres et al . , 2007 ) . These results indicate that transcriptional responses to cellular stress and slow proliferation also affect the proteome of aneuploid cells . To investigate whether additional protein responses are shared between aneuploid strains , we performed hierarchical clustering analysis of the protein changes for cells grown in rich medium after reducing the weight of the duplicated gene products ( ‘Materials and methods’ ) . We identified a novel signature of upregulated proteins in all of the disomes compared to wild-type cells ( Figure 6A , Figure 6—source data 1 ) . Here , we refer to this signature as the APS ( aneuploidy-associated protein signature ) . Importantly , this protein signature was not observed in three independent wild-type/wild-type control experiments10 . 7554/eLife . 03023 . 023Figure 6 . Identification of protein signature associated with aneuploidy . ( A ) Hierarchically clustered protein levels from strains grown in YEPD . Proteins encoded on duplicated chromosomes were down-weighted and all data were clustered using the program WCluster . Gene Ontology enrichment analysis of 92 proteins that are significantly upregulated in all 12 disomic strains is shown . We refer to this signature the aneuploidy-specific signature or APS . ( B ) Correlation of the average APS and chromosome size in the disomes . Linear fit is shown in dashed line . ( C ) Comparison of transcript ( left ) and protein levels ( right ) of the APS . Averaged gene ( blue bars ) or protein ( red bars ) expression of the APS of each disomic strain are shown below . Error bars represent SEM . ( D ) Proliferation capabilities of WT , disomes and cells harboring YACs on YEPD medium alone or in the presence of 0 . 75 or 1 mM diamide . ( E ) Relative ROS levels of the disomes grown in YEPD relative to wild-type cell . Error bars represent SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02310 . 7554/eLife . 03023 . 024Figure 6—source data 1 . List of the Aneuploidy-associated protein signature . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02410 . 7554/eLife . 03023 . 025Figure 6—figure supplement 1 . Protein signatures associated with aneuploidy . ( A ) Comparison of transcript ( left ) and protein levels ( right ) of the ESR in disomes relative to wild-type cells grown in synthetic medium . Down and upregulated genes defined as in Gasch et al . ( 2000 ) . ( B ) Comparison of transcript ( left ) and protein levels ( right ) of the ESR of disomes relative to wild-type cells grown in YEPD medium . Down and upregulated genes defined as in Gasch et al . ( 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02510 . 7554/eLife . 03023 . 026Figure 6—figure supplement 2 . APS in the meiotically generated aneuploid strains . ( A ) Comparison of protein levels of the APS in five aneuploidy strains from Pavelka et al . ( 2010 ) . Averaged protein ( red bars ) expression of the APS of each aneuploid strain are shown below . Error bars represent SEM . ( B ) Correlation of the average APS and chromosome size in five aneuploidy strains from Pavelka et al . ( 2010 ) . Linear fit is shown in dashed line . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02610 . 7554/eLife . 03023 . 027Figure 6—figure supplement 3 . Ribosome and proteasome levels in aneuploid cells . ( A ) Averaged protein levels of proteasome subunits of each disomic strain relative to wild-type cells grown in synthetic ( top ) or YEPD ( bottom ) medium . Error bars represent SEM . ( B ) Averaged protein levels of ribosome subunits of each disomic strain relative to wild-type cells grown in synthetic ( top ) or YEPD ( bottom ) medium . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02710 . 7554/eLife . 03023 . 028Figure 6—figure supplement 4 . Proliferation capabilities of aneuploid cells in the presence of 3% H2O2 . ( A ) Proliferation capabilities of WT , disomes and cells harboring YACs on YEPD medium alone or in the presence of 3% H2O2 . ( B ) FACS analysis of cells grown in YEPD medium stained CM-H2DCFDA . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 028 GO enrichment analysis of the APS revealed a group of proteins associated with cellular responses to oxidative stress , including thioredoxins Trx1 and Trx2 , oxidoreductases Grx1 and Grx5 , peroxiredoxins Ahp1 and Prx1 , and the superoxide dismutase Sod1 . In addition , the APS included proteins upregulated during oxidative stress such as the yeast orthologue of the translationally controlled tumor protein ( p23 ) Tma19 , the essential NTPase required for ribosome synthesis Fap7 , the 3′-5′-exodeoxyribonuclease YBL055C , and the polyamine synthases Spe3 and Spe4 ( Gasch et al . , 2000; Juhnke et al . , 2000; Chattopadhyay et al . , 2006 ) . These results indicate that aneuploid cells may be exposed to higher levels of intracellular reactive oxygen species ( ROS ) ( see below ) . Another GO category of APS-enriched genes is ‘metabolic processes’ , including functions such as amino acid biosynthesis and cellular bioenergetics . Interestingly , the intensity of the APS , measured as the average increase of its 92 proteins , correlated with the size of the additional chromosome ( Pearson r = 0 . 62 , Figure 6B ) , indicating that it may be a direct consequence of the cellular imbalances caused by the presence of the extra chromosome , rather than due to increased dosage of specific genes . In addition , we found that the APS is also present in aneuploid strains isolated from random meiosis and that its intensity also correlated with the size of the additional chromosomes in those aneuploid strains ( Figure 6—figure supplement 2 ) . Despite not finding a statistically significant enrichment for cellular processes associated with proteotoxic stress , we found several upregulated proteins involved in protein quality control pathways ( Figure 6—source data 1 ) . These include the Hsp90 regulators Sba1 and Hch1 , the cis-trans peptide isomerases Cpr3 , Fpr1 and Fpr3 , and three proteins involved in ubiquitination including the ubiquitin-conjugating enzyme Ubc1 , the ubiquitin interacting protein Duf1 and the ubiquitin-like protein Rub1 . In addition , APS genes included several proteins involved in protein trafficking including Arc1 , Sec53 , Ric1 , Vti1 and Ykt6 . The upregulation of these proteins is consistent with increases in flux through protein folding , trafficking , and turnover machinery in aneuploid cells and will form the basis for future investigations . In support of a proposed need for increased protein degradation , we found that the average levels of proteasome subunits in all the disomic strains showed a small but significant increase compared to wild-type cells in almost every disomic strain independent of growth conditions ( Figure 6—figure supplement 3 ) . Unexpectedly , the corresponding mRNA transcripts for most of the upregulated proteins were not increased . The average gene expression levels showed minimal changes ( Figure 6C ) indicating that the control of protein upregulation is posttranscriptional . Intriguingly , we did not detect the APS signature in cells grown in synthetic medium . We do not yet understand the reason for this difference but hypothesize that the larger changes in gene and protein expression due to the selective conditions may mask its detection . Our proteome analysis revealed a response to oxidative stress in aneuploid yeast strains . To test whether this was due to defects in redox homeostasis , we compared the viability of wild-type cells and disomes in the presence of diamide or hydrogen peroxide ( H2O2 ) . We found that most disomes show hypersensitivity to the reactive oxygen species ROS-inducing agents diamide ( 1 mM ) or H2O2 ( 3% ) ( Figure 6D , Figure 6—figure supplement 4A ) . To investigate whether the mere presence of chromosome-size amounts of DNA was responsible for hypersensitivity to diamide or H2O2 , we tested the viability of strains harboring a yeast artificial chromosome ( YAC ) varying in size containing human or mouse DNA . Cells harboring such YACs did not exhibit hypersensitivity to the ROS inducing agents ( Figure 6D , Figure 6—figure supplement 4A ) , indicating that the presence of the extra yeast genes and their products is responsible for the increased sensitivity to oxidative stress . To test whether the upregulation of oxidative stress response proteins was due to increased levels of intracellular ROS , we measured ROS levels in the disomes during exponential growth using a fluorescent , ROS-sensitive dye , 5- ( and-6 ) -chloromethyl-2′ , 7′-dichlorodihydrofluorescein diacetate ( CM-H2DCFDA ) ( Figure 6E , Figure 6—figure supplement 4B ) . Basal levels of intracellular ROS were higher in most disomes compared to wild-type cells . ROS levels in cells harboring a YAC with human or mouse DNA did not show such increases ( Figure 6E ) . Our results indicate that aneuploidy disrupts cellular redox homeostasis leading to the accumulation of intracellular ROS . Our data further suggest that aneuploid cells respond to these elevated ROS levels by maintaining higher protein levels of ROS scavengers such as thioredoxins and oxidoreductases . Our previous studies identified loss of function mutations in the deubiquitinating enzyme UBP6 as attenuating the proteomic changes of aneuploidy in two disomic strains ( Torres et al . , 2010 ) . In one strain ( disome V ) but not the other ( disome XIII ) , this attenuation was associated with improved proliferative abilities . The studies described here show that aneuploidy profoundly impacts the proteome of all aneuploid yeast strains . We performed gene expression and proteomic analyses of 12 disomic strains harboring the deletion of UBP6 ( ubp6Δ ) . We measured both mRNA and protein levels for ∼70–80% of all verified open reading frames ( ORFs ) in the disomes-ubp6Δ relative to wild-type cells ( Figure 7A , Figure 7—figure supplement 1A , Figure 7—source data 1 ) . Plots of the log2 ratios sorted by chromosomal position showed a strong correlation between mRNA and protein levels ( Figure 7A , Figure 7—figure supplement 1B ) . While analysis of the log2 ratios of proteins encoded by non-duplicated genes showed a normal distribution ( Figure 7B ) , log2 ratios of proteins encoded by duplicated genes fit a sum of two populations one of which was significantly attenuated ( Figure 7C ) . The mRNA levels of these duplicated genes , however , showed an average increase of ∼twofold with no signs of compensation ( Figure 7—figure supplement 1C ) . Importantly , loss of UBP6 did not further attenuate levels of proteins found to be dosage compensated in the disomic strains ( Pearson r = 0 . 75 , Figure 7D , E , Figure 7—figure supplement 1D ) . These results indicate that the UBP6 deletion does not significantly alter attenuation of subunits of macromolecular complexes . 10 . 7554/eLife . 03023 . 029Figure 7 . Loss of UBP6 function preferentially affects proteins overproduced in disome V and disome XIII cells relative to wild-type . ( A ) The plots show the log2 ratio of the relative protein abundance of disomes harboring the UBP6 deletion compared to wild-type cells grown in YEPD . Protein levels are shown in the order of the chromosomal location of their encoding genes . Protein levels of duplicated chromosomes are shown in red . ( B ) Histogram of the log2 ratios of the relative protein levels of non-duplicated genes of 12 disomes harboring the UBP6 deletion relative to wild-type grown in YEPD medium . Fit to a normal distribution is shown ( black line ) . ( C ) Histogram of the log2 ratios of the relative protein levels of duplicated genes of 12 disomes harboring the ubp6Δ relative to wild-type grown in YEPD medium . Fit to a sum of two normal distributions is shown ( black line ) . ( D ) Histograms of the log2 ratios of the relative protein levels of duplicated genes found in complexes from disomic cells harboring the UBP6 deletion grown in YEPD medium compared to wild-type . Fits to a sum of two normal distributions are shown ( black lines ) . ( E ) Comparison of the protein levels of subunits of complexes when present in a duplicated chromosome in disomic cells vs disomes harboring the UBP6 deletion grown in YEPD . Pairwise comparison show a Pearson correlation coefficient ( r ) = 0 . 75 . ( F ) Average relative levels of the most upregulated proteins , log2 ratios ≥ 0 . 4 , in disomes-UBP6 ( blue ) and disomes-ubp6Δ ( red ) compared to wild-type cells . Pair-wise t test was performed between disomes , * refers to p value = 0 . 01 and *** refers to p value < 1E-4 . ( G ) Average relative levels of the most downregulated proteins , log2 ratios ≤ −0 . 4 , in disomes-UBP6 ( blue ) and disomes-ubp6Δ ( red ) compared to wild-type cells . Pair-wise t test was performed between disomes , * refers to p value = 0 . 01 and *** refers to p value < 1E-4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 02910 . 7554/eLife . 03023 . 030Figure 7—source data 1 . Gene expression and proteome data of disomes-ubp6Δ . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 03010 . 7554/eLife . 03023 . 031Figure 7—figure supplement 1 . Analysis of mRNA and protein changes in disomes upon loss of UBP6 . ( A ) Gene expression and protein levels of aneuploid strains harboring the ubp6Δ relative to wild-type cells grown in YEPD medium , ordered by chromosome position . Experiments ( columns ) are ordered by the number of the chromosome that is present in two copies . ( B ) Pairwise comparison of the mRNA vs protein levels in disomes-ubp6Δ grown in YEPD medium . Pearson correlation coefficient ( r ) = 0 . 35 . ( C ) Histogram of the log2 ratios of the relative mRNA levels of duplicated genes of 12 disomes harboring the ubp6Δ relative to wild-type grown in YEPD medium . Fit to a normal distribution is shown ( black line ) . ( D ) Histogram of the log2 ratios of the relative protein levels of duplicated genes not known to be part of complexes of 12 disomes harboring the ubp6Δ relative to wild-type grown in YEPD medium . Fit to a sum of two normal distributions is shown ( black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 03110 . 7554/eLife . 03023 . 032Figure 7—figure supplement 2 . Protein levels of subunits of complexes and gene expression changes of the most up and downregulated genes in disomes-ubp6Δ . ( A ) Log2 ratios of subunits of complexes when encoded in a duplicated chromosome relative to wild-type . Complexes that show significant attenuation ( mean of their subunits < 0 . 6 ( dashed red line ) are shown in red . ( B ) Average gene expression levels of the most upregulated proteins , log2 ratios ≥ 0 . 4 , in disomes ( blue ) and disomes-ubp6Δ ( red ) compared to wild-type cells . Pairwise t test was performed between disomes , * refers to p value = 0 . 01 and *** refers to p value < 1E-4 . ( C ) Average gene expression levels of the most downregulated proteins , log2 ratios ≤ 0 . 4 , in disomes ( blue ) and disomes-ubp6Δ ( red ) compared to wild-type cells . Pair-wise t test was performed between disomes , * refers to p value = 0 . 01 and *** refers to p value < 1E-4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 032 Next , we extended our analysis to all proteins whose levels were significantly altered in the disomes relative to wild-type cells regardless of their chromosomal origin . To this end , we binned proteins into three categories: upregulated ( log2 ratio ≥ 0 . 4 ) , downregulated ( log2 ratio ≤ −0 . 4 ) , and those that do not significantly change ( −0 . 4 < log2 ratio < 0 . 4 ) . We then compared the average of each category to the average change of the same proteins in the disomes lacking UBP6 . We found that all 12 disomes showed significant attenuation in the levels of their most upregulated proteins upon loss of UBP6 ( Figure 7F ) . Importantly , analysis of the changes in gene expression of this set of proteins showed minimal attenuation of mRNA levels , indicating that the increased attenuation upon loss of UBP6 is mediated posttranscriptionally ( Figure 7—figure supplement 2 ) . All disomes also showed significant increases of downregulated proteins upon the deletion of UBP6 bringing their levels closer to wild-type cells ( Figure 7G ) . The effect on downregulated proteins impacts fewer proteins than the upregulation ( Figure 7F , G ) . These results indicate that protein attenuation upon loss of UBP6 occurs in all aneuploid strains examined . Importantly , it affects both downregulated , and to a greater extent , upregulated proteins . Increased proteasomal degradation due to the loss of UBP6 could be responsible for the downregulation of overexpressed genes . Which proteins are direct targets of Ubp6 remains to be investigated . So far , we found that deletion of UBP6 in wild-type cells did not affect the half-life of six proteins , Fap7 , Glc8 , Bna5 , Tma19 , Trx1 and Trx2 , whose increased abundance in disomic strains is attenuated when UBP6 is deleted ( Figure 8—figure supplement 1 ) . It is thus possible that the ubp6Δ-mediated attenuation of overexpressed proteins in the disomes is indirect . How deletion of UBP6 brings about an increase in the levels of proteins that are down-regulated in aneuploid strains is more difficult to explain and one must invoke indirect effects such as downregulation of negative regulators of gene expression . Consistent with oxidative stress responsive protein levels being upregulated in aneuploid cells and attenuated upon loss of UBP6 , we found that the APS was significantly reduced in the disomic yeast strains lacking UBP6 ( Figure 8A ) . Interestingly , disome V , whose fitness is significantly improved upon deletion of UBP6 , showed the strongest reduction in the APS . In contrast , we found that the ESR was not significantly affected in the disomic strains lacking UBP6 ( Figure 8—figure supplement 2 ) . This is consistent with the fact that only 2 disomes , disome V and XI , show significant improvements in fitness when grown in rich medium ( Torres et al . , 2010 ) . Our results indicate that loss of UBP6 ameliorates protein responses associated with altered redox homeostasis and metabolism . 10 . 7554/eLife . 03023 . 033Figure 8 . Loss of UBP6 attenuates cellular responses to aneuploidy . ( A ) Comparison of transcript ( left ) and protein levels ( right ) of the APS in disomes-ubp6Δ . Averaged gene ( blue bars ) or protein ( red bars ) expression of the APS of each disomic strain are shown below . Error bars represent SEM . For comparison , dashed lines show the corresponding averages in disomes-UBP6 . ( B ) Relative ROS levels of disomes-ubp6Δ grown in YEPD at 30°C . Error bars represent SD ( n = 3 ) . For comparison , dashed lines show the corresponding ROS levels in disomes-UBP6 . ( C ) Doubling times of cells at 37°C . * refers to p-value < 0 . 05 ( t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 03310 . 7554/eLife . 03023 . 034Figure 8—source data 1 . List of strains utilized . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 03410 . 7554/eLife . 03023 . 035Figure 8—figure supplement 1 . Cyclohexidime chases of Ubiquitin , Trx1 , Trx2 , Bna5 , Tma19 , Fap7 and Glc8 in wild-type cells or cells harboring the ubp6Δ . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 03510 . 7554/eLife . 03023 . 036Figure 8—figure supplement 2 . ESR , ribosome and proteasome levels in disomes-ubp6Δ . ( A ) Comparison of transcript ( left ) and protein levels ( right ) of the ESR of disomes-ubp6Δ relative to wild-type cells grown in YEPD medium . Down and upregulated genes are defined as in Gasch et al . ( 2000 ) . ( B ) Averaged protein levels of ribosome ( top ) and proteasome subunits ( bottom ) of each disome-ubp6Δ relative to wild-type cells grown in YEPD medium . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03023 . 036 Amelioration of the APS upon UBP6 loss of function ( Figure 8A ) suggests that the elevated intracellular ROS levels observed in the disomes may also be affected by the deletion . Indeed , loss of Upb6 function resulted in a significant decrease in the basal levels of intracellular ROS in 12 out of 13 disomic strains ( Figure 8B ) . We were not able to test whether deletion of UBP6 suppressed the diamide and H2O2 sensitivity of the disomic strains because deletion of UPB6 itself causes sensitivity to these compounds . However , we were able to assess the effects of deleting UBP6 on the overall fitness of the disomic strains . We previously found that loss of UBP6 improved the fitness of 2 disomic strains when grown in YEPD medium and of 4 disomic strains when grown in selective medium ( Torres et al . , 2010 ) . At high temperature the impact on fitness was more global . Deletion of UBP6 suppressed the proliferation defect of 11 out of 13 disomic strains at 37°C ( Figure 8C ) . Our results indicate that UBP6 loss of function leads to lower intracellular ROS and ameliorates the APS in all disomic strains analyzed . Importantly , this suppression is also associated with an improvement in fitness in most disomic strains , particularly under proteotoxic stress conditions , suggesting that defects in redox homeostasis contribute to the proliferation defect of aneuploid strains . Our studies show that , in general , the acquisition of an extra chromosome translates into proportional increases in protein levels encoded on that chromosome . Despite the mechanisms that exist for dosage compensation of sex chromosomes , eukaryotic cells do not seem to have evolved mechanisms to silence genes upon the acquisition of an extra copy of an autosome . Therefore , a direct consequence of gaining an extra chromosome is increased flux through the transcription and translation machineries . However , the near comprehensive quantitative proteomic assessment of 12 disomic yeast strains described here reveals that levels of a sizeable portion of the proteome ( ∼20% ) do not scale with gene copy number . Most of the attenuated proteins are members of multi-protein complexes . Importantly , we show that attenuation of protein levels is mediated by posttranslational mechanisms , mostly by protein degradation . We hypothesize that these proteins only acquire a stably folded state when they are incorporated into their native complexes . Cells produce stoichiometric amounts of individual subunits of multiprotein complexes ( Li et al . , 2014 ) . Approximately one third of all yeast genes , which are randomly scattered in the genome , encode subunits of macromolecular complexes . Therefore irrespective of which chromosome is duplicated , significant protein attenuation must take place ( Figure 2F ) . For example , the ribosome , which consists of 79 unstable proteins ( encoded by 137 genes ) , is one of the most abundant multi-subunit complexes in the cell . It is estimated that 50% of all RNA Polymerase II transcription is devoted to the production of ribosomal proteins ( Warner , 1999 ) . Except for chromosomes I , III and VI , every yeast chromosome encodes several subunits , ranging from 5 in disome V to 19 in disome IV . Thus even a single excess chromosome leads to a substantial increase in transcription and translation of individual subunits without upregulation of the total number of ribosomes . Our results indicate that the majority of excess subunits , not only of the ribosome but also of most complexes , are destined for degradation . Altogether , our results provide direct evidence that a major consequence of aneuploidy is an increased burden on the protein quality control pathways including protein degradation . Cells have evolved several mechanisms that facilitate complex assembly , such as co-transcriptional regulation and dedicated chaperone systems that help stabilize unstable subunits to prevent their degradation . Our analysis indicates that two of the most stable and long-lived complexes in cells , the ribosome and the nucleosome , consist of subunits that may not exist for long unless assembled into their complex ( elBaradi et al . , 1986; Gunjan and Verreault , 2003; Meeks-Wagner and Hartwell , 1986; Tsay et al . , 1988 ) . Most other complexes show a large range of subunit stabilities . Remarkably , almost every complex analyzed in our study contains at least one attenuated subunit . Conversely , most macromolecular complexes with the exception of the nucleosome and ribosome contain at least one subunit that appears to be stable on its own . The existence of unfolded or partially folded unstable subunits may provide the necessary free energy to drive complex formation ( Kiefhaber et al . , 2012 ) . Our results suggest that a stable scaffold protein may also be required for complex assembly . Deciphering the molecular mechanisms that dictate which subunits are degraded and which are not will significantly contribute to our understanding of the regulation of macromolecular complex formation . Aneuploidy hampers cellular proliferation and consistently elicits gene expression responses associated with slow proliferation and stress . Here , we showed that such gene expression responses affect the proteome content of cells . Paradoxically , one facet of the ESR is the downregulation of ribosomal protein genes leading to lower ribosome protein levels in the disomic strains compared to wild-type cells ( Figure 6—figure supplement 3B ) . This is despite the apparent increase in total translation in cells harboring an extra chromosome . In addition , ribosomal footprinting analyses of disomic strains did not reveal any signs of impairment in translation efficiency . Therefore , the functional consequences of downregulation of ribosomes may be due to slower proliferation rates and may not affect the translational capacity of the cell . Nonetheless , increased translation and downregulation of ribosomal genes may provide the molecular explanation for the increased sensitivities of aneuploid cells to drugs that target the translational machinery . In addition to the ESR-driven protein changes , we identified a novel aneuploidy-specific protein expression signature . This signature is present in all disomes analyzed and consists of 92 upregulated proteins involved in the regulation of redox homeostasis and metabolism . The upregulation of several of these proteins appears to occur in response to higher basal levels of intracellular ROS in the disomes . At present , we do not know the source of elevated intracellular ROS but our results indicate that disruption of protein homeostasis may be the culprit . Increased protein translation , folding and turnover create a high demand for ATP , which leads to the accumulation of ROS ( Gorrini et al . , 2013 ) . In addition , endoplasmic reticulum ( ER ) stress due to increased protein folding could also contribute to ROS accumulation ( Tu and Weissman , 2002 ) . Another not mutually exclusive possibility is that altered metabolism due to upregulation of anabolic processes alters redox homeostasis in aneuploid cells ( Gorrini et al . , 2013 ) . Consistently , several proteins involved in the biosynthesis of amino acids , nucleotides and lipids are upregulated in the disomes ( Figure 6—source data 1 ) . Lastly , our analysis indicates that the average increase in levels of the APS strongly correlates with the size of the extra chromosomes in the disomes , suggesting that this response may be a direct consequence of the acquisition of extra genes . An unexpected finding in our studies is that the APS is not associated with increases in corresponding mRNA levels . Several potential mechanisms could mediate this response . Increased translation could be one reason . However , we found no detectable changes in translational efficiency of the APS genes in disomes V or XVI . We note that a more comprehensive ribosomal footprinting analysis would be necessary to reveal small but significant changes in translational control for a particular gene . It is also possible that the APS is the result of protein stabilization . Consistently , stabilization of proteins following transcriptional downregulation has been observed in cells exposed to mild oxidative stress over the course of several hours ( Vogel et al . , 2011 ) . How stabilization occurs is not yet known but changes in posttranslational modifications such as phosphorylation are certainly one possibility . Most of the proteins of the APS have been shown to be ubiquitinated or phosphorylated ( 76 of 92 , Figure 6—source data 1 ) . Because aneuploidy alters cellular metabolism , posttranslational modifications involving metabolites could also play a role . Interestingly , 34 of 92 APS proteins have been shown to be acetylated and/or succinylated in yeast ( Henriksen et al . , 2012; Weinert et al . , 2013 ) . Increased protein degradation mediated by the loss of function of UBP6 suppresses several phenotypes associated with aneuploidy . Our results indicate that protein attenuation upon loss of UBP6 occurs independently of the identity of the extra chromosome and that it affects both down and upregulated proteins , although the latter to a greater extent . We hypothesize that deletion of UBP6 directly affects protein degradation and/or ameliorates protein responses indirectly by suppressing aneuploidy-associated phenotypes . Another possibility is that loss of UBP6 directly leads to increased degradation of a few transcriptional regulators thereby affecting the levels of both down and upregulated proteins . Loss of UBP6 suppresses the sensitivity to high temperature exhibited by most disomic strains . In addition , we found that loss of UBP6 suppresses the APS and reduces elevated basal levels of reactive oxygen species in most disomes . Importantly , analysis of the attenuated proteins revealed that attenuation of subunits of complexes was not increased; thus UBP6 does not appear to be involved in their degradation . These findings also suggest that UBP6 substrates are enriched for proteins involved in stress responses . Whether the most attenuated proteins in the disomes upon UBP6 loss are direct targets of Ubp6’s deubiquitinating activity is not clear and requires further investigation . Nonetheless , loss of UBP6 leads to the clearance of protein aggregates in aneuploid cells ( Oromendia et al . , 2012 ) . This raises the possibility that the removal of protein aggregates could contribute to the beneficial effects of UBP6 deletion . Because protein aggregates sequester numerous proteins with essential cellular functions ( Olzscha et al . , 2011 ) , their removal upon loss of UBP6 may release sequestered proteins and could account for increases of downregulated proteins in the disomes . Alternatively , protein aggregates may consist of other metastable proteins simply as a consequence of impaired folding and/or chaperone activity and attenuation in protein abundance mediated by UBP6 alleviates such stress . Establishing which proteins are direct targets of Ubp6 will help us understand the molecular mechanisms by which its loss of function suppresses aneuploidy-associated phenotypes . Our studies revealed that aneuploidy leads to higher levels of intracellular ROS . This increase in ROS may in part be responsible for the genomic instability observed in aneuploid cells ( Sheltzer et al . , 2011 ) . Consistent with this , aneuploid mouse embryonic fibroblasts as well as most cancer cells are characterized by high levels of reactive oxygen species ( Li et al . , 2010; Gorrini et al . , 2013 ) . Unexpectedly , cells respond to increases in ROS by maintaining elevated levels of ROS scavenger proteins by posttranscriptional mechanisms . Our studies raise an important question: how do cancer cells exploit posttranscriptional mechanisms to alter protein levels and respond to intrinsic genomic alterations ? Quantification of the cancer proteome remains a formidable challenge; but such efforts hold significant potential to reveal novel insights into the mechanisms by which cancer cells thrive despite their unbalanced genome . Deciphering such mechanisms could significantly impact our understanding of tumorigenesis . Finally , our studies indicate that attenuation of proteome changes and removal of protein aggregates significantly ameliorates the detrimental effects of aneuploidy . Aneuploidy causes Down syndrome and is thought to play an active role in neurodegenerative diseases ( Siegel and Amon , 2012 ) . Our studies indicate that targeting genes in the protein degradation pathway , such as UBP6 , holds significant potential to ameliorate the detrimental consequences of aneuploidy in humans . This opens the window for the design of novel approaches to improve the symptoms of Down patients and prevent or delay the onset of Alzheimer's or Huntington's disease . All stains are derivatives of W303 ( E187 ) and are listed in Figure 8—source data 1 . CEN-plasmids were isolated from the MOBY collection and introduced into wild-type cells by transformation . Gene expression analysis was performed as described in Torres et al . ( 2007 ) and is available in Figure 2—source data 1 . All aneuploid strains used in this study were subjected to comparative genomic hybridization ( CGH ) to ensure that the additional chromosome was present in its entirety . Disomic yeast strains were generated by a chromosome transfer strategy described in Torres et al . ( 2007 ) . Cells disomic for chromosomes III and VII were not obtained because the MAT locus and the CYH2 locus located on chromosome III and VII , respectively , are required for selection steps during chromosome transfer procedure . Cell disomic for chromosome VI could not be generated as two copies of ACT1 and TUB2 seem to cause lethally ( Anders et al . , 2009 ) . Cells disomic for chromosome IV , the largest chromosome in yeast , were not analyzed because they show poor cell viability ( Torres et al . , 2007 ) . Cells were grown overnight at 30°C in selective medium ( -Lys-His+G418 ) in the presence of ‘light’ or ‘heavy’ lysine ( 100 mg/ml ) . Batch cultures were diluted to OD600nm = 0 . 2 the next day and harvested once they reached an OD600nm = 1 . 0 . Analysis of the log2 ratios was performed utilizing the PRISM software ( v6 . 0 ) . Pearson mode skewness was calculated as follows: ( median − mean ) /SD . Scatterplots and their correlation values ( Pearson r ) were also calculated with the PRISM software . Hierarchical clustering was performed using the program WCluster ( http://function . princeton . edu/WCluster/ ) . WCluster takes both a data table and a weight table to allow individual measurements to be differentially considered by the clustering algorithm . Protein expression data were clustered by a Pearson correlation metric with equal weighting given to all data , or with no weight given to genes on the duplicated chromosomes . Cells were grown overnight at 30°C in selective medium ( -His+G418 ) . Batch cultures were diluted to OD600nm = 0 . 2 into YEPD medium the next day . Once they reached an OD600nm = 1 . 0 , cells were transferred to PBS buffer and incubated with 1 µM CM-H2DCFDA at 30°C for 60 min . Excess dye was washed three times and cell fluorescence was analyzed by FACS . Polysomes were prepared as described ( Clarkson et al . , 2010 ) . Briefly , 250-ml cultures were grown in YEPD at 30°C to an OD600nm of 0 . 5 . Cycloheximide was added to a final concentration of 0 . 1 mg/ml for 3 min . Cells were pelleted by centrifugation and lysed by vortexing with zirconia/silica beads in 1× PLB ( 20 mM 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid–KOH , pH 7 . 4 , 2 mM magnesium acetate , 100 mM potassium acetate , 0 . 1 mg/ml cycloheximide , 3 mM dithiothreitol [DTT] ) and treated with RNasin Plus RNase inhibitor ( Promega , Fitchburg , WI ) . Lysates were clarified by centrifugation , and 25 A260 units were resolved on 11-ml linear 10–50% sucrose gradients in 1× PLB by centrifugation in a Beckman SW41 rotor ( Beckman Coulter , Indianapolis , IN ) for 3 hr at 35 , 000 rpm . For SILAC experiments , cells grown in heavy and light media were mixed in equal numbers and lysed by bead beating in a buffer containing 8 M urea , 75 mM NaCl , 50 mM Tris-Cl , pH 8 . 2 , and a protease inhibitor cocktail ( complete mini , Roche , Germany ) using three cycles of 90 s separated by three minute incubation on ice . The lysates were cleared of unlysed cells and insoluble material by centrifugation at 14 , 000×g for 15 min at 4°C . Protein concentrations were determined by a dye binding assay ( Bio-Rad , Hercules , CA ) . Disulfide bonds were reduced by adding dithiothreitol ( Sigma , St . Louis , MO ) to a final concentration of 5 mM and incubating at room temperature for 40 min . Reduced cysteines were alkylated by the addition of iodoacetamide to 15 mM and incubation for 40 min in the dark at room temperature . Alkylation was quenched with an additional 10 mM dithiothreitol . Lysates were diluted 2 . 5-fold with Tris–HCl , pH 8 . 8 ( 25 mM final concentration ) . Lysyl endopeptidase ( lysC , Wako , Richmond , VA ) was added to a final concentration of 10 ng/ml and digests were allowed to proceed overnight at room temperature with gentle agitation . Digestion was stopped by the addition of formic acid ( FA ) to a final concentration of 1% and precipitates were removed by centrifugation at 14 , 000×g for 3 min . The supernatants were applied to pre-equilibrated Sep-Pak tC18 columns ( Waters , Milford , MA ) and the columns were washed with 1% formic acid . Bound peptides were eluted with 70% acetonitrile ( ACN ) , 1% FA and lyophilized . Cells were grown overnight at 30°C in selective medium ( -His+G418 ) . Batch cultures were diluted to OD600nm = 0 . 2 into YEPD medium the next day and harvested once they reached an OD600nm = 1 . 0 . 100 μg total peptide from each strain was resuspended in 100 μl of 0 . 2 M Hepes ( pH 8 . 5 ) . TMT six-plex reagents ( 0 . 8 mg per vial ) ( Thermo Fisher , Rockford , IL ) were resuspended in 41 μl of anhydrous ACN and 10 μl of each reagent was added to each sample . Reactions were allowed to proceed at room temperature for 1 hr , after which they were quenched by the addition of 8 μl of 5% hydroxylamine for 15 min and then acidified by the addition of 16 μl neat FA . Reaction products from all six differentially labeled samples were combined and 1 ml of 1% FA was added before desalting on a 200-mg tC18 Sep-Pak . Eluted peptides were dried in a SpeedVac and stored at −20°C . SILAC Peptides were separated by strong cation exchange ( SCX ) chromatography as described previously ( Villen and Gygi , 2008 ) with minor changes . Briefly , 500 µg of an equal mix of heavy and light peptides , were resuspended in 250 µl of SCX buffer A ( 7 mM KH2PO4 , pH 2 . 65 , 30% ACN ) . Peptides were separated on a 4 . 6 mm × 200 mm polysulfoethyl aspartamide column ( 5 µm particles; 200 Å pores; PolyLC ) using a 36 min gradient from 0% to 50% buffer B ( 7 mM KH2PO4 , pH 2 . 65 , 30% ACN , 350 mM KCl ) at a flow rate of 1 ml/min . Fractions were collected every 1 . 5 min , freeze-dried , resuspended in 1% FA , and desalted using self-packed C18 STAGE-tips ( Rappsilber et al . , 2003 ) . Peptides were eluted into glass inserts with 70% ACN/1% FA , dried , and resuspended in 100 µl of 5% FA . TMT-labeled peptides were separated by high-pH reverse-phase HPLC ( Wang et al . , 2011 ) . 600 µg of six-plex labeled peptides were resuspended in 250 µl buffer A ( 5% ACN , 10 mM NH4HCO3 , pH 8 ) and separated on a 4 . 6 mm × 250 mm 300Extend-C18 , 5 µm column ( Agilent ) using a 50 min gradient from 18% to 38% buffer B ( 90% acn , 10 mM NH4HCO3 , pH 8 ) at a flow rate of 0 . 8 ml/min . Fractions were collected over 45 min at 28 s intervals beginning 5 min after the start of the gradient in a 96-well plate and lyophilized . Fractions were resuspended in 30 µl 1% FA and pooled into 12 samples of four fractions each ( only 48 of 96 fractions were used ) by combining fractions 1/25/49/73 , 3/27/51/75 , 5/29/53/77 , 7/31/55/79 , 9/33/57/81 , 11/35/59/83 , 14/38/62/86 , 16/40/64/88 , 18/42/66/90 , 20/44/68/92 , 22/46/70/94 , 24/48/72/96 into glass vial inserts . This pooling strategy serves to minimize peptide overlap between fractions . The pooled samples were dried down and resuspended in 25 µl of 5% FA . For SILAC experiments , 2–4 µl ( ∼1–3 µg ) of each SCX fraction was analyzed by LC-MS/MS on a LTQ-Orbitrap , LTQ-Orbitrap Discovery , or LTQ-Velos hybrid linear ion trap ( ThermoFisher ) . Between 17 and 25 fractions were analyzed for each experiment . In some cases , depending on separation quality and/or instrument performance , samples were run twice pooling both sets of data . Peptides were introduced into the mass spectrometer by nano-electrospray as they eluted off a self-packed 18 cm , 100 µm ( ID ) reverse-phase column packed with either 5 µm or 3 µm , 200 Å pore size , Maccel C18 AQ resin ( The Nest Group , Southborough , MA ) . Peptides were separated using a 95 min or 65 min ( Velos , Germany ) gradient of 5–27% buffer B ( 97% ACN , 0 . 125% FA ) with an in-column flow rate of 0 . 3–0 . 5 µl/min . For each scan cycle , one high mass resolution full MS scan was acquired in the Orbitrap mass analyzer and up to 10 or 20 ( Velos ) parent ions were chosen based on their intensity for collision induced dissociation ( CID ) and MS/MS fragment ion scans at low mass resolution in the linear ion trap . Dynamic exclusion was enabled to exclude ions that had already been selected for MS/MS in the previous 60 s . Ions with a charge of +1 and those whose charge state could not be assigned were also excluded . All scans were collected in centroid mode . For TMT experiments , 2–4 µl of each fraction was analyzed on a LTQ Orbitrap Velos mass spectrometer ( Thermo Fisher Scientific ) equipped with an Accela 600 quaternary pump ( Thermo Fisher Scientific ) and a Famos Microautosampler ( LC Packings , Netherlands ) . Peptides were separated with a gradient of 6–24% ACN in 0 . 125% FA over 150 min and detected using a data-dependent Top10-MS2/MS3 ‘multi-notch’ method ( Ting et al . , 2011; McAlister et al . , 2014 ) . For each cycle , one full MS scan was acquired in the Orbitrap at a resolution of 30 , 000 or 60 , 000 at m/z = 400 with automatic gain control ( AGC ) target of 2 × 106 . Each full scan was followed by the selection of the most intense ions , up to 10 , for collision-induced dissociation ( CID ) and MS2 analysis in the linear ion trap for peptide identification and subsequent higher-energy collisional dissociation ( HCD ) and MS3 analysis in the Orbitrap for quantification of the TMT reporter ions . AGC targets of 4 × 103 and 2 × 104 were used for MS2 and MS3 scans , respectively . Ions selected for MS2 analysis were excluded from reanalysis for 90 s . Ions with +1 or unassigned charge were also excluded from analysis . A single MS3 scan was performed for each MS2 scan selecting the most intense ions from the MS2 for fragmentation in the HCD cell . The resultant fragment ions were detected in the orbitrap at a resolution of 7500 . Maximum ion accumulation times were 1000 ms for each full MS scan , 150 ms for MS2 scans , and 250 ms for MS3 scans . MS/MS spectra were matched to peptide sequences using SEQUEST v . 28 ( rev . 13 ) ( Eng et al . , 1994 ) and a composite database containing the translated sequences of all predicted open reading frames of Saccharomyces cerevisiae ( http://downloads . yeastgenome . org ) and its reversed complement . Search parameters allowed for two missed cleavages , a mass tolerance of 20 ppm , a static modification of 57 . 02146 Da ( carboxyamidomethylation ) on cysteine , and dynamic modifications of 15 . 99491 Da ( oxidation ) on methionine . For SILAC samples , parameters also included a dynamic modification of 8 . 01420 Da on lysine . For TMT samples a static modification of 229 . 16293 Da on peptide amino termini and lysines was added . Peptide spectral matches were filtered to 1% FDR using the target-decoy strategy ( Elias and Gygi , 2007 ) combined with linear discriminant analysis ( LDA ) ( Huttlin et al . , 2010 ) using the SEQUEST Xcorr and ΔCn' scores , precursor mass error , observed ion charge state , and the number of missed cleavages . LDA models were calculated for each LC-MS/MS run with peptide matches to forward and reversed protein sequences as positive and negative training data . The data were further filtered to control protein-level FDRs . Protein scores were derived from the product of all LDA peptide probabilities , sorted by rank , and filtered to 1% FDR . The FDR of the remaining peptides fell markedly after protein filtering . Further filtering based on the quality of quantitative measurements ( see below ) resulted in a final protein FDR < 1% for all experiments . Remaining peptide matches to the decoy database as well as contaminating proteins ( e . g . , human keratins ) were removed from the final data set . SILAC ratios were calculated automatically using the VISTA program ( Bakalarski et al . , 2008 ) , requiring either a minimum signal-to-noise ratio ≥ 2 for both heavy and light or signal-to-noise ≥ 5 for one of the two . For TMT experiments raw reporter ion intensities were denormalized by multiplying with the ion accumulation times for each MS3 scan and corrected for isotopic overlap between reporter ions by using empirically derived values . We required each peptide to have denormalized reporter ion intensities ≥ 20 for the zero time point and at least four of six TMT channels . In all experiments , protein ratios were normalized to account for small variations in cell mixing by recentering the log2 protein abundance ratio distributions over zero using the assumption that most proteins are present at a one-to-one ratio . Proteins coded on the duplicated chromosomes , which are more abundant in the disomes were excluded when calculating this normalization factor . Protein ratios from the SILAC experiment were calculated as described ( Torres et al . , 2010 ) using the median log2 ratio of all peptides for each protein . For TMT experiments , relative protein abundances were calculated as the weighted average of all peptides from each protein using the ratio of the summed reporter ion intensities in each channel . Ratios for both experiments were log2-transformed for all subsequent analysis . Total RNA was isolated from cells frozen on filters . Filters were incubated for 1 hr at 65°C in lysis buffer ( 10 mM EDTA , 0 . 5% SDS , and 10 mM Tris , pH 7 . 5 ) and acid phenol . The aqueous phase was further extracted twice with an equal volume of chloroform using phase lock gel ( Eppendorf , Germany ) . Total RNA was then ethanol precipitated and further purified over RNeasy columns ( Qiagen , Germany ) . RNA quality was checked using the Bioanalyzer RNA Nano kit , and 325 ng was used for microarray labeling with the Agilent Low RNA Input Fluorescent Linear Amplification Kit . Reactions were performed as directed except using half the recommended reaction volume and one quarter the recommended Cy-CTP amount . Dye incorporation and yield were measured with a Nanodrop spectrophotometer . Equal amounts of differentially labeled control and sample cRNA were combined such that each sample contained at least 2 . 5 pmol dye . Samples were mixed with control targets , fragmented , combined with hybridization buffer , and hybridized to a microarray consisting of 60mer probes for each yeast open reading frame ( Agilent ) . Microarrays were rotated at 60°C for 17 hr in a hybridization oven ( Agilent , Santa Clara , CA ) . Arrays were then washed according to the Agilent SSPE wash protocol , and scanned on an Agilent scanner . The image was processed using the default settings with Agilent Feature Extraction software . All data analysis was performed using the resulting log2 ratio data , and filtered for spots called as significantly over background in at least one channel . mRNA expression data for cells grown in synthetic medium were obtained from the GEO database with accession number GSE7812 . mRNA expression data for cells grown in YEPD medium have been deposited at the GEO database with accession number GSE55166 . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD001019 ( Vizcaino et al . , 2014 ) . Data available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . 65364 ( Dephoure et al . , 2014 )	Nearly all tumor cells contain abnormal number of chromosomes . This state is called aneuploidy , and can also cause embryos to be miscarried , or to be born with severe developmental disorders . Proteins are produced from the genes contained within chromosomes , and so cells with too many chromosomes produce too many of some proteins . How do these cells cope with this excess ? Previous work identified one strategy where a gene called UBP6 is mutated to prevent it from working correctly . The UBP6 gene normally encodes a protein that removes a small tag ( called ubiquitin ) from other proteins . This tag normally marks other proteins that should be degraded; thus , if UBP6 is not working , more proteins are broken down . Dephoure et al . investigated the effect of aneuploidy on the proteins produced by 12 different types of yeast cell , which each had an extra chromosome . In general , the amount of each protein produced by these yeast increased depending on the number of extra copies of the matching genes found on the extra chromosome . However , this was not the case for around 20% of the proteins , which were found in lower amounts than expected . Dephoure et al . revealed that this was not because fewer proteins were made , but because more were broken down . These proteins may be targeted for degradation because they are unstable , as many of these proteins need to bind to other proteins to keep them stable—but these stabilizing proteins are not also over-produced . Aneuploidy in cells also has other effects , including changing the cells' metabolism so that the cells grow more slowly and do not respond as well to stress . However , Dephoure et al . found that , as well as reducing the number of proteins produced , deleting the UBP6 gene also increased the fitness of the cells . Targeting the protein encoded by the UBP6 gene , or others that also stop proteins being broken down , could therefore help to reduce the negative effects of aneuploidy for a cell . Whether targeting these genes or proteins could also help to treat the diseases and disorders that result from aneuploidy , such as Alzheimer's and Huntington's disease , remains to be investigated .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2014	Quantitative proteomic analysis reveals posttranslational responses to aneuploidy in yeast
Hippocampal place cells encode an animal's past , current , and future location through sequences of action potentials generated within each cycle of the network theta rhythm . These sequential representations have been suggested to result from temporally coordinated synaptic interactions within and between cell assemblies . Instead , we find through simulations and analysis of experimental data that rate and phase coding in independent neurons is sufficient to explain the organization of CA1 population activity during theta states . We show that CA1 population activity can be described as an evolving traveling wave that exhibits phase coding , rate coding , spike sequences and that generates an emergent population theta rhythm . We identify measures of global remapping and intracellular theta dynamics as critical for distinguishing mechanisms for pacemaking and coordination of sequential population activity . Our analysis suggests that , unlike synaptically coupled assemblies , independent neurons flexibly generate sequential population activity within the duration of a single theta cycle . Cognitive processes are thought to involve the organization of neuronal activity into phase sequences , reflecting sequential activation of different cell assemblies ( Hebb , 1949; Harris , 2005; Buzsáki , 2010; Wallace and Kerr , 2010; Palm et al . , 2014 ) . During navigation , populations of place cells in the CA1 region of the hippocampus generate phase sequences structured around the theta rhythm ( e . g . , Skaggs et al . , 1996; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007 ) . As an animal moves through the firing field of a single CA1 neuron , there is an advance in the phase of its action potentials relative to the extracellular theta cycle ( O'Keefe and Recce , 1993 ) . Thus , populations of CA1 neurons active at a particular phase of theta encode the animal's recent , current , or future positions ( Figure 1A , B ) . One explanation for these observations is that synaptic output from an active cell assembly ensures its other members are synchronously activated and in addition drives subsequent activation of different assemblies to generate a phase sequence ( Figure 1C ) ( Harris , 2005 ) . We refer to this as the coordinated assembly hypothesis . An alternative possibility is that independent single cell coding is sufficient to account for population activity . According to this hypothesis , currently active assemblies do not determine the identity of future assemblies ( Figure 1D ) . We refer to this as the independent coding hypothesis . 10 . 7554/eLife . 03542 . 003Figure 1 . Phase sequences in a place cell population . ( A ) During navigation , place cells are sequentially activated along a route . ( B ) Within each theta cycle , this slow behavioral sequence of place cell activations is played out on a compressed timescale as a theta sequence . Theta sequences involve both rate and phase modulation of individual cells , but it remains unclear whether additional coordination between cells is present . ( C ) Internal coordination may bind CA1 cells into assemblies , and sequential assemblies may be chained together synaptically . This would require specific inter- and intra-assembly patterns of synaptic connectivity within the network . ( D ) Alternatively , according to the independent coding hypothesis , each cell is governed by theta phase precession without additional coordination . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 003 Since these coding schemes lead to different views on the nature of the information transferred from hippocampus to neocortex and on the role of CA1 during theta states , it is important to distinguish between them . While considerable experimental evidence has been suggested to support the coordinated assembly hypothesis ( e . g . , Harris et al . , 2003; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Maurer et al . , 2012; Gupta et al . , 2012 ) , the extent to which complex sequences of activity in large neuronal populations can be accounted for by independent coding is not clear . To address this we developed phenomenological models of independent and coordinated place cell activity during navigation . In the independent coding model , the spiking activity of each cell is generated by rate coding across its place field and phase precession against a fixed theta rhythm . We show that in this model phase coding generates a traveling wave which propagates through the population to form spike sequences . This wave is constrained by a slower moving modulatory envelope which generates spatially localized place fields . In the coordinated assembly model , the spikes generated by each cell are also influenced by the activity of other cells in the population . As a result , population spike patterns are further entrained by population interactions which counter the effects of single cell spike time variability and increase the robustness of theta sequences . The independent coding hypothesis predicts that a population of independent cells will be sufficient to explain the spatiotemporal dynamics of cell assemblies in CA1 . In contrast , the coordinated assembly hypothesis predicts that groups of cells show additional coordination beyond that imposed by a fixed firing rate and phase code ( Harris et al . , 2003; Harris , 2005 ) . We show that the independent coding model is sufficient to replicate experimental data previously interpreted as evidence for the coordinated assembly hypothesis ( Harris et al . , 2003; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Maurer et al . , 2012; Gupta et al . , 2012 ) , despite the absence of coordination within or between assemblies . Moreover , novel analyses of experimental data support the hypothesis that place cells in CA1 code independently . Independent coding leads to new and experimentally testable predictions for membrane potential oscillations and place field remapping that distinguish circuit mechanisms underlying theta sequences . In addition we show that , despite the apparent advantage of coordinated coding in generating robust sequential activity patterns , it suffers from an inability to maintain these patterns in a novel environment . Thus , a key advantage of sequence generation through independent coding is to allow flexible global remapping of population activity while maintaining the ability to generate coherent theta sequences in multiple environments . To test the independent coding hypothesis , we developed a phenomenological model which generates activity patterns for place cell populations during navigation . While a phenomenological model of CA1 phase precession has previously been developed ( Geisler et al . , 2010 ) , several features of this model limit its utility for investigation of coordination across neuronal populations . First , the previous model addresses only the temporal dynamics of single unit activity and population average activity , without addressing the spatiotemporal patterns of spiking activity within the population , the nature of which is a central question in the present study . Second , the previous model assumes coordination between cells in the form of fixed temporal delays and is formulated for a fixed running speed . In contrast , we wish to understand in detail the temporal relationships between cells arising in populations with no direct coordination and how these temporal relationships might depend on factors such as running speed . We therefore develop a model of a single cell with a given place field and phase code and proceed to derive the patterns of population activity under the independent coding hypothesis . To do this , we modeled the firing rate field for each neuron using a Gaussian tuning curve: ( 1 ) rx ( x ) =A exp ( − ( x−xc ) 22σ2 ) , where rx describes firing rate when the animal is at location x within a place field with center xc , width σ , and maximum rate A ( Figure 2A , top panel ) . Simultaneously , we modeled the firing phase using a circular Gaussian: ( 2 ) rϕ ( ϕ ( x ) , θ ( t ) ) =exp ( k cos ( ϕ ( x ) −θ ( t ) ) ) , where rϕ describes the firing probability of the neuron at each theta phase at a given location ( Figure 2B ) . Here , θ ( t ) = 2πfθt is the local field potential ( LFP ) theta phase at time t and ϕ ( x ) is the preferred firing phase associated with the animal's location x , termed the encoded phase . The encoded phase ϕ ( x ) is defined to precess linearly across the place field ( Figure 2A , bottom panel; Supplementary file 1 , Appendix: A1 ) . The phase locking parameter k determines the precision at which the encoded phase is represented in the spike output ( Figure 2B ) . The instantaneous firing rate of the cell is given by the product of these two components r = rxrϕ . The phase locking can be set so that the cell exhibits only rate coding ( at k = 0 , where r = rx ) , only phase coding ( as k → ∞ , where all spikes occur at exactly the encoded phase ϕ ( x ) ) or anywhere in between ( Figure 2C ) . 10 . 7554/eLife . 03542 . 004Figure 2 . Single cell coding model . ( A ) Firing rate and phase at different locations within a cell's place field are determined by a Gaussian tuning curve rx and linearly precessing encoded phase ϕ , respectively . ( B ) The dependence of single cell activity on the LFP theta phase θ is modeled by a second tuning curve rϕ which depends on the angle between the LFP theta phase θ and encoded phase ϕ at the animal's location . The phase locking parameter k controls the precision of the phase code . ( C ) The combined dependence of single cell activity on location and LFP theta phase . ( D ) Temporal evolution of the rate and phase tuning curves for a single cell as a rat passes through the place field at constant speed . ( E ) The total firing rate corresponding to ( D ) , and spiking activity on 1000 identical runs . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 00410 . 7554/eLife . 03542 . 005Figure 2—figure supplement 1 . Effect of normalization factor ( Nspikes ) . Firing rate vs time for runs with v = 50 cm/s , k = 0 . 7 , and three different values of Nspikes . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 005 To model place cell activity during navigation on a linear track , we set x ( t ) = vt , where v is the running speed ( Figure 2D , E ) . This causes the encoded phase ϕ ( t ) to precess linearly in time at a rate fϕ which is directly proportional to running speed and inversely proportional to place field size , as in experimental data ( Huxter et al . , 2003; Geisler et al . , 2007 ) . To generate spikes we used an inhomogeneous Poisson process with an instantaneous rate r = rxrϕ . We normalized the firing rate such that the average number of spikes fired on a pass through a place field is independent of running speed ( see Supplementary file 1 , Appendix: A2 ) ( Huxter et al . , 2003 ) . If the phase ϕ ( x ) at each location in the place field is fixed , the full rate and phase coding properties of a cell are encompassed by three independent parameters—the width of the spatial tuning curve σ , the degree of phase locking k , and the average number of spikes per pass Nspikes . Phase precession ( Figure 2C ) and firing rate modulation as a function of time in this model ( Figure 2E ) closely resemble experimental observations ( e . g . , Skaggs et al . , 1996; Mizuseki and Buzsaki , 2013 ) . Place cells often show variations in firing rate in response to nonspatial factors relevant to a particular task ( e . g . , Wood et al . , 2000; Fyhn et al . , 2007; Griffin et al . , 2007; Allen et al . , 2012 ) . In our model , such multiplexing of additional rate coded information can be achieved by varying the number of spikes per pass Nspikes without interfering with the other parameters ϕ ( x ) , σ , and k ( Figure 2—figure supplement 1 ) . It has been shown that the trial to trial properties of phase precession in individual cells are more variable than would be expected based on the pooled phase precession data ( Schmidt et al . , 2009 ) . While it is possible that such trial to trial variability could reflect coordination between cell assemblies , such variability is equally consistent with an independent population code , and our model can be readily extended to incorporate such properties ( Supplementary file 1 , Appendix: A2 ) . Given this single cell model and assuming an independent population code , we next investigated the spatially distributed patterns of spiking activity generated in a CA1 population . To map the spatiotemporal dynamics of the population activity onto the physical space navigated by the animal , we analyzed the distributions of the rate components rx and phase components rϕ of activity in cell populations sorted according to the location xc of each place field ( Supplementary file 1 , Appendix: A3 ) . Our model naturally generates population activity at two different timescales: the slow behavioral timescale at which the rat navigates through space and a fast theta timescale at which trajectories are compressed into theta sequences . While the rat moves through the environment , the spatial tuning curves rx ( x ) generate a slow moving ‘bump’ of activity which , by definition , is comoving with the rat ( Figure 3A , top , black ) . Simultaneously , the phasic component rϕ ( ϕ ( x ) , θ ( t ) ) instantiates a traveling wave ( Figure 3A , top , red ) . Due to the precession of ϕ ( t ) , the wave propagates forward through the network at a speed faster than the bump , resulting in sequential activation of cells along a trajectory on a compressed timescale . The slower bump of activity acts as an envelope for the traveling wave , limiting its spatial extent to one place field ( Figure 3A , bottom ) . The continuous forward movement of the traveling wave is translated into discrete , repeating theta sequences via a shifting phase relationship to the slow moving component ( Figure 3B–D , Video 1 ) . Moreover , this shifting phase relationship generates global theta oscillations at exactly the LFP frequency that cells were defined to precess against ( Figure 3B , top panel ) . Thus , our model can be recast in terms of the dynamics of a propagating wavepacket comprising two components , with network theta resulting from their interaction . While we define single cells to precess against a reference theta rhythm ( i . e . , the LFP ) , we now see that this same reference oscillation emerges from the population , despite the higher frequencies of individual cells . 10 . 7554/eLife . 03542 . 006Figure 3 . Spatiotemporal dynamics of CA1 populations governed by independent coding . ( A ) Top: Population dynamics during a single theta cycle on a linear track after ordering cells according to their place field center xc in physical space . The two components of the population activity are shown—the slow moving envelope ( black ) and the fast moving traveling wave ( red ) , which give rise to rate coding and phase coding , respectively ( cf . Figure 2 ) . Bottom: Resulting firing rates across the population . When the traveling wave and envelope are aligned , the population activity is highest ( middle panel ) . The dashed line shows the location of the rat at each instant . ( B ) Firing rate in the population over seven consecutive theta cycles . The fast and slow slopes are shown ( solid and dashed lines , respectively ) , corresponding to the speeds of the traveling wave and envelope as shown in part ( A ) . The top panel shows the LFP theta oscillations and emergent population theta oscillations , which are generated by the changing population activity as the traveling wave shifts in phase relative to the slower envelope ( see Video 1 ) . ( C and D ) The spiking activity for a population of 180 cells . All panels used v = 50 cm/s , so that vp = 350 cm/s and c = 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 00610 . 7554/eLife . 03542 . 007Figure 3—figure supplement 1 . CA1 population activity governed by coordinated assemblies . ( A ) The simulated place cells interact via a combination of asymmetric excitation and feedback inhibition . The weights plotted here govern how the spikes emitted by a given cell will influence the spiking activity of its peers depending on their relative place field locations . ( B ) Population firing rate on a single run along a linear track ( 180 cells with v = 50 cm/s and k = 0 . 5 ) . The firing rate in each cell is a product of the animal's location , the LFP theta phase and the influence of recent peer spiking activity . ( C ) The spiking activity , generated using an inhomogeneous Poisson process . ( D ) Comparison of the global population firing rate for an independent coding population ( black ) and a coordinated population ( red ) , with identical single cell properties . Interactions between cells amplify theta oscillations and introduce a shift in firing phase . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 00710 . 7554/eLife . 03542 . 008Video 1 . Traveling wave dynamics in populations of CA1 place cells . Top: Distribution of the rate ( black ) and phasic ( red ) tuning curves for a population of linear phase coding place cells during constant speed locomotion on a linear track ( cf . Figure 3A ) . The evolution in the population over 7 consecutive theta cycles is shown , slowed by a factor of approximately 16× . Bottom: The evolution of the overall firing rate distribution in the population , generated by multiplying the two tuning curves shown in the top panel . Note that the population firing rate undergoes oscillations at LFP theta frequency and the center of mass of the population activity shifts from behind the animal to ahead of the animal in each theta cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 008 Our model's prediction of global theta oscillations emerging in networks of faster oscillating place cells is consistent with a previous phenomenological model which assumed a fixed running speed and fixed , experimentally determined temporal delays between cells ( Geisler et al . , 2010 ) . However , in contrast to previous models , our model based on single cell coding principles allows an analysis in which only place field configurations and navigational trajectories are required to fully predict at any running speed both the global theta oscillation and the detailed population dynamics . Experimental data show that the frequency of LFP theta oscillations is relatively insensitive to the running speed of the animal , showing a mild increase with running speed compared to a larger single unit increase ( Geisler et al . , 2007 ) . We therefore investigated the relationship between the running speed of the animal , the temporal delays between cells and the frequency of population theta oscillations in the independent coding model . The spiking delays between cells in our model are determined by speed of the fast moving traveling wave vp , which is related to the rat's running speed v by: ( 3 ) vp=cv , where c is called the compression factor . This factor is equivalent to the ratio of the rat's actual velocity and the velocity of the representation within a theta cycle and has been quantified in previous experimental work ( Skaggs et al . , 1996; Dragoi and Buzsáki , 2006; Geisler et al . , 2007; Maurer et al . , 2012 ) , although the relationship to the traveling wave model developed here was not previously identified ( see Supplementary file 1 , Appendix: A2 for derivation ) . Analysis of our model demonstrates that for an independent population code the compression factor naturally depends on running speed . This change in compression factor with running speed ensures that the network maintains a fixed population theta frequency while running speed and single unit frequency vary: ( 4 ) vp−v=λfθ , where the constant λ is the wavelength of the traveling wave ( equal to the size of a place field , measured as the distance over which a full cycle of phase is precessed [Maurer et al . , 2006] ) and vp − v stays constant across running speeds due to the changing compression factor . Hence , independent coding predicts temporal delays which are dependent on running speed . Conversely , our analysis shows that models incorporating fixed temporal delays between cells ( e . g . , Diba and Buzsáki , 2008; Geisler et al . , 2010 ) cannot maintain an invariant relationship between spike phase and location without producing a population theta oscillation whose frequency decreases rapidly with running speed , in conflict with experimental observations ( Geisler et al . , 2007 ) . In order to compare activity patterns predicted by independent coding schemes with those predicted when interactions between cell assemblies are present , we developed a second model in which the spiking activity of each place cell influences the spiking activity of peer cells within the population . While single cell rate and phase tuning curves in this coordinated assembly model are identical to those in the independent coding model , a peer weight function also modulates the probability of a spike occurring in each cell depending on the spikes of its peers ( Figure 3—figure supplement 1A , Supplementary file 1 , Appendix: A4 ) . In this model , asymmetric excitation stabilizes the temporal relationship between sequentially activated assemblies , while feedback inhibition between place cells normalizes firing rates ( cf . Tsodyks et al . , 1996 ) . The resulting sequences are considerably more robust than those generated by independent coding with the same single cell properties ( Figure 3—figure supplement 1B–C ) . Assembly interactions also amplify theta oscillations in the network ( Figure 3—figure supplement 1D ) ( Stark et al . , 2013 ) . Hence , assembly coordination provides a potential mechanism for stabilizing the sequential activity patterns generated by noisy neurons , as interactions entrain cells in the population into coherent activation patterns within each theta cycle . While alternative forms of assembly coordination might also be considered , we choose the present model for two key reasons . First , this model is simple , containing relatively few adjustable parameters while capturing the essential features of sequence generation via assembly coordination . Second , as we will show below , the coordination between cells under this model is sufficient to evaluate statistical tests of independence , allowing a systematic framework with which to interpret the results of such tests on experimental data . We next investigated the extent to which models for population activity based on independent coding and coordinated assemblies can account for observations previously suggested to imply coordination within and between assemblies ( Harris et al . , 2003; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Maurer et al . , 2012; Gupta et al . , 2012 ) . We show below that , although these observations at first appear to imply assembly coordination , they can be accounted for by the independent coding model . We go on to establish the power of several tests to distinguish spike patterns generated by independent and coordinated coding models . By applying these tests to experimental data , we provide further evidence that CA1 population activity is generated through independent coding . We first assessed whether independent coding accounts for membership of cell assemblies . A useful measure of the coding properties of place cell populations is to test how accurately single unit activity can be predicted from different variables . If , after accounting for all known single cell coding properties , predictions of the activity of individual place cells can be further improved by information about firing by their peer cells , it is likely that such cells are interacting through cell assemblies ( Harris , 2005 ) . Initial analysis of CA1 place cell firing suggested this is the case , with coordination between cells at the gamma timescale being implicated ( Harris et al . , 2003 ) . Because this improved predictability directly implies interactions between CA1 neurons , it would constitute strong evidence against the independent coding hypothesis . However , in accounting for single cell phase coding properties , the prediction analysis of Harris et al . ( 2003 ) assumed that firing phase is independent of movement direction in an open environment . In contrast , more recent experimental data show that in open environments firing phase always precesses from late to early phases of theta , so that firing phase at a specific location depends on the direction of travel ( Huxter et al . , 2008; Climer et al . , 2013; Jeewajee et al . , 2014 ) . Therefore , to test if the apparent peer-dependence of place cell activity is in fact consistent with independent coding , the directionality of phase fields must be accounted for . To address this we first considered whether the assumption of a nondirectional phase field would lead to an erroneous conclusion of coordinated coding when analyzing spike patterns generated by the independent coding model . To do this , we extended the traveling wave model to account for phase precession in open environments ( Supplementary file 1 , Appendix: A6 ) . We then constructed phase fields from simulated spiking data following the approach of Harris et al . ( 2003 ) , in which firing phase is averaged over all running directions , and separately constructed directional phase fields consistent with recent experimental observations ( Huxter et al . , 2008; Climer et al . , 2013; Jeewajee et al . , 2014 ) . We then calculated the predictability of neuronal firing patterns generated by the independent coding model using each of these phase fields . For simplicity , we considered the problem in one dimension , treating separately passes from right to left , left to right , and the combined data in order to generate the directional and nondirectional phase fields ( Figure 4A , B , respectively ) . We ignored any shifts in place field centers for different running directions ( e . g . , Battaglia et al . , 2004; Huxter et al . , 2008 ) and assumed that the place cells did not engage in multiple reference frames ( Jackson and Redish , 2007; Fenton et al . , 2010 ) . 10 . 7554/eLife . 03542 . 009Figure 4 . Peer prediction analysis for an independent population code . ( A ) Combined place and phase fields constructed from simulated data using only runs with a single direction . ( B ) Place/phase field constructed from a combination of both running directions , as used by Harris et al . ( 2003 ) . ( C ) Predictability analysis , using various combinations of place , phase , and peer activity . When using the nondirectional phase field of Harris et al . ( 2003 ) , an additional peer predictability emerges ( black vs green and purple ) . However , this additional predictability is seen to be erroneous if the directional phase field is used to predict activity ( red ) . ( D ) Dependence of peer predictability on the peer prediction timescale and phase locking of individual cells , for an independent population code . The heat map shows the predictability of a cell's activity from peer activity ( cf . part C , green line ) . The optimal peer prediction timescale depends on the amount of phase locking . The 20 ms characteristic timescale of peer correlations reflects independent phase precession of single cells rather than transient gamma synchronization of cell assemblies . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 00910 . 7554/eLife . 03542 . 010Figure 4—figure supplement 1 . Change in information after addition of peer activity to prediction metrics . Distributions of information gain/loss in individual cells after including peer activity in addition to all other prediction metrics . For independent coding and experimental data , peer prediction causes a decrease in information on average ( p = 3 . 9 × 10−17 and p = 1 . 4 × 10−6 , respectively ) . For coordinated coding , peer prediction causes an increase in information on average ( p = 9 × 10−83 ) . The decrease in information observed for independent coding simulations when peer activity is included occurs due to overfitting on a dataset of finite size . Due to statistical fluctuations in the data , peer weights are generally estimated as non-zero . Both the peer weights and the change in information when peers are included would be expected to approach zero as the amount of data increases for independent coding simulations , but not for coordinated coding simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 01010 . 7554/eLife . 03542 . 011Figure 4—figure supplement 2 . Results of prediction analysis on individual sessions . Top: Number of cells for which prediction improved with peers after place fields , velocity modulation factors and directional phase fields had been fitted , shown for each session/running direction in the experimental dataset . Middle: The results when the same analysis was applied to data simulated with independent coding ( twice as many sessions were simulated for comparison ) . Bottom: The results when data were simulated with coordinated assemblies . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 011 For the independent coding model , we find that peer prediction provides a higher level of information about a neuron's firing than predictions based on place and nondirectional phase fields , despite the absence of intra-assembly coordination in our simulated data ( Figure 4C , green and purple ) . However , prediction based on place fields and directional phase fields outperforms both of these metrics ( Figure 4C , red ) . Therefore , previous evidence for intra-assembly coordination can be explained by a failure to account for the phase dependence of CA1 firing . Instead , our analysis indicates that independent phase precession of CA1 neurons is sufficient to account for observations concerning membership of CA1 assemblies . We also find that nondirectional phase fields ( Figure 4B ) , as assumed by ( Harris et al . , 2003 ) , yield little improvement in predictability of a neuron's firing compared with predictions based on the place field alone , and for high phase locking are detrimental ( Figure 4C , blue vs black ) . While Harris et al . ( 2003 ) found that nondirectional phase fields generally do improve prediction , this discrepancy may arise from more complex details of experimental data in open exploration , for example a nonuniform distribution of running directions through the place field , which would cause the information in nondirectional phase fields to increase . Because peers share a relationship to a common theta activity and implement similar rules for generation of firing , a cell's activity in the independent coding model can nevertheless be predicted from that of its peers in the absence of information about location or theta phase ( Figure 4C , green ) . The quality of this prediction is dependent on the timescale at which peer activity is included in the analysis , so that the optimal timescale for peer prediction provides a measure of the temporal resolution of assembly formation . In experimental data the optimal timescale for peer prediction is approximately 20 ms , which corresponds to the gamma rhythm and the membrane time constant of CA1 neurons ( Harris et al . , 2003 ) . We find that in the independent coding model the optimal peer prediction timescale depends strongly on phase locking ( Figure 4D ) . Even though the model does not incorporate gamma oscillations or neuronal membrane properties , high values of phase locking also show a striking peak in peer predictability around the 20 ms range ( Figure 4D ) . We show below that for running speeds in the range 35–75 cm/s phase locking is likely to lie within the range at which the observed 20 ms prediction timescale dominates . Thus , the 20 ms timescales found both here and experimentally are explainable as a signature of the common , independent phase locking of place cells to the theta rhythm , rather than transient gamma coordination or intrinsic properties of CA1 neurons . While the above analysis demonstrates that independent coding is consistent with previous experimental results , it does not exclude the presence of coordinated assemblies . In particular , it is not clear whether , when applied to experimental data , including information about peer activity would continue to improve prediction compared to place and directional phase fields alone . We therefore applied the prediction analysis based on directional phase fields to experimental datasets recorded from CA1 place cells ( Mizuseki et al . , 2014 ) . To provide benchmarks for the interpretation of experimental results , we also analyzed simulated datasets generated with either independent coding or coordinated assemblies . We simulated datasets with the same number of sessions and recorded cells per session as the experimental dataset in order to obtain measures of peer prediction performance expected under each hypothesis ( see ‘Materials and methods’ ) . In simulations of independent cells , we found that information about peer activity continues to improve predictability compared to prediction from place and directional phase fields alone . The source of this predictability was found to be the common modulation of firing rate in each cell with the running speed of the animal , which is a further single cell coding feature not previously accounted for in prediction analyses ( McNaughton et al . , 1984; Czurko et al . , 1999; Huxter et al . , 2003; Ahmed and Mehta , 2012 ) . We therefore included in our analysis an additional prediction factor , termed the velocity modulation factor ( see ‘Materials and methods’ ) . After accounting for rate fields , directional phase fields and velocity modulation factors , inclusion of peer information increased the predictability of 84% of place cells simulated through coordinated coding , but only 38% of cells simulated through independent coding ( see Table 1 for a summary of all prediction metrics ) . On average , information decreased by 0 . 047 bits/s for each cell simulated by independent coding and increased by 0 . 24 bits/s for coordinated coding when peer information was added ( Wilcoxon signed rank test , p = 3 . 9 × 10−17 and p = 9 × 10−83 , respectively , Figure 4—figure supplement 1 ) . Thus , this new prediction analysis which accounts for directional phase fields and velocity modulation can effectively distinguish between independent and coordinated coding . 10 . 7554/eLife . 03542 . 012Table 1 . Performance of prediction metrics on experimental and simulated dataDOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 012Prediction metricIndependent codingCoordinated codingExperimental dataLocation100%100%44 . 6% ( SEM 5 . 8% ) Running speed99 . 3%99 . 7%77 . 8% ( SEM 3 . 7% ) Phase field99 . 3%100%75 . 7% ( SEM 5 . 7% ) Peer activity38%84 . 3%32 . 5% ( SEM 11% ) The percentage of cells for which prediction performance increased with the addition of each metric . Percentages refer to the number of cells for which information increased when the specified metric was included in addition to those listed in rows above . Note that for velocity , phase and peer prediction , only those cells for which prediction performance improved with information about location were considered . Simulations demonstrate that , after taking into account place fields , velocity modulation factors and phase fields , information about peer activity improves prediction for the majority of cells when coordination is present , but not when cells are independent . Experimental data are consistent with independent coding . When we applied this prediction analysis to experimental data , prediction performance improved for 75 . 7% ( ±5 . 7% , SEM , n = 10 sessions ) of experimentally observed place cells when phase fields were included and 77 . 8% ( ±3 . 7% ) of place cells when velocity modulation factors were included . In contrast , prediction performance improved for only 32% ( ±11% ) of the experimentally observed place cells when peer information was included after accounting for single cell coding properties ( Figure 4—figure supplement 2 shows the results for individual experimental sessions ) . On average , addition of peer information decreased the predictability of each cell by 0 . 049 bits/s ( ±0 . 013 , SEM , n = 270 cells , Wilcoxon signed rank test , p = 1 . 4 × 10−6 ) , in agreement with independent coding simulations and in contrast to coordinated coding simulations . Hence , after fully accounting for the directional properties of phase fields and the dependence of firing rate on running speed , peer prediction analysis supports independent coding as the basis of experimentally observed place cells in CA1 . Therefore , based on comparison of simulated with experimental datasets , coordinated assemblies appear unlikely to account for the observed activity in CA1 . While the above analysis demonstrates that intra-assembly interactions are not required to account for membership of CA1 assemblies , several studies support a role for inter-assembly coordination in the generation of theta sequences ( Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Maurer et al . , 2012; Gupta et al . , 2012 ) . We therefore investigated whether the independent coding or coordinated assembly model would better account for the results of these studies . We focus initially on the path length encoded by spike sequences , which we define as the length of trajectory represented by the sequence of spikes within a single theta cycle . Experimental data show that this path length varies with running speed ( Maurer et al . , 2012; Gupta et al . , 2012 ) , but it is not clear whether this phenomenon is a feature of independent coding or instead results from coordination between assemblies . To address this we first derived analytical approximations to the sequence path length for strong phase coding , where k → ∞ ( Supplementary file 1 , Appendix: A2 ) . This analysis predicts a linear increase in sequence path length with running speed , but with a lower gradient than that found experimentally ( Maurer et al . , 2012 ) . Hence , independent coding with strong phase locking does not quantitatively explain the changes in sequence properties with running speed . We reasoned that independent coding might still explain observed sequence path lengths if a more realistic tradeoff between rate and phase coding is taken into account . To test this , we varied phase locking k and decoded the path length following the method of Maurer et al . ( 2012 ) , which decodes the location represented by the population at each time bin in a theta cycle to estimate the encoded trajectory . We found that a good match to the data of Maurer et al . ( 2012 ) can be obtained by assuming that the degree of phase locking increases with running speed ( Figure 5A ) . This is due to the dependence of the decoded path length on the strength of phase locking ( Figure 5—figure supplement 1A ) . 10 . 7554/eLife . 03542 . 013Figure 5 . Decoded sequence path lengths and population activity propagation speeds . ( A ) With constant phase locking , the decoded path length increases linearly with running speed , but to account for experimental data a dependence of phase locking on running speed is required . The shaded regions show lower and upper bounds ( k = 0 and k = ∞ ) . ( B ) Dependence of decoded fast slope on running speed ( cf . our Figure 3B; Figure 3 of Maurer et al . ( 2012 ) ) . Again , a match to the data requires a velocity dependent phase locking . ( C ) The decoded slow slope matches the analytical value , where the population travels at the running speed v . Bounds show LFP theta frequencies below 4 Hz ( upper bound ) and above 12 Hz ( lower bound ) at each given running speed . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 01310 . 7554/eLife . 03542 . 014Figure 5—figure supplement 1 . Dependence of decoded sequence path lengths , fast slopes , and slow slopes on phase locking . ( A ) The decoded path length depends on the phase locking of individual cells . For zero phase locking , the decoded path length is the distance traveled by the rat in a theta cycle . This is because the decoded location in each time bin is simply the location of the rat . As phase locking is increased the path length increases asymptotically towards our analytical result , which is the distance traveled by the rat plus one full place field . This effect arises due to the gradual separation of cells representing different locations into separate theta phases , as seen explicitly in Figure 3C , D . Phases within a single theta cycle represent past , present , and future locations along the track . Dashed lines show the phase locking values plotted in Figures 2 , 3 . ( B ) Dependence of decoded fast slope on phase locking . While the analytical result for vp is independent of phase locking , the decoded value shown here is consistent with the intuitive notion that the sequence path length D is equal to the distance traveled by the fast moving wave in a theta cycle . ( C ) The decoded slow slope does not depend on phase locking , which is expected given the separation of timescales involved . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 01410 . 7554/eLife . 03542 . 015Figure 5—figure supplement 2 . Results of shuffling analysis . ( A–D ) The analysis of Foster and Wilson ( 2007 ) and ( E–F ) a corrected analysis . ( A ) Spike phases were initially calculated by interpolation between theta peaks , shown as dotted lines . ( B ) After shuffling the phases of spikes , a new spike time is calculated by interpolation between the nearest two theta troughs ( dotted lines ) to the original spike , which often generates erroneous spike times . The shuffled spike in this case acquires a small phase jitter , but a large temporal jitter . ( C ) The unshuffled sequence correlations between cell rank order and spike times . The red line shows the mean correlation . ( D ) Shuffled sequence correlations remained greater than zero , but were significantly reduced relative to the unshuffled case as in experimental data ( Foster and Wilson , 2007 ) . ( E ) Results of a corrected shuffling procedure applied to simulated independent coding datasets and an experimental dataset ( height magnified for comparison ) . Displayed are the average changes in sequence correlations caused by shuffling for each simulated dataset . In 74% of simulated datasets , there was no significant difference between the original and shuffled distributions . ( F ) Results of the corrected shuffling procedure when applied to datasets simulated with coordinated assemblies . In 81% of simulated coordinated coding datasets , shuffling significantly changed the distribution of sequence correlations . The experimental dataset was not significantly affected by shuffling ( p = 0 . 28 , t-test , 2436 putative sequences ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 015 Maurer et al . ( 2012 ) found that the compression factor c , which measures the compression of an encoded trajectory into a single theta cycle , also depends on running speed . To test whether independent coding might account for this observation , we investigated the behavior of the fast and slow slopes of population activity ( as shown in Figure 3B ) , representing assembly propagation at theta timescales and behavioral timescales , respectively ( i . e . , vp and v ) . In the analysis of Maurer et al . ( 2012 ) , the compression factor was estimated as the ratio of these two quantities . Following again the methods used by Maurer et al . ( 2012 ) to decode the fast and slow slopes from spiking data , we found that the dependence of the decoded fast slope on running speed in our simulated data matches experimental data provided that phase locking is again made dependent on running speed ( Figure 5B , Figure 5—figure supplement 1B ) . However , the slower behavioral timescale dynamics did not match those reported by Maurer et al . ( 2012 ) . Our decoded values for the slow slope closely matched the true value based on the rat's running speed . In contrast , the values reported by Maurer et al . ( 2012 ) are considerably lower ( Figure 5C ) which , if correct , would suggest that the population consistently moved more slowly than the rat , even moving backwards while the animal remained still . Because of this discrepancy we could not reproduce the compression factor reported by Maurer et al . ( 2012 ) . Nevertheless , the independent coding model accurately reproduces the theta timescale activity reported by Maurer et al . ( 2012 ) . The above analysis has two important implications . First , both the decoded sequence path length and theta-compressed propagation speed in the independent coding model match experimental data provided the degree of theta modulation of spike output increases linearly with running speed . This dependence of phase locking on running speed is consistent with the observed increase in LFP theta amplitude ( McFarland et al . , 1975; Maurer et al . , 2005; Patel et al . , 2012 ) , and is a novel prediction made by our model . Second , since the temporal delays between cells are determined by the propagation speed vp , the close match of this quantity to experimental data confirms the dependence of temporal delays on running speed predicted by our model , and argues against models based on fixed delays ( Diba and Buzsáki , 2008; Geisler et al . , 2010 ) . Further experimental support for the notion of inter-assembly coordination has come from an analysis suggesting that single cell phase precession is less precise than observed theta sequences ( Foster and Wilson , 2007 ) . This conclusion relies on a shuffling procedure which preserves the statistics of single cell phase precession yet reduces intra-sequence correlations . However , performing the same shuffling analysis on data generated by our independent coding model also reduced sequence correlations ( t-test , p < 10−46 ) ( Figure 5—figure supplement 2 ) . The effect arises because the shuffling procedure does not preserve the temporal structure of single cell phase precession , despite preserving the phasic structure ( Figure 5—figure supplement 2A , B ) . Hence , the phase–position correlations are unaffected , while the time–position correlations and hence sequence correlations are disrupted ( Figure 5—figure supplement 2C , D ) . Thus , inter-assembly coordination is not required to account for these findings . Nevertheless , although these results are reproducible by the independent coding model , it remains possible that coordinated assemblies underly the observed theta sequences . In particular , it is unclear whether this shuffling procedure could be modified to obtain a test for assembly coordination with greater statistical specificity and if so , whether it would reveal assembly coordination within experimental datasets . To address these questions , we analyzed experimental data along with data generated by independent coding and coordinated assembly models , using a modified version of this shuffling procedure ( see ‘Materials and methods’ ) . We found that the new shuffling procedure successfully detected assembly coordination with a statistical power of 81% ( calculated for datasets containing the same number of sessions , cells , and sequences as our experimental dataset ) . When applied to experimental data from CA1 , the shuffling test failed to detect any significant effect of shuffling ( t-test , p = 0 . 28 , 2436 events ) , as in most ( 74% ) of the simulated independent coding datasets ( Figure 5—figure supplement 2E , F ) . This failure to detect evidence of assembly coordination gives further support to the independent coding hypothesis . In additional support for the coordinated assembly hypothesis , Dragoi and Buzsáki ( 2006 ) performed an analysis suggesting that , during continuous locomotion around a rectangular track , some cell pairs show a lap by lap covariation of firing rates ( termed the dependent pairs ) . These cell pairs were found to spike with a more reliable temporal lag within theta cycles than cell pairs whose firing rates are independent , which was interpreted as evidence for direct interactions between dependent neurons . To test whether these results are instead consistent with independent coding , we applied the analysis of lap by lap firing rate covariations to data from simulations of the independent coding model . We found a similar fraction of apparently dependent cell pairs to that reported by Dragoi and Buzsáki ( 2006 ) , despite the absence of any true dependencies between cells in the model ( see ‘Materials and methods’ ) . Hence , this analysis artificially separates homogeneous populations of place cells into apparently dependent and independent cell pairs . Moreover , these dependent and independent cell groups displayed different spatial distributions of place fields , with dependent cell pairs generally occuring closer together on the track ( Wilcoxon rank sum test , p = 1 . 8 × 10−16 ) . By separating a homogeneous population of cells into dependent and independent groups , the analysis therefore introduces a sampling bias , leading to dependent cell pairs having different properties . While we were unable to reproduce the analysis of the temporal lags in each group due to a lack of information provided within the original study ( see ‘Materials and methods’ ) , the emergence of dependent cell pairs with measurably different properties in independent coding simulations nevertheless demonstrates that these results are not indicative of interactions between neurons . Finally , precise coordination of theta sequences has been suggested on the basis that theta sequence properties vary according to environmental features such as landmarks and behavioral factors such as acceleration , with sequences sometimes representing locations further ahead or behind the animal ( Gupta et al . , 2012 ) . To establish whether independent coding could also account for these results , we generated data from our model and applied the sequence identification and decoding analysis reported by Gupta et al . ( 2012 ) . We found that , even for simulated data based on pure rate coding with no theta modulation ( k = 0 ) , large numbers of significant sequences were detected at high running speeds ( Figure 6A ) . Therefore , to test the performance of the full sequence detection and Bayesian decoding protocol used by ( Gupta et al . , 2012 ) , we analyzed two simulated datasets—one with a realistic value of phase locking ( k = 0 . 5 , Figure 6B–D , solid lines ) and another with zero phase locking ( i . e . , no theta related activity , Figure 6B–D , dashed lines ) . In both cases , applying the reported Bayesian decoding analysis yielded similar decoded path lengths to those found experimentally ( Figure 6C , D ) . Importantly , there was an inverse relationship between the ahead and behind lengths decoded from the simulated data , which reproduces the apparent shift in sequences ahead or behind the animal observed in experimental data ( cf . Figure 4c of Gupta et al . ( 2012 ) ) . This effect was dependent on the density of recorded place fields on the track and the threshold for the minimum number of cells in a theta cycle required for sequence selection ( Figure 6—figure supplement 1 ) . As these results were obtained both in the case with realistic phase coding and in the case with only rate coding ( and therefore no theta sequences ) , the properties of the decoded trajectories are not related to theta activity within the population . Hence , these data do not constrain models of theta activity in CA1 . 10 . 7554/eLife . 03542 . 016Figure 6 . Analysis of individual sequence statistics . ( A ) The fraction of theta cycles which are classified as ‘significant sequences’ according to the Gupta et al . ( 2012 ) analysis , as a function of running speed and phase locking ( for simulated data generated under the independent coding model ) . Large fractions of significant sequences are generated even without phase coding or theta sequences within the population ( i . e . , at k = 0 ) . The black line shows the fraction reported experimentally . ( B ) The distribution of significant sequences over running speed and decoded path length for simulated data with phase locking k = 0 . 5 , as calculated by Gupta et al . ( 2012 ) ( cf . their Figure 1c ) . ( C ) The relationship between decoded path length and decoded ahead and behind lengths for significant sequences , calculated for a dataset with no theta activity ( k = 0 ) and a dataset with realistic theta activity ( k = 0 . 5 ) . ( D ) The relationship between the ahead length of the sequence and the behind length of the sequence for these two datasets . Note that the properties of the decoded trajectories do not depend on the theta activity in the data . This replicates the experimental data ( cf . Figure 4a-c of Gupta et al . ( 2012 ) ) , showing that similar trajectories are decoded by this algorithm regardless of the presence of theta sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 01610 . 7554/eLife . 03542 . 017Figure 6—figure supplement 1 . Dependence of decoded trajectories on the number of cells in a sequence . ( A–C ) Distributions of the number of cells which spike in a theta cycle , for simulations of the independent coding model with different densities of place fields on the track ( i . e . , different numbers of place fields on a track of fixed length ) . ( A ) The cell density used to reproduce the results of Gupta et al . ( 2012 ) . ( B and C ) Simulations with higher place field densities in which more active cells are recorded in each theta cycle on average . ( D–F ) Relationship between decoded ahead and behind length , calculated as in Gupta et al . ( 2012 ) , shown for simulations with different place field densities and for different thresholds of the minimum number of cells required for a sequence to be included for analysis . ( D ) Simulations with 12 cells on the track and a threshold of three cells generate results similar to Gupta et al . ( 2012 ) . ( E–F ) The density of place fields on the track and the threshold for sequence selection affect the decoded trajectories , with higher values for either resulting in a smaller change in behind length as a function of ahead length . ( G–H ) Spearman's rank correlation between ahead length and behind length for different place field densities plotted as a function of the threshold for the minimum number active of cells . Although the magnitude of the effect shown in ( D–F ) is diminished as these quantities increase , the correlation between ahead and behind length stays constant . Moreover , this correlation remains significant despite the decreasing effect size . Only when the number of selected sequences becomes too low to maintain a reliable measure does the effect become insignificant . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 017 In total , our analysis demonstrates that a traveling wave model based on independent phase coding for CA1 theta states is consistent with existing experimental data . Thus , neither intra- nor inter-assembly interactions are required to explain spike sequences observed in CA1 during theta states . Our analyses of experimental data along with simulations from each hypothesis render it unlikely that assembly coordination significantly shapes the structure of theta sequences or CA1 cell assemblies . Below , we investigate some functional consequences of the independent coding and coordinated assembly hypotheses and show that , despite the advantage of assembly coordination in generating robust sequential activity patterns , it suffers from severe limitations in remapping and storage of multiple spatial maps . Independent coding offers a solution to this problem , allowing flexible generation of sequential activity over multiple spatial representations . What are the advantages of independent coding compared to sequence generation through interactions between cell assemblies ? When an animal is moved between environments , the relative locations at which place cells in CA1 fire remap independently of one another ( e . g . , O'Keefe and Conway , 1978; Wilson and McNaughton , 1993 ) . This global remapping of spatial representations poses a challenge for generation of theta sequences through coordinated assemblies as synaptic interactions that promote formation of sequences in one environment would be expected to interfere with sequences in a second environment . Indeed , in the coordinated assembly model , simulations of remapping reduced single cell phase precession to below the level of independent cells ( i . e . , of an identical simulation with interactions between cells removed ) . Remapping in the coordinated coding model also substantially reduced firing rate and population oscillations ( Figure 7—figure supplement 1 ) . This decrease in firing rate following remapping contradicts experimental data showing an increase in firing rate in novel environments ( Karlsson and Frank , 2008 ) . It is not immediately clear whether the independent coding model faces similar constraints on sequence generation across different spatial representations . We therefore addressed the feasibility of maintaining theta sequences following remapping given the assumptions that underpin our independent coding model . We first consider the possibility that following remapping the phase lags between cell pairs remain fixed—that is , while two cells may be assigned new firing rate fields , their relative spike timing within a theta cycle does not change . This scenario would occur if the phase lags associated with phase precession were generated by intrinsic network architectures ( e . g . , Diba and Buzsáki , 2008; Geisler et al . , 2010; Dragoi and Tonegawa , 2011 , 2013 ) or upstream pacemaker inputs . For fixed phase lags , place cells display linear phase coding , whereby a cell continues to precess in phase outside of its rate coded firing field at a constant rate ( Figure 7A ) . In this scenario , the phase lag between two neurons depends linearly on the distance between their place field centers , while cells separated by multiples of a place field width share the same phase ( Figure 7A ) . Each cell pair therefore has a fixed phase lag in all environments and all cells can in principle be mapped onto a single chart describing their phase ordering ( Figure 7A ) . If this mechanism for determining phase ordering is hardwired , then following arbitrary global remapping , cells with nearby place field locations will in most cases no longer share similar phases ( Figure 7B ) . As a result , theta sequences and the global population theta will in general be abolished ( Figure 7B ) . However , there exist a limited set of remappings which in this scenario do not disrupt the sequential structure of the population ( e . g . , Figure 7C ) . On a linear track , these remappings are: translation of all place fields by a fixed amount , scaling of all place fields by a fixed amount , and permuting the place field locations of any cell pair with zero phase lag . 10 . 7554/eLife . 03542 . 018Figure 7 . Properties of CA1 populations governed by linear phase coding . ( A ) On a linear track , cells which precess linearly in phase maintain fixed theta phase lags . This is illustrated as a phase ordering ( colored bar ) , which describes the relative phase of each cell ( arrows show locations of cells at each phase ) . Each cell has a constant , running speed dependent frequency and a fixed phase offset to each other cell . ( B ) A complete global remapping with phase lags between cells held fixed . Theta sequences and population oscillations are abolished . ( C ) In a constrained place field remapping , theta sequences are preserved . ( D ) In open environments , phase lags depend on running direction . The set of population phase lag configurations needed to generate sequences in each direction is called a phase chart . ( E ) If a population has a fixed phase chart , the possible remappings are restricted to affine transformations . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 01810 . 7554/eLife . 03542 . 019Figure 7—figure supplement 1 . Remapping with coordinated assemblies . ( A ) Comparison of single cell phase precession generated by coordinated assemblies ( before and after remapping ) and independent coding . For this simulation , single cell phase and rate fields were assumed to be perfectly remapped , so that any changes are purely due to assembly interactions . Note that , while assembly interactions improve phase coding in single cells in the initial environment , after remapping these same interactions disrupt phase precession and cause a lower ( circular-linear ) correlation between spike phase and animal location than that generated by independent cells . ( B ) Population firing rate on a single trial along a linear track . While assembly interactions initially entrain and amplify theta oscillations in the population compared to independent cells , after remapping these interactions disrupt theta activity and cause a lower overall activity level . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 019 When considering global remapping in an open environment similar constraints apply . Because the phase lag between any two cells depends on running direction ( e . g . , Huxter et al . , 2008 ) , the population phase ordering must always be aligned with the direction of movement ( Figure 7D ) . Hence , in open environments , the notion of a phase chart must be extended to include a fixed phase ordering for each running direction . Given such a fixed phase chart , a set of remappings known as affine transformations preserve the correct theta dynamics ( see Supplementary file 1 , Appendix: A7 ) . Such remappings consist of combinations of linear transformations ( scaling , shear , rotation , and reflection ) and translations ( Figure 7E ) . Remappings based on permutation of place field locations of synchronous cells , which are permissible in one dimensional environments , are no longer tenable in the two dimensional case due to constraints over each running direction . Since place cell ensembles support statistically complete ( i . e . , non-affine ) remappings ( e . g . , O'Keefe and Conway , 1978 ) while maintaining phase precession , CA1 network dynamics are not consistent with the model outlined above . Moreover , this analysis demonstrates that previous models based on fixed temporal delays between cells ( e . g . , Diba and Buzsáki , 2008; Geisler et al . , 2010 ) cannot maintain theta sequences following global remapping . Nevertheless , it remains possible that CA1 theta dynamics are based on fixed phase charts , provided that multiple such phase charts are available to the network , similar to the multiple attractor charts which have been suggested to support remapping of firing rate ( Samsonovich and McNaughton , 1997 ) . In this case , each complete remapping recruits a different phase chart , fixing a new set of phase lags in the population . The number of possible global remappings that maintain theta sequences is then determined by the number of available phase charts . Such a possibility is consistent with recent suggestions of fixed sequential architectures ( Dragoi and Tonegawa , 2011 , 2013 ) and has not been ruled out in CA1 . It is also of interest that affine transformations are consistent with the observed remapping properties in grid modules ( Fyhn et al . , 2007 ) , suggesting that a single phase chart might be associated with each grid module . The above analysis demonstrates that both coordination of assemblies and independent , linear phase coding pose severe restrictions on global remapping which appear at odds with experimental observations . We asked if it is possible to overcome these constraints so that phase sequences can be flexibly generated across multiple environments . We reasoned that experimental data on phase precession only imply that phase precesses within a cell's firing field and need not constrain a cell's phase outside of its firing field . We therefore implemented a version of the independent coding model in which firing phase has a sigmoidal relationship with location ( Figure 8A–B , solid line; Supplementary file 1 , Appendix: A5 ) , such that phase precesses within the firing field but not outside of the field . In this case , each cell's intrinsic frequency increases as the animal enters the spatial firing field and drops back to LFP frequency when the animal exits the firing field ( Figure 8C , solid line ) . This is in contrast to the linear phase model and previous work with fixed delays ( Geisler et al . , 2010 ) in which each cell's intrinsic frequency is always faster than the population oscillation , both inside and outside of the place field ( Figure 8C , dashed line ) . In a given environment , spike phase precession and sequence generation in a population of cells with sigmoidal phase coding ( Figure 8D–F ) are similar to models in which cells have linear phase coding . However , in addition , sigmoidal phase coding enables theta sequences to be generated after any arbitrary global remapping ( Figure 8G ) . This flexible global remapping is in contrast to the scrambling of theta sequences following global remapping when cells have linear phase coding ( Figure 8G ) . Thus , independent sigmoidal coding is able to account for CA1 population activity before and after global remapping . 10 . 7554/eLife . 03542 . 020Figure 8 . Properties of CA1 populations governed by sigmoidal phase coding . ( A–C ) Firing rate and intracellular phase and frequency in the linear ( dashed lines ) and sigmoidal models ( solid lines ) during the crossing of a place field . In the sigmoidal model , phase precession is initiated inside the place field by an elevation of intracellular frequency from baseline . ( D–F ) Firing rate and intracellular phase and frequency for a place cell population on a linear track . In the sigmoidal model , an intracellular theta phase lag between cell pairs develops as the animal moves through their place fields . Outside their place fields , cell pairs are synchronized . ( G ) Global remapping in the linear and sigmoidal models . The sigmoidal model allows arbitrary remapping without disrupting population sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 020 Linear and sigmoidal models of phase coding lead to distinct experimentally testable predictions . Recordings of the membrane potential of CA1 neurons in behaving animals show that although spikes precess against the LFP , they always occur around the peak of a cell's intrinsic membrane potential oscillation ( MPO ) ( Harvey et al . , 2009 ) . Therefore the intrinsic phase ϕ of each cell in our model ( Figure 2D , E ) can be interpreted as MPO phase . While data concerning the MPO phase outside of the firing field are limited , such data would likely distinguish generation of theta sequences based on a linear and sigmoidal phase coding . If CA1 implements linear phase coding , then the MPO of each cell should precess linearly in time against LFP theta at a fixed ( velocity dependent ) frequency , both when the animal is inside the place field and when the animal is at locations where the cell is silent ( Figure 8A–C , dashed line ) . Alternatively , sigmoidal phase coding predicts that precession of the MPO against the LFP occurs only inside the firing rate field ( Figure 8A , B , solid line ) and that the MPO drops back to the LFP frequency outside of the place field ( Figure 8C , solid line ) as reported by Harvey et al . ( 2009 ) . A further prediction of sigmoidal coding is that , in contrast to models based on fixed delays ( Diba and Buzsáki , 2008; Geisler et al . , 2010 ) , the phase lag between any two cells changes when the animal moves through their place fields . Outside their place fields the cells are synchronized with each other and with the LFP , whereas a dynamically shifting phase lag develops as the animal crosses the place fields ( Video 2 ) . Finally , phase precession under the sigmoidal model behaves differently to the linear model in open environments . In the linear model , the phase chart fixes a different population phase ordering for each running direction , so that spike phase depends on the location of the animal and the instantaneous direction of movement . In the sigmoidal model , however , each cell has a location dependent frequency , so that the spike phase depends on the complete trajectory through the place field and no explicit directional information is required ( see Supplementary file 1 , Appendix: A6 ) . Rather , the directional property of the sequence arises purely through a location dependent oscillation frequency in each cell combined with the trajectory of the animal through each place field . In summary , our analysis demonstrates how evaluation of theta sequences following global remapping and of theta phase within and outside of a cell's firing field will be critical for distinguishing models of CA1 assemblies and theta generation . 10 . 7554/eLife . 03542 . 021Video 2 . Population dynamics with sigmoidal phase coding . Top: Distribution of the rate ( black ) and phasic ( red ) tuning curves for a population of sigmoidal phase coding place cells during constant speed locomotion on a linear track . Bottom: The evolution of the overall firing rate distribution in the population . Again , the population firing rate undergoes oscillations at LFP theta frequency and the center of mass of the population activity shifts from behind the animal to ahead of the animal in each theta cycle . However , in this case cells with place field centers distant from the animal's current location are synchronized with zero phase lag . DOI: http://dx . doi . org/10 . 7554/eLife . 03542 . 021 Our analysis demonstrates how complex and highly structured population sequences can be generated without coordination between neurons . In contrast to previous suggestions ( Harris et al . , 2003; Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Maurer et al . , 2012; Gupta et al . , 2012 ) , we find that the theta-scale population activity observed in CA1 is consistent with phase precession in independent cells , without interactions within or between cell assemblies . We demonstrate that independent coding enables flexible remapping of CA1 population activity while maintaining the ability to generate theta sequences . These properties are consistent with maximization of the capacity of CA1 for representation of distinct spatial experiences . The independent coding hypothesis leads to a novel view of the CA1 population as a fast moving traveling wave with a slower modulatory envelope . This model implements an invariant phase code via a change in the frequency and temporal delay between cells with running speed . Amplitude modulation of the envelope provides a mechanism for multiplexing spatial with nonspatial information , such as task specific memory items ( Wood et al . , 2000 ) and sensory inputs ( Rennó-Costa et al . , 2010 ) . The independence of each neuron naturally explains the robustness of phase precession against intrahippocampal perturbations ( Zugaro et al . , 2005 ) , an observation which is difficult to reconcile with models based on assembly interactions . Depending on the exact nature of the single cell phase code , independent phase coding can enable theta sequences to be maintained with arbitrary global remapping . This flexibility may maximize the number and diversity of spatial representations that CA1 can provide to downstream structures , offering a strong functional advantage over mechanisms based on interactions between cell assemblies . Independent phase coding leads to new and experimentally testable predictions that distinguish mechanisms of CA1 function during theta states . First , an absence of coordination within or between assemblies has the advantage that neural interactions do not interfere with sequence generation after global remapping . Rather , for independent coding models the constraints on sequence generation following remapping arise from the nature of the phase code . With linear phase coding the set of sequences available to the network is fixed , resulting in a limited set of place field configurations with a particular mathematical structure ( Figure 7 ) . Interestingly , the remappings observed in grid modules ( Fyhn et al . , 2007 ) , but not CA1 , are consistent with those predicted for networks with a single fixed set of theta phase lags called a phase chart . These findings , together with the fact that the temporal delays between cells depend on running speed , argue against previous models based on fixed delays within CA1 populations ( Diba and Buzsáki , 2008; Geisler et al . , 2010 ) . Nevertheless , more complex scenarios with multiple phase charts could explain CA1 population activity during theta oscillations and ‘preplay’ , which suggests a limited remapping capacity for CA1 ( Dragoi and Tonegawa , 2011 , 2013 ) . Alternatively , sigmoidal phase coding massively increases the flexibility for global remapping as cells can remap arbitrarily while maintaining coherent theta sequences within each spatial representation ( Figure 8 ) . Second , linear and sigmoidal phase coding predict distinct MPO dynamics . With linear phase coding , the temporal frequency of each MPO is independent of the animal's location . With sigmoidal phase coding , the MPO frequency increases inside the place field , so that phase precession occurs inside but not outside the place field . In this case , only the spiking assembly behaves as a traveling wave , whereas the MPOs of cells with place fields distant from the animal are phase locked to the LFP . Sigmoidal phase precession could emerge due to inputs from upstream structures ( Chance , 2012 ) or be generated intrinsically in CA1 place cells ( Leung , 2011 ) . Finally , in contrast to linear phase coding populations , sigmoidal phase coding populations do not require additional information from head direction or velocity cells to generate directed theta sequences in open environments . Instead , sigmoidal theta sequences are determined solely by the recent trajectory of the rat through the set of place fields together with a location dependent oscillation frequency , consistent with recent observations of reversed theta sequences during backwards travel ( Cei et al . , 2014; Maurer et al . , 2014 ) . In summary therefore , these two models may be distinguished experimentally on the basis of observations of the number of non-affine remappings in CA1 , the intracellular frequency and delay between place cells as a function of location and of the dependence of firing phase on the trajectory through a place field in open environments . While theta sequences of CA1 activity are most commonly observed during spatial navigation , similar activity patterns associated with short term memory have been observed during wheel running ( Pastalkova et al . , 2008 ) . In this situation each cell's activity depends on the phase of the LFP theta rhythm and on the temporal location within an ‘episode field’ rather than a place field . Our model can be applied equally well to these internally generated sequences if the rate coded episode field is assumed to have a similar temporal structure to a place field . An entirely different class of sequences , however , is observed during non-theta states such as sharp wave ripples ( SWR ) ( Buzsáki et al . , 1992; Diba and Buzsáki , 2007 ) . In contrast to theta sequences , SWR sequences are generally observed during states of immobility and are believed to arise from synchronous discharge in CA3 ( Buzsáki et al . , 1983 ) . Because SWR sequences are generated without co-occurence of longer time-scale firing fields or theta oscillations , they cannot be accounted for by the independent coding schemes that we investigate here , in which rate and phase information determine the activity of each cell . Instead , the nature of cell assemblies in CA1 may be highly state dependent , operating in at least two modes . During theta states , sequences are generated by independently precessing neurons , whereas during SWRs sequences may result from interactions between consecutively activated cell assemblies . Can independent coding account for manipulations that modify place cell dynamics ? Administration of cannabinoids disrupts phase precession by CA1 neurons and impairs spatial memory , but does not appear to affect the rate coded place firing fields of CA1 neurons ( Robbe and Buzsáki , 2009 ) . This dissociation between rate and phase coding can be accounted for in our model by assuming that rate fields are maintained while phase fields are disrupted ( Figure 2A ) or the degree of phase locking ( k ) is substantially reduced ( Figure 2B ) . In contrast , increased in-field firing of place cells following optogenetic inactivation of hippocampal interneurons ( Royer et al . , 2012 ) can be accounted for in our model by increased Nspikes , while altered phase of place cell firing following inactivation of parvalbumin interneurons can be accounted for in our model by modifying the phase fields ( Figure 2A ) of the place cells . Important future tests of the independent coding model will include comparison of its predictions of sequence activity , remapping and intracellular dynamics to experimental measures made during these kinds of manipulations . Our independent coding model offers a comprehensive account of population activity in CA1 during theta states and makes new predictions for coordination of network dynamics and remapping at the population level , but it does not aim to distinguish cellular mechanisms for phase precession . Nevertheless , by demonstrating that existing observations of population sequences can be explained by independent coding our model argues against mechanisms for phase precession that rely on synaptic coordination at theta time scales ( e . g . , Tsodyks et al . , 1996; Maurer and McNaughton , 2007; Lisman and Redish , 2009 ) . In contrast , our model does not distinguish between specific single cell mechanisms for phase precession such as dual oscillators ( Lengyel et al . , 2003; Burgess et al . , 2007 ) , depolarizing ramps ( Mehta et al . , 2002 ) , intrinsic membrane currents ( Leung , 2011 ) or dual inputs from CA3 and entorhinal cortex ( Chance , 2012 ) . Our model is also consistent with inheritance of phase precession in CA1 from upstream circuits in CA3 and entorhinal cortex ( Jaramillo et al . , 2014 ) . However , it argues against the possibility that CA1 inherits coordinated sequences from CA3 ( Jaramillo et al . , 2014 ) . It is possible that CA3 nevertheless generates sequences by synaptic coordination . Because CA3 neurons are connected by dense recurrent collaterals ( Miles and Wong , 1986; Le Duigou et al . , 2014 ) , there are likely to be substantial correlations in their output to CA1 , which could induce deviations from the independent population code outlined here . However , feedback inhibition motifs such as those found in CA1 may counteract such correlations ( Renart et al . , 2010; Tetzlaff et al . , 2012; Bernacchia and Wang , 2013; King et al . , 2013; Sippy and Yuste , 2013 ) . Hence , the local inhibitory circuitry in CA1 may actively remove correlations present in its input in order to generate an independent population code ( Ecker et al . , 2010 ) . A major advantage of independently precessing cell populations is that they provide a highly readable , robust , and information rich code for working and episodic memory in downstream neocortex . In particular , a downstream decoder with access to an independent population code need only extract the stereotyped correlational patterns associated with traveling waves under a given place field mapping . In this way it can flexibly decode activity across a large number of spatial representations . Decoding in the presence of additional correlations would likely lead to a loss of information ( Zohary et al . , 1994 ) . While this loss can to some extent be limited by including knowledge of these additional correlations ( Nirenberg and Latham , 2003; Eyherabide and Samengo , 2013 ) , this likely requires a high level of specificity and therefore a lack of flexibility in the decoder . The flexibility afforded by an independent population code may therefore provide an optimal format for the representation and storage of the vast number of spatial experiences and associations required to inform decision making and guide behavior . In the independent coding model , we simulated data from a population of place cells with place field centers xc and width σ which precess linearly through a phase range of Δϕ over a distance 2R on a linear track using Equation ( A3 . 6 ) in Supplementary File 1 . The initial phase ψs was either taken as 0 , or a uniform random variable ψs ∈ [0 , 2π] set at the beginning of each run . In all simulations , parameters were set as: 2R = 37 . 5 cm ( Maurer et al . , 2006 ) , Δϕ = 2π , σ = 9 cm , fθ = 8 Hz , Nspikes = 15 . Finite numbers of place cells were simulated with place field centers xc which were either uniformly distributed along a linear track with equal spacing or randomly sampled from a uniform distribution over the track . All cells were therefore identical up to a shift in place field center xc . Simulations were performed using Matlab 2010b and 2013b . Simulations of population activity generated through coordinated assemblies used equations ( A4 . 1–4 . 5 ) in Supplementary File 1 , with the single cell properties simulated as for the independent coding model . The peer interaction timescale was set to τ = 25 ms , and the interaction length for asymmetric excitation was set to ℓ = 10 cm with an excitatory amplitude of wE = 1/4 . The amplitude of the inhibitory weights was varied until the same number of spikes were generated as in the independent coding simulation ( for the parameters used in these simulations , the inhibitory amplitude was wI = 1/18 ) . We used data recorded from CA1 during navigation along a linear track . For details of experimental data see Mizuseki et al . ( 2014 ) . For the analysis performed in this study , simultaneous recordings of a large number of coactive cells in CA1 are required , which restricted the number of suitable datasets . In particular , we used datasets ec016 . 233 , ec016 . 234 , ec016 . 269 , ec014 . 468 , ec014 . 639 . To replicate the results of Harris et al . ( 2003 ) , we simulated constant speed movement along a linear track , with lap by lap running speeds drawn from a normal distribution with mean 35 cm/s and standard deviation of 15 cm/s . We simulated motion in each direction , using the same set of place fields in each case . We estimated the preferred firing phase at each location from the simulated data using the methods stated in Harris et al . ( 2003 ) , using either single-direction data or data consisting of runs in both directions to generate nondirectional or directional phase fields . The prediction analysis was performed according to the methods given in Harris et al . ( 2003 ) . For these initial simulations ( Figure 4 ) , we used the simulated value of phase locking rather than estimating it from the data . To display the peer prediction performance shown in Figure 4C , the optimal prediction timescale for each phase locking value was chosen . This was done separately for the peer only case and the peer plus phase field case . We then performed additional , more detailed simulations to test the performance of simulated and experimental data when using the new directional phase fields . We separated datasets according to the running direction along a linear track , analyzing each direction individually . In addition to fitting the place field , phase field , and peer factor used by Harris et al . ( 2003 ) , we also fitted a velocity modulation factor given by: ( 5 ) A ( v ) =∑tntw ( \|v−vt\| ) ∑tr0 ( xt ) dtw ( \|v−vt\| ) , which estimates the changes in firing rate of a cell according to running speed . In the above equation , the notation follows that of Harris et al . ( 2003 ) ( their Supplementary Information ) , that is , w is a Gaussian smoothing kernel of width 3 . 5 cm/s , nt is the number of spikes fired by the cell in time bin t , r0 is the estimated firing rate field at location x , xt is the animal's location in time bin t , and vt is its velocity . Our simulations showed that , using the methods of Harris et al . ( 2003 ) , the phase locking parameter k was underestimated outside of the place field center . Misestimation of phase field parameters introduces false peer predictability in simulated datasets . We therefore replaced their location dependent estimation with a fixed value equal to the phase locking estimated in regions where the place field is over 2/3 its maximum value . We also found that the 5 cm spatial smoothing kernel used by Harris et al . ( 2003 ) resulted in a high level of spurious peer prediction in simulations based on independent coding , since it extended the boundaries of place fields , allowing non-overlapping peer cells to compensate via inhibitory weights . A smaller kernel of 3 . 5 cm reduced the rate of false positive for peer prediction and was therefore used instead . We simulated 300 cells in each session of which we randomly sampled 15 for analysis in order to match the number of place cells typically recorded experimentally . 28 laps were simulated for each session and 10 sessions were simulated in total ( representing the two running directions over the five experimental sessions we analyzed ) . Peer prediction was performed at a timescale of 25 ms ( the optimal timescale in Harris et al . ( 2003 ) ) . To compare the sequence path length in spiking data generated from the independent coding model to experimental data , we followed the decoding methods outlined in Maurer et al . ( 2012 ) . Briefly , this involves constructing trial averaged time by space population activity matrices in order to decode the location represented by the population in each time bin over an average theta cycle . The decoded path length is measured as the largest distance between decoded locations within the theta cycle . To test the influence of phase locking in this analysis , k was varied incrementally from 0 to 6 and the sequence path length for the resulting data was calculated in each case . We used the same spatial and temporal bins ( 0 . 7 cm and 20° of LFP θ ) as the original study . To calculate the fast and slow slopes , we generated the contour density plots described by Maurer et al . ( 2012 ) using the same parameters as the sequence path length analysis . We simulated 100 trials for each running speed . We then divided these 100 trials into 10 subsets of 10 and applied the contour analysis to each subset . We fitted the fast slope to the 95% contour of the central theta peak , and measured the slow slope as the line joining the maximum of the top and bottom peaks of the central 3 . We averaged over the results from each subset to obtain the final value . The analytical value for the fast slope in the limit of high phase locking is FS = vp/ ( 360fθ ) , where the denominator arises due to the normalization to cm/deg in the analysis of Maurer et al . ( 2012 ) . Similarly for zero phase locking , FS = v/ ( 360fθ ) . The analytical value for the slow slope is independent of phase locking , SS = v/ ( 360fθ ) . Upper and lower bounds for the slow slope were therefore fitted assuming the reported running speed is accurate , and that the LFP theta frequency is in the range 4 Hz < fθ < 12 Hz . To reproduce the results of Foster and Wilson ( 2007 ) , we generated data from 1000 theta cycles , each with a running speed drawn from the same distribution as for the prediction analysis . Following the protocol outlined by Foster and Wilson ( 2007 ) , we found the set of all spike phases for each cell when the rat was at each position and analyzed events defined as 40 ms windows around firing rate peaks . Spike phases were calculated by interpolation between LFP theta peaks . For the shuffling analysis , each spike in an event was replaced by another spike taken from the same cell while the animal was at the same location . The new spike time was then calculated from its phase by interpolation between the closest two LFP theta troughs of the original spike , as reported in the original study . 100 such shuffles were performed for each event , and the correlation between cell rank order and spike times was calculated in each case . For the corrected shuffling procedure , we followed the methods of Foster and Wilson ( 2007 ) but with the following adjustments: the correlations between spike times and place field rank order within an event calculated in the original study were replaced with circular-linear correlations between spike phase and place field peaks in order to remove issues arising from conversion between spike time and spike phase ( Kempter et al . , 2012 ) ; a minimum running speed of 20 cm/s and a maximum running speed of 100 cm/s were imposed; the LFP phase was measured using a Hilbert transform rather than a linear interpolation between theta peaks; spikes that occured more than 50 cm away from the place field peak were discarded from the analysis . The circular-linear correlation requires a mild restriction of the range of possible regression slopes between the circular and linear variables , which in this case describes the distance traveled by a theta sequence within a theta cycle ( Kempter et al . , 2012 ) . We set this range as 25–80 cm/cycle , that is , around the size of a place field . For simulations using this shuffling procedure , we simulated 300 cells in each session on a linear track and randomly sampled 15 of these for further analysis . We again simulated 10 sessions with 28 laps each , for which the number of detected events was similar to that of the experimental dataset . We generated a large number of such datasets in order to obtain a distribution of shuffling test results to compare against the experimental dataset . To reproduce the results of Dragoi and Buzsáki ( 2006 ) , we simulated population activity on a linear track . To recreate the experimental conditions of Dragoi and Buzsáki ( 2006 ) , we set the track length as 250 cm and simulated 8 sessions ( i . e . , four animals by two running directions ) , each with 25 place cells . As the original experiment consisted of continuous locomotion around a rectangular track , we wrapped the boundaries of the linear track and simulated continuous sessions rather than single laps . Place fields were randomly distributed over the track following a uniform distribution . Running speed on each lap was drawn from the same distribution as the prediction and shuffling analyses . Phase locking was set to 0 . 5 . We calculated the dependent and independent cell pairs following the methods of Dragoi and Buzsáki ( 2006 ) , which uses temporal bins of 2 s to calculate firing rate correlations and a shuffling procedure to find significantly correlated cells . Dragoi and Buzsáki ( 2006 ) did not state the number of dependent and independent cell pairs obtained from their analysis . Therefore , to compare the results of our simulations to their experimental data , we estimated the number of points in their CCG-lag plot for dependent and independent cell pairs ( their Figure 3B ) and compared the result to the same measure in our simulations . CCG plots were calculated using the methods described in Dragoi and Buzsáki ( 2006 ) . Using this method , we found that 33% of cell pairs were dependent compared to an estimated 30–35% in Dragoi and Buzsáki ( 2006 ) . To calculate the reliability of temporal lags between dependent and independent pairs , Dragoi and Buzsáki ( 2006 ) took the central cloud of the CCG-lag vs place field distance scatter plot ( their Figure 2A ) and calculated the correlation between these two variables . However , the method for isolating the central cloud from the surrounding clusters was not disclosed . Without this information , we were unable to reproduce this analysis . To test for differences between place field separations of dependent and independent cell pairs , we again considered only cell pairs whose CCG lags passed the inclusion criteria ( as described in Dragoi and Buzsáki ( 2006 ) ) . We compared the vectors of cell pair separations for each group . To reproduce the results of Gupta et al . ( 2012 ) , we used the significant sequence testing protocol and Bayesian decoding algorithm described therein , with spatial binning set as 3 . 5 cm , as in the original study . Briefly , the significant sequence testing analysis tests if population activity within a theta cycle has significant sequential structure , whereas the Bayesian decoding algorithm generates a time by space probability distribution which is used to decode the ahead and behind lengths represented by the theta sequence . For Figure 6A , we varied phase locking and running speed independently and generated spiking data for each pair of values . In the simulations used to generate Figure 6 , we assumed that the number of spikes fired per theta cycle does not vary with running speed , as such a dependence introduces an additional change of the decoded sequence path length with running speed . In order to best match the fraction of theta cycles with three or more cells active reported by Gupta et al . ( 2012 ) , each simulated theta cycle contained 12 place cells with place fields randomly distributed over a region of space 94 . 5 cm ahead or behind the rat . We then applied the significant sequence detection methods for each resulting data set to obtain the fraction of significant sequences in each case . For Figure 6B , we used k = 0 . 5 and generated 1000000 theta cycles , each with a running speed drawn from a normal distribution with mean 30 cm/s and standard deviation 10 cm/s . Running speeds less than 10 cm/s were discarded and the remaining theta cycles were tested for significant sequential structure . For Figure 6C , D , we applied the Bayesian decoding algorithm to these significant sequences in order to calculate the path length , ahead length , and behind length . In addition , we applied the same analysis to another dataset simulated with k = 0 . To simulate remapping in the coordinated assembly model , we simulated spiking activity for a population of 300 cells on a linear track with weights as described in Supplementary file 1 , Appendix: A4 . To simulate the remapped population , we left this set of weights intact but randomly reassigned the place and phase fields of each cell , such that phase coding and rate coding were perfectly remapped but peer interactions were preserved between environments . To simulate remapping in the linear phase coding model , we assumed that phase lags were preserved between environments . The remapped population was simulated by randomly permuting the place field centers of cells while leaving the phase fields of each cell intact . To simulate remapping in the sigmoidal phase coding model , we assumed that the field of elevated frequency is locked to the place field before and after remapping . Hence , place fields were randomly permuted and the single cell frequency was defined to increase within the new place field .	When we explore a new place , we naturally create a mental map of the location as we go . This mental map is stored in a region of the brain called the hippocampus , which contains cells called place cells . These cells can carry information about our past , present , and future location in the form of electrical signals . They connect to each other to form networks and it has been proposed that these connections can store the information needed for the mental maps . Real-time maps are represented in the information carried by the electrical signals themselves . A physical location is specified by the individual place cell that is activated , and by the timing of the electrical signal it produces relative to a ‘brain wave’ called the theta rhythm . Brain waves are patterns of electrical signals activated in sets of brain cells and the theta rhythm is produced in the hippocampus of an animal as it explores its surroundings . Previous experiments suggested that when a rat explores an area , several sets of brain cells in the hippocampus are activated in sequence within each cycle of the theta rhythm . As the rat moves forward , the sequence shifts to different sets of cells to reflect the upcoming locations ahead of the rat . It has been thought that these sequences are triggered by the individual connections between the place cells . Here , Chadwick et al . developed mathematical models of the electrical activity in the brains of rats as they explored . They used these models to analyze data from previous experiments and found that the sequences of electrical activity arise from the timing of each cell's activity relative to the theta rhythm , rather than from the connections between the cells . Chadwick et al . 's findings suggest that the mental map may be highly flexible , allowing vast numbers of distinct memories to be stored within the same network of place cells without interference . Future studies will involve investigating the role of brain waves in the forming new mental maps and creating new memories .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Independent theta phase coding accounts for CA1 population sequences and enables flexible remapping
When pathogens enter the host , sensing of environmental cues activates the expression of virulence genes . Opposite transition of pathogens from activating to non-activating conditions is poorly understood . Interestingly , variability in the expression of virulence genes upon infection enhances colonization . In order to systematically detect the role of phenotypic variability in enteropathogenic E . coli ( EPEC ) , an important human pathogen , both in virulence activating and non-activating conditions , we employed the ScanLag methodology . The analysis revealed a bimodal growth rate . Mathematical modeling combined with experimental analysis showed that this bimodality is mediated by a hysteretic memory-switch that results in the stable co-existence of non-virulent and hyper-virulent subpopulations , even after many generations of growth in non-activating conditions . We identified the per operon as the key component of the hysteretic switch . This unique hysteretic memory switch may result in persistent infection and enhanced host-to-host spreading . Bacterial populations spontaneously differentiate into distinct phenotypic states ( Avery , 2006; Dubnau and Losick , 2006 ) . This variability has been described as a bet-hedging strategy that results in subpopulations that will survive under unpredictable stress ( Fraser and Kaern , 2009 ) . It has also been suggested that phenotypic variability is a ‘division of labor’ strategy: essentially , the bacterial population diversifies in order to utilize nutrients more efficiently or to allow invasion and colonization of diverse niches in the host ( Ackermann et al . , 2008 ) . Diversification upon infection is also related to antigenic variation , which is a key strategy to eluding the acquired immune response of the host ( Kamada et al . , 2015; Stewart et al . , 2011 ) . The role of phase-variation mechanisms in phenotypic diversification through reversible genetic changes is well established ( see for example , Casadesús and Low , 2013; McClain et al . , 1991; Silverman and Simon , 1980 ) . Diversification processes , not linked to DNA alteration have been attributed to noise in gene expression that can be further amplified by regulatory motifs such as excitatory dynamics ( Süel et al . , 2006 ) positive feedback leading to bi-stability ( Ozbudak et al . , 2004 ) and ultrasensitivity ( Temme et al . , 2008; Rotem et al . , 2010; Levine et al . , 2012 ) . Interestingly , phenotypic diversification in microorganisms is frequently accompanied by growth rate variability . One striking example is that of bacterial persistence under antibiotic treatment ( Lewis , 2000 ) mediated by growth rate bimodality ( Balaban et al . , 2004; Brauner et al . , 2016; Helaine and Holden , 2013 ) . Pathogenic bacteria tightly regulate the expression of virulence machinery . Environmental conditions that are close to those in the host environment can induce the expression of the virulence genes ( ‘activating conditions’ ) ( Leverton and Kaper , 2005; Rosenshine et al . , 1996 ) . In contrast , ‘non-activating conditions’ include a broad spectrum of conditions that do not resemble the host environment . Variability in the expression of virulence genes is observed when the bacteria are exposed to activating conditions ( Nielsen et al . , 2010; De Angelis et al . , 2011; Somvanshi et al . , 2012; Atack et al . , 2015 ) . This process has been extensively studied in Salmonella ( Ackermann et al . , 2008; Temme et al . , 2008 ) . Salmonella employ a type III secretion system ( T3SS ) to inject the host cells with virulence factors . Interestingly , upon shifting from non-activating to activating conditions , Salmonella exhibit bimodal T3SS expression . The burden of T3SS expression , together with its bimodal expression , results also in growth rate bimodality ( Ackermann et al . , 2008; Diard et al . , 2013; Hautefort et al . , 2003; Sturm et al . , 2011 ) . The bimodality in the T3SS expression provides Salmonella with a fitness advantage in the host ( Diard et al . , 2013 ) , contributing also to antibiotic persistence ( Arnoldini et al . , 2014 ) , and to reduction in generation of non-virulent mutants termed ‘defectors’ ( Diard et al . , 2013 ) . The aim of this study was to examine whether phenotypic variability plays a role in the virulence of a model organism , enteropathogenic E . coli ( EPEC ) , a human specific pathogen , during infection and in the transition to non-activating conditions . The major virulence factors of EPEC are a T3SS , similar to that of Salmonella , and a type IV pili termed the bundle forming pili ( BFP ) ( Gaytán et al . , 2016; Hazen et al . , 2016 , 2015b ) . EPEC can cause symptoms ranging from asymptomatic infection to a devastating lethal disease in infants and spreads in the host population by the fecal-oral route ( Hazen et al . , 2016 ) . Whereas the phenotypic variability of virulence upon shifting from non-activating to activating conditions has been extensively studied ( e . g . Arnoldini et al . , 2014; Diard et al . , 2013; Sturm et al . , 2011 ) , the opposite process ( i . e . , the behavior of the pathogen population upon shifting from activating to non-activating conditions ) is poorly understood . In vivo , shifts from activating to non-activating conditions can occur transiently within the host and also in the process of host-to-host spread . Using our recently established ScanLag ( Levin-Reisman et al . , 2010 ) setup that can detect sub-populations lag time or growth rate variability by tracking single-colony appearance , we evaluated the phenotypic variability of growth of EPEC populations upon activation and in the transition from activating to non-activating conditions . Our analysis revealed a novel long-term hysteretic memory-switch in EPEC , which mediates bimodality in virulence expression . Whereas bimodal virulence expression is observed as a transient behavior in activating conditions , the transition from activating to non-activating conditions resulted in the stable co-existence of non-virulent bacteria and a hypervirulent subpopulation that continued to express full virulence even after many generation of growth in non-activating conditions . It is likely that this hysteretic switch is common in pathogenic bacteria , ensuring persistence of infection and improved host-to-host spreading . The expression of the EPEC virulence machinery is activated upon growth in Dulbecco's Modified Eagle's medium ( DMEM ) at 37°C to OD600 of 0 . 2–0 . 5 ( ‘activating conditions’ ) . In contrast , overnight growth in Luria-Bertani liquid medium ( LB ) is considered as ‘non-activating conditions’ ( Hazen et al . , 2015a; Leverton and Kaper , 2005; Puente et al . , 1996; Rosenshine et al . , 1996 ) . We searched for growth heterogeneity in EPEC upon transition from activating to non-activating conditions . As a negative control , the non-pathogenic E . coli K-12 was also evaluated . Cultures were grown under activating or non-activating conditions and plated on LB agar plates ( i . e . , non-activating conditions ) . Analysis of the colony growth dynamics , using ScanLag ( Levin-Reisman et al . , 2010 ) scanners , showed that EPEC and K-12 from the overnight LB cultures displayed unimodal distributions of colony appearance times ( Figure 1A , B ) . In contrast , colonies of EPEC that originated from activating culture conditions exhibited a bimodal distribution of appearance time ( Figure 1C , Figure 1—source data 1 ) ; the ‘activated’ K-12 culture maintained a unimodal distribution ( Figure 1D ) . Further analysis showed that the bimodality in EPEC colony appearance time was due to a slightly reduced growth rate of the bacteria in the late-appearing colonies ( Figure 1—figure supplement 1 ) . These differences in the growth rates resulted in bimodal colony size distribution at 17 hr post-plating ( Figure 1E–G ) . We refer to these two colony morphotypes as BIG and SMALL , for early- and late-appearing colonies , respectively ( Figure 1C , F ) . 10 . 7554/eLife . 19599 . 003Figure 1 . EPEC displays the bimodal colony size after virulence activation . Bacterial cultures of EPEC or E . coli K-12 were grown overnight in LB media ( Non-Activation ) or in DMEM for 3 hr at 37°C to OD600 ~0 . 3 ( Activation ) . Cultures were then diluted , plated on LB agar , and incubated at 32°C ( non-activating conditions ) . Colony appearance time was monitored by ScanLag at 15 min intervals . The resulting histograms show ( A , B , C , D ) the fractions of colonies detected at each time point and ( E , F ) colony size distributions 1000 min after plating . Colonies larger or smaller than 105 pixels were defined as ‘BIG’ and ‘SMALL’ morphotypes , respectively . ( A–C , E , F ) Experiments were repeated in at least four independent biological replicates . ( D ) shows the cumulative of 4 independent biological replicates . ( G ) Time-lapse microscopy phase-contrast images of the two co-existing morphotypes , BIG and SMALL . Time points are indicated . Similar results were obtained in at least 10 different locations and in two independent biological replicates . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 00310 . 7554/eLife . 19599 . 004Figure 1—source data 1 . This Source Data file contains appearance time histogram raw data for Figure 1A and C ( activated and non-activated EPEC cultures ) from ScanLag experiments . Data collected from four experiments . The data were collected by the ScanningManager software application and analyzed by TimeLapse analysis software application http://bio-site . phys . huji . ac . il/Materials . For figures histogram was fitted to total 100% bacteria . Data columns marked with * and ** are used for creating Figure 1A and C respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 00410 . 7554/eLife . 19599 . 005Figure 1—figure supplement 1 . EPEC growth on LB measured by ScanLag and time-lapse microscopy . Static cultures of EPEC were grown in DMEM for 3 hr at 37°C to OD600 ~0 . 3 ( activating conditions ) . Cultures were then diluted , plated on LB agar , and incubated at 32°C ( non-activating conditions ) . ( A–C ) Growth of colonies was monitored at 15 min intervals using the ScanLag system . ( A ) Histogram of the fraction of colonies detected at each time point . ( B ) Histogram of colony area growth time , i . e . the time taken to increase the colony area from 20 to 80 pixels . ( C ) Two-dimensional histogram of the data in ( A ) and ( B ) allows the visualization of the bimodal phenotype . Note the diagonal shift of the SMALL morphotype suggesting an increase in time to reach the given area and not an increased lag duration ( Levin-Reisman et al . , 2010 ) ( n = 1184 ) . ( A–C ) Experiments were repeated in at least four independent biological replicates . ( D ) Analysis of single-cell generation time by time-lapse phase microscopy . Bacteria from BIG and SMALL colonies were suspended in LB and were placed on LB agar pads and observed by microscopy . Data are presented as the median ± Median Absolute Deviation ( MAD ) of each morphotype . The p-value was calculated by Wilcoxon rank sum test , h = 1 . ( 2% of non-dividing cells were excluded from the analysis since their division was out of the observed time range ) . Similar results were obtained in two independent biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 005 To examine the inheritability of the SMALL and BIG phenotypes , we resuspended SMALL and BIG colonies and immediately re-plated the bacteria on LB agar ( Figure 2A ) . Most SMALL colony bacteria gave rise to SMALL colonies ( 96% ± 3%; mean ± s . d . ) ( Figure 2B , D ) , whereas the bacteria originated from BIG colonies gave rise to bimodal distribution in colony size with 79% ± 4% BIG and 21% ± 4% SMALL colonies ( Figure 2C , E ) . Remarkably , repeating this procedure for four consecutive cycles resulted in similar ratios of BIG to SMALL colonies ( Figure 2D , E ) . These findings suggested that the memory of the SMALL phenotype is inherited for tens of generations . 10 . 7554/eLife . 19599 . 006Figure 2 . Memory and reset of colony-size bimodality . ( A ) Scheme of the experimental procedure for measuring the stability of colony morphotypes: Colonies of EPEC grown on LB agar were picked 1000 min after plating , resuspended , re-plated on LB agar , and subjected to ScanLag analysis . ( B ) Histogram of the fraction of colonies detected at each time point for bacteria taken from a SMALL colony . ( C ) Histogram of the fraction of colonies detected at each time point for bacteria taken from a BIG colony . ( B–C ) Experiments were repeated in at least four independent biological replicates . ( D , E ) The same procedure was repeated for four consecutive cycles using bacteria taken from ( D ) SMALL or ( E ) BIG colonies , and in each cycle the fraction of BIG and SMALL colonies was determined . Data are presented as the means ± s . d . of five technical replicates . ( F ) Scheme of the experimental procedure for the ‘reset’ of the bimodality . ( G ) SMALL or ( H ) BIG colonies of EPEC were picked 1000 min after plating , resuspended in LB broth and grown overnight to stationary phase at 37°C . Cultures were then plated and subjected to ScanLag analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 006 To test whether the switching between BIG and SMALL morphotypes is mediated by DNA rearrangement , we extracted DNA from BIG and SMALL colonies and sequenced the genomes at high coverage ( Supplementary file 1 ) . The genome sequences of SMALL and BIG colonies were identical and very close to that of the published reference sequence of EPEC strain E2348/69 ( Iguchi et al . , 2009 ) . More advanced analysis , using a custom algorithm for the detection of DNA rearrangements typical to phase variation , identified three loci that were undergoing active phase variation by DNA inversion ( Goldberg et al . , 2014 ) . However , we did not detect any specific differences in these regions between the BIG and SMALL genomes . The only significant difference between the two genomes was an approximately two-fold lower coverage of pMAR2 ( EAF plasmid , Iguchi et al . , 2009 ) in the BIG variant genome relative to the SMALL genome ( Supplementary file 1 ) . These results suggest that genetic changes are not involved in the colony size bimodality , favoring the possibility that the morphotypes are produced through an epigenetic mechanism . We noticed a reduction in the SMALL morphotype inheritability when SMALL colonies were grown for more than 24 hr , before suspending and re-plating , suggesting that growth to stationary phase may affect the SMALL morphotype memory . To test this prediction , SMALL or BIG colonies were resuspended and grown in LB broth to stationary phase and then plated and analyzed . In both cases the SMALL morphotype disappeared ( Figure 2F–H ) , and the culture was ‘reset’ to form the unimodal colony-size distribution typical of that reported for stationary-phase cultures ( Figure 1A , Levin-Reisman et al . , 2010 ) . Taken together , these results show that the SMALL morphotype is extremely stable during growth , but disappears in stationary phase cultures . In order to characterize the switching process and measure the switching rates between the two morphotypes , we fitted the results of the SMALL and BIG colony fractions in non-activating conditions to a model based on switching between two phenotypes with different growth rates ( Balaban et al . , 2004 ) ( Figure 3A , Equations 1 and 2 , see Materials and methods-Mathematical model ) . BIG bacteria ( B ) have a higher growth rate and can switch to the SMALL morphotype with rate a , whereas SMALL bacteria ( S ) can switch to the BIG morphotype with rate b . The model reproduced our experimental observations and suggested that upon growth in non-activating conditions , the switching rate from BIG to SMALL is about 10 times higher than from SMALL to BIG , resulting in a stable co-existence of the two morphotypes . Furthermore , the time scale of the switching from SMALL to BIG was extremely long , requiring more than 100 hr and many generations to reach steady state ( see Materials and methods-Mathematical model ) . 10 . 7554/eLife . 19599 . 007Figure 3 . Model and measurements of bimodal switching rates . ( A ) Scheme and equations of a bimodal switching model . The two morphotypes , BIG and SMALL , are characterized by different growth rates , µB and µS , respectively , and different switching rates a and b . Note that these parameters depend on growth conditions . ( B ) Measurement and fit to the analytical solution of equations Equations 1 and 2 during exponential growth under activating conditions with initial conditions B ( t = 0 ) =1 , S ( t = 0 ) =0 , see Materials and methods-Mathematical model . Green and red lines are ScanLag measurements of the SMALL and BIG morphotype fractions , respectively ( means ± s . d . of three independent biological replicates ) . Solid lines are the fit to data using Equations 1 and 2 , resulting in a = 0 . 24 ± 0 . 13 h−1 and b<<a under activating conditions . These switching rates result in a population dominated by the SMALL morphotype after a few hours . Note that stationary phase caused resetting of the culture to the BIG morphotype . ( C ) Model ( solid line ) and experimental measurement ( dotted line and markers ) of the growth of the total population for the data presented in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 007 The switching rates extracted from the growth of EPEC bacteria in non-activating conditions result in slow dynamics of switching between the two morphotypes . However , already after 3 hr of growth in activating conditions , we observe a high proportion of SMALL morphotypes , suggesting that the switching rate from BIG to SMALL in activating conditions is higher than in non-activating conditions . Our model predicted that longer exposure to activating conditions would lead to a higher percentage of the SMALL morphotype . In order to evaluate the predictive value of the model and the switching rates under activating conditions , we diluted a stationary phase culture into DMEM and monitored BIG and SMALL colony ratios over time . We found that the switching from BIG to SMALL was tenfold faster under activating conditions than under non-activating conditions ( Figure 3B and Materials and methods-Mathematical model ) . As a result , within a few hours under activating conditions , the SMALL morphotype dominated the culture . Similarly to growth in LB , stationary phase resets the culture to unimodality ( Figure 3B , C ) . Taken together , these results show that although the bimodal switch generates variability both under activating and non-activating conditions , the higher switching rate under activating conditions results in a fast increase of the SMALL to BIG ratio . In both conditions , stationary phase resets the culture to a unimodal BIG population . Activating conditions , which were shown above to increase the SMALL/BIG ratio , are known to activate the expression of key transcriptional regulators of EPEC virulence including Ler , GrlA , PerA , and PerC ( Figure 4A , reviewed in Clarke et al . , 2003; Mellies et al . , 2007 ) . We tested for possible involvement of these regulators in the BIG to SMALL switch and found that Ler and GrlA are not required for the bimodal colony size phenotype ( Figure 4—figure supplement 1 ) . In contrast , the EPEC strain cured of the EAF plasmid , which encodes the perABC operon , lost the capacity to generate bimodality and produced only BIG colonies ( Figure 4B ) . Complementing this strain with the EAF plasmid restored the bimodal phenotype . To identify the EAF plasmid genes required to establish the bimodality , we examine mutant strains ΔperA , ΔperC , and ΔbfpA ( Figure 4B , Figure 4—figure supplement 1 ) . Notably , only the ΔperA mutant failed to exhibit bimodality . Importantly , PerA is the autoactivator and thus the perA mutant is deficient in expressing the entire perABC operon . Complementing the ΔperA strain with a low copy number plasmid , containing the perABC operon with its native regulatory region transcriptionally fused to gfp ( pPerABC-GFP ) restored the bimodality . In this case , the SMALL colonies appeared later and were more abundant than in the wt strain , probably due to excessive PerABC expression . These results show that the per operon is required for the co-existence of the BIG and SMALL morphotypes , whereas PerC , GrlA , Ler , T3SS biogenesis , and BFP formation were not required for colony size bimodality ( Figure 4—figure supplement 1 ) . To find whether PerA or PerB underlie the bimodality , we deleted different fragments from the pPerABC-GFP plasmid resulting in pPerA-GFP , pPerB-GFP or pPerAB-GFP . Notably , we kept the native regulatory region implying that in all cases PerA is required for expression . We transformed the ΔperA mutant with plasmids expressing PerA-GFP , PerB-GFP or PerAB-GFP . Whereas complementation of ΔperA with PerA-GFP or PerB-GFP expressing plasmids resulted in unimodal colony morphotypes ( Figure 5A ) , the PerAB-GFP plasmid restored bimodality , indicating that both PerA and PerB are required for the bimodality . Microscopic observations show that GFP expression is bimodal in pPerAB-GFP plasmids , while uniformly high in pPerA-GFP ( Figure 5B ) . PerB and GFP cannot be expressed from pPerB-GFP in the absence of PerA and accordingly no GFP was observed in ΔperA mutant containing this plasmid ( Figure 5 ) . Expression of GFP from this plasmid was restored in the wt EPEC ( Figure 5—figure supplement 1 ) . These results show that both PerA and PerB expression are required for the co-existence of the BIG and SMALL morphotypes . To determine whether other EAF plasmid factors are required for the growth bimodality , we transformed E . Coli K-12 MG1655 strain with the above pPer plasmids and found that co-expression of PerA and PerB is sufficient for induction of bimodality also in this strain . In agreement with the results obtained in the ΔperA EPEC strain , PerA or PerB alone failed to generate bimodality ( Figure 5 ) . These results show that PerAB expression generates a bimodality of growth also without the EAF plasmid , although the phenotype was milder ( i . e . the growth difference between the two morphotypes is smaller ) ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 19599 . 008Figure 4 . Per operon is essential for establishing colony-size bimodality . ( A ) Scheme of key regulatory genes of the EPEC virulence machinery . Ler is the T3SS master regulator and its expression is induced by two redundant positive regulators , PerC and GrlA ( Bustamante et al . , 2011 ) . PerA is a positive autoregulator ( Ibarra et al . , 2003; Martínez-Laguna et al . , 1999; Porter et al . , 2004 ) of perABC operon and positive transcription regulator of typeIV pilli ( bfpA ) ( Ibarra et al . , 2003; Tobe et al . , 1996 ) . Open arrows represent operons , thick arrows and filled boxes represent protein production . Dotted lines indicate regulatory circuits . PerA positive feedback loop is marked in green . ( B ) Histogram of colony appearance times for bacteria taken from activated cultures of indicated strains . Strains without perA , either by gene deletion ( ΔperA ) or EAF plasmid curing ( JPN15/EAF- ) , result in unimodality . These experiments were repeated in at least two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 00810 . 7554/eLife . 19599 . 009Figure 4—figure supplement 1 . ScanLag colony appearance phenotype of EPEC virulence pathway mutants . A histogram of the fraction of colonies detected at each time point for bacteria taken from indicated activated cultures . Deletions in genes ler , grlRA , bfpA , and perC , encoding key regulators of EPEC virulence ( shown in Figure 4A ) , did not prevent bimodality . This experiment was repeated in at least two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 00910 . 7554/eLife . 19599 . 010Figure 5 . Co-expression of PerA and PerB results in colony-size bimodality . ( A ) A histogram of the fraction of colonies detected at each time point for bacteria taken from indicated cultures of EPEC ΔperA mutant transformed with pPerA-GFP ( ΔperA/pPerA-GFP ) , pPerB-GFP ( ΔperA/pPerB-GFP ) and pPerAB-GFP plasmids ( ΔperA/pPerAB-GFP ) . Cultures were started from a single colony ( BIG ) and grown in activating conditions . Only pAB-GFP plasmid , co-expressing PerA and PerB , restored the bimodality . This experiment was repeated in at least two independent biological replicates . ( B ) Expression of perA , perB and perAB using a transcriptional GFP reporter by time-lapse microscopy of single cells extracted from BIG colonies as in ( A ) collected 1000 min after plating . Similar results were obtained in two independent biological replicates and in at least three different locations . Scale bar: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01010 . 7554/eLife . 19599 . 011Figure 5—figure supplement 1 . Expression control of pPerB-GFP in wild type EPEC . ( A ) Histogram of the fraction of colonies wild type EPEC transformed with pPerB-GFP plasmid ( wt/pPerB-GFP ) and grown as in Figure 5 . This experiment was repeated in at least two independent biological replicates . ( B ) Expression of perB using a transcriptional GFP reporter by time-lapse microscopy in BIG colonies of wild type EPEC ( wt/pPerB-GFP ) from ( A ) collected 1000 min after plating . PerB is expressed in a subpopulation of wild type EPEC . Similar results were obtained in two independent biological replicates and in at least three different locations . Scale bar: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01110 . 7554/eLife . 19599 . 012Figure 5—figure supplement 2 . Co-expression of PerA and PerB causes bimodality of colony growth in E . Coli K-12 MG1655 . A histogram of colony area growth time , i . e . the time taken to increase the colony area from 20 to 80 pixels for bacteria taken from indicated activated cultures of MG1655 E . Coli K-12 bacteria transformed with pZS11GFP ( MG1655/pPerA-GFP ) , pPerABC-GFP ( MG1655/pPerABC-GFP ) , pPerA-GFP ( MG1655/pPerA-GFP ) , pPerB-GFP ( MG1655/pPerB-GFP ) and pPerAB-GFP plasmids ( MG1655/pPerAB-GFP ) . Co-expression of PerA and PerB causes bimodality of colony growth in MG1655 . This experiment was repeated in at least two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 012 We next asked whether the colony-size bimodality correlates with bimodality in perABC expression in the progenitor cells , namely the founders of each colony . We subjected a culture of the ΔperA EPEC complemented with pPerABC-GFP to activating conditions and measured GFP expression by microscopy ( Figure 6A ) . We detected a bimodal expression of the perABC-gfp operon . Furthermore , the bacteria that did not express GFP were larger and divided more rapidly than those that expressed GFP . To correlate between perABC-gfp expression and colony size , we used Fluorescence-activated cell sorting ( FACS ) to collect separately the GFP-OFF and GFP-ON bacteria ( Figure 6B ) . Each subpopulation was then plated and analyzed by ScanLag ( Figure 6C , D ) . The results showed that the GFP-OFF bacteria grew in BIG colonies , whereas the GFP-ON bacteria generated almost exclusively SMALL colonies . Similar results were obtained using wild-type EPEC transformed with pPerABC-GFP , whereas no bimodality was observed with control pZS11GFP plasmid expressing GFP from a synthetic constitutive promoter ( Figure 6—figure supplement 1 ) . Taken together , these results show that during growth in activating conditions , a bimodal expression of the perABC operon is established in progenitor cells that lead to bimodality in colony size upon plating and growth in non-activating conditions . 10 . 7554/eLife . 19599 . 013Figure 6 . Bimodality of perABC expression during activation underlies colony-size bimodality . EPEC ΔperA containing the plasmid pPerABC-GFP was grown under activating conditions . ( A ) Time-lapse microscopy of the activated ΔperA/ pPerABC-GFP under non-activating conditions ( i . e . on LB-agar pads , at 32°C ) Scale bar: 15 μm . ( B ) Flow cytometry analysis ( t = 0 min ) for levels of GFP in the cells ( n = 10000 bacteria ) . Time points are indicated . Similar results were obtained in at least five different locations and in two independent biological replicates . ( C–D ) Sorted fractions of perABC GFP-ON and GFP-OFF ( from B ) populations were plated under non-activating conditions and analyzed by ScanLag . Histograms show appearance time of sorted ( C ) GFP-OFF and ( D ) GFP-ON subpopulations . ( B–D ) Experiments were repeated in at least two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01310 . 7554/eLife . 19599 . 014Figure 6—figure supplement 1 . Expression of perABC in wild-type EPEC during activation . ( A ) Expression of perABC in a wild-type EPEC population grown in activating conditions was analyzed using a transcriptional GFP reporter for perABC ( wt/pPerABC-GFP ) . Activated culture was placed on LB agar pads ( non-activating conditions . i . e . LB agar at 32°C ) and analyzed by time-lapse microscopy . Similar results were obtained in at least five different locations and in two independent biological replicates . ( B ) A wild type EPEC control strain , containing a plasmid expressing GFP from PLtetO-1 promoter ( wt/ pZS11GFP ) did not display bimodal GFP expression under the same conditions . Similar results were obtained in at least five technical replicates . Scale bars: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 014 Ler is the T3SS master regulator , and under activating conditions its expression is induced by two redundant positive regulators , PerC and GrlA ( Bustamante et al . , 2011; Gómez-Duarte and Kaper , 1995; Porter et al . , 2004 ) ( Figure 4A ) . Under activating conditions , we observed that Ler expression from chromosomal ler-gfp transcriptional fusion was unimodal ( Figure 7A–C ) , as previously reported ( Berdichevsky et al . , 2005; Roe et al . , 2004 ) . This was in contrast to the bimodal perABC-gfp expression in the same conditions ( Figure 6A , B ) . We predicted , however , that upon shifting the culture from activating to non-activating conditions , which suppresses GrlA activity , only the sub-population that expresses perABC will continue to express ler , resulting in bimodal Ler expression . As predicted , we found that upon shifting from activating to non-activating conditions , Ler-GFP expression was reduced in approximately 50% of the bacterial population and became bimodal ( Figure 7B , D ) . Time-lapse microscopy of bacteria taken from BIG and SMALL colonies showed that , as expected , Ler expression was uniformly high in the SMALL population but bimodal in the BIG population ( Figure 7—figure supplement 1 ) . These results suggest that the bimodal expression from perABC during activating conditions does not result in bimodal Ler expression since GrlA , which is redundant to PerC , activates ler expression regardless of whether PerC is expressed or not . However , upon shifting to non-activating conditions and the bimodal expression of perABC drives the bimodality of Ler expression . 10 . 7554/eLife . 19599 . 015Figure 7 . Ler expression is unimodal during activation but becomes bimodal when cells are shifted to non-activating conditions . EPEC ler-gfp was grown under activating conditions . ( A ) Flow cytometry analysis ( t = 0 min ) shows unimodal GFP ( Ler-ON state ) expression ( n = 10000 ) . ( B ) Time-lapse microscopy during growth under non-activating at indicated times . Scale bar: 15 μm . ( C–D ) Quantification of GFP levels from the images shown in panel ( B ) . Similar results were obtained in at least five different locations and in two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01510 . 7554/eLife . 19599 . 016Figure 7—figure supplement 1 . Ler is differentially expressed in BIG and SMALL colony morphotypes . ( A ) EPEC containing a chromosomal ler-gfp fusion was grown for 3 hr in DMEM at 37°C and plated on LB agar at 32°C . ScanLag analysis and colony appearance time was monitored and plotted . ( B ) BIG and ( C ) SMALL colonies of ler-gfp EPEC were picked separately after 1000 min , suspended in LB , re-plated on LB agar pads ( non-activating conditions . i . e . LB agar at 32°C ) , and time-lapse microscopy was used to measure the ler-gfp expression during 3 hr of growth . GFP , reflecting ler expression , in bacteria that originated from a SMALL morphotype colony is uniformly high but is bimodal in the bacteria that originated from a BIG morphotype colony . Similar results were obtained in at least five different locations and in two independent biological replicates . Scale bar: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 016 The stability of the SMALL morphotype suggests that the virulence switch mediated by the per operon is a hysteretic switch , maintaining long-term memory of the previous state as shown in Figure 8A . Accordingly , the entire population switches to a virulent state upon activation , characterized by unimodal and high ler expression ( Figure 8A , State 1 ) . In contrast , per expression during activation is typically bimodal , although prolonged activation eventually shifts the population to SMALL ( Figure 3B ) . When transferred to non-activating conditions , the per-ON bacteria remain hypervirulent , expressing both Ler and PerABC , leading to SMALL colony morphotype ( Figure 8 , State 2 ) . This hypervirulent state is maintained for an extremely long time but is ‘reset’ to the non-virulent state during stationary phase ( Figure 8 , State 3 ) . When a ‘reset’ population is subjected to non-activating conditions the majority remains in the BIG morphotype ( Figure 8 , State 4 ) . This insight leads to several predictions . Firstly , the SMALL morphotype should express virulence factors downstream of both ler and per ( i . e . , BFP and T3SS ) , even after many generations of growth in non-activating conditions . Secondly , the BIG colony morphotype should consist mainly of non-virulent bacteria with ~20% of hypervirulent ones . Finally , the deletion of the per operon should not prevent the activation of virulence ( i . e . , T3SS expression can be driven through the GrlA-Ler path ) , but erase the hysteretic switch , and thus ler expression and virulence of all bacteria , should decline as soon as the ΔperA bacteria are transferred to non-activating conditions ( Figure 8B ) . 10 . 7554/eLife . 19599 . 017Figure 8 . PerABC maintains long-term memory through a hysteretic switch . ( A ) Scheme of the hysteretic switch in wild-type EPEC . Subjecting a culture to activating conditions for several hours results in a majority of SMALL bacteria ( green ) ( State 1 ) ( Figure 3B ) . Even when transferred to non-activating conditions , the SMALL bacteria maintain their phenotype ( State 2 ) , unless subjected to stationary phase conditions results in BIG ( red ) colonies ( State3 , ‘Reset’ ) . Growth under non-activating conditions maintains a majority of the BIG phenotype ( State 4 ) . Shifting again to activating conditions regenerates the SMALL phenotype ( State 1 ) . ( B ) Deletion of perA abolishes the hysteretic switch but does not prevent ler activation . ( C ) Western blot analysis of proteins extracted from wild-type EPEC in the different states defined in ( A ) using antibodies raised against BfpA , EspB , Tir , and FliCH6 . The following conditions were used: Reset: LB overnight culture ( State3 ) ; Activation ( State1 ) ; SMALL colony ( State2 ) ; BIG colony ( State 4 ) . ( D ) Western blot analysis of proteins extracted from EPEC ΔperA grown in Activation ( State 1 ) and BIG colony ( State 4 ) . ( C , D ) Similar results were obtained in at least two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01710 . 7554/eLife . 19599 . 018Figure 8—figure supplement 1 . Loading control of proteins for Western blot analysis . Total protein analysis of Stain-Free Precast Gels ( Bio-Rad ) was used as loading control . Same gel was used for protein transfer to membrane and following Western blot procedure ( Figure 8 ) . Lane ( − ) was not used in this experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 018 In order to test these predictions , we extracted proteins from EPEC and ΔperA cultures grown under conditions that would lead to states shown schematically in Figure 8A , B and performed Western blot analysis using antibodies raised against BfpA , EspB , and Tir . BfpA , an important constituent of the type IV pilli , was used as readout for PerA activity , and the T3SS proteins EspB and Tir were used as readout for Ler activity . As an additional negative readout of virulence , we used flagellin ( FliCH6 ) , which is known to be repressed by GrlA ( Iyoda et al . , 2006 ) ( Figure 4A ) . As expected , in EPEC , EspB , Tir , and BfpA , but not FliC were highly expressed upon growth in activating conditions ( Figure 8A , C: State 1 ) . Similar expression patterns were seen in SMALL colony bacteria , despite the fact that at least 20 generations had passed since the transition from activating to non-activating conditions ( Figure 8A , C: State 2 ) . In contrast , bacteria from BIG colonies showed low expression levels of all virulence factors and high levels of FliC . The residual expression of BfpA , EspB , and Tir by the BIG colony bacteria is consistent with the prediction of ~20% SMALL variants in the BIG population ( Figure 8A , C: State 4 ) . Finally , the ΔperA mutant showed high expression of EspB and Tir when grown in activating conditions , consistent with activation of ler expression by GrlA ( Figure 8B , D: State 1 ) . However , expression of these proteins rapidly diminished when the mutant culture was shifted to growth under non-activating conditions ( Figure 8B , D: State 4 ) , leading to a pattern expected from a unimodal non-virulent population ( i . e . , high FliC , low EspB and low Tir ) and similar to the pattern of expression in the stationary phase LB culture ( Figure 8A , C: State 3 ) . Our results show that the SMALL colonies express a high level of functional T3SS and BFP and thus may be hypervirulent . To determine whether this expression pattern results in a hypervirulent phenotype , we first tested SMALL colony bacteria for BFP functionality by monitoring BFP-mediated self-aggregation ( Bieber et al . , 1998 ) . Time-lapse microscopy showed that , as expected , resuspended SMALL colony bacteria rapidly aggregated , whereas BIG colony bacteria remained mostly planktonic ( Figure 9A , Video 1 ( BIG ) , Video 2 ( SMALL ) , Figure 8—figure supplement 1A ) . In both cases , the aggregates disintegrated upon reaching stationary phase , consistent with the resetting of colony-size bimodality ( Figure 2F–H ) and the significant reduction in BfpA production in stationary phase ( Figure 8C ) . 10 . 7554/eLife . 19599 . 019Figure 9 . Bimodal perABC expression correlates with bimodality in microcolony formation and host cell attachment . ( A ) Phase-contrast images of the dynamics of self-aggregation observed by time-lapse microscopy of bacteria from BIG or SMALL colonies . Scale bar: 15 μm . See also Videos 1 and 2 . ( B ) Time-lapse microscopy of HeLa cells infected with EPEC ΔperA/pPerABC-GFP . Scale bar: 25 μm . ( C ) Fluorescent time-lapse microscopy of HeLa cells infected with a 1:1 mixture of wild-type EPEC from BIG and SMALL colonies tagged with mCherry and YFP , respectively . Scale bar: 25 μm . Enlarged image shows SMALL bacteria attached to the Hela cells whereas the BIG bacteria are planktonic resulting in a shift between phase-contrast and red fluorescent image . ( D , E ) Quantification of ( D ) attached and ( E ) planktonic bacteria in images taken from ( C ) . The area of bacteria was determined based on fluorescent signal . The attached bacteria area was normalized to the total area of HeLa cells in the frame . Planktonic bacteria area was normalized to the area free of cells . Data are presented as the means ± s . d . of 6 frames . The experiment was repeated three times . See also Videos 3 and 4 . Similar results were obtained in at least five different locations and in two independent biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 01910 . 7554/eLife . 19599 . 020Figure 9—figure supplement 1 . Bacteria from SMALL colonies have enhanced self-aggregation properties in liquid culture and increased formation of microcolonies on host cells . ( A ) BFP-related aggregation observed directly in EPEC cultures . EPEC bacteria from different conditions were diluted in LB to ~107 bacteria/ml and grown under non-activating conditions . Images were acquired with a scanner . The following cultures were used for inoculation: EPEC culture grown in DMEM for 3 hr at 37°C ( Activated ) , bacteria from suspended SMALL , or BIG colonies ( SMALL and BIG respectively ) , EPEC grown in LB to stationary phase at 37°C ( Reset ) . Note the low turbidity reflecting the absence of planktonic bacteria only in the SMALL well . Scale bar: 4 mm . Similar results were obtained in at least two independent biological replicates . ( B ) Fluorescent time-lapse microscopy of HeLa cells infected with a 1:1 mixture of wild-type EPEC from BIG and SMALL colonies tagged with YFP and mCherry , respectively ( this tagging is the opposite of that in Figure 9B ) . Similar results were obtained in at least five different locations and in two independent biological replicates . Scale bar: 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 02010 . 7554/eLife . 19599 . 021Figure 9—figure supplement 2 . Bacteria from SMALL colonies induce massive pedestal formation during HeLa infection . HeLa cells were infected for 2 hr with different inoculums of ler-GFP EPEC ( green ) . The following cultures were used: EPEC culture grown in DMEM for 3 hr at 37°C ( activated ) , and bacteria from suspended SMALL , or BIG , colonies ( SMALL and BIG respectively ) . Infected cells were fixed and stained for actin ( phalloidin rhodamine-red ) and DNA ( DAPI-blue ) . Experiments were carried out in duplicate and representative images are shown . Similar results were obtained in two technical replicates . Scale bar: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 02110 . 7554/eLife . 19599 . 022Figure 9—figure supplement 3 . Invasion of HeLa cells by EPEC from SMALL and BIG colonies . ( A ) Schematic illustration of gentamicin protection assay of HeLa cells infected with a 1:1 mixture of wild-type EPEC from SMALL and BIG colonies tagged with GFP and mCherry , respectively . ( B ) Plot of the results of the gentamicin protection assay for wild-type EPEC bacteria and for the ΔperA mutant . Data were normalized to initial CFU ( t = 0 min ) and are presented as the means ± s . d . of at least three technical repeats . ( C ) Schematic illustration of the microscopy-based gentamicin invasion assay together with induction dynamics . Induction of GFP fluorescence by IPTG 1 hr after gentamcin treatment enables the visualization of only live intracellular bacteria . ( D ) Confocal images ( top and side views ) of intracellular bacteria expressing GFP using the protocol described in ( C ) . HeLa cells were fixed and stained for actin ( red ) and DNA ( blue ) . Scale bar: 10 μm . ( E ) Time-lapse microscopy of HeLa cells infected according to the protocol shown in ( C ) showing growth of intracellular bacteria . Overlay of phase contrast and green fluorescence images are shown . Similar results were obtained in at least three different locations and in three independent biological replicates . Scale bar: 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 02210 . 7554/eLife . 19599 . 023Video 1 . Dynamics of self-aggregation observed by time-lapse microscopy in BIG bacteria . Bacteria were resuspended from a BIG colony and placed on a wet LB agarose pad for imaging bacteria in suspension . Bacteria divide and remain mostly planktonic . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 02310 . 7554/eLife . 19599 . 024Video 2 . The dynamics of self-aggregation observed by time-lapse microscopy in SMALL bacteria . Bacteria were resuspended from a SMALL colony and placed on a wet LB agarose pad for imaging bacteria in suspension . Bacteria divide and aggregate continuously until they reach stationary phase , which results in the disintegration of the aggregates . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 024 In order to determine whether high levels of per operon expression correlate with higher infectivity , we followed the infection of Hela cells with EPEC ΔperA bacteria transformed with pPerABC-GFP plasmid . We used microscopy to evaluate perABC-gfp expression , aggregation ( i . e . , microcolony formation ) , and attachment to host cells . The results showed that the bacteria that expressed PerABC ( GFP-ON , per-ON ) formed microcolonies that rapidly attached to the epithelial cells ( Figure 9B ) . In contrast , the GFP-OFF ( per-OFF ) bacteria remained mostly planktonic and unattached . These results show that the per-ON bacteria , which generate the SMALL colony morphotype , display higher infectivity than per-OFF bacteria . We next asked whether this high infectivity is maintained in bacteria taken from SMALL colonies ( i . e . , bacteria originating from per-ON bacteria but that were grown in non-activating conditions for many generations ) . We infected HeLa cells with a 1:1 mix of bacteria from SMALL and BIG colonies tagged with constitutive YFP and mCherry ( Gefen et al . , 2008 ) , respectively , and compared infectivity by time-lapse microscopy ( Figure 9C , Figure 9—figure supplement 1B ) . During the first hours of infection , the SMALL bacteria formed aggregates ( microcolonies ) and almost no planktonic single bacteria were found in the surrounding medium ( Figure 9C–E , Figure 9—figure supplement 1B , Videos 3 and 4 ) . Furthermore , these SMALL microcolonies attached to host cells , an indication of BFP function ( Figure 9—figure supplement 2 ) , induced actin rearrangement in the host , and invaded into the host cell , indicatives of T3SS functionality ( Figure 9—figure supplement 2 , Figure 9—figure supplement 3 ) . In contrast , only a few bacteria that originated from BIG colonies were organized into attached microcolonies , and formation of actin rearrengement as well as invasiveness were marginal ( Figure 9C–E , Figure 9—figure supplement 1B , Videos 3 and 4 , Figure 9—figure supplement 3 ) . These results show that the SMALL morphotype maintains high infectivity and expresses functional BFP and T3SS even after many generations of growth in non-activating conditions . 10 . 7554/eLife . 19599 . 025Video 3 . Dynamics of infection by SMALL ( green ) and BIG ( red ) bacteria on HeLa cells . The SMALL bacteria form microcolonies attached to the HeLa cells , whereas the BIG bacteria remain mostly planktonic ( same as Figure 9C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 02510 . 7554/eLife . 19599 . 026Video 4 . Dynamics of infection by SMALL ( red ) and BIG ( green ) bacteria on HeLa cells . The SMALL bacteria form microcolonies attached to the HeLa cells , whereas the BIG bacteria remain mostly planktonic ( same as Video 3 but with fluorescent markers opposite tagging ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19599 . 026 This study showed , for the first time to our knowledge , that the virulence machinery of a human pathogen , EPEC , is controlled by a hysteretic switch with long epigenetic memory . We showed that PerA and PerB are sufficient for this hysteretic switch . We found that when exposed to virulence-activating conditions all EPEC bacteria upregulate expression of T3SS virulence genes , unlike the bimodal virulence expression observed in Salmonella . However , we found that the EPEC virulent population is bimodal for expression of the per regulated genes , resulting in two coexisting virulent sub-populations of bacteria , planktonic ( per-OFF ) and aggregative ( per-ON ) , with different infection and invasion abilities . The latter population constitutively expresses both BFP and T3SS for many generations , rapidly attaches to host cells as microcolonies , delivers effectors into the host cell , and invades it . This rapid invasion may protect the bacteria against the host immune system and establish persistent infection ( Tuchscherr et al . , 2011 ) . The striking difference between the two phenotypes that we unveiled is in their abilities to maintain their virulence state when transferred to non-activating conditions . Whereas the per-OFF bacteria no longer express ler regulated genes upon transfer to non-activating conditions , the per-ON bacteria maintain high expression of the ler regulated virulence genes even after tens of generations of growth in non-activating conditions . This per-ON long-term memory may allow the pathogen to overcome transit through niches of non-activating conditions without a drop in virulence level . An extreme case of this type of transit is the host-to-host transit through the fecal-oral route . In addition , the long-term memory per-ON state may set the stage for further diversification within this subpopulation , possibly creating a range of infective phenotypes , each adapted to a different niche and/or stage of infection within the host intestine . Alternatively , the hysteretic switch that enables the coexistence of two different phenotypes may be attributable to bet-hedging ( i . e . to risk spreading in the absence of a predictable environment ) . The SMALL morphotype infects host cells more rapidly but bears the cost of expressing virulence genes and exposure to the immune system , whereas the BIG morphotype grows faster and is less immunogenic . Notably , bimodality is stable mainly under non-activating conditions , suggesting that insufficient cues from the environment regarding bacterial residence inside or outside the host may promote a bet-hedging strategy ( Kussell and Leibler , 2005 ) . We found that PerA is central for this switch and that the PerA-regulated perABC operon exhibits bimodal expression when co-expressed with PerB . Notably , the per operon is regulated by PerA auto-activation ( Figure 4A ) , a network motif that has been shown to lead to bimodality ( Smits et al . , 2006 ) and hysteresis ( Mitrophanov and Groisman , 2008 ) , which can lead to bistability by growth feedback mechanisms ( Deris et al . , 2013; Irwin et al . , 2010; Tan et al . , 2009 ) , or other mechanisms , for which PerB may be required . Specific environmental conditions ( termed here ‘activating conditions’ ) strongly increased the per-ON frequency in the population by enhancing the switching rate by a factor of ~10 compared to non-activating conditions . Thus , growth under activating conditions resulted in almost 100% of per-ON cells at steady state . Importantly , our results show that per expression is essential for establishing a hysteretic long-term memory switch , resulting in the co-existence of per-OFF and per-ON subpopulations , of which the latter remains stable even in the face of drastic changes in the environmental conditions , such as shifts in temperature and growth media . Given this stability , a single bacterium in the per-ON state generates a colony of the SMALL morphotype in which most of the bacteria remain per-ON and are primed for rapid infection of host cells . Interestingly , the per-ON memory vanished once growth reaches stationary phase , and the entire population switched to per-OFF . Bimodality in expression of virulence genes has been extensively studied in Salmonella and thus it is useful to compare the two pathogens . In EPEC , expression of the Ler master regulator , and thus expression of T3SS , is high and unimodal under activating conditions . Redundant activation of the ler promoter by independent regulators ( GrlA and PerC ) ensures that all bacteria express virulent genes during activation . In contrast , Salmonella displays bimodal T3SS expression upon growth under activating conditions and during initial infection ( Hautefort et al . , 2003; Sturm et al . , 2011 ) . The per-switch acts hysteretically , locking the expression of ler and activation of virulence in an ‘ON’ state , even when bacteria are switched back to non-activating conditions . Thus in EPEC , ler expression becomes bimodal only when the bacteria are transferred to non-activating conditions , resulting in the co-existence of non-activated bacteria and bacteria that are already primed for infection through constitutive expression of BFP and T3SS . Importantly , we were able to evaluate the rates of switching between the two EPEC phenotypes and found that the rate from ON to OFF is extremely slow ( several weeks ) , even under non-activating conditions . A long-lived ON state has been observed also in Salmonella ( Sturm et al . , 2011 ) , but upon shifting to non-activating conditions , regulators of virulence decay within 1 to 2 hr ( Temme et al . , 2008 ) . The importance of the described hysteretic switch for EPEC virulence is reflected by the conservation of the switch core: the perABC operon . Recent reports comparing the genome sequences of a large number of EPEC clinical isolates show that EPEC is an umbrella name for a collection of E . coli strains belonging to diverse phylogenetic branches that acquired independently , through horizontal gene transfer , a pathogenicity island encoding T3SS ( the LEE island ) ( Ingle et al . , 2016 ) . Notably , most of these EPEC strains also acquired plasmids containing the perABC operon ( Ingle et al . , 2016 ) . Furthermore , strains containing both the LEE and plasmids encoding perABC and bfp operons cause a more severe disease ( Hazen et al . , 2015b ) . Hazen et al . proposed that the contribution of PerA might be related to regulation of additional virulence-related genes ( Hazen et al . , 2015b ) . Our findings suggest that perABC also enhances the fitness of infecting EPEC by facilitating formation of long-term memory and stable phenotypic bimodality . The contribution of the per-switch to virulence in vivo could not be tested since an animal model for EPEC is not available . Pathogens closely related to EPEC , including enterohemorragic E . coli ( EHEC ) and Citrobacter rodentium ( CR ) , do not carry the perABC operon . Interestingly , however , heterogeneity in virulence is observed in these pathogens ( Roe et al . , 2004; Kamada et al . , 2015 ) , but the involved switch and whether it is also hysteretic have not been studied . In conclusion , we report here how a hysteretic switch controls the virulence traits of a human pathogen , EPEC . Our findings and approach should provide a framework to search for similar switches in other pathogens . Furthermore , this understanding may lead to the development of new strategies to interfere with the establishment of stable virulence-ON mechanisms and thus reduce virulence and pathogen spreading to new hosts . The used bacterial strains , plasmids and primers are listed ( Supplementary file 2 and 3 ) . Activating conditions: Following the procedure in ( Kenny et al . , 1997 ) , bacterial strains were grown overnight ( O/N ) in LB medium ( Sigma , Israel ) at 37°C without shaking , diluted 1:40 into DMEM-HEPES ( Gibco , Israel ) medium and incubated for 3 hr at 37°C without shaking to exponential phase ( O . D . ~0 . 3 ) . Non-activating conditions: bacteria were plated in LB agar at a concentration below 200 cfu/plate and incubated at 32°C . For analysis of bacteria isolate directly from colonies , BIG and SMALL colonies were collected according to their size at 17 hr ( 1000 min ) after plating and diluted in 0 . 9% NaCl to density of ~108 bacteria/ml . Deletion mutants were produced as described ( Datsenko and Wanner , 2000 ) . pZSGFP was created by replacing the hip promoter of pZS1HGFP plasmid ( Rotem et al . , 2010 ) with the synthetic PLtetO-1 promoter ( Lutz and Bujard , 1997 ) using 5'-phosphorylated PCR primer , followed by ligation . pPerABC-GFP was created with an isothermal cloning kit ( NEB , United States ) . The hip promoter of pZS1HGFP was replaced with the genomic region of the perABC operon including the upstream promoter region . The derivative plasmids: pPerA-GFP , pPerB-GFP and pPerAB-GFP were produced by single ligation step after excision of the PerAB , PerA/C and PerC respectively ( see Supplementary file 2 ) . HeLa cells ( Supplementary file 2 ) were grown in DMEM supplemented with 300 μg/ml L-Glutamine , 100 U/ml Penicillin , 100 μg/ml Streptomycin and 10% FCS at 37°C and 5% CO2 . For infection experiments , HeLa cells were seeded in 24-well plates . When cultures reached ~106 cells/well , they were washed twice with PBS and medium was changed to DMEM-HEPES ( Gibco , United States ) without supplements . HeLa cells were routinely tested for absence of mycoplasma contamination by EZ-PCR Mycoplasma Test Kit ( Biological Industries LTD . , Israel ) . Bacteria were diluted to a concentration of 103 bacteria/ml and plated on LB agar . The plates were placed in a 32°C incubator on EPSON Perfection 3490 scanners that scan the plates every 15 min with custom ScanLag software , as described ( Levin-Reisman et al . , 2010 ) . MatLab based applications were used to automatically detect colonies in each frame and to monitor the growth of individual colonies ( Software for controlling the scanners and for image analysis can be found at http://bio-site . phys . huji . ac . il/Materials ) . The area growth rate of each colony and its time of appearance were extracted as described ( Levin-Reisman et al . , 2010 ) EPEC was grown overnight in LB medium ( Sigma , Israel ) at 37°C , diluted to ~1000 cells/ml in DMEM and grown under activating conditions . The low culture density enabled follow up of many hours of exponential growth before reaching stationary phase . To eliminate possible artifacts due to the BFP-mediated-aggregation we induced disaggregation as follows: the culture was divided into tightly closed Eppendorf tubes , one for each time point . These tubes were subjected to intensive vortex and kept on ice for 10 min before plating . Disaggregation was confirmed by microscopy . The levels of BIG and SMALL progenitors in the populations at each time point during growth were determined by ScanLag . Time-lapse microscopy was performed using a Leica DMIRE2 inverted microscope system . Autofocus and image acquisition were done by using custom macros in μManager ( an open source software program ) to control the microscope , stage , shutters , and camera . The microscope was placed in a large incubator box ( Life Imaging Systems ) that controls the temperature to an accuracy of 0 . 1°C . GFP- or YFP-expressing bacteria were imaged using Yellow GFP filter ( Ex-500nm , Em-535nm , Chroma USA ) ; mCherry signal was measured by HcRed1 filter ( Ex-575nm , Em-640nm , Chroma USA ) . Excitation was performed with LEDs ( Coolled , United Kingdom ) and images were acquired with a cooled CCD camera ( −75°C ) ( Orca II , Back-illuminated , Hamamatsu ) and processed with ImageJ ( http://rsbweb . nih . gov/ij/ ) . X100 NA 1 . 40 oil objective was used for individual bacterial observation on agar pads; X63 NA 0 . 70 long-distance air objective NA for imaging Hela and bacteria in 24-well plates; X20 was used for imaging growing colonies on a very thin LB + 1 . 5% agar layer . To observe the growth of individual bacteria , a LB + 1 . 5% agarose pad was prepared in a polydimethylsiloxane ( PDMS ) square mold and dried for 10 min at 37°C . Bacterial samples of 1 μl ( ~105 cells ) were placed between the microscopic slide and the pad inserted into the same PDMS mould and covered with a coverslip . For microscopic observation of self-aggregation , agarose pads were not dried and bacteria ( 5 µl , ~105 cells ) originated from suspended colony were placed between the microscopic slide and the pad . The OD of the cultures was determined and density was adjusted if needed . Bacteria were collected by centrifugation and resuspended in 5 μl loading sample buffer per 0 . 1 O . D . Proteins were extracted by boiling of the samples and resolved by 12% Mini-PROTEAN TGX Stain-Free Precast Gels ( Bio-Rad , Israel ) . Total protein staining was used as a loading control ( see Figure 8—figure supplement 1 ) . Proteins were transferred to nitrocellulose membrane ( Bio-Rad ) for standard Western blot analysis with antibodies raised in rabbits against BfpA ( gift from Michael Donnenberg ) , Tir and EspB ( gift from Gad Frankel ) , or FliC-H6 ( gift from the Israeli Ministry of Health ) and secondary anti-rabbit HRP-conjugated antibody . Bacteria were diluted 1:50 in LB media in 24-well plates . The plates were incubated at 32°C with mild agitation under non-activating conditions and transferred manually every 30 min to an Epson Perfection V500 Scanner for imaging of aggregates . HeLa cells were grown on round coverslips within 24-well plates . In the next day cells were washed and infected with ~106 bacteria/well . Two hours after infection , the wells were washed twice with PBS and fixed with 4% formaldehyde in PBS , 10 min . The coverslips were then washed twice with TBS , and the cells were permeabilized with 0 . 25% Triton X-100 for 2 min . Actin was stained with Texas Red-phalloidin ( Molecular Probes , United States ) and DNA was stained with Dapi ( Molecular Probes , United States ) , at a 1:1000 dilution . The coverslips were washed twice with PBS , mounted with ImumMount ( Thermo Scientific , United States ) and viewed with a fluorescence inverted microscope with X100 oil objective . HeLa cells were seeded in 24-wells plates ( NUNC , United States ) and grown in 1 ml/well as mentioned above . EPEC BIG/pZA21mCherry and SMALL/pZA21YFP colonies ( harboring mCherry and YFP constitutive markers ) , were suspended , mixed 1:1 and added to the HeLa cells at a concentration of ~106 bacteria/well . Time-Lapse microscopy was performed directly on 24-wells plates ( NUNC , United States ) at 37°C . The opposite fluorescent markers ( i . e . SMALL/pZA21mCherry and BIG/pZA21YFP ) were measured in parallel to rule out effects of markers . The analysis of fluorescent intensities was performed with ImageJ . The total area occupied by bacteria was measured according to the yellow fluorescence signal . In order to compare different frames with variable Hela cell coverage , we normalized the total number of bacteria attached to cells with the total area of cells in the frame . This area was extracted from phase-contrast images . Planktonic bacteria area was normalized with the total area of the frame not covered by Hela cells . HeLa cells grown in 24-well plates and infected with EPEC bacteria were incubated at 37°C for 2 hr . Gentamicin at a final concentration of 25 µg/ml was added to kill extracellular bacteria with little effect on intracellular bacteria ( Benjamin et al . , 1995 ) and plates were returned to the 37°C incubation . Bacteria that survived the gentamicin treatment were counted at 30 min intervals after gentamicin addition . To this end cells were washed twice to remove gentymicine , lysed in 1 ml 1% Triton X-100 to free the intracellular bacteria , which were then pelleted by centrifugation ( 3 min at 1500g ) , resuspended in PBS and spread onto LB agar plates to evaluate bacterial CFU/ml . Isolated colonies were counted after overnight incubation at 37°C , and the progeny of BIG-YFP and SMALL-mCherry colonies were differentiated by color . The time point t = 0 sample was measured before gentamicin treatment . All bacterial counts were normalized to counts at t = 0 counts . The same results were obtained with swapped reporter colors , SMALL-YFP and BIG-mCherry . EPEC containing a plasmid with IPTG inducible GFP reporter ( pSA11 ) was used to generate SMALL colonies , which were suspended and used to infect HeLa cells . Two hours post infection gentamicin was added ( 25 µg/ml ) and after an additional hour IPTG ( 20 μg/ml ) was added to the infected cells . Only the metabolically active intracellular bacteria that are not exposed to gentamicin are able to produce GFP in response to IPTG , thus enabling to detect even a small fraction of intracellular bacteria by time-lapse microscopy . For immunostaining experiments , the gentamicin protection assay was performed as described above , except for the growth of HeLa which was here done on round coverslips inserted into 24-well plates . HeLa cells were infected with EPEC SMALL/pSA11 bacteria for 2 hr and then treated with 25 µg/ml gentamicin . After 1 hr wells were supplied with IPTG to visualize metabolically active intracellular bacteria . One hour later samples were washed twice with PBS and fixed with 4% formaldehyde-PBS . The coverslips were washed twice with TBS , the cells were permeabilized with 0 . 25% Triton X-100 and washed twice with PBS . Actin was stained with Texas Red-phalloidin ( Molecular Probes ) and DNA was stained with Hoechst ( Molecular Probes , United States ) , at a 1:1000 dilution . After washing twice with PBS , the coverslips were mounted with ImumMount ( Termo Scientific , United States ) and viewed with a FV-1200 Olympus ( Japan ) confocal microscope . The GFP signal was measured with a green filter ( Ex-500nm , Em-540nm ) , mCherry signal was measured with a red filter ( Ex-570nm , Em-620nm ) and Hoechst was measured with a Dapi filter ( Ex-430nm , Em–470nm ) . BIG or SMALL colonies were suspended , diluted in PBS to ~105 cells/ml and analyzed using a FACS Aria III cell-sorter equipped with 488 nm and 561 nm lasers ( BD Biosciences , San Jose , CA ) . Side and forward scatter of bacteria were determined using log scale SSC/FSC plots with respective thresholds of 200 and 2200 . Sorting was done at a minimal flow rate according to GFP intensity criteria . BIG and SMALL colonies were suspended , diluted 1:200 and bacteria were grown in LB to O . D . ~0 . 3 at 32°C . DNA was extracted with the DNeasy Blood and Tissue kit ( Qiagene ) according the manufacturer's instructions . Genomic extraction , Whole-Genome Sequencing and analysis was done as published previously ( Goldberg et al . , 2014 ) . The WGS raw data are available as NCBI BioProject PRJNA255355 ( Accessions: SRX757584 and SRX757585 for SMALL and BIG respectively ) We used a simple mathematical model to describe the expected dynamics of switching between the BIG and SMALL morphotypes . Fitting the experimental results to the data enabled the evaluation of the switching rates , a ( BIG to SMALL ) and b ( SMALL to BIG ) , under virulence activating conditions . Surprisingly , we found that what was considered as ‘non-activating conditions’ , namely growth in LB , does not abolish the switching but rather reduces its frequency . The two morphotypes , BIG ( B ) and SMALL ( S ) are characterized by different growth rates , µB and µs respectively , and switching rates , a and b , respectively ( Equations 1 and 2 in Figure 3A ) . The analytical solution of Equations 1 and 2 , as detailed in ( Balaban et al . , 2004 ) , is: ( 1 ) B ( t ) , S ( t ) =eμ¯t[αB , SeΩt+βB , Se−Ωt]{μB∗=μB−aμS∗=μS−bμ¯=μB∗+μS∗22Ω= ( μB∗−μS∗ ) 2+4ab{αB=B0 ( μ¯+Ω−μS∗ ) +S0b2ΩβB=−B0 ( μ¯−Ω−μS∗ ) −S0b2Ω{αS=S0 ( μ¯+Ω−μB∗ ) +B0a2ΩβS=−S0 ( μ¯−Ω−μB∗ ) −B0a2Ω where B0 and S0 are the numbers of the BIG and SMALL morphotypes at t = 0 .	Bacteria typically cope with harsh and changing environments by activating specific genes or accumulating those mutations that change genes in a beneficial way . Recently , it was also shown that the levels of gene activity can vary between otherwise identical bacteria in a single population . This provides an alternative strategy to deal with stressful conditions because it generates sub-groups of bacteria that potentially already adapted to different environments . Bacteria that enter the human body face many challenges , and this kind of pre-adaptation could help them to invade humans and overcome the immune system . However , this hypothesis had not previously been tested in a bacterium called enteropathogenic E . coli , which infects the intestines and is responsible for the deaths of many infants worldwide . Ronin et al . show that cells in enteropathogenic E . coli colonies spontaneously form into two groups when exposed to conditions that mimic the environment inside the human body . Once triggered , one of these groups is particularly dangerous and this “hypervirulent” state is remembered for an extremely long time meaning that the bacteria remain hypervirulent for many generations . In addition , Ronin et al . identified the specific genes that control the switch to the hypervirulent state . These findings have uncovered the existence of groups of enteropathogenic E . coli that are pre-adapted to invading human hosts . Finding out more about how the switching mechanism works and its relevance in other bacteria may help researchers to develop new therapies that can help fight bacterial infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2017	A long-term epigenetic memory switch controls bacterial virulence bimodality
Single-cell transcriptomes are established by transcription factors ( TFs ) , which determine a cell's gene-expression complement . Post-transcriptional regulation of single-cell transcriptomes , and the RNA binding proteins ( RBPs ) responsible , are more technically challenging to determine , and combinatorial TF-RBP coordination of single-cell transcriptomes remains unexplored . We used fluorescent reporters to visualize alternative splicing in single Caenorhabditis elegans neurons , identifying complex splicing patterns in the neuronal kinase sad-1 . Most neurons express both isoforms , but the ALM mechanosensory neuron expresses only the exon-included isoform , while its developmental sister cell the BDU neuron expresses only the exon-skipped isoform . A cascade of three cell-specific TFs and two RBPs are combinatorially required for sad-1 exon inclusion . Mechanistically , TFs combinatorially ensure expression of RBPs , which interact with sad-1 pre-mRNA . Thus a combinatorial TF-RBP code controls single-neuron sad-1 splicing . Additionally , we find ‘phenotypic convergence , ’ previously observed for TFs , also applies to RBPs: different RBP combinations generate similar splicing outcomes in different neurons . The complement of genes expressed in an individual cell type controls its identity , development , and function . While transcriptional regulation is a major component of gene expression , post-transcriptional regulation can further shape cellular attributes by , for example , determining which gene isoforms are expressed in a cell . Much recent work has gone into cataloging gene expression networks in single cells , particularly those of specific neuronal types ( Tanay and Regev , 2017; Zeng and Sanes , 2017 ) . Molecular studies have also identified mechanisms by which transcription factors ( TFs ) shape gene expression networks in single neurons . Due to technical limitations , less is known about post-transcriptional regulation at the level of single neurons , or about the RNA binding proteins ( RBPs ) mediating post-transcriptional regulation ( Gracida et al . , 2016 ) . It is also unknown to what extent transcriptional and post-transcriptional gene regulatory networks are coordinated in single cells . A number of studies have identified individual RBPs that affect the splicing of a TF , thus altering the activity or specificity of that TF ( Calarco et al . , 2009; Han et al . , 2013; Linares et al . , 2015; Raj et al . , 2011 ) . These results suggest that there may be extensive cross-talk between transcriptional and post-transcriptional regulatory layers . The nematode Caenorhabditis elegans has been used extensively as a model to reveal underlying principles by which TFs shape the transcriptomes of individual neurons . The worm’s invariant cell lineage , coupled with genetic tools and a transparent body , enables systematic in vivo analysis of gene expression in single neurons , and identification of TFs responsible for cell-specific gene expression . This type of analysis has revealed a number of gene regulatory principles , including the concept of a ‘combinatorial code’ of TFs which can be re-used in different neuron types , with particular combinations of TFs determining specific cell fates ( Gendrel et al . , 2016; Gordon and Hobert , 2015; Pereira et al . , 2015 ) . Another example is the concept of ‘phenotypic convergence’ by which various neurons express similar gene networks but the TFs driving the networks are different for each neuron type ( Gendrel et al . , 2016; Pereira et al . , 2015 ) . These principles appear to apply to the nervous systems of other organisms as well ( Konstantinides et al . , 2018 ) . However , it remains unknown whether similar mechanistic principles apply to post-transcriptional regulation by RBPs in the nervous system . Here we use single-cell in vivo fluorescent splicing reporters to investigate the cell-specific splicing of sad-1 , a conserved neuronal kinase . The C . elegans sad-1 gene encodes two isoforms that differ in their ability to interact with the F-actin binding protein NAB-1/Neurabin ( Hung et al . , 2007 ) , and have different roles in synapse formation and development ( Kim et al . , 2010 ) . We find that sad-1 undergoes unique splicing patterns in various neuron types , and that developmentally-related cell types ( the ALM touch-sensing neuron and the BDU neuron ) exhibit opposing patterns of splicing ( exon inclusion vs . exon skipping ) . A combination of unbiased genetic screens and candidate targeted mutations identified a cascade of three cell-fate determining TFs and two neuronal RBPs required for proper splicing of sad-1 in ALM neurons . Mechanistic dissection revealed that the three TFs function to establish cell-specific expression of the two RBPs in the ALM neuron , and that the two RBPs in turn directly bind to sad-1 intronic regions to mediate exon inclusion in the ALM neuron . Finally , we find that in other neuron types , similar principles apply but with different combinations of TFs and RBPs mediating sad-1 exon inclusion . These results indicate that neuronal RBPs , like TFs , are employed in a combinatorial code to shape neuron-specific splicing patterns , and demonstrate phenotypic convergence by which different RBPs mediate similar splicing outcomes in various neurons . To identify alternative splicing regulation in individual neuronal cell types , we created two-color splicing reporters that provide a fluorescent readout of splicing regulation in vivo in single cells ( Kuroyanagi et al . , 2006; Orengo et al . , 2006 ) . A minigene representing an alternative splicing event of interest is cloned upstream of a dual GFP/RFP cassette ( Figure 1A–B ) . The GFP and RFP coding sequences reside in alternative reading frames . The alternative exon is engineered to shift the reading frame by +1 nucleotide such that splicing of the alternative exon determines the reading frame , and therefore the translation of GFP versus RFP . Application of two-color fluorescent reporters to transparent organisms such as C . elegans enables in vivo imaging of alternative splicing in individual cells . We have created reporters for splicing events in a number of neuronal genes , and uncovered a rich variety of splicing patterns in single neurons ( Norris et al . , 2014 ) . One intriguing example of neuron-specific alternative splicing is in the conserved neuronal kinase sad-1 , which plays important roles in neuronal development in both worms and mice ( Kim et al . , 2008; Kishi et al . , 2005 ) . In C . elegans , sad-1 is encoded by seventeen exons , and the fifteenth exon is an alternative cassette-type exon ( Figure 1A ) . Alternative splicing of this exon changes the coding sequence and length of the sad-1 C-terminus ( Kim et al . , 2010 ) . This presents an interesting parallel with mice and human genomes , which encode two separate genes homologous to sad-1 ( SAD-A and SAD-B ) that are nearly identical except for their C-terminal coding sequence and length . A two-color splicing reporter for sad-1 in C . elegans revealed that many neurons express both the skipped and included isoforms ( Figure 1C , Figure 1—figure supplement 1 ) . For example , motor neurons in the ventral nerve cord express both isoforms of sad-1 ( Figure 1C ) . On the other hand , the ALM touch-sensing neuron expresses only the included isoform , while the BDU neuron , which is the sister cell to the ALM neuron , expresses only the skipped isoform ( Figure 1C–F ) . While different neurons exhibit differences in sad-1 splicing , the splicing pattern in a given neuron is reproducible and invariant from one animal to the next , suggesting that sad-1 splicing in various neurons is under strict regulatory control . These results led us to ask how ALM and BDU neurons , which are developmentally related ( Figure 1F ) and share a number of anatomical and gene-expression features , specify opposite splicing regimes . To identify regulators of sad-1 splicing in the ALM touch neuron , we performed an unbiased forward genetic screen . Parental worms harboring the sad-1 splicing reporter were mutagenized with EMS . We then screened for F2 animals ( potential homozygotes ) with aberrant expression of the skipped ( GFP ) isoform in the ALM neuron ( Figure 2A ) . This screen identified three distinct loci that transform the splicing pattern from the ALM neuron pattern ( full exon inclusion ) to resemble the pattern in their BDU sister cells ( full exon skipping ) . Whole-genome resequencing of the mutant strains identified loss-of-function mutations in three conserved TFs: unc-86 , mec-3 , and alr-1 ( Figure 2B–F , Figure 2—figure supplement 1 ) . All three genes have previously been identified as key regulators of touch-neuron cell fate ( Gordon and Hobert , 2015; Topalidou et al . , 2011 ) . The three TFs function in a transcriptional cascade ensuring cell-specific expression of mec-3 in touch neurons , which then results in expression of a battery of touch-neuron specific genes ( Figure 2G ) . Loss of the TF mec-3 results in touch neurons ( ALMs ) adopting certain gene-expression characteristics of their sister cells ( BDUs ) ( Gordon and Hobert , 2015 ) , mirroring our observation that loss of mec-3 transforms sad-1 splicing from an ALM ( exon 15 included ) to a BDU ( exon 15 skipped ) pattern . Previous work demonstrates that the MEC-3 TF is expressed only in touch neurons , while UNC-86 and ALR-1 are expressed in various neuron types ( Topalidou et al . , 2011 ) . However , we find that unc-86 and alr-1 mutants affect sad-1 splicing only in the touch neurons ( Figure 2D–F ) . This is in accordance with previous work indicating that a major function of unc-86 and alr-1 in touch neurons is to combinatorially ensure appropriate expression of mec-3 , and that all three TFs are needed for proper differentiation of touch neurons ( Topalidou et al . , 2011 ) . We therefore conclude that the combinatorial activity of all three TFs is required for proper sad-1 splicing in the ALM neuron . We were surprised to identify TFs , but not RBPs , in our forward genetic screen for regulators of sad-1 alternative splicing . We hypothesized that multiple RBPs might co-regulate sad-1 alternative splicing in the ALM neuron and therefore mutations in individual RBPs might result in mild splicing defects . We therefore examined the sequence surrounding the sad-1 alternative exon for conserved cis-elements corresponding to known in vitro RBP sequence preferences ( Ray et al . , 2013 ) . We identified three candidate elements: one corresponding to the mbl-1/Mbnl1 consensus binding motif , and two corresponding to the mec-8/RBMS motif ( Figure 3A–C ) . To test whether these RBPs affect sad-1 alternative splicing , we created deletions for each gene with CRISPR/Cas9 ( Norris et al . , 2017 ) . Both mec-8 and mbl-1 mutants result in aberrant sad-1 splicing in the ALM neuron , displaying partial skipping and partial inclusion ( Figure 3D–F , Figure 3—figure supplement 1 ) . As in the case of the TF mutants , mec-8 mutants affect sad-1 splicing specifically in the ALM neurons , whereas mbl-1 mutants affect sad-1 splicing in ALM neurons as well as specific neurons in the ventral nerve cord ( see Figure 6 , below ) . To verify that the phenotypes of our CRISPR mutants were on-target effects , we crossed the sad-1 splicing reporter into existing alleles for mec-8 ( e398 , premature stop codon [Davies et al . , 1999; Lundquist et al . , 1996] ) and mbl-1 ( wy560 , large deletion affecting multiple genes including mbl-1 [Spilker et al . , 2012] ) . We found these alleles to affect splicing of sad-1 exactly as our CRISPR mutations ( Figure 3—figure supplements 1–2 ) . Whereas TF mutants result in full skipping of the sad-1 alternative exon , RBP mutants result in only partial skipping . This provides a probable explanation for not identifying these RBPs in our genetic screen: partial exon skipping leads to dim GFP expression , which is not sufficiently bright to be noticed upon brief visual inspection . We therefore tested whether simultaneous loss of both RBPs recapitulates the full skipping of sad-1 exon 15 observed in TF mutants . We created mec-8; mbl-1 double mutants expressing the sad-1 splicing reporter . These double mutants result in complete loss of sad-1 exon inclusion in the ALM neuron , recapitulating the splicing phenotype of the single TF mutants ( Figure 3G ) . These results led us to hypothesize that the TFs identified in our screen exert their effects on sad-1 splicing by controlling expression of both mec-8 and mbl-1 . To examine whether the neuronal TFs alter expression of mec-8 and mbl-1 RBPs in the ALM neuron , we created reporter lines for each RBP . To this end , each RBP was C-terminally tagged in a fosmid containing large regions of surrounding genomic context ( Poser et al . , 2008; Spilker et al . , 2012 ) ( Figure 4A–E ) . Compared to traditional transgenic reporters , fosmids are more likely to contain all regulatory information needed to drive normal expression of the gene in question . This is demonstrated in the case of the mec-8 RBP . The classical mec-8::GFP promoter fusion drives expression in a number of cells , but not in the ALM neuron ( Figure 4—figure supplement 1 ) ( Spike et al . , 2002 ) . On the other hand , we detected expression of the mec-8 fosmid reporter in many of the same cells , both neuronal and non-neuronal , plus strong expression in the ALM neuron ( Figure 4A–B ) . A similar fosmid reporter for mbl-1 likewise exhibits expression in the ALM neuron , as well as many other neurons in the nervous system ( Figure 4D , Figure 4—figure supplement 1 ) . This is in line with previous reports on mbl-1 expression ( Spilker et al . , 2012 ) . We tested expression of our reporters in the context of a mec-3 mutant to determine whether expression of mec-8 and mbl-1 in ALM neurons depends on the TF cascade uncovered in our screen . The mec-3 TF is expressed only in touch neurons , and therefore we would expect mec-3 mutants to affect RBP expression only in the touch neurons . Indeed , in mec-3 mutants , expression of both mec-8 and mbl-1 RBPs are abolished in the ALM neuron , while expression in the surrounding neurons and tissues remains unchanged ( Figure 4B–E ) . Together these results indicate that the expression of mec-8 and mbl-1 RBPs are under the control of neuron subtype-specific TFs . To examine whether mec-8 and mbl-1 RBPs might be under direct transcriptional control by one or more of the TFs , we used existing ChIP data for ALR-1 ( Niu et al . , 2011 ) , in vitro derived consensus binding motifs for UNC-86 ( Weirauch et al . , 2014 ) , and a previously-defined UNC-86/MEC-3 heterodimer binding motif ( Röhrig et al . , 2000; Xue et al . , 1993 ) . We did not find conserved UNC-86 binding motifs or an UNC-86/MEC-3 heterodimer binding motif in the promoters for mec-8 or mbl-1 , but did find ALR-1 ChIP peaks in both promoters ( Figure 4—figure supplement 2 ) . This data suggests that alr-1 may directly control transcription of mec-8 and mbl-1 RBPs . The observations that ( 1 ) mec-8; mbl-1 RBP double mutants recapitulate the phenotype of the TF mutants , and ( 2 ) the TFs are necessary for expression of both RBPs in the ALM neuron , together suggest that the splicing defects in the TF mutants are mediated by effects on expression of the two RBPs . Further support for this hypothesis arose indirectly in the course of crossing TF and RBP mutants together . We found that while TF or RBP mutant heterozygotes exhibit normal sad-1 splicing in the ALM neuron , double heterozygotes ( for example alr-1/+; mbl-1/+ , or mec-3/+; mec-8/+ ) exhibit partial exon skipping in ALM , similar to the RBP single mutants ( Figure 4—figure supplement 3 ) . Such ‘non-allelic non-complementation’ is often interpreted to mean that the two genes function in the same complex , or , more likely in this case , function in the same pathway ( Yook et al . , 2001 ) . This indirect evidence further suggests that the TFs and RBPs affect sad-1 splicing as part of the same molecular pathway . If sad-1 splicing is controlled in a linear pathway as suggested by the above series of experiments , with upstream TFs affecting RBP expression in the ALM neuron , then over-expressing an RBP in the context of a TF mutant should partially restore splicing in ALM . To test this hypothesis we created a strain over-expressing a mec-8 transgene specifically in the touch neurons ( pmec-3::mec-8 ) . When introduced into an alr-1 mutant , this transgene partially rescues the splicing of sad-1 in the ALM neuron ( Figure 4F–H ) . Likewise , over-expression of mbl-1 in an alr-1 mutant partially rescues splicing in the ALM neuron ( Figure 4I ) . These results further support a linear gene regulatory pathway in which neuronal fate-determining TFs control neuron-specific expression of RBPs , which then control alternative splicing of sad-1 ( Figure 4J ) . To test whether mec-8 and mbl-1 directly affect splicing by binding to the sad-1 pre-mRNA , we created two-color splicing reporters in which the putative mec-8 or mbl-1 cis-elements are mutated ( Figure 3A and Figure 5 ) . If the RBPs act directly by binding the cis-element , then mutation of the cis-element should affect the splicing pattern in a manner resembling the wild-type splicing reporter in the context of the RBP deletion mutant . If the RBPs act indirectly , mutating the cis-element should have no effect on the splicing pattern . Mutation of the mbl-1 cis-element resulted in ALM neurons with altered sad-1 splicing in which the exon is partially skipped and partially included ( Figure 5A–B ) . This recapitulates the phenotype of mbl-1 null mutations ( Figure 3F ) , suggesting that mbl-1 exerts its effects on splicing directly through binding a conserved cis-element in the upstream intron . We identified two consensus mec-8 binding motifs in conserved regions in the intron downstream of the cassette exon . We therefore created splicing reporters mutant for both cis-elements as well as for each element individually . The splicing reporter mutant for both elements recapitulates the splicing phenotype of mec-8 null mutants ( Figure 5E ) . Likewise , mutating either mec-8 binding site in isolation recapitulates a mec-8 null mutation ( Figure 3E and Figure 5C–D ) , suggesting that mec-8 binding to both cis-elements is required for appropriate sad-1 splicing . We tested whether mutation of a putative cis-element could be rescued by over-expression of its cognate RBP , and found that cis-element mutants were not rescued by RBP over-expression ( Figure 5—figure supplement 1 ) , providing further evidence that the RBPs act directly on the sad-1 pre-mRNA . Together these results indicate that mec-8 and mbl-1 RBPs combinatorially ensure sad-1 exon inclusion in ALM neurons through direct interactions with the neighboring introns . Having identified regulatory mechanisms controlling sad-1 splicing in the ALM neuron , we next wondered whether similar principles apply in other neuron types . Most neurons besides the ALM and BDU neurons express both skipped and included sad-1 isoforms . This could represent the neuronal ‘ground state’ of splicing in the absence of cell-specific splicing regulators . On the other hand , our observations that loss of both mec-8 and mbl-1 in the ALM neuron results in full exon skipping suggest that the ground state may be complete exon skipping . This hypothesis predicts that other neurons in which sad-1 is partially included express one or more RBPs mediating exon inclusion . In the course of examining sad-1 splicing in ALM neurons , we noticed that mbl-1 mutants affect sad-1 splicing not only in ALM , but also in the excitatory cholinergic motor neurons of the ventral nerve cord ( Figure 6A–D ) . Whereas mbl-1 mutants cause a change in sad-1 splicing from full inclusion to partial inclusion in ALM neurons , in excitatory motor neurons mbl-1 mutants shift from partial inclusion to no inclusion ( Figure 6C–D ) . On the other hand , the inhibitory motor neurons remain unaffected in mbl-1 mutants , expressing both the included and skipped isoforms ( Figure 6D , arrowheads ) . This is consistent with our mbl-1 gene expression reporter , which reveals expression of mbl-1 in the excitatory motor neurons , but not in the inhibitory motor neurons ( Figure 6—figure supplement 1 ) . We did not detect mec-8 expression in motor neurons of the ventral nerve cord , and mec-8 mutants had no effect on splicing of sad-1 in motor neurons ( Figures 3E and 4B ) . It therefore seems that in neurons expressing mbl-1 such as excitatory motor neurons , the presence of mbl-1 mediates partial exon inclusion . In neurons expressing both mbl-1 and mec-8 such as ALM touch neurons , the two RBPs together mediate full inclusion . In mbl-1 mutants , sad-1 exon inclusion is lost in excitatory neurons but remains in inhibitory motor neurons . We therefore wondered whether there was an additional RBP expressed in inhibitory motor neurons mediating sad-1 inclusion . mec-8 was ruled out because it is not expressed in inhibitory motor neurons and does not affect sad-1 splicing in the nerve cord . On the other hand , the RBP msi-1/Musashi has been reported to be expressed in inhibitory but not excitatory neurons of the nerve cord ( Yoda et al . , 2000 ) , which is a mutually exclusive pattern with mbl-1 . We therefore tested msi-1 as a candidate for the RBP mediating sad-1 exon inclusion in the inhibitory motor neurons . We generated a msi-1 deletion mutant , which shows loss of sad-1 inclusion specifically in the inhibitory motor neurons ( Figure 6E ) . Furthermore , msi-1; mbl-1 double mutants result in complete loss of exon inclusion in the ventral nerve cord ( Figure 6F ) . These results indicate that mbl-1 and msi-1 act in distinct cell types to achieve partial sad-1 exon inclusion throughout the ventral nerve cord . We suspect that msi-1 , like mbl-1 and mec-8 , directly affects sad-1 splicing by binding in the intronic regions surrounding the alternative exon . in vitro experiments have identified a UAG motif ( Figure 6H ) ( Ray et al . , 2013 ) , usually in bipartite form ( e . g . UAGNNUAG ) ( Dominguez et al . , 2018 ) , as the consensus binding motif for msi-1 . There is a conserved bipartite UAG motif in the intron downstream of the sad-1 cassette exon ( Figure 6G–H ) , and we hypothesize that msi-1 binds there to mediate exon inclusion in inhibitory motor neurons . Together the results from three different neuronal cell types ( ALM neuron , excitatory motor neurons , and inhibitory motor neurons ) constitute an example of ‘phenotypic convergence , ’ in which phenotypic similarity between cells is generated by distinct molecular mechanisms . Substantial evidence of such phenotypic convergence exists for TFs controlling neuronal properties in worms and flies ( Gendrel et al . , 2016; Konstantinides et al . , 2018; Pereira et al . , 2015 ) . Our results now extend this principle to RBPs and their control of alternative splicing , revealing phenotypic convergence in which similar splicing patterns ( i . e . sad-1 exon inclusion ) are generated in various neurons by diverse RBPs acting in specific neuronal subtypes ( Figure 7 ) . To further examine this principle , we tested whether ectopic expression of an RBP in a neuron type in which it is not normally expressed would be sufficient to alter sad-1 splicing in that neuron . We expressed mec-8 in excitatory motor neurons ( where normally only mbl-1 is expressed ) and found that mec-8 expression is sufficient to alter sad-1 splicing patterns from partial inclusion to full inclusion specifically in the excitatory motor neurons ( Figure 6—figure supplement 2A ) . Similarly , mbl-1 expression in inhibitory motor neurons ( where normally only msi-1 is expressed ) results in full exon inclusion [Figure 6—figure supplement 2A] ) . Finally , we asked whether phenotypic convergence occurs simultaneously at multiple levels ( TFs and RBPs ) with regard to sad-1 splicing . To do so we examined mutants for the TF unc-3 , which controls the fate of excitatory motor neurons in the ventral nerve cord ( Kratsios et al . , 2012 ) , analogous to ALM cell fate determination by unc-86/mec-3/alr-1 . In unc-3 mutants , sad-1 exon inclusion is lost in excitatory motor neurons , similar to mbl-1 RBP mutants ( Figure 6D , Figure 6—figure supplement 2B ) . However , whereas unc-86/mec-3/alr-1 mutants exhibit completely-penetrant loss of sad-1 exon inclusion , unc-3 mutants exhibit partially-penetrant defects , ranging from moderate to complete loss of sad-1 inclusion in excitatory motor neurons . Similarly , loss of unc-3 results in partially-penetrant defects in mbl-1 expression ( Figure 6—figure supplement 2C ) . Together these results demonstrate that phenotypic convergence among different neuron types occurs simultaneously at multiple layers of gene regulation: different TFs ( e . g . mec-3 and unc-3 ) specify expression of different RBP complements ( e . g . mbl-1 and mec-8 ) which have a common function of mediating sad-1 exon inclusion . In this study we find that sad-1 splicing undergoes precise regulation in numerous neuronal types . Although ALM and BDU neurons are sister cells , express many of the same genes , and share a number of cell-specific TFs , they have opposing patterns of sad-1 splicing . This highlights the fact that post-transcriptional control can further diversify attributes of single cells on top of the more well-known role of transcriptional control . Our results demonstrate that sad-1 splicing is regulated according to a combinatorial RBP code , with different splicing outcomes depending on whether a cell expresses zero , one , or two neuron-specific RBPs ( Figure 7 ) . This suggests that the ‘default’ outcome of sad-1 splicing is full skipping of the cassette exon , as observed in the BDU neuron which does not express any of the sad-1-regulating RBPs . Only cells with at least one RBP mediating exon inclusion express sad-1 included isoforms . Cells with multiple such RBPs ( e . g . the ALM neuron ) express only the included isoform . In previous work we found that alternative splicing of the kinase unc-16/ JIP3 in motor neurons is likewise controlled by a pair of RNA binding proteins ( Norris et al . , 2014 ) . However , unc-16 splicing and sad-1 splicing in motor neurons are regulated by distinct pairs of RBPs . Whereas sad-1 splicing in motor neurons is regulated by mbl-1 and msi-1 RBPs , unc-16 is combinatorially regulated by unc-75 and exc-7 in motor neurons ( Norris et al . , 2014 ) . This suggests that even within a single neuron type , different splicing events are regulated by different complements of RBPs . The importance of TFs controlling gene expression networks in single neurons is well established , and the importance of RBPs controlling post-transcriptional networks in single cells is gaining wider appreciation ( Norris and Calarco , 2012; Norris et al . , 2014; Song et al . , 2017; Wamsley et al . , 2018 ) . How these two modes of regulation might interact remains understudied . Here we show that the two modes of regulation interact in a traditional linear type of pathway . A combination of cell-specific TFs establishes a transcriptional network in a single neuron type . This network includes a specific combination of neuronal RBPs , and the particular combination of RBPs in a given neuron then establish a unique post-transcriptional gene regulatory network in that neuron . Multiple layers of regulatory control can thus increase the diversity of single neuron transcriptomes and fine-tune the properties of individual neurons . In the present study we have identified a linear pathway in which TFs influence the expression of RBPs , which then influence alternative splicing in single neurons . This adds to a substantial body of literature finding that RBPs can affect the function of specific TFs by modulating their alternative splicing ( Calarco et al . , 2009; Han et al . , 2013; Linares et al . , 2015; Raj et al . , 2011 ) . In the future it will be interesting to see whether additional regulatory logics exist between TFs and RBPs . Single-neuron TF combinations have been identified with a variety of feedback and feedforward mechanisms resulting in interesting regulatory properties ( Mangan and Alon , 2003 ) , and in principle TFs and RBPs could likewise interact in complex ways , leading to an even greater array of diversification strategies ( Han et al . , 2017 ) . Together this study highlights the importance of considering neuron-specific ‘combinatorial codes’ not only from the perspective of TF combinations , but the specific complement of both TFs and RBPs shaping the transcriptome of a given neuron . A theme emerging from recent studies of single-neuron transcriptomes is ‘phenotypic convergence , ’ in which multiple neurons share gene expression similarities , but the regulatory mechanisms by which they do so are distinct in each neuron . For example , in worms , cholinergic neuron cell fate and core cholinergic gene expression properties are controlled by different combinations of TFs in different cholinergic neuron sub-types ( Pereira et al . , 2015 ) . This is also the case for other neuron types in C . elegans ( Gendrel et al . , 2016 ) . More recently , phenotypic convergence has been reported for TFs in neurons of the Drosophila optic lobe ( Konstantinides et al . , 2018 ) , indicating that phenotypic convergence mediated by TFs is a widespread phenomenon . We now extend this principle of phenotypic convergence to the regulation of splicing by RBPs as well . sad-1 exon inclusion is mediated in various neuron types , with a unique complement of RBPs responsible for exon inclusion in each specific type that we have studied ( ALM neuron , inhibitory motor neurons , and excitatory motor neurons; Figure 7 ) . This likely represents phenotypic convergence on multiple levels , as the RBPs regulating splicing are different in each neuron , and the TFs regulating RBP expression are likewise different in each neuron . Each of these levels coordinately converges upon appropriate splicing of sad-1 in each neuron type . Additional neuron types with similar sad-1 splicing patterns ( see Figure 1 and Figure 1—figure supplement 1 ) may represent additional examples of phenotypic convergence whose underlying mechanisms remain unexplored . C . elegans were maintained under standard conditions ( Brenner , 1974 ) at 20°C on nematode growth media ( NGM ) plates seeded with OP50 E . coli bacteria . New transgenic worms were generated by microinjection with 15 ng/μl transgene and 15 ng/μl co-injection marker ( either rgef-1 , unc-17 , or unc-25 promoter driving BFP ) . The forward mutagenesis screen was performed on animals harboring the sad-1 exon 15 splicing reporter with EMS at 47 mM for 4 hr . F1s were picked onto new plates , 10 F1s per plate . After 3–4 days of growth , F2s were screened by eye on the Zeiss Axiozoom . V16 for touch cells appearing in the GFP channel ( representing aberrant exon skipping ) and were then verified for a concomitant loss of RFP ( representing loss of exon inclusion ) . Such worms were picked individually onto a new plate to verify the phenotype in the F3 generation and to establish a clonal population . After outcrossing , strains were subjected to whole-genome resequencing ( Illumina , 1 × 75 bp ) and potential causative mutations were identified using the CloudMAP workflow on the Galaxy web platform ( Minevich et al . , 2012 ) . A total of approximately 6000 haploid genomes were screened . Targeted mutant strains were generated using CRISPR/Cas9 as previously described ( Calarco and Norris , 2018; Norris et al . , 2015 ) , such that the gene of interest is deleted and is replaced with a heterologous GFP reporter under the control of a pharyngeal promoter ( pmyo-2 ) which does not interfere with the visualization of the sad-1 splicing reporter in the ALM , BDU or ventral nerve cord neurons . Seamless gene replacement was verified by PCR amplification and Sanger sequencing of both junction boundaries . Images were obtained with a Zeiss Axio Imager . Z1 and processed in ImageJ . sad-1 minigenes were created using the following primers: Forward 5’ GATAAAACTGAAACAACTTCTGC and Reverse 5’ GGGGTTGGCGATTTGTATGAGaTAGC . Restriction sites were appended to both the forward primer ( XhoI ) and reverse ( NotI ) primers to facilitate cloning into a Gateway-compatible vector as previously described ( Norris et al . , 2014 ) . The reporter was then cloned downstream of a pan-neuronal rgef-1 promoter , as endogenous sad-1 has been detected broadly throughout the nervous system ( Crump et al . , 2001 ) . Mutant versions of the splicing reporter were synthesized de novo then cut with XhoI and NotI and cloned as above . Some strains were provided by the Caenorhabditis Genome Center , which is funded by the NIH Office of Research Infrastructure Programs ( P40 OD010440 ) . Other strains were provided by the National BioResource Project ( Tokyo ) .	All the cells in the human nervous system contain the same genetic information , and yet there are many kinds of neurons , each with different features and roles in the body . Proteins known as transcription factors help to establish this diversity by switching on different genes in different types of cells . A mechanism known as RNA splicing , which is regulated by RNA binding proteins , can also provide another layer of regulation . When a gene is switched on , a faithful copy of its sequence is produced in the form of an RNA molecule , which will then be ‘read’ to create a protein . However , the RNA molecules may first be processed to create templates that can differ between cell types: this means that a single gene can code for slightly different proteins , some of them specific to a given cell type . Yet , very little is known about how RNA splicing can generate more diversity in the nervous system . To investigate , Thompson et al . developed a fluorescent reporter system that helped them track how the RNA of a gene called sad-1 is spliced in individual neurons of the worm Caenorhabditis elegans . This showed that sad-1 was turned on in all neurons , but the particular spliced versions varied widely between different types of nerve cells . Additional experiments combined old school and cutting-edge genetics technics such as CRISPR/Cas9 to identify the proteins that control the splicing of sad-1 in different kinds of neurons . Despite not directly participating in RNA splicing , a number of transcription factors were shown to be involved . These molecular switches were turning on genes that code for RNA binding proteins differently between types of neurons , which in turn led sad-1 to be spliced according to neuron-specific patterns . The findings by Thompson et al . could provide some insight into how mammals can establish many types of neurons; however , a technical hurdle stands in the way of this line of research , as it is still difficult to detect splicing in single neurons in these species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "genetics", "and", "genomics" ]	2019	Splicing in a single neuron is coordinately controlled by RNA binding proteins and transcription factors
Hallucinations occur in both normal and clinical populations . Due to their unpredictability and complexity , the mechanisms underlying hallucinations remain largely untested . Here we show that visual hallucinations can be induced in the normal population by visual flicker , limited to an annulus that constricts content complexity to simple moving grey blobs , allowing objective mechanistic investigation . Hallucination strength peaked at ~11 Hz flicker and was dependent on cortical processing . Hallucinated motion speed increased with flicker rate , when mapped onto visual cortex it was independent of eccentricity , underwent local sensory adaptation and showed the same bistable and mnemonic dynamics as sensory perception . A neural field model with motion selectivity provides a mechanism for both hallucinations and perception . Our results demonstrate that hallucinations can be studied objectively , and they share multiple mechanisms with sensory perception . We anticipate that this assay will be critical to test theories of human consciousness and clinical models of hallucination . Hallucinations occur across a wide range of pathologies and are also common in non-clinical populations ( Barrett , 1993; 1994 ) . Generally , hallucinations are defined as an involuntary percept-like experience in the absence of an appropriate direct stimulus ( Bentall , 1990 ) . However , very little is known about the mechanisms underlying hallucinations , due largely to the methodological constraints caused by their inherently subjective , constantly changing heterogeneous content . Visual hallucinations are thought to arise in exceptional circumstances when external stimuli are overwhelmed by internally generated spontaneous patterns of neural activity . This situation occurs when the parameters governing normal visual function are altered due to changes in brain anatomy or physiology ( Ffytche , 2008; Butler et al . , 2012 ) , state changes such as dreaming or migraines ( Llinás and Ribary , 1993; Aurora and Wilkinson , 2007 ) , psychotropic drugs that temporarily perturb normal cortical function , or empty full field luminance flicker ( Passie et al . , 2008; Billock and Tsou , 2012 ) . However , across these classes of hallucination , understanding has been severely limited by the multi-feature ( color , form and motion ) heterogeneous content that changes unpredictably over time , and typically requires subjective reports or subsequent depiction such as drawing to communicate subjective experience ( Allefeld et al . , 2011 ) . To study visual hallucinations , we constrained empty-field flicker to a thin annulus that was centred on the fovea ( Figure 1A ) . This stimulus effectively constrained the hallucinated forms to one spatial dimension . 10 . 7554/eLife . 17072 . 003Figure 1 . Hallucination stimulus and data . ( A ) Physical stimulus and depiction of the percept . ( B ) Depiction of the stimulus used to measure the effective contrast . The small inner annulus is the perceptual , while the larger outer annulus shows a depiction of the hallucinated content . ( C ) Hallucination depiction and nonius lines used to measure the effective rotation propagation times . ( D ) Effective contrast data showing mean point of subjective equivalence between the perceptual and hallucinated content as a function of flicker frequency ( Exp . 1; N = 42 , 56 trials per frequency ) . ( E ) Data showing interocular interaction ( Exp . 2; N = 20 , 56 trials per combination of synchrony and frequency ) . Synchronous and asynchronous flickering annuli give different contrasts measures . ( F ) Hallucination motion speed measures using stimulus in C , as a function of flicker rate ( Exp . 3A; N = 6 , 40 trials per frequency ) . ( G ) Dependence of propagation times on cortical distance ( Exp . 3B; N = 6 , 40 trials per eccentricity condition ) . Distance around the annulus was converted into cm across cortex using the formulae from ( Horton and Hoyt , 1991 ) . Main effect of distance F ( 2 , 5 ) = 13 . 74; p=0 . 001 . ( H ) The effect of number of physical gaps in the annulus ( Exp . 4A; N = 4 , 10 trials per stimulus ) . ( I ) The effect of the width of physical gaps in the annulus ( Exp . 4B; N = 4 , 10 trials per stimulus ) . ( J ) Hallucinated structure and motion is reduced at isoluminance ( Exp . 5; N = 5 , 20 trials per luminance condition ) . All error bars show ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 003 When a white annulus was flickered on/off on a black background ( ~2–30 Hz ) , we noticed that light grey blobs appeared and rotated around the annulus , first in one direction then the other ( Figure 1A; see Videos 1 and 2 ) . Unlike full field luminance flicker stimulation whose content changes as a function of flicker frequency ( Allefeld et al . , 2011; Mauro et al . , 2015 ) , the light grey blobs remained clearly observable across the range of oscillation frequencies tested . This reduces the visual feature dimensions and overcomes many of the prior limitations set by the multi-feature heterogeneous content in pathological , spontaneous and full field flicker induced hallucinations . 10 . 7554/eLife . 17072 . 004Video 1 . An animated movie representation of one of our stimuli . Under the right viewing conditions , you may experience light grey blobs ( that are not physically presented in the movie ) appearing around the flickering annulus . This video contains flashing and alternating images , and therefore might not be suitable for readers with photosensitive epilepsy . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 00410 . 7554/eLife . 17072 . 005Video 2 . A perceptual , retina-based representation approximating what some see in the otherwise empty flickering white annulus of Video 1 . Note , this is only one interpretation , the individual hallucination experience may vary from individual to individual . This video contains flashing and alternating images , and therefore might not be suitable for readers with photosensitive epilepsy . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 005 To measure the strength of this hallucination , we devised a technique allowing us to assess the effective contrast with a two alternative forced choice procedure . We presented an interior annulus housing physical sinusoidal luminance modulation simultaneously with the flickering annulus ( Figure 1B ) . We presented this physical retina-sourced annulus at a range of different contrasts and participants reported whether it was higher or lower in contrast than any content in the flickering empty annulus . Across two experiments , data from 28 and 24 subjects were fit with cumulative Gaussian functions to give an estimate of the point of subjective contrast equivalence . This gave us a proxy of the effective contrast of the hallucinated blob structures across different flicker frequencies . Figure 1D shows the effective contrast of the hallucinations as a function of flicker frequency . Contrast estimates peaked around 11 Hz , then continued to decline slowly as a function of frequency ( main effect frequency: exp . 1A F ( 3 , 81 ) = 3 . 42 , p=0 . 021; exp . 1B: F ( 3 , 69 ) = 5 . 81 , p=0 . 001 ) . One proposition is that flicker induced hallucinations might be largely due to an interaction between perception and retinal after effects ( Bidwell , 1897 ) . To test for a cortical contribution to the hallucinated blobs , we devised a version of the hallucination-contrast experiment that depended on cortical interocular cross-talk . We presented two small annuli , one to each eye using a mirror stereoscope , and flickered both rings either synchronously or asynchronously at 2 . 5 Hz to 20 new participants . In the asynchronous condition binocular neurons should receive flicker stimulation at ~5 Hz ( rather than 2 . 5 Hz ) . Accordingly , if the hallucinated content is the product of binocular neurons , viewers should experience higher contrast in the asynchronous condition as 5 Hz is closer to the 11 Hz peak in contrast , we found in the first experiment . Figure 1E shows exactly this; hallucinations in the asynchronous 2 . 5 Hz condition were perceived at a higher contrast than in the synchronous condition ( t ( 19 ) = 2 . 60 , p=0 . 018 ) . Likewise , the same experiment performed at 21 Hz yielded the opposite pattern of results , the synchronous condition gave higher contrast values ( t ( 19 ) = 2 . 28 , p=0 . 034; interaction: F ( 1 , 19 ) = 8 . 06 , p=0 . 011 ) , as hallucinations produced by >21 Hz flicker were perceived as lower in contrast in our first experiment ( Figure 1D ) . Together these data suggest that flicker induced hallucinations transpire at or beyond binocular neurons and cannot be the sole product of retinal afterimages . To measure the motion dynamics of rotation in this hallucination , we devised a technique allowing us to quantify the speed of rotation similar to that used in ( Wilson et al . , 2001 ) . Observers depressed and held a key when a monitored section of the rotating hallucinatory pattern passed a nonius line at the top of the annulus ( Figure 1C ) . Observers then released the key only when the monitored section of the pattern reached a bottom nonius line , marking a travelled distance of half a rotation . This technique gave us propagation times for a fixed annulus distance , hence the speed of the travelling hallucination . The rotational speed was dependent on the annulus flicker rate ( Figure 1F ) , with higher frequencies giving faster rotation speeds ( F ( 2 , 10 ) = 24 . 75 , p<0 . 001 ) . To learn how rotation speed varies with eccentricity , we scaled our entire stimulus across three different mean annular radii and using the same timing procedure mapped speed in 6 observers . Using the same speed procedure as above , the mean speed for eccentricities of 6 . 19° , 7 . 96° , and 9 . 67° ( mean radii ) were 34 , 33 and 33° s−1 . Based on the hypothesis that the hallucinated structure and motion originates in primary visual cortex we converted visual distance into physical distance on the cortex in cm using the detailed surface map of human V1 ( Horton and Hoyt , 1991 ) . Figure 1G shows longer propagation times as a function of greater neural distance , corresponding to a mean propagation speed of 5 . 6 cm s−1 over V1 surface . Next we tested whether the hallucinations could propagate across gaps in the flickering annulus stimulus . We added 4 , 8 , and 12 permanent gaps ( 0 . 59° width ) into the flickering annulus at cardinal locations and sub-cardinal divisions ( Figure 1H ) . Eight observers monitored the direction of motion of the hallucinated blobs by holding one of the designated keys down or reported no clear direction by releasing keys for multiple 60-second durations . Figure 1H shows the percentage of time observers reported rotational motion vs . stationary patterns as a function of gap number . There was a clear trend of less rotation with a greater number of gaps ( F ( 2 , 6 ) = 14 . 68 , p=0 . 005 ) . Next we held the number of gaps constant at four , and manipulated gap size across three different values ( 0 . 59° , 1 . 78° , 2 . 97° ) , while observers again tracked hallucinatory motion or its absence . Again , the percentage of time rotational motion was perceived went down as a function of gap size ( F ( 2 , 6 ) = 15 . 93 , p=0 . 004 ) ( Figure 1I ) . To learn if these hallucinations are specific to luminance flicker or generalize to isoluminant color flicker , we ran a new experiment with five observers who tracked static shapes , moving shapes , or no structure at all , for both the standard luminance flicker and subjective isoluminant red and green annuli . We first assessed subjective isoluminance in each observer using the flicker fusion technique ( Wagner and Boynton , 1972 ) . Using these color values we flickered annuli at 6 or 12 Hz for 30 s . Figure 1J shows the percentage of time reported for each category ( no shape , shape only , and shape + motion ) for luminance and isoluminance stimuli collapsed across flicker frequency ( main effect and all interactions for flicker frequency: all ps > 0 . 29 ) . Strikingly , unlike luminance stimuli , observers reported no pattern for over 60% of the time at isoluminance ( F ( 2 , 8 ) = 39 . 67 , p<0 . 001 ) . These data suggest a neural locus sensitive to luminance , but not color flicker , likely the dorsal visual processing stream , which has a higher proportion of cells blind to isoluminance ( Shapley , 1990 ) . However , we only tested isoluminant red/green , so it remains unknown if this pattern would extend to blue/yellow stimuli . Over the past several decades researchers have been fascinated by perceptual bistability as a method to study the neural correlates of consciousness and how the brain makes decisions ( Blake and Logothetis , 2002 ) . Accordingly , we wondered if the motion in this hallucination might indeed be bistable and hence open a new window into the study of how the brain makes choices for conscious experience . First , to investigate if people hallucinated clockwise and counter-clockwise motion equal proportions of time , 9 observers tracked rotation direction for 10 min . Figure 2A shows equivalent percentages of time reported for each motion direction ( CW vs CCW: t ( 8 ) < 1 , p=0 . 53 ) . One classic hallmark of bistability is that the distribution of dominance durations exhibit a long tail ( occasional long durations ) , forming a gamma-like distribution ( Blake and Logothetis , 2002 ) . Figure 2B shows a clear long tail for dominance durations for the hallucinated bistability , consistent with the core characteristic of other bistable perceptual phenomena . 10 . 7554/eLife . 17072 . 006Figure 2 . Bistability of the hallucination . ( A ) Data showing that subjects experience the hallucination rotation equally in clockwise and counter-clockwise directions ( Exp . 6; N = 9 , 10 min of tracking ) . ( B ) A histogram of dominance durations , showing the long tail and fit with a gamma function , a hallmark of perceptual bistability ( Exp . 6; from the gamma fit: a = 2 . 49; b = 0 . 40 ) . ( C ) Depiction of the motion probe stimulus . Motion probes were presented moving clockwise or counter-clockwise , while subjects tracked hallucination alternations . ( D ) Contrast thresholds from the probe stimulus for congruent and incongruent probes ( Exp . 7; N = 13 , 384 trials ) . ( E ) Longer dominance durations over viewing time suggest a form of adaptation that is local in visual space , the second ring ‘resets’ the adaptation ( Exp . 8A; N = 7 , 10 min tracking per stimulus order ) . ( F ) Dominance durations over time for three different flicker frequencies ( Exp . 8B; N = 6 , 10 trials per frequency ) . ( G ) 8 individual subjects and the mean stability for intermittent physical presentations ( Exp . 9; 100 trials ) . ( H ) Comparing dominance durations in binocular rivalry , rotation globe-stimulus and flicker hallucinations ( Exp . 10; N = 84 , 2 trials of two minutes tracking ) . Retest reliability: flicker: r =0 . 736; globe: r =0 . 741; rivalry: r =0 . 744; all ps <0 . 0001 . All error bars show ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 006 To obtain more objective performance based measures of this rotational bistability we tested whether sensitivity to retina-sourced physical motion stimuli presented within the annulus might differ when presented congruently or incongruently with the hallucinated motion . This would demonstrate an interaction between hallucinated content and physical discrimination – supporting a common mechanism . A new set of 10 participants tracked hallucinated motion alternations for periods of 10 s . At a random time-point during the final 5 s of tracking a physical motion probe ( Figure 2C; see Materials and methods ) was presented in the annulus at either left or right of fixation and participants had to report on which side the probe was presented ( a two alternative forced choice task ) . The probe was presented at one of six different contrasts ( 9 . 5% , 10% , 11% , 13% , 17% , 25% ) , set during pilot tests . Probe data was then separated based on probe direction , congruent or incongruent with the concurrent hallucination direction , and probe accuracy was fit with a non-linear function to give a threshold estimate of 70% accuracy ( see Materials and methods ) . Figure 2D shows a significant difference between mean contrast thresholds for congruent and incongruent trials , with greater sensitivity to the probe stimulus when it was rotating in a congruent direction with the hallucinatory percept ( t ( 9 ) = 2 . 8 , p=0 . 02 ) . This suggests that the hallucinatory motion percept ( or tracking it ) weakly suppresses or boosts retina-based motion depending on the motion congruity . Next we wondered if hallucinated bistability , like perceptual retina-sourced bistability such as binocular rivalry , undergoes a local form of adaptation resulting in longer dominance durations over time ( Suzuki and Grabowecky , 2007 ) . For example , binocular rivalry alternations slow during viewing or when the stimulus is moved through visual space ( Blake et al . , 2003 ) . Participants continuously tracked hallucinated alternations for a 5-min period in a small annulus , immediately followed by a further 5 min of tracking in a larger annulus , that did not spatially overlap with the prior stimulus ( order was counter-balanced ) . Figure 2E shows that dominance durations increased over time during a session of same-sized stimulus . Further , in the second period of hallucination tracking , the new different-sized and non-overlapping annulus did not ‘follow on’ from the slower dynamics , but returned to the original shorter durations . These data suggest that hallucinated bistability can undergo a form of sensory adaptation that is local in visual space . To test if these changes in alternation rate were due to adaptation to the hallucinated content or due to adaptation to the actual perceptual flicker , six participants tracked alternations for 10 consecutive 30 s periods at three different flicker rates ( 8 . 5 , 10 . 6 and 14 . 2 Hz ) . Figure 2F again shows adaptation , with an increase in dominance durations over time , however across the whole period the alternation rate was not statistically different between the three flicker rates ( F ( 2 , 24 ) = 1 . 76 , p=0 . 19 ) , suggesting the change in alternation rate was not due to flicker adaptation , but most probably due to neural fatigue in neurons representing the hallucinatory bistable structure . Another hallmark of perceptual bistability is the striking stabilization of the normal continuous stochastic dynamics by a sensory memory between intermittent presentations ( Pearson and Brascamp , 2008 ) . To learn if these bistable hallucinations show a similar sensory memory we presented the flickering annulus to eight participants for 2 s followed by 4 s without flicker ( repeating intermittent presentation ) . Participants reported the dominant rotation direction on each 2 s presentation for 100 trials . Figure 2G shows the hallucination motion stability as the percentage of reported motion direction consecutively the same ( e . g . clockwise , clockwise ) , such that reporting the same percept on every trial would result in 100% stability ( Pearson and Brascamp , 2008 ) . All individual subjects show a stability measure above the chance score of 50% ( two rotation directions ) , with the mean significantly above chance ( 75%; compared to 50%: p<0 . 001 ) , suggesting that hallucinated bistability can be stabilized by a novel form of memory across intermittent presentations . Finally , to probe for potential overlapping mechanisms between hallucinated and perceptual or retina-sourced bistability 74 new participants tracked alternations in binocular rivalry , a bistable rotating sphere ( see Materials and methods ) and the flicker-induced hallucinations . Figure 2H shows a scatter plot of the data , the rotating sphere ( red ) significantly predicted hallucinated alternation rates ( r = 0 . 312; p=0 . 006 ) , while binocular rivalry ( blue ) did not ( r = 0 . 133; p=0 . 228 ) . We propose that the predictive relationship between the rotating sphere and hallucinated bistability , but not rivalry , might be due to the former two both involving a common neurophysiological mechanism for motion . Next we extend a model developed for full field flicker hallucination to explain both the constrained hallucinated content and its bistability ( Rule et al . , 2011 ) . This model is based on the idea that uniform luminance flicker stimulation resonates with the natural frequency of cortical cells to evoke standing waves of activity in primary visual cortex that induce the conscious experience of the hallucination ( Billock and Tsou , 2012; Rule et al . , 2011; Ermentrout and Cowan , 1979 ) . We modeled the region of visual cortex that was stimulated by the flickering annulus as a ring of tissue in one spatial dimension ( Figure 3A ) . The neural tissue was modelled using an established neural field model ( see Appendix 1: Equations 1–3 ) comprising spatially-coupled populations of excitatory ( E ) and inhibitory ( I ) cells ( Figure 3B , top ) . The standing waves/hallucinations arise when flicker frequencies are approximately twice the natural frequency of the damped oscillations in the neural dynamics ( Rule et al . , 2011 ) . The resonant frequency of simple cells in the primary visual cortex of cat typically peaks near 5 Hz ( Movshon et al . , 1978 ) . Here we tuned the natural frequency of the model to 5 . 5 Hz so that it elicited prominent standing waves with 11 Hz flicker to match our behavioral data ( Figure 1D ) . 10 . 7554/eLife . 17072 . 007Figure 3 . A neural field model of flicker-induced bistable motion . ( A ) The log-polar retinotopic map of human visual cortex . The fovea is located at the centre . The annulus stimulus maps onto a thin strip of tissue ( shaded ) that spans both hemispheres . Inter-hemispheric fibres ( dotted lines ) connect the strips to form a contiguous ring . ( B ) Schematic of the model . The ring of tissue is modelled in one spatial dimension using an established neural field model ( Rule et al . , 2011 ) . that comprises local populations of excitatory ( E ) and inhibitory ( I ) cells . Flicker and counter-phase stimulation both induce counter-phase responses in this model . Motion within the cortical response patterns was detected by banks of Gabor filters ( large arrows ) following the motion-energy model . The motion-energy signals ( boxes ) represented the percepts of LEFT ( anti-clockwise ) and RIGHT ( clockwise ) motion . Those percepts were subject to rivalry through mutual inhibition and firing rate adaptation . ( C ) Space-time plots of flicker . ( D ) Cortical responses to flicker . ( E ) Time-averaged LEFT and RIGHT motion-energy responses to flicker stimulation . Error bars are ±1 standard deviation . ( F ) Time course of the perceptual decisions evoked by flicker . ( G ) Histogram of dominance durations for switching between left and right motion percepts fit with a gamma function . The variation in switch times is due to injected noise in perceptual rivalry model . ( H–L ) The analogous representations for the counter-phase retina-sourced stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 00710 . 7554/eLife . 17072 . 008Figure 3—figure supplement 1 . Motion percepts in the model under four different stimulus conditions . ( A–D ) Space-time plots of the four stimuli . ( E–H ) Space-time plots of the corresponding cortical responses . ( I–L ) Time- averaged responses of the LEFT and RIGHT motion detectors . ( M–P ) Time courses of the perceptual decisions . The two left most columns reproduce the results of the main text , namely the comparison of illusory motion induced by flicker and counter-phase stimulation . The two rightmost columns demonstrate the absence of illusory motion in response to true motion and slow strobe-like flicker stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 17072 . 008 Motion signals were then extracted from the space-time signatures of the standing waves in cortex using two banks of direction-selective motion detectors ( Figure 3B , middle ) . These detectors ( see Appendix 1: Equations 8–10 ) were implemented using Gabor filters ( Jones and Palmer , 1987 ) according to the classic motion-energy model ( Adelson and Bergen , 1985 ) . Given that our interest was in hallucinated motion , we applied the motion-energy model directly to the cortical activity patterns rather than to the stimulus pattern . The output of the motion detectors within each bank were linearly combined to form a net motion-energy signal that represented the valence of motion in the detector’s preferred direction . These net motion signals were then fed into distinct neural populations that encoded the perceptual states of Left ( counter-clockwise ) and Right ( clockwise ) motion respectively ( Figure 3B , bottom ) . Perceptual rivalry between these two neural populations was achieved through mutual inhibition and firing rate adaptation ( see Appendix 1: Equations 11–13 ) . Models of this class replicate spontaneous perceptual switching between ambiguous stimuli and the association of higher switching rates with higher stimulus contrast/energy ( Laing and Chow , 2002; Shpiro et al . , 2007; Wilson , 2003 ) . Independent noise was injected into the LEFT and RIGHT neural populations to induce variability in the dominance times of the rival motion percepts . When stimulated with 11 Hz uniform flicker ( the stimulus parameters that gave the highest hallucinated contrast; Figures 1D and 3C ) , the model generated ‘cortical’ standing waves , that is , oscillations in modelled cortex , at half the flicker rate ( Figure 3D ) , in accordance with previous findings . Standing waves were observed for flicker frequencies in the range 8–18 Hz , which we interpret as hallucinatory . Further , it was also possible to induce spatially irregular patterns by adding in weak random connections between the E cells shown in Figure 3B ( data not shown ) . Beyond those frequencies the cortical model produced spatially uniform responses ( Figure 3—figure supplement 1H ) which we interpreted as non-hallucinatory . The space-time signatures of the flicker-induced standing waves evoked identical responses from both the Left and Right motion detectors ( Figure 3E ) . Consequently , those ambiguous motion signals induced spontaneous switching between the two populations ( Figure 3F ) . The dominance times of each perceptual decision ( Figure 3G ) exhibited a long-tailed distribution ( M = 5 . 1 s , SD = 1 . 6 s ) that was qualitatively similar to those observed experimentally ( Figure 2B ) . However , it is interesting to note the differences in the shape-parameter ( a ) of the gamma fits , between the behavioral and model data . This difference could be summed up by describing the model data as being closer to a normal distribution than the behavioral data . One possible reason for the difference could be our choice of Gaussian noise in the rivalry model . Future work could probe different noise distributions to fine tune a model of bistable hallucinations . When the same model is presented with a counter-phase retina-sourced physical patterned stimulus ( Figure 3H ) , with identical spatial ( fx = 0 . 11 cycles/mm ) and temporal ( ft = 5 . 5 Hz ) frequencies to the flicker-induced standing waves ( hallucinations ) , standing waves were also produced in model-cortex ( Figure 3I; see Appendix 2 ) . The net motion signals were stronger than those induced by the blank flicker stimulation ( Figure 3E and J ) , hence the spontaneous switching between LEFT and RIGHT perceptual decisions was somewhat faster ( M = 4 . 3 , SD = 1 . 5 , Figure 3K and L ) . Our data implicate retinotopically organized visual cortex as the site of bistable flicker induced hallucinations . Unlike previous work on visual hallucinations that suffered methodological limitations due to the almost infinite array of unpredictable hallucinated features ( e . g . multiple combinations of color , form and motion ) , our technique essentially limits the space to a one dimensional annulus and the one set of visual features: moving light grey blobs . The intriguing phenomenon of perceptual bistability has fascinated thinkers for centuries ( Blake and Logothetis , 2002 ) , both because of the phenomenological experience of oscillating consciousness , but also as it provides the informative dissociation between low-level stimulation and awareness . Here we show a hallucinated bistable stimulus , in which both the content ( form and motion ) and alternations in motion are endogenously generated . Counter-phase physical stimuli are known to induce motion percepts that spontaneously switch between the two candidate directions of motion . Our model suggests that uniform flicker can induce spatiotemporal responses in primary visual cortex that are very similar to those induced by counter-phase physical stimuli , given the appropriate choice of spatial and temporal frequencies . We argue that the same neural mechanism that contributes to apparent motion of a counter-phase physical stimulus also contributes to hallucinated experience . It is up to future research to divulge if sensory decisions in endogenously generated content are resolved using the same neural machinery as exogenously sourced perceptual information . A neural field model based on the idea that standing waves of neural activity form visual hallucinations , provided a quantitative mechanism for these bistable hallucinations . It is interesting that in general , theories based on so-called symmetry-breaking standing waves have been proposed to explain the complex spatio-dynamics of the human brain ( Atasoy et al . , 2016 ) , as well as in many other physical processes such as convection in fluids , animal coat markings , and cellular division ( Stewart , 1999; Turing , 1952; Kondo and Miura , 2010 ) . The exceptional circumstances in which externally sourced stimuli are overwhelmed by internally generated spontaneous patterns of neural activity and the accompanying conscious experience ( hallucinations ) , are notoriously difficult to study scientifically . Accordingly , almost no specific treatments have been developed for clinical use . Our technique for controlling and objectively measuring the range of hallucinated features ( contrast , motion , bistability ) and the corresponding neural model should prove useful in probing the mechanism ( s ) that allows such a range of non-ordinary function to produce hallucinations . Participants were students from the University of New South Wales who participated as part of a course requirement or were reimbursed for the time financially . No participants had a history of migraines , psychiatric or neurological disorders , and all had normal or corrected to normal eyesight . Informed written consent was obtained according to procedures approved by the ethics committee of the School of Psychology at the University of New South Wales . Experiments 1 , 2 and 7 were performed in a blackened room using a linearized CRT monitor at a resolution of 1600x1200 pixels and a refresh rate of 85 Hz . A chin rest was used to maintain a fixed viewing distance of 57 cm . Experiments 3 , 4 , 5 , 6 , 8 , and 9 were performed on a different CRT monitor with a resolution of 1280x1024 , a refresh rate of 85 Hz , and at a viewing distance of 57 cm . Experiment 10 was performed on a third CRT monitor with a resolution of 1280 x 960 pixels , a refresh rate of 85 Hz , and at a viewing distance of 47 cm .	Hallucinations can occur in both healthy and unwell people . Drugs , sleep deprivation , loss of vision , and migraines can all trigger visual hallucinations in people with no psychiatric illness . We have known for more than 200 years that flickering light can induce hallucinations in almost anyone . However , the unpredictability , complexity and personal nature of hallucinations make them difficult to measure scientifically , and previous studies often had to rely on drawings and verbal descriptions . Pearson et al . now show how to induce visual hallucinations in anyone , and how to measure them objectively and reliably without relying on subjective reports or drawings . The participant volunteers were university students with no history of migraines or psychiatric disorders . The students watched an image of a plain white ring flicker on and off around 10 times per second against a black background . All individuals reported seeing pale grey blobs appear in the ring and rotate around it , first in one direction and then the other . These grey hallucinations are much simpler than the complex multi-shape hallucinations people generally experience and so they are easier to study objectively . To measure the hallucinations , Pearson et al . placed a second ring marked with permanent perceptual grey blobs inside the white ring . By stating whether the hallucinated blobs were lighter or darker than the real blobs , the participants were able to communicate the strength of their hallucinations . Similarly , by indicating when the hallucinated blobs had moved past fixed lines at the top and bottom of the white ring , the subjects were able to convey the speed of the hallucinated motion . The hallucinated blobs and ‘real’ perceived blobs had many of the same properties , and seemed to arise in the same part of the brain , the visual cortex . By using the data to construct a neural computer model of visual cortex , Pearson et al . propose a mechanism that can explain both normal vision and hallucinations . The next step is to investigate whether the experimental methods can also model the hallucinations produced by psychiatric disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Sensory dynamics of visual hallucinations in the normal population
Inserted ( I ) domains function as ligand-binding domains in adhesins that support cell adhesion and migration in many eukaryotic phyla . These adhesins include integrin αβ heterodimers in metazoans and single subunit transmembrane proteins in apicomplexans such as TRAP in Plasmodium and MIC2 in Toxoplasma . Here we show that the I domain of TRAP is essential for sporozoite gliding motility , mosquito salivary gland invasion and mouse infection . Its replacement with the I domain from Toxoplasma MIC2 fully restores tissue invasion and parasite transmission , while replacement with the aX I domain from human integrins still partially restores liver infection . Mutations around the ligand binding site allowed salivary gland invasion but led to inefficient transmission to the rodent host . These results suggest that apicomplexan parasites appropriated polyspecific I domains in part for their ability to engage with multiple ligands and to provide traction for emigration into diverse organs in distant phyla . Domains with similar overall structures , initially described in von Willebrand factor A ( VWA domains ) , are found in cell-surface proteins including integrins , extracellular matrix , and complement components , and mediate a diversity of functions including cell adhesion , migration , and signaling ( Whittaker and Hynes , 2002 ) . Here , we study a subset of VWA domains termed I domains because they are inserted in other domains in integrins . I domains differ from VWA domains in the position of their ligand binding sites and in the presence of a metal ion-dependent adhesion site ( MIDAS ) at the center of their ligand binding site ( Liddington , 2014 ) . Within integrins , I domains switch between closed and open states coordinately with conformational change in neighboring domains . This switch from closed to open conformation in the I domain alters the ligand-binding site around the MIDAS and increases affinity for ligand by ~1 , 000 fold ( Schürpf and Springer , 2011 ) . I domains are key modules in adhesins employed by apicomplexan pathogens . I domain-containing , membrane-spanning surface glycoproteins have been shown to be essential for tissue traversal and cell invasion by Toxoplasma gondii and Plasmodium spp . and are present in all known apicomplexans ( Sultan et al . , 1997; Morahan et al . , 2009 ) . In Plasmodium ( P . ) , a protein named CTRP containing six I domains is required for invasion of the mosquito midgut by the ookinete ( Ramakrishnan et al . , 2011; Dessens et al . , 1999 ) . Once the ookinete crosses the midgut epithelium it forms an oocyst wherein it differentiates into hundreds of sporozoites ( Frischknecht and Matuschewski , 2017 ) . Sporozoites use proteases ( Aly and Matuschewski , 2005 ) and active motility ( Klug and Frischknecht , 2017 ) to egress from the oocyst into the hemolymph and subsequently enter the salivary glands from where they can be transmitted back to a vertebrate host . Once deposited in the skin during a blood meal by an infected mosquito , sporozoites migrate rapidly to find and enter blood vessels ( Amino et al . , 2006 ) . Within the blood stream parasites are passively transported to the liver , where they infect hepatocytes and develop into liver stages ( Douglas et al . , 2015 ) . The subsequent blood stages cause the typical symptoms of malaria by triggering a massive immune response , clogging capillaries and lysing red blood cells ( Cowman et al . , 2016 ) . Sporozoites express two adhesins with I domains , TRAP ( thrombospondin related anonymous protein ) and TLP ( TRAP-like protein ) . These proteins are stored in secretory vesicles called micronemes , at the apical end of the highly polarized sporozoite ( Tomley and Soldati , 2001 ) . After fusion of micronemes with the plasma membrane , TRAP and TLP decorate the surface of the sporozoite and form a bridge between extracellular ligands and the membrane-subtending actin-myosin motor that drives gliding motility and invasion ( Heintzelman , 2015; Frischknecht and Matuschewski , 2017 ) . Deletion of tlp causes only a mild phenotype in tissue traversal while deletion of trap yields sporozoites that cannot move productively in vitro , fail to enter salivary glands , and are unable to infect mice if isolated from mosquitoes and injected intravenously ( Sultan et al . , 1997; Moreira et al . , 2008; Hellmann et al . , 2013; Quadt et al . , 2016 ) . Mutations of amino acids within the MIDAS motif of the single I domain in TRAP decreased the capacity of sporozoites to enter salivary glands and liver cells as well as to infect mice ( Wengelnik et al . , 1999; Matuschewski et al . , 2002 ) . However , these mutant sporozoites were still able to migrate in vitro . This suggests that the MIDAS is important for ligand binding but not for productive motility . Crystal structures of the N-terminal portion of TRAP in Plasmodium spp . and a TRAP orthologue in Toxoplasma gondii , the micronemal protein 2 ( MIC2 ) , revealed the I domain in both open and closed conformations in association with a thrombospondin type-I repeat domain ( Song et al . , 2012; Song and Springer , 2014; Figure 1 ) . The apicomplexan I domains structurally resemble I domains found in integrin α-subunits ( αI domains ) much more than I domains in integrin β-subunits ( βI domains ) . Between the closed and open states of both apicomplexan I domains and integrin αI domains , the Mg2+ ion at the MIDAS similarly moves ~2 Å closer to one coordinating sidechain and away from another . This movement is linked to essentially identical pistoning of the C-terminal , α7-helix toward the ‘bottom’ of the domain ( Figure 1A compared to 1B and Figure 1D compared to 1E ) . The distance pistoned is equivalent to two turns of an α-helix . Uniquely in the apicomplexan I domains , a segment N-terminal to the I domain is disulfide-linked to the last helical turn of the α7-helix in its closed conformation ( Figure 1A–C ) . As this segment pistons out of contact with the remainder of the I domain in the open conformation , the last two turns of the α7-helix with its cysteine unwind , the N-terminal segment with its cysteine moves in a similar direction , and these segments reshape to form a β-ribbon ( Figure 1A ) . Because of close structural homology between human integrin αI- and apicomplexan adhesin I domains in the regions shown in green in Figure 1A–E , we were able to engineer exchanges between them in this study ( Figure 1F ) . P . berghei parasites expressing TRAP without an I domain phenocopy trap ( - ) parasites and show that the I domain is essential for motility and invasion . Parasites expressing TRAP with the I domain of MIC2 from the related apicomplexan Toxoplasma gondii , show rescued motility and invasion . By reversing the charges of amino acid residues around the MIDAS motif we could partially uncouple the function of TRAP during salivary gland invasion and rodent infection . Our results show that I domains have the capacity to be poly-specific and permit TRAP to function as an adhesin in both vertebrate and arthropod hosts . The importance of the I domain for TRAP function was tested using the trapΔI parasite line ( Figure 2A , Figure 2—figure supplement 1 ) . In several independent experiments in which mosquitoes fed on infected mice , only few trapΔI or trap ( - ) sporozoites could be observed within the mosquito salivary glands , whereas for trap wild type ~10 , 000 sporozoites were observed per mosquito ( Figure 2B , Table 1 ) . Thus , trapΔI sporozoites are similarly impaired in salivary gland invasion as trap ( - ) sporozoites . To determine whether mutant parasites retained the ability to migrate steadily on microscope slides , that is to glide in circles ( Vanderberg , 1974 ) , sporozoites were isolated from hemolymph and activated by addition of 3% bovine serum albumin ( BSA ) . Hemolymph sporozoites , like midgut sporozoites , were present in all mutants studied here ( Table 1 ) . Sporozoites were defined as gliding and productively motile if they were able to complete at least one circle within 3 or 5 min , depending on the experiment . Sporozoites exhibiting other types of motion were classified as unproductively motile , while sporozoites that were attached but were not moving or were not attached were classified as non-motile ( Figure 2—figure supplement 2; Münter et al . , 2009 ) . Hemolymph sporozoites were ~19% productively motile in wild type while none were motile in trapΔI ( Figure 2C ) and trap ( - ) mutants ( Münter et al . , 2009; Hegge et al . , 2010; Figure 2C ) . These results show that the I domain is required for productive motility . Infectivity of mutant sporozoites was tested by exposing naive mice to infected mosquitoes or by intravenously injecting sporozoites obtained from the midgut , hemolymph or even salivary glands ( Table 1 ) . Upon infection with wild type sporozoites , the first blood stage parasites were visible as expected after 3 to 8 days; in contrast , no infections could be observed for trapΔI and trap ( - ) parasites ( Figure 2—figure supplement 3 , Table 2 , Table 3 ) . Immunofluorescence assays on midgut sporozoites using an antibody recognizing the repeat region of TRAP showed specific fluorescence in most sporozoites at one end with no recognizable difference between trapΔI and wild type sporozoites . This suggests that the mutated TRAP is also localized in the secretory micronemes . In contrast , TRAP-specific fluorescence was absent in trap ( - ) sporozoites . TRAP could also be observed on the surface of unpermeabilized trapΔI sporozoites indicating that micronemal secretion is not abolished in these parasites ( Figure 2D ) . We next tested whether the lack of productive motility and infectivity as well as the severely impaired salivary gland invasion rate in trapΔI parasites could be rescued by replacing the deleted TRAP I domain with an I domain from a foreign species . We selected structurally characterized I domains of MIC2 from Toxoplasma gondii and the I domains of the human integrin α-subunits αX ( CD11c ) and αL ( CD11a ) ( Figure 1C–E ) . Sequence identity among I domains is 36% between αL and αX , 18% between these integrins and P . berghei TRAP , and 28% between TRAP and MIC2 ( Supplementary file 2 ) . The I domains from TRAP and αX are basic , with pI values of 9 . 7 and 8 . 9 , respectively , while those of MIC2 and αL are acidic , with pI values of 6 . 1 and 5 . 8 , respectively ( Song et al . , 2012; Song and Springer , 2014 ) . Furthermore , the αX I domain is poly-specific as shown by binding to multiple glycoproteins and proteolytic fragments as well as heparin ( Vorup-Jensen et al . , 2005; Vorup-Jensen et al . , 2007 ) , while the αL I domain is highly specific for the ligand intercellular adhesion molecule ( ICAM-1 ) and its homologues ICAM-2 , ICAM-3 , and ICAM-5 ( Grakoui et al . , 1999 ) . The fluo line , with eGFP constitutively expressed in all parasite stages and mCherry specifically expressed in sporozoites , was used to generate some of the respective transgenic parasite lines to simplify analysis of TRAP I domain replacements throughout the parasite life cycle ( Bane et al . , 2016 ) . Parasite lines expressing either codon modified wild type TRAP ( TRAP-I ) or codon modified TRAP with foreign I domains ( MIC2-I , αX-I or αL-I ) replacing the P . berghei TRAP I domain ( Supplementary file 2 ) were generated with both fluo and non-fluo trap ( - ) parasite lines by homologous recombination ( Figure 3A and Figure 3—figure supplement 1 ) . TRAP shows the same localization in MIC2-I , αX-I and αL-I midgut sporozoites as in TRAP-I midgut sporozoites ( Figure 3B ) . Western blots showed similar expression levels of TRAP in hemolymph sporozoites of TRAP-I , trapΔI , αX-I and αL-I while no TRAP could be detected for trap ( - ) ( Figure 3C ) . Similar localization and expression of TRAP was also observed for MIC2-I and TRAP-I salivary gland sporozoites ( Figure 3D , E ) . MIC2-I fluo and non-fluo lines entered salivary glands at similar rates as wild type sporozoites ( Figure 3F , G , H , Table 1 ) . αX-I and αL-I sporozoites were capable of invading the salivary glands , albeit at very low rates ( Table 1 and Figure 3F and G , note the log scale of the y axis ) . The numbers for αX-I ranged from ~100 ( fluo ) to ~200 ( non-fluo ) and for αL-I from ~30 ( non-fluo ) to ~90 ( fluo ) sporozoites per gland ( Figure 3F , G and Table 1 ) compared to 0–200 for trap ( - ) sporozoites and 0–800 for trapΔI . These small numbers might also be due to contaminants from the hemolymph and hence need to be treated with caution . Yet in all ( 8/8 ) αX-I mosquito infections sporozoites were detected in salivary glands , whereas this was only the case in 50% ( 4/8 ) of αL-I infections . In contrast for mosquitoes infected with trap ( - ) or trapΔI sporozoites were only detected in 13% ( 1/8 ) and 21% ( 3/14 ) experiments , respectively . ( Table 1 ) . In line with this observation , αX-I but not αL-I sporozoites could be detected in isolated salivary glands by live fluorescence microscopy ( Figure 3H and Video 1 ) that indeed sporozoites expressing αX-I can enter this organ more efficiently than those expressing αL-I . We next analyzed motility of the parasite lines in vitro using the classification scheme shown in Figure 2—figure supplement 2 . Gliding assays of hemolymph sporozoites revealed that ~24% of TRAP-I but only ~4% of MIC2-I sporozoites were productively motile ( Figure 4A ) . As expected , a higher proportion of salivary gland sporozoites were productively motile;~53% for TRAP-I and ~15% for MIC2-I sporozoites ( Figure 4B ) . Among hemolymph sporozoites , ~1% of αX-I ( Video 2 ) but none of the >3000 observed αL-I sporozoites showed productive movement ( Figure 4A ) . Motile TRAP-I and MIC2-I salivary gland sporozoites moved with a similar speed of ~1 . 5 µm/s ( Figure 4C ) , showed similar trajectories ( Figure 4D ) , and showed similarly persistent gliding ( Figure 4E ) . Owing to the low numbers of αX-I and αL-I sporozoites in the glands ( Figure 3 ) , similar quantitation of sporozoite motility was not possible . To probe the infectivity of mutant sporozoites , mice were exposed to infected mosquitoes . This revealed infection rates of 100% with a similar prepatent period ( time until an infection could be detected in the blood ) for TRAP-I and MIC2-I sporozoites ( Table 2 , Table 3 ) . In contrast , no transmission could be observed for αX-I and αL-I sporozoites ( Figure 3—figure supplement 2 , Table 2 ) . Intravenous injection of 10 , 000 TRAP-I or MIC2-I salivary gland sporozoites also infected all mice with a prepatency of three days . Unfortunately , the numbers of αX-I and αL-I salivary gland sporozoites were too low for comparative tests ( Table 1 ) . We therefore injected 10 , 000 hemolymph sporozoites , which again resulted in similar infection rates as seen for TRAP-I and MIC2-I sporozoites with a slightly longer prepatency when compared to salivary gland sporozoites ( Figure 3—figure supplement 3 , Table 2 ) . Interestingly , 50% ( 4/8 ) of mice injected with 10 , 000 αX-I hemolymph sporozoites became blood stage patent after a delayed prepatency of >5 days , while no infection was observed for the same number of αL-I hemolymph sporozoites ( 0/8 ) ( Figure 3—figure supplement 3 , Table 2 ) . When injecting 25 , 000 hemolymph sporozoites , 5 of 8 mice injected with αX-I sporozoites became blood stage patent and additionally , 1 of 6 mice injected with αL-I sporozoites became infected ( Figure 3—figure supplement 3 , Table 2 , Table 3 ) . Finally , we injected mice with 500 , 000 midgut sporozoites to compare αX-I and αL-I side by side with trap ( - ) and trapΔI . Injections of αX-I midgut sporozoites infected 2 out of 4 mice with a prepatent period of 8 days . In contrast no infections in mice could be observed when αL-I ( 0/2 ) , trap ( - ) ( 0/4 ) or trapΔI ( 0/6 ) midgut sporozoites were injected ( Table 2 ) . Mice infected with the same number of wild type or wild type-like midgut sporozoites become blood stage patent after 6–8 days ( Table 2; Klug and Frischknecht , 2017 ) . To exclude spurious results from contamination with other parasite lines , some of the transmitted parasites were propagated in mice and analyzed via PCR for the correct genotype ( Figure 3—figure supplement 4 ) . Additionally , the TRAP locus of these parasites was sequenced to ensure that the correct I domain was present . In all tested cases , the expected αX-I and αL-I genotype was confirmed . Thus , TRAP-I and MIC2-I sporozoites are comparably infectious to mice , αX-I sporozoites are moderately infectious and αL-I , while nearly completely deficient in infectivity , could in one case still cause an infection . While Plasmodium sporozoites can infect different types of cells they have a strong tropism for the liver . In contrast , Toxoplasma gondii can infect any nucleated cell from a warm-blooded animal ( Boothroyd , 2009 ) . Hence , we tested whether an exchange of the I domain affected host cell invasion or tissue tropism of sporozoites . First , we tested the capacity of TRAP-I and MIC2-I salivary gland sporozoites to invade HepG2 cells . After exposure for 1 . 5 hr , 2-fold fewer MIC2-I sporozoites than TRAP-I sporozoites were intracellular ( Figure 5A ) . At 48 hr after infection , 5-fold fewer MIC2-I sporozoites had developed into liver stage parasites when compared to TRAP-I sporozoites ( Figure 5B ) . However , the size of liver stage parasites after 48 hr was comparable implying that an exchange of the I domain affects parasite invasion but not intracellular development ( Figure 5C , D ) . To test in vivo tropism of sporozoites , mice were infected by intravenous injection of 20 , 000 TRAP-I and MIC2-I salivary gland sporozoites and after 42 hr liver , spleen , lung and a part of the small intestine were harvested . Parasite load was measured by quantitative RT-PCR . Liver tropism of both TRAP-I and MIC2-I salivary gland sporozoites was pronounced , with ~16 fold more parasites localizing to the liver than to any other organ ( Figure 5E , Figure 5—figure supplement 1 ) . However , the liver burden of MIC2-I parasites was reduced by ~40% relative to TRAP-I parasites , reflecting the similar decrease observed during in vitro infection experiments . Based on the results with different I domains , we sought a common pattern explaining the observed phenotypes . One parameter important for protein-protein interactions is surface charge . Interestingly , the surface of the TRAP I domain is very basic , with a pI of 9 . 7 ( Figure 6A ) , while the surface of the second best functioning I domain , MIC2 , is acidic ( pI 6 . 1 ) ( Figure 6B ) . Of the human integrin I domains , the one from αX has a basic charge ( pI 8 . 9 ) , similar to the TRAP I domain ( Figure 6C ) , while the one of αL is even slightly more negatively charged ( pI 5 . 8 ) ( Figure 6D ) than the I domain of MIC2 . To test if surface charge could be important for P . berghei TRAP I-domain function , we rendered it anionic ( pI of 6 . 8 ) with seven charge reversal mutations ( H56E , H62E , H123E , K164Q , K165D , R195E and K202E ) around the perimeter of the putative ligand binding site , distal from the MIDAS ( Figure 6E , Figure 6—figure supplement 1 ) . Secretion of TRAP to the surface was not altered in these RevCharge sporozoites and no difference in protein sub-cellular distribution or expression was observed ( Figure 6F , G ) . Strikingly , salivary gland invasion was also not affected ( Figure 6H , Table 1 ) . However , only ~1% of RevCharge salivary gland derived sporozoites were productively motile in gliding assays compared to ~53% of TRAP-I parasites ( Figure 6I ) . Infection by mosquito bite revealed that only 2 out of 8 mice in two independent experiments became blood stage positive with a delay in prepatency of >2 days ( Figure 6J , Table 2 , Table 3 ) indicating a decreased infectivity of salivary gland sporozoites by over 99% . In contrast , injection of 10 , 000 salivary gland sporozoites in three independent experiments ( nine mice in total ) led to an infection of all mice with the prepatency period of the RevCharge mutant being delayed by 0 . 5–1 day compared to controls ( wild type/TRAP-I ) ( Figure 6K , Table 2 ) . This corresponds to a decreased infectivity of 50–90% . In vitro invasion of liver cells also showed a severe impairment of the RevCharge mutant compared to TRAP-I . While after 2 hr 54% of TRAP-I sporozoites showed an intracellular localization only 3% of RevCharge sporozoites were observed inside cells ( Figure 6—figure supplement 2A ) . In line with this result very few growing liver stages of the RevCharge mutant were observed while on average >250 liver stages per well were counted for TRAP-I ( Figure 6—figure supplement 2B ) . However , liver stages of the RevCharge mutant developed normally ( Figure 6—figure supplement 2C ) . These results suggest that the basic charge around the MIDAS of the TRAP I domain is a key determinant of sporozoite motility and infectivity during natural transmission from mosquito to mammal . Most intriguingly , the introduced mutations appear to uncouple the capacity of the P . berghei sporozoite to infect insect salivary glands from efficiently infecting the rodent host . Previous studies on the function of the TRAP I domain focused on invariant residues that coordinate the MIDAS metal ion in the I domain ( Wengelnik et al . , 1999; Matuschewski et al . , 2002 ) . The sidechains of these five invariant I domain residues coordinate the MIDAS Mg2+ ion either directly or indirectly through water molecules . In contrast to the closed conformation , in the open conformation of TRAP and integrin αI domains , neither of the two Asp residues directly coordinate the MIDAS metal ion . The metal ion is therefore thought to have high propensity in the open conformation to bind an acidic residue in the ligand . The MIDAS residues and their bound water molecules occupy five of the six coordination positions around the MIDAS Mg2+ ion . The remaining , sixth coordination position is occupied when a critical Asp or Glu sidechain in the ligand binds through a carboxyl oxygen to the MIDAS Mg2+ ion ( Liddington , 2014 ) . Mutation of single MIDAS residues or removal of the Mg2+ ion by chelation abolishes ligand binding by integrins ( Michishita et al . , 1993; Kern et al . , 1994; Kamata et al . , 1995 ) . Similarly , mutation of the MIDAS motif of TRAP severely impairs salivary gland invasion and infectivity of the vertebrate host while gliding motility is decreased in salivary gland but not in hemolymph sporozoites ( Matuschewski et al . , 2002 ) . In contrast , deletion of TRAP completely abrogates salivary gland invasion , infectivity and productive motility ( Sultan et al . , 1997; Münter et al . , 2009 ) and severely affects substrate adhesion ( Münter et al . , 2009; Hegge et al . , 2010 ) . To elucidate the role of the I domain in TRAP function , we generated the parasite line trapΔI expressing TRAP without its I domain . TRAPΔI was expressed and correctly localized in sporozoites as shown by western blotting and immunofluorescence . Interestingly , this line was severely deficient in salivary gland invasion and could not infect mice . Furthermore , we observed no productive movement and only some back-and-forth motility in trapΔI hemolymph sporozoites , similar to trap ( - ) hemolymph sporozoites . These results suggest that the biological functions of TRAP are dependent on its I domain . It was therefore remarkable that TRAP function was largely restored by replacement of its I domain by its homologue from Toxoplasma gondi with only 28% amino acid sequence identity . Compared to cell surface receptors that engage in typical protein-protein interactions , that is those that are not dependent on Mg2+ ions , these sequence identities are substantially below the level of 40% to 50% identity generally required for members of the same receptor family to recognize the same ligand . The strength of the Mg2+ ion bond to the oxygen in the ligand , which has a bond distance of only 2 Å and is partially covalent , may overcome the lack of complementarity in other regions of the ligand binding site , and explain the ability of I domains of such remarkably low sequence identity to function in TRAP . In addition to the low sequence identity , there are six to seven sequence positions where residues are inserted or deleted in Toxoplasma MIC2 or human integrin αI domains compared to TRAP ( Figure 1F ) . Nonetheless , all regions involved in conformational change around the MIDAS and in the β6-α7 loop are highly conserved in conformation in both the open and closed conformations of the apicomplexan and human I domains ( Song et al . , 2012; Song and Springer , 2014 ) . Western blotting showed that the αL-I , αX-I , and MIC2-I TRAP fusions were as well expressed as TRAP-I; furthermore , immunofluorescence showed normal localization in sporozoites , suggesting that trafficking was not impaired . Although deletion of the TRAP I domain completely abolished mosquito salivary gland invasion and mouse infection by midgut and hemolymph sporozoites , the MIC2 I domain completely restored these functions . Moreover , no differences in mouse infection were observed between TRAP and MIC2 I domain replacements in infection by mosquito bite or intravenous injection with salivary gland sporozoites , including the length of the prepatent period . Fewer MIC2-I than TRAP-I sporozoites were able to glide productively in vitro; however , the gliding speed of motile sporozoites and persistence of gliding was similar . Infectivity of HepG2 liver cells was significantly decreased for MIC2-I compared to TRAP-I sporozoites; yet , quantitative RT-PCR showed that there was little difference in ability to infect liver cells in vivo between intravenously injected MIC2-I and TRAP-I sporozoites . Moreover , there was no change in in vivo tropism . We also examined replacement with two mammalian integrin I domains , from the promiscuous αXβ2 integrin and the much more selective αLβ2 integrin . Salivary gland invasion was decreased 100 to 400 and 500 to 1 , 000-fold in αX-I and αL-I compared to TRAP-I sporozoites , respectively . Infectivity of hemolymph αX-I sporozoites in mice was 100-fold deacreased compared to controls while αL-I hemolymph sporozoites showed nearly no infectivity as trap ( - ) sporozoites . The greater efficacy of αX-I than αL-I sporozoites in invasion of mosquito salivary glands , infection of mice , and productive motility in vivo confirmed our hypothesis that the polyreactive αX I domain would better support TRAP function than the highly specific αL I domain , albeit at low levels . Multiple factors may account for the substantially lesser efficacy of mammalian integrin I domains than the MIC2 I domain . MIC2 has a similar function to TRAP in Toxoplasma and may be its orthologue . Each has an I domain that is inserted in an extensible β-ribbon in tandem with a TSR domain and a long-range disulfide bond that moves in allostery ( Figure 1 ) . Integrins , TRAP , and MIC2 connect to the actin cytoskeleton , which applies force to their cytoplasmic domains that is resisted by ligands and provides traction for motility and cell invasion . The force ( F ) transmitted through the I domain , times the difference in extension of the I domain in the closed and open states ( Δx ) , gives an energy ( F•Δx = E ) that tilts the energy landscape toward the open , high affinity state ( Li and Springer , 2017 ) . Because of their extensible β-ribbons , Δx is substantially greater for TRAP and MIC2 than for integrin I domains , and cytoskeletal forces may also differ considerably . To compensate for these differences , the energy differences between the closed and open states , as well as the kinetics for crossing between them , may be tuned differently for TRAP and MIC2 than for integrin I domains . Their degree of polyspecificity is also likely to vary . Integrins αXβ2 and αMβ2 have each been reported to bind >30 ligands , including glycoproteins and heparan ( Yakubenko et al . , 2002 ) . However , these integrin I domains can also bind specifically , as shown by binding to distinct high-affinity sites within their common ligand , iC3b ( Xu et al . , 2017 ) . The most polyreactive integrin I domains , αX and αM are basic with higher pI ( 8 . 9 and 9 . 5 , respectively ) than the highly specific αL ( pI 5 . 6 ) , αE ( pI 4 . 6 ) , and collagen-binding α1 , α2 , α10 , and α11 I domains ( pI 5 . 1–5 . 6 ) . As the surfaces of all cells are negatively charged and almost all proteins are acidic , we tested whether the functions of the basic TRAP I domain ( pI = 9 . 7 ) would be altered with seven substitutions that made it neutral ( pI 6 . 8 ) . This RevCharge mutant was well expressed and localized . RevCharge sporozoites accumulated well in mosquito salivary glands and were infectious when injected intravenously . However , gliding in vitro and infectivity by mosquito bite were greatly decreased . Although the pI of RevCharge of 6 . 8 was close to that of MIC2 , its mutations were clustered around the periphery of the MIDAS site and the electrostatic surface of its ligand-binding face more closely resembled that of αL than MIC2 , TRAP , or αX ( Figure 6 ) . These results suggest that the distribution of electrostatic charge on the TRAP I domain is an important variable . The difference between infection by mosquito bite and intravenous injection further suggests that electrostatics may have an important role in successful sporozoite exit from the dermis into the circulation . The findings that the I domain from Toxoplasma MIC2 that encounters different ligands as TRAP can replace its function in mediating salivary gland invasion and infection of mice strongly suggest that the I domain functions to provide traction for motility and cell invasion rather than tissue tropism . Binding of the TRAP I domain to multiple distinctive ligands is required by its function in mosquitos in migration into salivary glands and in vertebrate hosts in emigration from the dermis into the circulation and from the circulation into the liver , followed by productive invasion of liver parenchymal cells . The ligands in the mosquito must differ from those in the vertebrate; those encountered in the vertebrate in the dermis , endothelium and liver may also differ . Furthermore , Toxoplasma lives in epithelia in the gastrointestinal system of cats and other mammals , and lacks an arthropod host , yet its MIC2 I domain functions robustly in mosquito salivary gland invasion . Our results are consistent with TRAP binding to the many ligands that have been identified for it , including liver-specific fetuin-A ( Jethwaney et al . , 2005 ) , proteoglycans ( Robson et al . , 1995; Pradel et al . , 2002 ) , glycosaminoglycans ( Matuschewski et al . , 2002 ) , and the integrin αV subunit ( Dundas et al . , 2018 ) in vertebrates and saglin in mosquitos ( Ghosh et al . , 2009 ) . Abolition of TRAP function by deletion of its I domain as shown here , together with the negligible effects of mutation or exchange of other domains ( Matuschewski et al . , 2002; Ejigiri et al . , 2012 ) strongly suggests that the I domain is the sole ligand binding domain . A model in which TRAP does not provide liver tropism is also consistent with the expression of TRAP in sporozoites of Plasmodium species that infect birds and initially infect skin rather than liver cells ( Böhme et al . , 2018 ) . Our results support a model in which TRAP does not determine tropism for salivary glands in mosquitos or liver in mammals and instead acts as a poly-specific receptor that provides traction for sporozoite emigration into these tissues and infection of cells that is triggered by other receptors . Similarly , integrins , from which the TRAP I domain may have been borrowed , are activated by cytoskeletal activity , which is generally signaled by G protein-coupled receptors or receptors that couple to tyrosine kinases . This model has the advantage that the signaling receptors can be extremely sensitive and ATP-dependent cytoskeleton polymerization or actomyosin contraction through activation of adhesins can greatly amplify those signals and provide ultrasensitive regulation of adhesiveness ( Li and Springer , 2017 ) . This might be essential during the stick-and-slip gliding of sporozoites ( Münter et al . , 2009 ) . Chemoattractants have the advantage that they can drive directional migration through multiple layers of distinctive cell types as in the liver . In vertebrates , chemoattractants and their receptors cooperate with integrins and their ligands to enable highly specific leukocyte emigration from the vasculature into organs or sites of inflammation ( Springer , 1994 ) . Chemoattractant receptors evolved distinctly in prokaryotes and eukaryotes and perhaps take yet a different form in apicomplexans . The ability of chemoattractants to activate motility , termed chemokinesis , may be related to the use here of BSA , a known carrier of fatty acids and other hydrophobes , to activate sporozoite gliding . Our finding that the I domain of TRAP is required for gliding motility and invasion by sporozoites of mosquito salivary glands and mammalian liver , without a strong requirement for evolved specificity , suggests that new paradigms may be needed to understand sporozoite activation and homing of sporozoites to specific organs in their hosts . Plasmodium sequences were retrieved from PlasmoDB ( http://plasmodb . org/plasmo/ , version 26–34 ) ( Aurrecoechea et al . , 2009 ) and multiple sequence alignments were performed with Clustal Omega ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) ( Sievers et al . , 2011 ) . To change the codon usage of open reading frames ( ORFs ) we applied the tool OPTIMIZER ( http://genomes . urv . es/OPTIMIZER/ ) ( Puigbò et al . , 2007 ) . TRAP knockout ( trap ( - ) ) parasites were generated with the PlasmoGem ( Schwach et al . , 2015 ) vector ( PbGEM-107890 ) using standard protocols ( Janse et al . , 2006 ) . Isogenic trap ( - ) parasites were subsequently negatively selected with 5-fluorocytosine ( 1 . 0 mg/mL in the drinking water ) to give rise to selection marker free trap ( - ) parasites ( Braks et al . , 2006 ) . For the generation of trapΔI parasites we made use of the Pb238 vector ( Deligianni et al . , 2011; Klug and Frischknecht , 2017 ) . In a first step the trap 3’UTR ( 970 bp ) was amplified with the primers P165/P166 and cloned ( BamHI and EcoRV ) downstream of the resistance cassette in the Pb238 vector . In a next step the coding sequence of the trap gene including the 5’ and 3’ UTR was amplified with the primers P508/P509 and cloned in the pGEM-T-Easy vector giving rise to the plasmid pGEM-TRAPfull . Subsequently the pGEM-TRAPfull plasmid was mutated with the primers P535/P536 and P537/P538 to introduce a restriction site for NdeI directly in front of the start codon ATG and a restriction site for PacI directly after the stop codon TAA . The mutated sequence was cloned ( SacII and EcoRV ) in the Pb238 intermediate vector that contained already the trap 3’UTR downstream of the selection marker and the resulting plasmid was named Pb238-TRAP-NdeI/PacI . The designed DNA sequence lacking the coding region of the I domain was codon modified for E . coli K12 and synthesized at GeneArt ( Invitrogen ) . Subsequently the designed sequence was cloned ( NdeI and PacI ) in the Pb238-TRAP-NdeI/PacI by replacing the endogenous trap gene . Final DNA sequences were digested for linearization ( NotI , PbGEM-107890; SacII and KpnI , Pb238-TRAP∆I ) , purified and transfected into wild type parasites ( wt ) using standard protocols ( Janse et al . , 2006; Figure 2—figure supplement 1 ) . To generate parasite lines expressing TRAP with different I domains , bp 115 to bp 696 ( 581 bp; I42 to V228; 194 aa in total ) of the wild type trap gene ( Plasmodium berghei ANKA strain ) were exchanged with sequences from the micronemal protein 2 ( MIC2 ) of Toxoplasma gondii ( 567 bp , L75 to V263 , 189 aa ) , the integrin CD11c/αX ( 552 bp , Q150 to I333 , 184 aa ) and the integrin CD11a/αL ( 531 bp , V155 to I331 , 177 aa ) of Homo sapiens . Chimeric sequences as well as the coding sequence of the wild type trap gene that served as a control , were codon modified for E . coli K12 to prevent incorrect homologous recombination with the trap coding sequence downstream of the I domain coding region and to avoid changes of the codon usage within the open reading frame caused by the exchanged I domain coding sequence . This enabled also simple differentiation between wild type and transgenic parasites by PCR . For the generation of RevCharge parasites seven mutations of non-conserved amino acids ( H56E , H62E , H123E , K164Q , K165D , R195E and K202E; P . berghei ANKA strain ) were introduced into the codon modified wild type trap gene . These mutations were expected to shift the surface charge at the apical side of the I domain from a pI of 9 . 7 to 6 . 8 while leaving the MIDAS intact and the structural integrity of the domain unaffected . All designed sequences were synthesized at GeneArt ( Invitrogen ) and cloned in the Pb238-TRAP-NdeI/PacI vector ( NdeI/PacI ) that was already used to generate the trapΔI line . Constructs were digested for linearization ( ScaI-HF ) and transfected using standard protocols ( Janse et al . , 2006 ) . Transfections were performed in the negatively selected TRAP knockout line trap ( - ) as well as in the fluorescent background line fluo to independently generate fluorescent ( fluo ) and non-fluorescent ( non-fluo ) sets of mutants ( Figure 3—figure supplement 1 ) . Isogenic parasite lines were generated by serial dilution of parental populations obtained from transfections . Per transfection one mouse was infected by intraperitoneal injection of ~200 µL frozen parasites of the parental population . To increase the number of transfected parasites within infected mice pyrimethamine ( 0 . 07 mg/mL ) was given within the drinking water 24 hr post injection ( hpi ) . Donor mice were bled two to three days post injection once parasitemia reached 0 . 5–1% . Parasites were diluted in phosphate buffered saline ( PBS ) to 0 . 7–0 . 8 parasites per 100 µL and the same volume was subsequently injected into 6–10 naive mice . Parasites were allowed to grow for 8–10 days until parasitemia reached 1 . 5–2% . Blood of infected mice was taken by cardiac puncture ( usually 600–800 µL ) and used to make parasite stocks ( ~200 µL infected blood ) and to isolate genomic DNA with the Blood and Tissue Kit ( Qiagen ) . Naive mice were infected with 100–200 µL frozen parasite stocks and parasites allowed to grow for four to five days . Infected mice were either directly fed to mosquitoes or used for a fresh blood transfer of 2 × 107 parasites by intraperitoneal injection into two to three naive mice . Parasites within recipient mice were allowed to grow for further three to four days , depending on the number of exflagellation events observed . To determine the number of exflagellation events , and subsequently the number of male gametocytes , a drop of tail blood was placed on a microscope slide , covered with a coverslip and incubated for 10–12 min at room temperature ( 20–22°C ) . The number of exflagellation events was counted with a light microscope ( Zeiss ) and a counting grid by using 40-fold magnification with phase contrast . If at least one exflagellation event per field could be observed mice were fed to mosquitoes . Mosquitoes were starved overnight prior to the feeding to increase the number of biting mosquitoes . Per mosquito cage ( approximately 200–300 female mosquitoes ) two to three mice were used for feeding . To estimate the number of sporozoites , infected Anopheles stephensi mosquitoes were dissected on day 13 , 14 , 18 , and 22 post infection . For the collection of hemolymph sporozoites , infected A . stephensi mosquitoes were cut with a needle to remove the last segment of the abdomen . Subsequently the thorax was pierced with a finely drawn Pasteur pipette filled with RPMI/PS solution . While gently pressing the pipette the haemocoel cavity was flushed with solution which dripped off the abdomen . The sporozoite solution was collected on a piece of foil and transferred into a plastic reaction tube . To determine the number of midgut sporozoites , the abdomen of infected mosquitoes was dissected with two needles and the midgut was extracted . Isolated midguts were transferred into a plastic reaction tube containing 50 µL RPMI/PS solution . For the isolation of salivary gland sporozoites , the head of infected mosquitoes was gently pulled away with a needle while fixing the mosquito in place with a second needle . Ideally the salivary glands stayed attached to the head and could be easily isolated . Salivary glands were transferred into a plastic reaction tube with 50 µL RPMI/PS . To release sporozoites , pooled midguts and salivary glands were homogenized using a plastic pestle for 2 min . To count the parasites in each sample , 5–10 µL of the sporozoite solution ( 1:10 dilution ) was applied on a hemocytometer . Sporozoites were counted using a light microscope ( Zeiss ) with 40-fold magnification and phase contrast . For each counting experiment at least 10 mosquitoes were dissected . However , the number was adapted depending on the infection rate of the mosquitoes and the experiment that was performed . To determine the prepatency of parasite lines during sporozoite transmission two different routes of infection were tested . Female C57BL/6 mice were either exposed to infected mosquitoes or infected by intravenous injection of midgut , hemolymph or salivary gland sporozoites . To infect mice by mosquito bites , mosquitoes infected with fluorescent parasite lines were pre-selected for fluorescence of the abdomen by using a stereomicroscope ( SMZ1000 , Nikon ) with an attached fluorescence unit . Subsequently parasite positive mosquitoes were separated in cups to 10 each and allowed to recover overnight . Approximately six hours prior to the experiment mosquitoes were starved by removing salt and sugar pads . Mice were anesthetized by intraperitoneal injection of a mixture of ketamine and xylazine ( 87 . 5 mg/kg ketamine and 12 . 5 mg/kg xylazine ) and placed with the ventral side on the mosquito cups . Mosquitoes infected 17–24 days prior to the experiment were allowed to feed for approximately 15 min before mice were removed . During this time eyes of mice were treated with Bepanthen cream ( Bayer ) to prevent dehydration of the cornea . After the experiment mice were allowed to recover and tested for blood stage parasites on a daily basis by evaluation of Giemsa stained blood smears . If non-fluorescent parasite lines were tested infected mosquitoes were not pre-sorted . In these experiments midguts of mosquitoes that had taken a blood meal were dissected after the experiment and the number of midgut sporozoites was counted as described previously . Mice that were bitten by mosquitoes that contained no midgut sporozoites were excluded from the analysis . For injections hemolymph or salivary gland sporozoites were isolated either 13–16 days ( hemolymph ) or 17–24 days ( salivary gland ) post infection . Sporozoite solutions were diluted to the desired concentration with RPMI/PS ( either 10 , 000 or 25 , 000 sporozoites ) and injected intravenously into the tail vein of naive mice . The presence of blood stage parasites was evaluated on a daily basis . To analyze speed and movement pattern of sporozoites , in vitro gliding assays were performed in glass-bottom 96-well plates ( Nunc ) . Hemolymph and salivary gland sporozoites were obtained by dissecting infected A . stephensi mosquitoes . To free the sporozoites from salivary glands , samples were grounded with a pestle . Subsequently salivary gland samples were centrifuged for 3 min at 1 , 000 rpm ( Thermo Fisher Scientific , Biofuge primo ) to separate sporozoites from tissue . Afterwards ~40 µL of the supernatant was transferred into a new 1 . 5 mL plastic reaction tube and diluted with a variable volume of RPMI/PS depending on the planned number of assays and the sporozoite concentration resulting in a minimum of 50 , 000 sporozoites per well . For each assay about 50 µL of the sporozoite suspension was mixed with 50 µL RPMI medium containing 6% bovine serum albumin ( BSA ) to initiate activation . Subsequently sporozoites were allowed to attach to the bottom by centrifugation at 800 rpm for 3 min ( Heraeus Multifuge S1 ) . Using fluorescence microscopy ( Axiovert 200M ) with a 10x objective movies were recorded with one image every three seconds for 3 to 5 min depending on the experiment . Movies were analyzed manually using the Manual Tracking Plugin from ImageJ ( Schindelin et al . , 2012 ) to determine speed and trajectories of moving sporozoites . Sporozoites that were able to glide at least one full circle during a 3 min movie were considered to be productively moving while all other sporozoites were classified as non-productively moving ( moving less than one circle ) or non-motile . For the imaging of sporozoites within salivary glands infected mosquitoes were dissected 17–24 days post infection as described previously . Isolated salivary glands were transferred with a needle to a microscope slide containing a drop of Grace’s medium ( Gibco ) and carefully sealed with a cover slip . Samples were imaged with an Axiovert 200M ( Zeiss ) using 63x ( N . A . 1 . 3 ) and 10x ( N . A . 0 . 25 ) objectives . For immunofluorescence assays we made use of antibodies directed against the circumsporozoite protein ( CSP ) and the thrombospondin related anonymous protein ( TRAP ) . In all assays the anti-CSP antibody mAb 3D11 ( Yoshida et al . , 1980 ) was applied as unpurified culture supernatant of the corresponding hybridoma cell line ( 1:5 diluted for immunofluorescence assays ) . TRAP antibodies were generated against the peptide AEPAEPAEPAEPAEPAEP by Eurogenetec and the purified antibody was applied as 1:100 dilution in immunofluorescence assays . Antibodies against the same peptide have been shown previously to specifically detect TRAP by immunofluorescence and western blotting ( Ejigiri et al . , 2012 ) . Secondary antibodies coupled to AlexaFluor 488 or Cy5 ( goat anti-mouse or goat anti-rabbit ) directed against primary antibodies were obtained from Invitrogen and always used as 1:500 dilution . To visualize the expression and localization of TRAP in sporozoites , infected salivary glands were dissected as described previously and pooled in plastic reaction tubes containing 50 µL PBS or RPMI/PS . Afterwards salivary glands were mechanically grounded with a plastic pestle to release sporozoites from tissue . Immunofluorescence assays were performed by two different methods either fixing sporozoites in solution or on glass cover slips . To fix the parasites on glass , salivary glands were dissected in RPMI/PS and treated as described . Sporozoite solutions were transferred into 24-well plates containing round cover slips , activated with an equal volume RPMI/PS containing 6% BSA and allowed to glide for approximately 30 min at RT . Subsequently the supernatant was discarded and sporozoites were fixed with 4% PFA ( in PBS ) for 1 hr at RT . Fixed samples were washed three times with PBS for 5 min each . If immunofluorescence was performed on sporozoites in solution , salivary glands were dissected in PBS and treated as described previously . Sporozoite solutions were directly fixed by adding 1 mL of 4% PFA ( in PBS ) for 1 hr at RT . After fixation samples were washed as described for samples fixed on glass while samples in solution had to be additionally pelleted after each step by centrifugation for 3 min at 10 , 000 rpm ( Thermo Fisher Scientific , Biofuge primo ) . Subsequently sporozoites were blocked ( PBS containing 2% BSA ) or blocked and permeabilized ( PBS containing 2% BSA and 0 . 5% Triton X-100 ) over night at 4°C or for 1 hr at RT , respectively . Samples were incubated with primary antibody solutions for 1 hr at RT in the dark and subsequently washed three times with PBS . After the last washing step , samples were treated with secondary antibody solutions and incubated for 1 hr at RT in the dark . Stained samples were washed three times in PBS and the supernatant was discarded . If the immunofluorescence assay was performed in solution , sporozoite pellets were resuspended in 50 μL of remaining PBS , carefully pipetted on microscopy slides and allowed to settle for 10–15 min at RT . Remaining liquid was removed with a soft tissue and samples were covered with cover slips which had been prepared with 7 μL of mounting medium ( ThermoFisher Scientific , ProLong Gold Antifade Reagent ) . If the immunofluorescence assay was performed on sporozoites that were fixed on glass , cover slips were removed with forceps , carefully dabbed on a soft tissue and placed on microscopy slides that had been prepared with 7 μL of mounting medium . Samples were allowed to set overnight at RT and kept at 4°C or directly examined . Images were acquired with a spinning disc confocal microscope ( Nikon Ti series ) with 60-fold magnification ( CFI Apo TIRF 60x H; NA 1 . 49 ) . Salivary glands of infected mosquitoes were dissected in 100 µL PBS on ice and subsequently smashed with a pestle to release sporozoites . For the isolation of hemolymph sporozoites infected mosquitoes were flushed with PBS as previously described . Sporozoite solutions were kept on ice , counted using a haemocytometer and distributed to 30 , 000 sporozoites per reaction tube . Samples were centrifuged for 1 min with 13 , 000 rpm ( Thermo Fisher Scientific , Biofuge primo ) at 4°C to pellet sporozoites . Subsequently the supernatant was discarded , and pellets were lysed in 20 µL RIPA buffer ( 50 mM Tris pH 8 , 1% NP40 , 0 . 5% sodium dexoycholate , 0 . 1% SDS , 150 mM NaCl , 2 mM EDTA ) supplemented with protease inhibitors ( Sigma-Aldrich , P8340 ) for ≥1 hr on ice . Lysates were mixed with Laemmli buffer , heated for 10 min at 95°C and centrifuged for 1 min at 13 , 000 rpm ( Thermo Fisher Scientific , Biofuge primo ) . Samples were separated on precast 4–15% SDS-PAGE gels ( Mini Protein TGX Gels , Bio-Rad ) and blotted on nitrocellulose membranes with the Trans-Blot Turbo Transfer System ( Bio-Rad ) . Blocking was performed by incubation in PBS containing 0 . 05% Tween20% and 5% milk powder for 1 hr at RT . Afterwards , the solution was refreshed and antibodies directed against TRAP ( rabbit polyclonal antibody , 1:1000 diluted ) or the loading control CSP ( mAb 3D11 , cell culture supernatant 1:1000 diluted ) were added . Membranes were washed three times ( PBS with 0 . 05% Tween20 ) and secondary anti-rabbit antibodies ( Immun-Star ( GAR ) -HRP , Bio-Rad ) or anti-mouse antibodies ( NXA931 , GE Healthcare ) conjugated to horse radish peroxidase were applied for 1 hr ( 1:10 , 000 dilution ) at room temperature . Signals were detected using SuperSignal West Pico Chemiluminescent Substrate or SuperSignal West Femto Maximum Sensitivity Substrate ( Thermo Fisher Scientific ) . After the detection of TRAP , blots were treated with stripping buffer ( Glycine 15 g/L , SDS 1 g/L , Tween20 10 ml/L , pH 2 , 2 ) for 15 min prior to incubation with anti-CSP antibodies used as loading control . HepG2 cells were a gift from our virology department who had obtained the cells from ATCC . Cell line identity was regularly confirmed by SNP sequencing and visual observations of cell morphology . Cells were routinely tested for mycoplasma contamination using the MycoAlert mycoplasma detection kit ( Lonza , Basel , Switzerland ) . We culture cells for a maximum of 10 passages . Two days prior to the experiment 50 , 000 HepG2 cells/well were seeded in an 8-well Permanox Lab-Tek chamber slide ( Nunc ) . On day zero salivary glands were isolated from infected female mosquitoes and collected in RPMI medium within a 1 . 5 mL reaction tube . Sporozoites were released by mechanically disrupting the salivary glands with a polypropylene pestle . The solution was centrifuged in a tabletop centrifuge at 1 , 000 rpm for 3 min at RT and the supernatant was transferred to a new 1 . 5 mL reaction tube . The salivary gland pellet was resuspended in 100 µL RPMI medium , smashed again using a pestle , centrifuged ( 1 , 000 rpm , 3 min , RT ) and the supernatant was pooled with the first one . A 1:10 dilution of the sporozoite solution was counted in a Neubauer counting chamber and 10 , 000 salivary gland sporozoites were used to infect HepG2 cells per well . After 1 . 5 hr wells were washed twice with complete DMEM medium and HepG2 cells were allowed to grow in complete DMEM medium supplemented with 1x Antibiotic-Antimycotic ( Thermo Fisher Scientific ) . At 24 hr and 48 hr post infection cells were fixed using ice cold methanol for 10 min at RT , followed by blocking with 10% FBS/PBS overnight at 4°C . Staining with primary antibody α-PbHSP70 1:300 in 10% FBS/PBS for 2 hr at 37°C was succeeded by two washing steps with 1% FBS/PBS . Incubation with secondary antibody α-mouse Alexa Fluor 488 1:300 in 10% FBS/PBS was performed for 1 hr at 37°C . Hoechst 33342 was added and incubated for 5 min at RT followed by two washing steps with 1% FBS/PBS . The assay was mounted in 50% glycerol and sealed using a glass cover slip . Samples were imaged with an Axiovert 200M ( Zeiss ) microscope and subsequently analyzed using ImageJ ( Schindelin et al . , 2012 ) . In brief , the perimeter of single liver stages was encircled , and the area measured using the internal measurement tool . Two days prior to the experiment 180 , 000 HepG2 cells/well were seeded in an 8-well Permanox Lab-Tek chamber slide ( Nunc ) . Sporozoites were isolated as described above and 10 , 000 sporozoites/well were used to infect HepG2 . At 1 . 5 hr post infection cells were washed twice with complete DMEM medium and fixed using 4% PFA/PBS 20 min at RT . Blocking was performed with 10% FBS/PBS o/n at 4°C followed by incubation with primary antibodies α-PbCSP 1:100 in 10% FBS/PBS ( 2 hr 37°C ) , two washing steps with 1% FBS/PBS and incubation with secondary antibodies α-mouse Alexa Fluor 488 1:300 in 10% FBS/PBS ( 1 hr 37°C ) . After two washing steps with 1% FBS/PBS , cells were permeabilized by addition of ice cold methanol and incubation at RT for 10 min . Blocking with 10% FBS/PBS ( 4°C and overnight ) was followed by an incubation with primary antibodies α-PbCSP 1:100 in 10% FBS/PBS ( 2 hr at 37°C ) , two washing steps with 1% FBS/PBS and incubation with primary antibodies α-mouse Alexa Fluor 546 1:300 in 10% FBS/PBS ( 1 hr at 37°C ) . The assay was mounted in 50% glycerol after two washing steps with 1% FBS/PBS . To measure the parasite load of different organs C57BL/6 mice were infected by intravenous injection of 20 , 000 salivary gland sporozoites . At 42 hr after infection organs were harvested , homogenized and the RNA was isolated using Trizol according to the manufacturer's protocol . Isolated RNA was treated with DNase using the Turbo DNA-free Kit ( Invitrogen ) . Subsequently RNA content of generated samples was measured using a NanoDrop and liver , intestine , spleen and lung samples were pooled in equal amounts of RNA to generate single samples for each parasite line and harvested organ . RNA pools were used to synthetize cDNA using the First Strand cDNA Synthesis Kit ( Thermo Fisher Scientific ) . Quantitative RT-PCR was performed in triplicates on an ABI7500 ( Applied Biosystems ) using a 2x SYBR green Mastermix ( Applied Biosystems ) . Plasmodium berghei 18S rRNA was used to quantify parasites and mouse specific GAPDH was utilized as housekeeping gene for normalization . Subsequently the ∆cT was plotted as mean of all replicates per parasite strain and harvested organ ( Schmittgen and Livak , 2008 ) . For all experiments female 4–6 week-old Naval Medical Research Institute ( NMRI ) mice or C57BL/6 mice obtained from Janvier laboratories were used . Transgenic parasites were generated in the Plasmodium berghei ANKA background ( Vincke and Bafort , 1968 ) either directly in wild type or from wild type derived strains ( e . g . trap ( - ) and fluo ) . Parasites were cultivated in NMRI mice while transmission experiments with sporozoites were performed in C57Bl/6 mice only . Statistical analysis was performed using GraphPad Prism 5 . 0 ( GraphPad , San Diego , CA , USA ) . Data sets were either tested with a one-way ANOVA or a Student’s t test . A value of p<0 . 05 was considered significant .	Malaria is an infectious disease caused by single-celled parasites known as Plasmodium . Humans and other animals with backbones – such as birds , reptiles and rodents – can become hosts for these parasites if an infected female mosquito feeds on their blood . Likewise , healthy mosquitoes can in turn become infected with Plasmodium if they feed on the blood of an infected animal . To complete their life cycle , Plasmodium parasites within a mosquito must become spore-like cells called sporozoites . These sporozoites are highly mobile and can get into the mosquitoes’ salivary glands , meaning they can be passed on to a new host when the insect feeds . During a mosquito bite the sporozoites are spat into the skin of the potential host , where they then need to migrate rapidly to enter the bloodstream . Once in the blood , the sporozoites can then get into liver cells and begin a new infection . One protein called TRAP , which is found on the surface of the sporozoites , is important for their migration and the infection of the salivary glands or liver . Yet it was not known how this happens at the level of the individual proteins involved . Klug et al . have now tested how a part of the TRAP protein , called the I domain , contributes to the infection process . In the experiments , the I domain of TRAP was deleted which showed that the sporozoites need this domain to be able to move around and get into the host tissues . Without the I domain the sporozoites were stuck and could not successfully infect either the mosquitoes , the livers of mice , or human liver cells grown in the laboratory . Klug et al . then replaced the Plasmodium I domain of TRAP with the I domain from a distantly related parasite called Toxoplasma gondii , which causes a condition known as toxoplasmosis . The I domain from Toxoplasma allowed the Plasmodium parasites to infect the host tissues again . This observation was unexpected because Toxoplasma and Plasmodium parasites have evolved separately over the last 800 million years and Toxoplasma does not infect insects . These findings suggest that the I domain of TRAP evolved to bind several other proteins in different tissues and hosts . Future studies will investigate which other parasite proteins TRAP works with to guide sporozoites to the salivary glands or liver . Knowledge of how these proteins act together may lead to new approaches for treating or preventing malaria . For example , some treatments could stop sporozoites from entering liver cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2020	Evolutionarily distant I domains can functionally replace the essential ligand-binding domain of Plasmodium TRAP
The robust representation of the environment from unreliable sensory cues is vital for the efficient function of the brain . However , how the neural processing captures the most reliable cues is unknown . The interaural time difference ( ITD ) is the primary cue to localize sound in horizontal space . ITD is encoded in the firing rate of neurons that detect interaural phase difference ( IPD ) . Due to the filtering effect of the head , IPD for a given location varies depending on the environmental context . We found that , in barn owls , at each location there is a frequency range where the head filtering yields the most reliable IPDs across contexts . Remarkably , the frequency tuning of space-specific neurons in the owl's midbrain varies with their preferred sound location , matching the range that carries the most reliable IPD . Thus , frequency tuning in the owl's space-specific neurons reflects a higher-order feature of the code that captures cue reliability . Perception relies on sensory cues that are used by the brain to infer properties of the environment . For example , the ability to see the world in three dimensions depends on cues that signal depth ( Howard , 2012 ) . Similarly , sound localization , relies on auditory spatial cues including phase differences of sounds between the ears ( Moiseff and Konishi , 1983; Grothe et al . , 2010 ) . Variability of sensory cues is intrinsic to the physics of stimuli and sensory organs . For instance , light and sound waves are reflected and absorbed differently by various media depending on wavelength and location ( Carlile , 1996 ) . Multiple light and sound sources can also physically interfere with each other . In the auditory system , contexts such as whether the environment is reverberant , quiet or noisy can influence spatial cues greatly . The presence of concurrent sounds can shift auditory cues used for localizing a target sound ( Keller and Takahashi , 2005 ) . This shift makes cues differ from what would be measured if the sound was emitted in a quiet environment . To be considered reliable , cues associated with a given location must be similar across different contexts . Unreliable cues , on the other hand , vary across contexts . The brain must take into account the reliability of sensory cues in order to make perception robust to natural variations . A possible strategy to overcome these variations is to integrate sensory cues with reference to their variability . Indeed , when cues provide ambiguous or conflicting information , those cues that vary less are weighted more heavily in perceptual judgments ( Ernst and Banks , 2002; Jacobs , 2002; Alais and Burr , 2004 ) . Humans and other animals use the interaural time difference ( ITD ) for sound localization ( Grothe et al . , 2010 ) . ITD is the difference in the arrival time of a sound at the ears . ITD results from unequal distances of a sound source to the two ears when the source is to the left or to the right of the listener . In barn owls , ITD is the main cue for localizing in the horizontal space ( Moiseff and Konishi , 1983; Moiseff , 1989 ) . ITD is initially detected by brainstem neurons tuned to narrow frequency bands in both mammals and birds ( Carr and Konishi , 1990; Schnupp and Carr , 2009; Figure 1A ) . These neurons compare the timing of inputs from the left and right sides of the brain ( Carr and Konishi , 1990 ) . Due to the periodicity of sound signals when narrow frequency channels are considered , the shift in time between the left and right ears for each frequency component is more precisely expressed in terms of phase , referred to as the interaural phase difference ( IPD ) . The spectrum of IPDs across frequency serves as a set of cues used for sound localization . 10 . 7554/eLife . 04854 . 003Figure 1 . Cue variability in the sound localization system . ( A ) Tuning to interaural time difference ( ITD ) emerges from the convergence of inputs selective for the same ITD but different frequencies ( F1–F3 ) ( Takahashi and Konishi , 1986; Mazer , 1998; Pena and Konishi , 2000; best ITD indicated by the dashed line ) . While ITD-selective neurons respond to a broad range of frequencies ( here the black bold curve represents the neuron's combined response for frequencies F1 , F2 and F3 ) , their inputs are narrowly tuned to frequency ( each input only responding to F1 , F2 or F3 ) . Because the inputs are narrowly tuned to frequency , the responses at each input vary with the phase difference between the left and right ears ( IPD ) of their preferred frequency , as shown by the sinusoidal curves . ( B ) A sound emitted by a single source ( B , left ) in front of the owl is filtered by the head and decomposed in narrow frequency channels by the cochlea . The localization cue corresponds to an IPD in each frequency channel ( F1–F3 ) . In a different context ( B , right ) , the target frontal sound ( yellow ) is emitted concurrently with another sound source from a different location ( blue ) . The blue source interferes with the yellow target and shifts the resultant IPDs in each frequency channel ( shown in green ) . Black dotted lines indicate IPD responses for the target frontal source alone , for comparison . While IPD shifts greatly for some frequencies ( F1 and F3 ) , in others ( F2 ) IPD is more robust to the presence of another source . Thus in this example , F2 carries the most reliable IPD cue . To provide a clearer visualization that IPD is encoded at different frequencies , F1–F3 inputs remain plotted as a function of ITD in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 003 Previous studies of sound localization suggest that the emergence of space-specific neurons in the midbrain is due to a wide convergence of frequency channels ( Brainard et al . , 1992 , Mazer , 1998 , Saberi et al . , 1998; Figure 1A ) . These studies focused on how the brain infers sound location from IPD without taking into account that IPD at a given sound location may vary between contexts in a frequency-dependent manner . In fact , a direct consequence of the filtering effect of the head is that the IPD induced by different acoustic objects varies in each frequency channel depending on context ( Blauert , 1997; Keller and Takahashi , 2005 ) . For example , the phase of a target sound can be shifted by the presence of a concurrent sound and the amount of phase shift is frequency dependent ( Keller and Takahashi , 2005; Figure 1B ) . Thus , for a given location of a sound source , the IPD cue may vary in a frequency-dependent manner in the presence of another sound . In other words , IPD may not be equally reliable as a sound localization cue at every frequency band . Whether the neural code for sound localization captures the reliability of auditory cues across contexts is unknown . A common strategy for dealing with the natural variation of sensory cues is to weight cues in proportion to their reliability ( Ernst and Banks , 2002; Jacobs , 2002; Alais and Burr , 2004; Fetsch et al . , 2012 ) . For sound localization based on IPD , this would mean that those frequencies that elicit more reliable IPD ( i . e . , where the variability of IPD is small ) at a given location across different contexts should be favored in the process of estimating this particular location . We will refer to this mechanism as weighting by reliability . To test whether weighting by reliability occurs in sound localization , one can examine populations of neurons that integrate localization cues . A neural representation of auditory space emerges in the barn owl's external nucleus of the inferior colliculus ( ICx ) . Thus , the barn owl's ICx provides an opportunity to test whether weighting by reliability is used to map sound location from variable cues . In the present study , we demonstrate that frequency tuning of ICx neurons changes with their preferred ITD . This coding captures the variability of IPD across frequency channels in a manner consistent with weighting neural responses by cue reliability . To test whether weighting by reliability occurs in the owl's sound localization pathway , we first mapped the spatial and frequency tunings of the entire ICx . 177 single units obtained from 138 recording sites in the ICx of two adult barn owls were included in this analysis . Single unit recordings were validated by spike sorting software ( Quiroga et al . , 2004 ) in 99 of the 138 recordings sites , whereas in 39 recording sites spike sorting separated two different units . Therefore , while the majority of recordings consisted of single units , some of them yielded two units . ITD and frequency tuning were measured for each neuron . We estimated best ITD from the peak of the rate-ITD tuning curve ( Figure 2A , top row ) and the best frequency ( BF ) from the center of the rate-frequency tuning curve ( Figure 2A , bottom row ) . The neurons' best ITD is correlated with the preferred azimuth in the map of auditory space of the owl's ICx ( Moiseff and Konishi , 1983 ) . To achieve a representative assessment of the neural population , we recorded responses of ICx neurons with best ITDs spread over the entire map . Best ITD varied from 0 to 249 µs and BF varied from 920 Hz to 6168 Hz . To the best of our knowledge , neurons tuned to such large ITDs or low BFs have not been reported in the owl's ICx ( Perez and Pena , 2006; Wagner et al . , 2007; Vonderschen and Wagner , 2009 ) . As shown in Figure 2 , best ITD was tightly correlated with BF ( r2 = 0 . 75 , p < 0 . 001 ) . Neurons preferring small ITDs were tuned to higher frequencies and , conversely , those preferring large ITDs were tuned to lower frequencies . The strong correlation between best ITD and BF was not affected whether the analysis was performed using recording sites with a single unit ( n = 99 , r2 = 0 . 71 , p < 0 . 001 ) , pooled recordings of single and multi-units ( n = 138 , r2 = 0 . 72 , p < 0 . 001 ) or all sorted single units ( n = 177 , r2 = 0 . 75 , p < 0 . 001 ) . All further analyses were thus performed on sorted single-unit data . 10 . 7554/eLife . 04854 . 004Figure 2 . Spatial-dependence of frequency tuning in the population of ICx neurons . ( A; top ) ITD tuning measured with broadband noise of three example neurons tuned to 0 µs ( blue ) , 100 µs ( red ) and 200 µs ( yellow ) . ( A; bottom ) The frequency tuning , measured with tones at the best ITD shows that best frequencies ( BF ) decrease as best ITD increases for each neuron ( BFs of the shown examples are 6 kHz ( blue ) , 4 kHz ( red ) and 2 kHz ( yellow ) ) . ( B ) BF decreases with best ITD across the sample of ICx neurons . Linear regression is indicated by a solid line . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 004 To investigate IPD reliability , we examined the statistics of the auditory input for the owl . Specifically , we considered how IPD varies when concurrent sources are present . The presence of concurrent sounds is likely a primary source of variability of IPD in the owl's natural environment because sounds made by prey are often faint ( Konishi , 1973 ) and concurrent sounds from different locations can dramatically shift the measured IPD ( Keller and Takahashi , 2005 ) . In the owl , and other species , sounds are modified in a location-dependent manner by facial structures . In barn owls , which lack a pinna , the facial ruff and ear canal act as filters ( Keller et al . , 1998; von Campenhausen and Wagner , 2006 ) embodied in the head-related transfer functions ( HRTFs; Keller et al . , 1998 ) . These filters change both the phase and the magnitude of each frequency component of the sound in a location-dependent manner . When sounds are coming from multiple sources , the sound waves from each source will add in the ears and alter binaural cues . As described in Keller and Takahashi ( 2005 ) , the binaural cues resulting from the mixture of multiple sound sources are dictated by the relative intensity of each source within each frequency band . If two sources emitting sounds at the same intensity differ in the IPD within a frequency band , the resultant IPD is the average of the IPDs from the individual sources . When a target source carries more power in a particular frequency band , the resulting IPD will shift closer to this source . If the power of the second source is larger at a given frequency band , the IPD at this band is drawn away from the target source . Figure 3A shows an example of the relative variation in IPD around the mean IPD caused by the presence of concurrent sources . In this example the IPDs generated by one source located at 5° of azimuth and 0° of elevation ( black dots ) is contrasted with IPDs obtained when a second source , also at 0° elevation , is added at azimuths ranging between −90 and 90° ( grey dots ) . For this azimuthal location , IPD variability induced by a second source is greatest between 3 and 4 kHz . 10 . 7554/eLife . 04854 . 005Figure 3 . IPD variability computed from the head filters . ( A ) Example IPD variation around the mean ( normalized to zero for clarity ) as a function of frequency when a sound is presented near the front ( black dots ) and when a second sound source is added from various locations between −90 and 90° ( gray dots ) . Note the larger scatter of gray dots in the lower frequencies ( B ) IPD standard deviation over concurrent sound sources across frequency at the location of the target sound . The dotted lines show the location of the target sound in ( A ) . In ( A ) and ( B ) units are percent of cycle . ( C ) IPD reliability across location and frequency , normalized by the maximum at each location . ( D ) Standard deviation of the IPD reliability across HRTFs from 10 different owls . Note that the same colors map the values 0 ( least reliable IPD ) to 1 ( most reliable IPD ) in C and 0–0 . 25 in D . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 005 We measured IPD variability induced by the presence of concurrent sound sources using the HRTFs of 10 owls . We computed IPD variability on a frequency-by-frequency basis as the standard deviation of IPD ( expressed in percent of cycle ) over different locations of the second source ( within ±90deg ) . The presence of a concurrent sound had a powerful effect on IPD , accounting for variations of up to 30% of a cycle . The IPD variability was highest for locations in the periphery , especially at high frequencies ( Figure 3B ) . The presence of two sources also increased IPD variability at low frequencies for target locations near the center ( Figure 3A , B ) . This pattern of IPD variability with concurrent sounds can be explained by the direction and frequency dependence of the intensity gain of the HRTFs . The intensity gain is largest for azimuths near the front and decreases significantly for eccentric locations ( Keller et al . , 1998 ) . Therefore , the IPD of a target source placed near the center will not be shifted greatly when a second source is placed at eccentric directions . Conversely , the IPD of a target source in the periphery will be shifted significantly when a second source is placed at a central location with high intensity gain . The difference in gain between frontal and peripheral locations is highest at high frequencies . Therefore , the largest variability was observed for high frequencies in the periphery . We then took the inverse of the IPD variance as a measure of the reliability of IPD ( Figure 3C ) . We assumed that an ICx neuron with a given preferred location will weight the inputs from different frequencies according to their relative reliability at that location . We therefore normalized the reliability within each location . We found that IPD reliability depends on the filtering of sounds by the head in a systematic manner across frequency and locations . Overall , IPD reliability was greater at high frequencies for locations in the front and at lower frequencies for locations in the periphery ( Figure 3C ) . The overall pattern of IPD reliability was consistent across HRTFs from 10 different owls as illustrated by the small variance across animals ( Figure 3D ) . If the frequency tuning of ICx neurons were driven by cue reliability , then we would expect a strong relationship between frequency tuning and IPD reliability . The dependence of BF on ITD tuning ( Figure 2 ) was predicted by the IPD reliability . Across neurons , their BF and the frequency at which IPD reliability was maximal at their preferred location were highly correlated ( r2 = 0 . 81 , p < 0 . 001 ) . Additionally , we compared the lower- and upper- edges of the neurons' frequency tuning curves with the range of frequency that carried the most reliable IPD ( see ‘Materials and methods’ ) . The upper- and lower-frequency edges of the frequency tuning curves followed the upper ( r2 = 0 . 72 , p < 0 . 001 ) and lower edges ( r2 = 0 . 52 , p < 0 . 001 ) of the range of most reliable IPD ( Figure 4A ) . Thus , the measured frequency tuning fell within the boundaries predicted by the IPD reliability . Since the match between frequency tuning curves and IPD reliability was assessed for each location , the normalization of the IPD reliability had no effect on these calculations . 10 . 7554/eLife . 04854 . 006Figure 4 . Frequency tuning in ICx matches IPD reliability . ( A ) Upper ( white circles ) and lower ( black circles ) edges of frequency tuning of ICx neurons superimposed on the plot of IPD-reliability . ( B ) Upper and lower boundaries of frequency tuning superimposed on the average gain normalized at each location ( from 0 to 1 ) . ( C ) The correlation coefficients between frequency tuning curves and frequency tuning predicted by the IPD-reliability ( gray ) are higher than the correlation coefficients between frequency tuning curves and gain alone ( black ) . ( D ) Upper and lower edges of frequency tuning superimposed on the average gain normalized successively at each frequency and at each location ( from 0 to 1 ) . ( E ) The correlation coefficients between frequency tuning curves and frequency tuning predicted by the IPD-reliability ( gray ) are higher than the correlation coefficients between frequency tuning curves and the normalized gain ( black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 006 To verify that additional sound sources did not change the pattern of IPD variability , we computed the IPD reliability with three concurrent sounds . The measured BF and the center frequency at which IPD reliability from three concurrent sounds was maximal were also highly correlated ( r2 = 0 . 77 , p < 0 . 001 ) . ICx neurons are highly sensitive to interaural correlation , defined as the cross-correlation of the signals at the two ears ( Albeck and Konishi , 1995 ) . The addition of a concurrent source may decrease the interaural correlation at the preferred location of ICx neurons . Thus , in addition to the shift in IPD , the presence of concurrent sounds could modulate the firing rate of ICx neurons by changing interaural correlation . To test whether the IPD reliability is consistent with the pattern of interaural correlation induced by concurrent sounds , we examined the mean interaural correlation at the ITD of the target source within each frequency channel while concurrent sounds from other locations were presented . We found that the pattern of interaural correlation as a function of frequency and location was very similar to the IPD reliability ( r2 = 0 . 81 , p < 0 . 001 ) . Like IPD reliability , the mean interaural correlation was greater at high frequencies for locations in the front and at lower frequencies for locations in the periphery . The measured BF and the center frequency at which the mean interaural correlation was maximal were also highly correlated ( r2 = 0 . 88 , p < 0 . 001 ) . This shows that weighting frequencies by their reliability maintains frequencies with high input coherence to ICx neurons at each location . The change in sound-level due to filtering by the head , referred to as gain , is also frequency- and location-dependent ( Keller et al . , 1998; Figure 4B ) and could drive frequency tuning . To investigate whether reliability of IPD or simply sound level was more important in driving the tuning of ICx neurons , we asked whether frequency tuning was better predicted by IPD reliability ( Figure 4A ) or by the gain ( Figure 4B ) . Gain and IPD reliability patterns displayed similarities , as expected from the fundamental relationship between the interaction of concurrent sound sources and their gain ( Keller and Takahashi , 2005 ) ; IPD at locations with lower gain is more susceptible to variation in the presence of other sounds than is IPD at locations with higher gain . Additionally , correlation between IPD reliability and gain may also arise from common causes of variation such as acoustic reflections at the facial ruff ( von Campenhausen and Wagner , 2006 ) . However , the frequency tuning of neurons did not match the gain pattern as well as it matched the IPD reliability as a function of frequency at the best ITD ( Figure 4A , B ) . We computed the correlation coefficients between the frequency tuning curves and both the IPD reliability and gain as a function of frequency at the best ITD of each neuron . Out of 177 neurons , the frequency tuning curves of 145 neurons ( 82% ) were significantly correlated with IPD reliability as a function of frequency at the best ITD ( Figure 4C , gray histogram ) whereas only 85 neurons ( 48% ) had their frequency tuning curves significantly correlated with the gain as a function of frequency at the best ITD ( Figure 4C , black histogram ) . Further , the correlation between frequency tuning curves and IPD reliability yielded higher correlation coefficients than between frequency tuning and gain ( p < 0 . 001 , Wilcoxon rank-sum test ) . It has also been shown that the relative gain of different frequency bands can be a cue for sound localization ( Butler , 1987; Butler and Musicant , 1993 ) . A critical feature of the owl's facial ruff is its ability to increase the intensity of the sound of high over low frequencies in the frontal space ( Keller et al . , 1998 ) . At low frequencies , the owl's facial ruff has a relatively small location-dependent effect on sound level ( Hausmann et al . , 2010 ) . Thus , the relative gain of high and low frequencies at different locations could provide a cue for stimulus location . To test whether the relative gain of each frequency could predict the ITD tuning , we normalized the gain across locations for each frequency separately before normalizing the gain across frequency for each location ( Figure 4D ) . Once again , the results for the gain did not correlate with the experimental frequency tuning as well as the IPD reliability did ( Figure 4D , E , p < 0 . 001 Wilcoxon rank-sum test ) . Only 44 neurons ( 25% ) were significantly correlated with the normalized gain ( Figure 4E , black histogram ) . Thus IPD reliability yielded a better prediction of the frequency tuning than did the gain . To test whether changes in neural responses when concurrent sounds are present were consistent with predictions made from the HRTFs , we examined spatial tuning in neurons of the core of the central nucleus of the inferior colliculus ( ICCc ) . ICCc is located earlier in the pathway leading to ICx and contains ITD-sensitive neurons that are narrowly tuned to frequency ( Wagner et al . , 2002 ) . While ITD tuning varies with frequency in ICx , such dependence is not observed in ICCc ( Wagner et al . , 2002 , 2007 ) . Thus , recording in ICCc allowed us to assess sensory-input variability before frequency convergence occurs . Because the predictions due to reliability or to gain differed most between high and low frequencies in frontal locations ( Figure 4A , B , D ) , we explored this range in ICCc using concurrent sounds . We measured the spatial tuning of ICCc cells using single sources and while another sound was presented from different locations covering the frontal hemisphere on the horizontal plane ( Figure 5A , see ‘Materials and methods’ ) . To quantify the effect of concurrent sounds in altering the spatial tuning of ICCc cells , we cross-correlated the spatial tuning curves obtained with single sources with those measured with concurrent sounds ( Figure 5B ) . We found that the spatial tuning of ICCc cells tuned to the front ( preferred location between 5 and 20° ) was more affected by the presence of another sound , that is , more variable , at lower than at higher frequencies ( Figure 5A , B , r2 = 0 . 6 , p < 0 . 001 ) . Experiments were performed in a high quality anechoic chamber ( see ‘Materials and methods’ ) , thus we consider it unlikely that the variability in spatial tuning of low frequency ICCc neurons is due to room reflections . Therefore the data from ICCc neurons confirm our prediction that the variability induced by the HRTFs is carried over the sound localization pathway . 10 . 7554/eLife . 04854 . 007Figure 5 . Testing the weighting by reliability hypothesis . ( A ) The colored curves show the normalized spatial tuning of two example ICCc neurons whose best frequencies are 1 kHz ( blue ) and 4 kHz ( yellow ) . The spatial tuning was measured by varying the location of single sound sources using an array of speakers . The gray curves show the normalized spatial tuning of the same neurons measured with an additional concurrent sound at another location ( nine different locations of concurrent sounds were tested in total ) . Examples of the spatial tuning measured in the presence of a concurrent sound source at 10° , 50° and 90° are displayed . The tuning of the low frequency neuron ( top row ) is more affected by concurrent sounds than the tuning of the higher frequency neuron ( bottom row ) . ( B ) For each neuron , we correlated the curves measured with single sound sources with the curves measured with concurrent sound sources . The mean correlation coefficients between spatial tuning curves using a single sound and the tuning curves using concurrent sounds increases with best frequency . The linear regression is indicated by a solid line . The similarity between the curves in different contexts increases with best frequency . ( C ) Box-plot showing median ( blue line ) and quartiles of the Fano factor distribution . The Fano factors spread around 1 , indicating trial-to-trial variability similar to a Poisson response . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 007 It has been proposed that the interaural canal connecting the middle ear cavities in birds could affect ITD at low frequencies ( Calford and Piddington , 1988 ) . Because HRTFs are measured using microphones positioned at the ear canal , they are blind to the effect of the interaural canal . The largest ITD predicted by the HRTFs at low frequencies was similar to the largest best ITD we recorded in ICx . Therefore it does not appear that the interaural canal has an important effect increasing the magnitude of ITD . However , if the interaural canal increased the gain of low frequencies dramatically in the frontal space , it could change the variability induced by concurrent sound sources . Yet , our results in ICCc are consistent with predictions made from the HRTFs without the need to invoke an effect of the interaural canal . There are a number of possible sources of neural noise that may contribute to the frequency dependence of IPD variability . While neural noise would not affect our measurements of BF or best ITD , which rely on tuning functions using mean firing rate , frequency-dependent neural noise during development could influence the learned connectivity in the auditory system that establishes the correlation between BF and best ITD . To assess the frequency dependence of neural noise we computed the Fano factors , a measure of trial-to-trial response variability , for ITD tuning curves obtained with tonal stimulation . We measured ITD tuning curves using tones from 1 to 7 kHz in 70 ICx cells ( total of 342 curves ) . If neural noise were correlated with frequency , Fano factors should vary with stimulus frequency . We found that the Fano factors spread around 1 ( Figure 5C , median = 1 . 06 ) and were negligibly correlated with the stimulus frequency ( r2 = 0 . 1 , p = 0 . 03 ) . We also assessed whether neural noise is frequency-dependent when examined at different locations . We split the neurons into 4 groups according to their best ITDs ( 0–50 µs; 50–100 µs; 100–150 µs; larger than 150 µs ) . For each group we averaged the Fano factors of the different neurons across the same stimulating frequency . We found no relationship between the Fano factor and frequency as a function of best ITD in the first 3 groups ( Group1: r2 =−0 . 08 , p = 0 . 68; Group2: r2 = 0 . 007 , p = 0 . 92; Group3: r2 = 0 . 1 , p = 0 . 62 ) . For best ITDs larger than 150 µs , the Fano factor decreased as frequency increased ( r2 =−0 . 7 , p < 0 . 001 ) . However , if neural noise drove the low-pass frequency tuning at eccentric locations , the opposite relationship would be expected ( higher frequencies noisier than lower frequencies ) . Thus , frequency-dependent trial-to-trial variability in neural responses cannot explain the correlation between best ITD and frequency tuning . Phase locking often weakens at higher frequencies , thus potentially increasing IPD variability at these frequencies ( Koppl , 1997b ) . To test this , we examined whether the strength of IPD tuning varied with the stimulation frequency using a synchronization coefficient ( Goldberg and Brown , 1969; Kuwada et al . , 1987 ) . In the same dataset used to assess the Fano factor ( n = 342 ) , ITD curves were folded into IPD curves ( see ‘Materials and methods’ ) . The response at each IPD was treated as a vector with direction given by the IPD and length given by the mean firing rate at that IPD . The synchronization coefficient was the amplitude of the mean IPD vector divided by the sum of the mean firing rates for the entire period . Coefficients decreased minimally with frequency ( r2 = 0 . 14 , p < 0 . 001 ) below 7 kHz . In sum , we found no evidence consistent with neural noise explaining the frequency-dependent ITD tuning . Thus , context dependence of the auditory cues resulting from the filtering properties of the HRTFs appeared to be the main source of location-dependent variability and the primary mechanism to adjust tonotopy along with spatial tuning . In the present study we linked reliability of sensory cues with tuning properties of auditory neurons . Sounds at a given position do not yield identical localization cues over different contexts . However , for each location , there is a frequency range within which the localization cue IPD is most reliable . We showed that ICx neurons limit their frequency tuning to this range ( Figure 6 ) . Thus , the frequency tuning in the space-specific neurons of ICx is not simply due to tonotopy inherited from upstream neurons , but rather reflects a higher-order aspect of the neural code that may contribute to a more robust representation of sound location . Our study provides a case for how stimulus statistics can be captured by the neural processing . 10 . 7554/eLife . 04854 . 008Figure 6 . Adjusting tonotopy through weighting by reliability to represent space . Neurons receive inputs where different frequencies ( F1–F6 ) are weighted by the IPD variance . For neurons tuned to frontal space ( left ) , a larger weight is assigned to high frequencies where IPD is less variable , while neurons tuned to more peripheral space ( middle and right ) receive stronger input at lower frequencies . The effect of the head on IPD variability is indicated by the color ( green is less variable ) and superimposed sine waves ( superimposed sinusoids shifted in phase indicate more variability ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04854 . 008 Consistent with a trend reported in the optic tectum ( Knudsen , 1984 ) , we showed that the distribution of preferred frequency varies systematically with preferred ITD in ICx . This means that at each frequency , preferred ITDs are limited to a narrow part of the natural range . This is in stark contrast to the broad distribution of best ITDs across frequency reported in the upstream nucleus ICCc ( Wagner et al . , 2002 , 2007 ) . Evidence for best ITDs covering the natural range in nucleus laminaris ( NL ) comes from studies of neurophonic potentials ( Sullivan and Konishi , 1986 ) and axonal delays of input fibers to NL ( Carr and Konishi , 1990 ) . However , recent studies show that the methodology relying on the neurophonic to estimate the best ITDs of actual neurons , as opposed to the compound field potential from input fibers , may lead to spurious conclusions ( Mc Laughlin et al . , 2010; Kuokkanen et al . , 2013 ) . This suggests that the range of best ITDs at each frequency in NL deserves further investigation . Selectivity for reliable localization cues could develop by Hebbian mechanisms favoring the least variable inputs . Indeed , the tuning of ICx cells adjusted to modifications of sound localization cues in juvenile and adult owls that had their facial ruff removed ( Knudsen et al . , 1994 ) . Frequency-specific plasticity was also observed in owls raised with an acoustic filtering device that altered ITD in a frequency-dependent manner ( Gold and Knudsen , 2000 ) . When the device shifted the ITD by more than 50 µs , the ITD tuning of neurons also shifted by more than 50 µs compared to control owls . These studies indicate that ITD and frequency tunings can be adjusted by experience . Frequency-dependent ITD tuning has been demonstrated in several other species and different brain regions , including the inferior colliculus of the guinea pig ( McAlpine et al . , 2001 ) and cat ( Hancock and Delgutte , 2004; Joris et al . , 2006 ) , the medial superior olive of the gerbil ( Brand et al . , 2002; Day and Semple , 2011 ) , and the lateral lemniscus of the chinchilla ( Bremen and Joris , 2013 ) . Thus , frequency-dependent ITD tuning is observed not only in cases where best ITDs fall largely out of the physiological range , such as in small rodents ( McAlpine et al . , 2001; Brand et al . , 2002 ) , but also in maps of auditory space where neurons are tuned to physiological ITDs , such as in the owl ( Wagner et al . , 2007 ) . The correlation between ITD and frequency tuning thus represents a general organizational principle across species and coding schemes ( Schnupp and Carr , 2009 ) for sound localization . It has been proposed that owls and small mammals use different strategies to encode IPDs based on head size and the frequency range over which ITD can be encoded ( Harper and McAlpine , 2004 ) . The model of Harper and McAlpine ( Harper and McAlpine , 2004 ) proposes that the optimal distribution of preferred IPDs at each frequency will maximize the information that the population provides about IPD . In this framework , the optimal distribution of preferred IPDs at each frequency depends on the statistical distribution of IPD from the environment . This theory predicts that preferred IPDs should cover the natural range of IPD at each frequency that is relevant for sound localization in the owl . Harper and McAlpine took into account the statistics of the human acoustic cue to predict the optimal relationship between best IPD and frequency in humans . However , their predictions about the optimal neural code in the owl used a distribution of IPD that depends only on head size , and not on the IPD distribution occurring in natural environments . While this optimal coding model suggests that the distribution of IPD in natural environments may be an important factor influencing the neural code for sound localization , the model only addresses the representation of IPD at each frequency , and does not address how sound localization cues should be integrated over frequency to produce spatially selective auditory neurons . Here , we show that in the owl , spatial dependence of frequency tuning can be explained by a code that captures the range of frequency that carries the most reliable IPD . This study sheds light on an outstanding question in sound localization: Do neurons match the pattern of IPD across frequency experienced by each species ? Our results suggest that neurons are tuned to the frequency range within which IPD varies least for each location over different environmental conditions . Thus , rather than neurons matching the IPD spectrum over the audible frequency range ( Brainard et al . , 1992; Goodman et al . , 2013 ) , they exclude ranges where the cue is most unreliable . It has been suggested that the owl's brain represents the probability distribution of features in natural scenes ( Fischer and Pena , 2011 ) . Consistent with this idea is that the oblique effect in humans can be explained by a neural representation of the visual scene where vertical and horizontal orientations are more likely ( Girshick et al . , 2011 ) . Our study further strengthens the idea that the brain represents the likelihood of natural features by showing that it synthesizes an internal model of stimulus reliability . The surgeries were performed as described previously ( Wang et al . , 2012 ) . Briefly , two female adult barn owls were anesthetized with IM injections of ketamine hydrochloride ( 20 mg/kg; Ketaset ) and xylazine ( 4 mg/kg; Anased ) . It has been shown that the responses of midbrain neurons are remarkably stable under ketamine anesthesia ( Ter-Mikaelian et al . , 2007; Schumacher et al . , 2011 ) . These procedures complied with National Institutes of Health and the Albert Einstein College of Medicine's Institute of Animal Studies guidelines . Responses were recorded with 1 MΩ tungsten electrodes ( A–M Systems , Sequim , WA ) . Tucker Davis Technologies System 3 ( Alachua , FL ) and Matlab software were used to record neural data . ICx and ICCc neurons were identified by well-established physiological criteria based on their tuning to ITD , interaural level difference ( ILD ) and frequency , which permit unequivocal identification of recording sites ( Singheiser et al . , 2012 ) . All experiments were performed inside a double-walled sound-attenuating chamber ( Industrial Acoustics 120a , Bronx , NY ) lined with echo-absorbing acoustical foam ( Sonex 4″ wedge , Minneapolis , MN ) . These are rated to absorb sounds with highest efficiency down to below 1 kHz . Auditory stimuli delivered through calibrated earphones , consisting of a speaker ( Knowles model 1914 , Itasca , IL ) and a microphone ( Knowles model EK-23024 ) housed in a cylindrical metal earpiece and inserted in the owl's ear canal , consisted of 100 ms signals with a 5 ms rise-fall time at 10–20 dB above threshold . ITD tuning was initially estimated with broadband noise ( 0 . 5–10 kHz ) . ITD varied between ±300 μs in 30 µs steps over five trials . Frequency tuning was estimated with tones at the best ITD ranging between 600 Hz to 9000 kHz in 200 Hz steps , repeated over 15 to 20 trials . ITD tuning within the main peak of the rate-ITD curve was recorded at a finer resolution ( 10 µs steps; 20 repetitions ) at the best ILD . The best ILD was determined as the ILD at the peak of the rate-ILD curve . ITD tuning to tonal stimulation was also measured . The range of ITD and sampling steps was adjusted by the stimulus frequency such that roughly three periods of the stimulating frequency were tested and 21 or more different ITDs were sampled . Each trial was repeated 20 times . Stimuli within all tested ranges were randomized during data collection . To measure the effect of concurrent sounds on the spatial tuning of neurons in ICCc we used an array of 21 calibrated speakers placed inside a sound-attenuating chamber ( Wang et al . , 2012 ) . The speaker array covered a range of ±100° in azimuth with an angular separation between speakers of 10° . To measure the spatial tuning with concurrent sound sources , we simultaneously played a broadband noise ( 0 . 5–10 kHz ) at a given location while randomly activating other speakers of the array with another broadband noise ( within ± 90° from the preferred location in 20° steps ) . Dichotic and free-field recordings were performed in the same animals without disrupting the facial ruff . Recordings were performed at 138 ICx sites . Wave_clus was used for spike sorting ( Quiroga et al . , 2004 ) . Briefly , spikes were detected using a voltage threshold set at five times the estimated standard deviation of the signal . To avoid double detection , spikes were separated by at least 1 ms . Neurons were considered isolated based on the presence of a refractory period of more than 1 ms in the inter-spike interval histogram and the similarity of spike shape . A complementary quality metric was the non-overlap of wavelet coefficients . Additionally , the results of the sorting algorithm were visually inspected to confirm the quality of the sorting . Consistent with previous reports ( Winkowski and Knudsen , 2006 ) , no significant differences were found between the results of sorted and non-sorted traces as neighboring ICx neurons have very similar tunings . For each stimulus parameter , a rate curve was computed by averaging the firing rate across stimulus repetitions and subtracting the spontaneous rate . The best ITD was the ITD corresponding to the center of the main peak in the tuning curve . ITD tuning curves in ICx typically have a main peak and several smaller side peaks . We identified the main peak in the ITD tuning curve , then measured the ITD range where the firing rate was at least half the difference between the minimum and the maximum response . The best ITD was the center of this range of ITDs . We used the absolute value of the best ITD to combine data from contra- and ipsi-lateral sides as a function of the eccentricity of the receptive field . For assessment of frequency tuning , the response area was defined as the frequency range that elicited more than 50% of the maximum response . The lower and upper edges of the frequency tuning curve were respectively the lowest and the highest frequencies of this range . Frequency tuning curves in ICx are broad and often lack an unequivocal peak . We thus defined the BF as the frequency at the center between the lower and upper edges . We also calculated the BF with a threshold at 30% and 75% of the maximum response or as the center of mass of the frequency tuning curve . No significant differences were found compared to the BF estimated with a threshold at 50% of the maximum response . The gain ( in dB SPL ) was computed as the average of the left and right monaural gains . The monaural gains at each frequency represented the relative attenuation of the sound level by the HRTFs at each ear . To compare the lower- and upper- edges of the frequency tunings with IPD reliability and gain , we computed the lower- and upper edges of the IPD reliability and gain . The lower- and upper- edges of the IPD reliability and gain were the lowest and the highest frequencies of the range that elicited more than 50% of the maximum IPD reliability and gain , respectively . To quantify the neural variability across frequency , we calculated the Fano factors across trials in ITD curves with tonal stimulation . The Fano factor of an ITD curve is the ratio of the variance to the mean of the spike count . We measured the strength of IPD tuning using circular statistics ( Goldberg and Brown , 1969 ) . The range of ITD used for this analysis was adjusted by the stimulus frequency , such that the number of data points relative to the period of the stimulus frequency was constant . We converted ITD to IPD by folding ITD curves from tonal stimulation into a single period of the stimulating frequency . IPD curves were fitted by a Gaussian function to achieve a uniform sampling of IPD over one period ( Perez and Pena , 2006 ) . The average response at each IPD was treated as a vector and the goodness of IPD tuning was estimated by a synchronization coefficient ( Goldberg and Brown , 1969; Kuwada et al . , 1987 ) . The synchronization coefficient varies from 0 to 1 , indicating no selectivity to IPD or perfect phase synchrony , respectively . We reported the p-value and r² of the linear regression for estimating the relationships between variables . Pearson correlation coefficients were used to compare the smoothed frequency tuning curves with the estimated frequency from the IPD reliability and gain . The Pearson correlation coefficients of the spatial tuning curves obtained with single sources and those measured with concurrent sounds were averaged to quantify the overall effect of concurrent sounds in altering the spatial tuning of ICCc cells . The HRTFs of ten barn owls were provided by Dr Keller ( Keller et al . , 1998 ) . The gain of the loudspeaker used by Keller et al . , 1998 decreased sharply below about 2 kHz . In the present study , the HRTFs were reprocessed by Dr Keller , equalizing the gain of the loudspeaker down to 1 kHz . As in the original paper , in order to remove the effects of the loudspeaker , an inverse filter for the loudspeaker was constructed and convolved with the head-related impulse responses . The new inverse filter , now optimized to include frequencies down to 1 kHz , removes most of the effects of the loudspeaker . To align HRTFs across owls , IPD variability was computed at an elevation centered between the acoustic axes of left and right ear-canals ( the locations of maximum intensity gain of the HRTFs for left and right ears ) . This accounts for slight differences in head placement of each owl that might lead to differences in the definition of zero elevation across owls . We calculated the left and right acoustic axes by taking the median gain across the different frequencies and determined the center of gravity of the top 90% elevations . To compute IPD variability for concurrent sound sources , broadband noise signals with flat spectra between 0 . 5 and 9 kHz and equal gain were each convolved with head-related impulse-responses at the appropriate source location . The outputs were passed through a gammatone filter-bank with center frequencies ranging from 1 kHz to 8 kHz in 0 . 2 kHz steps . The time constants of the filters were specific to the owl and estimated from Koppl ( 1997a ) to model cochlear filters , as described in Fischer et al . ( 2009 ) . Because variability was estimated within narrow frequency bands , we used IPD , rather than ITD , by converting time into phase . The IPD in narrow frequency channels was calculated as the phase delay with the highest value of the cross-correlation between the left and right outputs of the gammatone filter . For each target location , we obtained 37 estimates of IPD , each with a second source located at one of the locations covering the frontal hemisphere ( between −90 and 90° in steps of 5° at elevation zero ) . The circular standard deviation of IPD over locations of the second sound source was used as the estimate of IPD variability at each location . To test whether reliability weighting allows ICx neurons to integrate coherent inputs in each frequency channel , we compared the average interaural correlation across locations of concurrent sounds with the IPD reliability . Concurrent broadband noise signals with flat spectra between 0 . 5 and 9 kHz and equal gain were convolved with head-related impulse-responses at the appropriate source locations . The outputs were passed through a gammatone filter-bank with center frequencies ranging from 1 kHz to 8 kHz in 0 . 2 kHz steps as described above . The cross-correlation was computed within each frequency channel . For a single sound source the interaural correlation was close to 1 and decreased when a concurrent sound was added . The interaural correlations with concurrent sounds were averaged across different locations of concurrent sounds .	The ability to locate where a sound is coming from is an essential survival skill for both prey and predator species . A major cue used by the brain to infer the sound's location is the difference in arrival time of the sound at the left and right ears; for example , a sound coming from the left side will reach the left ear before the right ear . We are exposed to a variety of sounds of different intensities ( loud or soft ) , and pitch ( high or low ) emitted from many different directions . The cacophony that surrounds us makes it a challenge to detect where individual sounds come from because other sounds from different directions corrupt the signals coming from the target . This background noise can profoundly affect the reliability of the sensory cue . When sounds reach the ears , the head and external ears transform the sound in a direction-dependent manner so that some pitches are amplified more than other pitches for specific directions . However , the consequence of this filtering is that the directional information about a sound may be altered . For example , if two sounds of a similar pitch but from different locations are heard at the same time , they will add up at the ears and change the directional information . The group of neurons that respond to that range of pitches will be activated by both sounds so they cannot provide reliable information about the direction of the individual sounds . The degree to which the directional information is altered depends on the pitch that is being detected by the neurons; therefore detection of a different pitch within the sound may be a more reliable cue . Cazettes et al . used the known filtering properties of the owl's head to predict the reliability of the timing cue for sounds coming from different directions in a noisy environment . This analysis showed that for each direction , there was a range of pitches that carried the most reliable cues . The study then focused on whether the neurons that represent hearing space in the owl's brain were sensitive to this range . The experiments found a remarkable correlation between the pitch preferred by each neuron and the range that carried the most reliable cue for each direction . This finding challenges the common view of sensory neurons as simple processors by showing that they are also selective to high-order properties relating to the reliability of the cue . Besides selecting the cues that are likely to be the most reliable , the brain must capture changes in the reliability of the sensory cues . In addition , this reliability must be incorporated into the information carried by neurons and used when deciding how best to act in uncertain situations . Future research will be required to unravel how the brain does this .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Spatial cue reliability drives frequency tuning in the barn Owl's midbrain
Regulation of animal development in response to nutritional cues is an intensely studied problem related to disease and aging . While extensive studies indicated roles of the Target of Rapamycin ( TOR ) in sensing certain nutrients for controlling growth and metabolism , the roles of fatty acids and lipids in TOR-involved nutrient/food responses are obscure . Caenorhabditis elegans halts postembryonic growth and development shortly after hatching in response to monomethyl branched-chain fatty acid ( mmBCFA ) deficiency . Here , we report that an mmBCFA-derived sphingolipid , d17iso-glucosylceramide , is a critical metabolite in regulating growth and development . Further analysis indicated that this lipid function is mediated by TORC1 and antagonized by the NPRL-2/3 complex in the intestine . Strikingly , the essential lipid function is bypassed by activating TORC1 or inhibiting NPRL-2/3 . Our findings uncover a novel lipid-TORC1 signaling pathway that coordinates nutrient and metabolic status with growth and development , advancing our understanding of the physiological roles of mmBCFAs , ceramides , and TOR . Regulation of animal growth and development in response to nutritional cues is an intensely studied problem ( Hietakangas and Cohen , 2009; Zoncu et al . , 2011 ) . In animals , nutrient signals are perceived in specialized tissues and are then communicated to all other tissues to coordinate growth and development . The target of rapamycin ( TOR ) complexes ( TORC1 and TORC2 ) are known to function in sensing various nutrient signals ( Ma and Blenis , 2009; Laplante and Sabatini , 2012; Zoncu et al . , 2012 ) , and their roles are connected to growth , metabolism , stress responses , and cancers ( Hansen et al . , 2008; He and Klionsky , 2009; Howell and Manning , 2011 ) . While amino acids , energy , and growth factors have been described as nutrient inputs to TOR complexes , roles of lipid molecules as signals to these systems in controlling postembryonic growth and development were not known . Clearly , it is also essential to use whole-animal models to investigate how different signaling systems in different tissues interact to specify decisions regarding postembryonic development and behaviors . In Caenorhabditis elegans , the first larval stage has been established as a model system for the study of animal growth and development in response to food availability . When hatched in food-free surroundings , this nematode enters a quiescent state , termed L1 diapause ( Johnson et al . , 1984; Munoz and Riddle , 2003; Baugh and Sternberg , 2006 ) . An insulin/IGF-1 receptor signaling ( IIS ) pathway plays a critical role in the induction of L1 diapause and survival of the developmentally arrested animals ( Gems et al . , 1998; Baugh and Sternberg , 2006; Kniazeva et al . , 2008; Lee and Ashrafi , 2008; Jones et al . , 2009; Soukas et al . , 2009 ) . Moreover , TOR complexes have also been indicated to play prominent roles in regulating postembryonic growth and lifespan in C . elegans ( Vellai et al . , 2003; Syntichaki et al . , 2007; Hansen et al . , 2008; Honjoh et al . , 2009; Lucanic et al . , 2011 ) . Monomethyl branched-chain fatty acids ( mmBCFAs ) are widely present in bacteria , plants , and animals , including humans ( Nicolaides and Ray , 1965; Ran-Ressler et al . , 2008 ) . In mammals , mmBCFAs are derived from branched-chain amino acids ( BCAAs ) ( Morii and Kaneda , 1982; Oku et al . , 1994 ) , although the remainder of the de novo pathway has not been delineated . The physiological roles of these FA variants are essentially unknown , even though they were found to be present at very high levels in certain tissues ( Nicolaides and Ray , 1965; Ran-Ressler et al . , 2008 ) . In C . elegans , readily detectable mmBCFAs C15ISO and C17ISO are synthesized from the BCAA leucine ( Kniazeva et al . , 2004 ) . The key enzymes in this de novo synthesis pathway , including the branched-chain ketoacid dehydrogenase complex ( BCKDC ) , an FA elongase ( ELO-5 ) , and an acyl CoA synthetase ( ACS-1 ) , are evolutionarily conserved ( Kniazeva et al . , 2004 , 2008 ) . We have previously shown that newly hatched C . elegans that are deficient for mmBCFAs cannot initiate postembryonic growth and development , and instead enter L1 diapause . Further genetic analysis suggested that this developmental arrest is independent of the IIS pathway ( Kniazeva et al . , 2004 , 2008 ) . It was not clear whether the essential roles of mmBCFAs and their derived lipids were due to structural requirements for animal development , as was suggested by other studies , or due to regulatory functions specific to cellular signaling processes . Testing these hypotheses using model organisms is highly significant because roles of fatty acids and lipids function as nutrient signals for postembryonic development are in general poorly understood . In this study , we discovered that ( 1 ) an mmBCFA-derived glucosylceramide ( d17iso-GlcCer ) mediates the function of mmBCFAs in promoting postembryonic growth and development and ( 2 ) d17iso-GlcCer acts through a signaling system that includes the NPRL-2/NPRL-3 protein complex ( negative factor ) and TORC1 ( positive factor ) to promote postembryonic development . elo-5 loss-of-function mutants ( termed elo-5 ( − ) hereafter ) are deficient for mmBCFAs and are developmentally arrested at the early L1 stage . This phenotype is completely rescued by dietary supplementation of mmBCFAs ( Kniazeva et al . , 2004 , 2008 ) . We found that in mmBCFA-deficient elo-5 ( − ) larvae , the amount of mmBCFA-containing sphingolipids were dramatically reduced and thus asked if this mmBCFA function is mediated by a sphingolipid . Sphingolipids are a class of well-known bioactive lipids ( Hannun and Obeid , 2008 ) that are composed of an aliphatic amino alcohol backbone called the sphingoid base or long chain base ( LCB ) , and usually an N-acylated fatty acid ( FA ) side chain ( Figure 1A ) . In C . elegans , both the LCB and the FA side chain may derive from mmBCFAs , such as C15ISO and C17ISO ( Chitwood et al . , 1995; Gerdt et al . , 1997; Figure 1A ) . In this study , a sphingolipid with a mmBCFA-derived LCB is termed d17iso-sphingolipid , whereas a sphingolipid with a C17ISO-derived FA side chain and a straight LCB is termed c17iso-sphingolipid . The ‘d’ and ‘c’ refer to carbon atoms on the LCB and FA chain , respectively . 10 . 7554/eLife . 00429 . 003Figure 1 . Iso-branched d17iso-sphinganine rescues elo-5 ( - ) L1 arrest . ( A ) Sphingolipid biogenesis pathway in C . elegans , including the catalytic enzymes ( blue ) and corresponding genes ( red ) used in this study . Molecular structures in green and in light blue indicate the long chain base and side chain fatty acid , respectively . ( B ) Summary of growth phenotype of various mutants with indicated lipid supplementations . elo-5 ( gk208 ) , cgt-1 ( tm1027 ) , and cgt-3 ( tm504 ) were the alleles of the ( − ) mutants . LCB: long chain base; SPA: sphinganine . See Supplementary file 1 for numerical data or more detailed description of the phenotypes . ( C ) Mass spectrometry of iso-branched d17iso-sphinganine ( d17iso-SPA ) isolated from the bacteria S . spiritivorum . The major peak ( m/z = 288 . 4 ) corresponds to d17iso-SPA in each panel . The lower panel shows fragmentation of d17iso-SPA by MS-MS scan . The fragment peaks are labeled with the name of lost residues . ( D ) d17iso-SPA rescues elo-5 ( − ) animals more efficiently than C15ISO and C17ISO . Error bar , SD . ( E ) Gas chromatography of methyl-esterified fatty acid extracts from rescued elo-5 ( − ) animals fed with d17iso-SPA . Low concentration of d17iso-SPA ( second and third panels ) supplement did not restore C15iso or C17iso fatty acid levels . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 003 We then examined if dietary supplementation with either of these two types of sphingolipids could substitute for mmBCFAs to rescue elo-5 ( − ) animals from L1 arrest . Because the d17iso-LCB was not commercially available , we purified d17iso-sphinganine ( d17iso-SPA; Figure 1A ) from a d17iso-sphingolipid producing bacteria S . spiritivorum ( Yabuuchi et al . , 1983 ) using mass spectrometry analysis . This purified d17iso-SPA fraction , but not a chemically synthesized c17iso-sphingolipid with a straight-chain d18:1-LCB or a straight-chain LCB d16-SPA , was found to be sufficient to overcome elo-5 ( − ) L1 arrest ( Figure 1B , C; Supplementary file 1 ) . Additionally , efficient rescue was also achieved with a custom synthesized d17iso-SPA ( Figure 1D ) . Therefore , d17iso-sphingolipid is sufficient to promote the postembryonic growth and development of mmBCFA-deficient animals . We then carried out further tests to confirm that this rescue effect of d17iso-SPA supplementation resulted from the function of this metabolite itself , but not from its catabolic conversion back to C15ISO or C17ISO . First , we fed elo-5 ( − ) larvae with various quantities of C15ISO , C17ISO , or d17iso-SPA . We observed that d17iso-SPA suppressed the L1 arrest at a much lower concentration than C15ISO or C17ISO ( Figure 1D ) . Second , we observed that the low concentrations of d17iso-SPA suppressed the L1 arrest without restoring mmBCFA levels ( Figure 1E ) . Finally , we found that supplementation with d17iso-SPA , but not mmBCFAs , also suppressed the L1 arrest phenotype caused by feeding RNAi of sptl-1 , a gene encoding a serine palmitoyltransferase homolog in C . elegans required for d17iso-SPA biosynthesis ( Figure 1A , B; Seamen et al . , 2009 ) . We thus conclude that mmBCFAs promote postembryonic development through d17iso-SPA . This growth-promoting function of d17iso-SPA may be mediated by a more complex sphingolipid derived from it . Because it is technically difficult to purify or synthesize sphingolipids of more complex structure ( e . g . , glucosylceramide [GlcCer] ) and directly test their effect by dietary supplementation , we examined additional enzymes involved in sphingolipid synthesis . Specifically , fath-1 ( C25A1 . 5 ) encodes a homolog of mammalian fatty acid 2-hydrolase ( FA2H ) responsible for the synthesis of the FA side chain of ceramides , whereas cgt-1 and cgt-3 encode two ceramide glucosyltransferases that glucosylate ceramides to GlcCer ( Figure 1A ) . Both fath-1 ( RNAi ) -treated animals ( Figure 1B; Supplementary file 1 ) and the cgt-1 ( − ) ;cgt-3 ( − ) double mutant ( Marza et al . , 2009 ) cause larval arrest . We found the arrest under either condition could not be overcome by feeding d17iso-SPA ( Figure 1B; Supplementary file 1 ) . Disrupting the function of glycosyltransferases that further modify GlcCer does not cause larval arrest ( Griffitts et al . , 2003 ) . These results suggest that d17iso-glucosylceramide ( d17iso-GlcCer ) ( Figure 1A ) may be the lipid molecule that mediates the role of mmBCFAs and d17iso-SPA in promoting postembryonic growth and development in C . elegans . d17iso-sphingolipid may play either a structural or regulatory role for the initiation of postembryonic development . If it is the latter , the requirement of d17iso-sphingolipid for postembryonic development might be bypassed by genetically activating the downstream regulatory pathway . To test this possibility and identify the mechanism , we performed a genetic screen , modified from a previous screen ( Seamen et al . , 2009 ) , for mutations that could suppress the L1 arrest of elo-5 ( − ) mutants ( Figure 2A ) . Among four suppressors identified in this screen , and the tat-2 mutations from a previous elo-5 ( − ) suppressor screen ( Seamen et al . , 2009 ) , ku540 was the only mutation that rescued elo-5 ( − ) larvae to adults and permitted growth for more than one generation without mmBCFA supplementation or restoration of mmBCFA synthesis ( Figure 2B–E; see ‘Materials and methods’ ) . In fact , elo-5 ( − ) ;ku540 homozygous animals could propagate continuously , even though they displayed slow growth , smaller body size , and smaller brood size ( Figure 2C ) . 10 . 7554/eLife . 00429 . 004Figure 2 . ku540 mutant suppresses L1 arrest of mmBCFA and sphingolipid biosynthetic mutants without restoring the levels of mmBCFAs . ( A ) Screening strategy to isolate elo-5 ( − ) suppressors . Green-colored C . elegans carry the extrachromosomal array containing copies of the elo-5 ( + ) , sur-5-gfp , and rol-6 ( dn ) genes . ( B ) Percentages of animals reaching adulthood ( bypass L1 arrest ) when fed with/without C17ISO supplement . elo-5 ( − ) ;ku540 mutant animals reach adulthood without C17ISO supplement . Error bar , SD . ( C ) Images showing that elo-5 ( − ) animals arrested at L1 and elo-5 ( − ) ;ku540 animals reached adulthood . ( D ) Mass spectrometry by precursor scan m/z = −241 . 2 showing no detectable C15FA-containing lipids in elo-5 ( − ) ;ku540 animals . Numeric data for the levels of PA and PE are shown in Figure 2—figure supplement 1A . ( E ) Bar graph of relative FA composition by gas chromatography ( GC ) of methyl-esterified fatty acid extracts . These data indicate no detectable C15ISO or C17ISO was restored in elo-5 ( − ) ku540 animals . GC graph is shown in Figure 2—figure supplement 1B . ( F ) and ( G ) Mass spectrometry of d17iso-ceramide–containing lipids by precursor scan m/z = +250 . 3 , and d16-ceramide–containing lipids by precursor scan m/z = +236 . 3 ( F ) . d17iso-ceramides and d17iso-glucosylceramides are detectable in wild-type but not elo-5 ( − ) ;ku540 animals ( G ) . In contrast , d16-ceramides and d16-glucosylceramides are present in elo-5 ( − ) ;ku540 , but not wild-type , animals . Numeric data for relevant peaks are shown in Figure 2—figure supplement 1C , D . ( H ) and ( I ) Percentages of animals of the indicated genotypes and treatment that reached adulthood . Error bar , SD . ( H ) The addition of ku540 dramatically suppressed the L1 arrest phenotype of cgt-1 ( − ) ;cgt-3 ( RNAi ) with and without the C17ISO supplement . The inclusion of elo-5 ( − ) was due to its close linkage with ku540 . When C17ISO was added to remove the negative effect of elo-5 ( − ) , about 90% of elo-5 ( − ) ;ku540; cgt-1 ( − ) ;cgt-3 ( RNAi ) animals reached adulthood . C17ISO itself does not rescue the phenotype . ( I ) . ku540 suppressed the L1 arrest phenotype the cgt-1 ( − ) ;cgt-3 ( − ) double mutants ( 83 . 4% , n = 126 ) . In this particular test , elo-5 ( − ) ku540;cgt-1 ( − ) ;cgt-3 ( − ) homozygous animals were the progeny of elo-5 ( −/+ ) ku540 ( −/+ ) ;cgt-1 ( −/+ ) ;cgt-3 ( − ) heterozygous mothers , and the presented data were obtained after normalizing against heterozygous populations ( see ‘Materials and methods’ ) . In all the other tests shown in Figure 2 , elo-5 ( − ) ku540 homozygous animals were the progeny of homozygous mothers . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 00410 . 7554/eLife . 00429 . 005Figure 2—figure supplement 1 . Quantification of GC and mass spectrometry analyses of elo-5 ( − ) ku540 animals . ( A ) Tables show relative intensities of the five strongest signals by mass spectrometry precursor scan ( m/z = −241 . 2 ) in wild-type animals ( second right column ) and their normalized relative intensities in elo-5 ( − ) ku540 animals ( right column ) . The low intensities of these lipids in elo-5 ( − ) ku540 animals indicate that the C15ISO mmBCFA level was not restored elo-5 ( − ) ku540 animals . ( B ) Gas chromatography of methyl-esterified fatty acid extracts showing no detectable C15ISO or C17ISO in elo-5 ( − ) ku540 animals . ( C ) Table shows relative intensities of normal d17iso-sphingolipid by mass spectrometry precursor scan ( m/z = +250 . 3 ) in wild-type animals ( second right column ) and their normalized relative intensities in elo-5 ( − ) ku540 animals ( right column ) . The low intensities of these lipids in elo-5 ( − ) ku540 animals indicate that the d17iso-sphingolipid level was not restored elo-5 ( − ) ku540 animals . ( D ) Table shows relative intensities of abnormal d16-sphingolipid by mass spectrometry precursor scan ( m/z = +238 . 3 ) in elo-5 ( − ) ku540 animals ( second right column ) . These sphingolipids are not present in wild-type animals ( right column ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 00510 . 7554/eLife . 00429 . 006Figure 2—figure supplement 2 . Suppression of fath-1 ( − ) by nprl-3 ( ku540 ) . Microscopic images showing that nprl-3 ( ku540 ) suppresses the L1 arrest phenotype of fath-1 ( injection RNAi ) with or without C17ISO supplementation . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 006 Mass spectrometry ( MS ) and gas chromatography ( GC ) analyses indicated that elo-5 ( − ) ;ku540 did not restore mmBCFA levels , and consequently the levels of d17iso-sphingolipids ( Figure 2D–G ) . elo-5 ( − ) ;ku540 animals had no detectable C15ISO-containing or C17ISO-containing lipids ( Figure 2D , E , Figure 2—figure supplement 1A , B ) . In contrast to wild type animals , elo-5 ( - ) ;ku540 animals had ceramides and glucosylceramides containing straight chain d16LCB rather than d17isoLCB ( Figure 2F , G , Figure 2—figure supplement 1C , D ) , which is similar to the elo-5 ( − ) single mutant ( Entchev et al . , 2008 ) . These data indicated that the suppression of L1 arrest by the ku540 mutation is neither due to a change in mmBCFA nor the metabolism of the derived sphingolipids . Therefore , our data suggest that ku540 renders mmBCFAs nonessential for C . elegans development . The above results suggested that ku540 might modify a signaling pathway downstream of d17iso-sphingolipids . If so , ku540 should also suppress mutations disrupting the biosynthesis of this sphingolipid . This question was addressed by the following tests . As shown above , RNAi of fath-1 , or mutating both cgt-1 and cgt-3 , caused larval arrest that could not be rescued by feeding with either mmBCFAs or d17iso-SPA ( Figure 1A , B ) . We found that the L1 arrest under both conditions , just like the arrest caused by elo-5 ( − ) , was effectively suppressed by ku540 ( Figure 2H , I and Figure 2—figure supplement 2 ) . These results provided strong evidence that d17iso-GlcCer mediated the roles of mmBCFAs and d17iso-SPA in postembryonic development . Taken together , our biochemical and genetic data indicate that ku540 bypasses the requirement of mmBCFAs and d17iso-GlcCer in development and does so by modifying a regulatory function downstream of d17iso-GlcCer . We cloned the gene containing the ku540 mutation by genetic mapping and whole-genome sequencing ( Figure 3—figure supplement 1A ) . The gene was annotated as F35H10 . 7 ( wormbase . org ) that encodes a protein structurally conserved from budding yeast ( NPR3 , Nitrogen Permease Regulator 3 ) to humans ( NPRL-3 , Nitrogen Permease Regulator Like 3; Figure 3B ) . We named the gene nprl-3 ( NPRL-3 for protein ) . nprl-3 ( ku540 ) is a C-T substitution that changes a conserved proline to serine ( Figure 3A ) . RNAi of nprl-3 by dsRNA injection also rescued the L1 arrest of elo-5 ( − ) and cgt-1 ( − ) ;cgt-3 ( − ) animals ( Figure 3C ) , indicating that ku540 is likely a partial loss-of-function or reduction-of-function mutation . A transcriptional GFP fusion transgene ( nprl-3::GFP ) was observed to express ubiquitously throughout development and adulthood ( Figure 3—figure supplement 1B ) . 10 . 7554/eLife . 00429 . 007Figure 3 . ku540 is a loss-of-function missense mutation of nprl-3 . ( A ) Predicted structure and position of the ku540 mutation in the nprl-3 gene ( F35H10 . 7 ) . ( B ) Abbreviated alignment of C . elegans NPRL-3 with its orthologs in other organisms . ( C ) C . elegans images showing nprl-3 ( RNAi ) mimics the effect of the ku540 mutation in rescuing the L1 arrest phenotype caused by blocking mmBCFA or glucosyl-ceramide biosynthesis . nprl-3 dsRNA injection rescued 61 . 0% of elo-5 ( − ) ( n = 123 ) and 6 . 8% of cgt-1 ( − ) cgt-3 ( − ) ( n = 71 ) animals to or beyond L3 stage . nprl-2 dsRNA injection also rescued 42 . 4% of elo-5 ( − ) animals to or beyond L3 stage ( n = 128 ) . All data have been normalized to heterozygous populations ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 00710 . 7554/eLife . 00429 . 008Figure 3—figure supplement 1 . Mapping and expression of nprl-3 . ( A ) A simplified diagram of the mapping process . ku540 is genetically linked to the E03H12 SNP marker on chromosome IV ( upper section ) . Further three-point mapping narrowed the ku540 locus to near unc-24 , by visible markers ( red ) and by physical mutations ( blue ) found by genomic deep sequencing ( middle section ) . The three candidate gene mutations resulting in amino acid changes are shown ( lower section ) . ( B ) DIC and fluorescence images showing two representative animals carrying a nprl-3 promoter::GFP transgene . The head is indicated by a black arrow . GFP is visible in most tissues with stronger expression seen in the head and tail regions . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 008 In Saccharomyces cerevisiae , NPR3 was shown to form a heterodimer with NPR2 . Mutations in both proteins cause multiple growth defects when cells were cultured in low-quality amino acid conditions ( Neklesa and Davis , 2009 ) . We found that RNAi by dsRNA injection of the C . elegans NPR2 homolog nprl-2 ( F49E8 . 1 ) also suppressed the developmental arrest of elo-5 ( − ) larvae ( Figure 3C ) . This suggested that NPRL-2 and NPRL-3 in C . elegans may also function in a complex to negatively impact mmBCFA/d17iso-GlcCer-mediated L1 growth . The NPR2/3 complex was proposed to be a negative regulator of the TOR pathway in S . cerevisiae , based on the observations that mutations in these two genes caused growth defects in an amino acid scarce environment and that the defects could be effectively suppressed by blocking the TOR pathway ( Neklesa and Davis , 2009 ) . We thus explored the possibility that TOR Complex 1 ( TORC1 ) is also the downstream target of NPRL2/3 in growth regulation in C . elegans and that nprl-3 ( ku540 ) alleviates d17iso-GlcCer deficiency-induced growth arrest by activating the TORC1 pathway . We reasoned that if reducing nprl-2/3 activity rescued elo-5 ( − ) L1 growth by up-regulating the TORC1 pathway in C . elegans , then reducing , but not eliminating , TORC1 activity may reverse the rescue effects of nprl-3 ( ku540 ) on elo-5 ( − ) mutant background . Caenorhabditis elegans has orthologs of key components of mammalian TORC1 , including regulatory elements raga-1 ( RagA ) and rheb-1 ( Rheb ) ( Long et al . , 2002; Schreiber et al . , 2010 ) . Feeding RNAi of raga-1 and rheb-1 have been shown to be specific and effective to reduce , but not eliminate , TORC1 function ( Hansen et al . , 2007; Syntichaki et al . , 2007; Lemire et al . , 2009; Figure 4—figure supplement 1; see ‘Materials and methods’ ) . We thus tested the effect of raga-1 ( RNAi ) and rheb-1 ( RNAi ) on elo-5 ( − ) nprl-3 ( ku540 ) mutants . While nprl-3 ( ku540 ) effectively suppressed the L1 arrest phenotype of elo-5 ( − ) , the suppression dramatically decreased when raga-1 ( RNAi ) or rheb-1 ( RNAi ) were applied to the elo-5 ( − ) nprl-3 ( ku540 ) double mutant ( Table 1 ) . This indicates that the rescue effect of nprl-3 ( ku540 ) on elo-5 ( − ) depends on intact TORC1 activity . 10 . 7554/eLife . 00429 . 009Table 1 . Intact TORC1 function is necessary for mmBCFA-mediated growth regulationDOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 009GenotypeRNAiDietary C17ISONormalized % of F1 reached adulthoodNpelo-5 ( − ) Vector−0208elo-5 ( − ) Vector+86 . 92050elo-5 ( − ) nprl-3 ( − ) Vector−71 . 569elo-5 ( − ) nprl-3 ( − ) Vector+111 . 51030 . 24elo-5 ( − ) nprl-3 ( − ) raga-1 ( a ) −35 . 556elo-5 ( − ) nprl-3 ( − ) raga-1 ( a ) +102 . 5830 . 031elo-5 ( − ) nprl-3 ( − ) raga-1 ( b ) −25 . 5214elo-5 ( − ) nprl-3 ( − ) raga-1 ( b ) +88 . 02790 . 00001elo-5 ( − ) nprl-3 ( − ) rheb-1 ( a ) −14 . 2351elo-5 ( − ) nprl-3 ( − ) rheb-1 ( a ) +64 . 65260 . 0000001elo-5 ( − ) nprl-3 ( − ) rheb-1 ( b ) −32 . 5201elo-5 ( − ) nprl-3 ( − ) rheb-1 ( b ) +86 . 01160 . 0037elo-5 ( − ) nprl-3 ( − ) rsks-1 ( a ) −15 . 068elo-5 ( − ) nprl-3 ( − ) rsks-1 ( a ) +72 . 5830 . 022elo-5 ( − ) nprl-3 ( − ) rsks-1 ( b ) −19 . 0183elo-5 ( − ) nprl-3 ( − ) rsks-1 ( b ) +61 . 01800 . 0034elo-5 ( − ) nprl-3 ( − ) ife-2 ( a ) −10 . 5144elo-5 ( − ) nprl-3 ( − ) ife-2 ( a ) +71 . 5490 . 003elo-5 ( − ) nprl-3 ( − ) ife-2 ( b ) −0103elo-5 ( − ) nprl-3 ( − ) ife-2 ( b ) +103 . 51500elo-5 ( − ) nprl-3 ( − ) let-363−0>100elo-5 ( − ) nprl-3 ( − ) let-363+0>100NAPercentages of elo-5 ( − ) ;nprl-3 ( ku540 ) homozygotes with the indicated RNAi treatments that reached adulthood , where ( a ) and ( b ) indicate two different RNAi constructs targeting different parts of the same gene . The presented percentage of elo-5 ( − ) nprl-3 ( ku540 ) animals that reached adulthood was calculated by normalizing against the percentage of elo-5 ( −/+ ) nprl-3 ( ku540 ) /+ heterozygotes ( see ‘Materials and methods’ for detail ) . Without C17ISO supplementation , RNAi knockdown of multiple TORC1 components reverted elo-5 ( − ) ;nprl-3 ( ku540 ) animals to larval arrest . We carried out further analyses by repeating the above tests in the presence of dietary C17ISO that rescues the defects of elo-5 ( − ) . We found that the effects of raga-1 ( RNAi ) and rheb-1 ( RNAi ) in the above tests were essentially eliminated by C17ISO supplementation ( Table 1 ) . This result further indicates that the reversal of the nprl-3 ( ku540 ) suppression by raga-1 ( RNAi ) or rheb-1 ( RNAi ) depends on mmBCFA deficiency . In other words , reducing , but not eliminating , TORC1 activity neutralized the effect of nprl-3 ( ku540 ) and restored the L1 arrest phenotype of elo-5 ( − ) . To further examine whether the rescuing effect of nprl-3 ( ku540 ) on elo-5 ( − ) L1 arrest required the canonical activity of TORC1 , we used RNAi to knock down C . elegans orthologs of elf4E ( ife-2 ) and p70S6K ( rsks-1 ) , two well-known downstream targets of TORC1 ( Long et al . , 2002; Syntichaki et al . , 2007; Anjum and Blenis , 2008; Sheaffer et al . , 2008 ) . We found that elo-5 ( − ) nprl-3 ( ku540 ) animals treated with ife-2 ( RNAi ) or rsks-1 ( RNAi ) , with or without C17ISO supplement , yielded results similar to those obtained with raga-1 ( RNAi ) and rheb-1 ( RNAi ) ( Table 1 ) . Taken together , these results indicate that TORC1 acts downstream of d17iso-GlcCer and is negatively regulated by NPRL-3 . The let-363/TOR ( − ) mutations cause late larval arrest and other severe defects ( Long et al . , 2002 ) , but injection RNAi causes an early embryonic lethal phenotype ( Sonnichsen et al . , 2005 ) , indicating that LET-363/TOR , a key component of both TORC1 and TORC2 , plays critical regulatory roles in a broad range of developmental stages and a maternal effect largely masks its roles in earlier stages , including its likely functions during L1 growth . The defects caused by feeding RNAi of let-363/TOR ( Long et al . , 2002 ) are more severe than that by feeding RNAi of other TORC1 components , described above . elo-5 ( − ) nprl-3 ( − ) ;let-363 ( RNAi ) animals displayed pleiotropic phenotypes that could not be rescued by C17ISO supplement ( Table 1 ) , which is consistent with TORC1 acting downstream of ELO-5 . If d17iso-GlcCer mainly acts through TORC1 for the growth regulation function , constitutive activation of TORC1 should be sufficient to overcome the L1 arrest of elo-5 ( − ) animals , mimicking the effect of nprl-3 ( ku540 ) . We used three different established methods to test this possibility . We first followed the scheme by Sancak et al . ( 2010 ) , in which fusion of one of the TOR binding partners , Raptor , with a C-terminal lysosome localization signal from Rap-1 can localize mTOR to the lysosome and thus constitutively activate the TOR pathway . Both Rap-1 and Raptor proteins are highly conserved in C . elegans ( encoded by rap-1 and daf-15; Long et al . , 2002; Pellis-van Berkel et al . , 2005 ) . We fused a 22-amino acid fragment of the C-terminal end of rap-1 to the C-terminal end of daf-15 and named it daf-15::rap-1 ( 22 ) . We found that 56 . 0% of elo-5 ( − ) mutants carrying daf-15::rap-1 ( 22 ) bypassed L1 arrest and reached L3 to adult stage ( n = 161; normalized against elo-5 ( −/+ ) heterozygotes containing the transgene; Figure 4A , B ) . This indicates that hyperactivation of TORC1 was sufficient to support L1 growth in mmBCFA-depleted animals . 10 . 7554/eLife . 00429 . 010Figure 4 . TORC1 activation is sufficient for mmBCFAs-mediated growth regulation . ( A ) Representative florescent images showing that elo-5 ( − ) animals with each of four RFP-marked transgenes , which constitutively activated TORC1 , bypass L1 arrest to reach beyond L3 stage ( statistical data are described in the text and Figure 4B ) . The rpl-28 promoter drives ubiquitous expression , whereas the ges-1 promoter drives the expression specifically in the intestine ( Edgar and McGhee , 1986 ) . Arrow in the upper left panel marks an arrested L1 . ( B ) Percentage of homozygous elo-5 ( − ) animals carrying the daf-15::rap-1 ( 22 ) transgene reached L3-adult stages ( n = 161 ) . The data were normalized against that of elo-5 ( −/+ ) heterozygous animals . Error bar , SD . ( C ) . Immunofluorescence images showing FIB-1 expression and localization in intestinal cells of L3 larvae . DAPI-stained nuclei are blue . Green fluorescence indicates the staining of antibody against FIB-1 . FIB-1 localization in the nucleoli is largely abolished in elo-5 ( − ) animals and restored by the nprl-3 ( − ) mutation ( the percentages of condensed nucleoli localization for the three genotypes from top to bottom are 95% [n=42] , 24% [n=112] and 93% [n=68] ) . ( D ) A model for the regulation of postembryonic growth and development by mmBCFAs and GlcCer . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 01010 . 7554/eLife . 00429 . 011Figure 4—figure supplement 1 . Microscopic images of C . elegans with raga-1 ( RNAi ) treatment . Animals without C17ISO supplement were developmentally arrested at L1 stage ( arrows ) . Animals with C17ISO supplement reached adulthood ( arrowhead ) , suggesting that the arrest depends on mmBCFA deficiency and is not caused by RNAi itself . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 01110 . 7554/eLife . 00429 . 012Figure 4—figure supplement 2 . Leucine could not promote postembryonic development independent of the mmBCFA/d17isoGlcCer/TORC1 pathway . Normalized percentages of elo-5 ( − ) animals that reached adulthood on various supplements . While 1 mM C17ISO could suppress the L1 arrest of elo-5 ( − ) , 10 mM leucine could not . Error bar: SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 01210 . 7554/eLife . 00429 . 013Figure 4—figure supplement 3 . mmBCFA/GlcCer/TORC1 pathway is independent of the DAF-7/TGF-β pathway . ( A ) Cartoon illustration of a simplified DAF-7/TGF-β pathway in C . elegans . Mutations in daf-3 , daf-5 , or bra-1 have been shown to cause constitutive activity of the TGF-β pathway and suppress dauer formation ( Patterson and Padgett , 2000 ) . ( B ) Percentages of animals of the indicated genotypes that reached adulthood on elo-5 ( RNAi ) plates . Mutation in none of these negative regulators of the daf-7 pathway permitted the elo-5 ( RNAi ) –treated animals to bypass L1 arrest , suggesting that the TGF-β pathway does not act downstream of mmBCFAs . ( C ) Percentages of daf-7 ( − ) animals that exited dauer stage to reach adulthood on indicated RNAi plates . elo-5 ( RNAi ) enhanced the constitutive dauer formation phenotype of a daf-7 ( − ) mutant . ( D ) Percentages of daf-7 ( − ) animals that exited dauer stage to reach adulthood on various branched lipid supplements . Neither C17ISO nor d17iso-SPA could suppress the constitutive dauer formation of daf-7 ( − ) , suggesting that DAF-7/TGF-β does not act upstream of mmBCFAs or d17iso-sphingolipid . ( E ) Percentages of asna-1 ( − ) animals that reached adulthood on plates with various branched-chain lipid supplements . asna-1 encodes a protein required for proper DAF-7/TGF-β function and an asna-1 ( − ) mutation causes L1 arrest ( Kao et al . , 2007 ) . Neither C17ISO nor d17iso-SPA permits asna-1 ( − ) animals to bypass L1 arrest . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 01310 . 7554/eLife . 00429 . 014Figure 4—figure supplement 4 . Lysosome integrity is not disrupted in mmBCFA-deficient animals . ( A ) – ( H ) DIC and GFP images illustrating LMP-1::GFP ( A–D ) and GLO-1::GFP ( E–H ) expression patterns are similar throughout the intestine of young wild-type or elo-5 ( RNAi ) larvae . Both LMP-1::GFP ( A–D ) and GLO-1::GFP are lysosomal markers in C . elegans . These data indicate that lysosomal integrity is not disrupted in mmBCFA-deficient animals . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 01410 . 7554/eLife . 00429 . 015Figure 4—figure supplement 5 . Neutral red staining of let-363 -and elo-5-deficient animals . ( A ) – ( H ) DIC and Rhodamine channel fluorescence images of larvae stained with Neutral red . ( A–D ) let-363 homozygous L3 animals have increased size and intensity of Neutral red stained lysosomes ( D ) compared to the heterozygous control ( B ) . ( E–H ) elo-5 homozygous L1 arrested animals show similar Neutral red intensity and staining pattern ( H ) when compared to the heterozygous control ( F ) . These data indicate that ( 1 ) lysosome integrity is not disrupted in elo-5 ( − ) animals , and ( 2 ) unlike let-363 ( − ) , elo-5 ( − ) animals do not show increased Neutral red staining in the intestine . DOI: http://dx . doi . org/10 . 7554/eLife . 00429 . 015 The second TORC1-activating transgene used was raga-1Q63L that has been proven to constitutively activate TORC1 both in mammals and C . elegans ( Kim et al . , 2008; Schreiber et al . , 2010 ) . We found that 50 . 4% of elo-5 ( − ) mutants containing the raga-1Q63L transgene bypassed L1 arrest and reached L3–L4 stage ( n = 248 , p=1 . 0 × 10−6; normalized against elo-5 ( −/+ ) heterozygotes containing the transgene; Figure 4A ) while none of elo-5 ( − ) mutants without this transgene did ( n = 136 ) . Finally , we also tested a third TORC1-activating transgene , rheb-1Q71LQ72D , and found that it also permitted 41 . 4% of elo-5 ( − ) animals to grow beyond the L1 stage ( n = 169; normalized against elo-5 ( −/+ ) heterozygotes containing the transgene; Figure 4A ) . These results led to the conclusion that TORC1 acts downstream of d17iso-GlcCer to promote L1 growth and development and that this activity is negatively regulated by NPRL-2/3 . Previous studies on elo-5 , acs-1 , tat-2 , and cgt-1/cgt-3 mutants suggested that biosynthesis and localization of mmBCFAs and their derived sphingolipids in the intestine are essential and sufficient for postembryonic development ( Marza et al . , 2009; Seamen et al . , 2009; Kniazeva et al . , 2012 ) . The fact that elo-5 ( − ) ;nprl-3 ( ku540 ) animals treated with ife-2 RNAi , that targets intestinal but not neuronal expression of the gene ( Syntichaki et al . , 2007 ) , are L1 arrested ( Table 1 ) also points to the requirement of TORC1 activity in the intestine . To directly evaluate a role of intestinal TORC1 in mmBCFA/d17iso-GlcCer regulated growth , we made a raga-1Q63L TORC1 hyperactivating transgene driven by the intestinal-specific promoter ges-1 ( Edgar and McGhee , 1986 ) . We found that elo-5 ( − ) animals containing this transgene bypassed the L1 arrest ( 37 . 9% , n = 132 , normalized against elo-5 ( −/+ ) heterozygotes containing the transgene; Figure 4A ) . The rescue with the intestinal expression was similar to that by a ubiquitously expressed raga-1Q63L transgene ( above ) . A neuronal-specific rgef-1 promoter–driven ( Brignull et al . , 2006 ) raga-1Q63L , however , could not suppress the L1 arrest of elo-5 ( − ) animals ( 0% , n > 100 ) . These results suggest that the intestinal d17iso-GlcCer/TORC1 pathway regulates postembryonic growth . Although the genetic data described above sufficiently indicate that TORC1 acts downstream of d17iso-GlcCer and NPRL-2/3 for L1 growth , we carried out further tests to observe downstream activity of TORC1 in the intestine . Because the biochemical assay for p70S6K phosphorylation has not been established in C . elegans’ TOR-related studies , we adopted the strategy by Sheaffer et al . ( 2008 ) , where localization of FIB-1 , a Box C/D small nucleolar ribonucleoprotein ( snoRNP ) was used as a marker of TORC1 activation in C . elegans . In wild-type animals , FIB-1 is highly expressed and localized in the nucleolus , where it methylates pre-rRNAs during ribosome maturation , but the expression is decreased and no longer localized to the nucleolus in let-363/TOR ( − ) animals ( Sheaffer et al . , 2008 ) . We found that the nucleolar localized FIB-1 was dramatically reduced in intestinal cells of mmBCFA-deficient [elo-5 ( RNAi ) ] animals ( Figure 4C ) . Furthermore , nprl-3 ( − ) rescued the FIB-1 expression and nucleolar localization in elo-5 ( − ) back to that of wild type ( Figure 4C ) . Combining the results from the functional and biochemical assays described above , we conclude that TORC1 is the key downstream factor mediating mmBCFA and d17iso-GlcCer functions in the intestine to promote postembryonic growth and development ( Figure 4D ) . In this study , we uncovered and characterized a novel GlcCer-stimulated TORC1 pathway that promotes postembryonic development ( Figure 4D ) . This discovery reveals a specific link between lipids and the TORC1 signaling pathway . Both ceramide and TORC1 have been individually implicated for roles in stress response , apoptosis , cancers , and other cellular processes by studies in C . elegans and mammals ( Hannun and Obeid , 2008; Hansen et al . , 2008; He and Klionsky , 2009; Howell and Manning , 2011; Laplante and Sabatini , 2012; Menuz et al . , 2009; Zoncu et al . , 2011 ) . It is conceivable that some of these functions are mediated by this GlcCer/TORC1 signaling pathway . Moreover , our data suggest that this GlcCer/TORC1 pathway may serve as a ‘check point’ to coordinate metabolic status in the intestine with postembryonic growth and development . d17iso-GlcCer is a sphingolipid metabolite rather than a nutrient directly absorbed from food ( Merrill et al . , 1997 ) ; it is the product of a long biosynthetic pathway involving many enzymatic steps and thus may reflect the availability of many metabolites and enzymes . Although leucine is a precursor molecule of mmBCFA biosynthesis , this long lipid synthesis pathway is unlikely part of a mechanism to specifically sense the level of essential amino acids . Furthermore , mmBCFA levels are stable in starved L1 larvae ( Kniazeva et al . , 2008; Figure 4—figure supplement 4C ) , suggesting that mmBCFA levels do not directly reflect the feeding status through a direct substrate–product relationship . There are several reported leucine-sensing mechanisms , based on studies using tissue culture cells ( Bonfils et al . , 2012; Duran et al . , 2012; Han et al . , 2012; Zoncu et al . , 2012 ) . To test if such a pathway acts in parallel to the mmBCFA/d17iso-GlcCer pathway to promote TORC1 in the C . elegans intestine , we supplemented elo-5 ( − ) animals with a high level of dietary leucine ( 10 mM ) and failed to observe any suppression of the L1 arrest phenotype ( Figure 4—figure supplement 2 ) , consistent with the idea that TORC1 activation by mmBCFA/d17iso-GlcCer is essential in the intestine of L1 larvae and cannot be effectively compensated by another leucine/TORC1 mechanism in C . elegans . In C . elegans , both the IIS and DAF-7/TGF-β pathways have been shown to play critical roles in the regulation of postembryonic growth and development in response to nutrient/food availability ( Baugh and Sternberg , 2006; Mukhopadhyay and Tissenbaum , 2007; Fielenbach and Antebi , 2008; Lee and Ashrafi , 2008; Jones et al . , 2009; Soukas et al . , 2009 ) . Our previous data and this study indicated that the GlcCer/TORC1 pathway is initiated in the intestine ( Figure 4 ) and is independent of the IIS pathway ( Kniazeva et al . , 2008 ) . Our additional analysis also suggests that the GlcCer/TORC1 pathway is DAF-7/TGF-β pathway independent ( Figure 4—figure supplement 3 ) . This independence may facilitate the role of such a pathway to promote specific developmental events under specific physiological conditions . There are several reasons for us to believe that such a GlcCer/TORC1 pathway is likely conserved in mammals . First , the components of TORC1 are conserved between C . elegans and mammals . The NPRL-3 protein we identified in this pathway is conserved in all eukaryotes , and its negative regulatory role on TOR has been characterized in budding yeast ( Neklesa and Davis , 2009 ) . Several proteins ( such as NPR2 and IML1 ) that are reported to form a complex with NPR3 in budding yeast have orthologs in C . elegans as well as mammals ( Neklesa and Davis , 2009; Wu and Tu , 2011 ) . NPRL2 ( human ortholog of NPR2 ) and DEPDC5 ( human ortholog of IML1 ) have been reported to function as tumor suppressors ( Merrill et al . , 1997; Li et al . , 2004; Neklesa and Davis , 2009 ) . This property is consistent with the role of this complex in repressing TOR-mediated growth regulation that was identified in budding yeast and this study . Second , a role for GlcCer in nutrient sensing is consistent with studies showing that ceramide and sphingolipids play roles in cell signaling and growth control in both invertebrates and mammals ( Deng et al . , 2008; Hannun and Obeid , 2008; Menuz et al . , 2009 ) . The finding that mouse mutants with blocked GlcCer biosynthesis die as early embryos suggests the essential role of these lipids in growth and development ( Yamashita et al . , 1999 ) . Third , the TOR pathway is down-regulated by inhibiting the abnormally high GlcCer biosynthesis in polycystic kidney disease in mouse models ( Natoli et al . , 2010 ) , suggesting a possible conserved link between GlcCer and TOR activities . It is currently unknown whether an iso-branched LCB is also required for the potential role of GlcCer in mammals , given that most LCBs in mammals are straight-chain LCBs . However , it is important to note that mmBCFAs ( derived from branched-chain amino acids ) are also present in mammals . For example , they are constituents of sphingolipids in skin cells and the intestinal tract of human newborns . They are also found to be incorporated into LCBs ( to form an iso- or an anteiso-LCB similar to that in C . elegans ) in multiple mammalian tissues ( Aungst , 1989; Karlsson , 1997; Oku et al . , 2000; Ran-Ressler et al . , 2008 ) . Therefore , their important physiological functions , including possible roles in ceramide-involved growth regulation , may be assumed , albeit not yet uncovered experimentally . How does d17iso-GlcCer regulate the activity of TORC1 ? Recent work in mammalian cells has indicated that TORC1 , for its role in sensing amino acids , localizes at the surface of the lysosome , likely in lipid rafts ( Nada et al . , 2009 ) . If such a mechanism is conserved in all animals , conceptually , d17iso-GlcCer could act as a ligand that binds to a ‘receptor’ to either repress the NPRL-2/3 complex or activate the TOR complex at the surface of the lysosome . Alternatively , this lipid could be required for lysosome biogenesis or could be an essential constituent of the membrane microdomain that permits the proper localization or activity of TORC1 on the lysosome . By examining lysosomal markers LMP-1::GFP ( Artal-Sanz et al . , 2006; O’Rourke et al . , 2009; Rabbitts et al . , 2008 ) , GLO-1::GFP ( Schroeder et al . , 2007; Zhang et al . , 2010 ) , and Neutral red ( Long et al . , 2002 ) , we did not observe obvious defects in lysosome formation in elo-5 ( − ) mutants ( Figure 4—figure supplements 4 and 5 ) . Major defects in lysosome formation would also be inconsistent with our genetic suppression data and previous studies on GlcCer by others ( Entchev et al . , 2008; van der Poel et al . , 2011 ) . One possible mechanism is that d17iso-GlcCer acts through the V-ATPase pathway , since the lysosomal V-ATPase has been shown to be stimulated by GlcCer and to play an essential role in TORC1 activation in studies using tissue culture cells ( van der Poel et al . , 2011; Zoncu et al . , 2011; Bar-Peled et al . , 2012 ) . Further biochemical and genetic analyses are needed to test this hypothesis in C . elegans . It may be important to point out again that activation of TORC1 bypasses the robust L1 arrest phenotype caused by either mmBCFA or d17iso-GlcCer deficiency . Remarkably , the elo-5 ( − ) ; nprl-3 ( ku540 ) double mutants , still deficient for these lipids , propagate continuously . Therefore , the only essential biochemical role of d17iso-GlcCer might be to activate TORC1 or repress NPRL-3 . In other words , the biochemical mechanism underlying this role of d17iso-GlcCer appears to be very specifically connected to TORC1 activation . The following strains were obtained from the Caenorhabditis Genetics Center Database ( CGC ) or as indicated; wild-type N2 Bristol , elo-5 ( gk208 ) , rrf-3 ( pk1426 ) , daf-3 ( mgDf90 ) , bra-1 ( nk1 ) , daf-5 ( e1386 ) , asna-1 ( ok938 ) /hT2[bli-4 ( e937 ) let ? ( q782 ) qIs48] , let-363 ( ok3018 ) , pwIs50[lmp-1::GFP + Cbr-unc-119 ( + ) ] , hjIs9 [ges-1p::glo-1::GFP + unc-119 ( + ) ] . The cgt-1 ( tm1027 ) and cgt-3 ( tm504 ) mutants were provided by the Mitani Lab ( National BioResource Project , Tokyo , Japan ) . Caenorhabditis elegans were maintained at 20°C on NGM plates ( referred to as standard plates ) with Escherichia coli OP50 bacterial food ( OP50/NGM ) . Washes and bleaching were done according to standard protocols ( Stiernagle , 2006 ) . Fatty acids C13ISO , C17ISO ( Larodan ) , C16:1 n7 , C18:1 n7 , and C20:4 n6 ( Sigma ) , as well as C17iso-d18:1-Ceramide and d17iso-SPA ( custom synthesis; Larodan ) were prepared as 10 mM stocks in DMSO . Leucine ( Sigma ) was prepared as a 100 mM stock in water . A stock solution was mixed with 500 µl of OP50 overnight bacterial suspension in a 1:10 ratio . Using elo-5 ( gk208 ) mutants . elo-5 ( gk208 ) adults maintained on the plates with S . maltophlia ( Kniazeva et al . , 2008 ) were washed off in M9 buffer , bleached , and eggs were plated on NGM plates spotted with 300 µl of OP50 overnight liquid culture supplemented with 1 mM C13ISO , which is a less efficient mmBCFA supplement than C17ISO and allows apparent normal growth only in the first generation followed by uniformly arrested C17ISO-deficient L1s in the second generation . Control animals were prepared in the same way , except C17ISO was used as a supplement for the N2 strain instead of elo-5 ( gk208 ) . Using RNAi . Adult animals of the corresponding strains were washed off OP50/NGM plates , bleached , and eggs plated on elo-5 ( RNAi ) plates prepared according to the standard protocol . Adults of the next generation were bleached for eggs , producing C17ISO-deficient larvae . A mixed population of elo-5 ( gk208 ) mutants maintained on plates with C17ISO supplement , which promoted wild-type growth and proliferation , were collected , washed , and incubated for 1 hr in M9 before replating on OP50/NGM plates without supplement . The next day , animals were washed off and used in the corresponding experiments . Depletion of C17ISO was confirmed by GC analysis of the FA composition in total lipid extracts from a representative group of animals . Gas chromatography method was described in a previous report ( Kniazeva et al . , 2008 ) . Mass spectrometry sample preparation was described in our previous report ( Kniazeva et al . , 2012 ) . Briefly , lipid extracts were dissolved in 1 ml of methanol with 1 mM formic acid and subjected to quantitative lipid analysis using a 4000 Q-Trap mass spectrometer ( AB Sciex ) . Samples were infused at a flow rate of 8 μl/min using a Harvard Apparatus syringe pump ( Harvard Apparatus ) . The detailed scan modes are described in the figure legends . The S . spiritivorum strain was obtained from ATCC ( #33861 ) and cultured under the conditions suggested by www . atcc . org . The total lipids were extracted by the method of Bligh and Dyer ( 1959 ) . The alkaline stable lipid fraction was purified as described previously ( Naka et al . , 2003 ) . The hydrolysis of sphingolipids to generate side chain fatty acid and LCB fractions was done as described previously ( Naka et al . , 2003 ) , with an additional last step using 100% methanol to extract the dry LCB fraction after evaporation of the chloroform solvent . The final fraction of LCB was verified by mass spectrum . We performed a screen that was significantly different from our previously published screen that resulted in isolation of three tat-2 ( − ) alleles that suppress elo-5 ( gk208 ) ( Seamen et al . , 2009 ) . The original strain used in this genetic screen was elo-5 ( gk208 ) injected with elo-5 ( genomic rescuing ) , elo-5Promoter::GFP ( Kniazeva et al . , 2004 ) , and rol-6 ( dn ) constructs . L4-staged P0 animals were mutagenized with standard EMS treatment and then single cloned to 10-cm plates containing 1 mM C13ISO . Suppressor candidates were determined in the F2 generation by the ability to reach gravid adulthood without carrying the GFP or rol-6 marker and the ability to grow to gravid adulthood without any supplement . From ∼2000 haploid genomes , 4 suppressor candidates were isolated . Among these four , and several suppressor mutations isolated in a previous screen ( Seamen et al . , 2009 ) , only ku540 permits elo-5 ( − ) animals to grow indefinitely without any mmBCFA supplementation and without recovering the production of mmBCFAs . In our previously published screen ( Seamen et al . , 2009 ) , we isolated three mutations in tat-2 that can suppress elo-5 ( − ) –induced L1 arrest in the presence of C13ISO , or temporarily suppress the L1 arrest for one generation without C13ISO supplement . The hypothesis was that tat-2 ( − ) alters the subcellular localization of certain mmBCFA-containing lipids so that their levels ( from a slow C13ISO to C17ISO conversion or from maternal sources ) are sufficiently maintained for one more generation . Therefore , tat-2 ( − ) does not bypass the requirement of mmBCFAs for growth . The tat-2 ( − ) mutations were also found to partially and temporarily suppress the lethality associated with disrupting the function of the enzyme ( SPTL-1 ) in the first step of sphingolipid biosynthesis , suggesting a potential link between sphingolipids and mmBCFAs . However , this observation may not be interpreted as the proof that mmBCFAs function through sphingolipids for the following reasons . ( 1 ) Unlike feeding d17iso-SPA or the downstream suppressor nprl-3 ( ku540 ) that fully suppresses mmBCFA deficiency [elo-5 ( − ) ]–induced L1 arrest ( this study ) , tat-2 ( − ) does not bypass the requirement of mmBCFAs for growth for more than one generation . ( 2 ) sptl-1 ( − ) disrupts all sphingolipid biosynthesis and its phenotype is , in contrast to elo-5 ( − ) , highly pleiotropic , as animals showed various morphological and growth defects and die at various larval stages . ( 3 ) tat-2 does not exclusively function with mmBCFA-containing lipids; it is also a critical player in steroid metabolism as exemplified by a recent article ( Liu et al . , 2012 ) . Therefore , the link between mmBCFA and sphingolipids through their interactions with tat-2 could be indirect and the partial suppression of sptl-1 ( − ) by tat-2 ( − ) could be due to suppression of mmBCFA-unrelated functions of sphingolipids . For SNP mapping of ku540 , we used a Hawaii strain based elo-5 ( gk208 ) mutant ( Seamen et al . , 2009 ) to cross with elo-5 ( gk208 ) ku540 animals and determined the locus of ku540 by linkage analysis ( Davis et al . , 2005 ) . The three-point-mapping strain dpy-13 ( e184 ) elo-5 ( gk208 ) unc-24 ( e138 ) was also used in this study . The elo-5 ( − ) ku540 genomic DNA was prepared using the Genomic DNA Sample Preparation Kit ( Illumina ) and sent for Illumina deep sequencing ( High Throughput Next-Generation Sequencing Core , University of Colorado ) . The raw data was analyzed using Maqgene ( Bigelow et al . , 2009 ) , and candidate mutations were confirmed by PCR and sequencing . Homologs of NPRL-2 and NPRL-3 were identified at ensembl . org . The alignment among NPRL3 homologs was done using MacVector software . Except for Figure 2I , all elo-5 ( − ) nprl-3 ( ku540 ) animals being tested for Figure 2 were from homozygous mothers . For Figure 2I , because the control cgt-3 ( − ) ;cgt-1 ( − ) animals were fully arrested at L1 and were derived from cgt-3 ( − ) ;cgt-1 ( − ) /nT1[qIs51] heterozygotes , we had to use heterozygous cgt-3 ( − ) ;elo-5 ( − ) nprl-3 ( ku540 ) /nT1[qIs51];cgt-1 ( − ) /nT1[qIs51] P0 animals to generate cgt-1 ( − ) ;cgt-3 ( − ) ;elo-5 ( − ) nprl-3 ( ku540 ) homozygous animals to generate comparable data . Because it is difficult to determine the genotype of arrested L1 animals , the ratios of homozygous cgt-1 ( − ) ;cgt-3 ( − ) ;elo-5 ( − ) nprl-3 ( ku540 ) animals that reached adulthood were calculated by the ratio of the homozygous animals in the total adult population , with a normalization by dividing by the expected ratio of 20% , from the Mendelian distribution of strains containing a recessive lethal translocation balancer ( Edgley et al . , 2006 ) . Generation of elo-5 ( − ) nprl-3 ( ku540 ) homozygous animals for the tests shown in Table 1 and Figure 4B is described below . All RNAi by feeding , except cgt-3 ( RNAi ) and let-363 ( RNAi ) , used bacterial clones from the MRC RNAi library ( Kamath et al . , 2003 ) or the ORF-RNAi Library ( Open Biosystems ) ( Figures 1–3 ) . cgt-3 ( RNAi ) and let-363 ( RNAi ) constructs were made as described ( Marza et al . , 2009; Honjoh et al . , 2009 ) . Feeding RNAi experiments were done as previously described ( Kniazeva et al . , 2008 ) . The DNA templates for nprl-3 and nprl-2 dsRNA synthesis were amplified from the RNAi-containing bacterial strain ( MRC RNAi library ) by PCR using T7 primers . dsRNA was synthesized using a MEGAscript RNAi Kit ( Life Technologies ) and then injected into adults of elo-5 ( − ) /nT1[qIs51] , elo-5 ( - ) nprl-3 ( ku540 ) /nT1[qIs51] , or cgt-3 ( - ) ;cgt-1 ( - ) /nT1[qIs51] . The eggs were collected from the 8th to 24th hr after injection . Nongreen F1 adult animals were verified by PCR to confirm the homozygosity of elo-5 ( − ) or cgt-1 ( − ) ;cgt-3 ( − ) animals . All RNAi ( raga-1 , rheb-1 , rsks-1 , ife-2 ) by feeding used sequence confirmed bacterial clones from both the MRC RNAi library ( Kamath et al . , 2003 ) or the ORF-RNAi Library ( Open Biosystems ) ( Table 1 ) . These RNAi clones have been extensively used for TORC1-related studies in many publications/meeting abstracts , and their efficiency and specificity have been well established . Furthermore , results from those references and our experiments have shown RNAi feeding of these genes did not reproduce the strong larval lethal phenotype ( Hansen et al . , 2007; Syntichaki et al . , 2007; Lemire et al . , 2009; Ching et al . , 2010; Polley and Fay , 2012 ) . This difference indicates that TORC1 function is not completely eliminated by RNAi feeding of these genes ( raga-1 , rheb-1 , rsks-1 , ife-2 ) . The elo-5 ( − ) nprl-3 ( ku540 ) animals were balanced by a GFP-labeled nT1[qIs51] balancer and treated with feeding RNAi . In the next generation , similar to the method described above , the ratios of homozygous elo-5 ( − ) nprl-3 ( − ) animals that reached adulthood were calculated by the ratio of homozygous elo-5 ( − ) nprl-3 ( − ) animals in the total adult population , with a normalization by dividing by the expected ratio 20% . As reported , RNAi knockdown of TORC1 components and the downstream target genes would affect the normal development of C . elegans ( Long et al . , 2002; Syntichaki et al . , 2007; Zoncu et al . , 2011 ) . In our experiments , the usage of the heterozygous elo-5 ( − ) nprl-3 ( ku540 ) /nT1[qIs51] animals for the RNAi knockdown experiment allowed us to exclude the potential negative effect from RNAi treatment of TORC1 components and TORC1 target genes themselves . Any of those elo-5 ( − ) nprl-3 ( − ) –independent negative effects from those RNAi treatments would affect the heterozygous as well as the homozygous elo-5 ( − ) nprl-3 ( − ) animals , and therefore not be included in the ratio in the data presented in Table 1 . For the pPD95 . 77-nprl-3::GFP plasmid , we cloned the potential 1 kbps promoter region upstream of the operon containing nrpl-3 and inserted it into the pPD95 . 77 vector ( Figure 4A , B ) . For raga-1Q63L , the genomic DNA including the full coding region was cloned into pPD95 . 77 , driven by a ubiquitous RPL28 promoter . For ges-1::raga-1Q63L , the genomic DNA including the full coding region was cloned into pPD95 . 77 , driven by an intestinal-specific ges-1 promoter . For rheb-1Q71LQ72D , the genomic DNA including the full coding region and about 1 kbps of upstream sequence was cloned into pPD95 . 77 . Amino acid mutations in both mutants were introduced by replacement with PCR-generated DNA fragments containing the designed mutations . For daf-15::rap-1 ( 22 ) , the genomic DNA of daf-15 , including the full coding region and the 1 kbps potential promoter region upstream of daf-15 , was cloned from the fosmid WRM061cH04 . After the stop codon of daf-15 was removed , it was fused with the genomic DNA encoding the last 22 amino acids and the 3′ UTR of rap-1 . For pPD95 . 77-nprl-3::GFP , 25 ng/μl of plasmid was injected in wild-type animals . For the daf-15::rap-1 ( 22 ) rescue experiment , elo-5 ( − ) /nT1[qIs51] animals were injected with 10 ng/µl daf-15::rap-1 ( 22 ) and 25 ng/µl psur-5::RFP plasmid . In the next generation , the ratios of homozygous elo-5 ( − ) animals that reached L3-young adulthood were calculated by the ratio of homozygous elo-5 ( − ) nprl-3 ( − ) animals in RFP-positive L3-young adult population , with a normalization by dividing by the expected ratio 20% , as described above . The reason for using heterozygous elo-5 ( − ) /nT1[qIs51] animals for this experiment is similar to that for the RNAi knockdown experiment we described above . By this method , we exclude the negative effect we observed from constitutively active TORC1 transgenes ( daf-15::rap-1 ( 22 ) , raga-1Q63L , ges-1::raga-1Q63L or rheb-1Q71LQ72D ) for their rescue effects . The Neutral red staining was performed following Long et al . ( 2002 ) . Animals were fed with Neutral red containing bacteria food for <10 min before evaluation by microscopy . Analysis of GFP expression and phenotypic abnormalities were performed with Nomarski optics using a Zeiss Axioplan2 microscope and a Zeiss AxioCam MRm CCD camera . Plate phenotypes were observed using a Leica MZ16F dissecting microscope , and pictures were taken with a Hamamatsu C4742-95 CCD camera . All statistical analyses , except the dsRNA feeding for TOR-related genes , were performed using Student’s t-test , and p<0 . 05 was considered a significant difference . The Fisher’s exact test was used for analysis of the TOR-related dsRNA feeding experiments and raga-1Q63L rescue experiment , and p<0 . 05 was considered a significant difference .	Animals require nutrients , including carbohydrates , lipids , and amino acids , for development and growth , and to maintain the normal functioning of cells . However , in most natural environments , the availability of food tends to fluctuate . Some animals have therefore acquired the ability to dramatically reduce their metabolic activity , and thus their energy and nutrient needs to survive fasting conditions . Caenorhabditis elegans is a transparent nematode worm that is used extensively as a model organism . When C . elegans larvae hatch in a food-free environment , they enter a quiescent state in which they suspend growth and cell division to conserve energy . However , the mechanisms that underlie this ability are not fully understood . Here , Zhu et al . reveal that a type of lipid called a sphingolipid is required for C . elegans larvae to begin postembryonic development . When this lipid is absent in the environment and not synthesized internally , the larvae remain in a state of arrested development , which can be overcome by resupplying the lipid . Zhu et al . show that the lipid acts through a signaling pathway involving an enzyme complex called TORC1 and that the effect of the lipid can be blocked by another protein complex called NPRL-2/3 . TORC1 is well known for its role in sensing amino acids and growth factors , but this is the first time that it has been shown to be involved in detecting lipids . Strikingly , Zhu et al . also show that , in the absence of the lipid , postembryonic growth and development can be initiated by activating TORC1 or inhibiting NPRL-2/3 . The work of Zhu et al . thus reveals a novel regulatory function of a specific fatty acid and sphingolipid variant that is used by C . elegans to coordinate its growth and development with its metabolic status or the availability of nutrients . Since all components of the pathway are conserved in mammals , the results could help to improve our understanding of how caloric restriction influences human health and aging .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2013	A novel sphingolipid-TORC1 pathway critically promotes postembryonic development in Caenorhabditis elegans
Biological membranes create compartments , and are usually formed by lipid bilayers . However , in hyperthermophilic archaea that live optimally at temperatures above 80°C the membranes are monolayers which resemble fused bilayers . Many double-stranded DNA viruses which parasitize such hosts , including the filamentous virus AFV1 of Acidianus hospitalis , are enveloped with a lipid-containing membrane . Using cryo-EM , we show that the membrane in AFV1 is a ~2 nm-thick monolayer , approximately half the expected membrane thickness , formed by host membrane-derived lipids which adopt a U-shaped ‘horseshoe’ conformation . We hypothesize that this unusual viral envelope structure results from the extreme curvature of the viral capsid , as ‘horseshoe’ lipid conformations favor such curvature and host membrane lipids that permit horseshoe conformations are selectively recruited into the viral envelope . The unusual envelope found in AFV1 also has many implications for biotechnology , since this membrane can survive the most aggressive conditions involving extremes of temperature and pH . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 001 Many viruses , including some of the most devastating human pathogens such as Ebola virus , are enveloped with a lipid membrane . The membrane is considered to be an adaptation to the host that has been convergently acquired in different virus orders ( Buchmann and Holmes , 2015 ) . Some groups of evolutionarily related viruses contain both enveloped and non-enveloped members . One example is provided by hyperthermophilic archaeal viruses of the order Ligamenvirales ( Prangishvili and Krupovic , 2012 ) which contain non-enveloped , rigid rod-shaped viruses of the family Rudiviridae and enveloped , flexible filamentous viruses of the family Lipothrixviridae . Viruses from these two families have many homologous genes and build their virions using structurally similar major capsid proteins . The ~4 Å-resolution cryo-EM structure ( DiMaio et al . , 2015 ) of a less-complex rudivirus , the Sulfolobus islandicus rod-shaped virus 2 ( SIRV2 ) ( Prangishvili et al . , 1999 ) , has shown that the viral double-stranded ( ds ) DNA is completely insulated from the external medium by the single capsid protein which transforms the viral DNA into A-form , explaining the stability of rudiviruses in extremely aggressive ( 80°C , pH 3 ) natural habitats . Virions of lipothrixviruses are more complex and are built from two paralogous major capsid proteins ( MCP1 and MCP2 ) which bind dsDNA to form the nucleocapsid ( Goulet et al . , 2009 ) . The nucleocapsid is enveloped by a lipid membrane and the termini of the virion are decorated with specialized structures involved in host recognition ( Bettstetter et al . , 2003 ) . Membranes of hyperthermophilic archaea often consist of unusual tetraether lipids , which form monolayers rather than bilayers typical of bacterial and eukaryotic membranes ( De Rosa and Gambacorta , 1988; Valentine , 2007 ) . It has been demonstrated that such membranes are more rigid and stable than typical bilayers , a property with considerable biotechnological appeal . However , molecular details of membrane remodeling in archaea remain grossly understudied , and how a virus captures a non-bilayer membrane from its host is virtually unknown and is conceptually difficult to reconcile with our current understanding of membrane remodeling by viruses infecting eukaryotes ( Harrison , 2008 , 2015; Hurley and Hanson , 2010 ) . Comparison of the organization of rudiviruses and lipothrixviruses provides an opportunity to probe structural changes permitted by envelope acquisition and can also explain how the lipid envelope may contribute to resisting stresses posed by the extreme geothermal environment . To better understand the evolutionary relationship between enveloped and non-enveloped viruses , we used cryo-EM ( Figure 1 ) to determine the structure of Acidianus filamentous virus 1 ( AFV1 ) , the prototypical lipothrixvirus infecting the hyperthermophilic and acidophilic archaeon Acidianus hospitalis ( Bettstetter et al . , 2003 ) , and compared the resultant structure to that of SIRV2 . 10 . 7554/eLife . 26268 . 003Figure 1 . Cryo-EMs of AFV1 . Arrows point to regions where the virions have been demembranated . This leads to a narrower diameter of the virions and a greatly increased flexibility , as seen by the loops in a , b . The loss of the membrane does not necessarily occur over the whole virion , as can be seen in c where a membrane-enveloped region in the center is bracketed by two regions with no membrane . The scale bar ( b ) is 1 , 000 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 003 Determining the structure of AFV1 was complicated by the fact that the virions are significantly more flexible , both with respect to bending as well as extension and compression , than those of SIRV2 . Segments could be classified by the pitch of the prominent helix which ranged from 39 to 47 Å ( Figure 2 ) , in contrast to the rather fixed pitch of 42 Å in SIRV2 ( DiMaio et al . , 2015 ) . A three-dimensional reconstruction ( Figure 3 ) of AFV1 not only reveals the gross morphology but also has allowed us to build a full atomic model for both the two MCP subunits and the DNA . While the Fourier Shell Correlation ( FSC ) is frequently used as the measure of resolution , numerous concerns have been raised about this metric since it is really a measure of self-consistency and not resolution ( Subramaniam et al . , 2016 ) . Nevertheless , the ‘gold standard’ FSC ( Figure 4 ) yields an overall resolution of 4 . 1 Å . We think that this is overly optimistic , and may arise from strong features in the DNA ( Figure 5 ) ( given the higher MW of the phosphates , the contrast is greater than for protein ) . A reasonable estimate ( based upon comparison with the atomic model ) is ~4 . 5 Å , but it is clear that parts of the complex ( such as the outer helices facing the membrane , Figure 3b ) are at a worse resolution , while other parts ( such as the helix-turn-helix motifs on the very inside , Figure 3c ) are at a better resolution . 10 . 7554/eLife . 26268 . 004Figure 2 . The distribution of segments sorted against references containing 1-start helices with different pitch values . The validity of this sorting was confirmed by taking power spectra from different bins , which behaved as expected and showed the helical pitch of the corresponding reference . The reconstruction was generated using segments from the central three bins . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 00410 . 7554/eLife . 26268 . 005Figure 3 . Three-dimensional reconstruction of AFV1 . ( a ) A slice perpendicular to the filament axis . The red arrows define a distance of 20 Å , the approximate thickness of the membrane enveloping the virions . The membrane has a denser outer component and a less dense inner part , separated by a region of lower density . ( b ) A view of the protein core , looking from the membrane . The asymmetric unit in the virus is a pseudo-symmetric heterodimer of MCP1 ( red ) and MCP2 ( yellow ) . ( c ) A view looking down the filament axis ( perpendicular to that in b ) with the model for the DNA phosphodiester backbone underneath the protein in blue . The helix-turn helix motif of each subunit faces into the narrow lumen . The resolution is good enough in this region that some bulky amino acids can be unambiguously located , and three Tyr21 residues are labeled . ( d ) The heterodimer in AFV1 has a pseudo-2-fold symmetry , in contrast to the homodimer in SIRV2 ( e ) which has strict 2-fold symmetry . In both , A-form DNA is bound within the central cleft . The N- and C-terminal ends in both ( d ) and ( e ) are labeled Nt and Ct , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 00510 . 7554/eLife . 26268 . 006Figure 4 . ( a ) An FSC curve between two reconstructions from completely independent sets of segments ( having no overlap ) , each started independently from a full reconstruction filtered to 7 Å resolution . The FSC falls to 0 . 143 at 1/ ( 4 . 1 Å ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 00610 . 7554/eLife . 26268 . 007Figure 5 . Packing of the DNA in the virion . ( a ) The phosphate backbone of the DNA model can be fit nicely into the density map , as most of the phosphate groups are well resolved . ( b ) A slice perpendicular to the DNA axis through the map and model . As expected for A-form DNA ( and in contrast to B-form ) , a hole is seen along the DNA axis , with the bases surrounding this cavity . ( c ) The Coulombic potential for the AFV1 capsid heterodimer shows significant positive regions ( blue ) surrounding the DNA , with negative regions ( red ) away from the DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 007 The outer diameter of the virion is ~185 Å , while the diameter of the nucleoprotein core alone ( excluding the membrane ) is ~135 Å . Surprisingly , the membrane is only ~20–25 Å thick , compared to ~40 Å found for archaeal tetraether monolayer ( Valentine , 2007 ) membranes ( Chong et al . , 2003; Chugunov et al . , 2014 ) and the 40–60 Å found for other cellular membranes and the viral envelopes derived from them ( Hollinshead et al . , 1999 ) , but a crude calculation done by integrating the cryo-EM density ( which corresponds to the Coulombic potential but will be roughly proportional to mass ) suggests that ~40% of the total mass of the virion is due to the membrane . The buoyant density for the AFV1 virions was previously determined using a CsCl gradient ( Bettstetter et al . , 2003 ) , and it was consistent with other membrane-enveloped dsDNA viruses ( King et al . , 2011 ) . The helically-averaged membrane shows two clear density peaks , with the highest one on the outside and a lower one at inner radius ( Figure 3a ) . This can be seen more easily in the cylindrically-averaged density distribution of the virion , which yields the radial mass distribution ( Figure 7a ) . 10 . 7554/eLife . 26268 . 008Video 1 . Conformational dynamics of 20 lipids from the simulated AFV1 envelope . Rendered in this movie are 20 sequential lipids from the simulated AFV1 envelope in stick form and the capsid protein in cartoon form . Frames are at nanosecond intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 00810 . 7554/eLife . 26268 . 009Video 2 . Conformational dynamics of AFV1 envelope lipids and interfacial water . Rendered in this movie are the simulated AFV1 envelope in stick form , all water molecules within 10 Å , and the capsid protein in cartoon form . Frames are at nanosecond intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 009 It is unlikely that any details of the membrane structure are lost due to the fact that either the helical symmetry of the nucleocapsid has been imposed upon the membrane ( Figure 3a ) or the membrane has been cylindrically averaged ( Figure 7a ) since the membrane is most likely a two-dimensional fluid . If there were some fine structure in the membrane ( e . g . , a liquid-crystalline phase with a spacing of ~5 Å ) then this would be lost by the helical symmetrization ( lost as well , of course , by cylindrical averaging ) but would appear in averaged power spectra . Since the membrane accounts for ~40% of the mass of the virions , and since there are no features in the averaged power spectra arising from any liquid crystalline features of the membrane , all evidence suggests that it is fluid . What we have been able to observe is that if we compare the membrane with helical symmetrization with a membrane generated by cylindrically-averaging the helical density , or with that obtained from an ab initio cylindrical symmetry reconstruction ( see Materials and methods ) , we see no systematic differences . This excludes the possibility that the membrane is deformed locally by the protein in a way that the membrane would deviate from cylindrical symmetry by the presence of certain amino acid residues either facing the membrane or inserted into the membrane . In SIRV2 , the radius of the DNA is ~60 Å , while in AFV1 the supercoiling is tighter and the radius is ~30 Å . The twist of the A-form DNA in SIRV2 is 11 . 2 bp/turn ( DiMaio et al . , 2015 ) while in AFV1 it is 10 . 8 bp/turn: in 10 right-handed turns of the 43 Å pitch AFV1 helix , there are 93 repeats of the DNA , each with 12 bp , so there are 1116 bp per 103 ( =93 + 10 ) right handed turns . These two values ( 11 . 2 and 10 . 8 ) bracket the ‘canonical’ value of 11 bp/turn frequently given for A-DNA ( Vargason et al . , 2001 ) . Interestingly , the helical pitch in both SIRV2 and AFV1 is ~42–43 Å , but in SIRV2 there are 14 . 7 homodimers per turn , while in AFV1 there are only 9 . 3 heterodimers . It is this looser packing in AFV1 that appears responsible for the greater flexibility and disorder . It also explains why the two viruses with linear dsDNA genomes of very different sizes – 35 , 450 kb for SIRV2 ( Peng et al . , 2001 ) and 20 , 080 for AFV1 ( Bettstetter et al . , 2003 ) – have virions of approximately the same length ( about 900 nm ) . In SIRV2 there are tight contacts across the helical turns , while in AFV1 such contacts are absent ( Figure 3b ) , allowing the virions to bend , extend , and compress . At the same time , due to looser protein packing , the AFV1 genome is not completely covered by the protein , while it is in SIRV2 . Consequently , the lipid envelope provides a necessary protection to the AFV1 genome in the highly acidic environment of the natural habitat , rationalizing the presence of the envelope in lipothrixviruses . When the membrane is removed ( we assume as an artifact of specimen preparation ) the virions become much more flexible ( Figure 1 ) . Since the membrane , which has fluid-like properties , is unlikely to be directly responsible for the increased rigidity of the enveloped virions , it suggests that the presence of the membrane constrains the protein and thus indirectly imparts rigidity to the structure . It was originally proposed from a crystallographic study that the two capsid proteins MCP1 and MCP2 would be packed very differently in the virion ( Goulet et al . , 2009 ) . A model , based upon crystal structures of most of one capsid protein and a fragment of the second one , proposed that one of the capsid proteins formed an inside core of the virion , with DNA wrapping around it , while the other subunit was on the outside of the DNA and partially inserted into the membrane . Surprisingly , we find that the two capsid proteins form a pseudo-symmetric heterodimer ( Figure 3d ) that resembles in many ways the symmetric homodimer found in SIRV2 ( Figure 3e ) , and that both interact with the DNA in an equivalent manner . We have accounted for all of the amino acids in the two capsid proteins with the exception of 5 N-terminal residues in both MCP1 and MCP2 . However , these residues would be too far from the membrane to contact it . Further , we see no density extending from the protein to the membrane . There are two main differences between the AFV1 and SIRV2 dimers: ( 1 ) In SIRV2 the N-terminal tail forms a long helix with a kink that allows it to continuously wrap around the DNA ( Figure 3e ) , while in AFV1 the N-terminal region of both MCP1 and MCP2 form helix-turn-helix motifs which fold back to cover the DNA on both sides ( Figure 3d ) ; ( 2 ) In SIRV2 the 2-fold axis of the dimer is perpendicular to the helical axis ( and goes through the 2-fold axis in the DNA ) , generating an overall bipolar symmetry for the virion , while in AFV1 the pseudo-2-fold axis of the heterodimer is tilted by 25 . 7° and does not intersect the helical axis , so that the virion has an overall polarity visible at fairly low resolution . Details of the wrapping of the A-form DNA by the heterodimer are shown in Figure 5 , where it can be seen that a positive Coulombic potential would surround the negatively-charged phosphate backbones of the DNA . The fact that the membrane is only 20–25 Å thick , half of regular lipid membranes , has led us to investigate the membrane further . Since the membrane lipids would not be synthesized by the virus but must come from the host , we first compared the distribution of lipids ( Figure 6a ) found in the host with those found in the virion membrane ( Figure 6b ) . There is a striking difference in the distributions showing that the incorporation of the glycerol dibiphytanyl glycerol tetraether ( GDGT ) lipids from the host is highly selective . While the single most dominant species in the host is GDGT-4 ( containing 4 cyclopentane moieties ) , in the virion it is GDGT-0 ( containing no cyclopentane moieties ) , found as only a few percent of the total host membrane lipids . Nevertheless , GDGT-0 is actually one of the most common archaeal membrane lipids ( Schouten et al . , 2013; Villanueva et al . , 2014 ) . Furthermore , it is generally the dominant , or one of the dominant , archaeal lipids in environmental samples taken from soils , rivers , lakes and oceans accounting for >40% of all GDGTs detected ( Schouten et al . , 2013 ) . The selective incorporation of host lipids in a viral membrane has previously been described , for example , in influenza budding from mammalian cells ( Gerl et al . , 2012 ) , or in a virus budding from algae ( Maat et al . , 2014 ) . Such selective incorporation could be driven by direct lipid binding by capsid proteins , enrichment of certain lipid species at sites of viral budding , or physical properties of the viral envelope that cause partitioning of lipids into or out of the nascent envelope during viral budding . Because GDGT-0 is more flexible than the cyclopentane-containing GDGT lipids ( Schwarzmann et al . , 2015 ) , it can better adopt the horseshoe conformations that have a lower free energy in the highly curved AFV-1 envelope ( Galimzyanov et al . , 2016 ) . We therefore hypothesize that selective partitioning of GDGT lipids due to the curvature of the envelope is the mechanism for GDGT-0 enrichment in the AFV-1 membrane . 10 . 7554/eLife . 26268 . 010Figure 6 . Lipid distribution of virions different from host cells . ( a ) Chemical diagrams for the lipids found in Acidianus hospitalis and AFV1 . ( b ) The distribution of lipids in the host membrane ( red ) differs significantly from the distribution found for AFV1 ( blue ) . The scale is in percentage . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 010 Knowing the lipid composition of the virion , we used molecular dynamics ( MD ) to model the viral membrane . Multiple simulations were performed of GDGT-0 lipids arranged cylindrically around the capsid assembly in different densities and orientations . Lipids were modeled with a single phosphoinositol headgroup , as this is the smallest headgroup commonly found on GDGT lipids in the A . hospitalis host ( the others are dihexose and sulfonated trihexose headgroups ) . These lipids frequently adopted a U-shaped or horseshoe conformation , and these horseshoe-rich envelopes with a mix of ‘inward-facing’ and ‘outward-facing’ lipids were the only ones that stably maintained the curvature and thickness observed in the radial density profile from cryo-EM . These lipids still form a monolayer one lipid thick , but the lipids in the monolayer have a mixed orientation . A horseshoe lipid conformation from simulation was therefore used to fit the cryo-EM radial density profile; the best-fit arrangement features 40% of lipids with headgroups facing inwards towards the nucleocapsid and 60% of lipids with headgroups facing towards the outside ( Figure 7 ) . Structural models simulated with this lipid orientation maintained a stable envelope structure with the thickness , curvature , propensity to horseshoe conformations , and the slight ~8 Å water-filled gap between envelope and capsid observed by cryo-EM , similar to a surface-supported membrane . The density in this gap observed by cryo-EM was the same as the solvent outside the virus , further suggesting that the region between the envelope and the polar capsid surface and the envelope is similar to that between a supported lipid bilayer and its planar support ( Ajo-Franklin et al . , 2005; Koenig et al . , 1996 ) . 10 . 7554/eLife . 26268 . 011Figure 7 . Modeling the viral membrane . ( a ) The cylindrically averaged density profile from EM ( blue curve ) is well fit by a cylindrical envelope ( green curve ) of phosphoinositol-GDGT0 lipids in horseshoe conformations ( b ) with 60% having headgroups facing away from the capsid and 40% having headgroups facing towards the capsid . Molecular dynamics simulations of the protein capsid and phosphoinositol-GDGT0 lipids constructed in this arrangement produced a stable envelope rich in horseshoe-conformation lipids ( c ) , while all other envelope arrangements tested failed to maintain the experimentally-derived thickness of 20–25 Å . The density peak at ~30 Å radius ( a ) arises from the DNA . The central cryo-EM density ( radius <15 Å ) could not be explained by the capsid proteins , and most likely involves either a minor viral protein or a host protein ( Figure 7—figure supplement 1 ) . Since the symmetry of the virion was imposed on this density , which likely does not have such a symmetry , the density was uninterpretable and removed from the other figures . Reconstructed density profiles from the simulations are shown in Figure 7—figure supplement 2 , accompanied by movies of 20 envelope lipids in Video 1 and of the entire envelope and interfacial water layers in Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 01110 . 7554/eLife . 26268 . 012Figure 7—figure supplement 1 . A central disconnected density ( yellow arrow ) cannot be explained by the two capsid proteins and most likely involves either a minor viral protein or a host protein . Since the symmetry of the capsid has been imposed upon this density , it is uninterpretable . In ( a ) , a thick cross-section is shown containing more than one helical turn of the capsid . In ( b ) , a higher magnification view is shown of a thin slice containing less than a helical turn . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 01210 . 7554/eLife . 26268 . 013Figure 7—figure supplement 2 . Computed radial density from molecular dynamics simulations . Radial density functions are plotted from the cryo-EM data and from 10-ns intervals in a molecular dynamics simulation of the AFV1 capsid and envelope . Simulations were performed with protein , lipid , and solvent; to compare with the observed electron density , DNA was modeled into each snapshot . Major envelope features reproduced by the simulations—the thickness and the water gap between capsid and envelope—were stably maintained over the course of the simulations . Although the envelope remained rich in horseshoe-conformation lipids , the highly ordered double-peak envelope structure that we can analytically match using electron density from horseshoe lipids was not maintained . This structure is likely quite sensitive to lipid density in the capsid , although it may also result from larger headgroup moieties , as discussed in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 26268 . 013 Simulations very robustly reproduced the width and placement of the envelope density compared to the cryo-EM data , and the density was stable over the course of multiple independent simulations . However , the lipids in the simulations were somewhat more disordered than suggested by the cryo-EM density , such that the double-peak density profile from cryo-EM was smoothed into a broader single peak , and most but not all simulated lipids were in horseshoe conformation . This could result from one of three factors: ( 1 ) a slight mismatch in the estimated density of lipids in the viral envelope leading to lateral pressure stresses in the envelope , ( 2 ) a larger lipid headgroup present in AFV-1 envelopes than those used in simulations—the simulations used a phosphoinositol headgroup and glycerol backbone as the minimal headgroups found on host lipids , but larger headgroups are also possible , or ( 3 ) factors internal to the simulation such as insufficient sampling time or slight mismatches in lipid parameterization . Despite this minor disordering of the lipid tails , the simulated AFV-1 envelope stably maintained a thickness consistent with a single horseshoe-conformation lipid with a thin layer of water between the capsid and envelope . Control simulations that used either incorrect lipid density or single-orientation lipids rather than an ‘in/out mix’ of headgroup orientations did not maintain these features over equivalent simulation timescales . These findings are thus robust and highly consistent with the experimental data . The simulation models therefore suggest that a mixture of inward-facing and outward-facing lipids primarily in horseshoe conformations is physically stable surrounding a highly curved polar capsid . These simulations match the thickness of the envelope in the electron-density profile and well explain the gap between capsid and envelope as a water layer , but they are not sufficiently powered to distinguish between some-horseshoe and all-horseshoe conformational distributions due to slow conformational relaxations of the lipids and initial-value sensitivity . We have tested sufficient initial conditions to say with confidence that ( 1 ) a canonical ‘straight’ tetraether lipid conformation is not compatible with the cylindrical curvature of the capsid; ( 2 ) an in/out orientational mix is necessary to capture the gap in density between the capsid and the envelope; and ( 3 ) multiple starting conditions with in/out horseshoe start states all produce a stable envelope with a thickness matching that observed experimentally . These simulations thus provide a specific structural model for the lipids to fit our experimental findings that GDGT-0 lipids in the envelope must occupy a horseshoe conformation based on the cryo-EM density profile of the envelope . The models also predicted that the membrane would account for ~43% of the total mass of the virion , in excellent agreement with the ~40% estimate from the cryo-EM density integration . AFV-1 is striking because its envelope differs substantially in composition and structure from those of previously described viruses . Because we have combined multiple experimental approaches with computational modeling to analyze the AFV-1 envelope , we briefly recapitulate the lines of evidence for each major finding before going on to discuss some of the important implications of this previously unappreciated envelope structure . Our cryo-EM density data show an envelope surrounding the viral capsid that is ~20 Å in thickness; our mass spectrometry data show that this envelope is composed predominantly of lipids with a GDGT-0 core . The only known conformation of GDGT-0 lipids to form a layer that is <40 Å in thickness is the horseshoe conformation , which has been characterized previously at fluid-air interfaces . Our cryo-EM density data further show an 8 Å region with density corresponding to water between the protein capsid and the lipid , and the structural refinement of the capsid yields a hydrophilic surface . Our computational models then explain these findings via a horseshoe-conformation monolayer with mixed orientation such that ( 1 ) GDGT-0 acyl tails are not fully exposed to the external solvent or the polar capsid and ( 2 ) the tight curvature of the capsid is well matched by the envelope ( which has a radius of curvature on the inside of the membrane of ~70 Å ) . Thus , the horseshoe-conformation lipid envelope is strongly supported by the experimental data themselves , while the computational model specifies the likely orientation of lipids within this envelope and provides a detailed model of molecular structure . Despite extensive exploitation of archaeal tetraether lipids for therapeutic purposes , such as archaeosome-based delivery of drugs , cancer vaccines , antigens , genes , etc . ( Kaur et al . , 2016 ) remarkably little is known about the actual structure of membranes in Acidianus specifically and Sulfolobales in general . The studies on membranes of Sulfolobales have thus far largely focused on lipid composition in different organisms and on investigation of lipid mixtures in in vitro systems . The lipid composition of Acidianus hospitalis , which we report in this study , is very similar to that previously determined in a related organism , Sulfolobus solfataricus ( Quemin et al . , 2015 ) which has a ~5 nm-thick membrane . Interestingly , it has been shown that a spindle-shaped virus SSV1 released from Sulfolobus cells by budding , similar to AFV1 , has a membrane enriched in GDGT-0 which is also considerably thinner compared to the host membrane ( Quemin et al . , 2016 ) . However , this observation remained unexplained . It thus appears that the lipid conformation described in our current study might be more general in enveloped viruses of archaea . A number of in vitro studies have described archaeal lipids forming U-shaped structures at an air-water interface ( Gliozzi et al . , 1994; Köhler et al . , 2006; Melikyan et al . , 1991; Patwardhan and Thompson , 2000; Tomoaia-Cotisel et al . , 1992 ) . Since air is extremely hydrophobic , the acyl chains face the air while the polar headgroups of these lipids face the water . The presence of cyclopentane rings ( found in the main host species , GDGT-4 ) has been suggested to rigidify the lipids , making them unable to form a horseshoe ( Gliozzi et al . , 1994 ) , in agreement with another study which found that more rigid tetraether lipids could not form a horseshoe at the air-water interface while the more flexible lipids did ( Patwardhan and Thompson , 2000 ) . Furthermore , theoretical studies suggest that membranes formed from horseshoe conformation lipids have lower curvature energies ( Galimzyanov et al . , 2016 ) and would thus be energetically favored on the highly curved AFV-1 surface . This may be the main driving force for the exquisite selectivity seen for the incorporation of host lipids in the viral membrane . A biological role for such a horseshoe conformation has not previously been suggested or found . Our study demonstrates that besides the canonical bacterial/eukaryotic membrane bilayer and archaeal monolayer , there is a third type of biological membrane , the viral horse-shoe membrane layer . The observation that a membrane that envelops a virus can be formed from lipids in such a conformation opens the door to designing such membranes for applications from drug delivery to nanotechnology . Since the archaeal lipids have been shown to resist phospholipases , extremes of temperature and pH , and can even survive autoclaving , the membrane described here has many potential applications ( Patel and Sprott , 1999 ) . The virus AFV1 was purified for electron microscopy as described earlier ( Bettstetter et al . , 2003 ) ; for lipid analysis additional purification was performed on a sucrose gradient ( Quemin et al . , 2015 ) . The purified virus preparation ( 3 μL , 1–2 μg/μl ) was applied to lacey carbon grids that were plasma cleaned ( Gatan Solarus ) and vitrified in a Vitrobot Mark IV ( FEI , Inc . ) . Grids were imaged in a Titan Krios at 300 keV , and recorded with a Falcon II direct electron detector at 1 . 05 Å per pixel , with seven ‘chunks’ per image . Each chunk , containing multiple frames , represented a dose of ~20 electrons/Å2 . A total of 557 images ( each 4 k x 4 k ) were selected that were free from drift or astigmatism , and had a defocus less than 3 . 0 μm . The program CTFFIND3 ( Mindell and Grigorieff , 2003 ) was used for determining the Contrast Transfer Function ( CTF ) and the range used was from 0 . 6 to 3 . 0 μm . The SPIDER software package ( Frank et al . , 1996 ) was used for most subsequent steps . The CTF was corrected by multiplying each image by the theoretical CTF , both reversing phases where they need to be reversed and improving the Signal-to-Noise ratio . This multiplication of the images by the CTF is actually a Wiener filter in the limit of a very poor SNR . The program e2helixboxer within EMAN2 ( Tang et al . , 2007 ) was used for boxing long filaments from the micrographs , and 546 such boxes of varying length were extracted . Overlapping boxes , 384 px long with an 8 px shift between adjacent boxes ( ~twice the axial rise per subunit ) were extracted from these long filaments , yielding 215 , 549 segments that were padded to 384 × 384 px . The CTF determination and particle picking came from the integrated images ( all seven chunks ) , while the segments used for the initial alignments and reconstruction came from the first two chunks . The determination of the helical symmetry was by trial and error , searching for a symmetry which yielded recognizable secondary structure ( Egelman , 2014 ) . The IHRSR algorithm ( Egelman , 2000 ) was used for the helical reconstructions , starting from a solid cylinder as an initial model . Once the correct symmetry was determined ( an axial rise of 4 . 6 Å and a rotation of 38 . 7° per subunit ) it was apparent that the pitch was quite variable , and segments were sorted using references that had a continuous 1-start helix with different pitch . Segments were excluded if they corresponded to a pitch less than or equal to 40 . 5 Å or greater than or equal to 44 . 1 Å ( Figure 2 ) , reducing the number of segments to 119 , 495 . The final reconstruction was generated by imposing the helical parameters found for each segment using the first two chunks on segments containing only the first chunk ( ~20 electrons/Å2 ) and using these for the back-projection in SPIDER . The variability in the structure was further overcome by only symmetrizing the central third ( 128 px ) of the 384 px long asymmetric reconstruction . Since the images had been multiplied by the CTF twice ( once by the microscope and once by us in phase correction ) , the amplitudes of the final volume were divided by the sum of the squared CTFs . The reconstructed volume ( which has a very high SNR from the extensive averaging ) is corrected only in Fourier amplitudes by dividing by the sum of the squared CTFs . This is a Wiener filter in the limit of a very high SNR . The map was also sharpened using a negative B-factor of 220 . Resolution was estimated by dividing the data set into two independent halves , such that there was no overlap in segments between one set and the other . These were used to iteratively generate ( after 20 cycles ) two reconstructions for a Fourier Shell Correlation ( FSC ) , starting with a reference volume for each set that was filtered to 7 Å . Choosing the FSC = 0 . 143 threshold yielded a resolution of 4 . 1 Å ( Figure 4 ) . The radial density profile was generated by cylindrically averaging the reconstructed volume , since the cylindrically-averaged radial density distribution after helical symmetrization is actually the same as the unsymmetrized mean radial density distribution . Cylindrically-symmetrizing the IHRSR helical reconstruction ( Figure 7a ) leaves only the equatorial terms in a Fourier-Bessel synthesis . That is the same as the approach of reconstructing by assuming cylindrical symmetry , which is the J0 Fourier-Bessel transform of the equator using Fourier methods . We have done this in real space using ~9000 segments classified as having an out-of-plane tilt of <0 . 5° , and treated these as cylindrically-symmetrical objects , reconstructed them using standard back-projection methods and then corrected for the CTF . The density profile for the membrane is the same as in Figure 7a . A small central density within the lumen of the reconstruction ( Figure 7—figure supplement 1 ) with no connectivity to the nucleocapsid could not be explained by the capsid proteins , and most likely involves either a minor viral protein or a host protein . Since the symmetry of the virion was imposed on this density , which likely does not have such a symmetry , the density was uninterpretable and removed from the other figures .	Virtually every environment on the planet is home to some form of life , even places that , at first glance , appear to be too harsh for any organism to survive in . For example , a microscopic organism known as Acidianus hospitalis thrives in highly acidic environments that are hotter than 80°C , conditions that would kill humans and many other species . Acidianus hospitalis has many adaptations that allow it to survive in its extreme environment . For example , the membrane that surrounds its cells has a different structure to the membranes that surround the cells of most other species . Membranes are made of molecules known as lipids . Generally these lipids assemble into two distinct layers ( known as a bilayer ) to form the membrane . However , in A . hospitalis the membrane contains only a single layer of lipids that resembles a bilayer in which lipids in opposite layers have fused together to make longer molecules . A virus known as AFV1 is able to infect A . hospitalis . Like many other viruses , AFV1 steals part of its host cell’s membrane when it leaves the cell in search of new cells to infect . This stolen membrane helps to protect the virus from its surroundings , however , the structure of the membrane surrounding AFV1 was not known . Kasson et al . combined a technique called cryo-electron microscopy with computer simulations to study the membrane surrounding AFV1 . The study shows that this membrane is only half as thick as the membrane that surrounds A . hospitalis . To make this thinner membrane , flexible lipid molecules from the A . hospitalis membrane bend into a U-shape . These findings reveal a new type of membrane structure not previously seen in the natural world . In the future , this thinner membrane could have many uses in biotechnology , such as to make probes for medical imaging in patients or to deliver drugs to specific sites in the body . Enveloped by this unusual membrane , these structures may be more resistant to the normal processes that degrade and destroy foreign materials in humans and other organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2017	Model for a novel membrane envelope in a filamentous hyperthermophilic virus
The adult mammalian brain is mainly composed of mature neurons . A limited amount of stem cell-driven neurogenesis persists in postnatal life and is reduced in large-brained species . Another source of immature neurons in adult brains is cortical layer II . These cortical immature neurons ( cINs ) retain developmentally undifferentiated states in adulthood , though they are generated before birth . Here , the occurrence , distribution and cellular features of cINs were systematically studied in 12 diverse mammalian species spanning from small-lissencephalic to large-gyrencephalic brains . In spite of well-preserved morphological and molecular features , the distribution of cINs was highly heterogeneous , particularly in neocortex . While virtually absent in rodents , they are present in the entire neocortex of many other species and their linear density in cortical layer II generally increased with brain size . These findings suggest an evolutionary developmental mechanism for plasticity that varies among mammalian species , granting a reservoir of young cells for the cerebral cortex . Structural changes occurring in the adult brain are important for physiological plasticity ( adaptation to changing environment ) , protection against age-related dysfunction ( e . g . , dementia ) , and possibly brain repair ( Martino et al . , 2011; Bond et al . , 2015; Bao and Song , 2018 ) . Brain structural plasticity consists of synaptic plasticity ( synapse formation/elimination; Forrest et al . , 2018 ) and genesis of new neurons driven by neural stem cells ( Aimone et al . , 2014; Lim and Alvarez-Buylla , 2016 ) . The latter process , known as adult neurogenesis , is widely present in the brain of non-mammalian vertebrates ( including most of the telencephalon; Ganz and Brand , 2016 ) and becomes spatially restricted to a few , subcortical neurogenic niches in mammals ( Bond et al . , 2015; Lim and Alvarez-Buylla , 2016; Aimone et al . , 2014; Feliciano et al . , 2015 ) . Adult neurogenesis is substantially absent in the neocortex , with only a very small amount of postnatal addition of interneurons described in mice ( Dayer et al . , 2005 ) . The highly expanded neocortex of large-brained mammals is located far from the neurogenic sites . Additionally , the mammalian neocortex requires substantial structural stability , which is thought to be related to the retention of long-term memories ( Parolisi et al . , 2018; Koketsu et al . , 2003 ) . Recently , attention has been focused on a different population of cortical cells that might also be involved in plasticity , the ‘immature’ neurons ( cINs; Gómez-Climent et al . , 2008; Piumatti et al . , 2018; Rotheneichner et al . , 2018; Benedetti et al . , 2020 ) , which are generated prenatally , but continue to express typical markers of immaturity during adulthood , including doublecortin ( DCX; Gómez-Climent et al . , 2008; Luzzati et al . , 2009 ) and ‘polysialylated’ or ‘embryonic’ Neural Cell Adhesion Molecule ( PSA-NCAM; Seki and Arai , 1991 ) . The cINs can progressively mature through the lifespan , ultimately losing the markers for immaturity ( Rotheneichner et al . , 2018; Benedetti et al . , 2020 ) . They are considered as a potential reservoir of young , plastic neuronal phenotypes ( Piumatti et al . , 2018; Bonfanti and Nacher , 2012; La Rosa et al . , 2019 ) , which might represent a form of slow , delayed neurogenesis ( ‘neurogenesis without division’ ) if ultimately integrated into circuits ( Rotheneichner et al . , 2018; Benedetti et al . , 2020; König et al . , 2016 ) . These immature cells were initially discovered in cortical layer II of the mouse and rat piriform cortex ( reviewed in Bonfanti and Nacher , 2012 ) . While in rodents they are confined to the paleocortex ( three-layered allocortex ) , some mammals also host them in the neocortex ( six-/five-layered isocortex; Luzzati et al . , 2009; Zhang et al . , 2009; Cai et al . , 2009 ) . Current knowledge on this cell population is fragmentary , due to the complexity of systematic studies involving large-sized and wild animal species ( Bonfanti and Nacher , 2012; La Rosa et al . , 2019 ) . The recent qualitative observation that cINs are widespread in the cerebral cortex of sheep , a mammal endowed with a relatively large and gyrencephalic brain ( Piumatti et al . , 2018 ) , invites the hypothesis that they might be important more generally in mammalian species with larger brain sizes than rodents’ ( Piumatti et al . , 2018; Palazzo et al . , 2018 ) . To test this idea , here we systematically studied the occurrence , anatomical distribution , morphology , protein expression profile , maturity/immaturity state , and density ( linear density: number of DCX+ cells/mm of cortical layer II ) of cINs in the cerebral cortex of 80 brains across a phylogenetic diversity of eutherian mammals ( spanning three of the four mammalian superorders; Nishihara et al . , 2009 ) which are widely different in their neuroanatomy ( brain size , gyrencephaly , encephalization; Figure 1—figure supplement 1 and Supplementary file 2 ) and other life history and socioecological features ( lifespan , habitat , food habit ) . Though postmortem interval ( PMI; the time between death and fixation of the brain ) , fixation procedure and other conditions can influence the detection of DCX+ cells in brain tissues , as described for neurogenesis in the hippocampus ( Chawana et al . , 2020; Kempermann et al . , 2018 ) , no substantial variation was observed in our specimens in the quality and intensity of DCX staining of cINs ( Figures 2 and 3 ) , regardless of slight differences in the fixation procedures across specimens in the sample ( Supplementary file 1 ) . The PMI length , considered as one of the main limits to brain tissue quality ( Moreno-Jiménez et al . , 2019 ) , was generally very short ( between a few minutes and 1 hr in all specimens analysed here , apart from chimpanzee , which was less than 14 hr; Supplementary file 1 ) . In order to confirm that we were localizing the cINs as they have been described previously in rodents and sheep ( Gómez-Climent et al . , 2008; Rubio et al . , 2016; Piumatti et al . , 2018; reviewed in Bonfanti and Nacher , 2012; König et al . , 2016 ) in all species studied , unbiased by differences in fixation procedure , we considered multiple features of this neuronal population , such as: i ) the morphology of the DCX+ cells ( type 1 and type 2 cells , based on cell soma size and dendritic arborisation ) ; ii ) the staining of other markers ( PSA-NCAM , NeuN ) ; iii ) the possible co-expression with markers for cell proliferation . In addition , we checked for the occurrence of staining for all markers in other anatomical regions of the same animals where DCX expression is known to be at a high level: the SVZ and hippocampal neurogenic zones ( Figure 3A ) , and the piriform cortex ( Figure 2—figure supplement 1 ) . The anatomical distribution of the DCX+ neurons was investigated in both paleo- and neocortex ( Figure 2B ) . DCX expression was consistently found in cells located in the upper part of layer II of the paleocortex , in all the species considered , at all ages evaluated ( Figure 2—figure supplement 1 ) . In rodents ( mouse and naked mole rat ) , only occasional DCX+ neurons were detectable in layer II of the neocortex . In other small-brained , lissencephalic species ( sengi and bats ) they were also found in the lateral part of the neocortex , with variable distribution along the brain dorsal-ventral axis . In all other mammals in our study , DCX+ neurons were observed throughout the entire neocortex ( Figure 2B ) . Spatial distributions were rather similar in the different anterior-posterior brain levels ( Figure 4—figure supplement 1A ) . The morphology of the layer II DCX+ cells fell into the two main types in all mammals investigated , as reported previously ( type 1 , small cell soma , bipolar; type 2 , large cell soma with ramified dendrites; Figure 2D , E; Piumatti et al . , 2018; Bonfanti and Nacher , 2012; König et al . , 2016 ) . The cell soma size range was 2–9 µm in diameter for type 1 cells , and 7–17 µm for type 2 cells ( ranges for each species in Figure 2E ) . Since type 2 cells are known to be less immature than type 1 ( Piumatti et al . , 2018; Rotheneichner et al . , 2018 ) , each was counted separately ( Figure 2D ) . In all mammals considered , type 1 cells were more abundant than type 2 cells , the latter representing 3–16% of the total DCX+ cells ( apart from rodents , Figure 2E ) . Among rodents , mice showed the highest percentage of type 2 cells ( around 40% ) and naked mole rats the lowest ( 1–4% ) ( Figure 2E ) . On the whole , morphology and cell type proportions of the layer II DCX+ neurons were rather constant in the different species and ages studied , whereas their anatomical distribution in the neocortex was variable across phylogeny . We examined staining for markers of cell proliferation ( Ki-67 antigen; PCNA in bats ) in adult cerebral cortex samples from each brain level , with particular reference to layer II and DCX+ neurons . In rabbit and sheep , bromodeoxyuridine ( BrdU ) was injected in adult animals to determine whether DCX+ cells could have been generated in adulthood . The periventricular ( SVZ ) and hippocampal ( SGZ ) neurogenic zones were used as internal , positive controls ( Figure 3A ) . No DCX/Ki-67 or DCX/BrdU double-labelled cells were ever observed in the cortex of any of the species investigated ( Figure 3A ) . Only rare Ki-67 , PCNA or BrdU immunoreactive nuclei were detectable in the cortex ( with an average number of proliferating cells from 0 to 5 cells for the cortical area considered per cryostat section analysed ) , never in association with DCX+ cells ( Figure 3A and Figure 3—figure supplement 1 ) . These proliferating nuclei were identified as glial cells in double staining carried out for astrocytic and oligodendrocytic glial markers ( Figure 3A; Selinfreund et al . , 1991; Boda et al . , 2015 ) . Two additional markers were used in association with DCX to assess the neuronal maturational stage of the cINs in mouse , cat , rabbit , and marmoset ( Figure 3B; Piumatti et al . , 2018 ) : PSA-NCAM , a low-adhesive form of N-CAM widely present in neurons during the development of the nervous system ( Bonfanti , 2006 ) and expressed by cells retaining plasticity during adulthood ( Hoffman et al . , 1982; Bonfanti et al . , 1992; Bonfanti and Nacher , 2012 ) ; NeuN , an RNA-binding protein expressed by postmitotic neurons that start differentiation ( Mullen et al . , 1992 ) , which can identify most types of mature neurons with some exceptions ( Gusel'nikova and Korzhevskiy , 2015 ) , and is expressed in type 2 cINs ( Piumatti et al . , 2018; referred to as ‘complex cells’ in Rotheneichner et al . , 2018 ) . Across the mammalian species in our sample , only 17–18% of the DCX+ cells co-expressed NeuN , with the NeuN+/DCX+ neurons ( Figure 3C ) mostly characterized by the type 2 cell morphology , in accord with previous observation in mouse and sheep ( Piumatti et al . , 2018; Rotheneichner et al . , 2018 ) . About 14–39% of the DCX+ cells were immunoreactive for PSA-NCAM ( with no particular relation with cell morphology; Figure 3C ) . We assessed the numbers of DCX+ cells per mm of layer II perimeter ( linear density: calculated in the cerebral cortex , in paleocortex and neocortex; Figure 4—figure supplement 1B ) , at the four brain levels ( L1–L4 ) at the ages shown in Figure 1C ( total number of brains analysed = 80 ) . Linear density was measured since cINs are arranged in a monolayer-like row within cortical layer II ( hosting all DCX+ cells of the cortex ) . Such density , calculated on the real cortical layer II length measured in entire brain coronal sections ( highly varying in different species and ages ) , represents a comparable value , allowing inferences across different mammals . The total count of DCX+ cells was performed in each coronal section ( total number of sections = 960 ) from both paleocortex and neocortex , in order: i ) to establish in each species , the exact anatomical location of the cINs in the cortex ( considering two extremes , in chimpanzees , a total of 1774 . 25 cm of cortical layer II were evaluated with an average of 18 . 88 cm for each cryostat section , while in mice , 167 . 14 cm of cortical layer II were evaluated with an average of 0 . 86 cm for each cryostat section; ii ) to identify each cell as belonging to either type 1 or type 2 morphology ( total number of cells counted = 414 . 008; Figure 2D and Figure 1—figure supplement 1 ) . The number of layer II cortical DCX+ cells was investigated in all species for which a brain hemisphere was available from 4 individuals ( n = 10 species; qualitative analysis only was performed on two additional species: sengis and horses ) . By comparing pre-puberal specimens to adult ones , in rodents a dramatic decrease in the linear density was found ( nonparametric Mann Whitney test , p<0 . 01; data referred to cerebral cortex; Figure 4A ) . In rabbit , a slight age-related decrease was observed ( nonparametric Mann Whitney test , p<0 . 05 ) , whereas in sheep no significant differences were detectable between the two age groups . These data suggest that maturation of the cINs , while rapidly occurring at young ages in rodents ( thus eroding the reservoir of immature cells ) , progressively slows in a larger brained mammal , the sheep , leaving a greater remaining population of immature neurons in the adult . To define the abundance of this cINs reservoir , we quantified the cortical layer II DCX+ cells in adult specimens from a diverse group of species ( pooled across three stages: young-adult , middle age and aged; Figure 1C ) . A high degree of interspecific variability in the density of cINs was found in cerebral cortex , with notable differences between paleocortex and neocortex ( Figure 4B ) . When data were organized in two groups ( Figure 1C and Supplementary file 2 ) , to compare small ( brain size range 0 , 5 to 12 g ) to large brains ( brain size range 30 to 384 g; which also have more neocortical surface; Supplementary file 3 ) , a significant higher density was observed in the latter group with respect to the former ( nonparametric Mann Whitney test , p<0 . 0001; Figure 4—figure supplement 2A ) . When considering the neocortex and paleocortex separately , the same trend in DCX+ cell linear densities was observed ( see Figure 4C , D and Figure 4—figure supplement 2B , C ) . The linear density in paleocortex was higher than in neocortex for all species ( a significant correlation in densities from paleo and neocortex was present - nonparametric Spearman correlation , p<0 . 001; Figure 4—figure supplement 3 ) ; variation among species is more evident in neocortex ( median linear density in neocortex in cat is 17 . 58 cells/mm and in mouse 0 , while in paleocortex in cat is 58 . 41 cells/mm and in mouse 1 . 18 cells/mm; Figure 4C and Figure 4—figure supplement 2B ) . To investigate if the occurrence of cINs might be linked to other species-specific factors , the linear density of DCX+ cells in cerebral cortex was correlated with encephalization quotient and lifespan , but no correlations were found ( not shown ) . To investigate variance in density of DCX+ neurons in relation to brain weight , neocortical surface area , layer II perimeter , brain length , we performed a Principal Component Analysis ( PCA ) . The first principal component explained 78% of the variance and the second principal component explained 19% of the variance . In particular , measures of brain size contributed more to the loading of the first component , whereas density of DCX+ neurons loaded most on the second component , showing that species grouped according to these biological features ( Figure 5B and Discussion ) . To determine the scaling relationships in our datasets , phylogenetic generalized least squares ( PGLS ) regression analysis was performed . There was a significant relationship between linear density of DCX+ neurons and species mean brain weight ( adjusted r2 = 0 . 618 , p=0 . 02 ) , with a moderate phylogenetic signal ( Pagel’s lambda = 0 . 24 ) ( Figure 5D ) . There was also a significant relationship between DCX+ neuron density and layer II perimeter as measured from the same brains in our sample ( adjusted r2 = 0 . 595 , p=0 . 03 , Pagel’s lambda = 0 . 00 ) . The relationship between DCX+ neuron density and gyrification index approached significance ( adjusted r2 = 0 . 328 , p=0 . 10 , Pagel’s lambda = 0 . 00; Figure 5—figure supplement 1 ) . Finally , to determine whether the distribution of DCX+ immature neurons in cortical layer II might be heterogeneous through the anterior-posterior extension of the brain , the neocortical linear densities obtained in the four coronal levels ( L1-L4 ) were compared in each species . Two-way ANOVA with Bonferroni post-hoc tests found no significant differences among brain levels in any species , except for cats ( lower cell density in L4: L1 vs L4: p<0 . 001; L2 vs L4: p<0 . 001; L3 vs L4: p<0 . 01; Figure 4E ) . In a heatmap analysis , animals belonging to the same orders ( chimpanzee and marmoset , fox and cat , mouse and NMR ) were clustered together ( Figure 4F ) . The linear density found in large-brained mammals ( chimpanzee , fox , sheep , cat ) was as much as one order of magnitude higher in comparison to the species with small , lissencephalic brains ( with the only exception of rabbits , which are known to display unexplained high levels of structural plasticity; Luzzati et al . , 2003; Luzzati et al . , 2006; Ponti et al . , 2008 ) . Such an increase is far more evident in the neocortex: if considering an estimation of the absolute number of cINs , a nearly 2 million-fold difference emerges between mouse and chimpanzee ( Figure 5A , right , and Supplementary file 4 ) . The residual variance in these relationships may be due to differences in various factors that we were not able to control in the sample , such as rearing history or lifestyle of the animals ( e . g . , captivity or wild ) . Our current results , nevertheless , demonstrate that mammals show considerable variability in numbers of cINs across their cerebral cortices and that densities tend to generally be associated with brain size . The finding of a greater immature neuron population present in the neocortex of some large-brained mammals reveals a reservoir of undifferentiated cells in an expanded brain region characterized by higher computational capacities ( Roth and Dicke , 2005; Roth , 2015; Zilles et al . , 2013 ) . The occurrence of more inter-individual variations in large-brained , gyrencephalic mammals suggests that having a greater reservoir of cINs might favor the possibility of their modulation through life . Future studies focused on single species involving groups of animals kept in different , highly controlled , environmental conditions for long periods of time are required to reveal if external factors can modulate the population ( reservoir ) of cINs in individuals . No correlations emerged by considering lifespan or categorical aspects of ecological specializations , such as habitat or dietary preferences . Considering parameters linked to brain size and allometric scaling , PCA confirmed that species tend to be clustered along two main axes of variance that separate on the basis of brain size measures and cIN density , respectively ( Figure 5B ) . In addition , PGLS regression analysis indicates that linear density of DCX+ neurons covaried with brain weight , layer II perimeter , and gyrification index ( Figure 5D ) . The fact that cINs appear to increase significantly in association with mammalian brain size among phylogenetically divergent taxa such as primates , artiodactyls , and carnivores suggests the independent evolution of this phenotype . Aside from their cortical distribution ( in terms of anatomical distribution within the entire neocortex of each species ) and density , other features of the cINs investigated here ( morphology , occurrence and relative amount of cell types , degree of maturity/immaturity , non-proliferative state ) were substantially similar regardless of the species considered ( Figure 5A , left ) . Even the spatial distribution within the cortex ( considering the anterior-posterior brain levels in the neocortex of each species ) ; Figure 5A , left ) , showed no substantial variation , thus indicating that there is not likely a strong link between occurrence of cINs and specific functional cortical areas . In a heatmap analysis cats were the only species to show minor differences among anatomical levels that we sampled in the brain , the disparity being limited to a drop in cell density in the occipital region ( Figure 4E , F ) . Finally , no substantial differences in cIN density and distribution were found by comparing species with five-layered ( e . g . , sheep , which lacks layer IV and is generally accompanied by expansion of layers II and III; Cozzi et al . , 2017 ) , and six-layered cortex ( the remaining species ) . Accordingly , supragranular layer II ( but not layer IV ) persists through the evolution of the mammalian brain independently of the organization of the cortex in five or six layers ( Cozzi et al . , 2017 ) . On the whole , the analyses carried out in the present study indicate that cINs are a cell population with a set of phylogenetically conserved features independent from cortical anatomy and its functional specializations , yet , with increased distribution in mammals with enlarged neocortices . Our results strongly suggests that this trait and the mechanisms and processes that are underpinning it has evolved independently several times in mammals ( with convergent gains and losses ) . Our sampling regime encompassed both large and small brained species in three of the four eutherian mammal superorders: Laurasiatheria , Euarchonoglires and Afrotheria , thought to have initially diverged in the Cretaceous Period ( Nishihara et al . , 2009 ) . As such , it would be of great interest to establish whether common pathways have been selected to produce the phenotypes we observe . Whether and how the abundance of cINs in large-brained mammals could be linked to cortical function and higher-order cognitive abilities merits further investigation . The complex relationship between expansion of the brain and the increase of computational capabilities remains to be understood ( Healy and Rowe , 2007 ) . Nevertheless , the overall phylogenetic distribution of cINs suggests reconsideration of the mechanisms of plasticity in large-brained mammals . The mammalian brain has low capacity for cell renewal , with most neurons being lifelong , mature elements . The exception represented by adult stem cell niche-depending neurogenesis is spatially restricted ( Bond et al . , 2015; Bao and Song , 2018; Forrest et al . , 2018; Lim and Alvarez-Buylla , 2016 ) , does not serve the neocortex and is thought to be reduced in large-brained species ( Sanai et al . , 2011; Paredes et al . , 2016 ) , the issue being still debated for the hippocampus ( Kempermann et al . , 2018; Petrik and Encinas , 2019 ) . The cINs , as a special type of undifferentiated cells generated before birth but retaining molecular profiles of immaturity ( Bonfanti and Nacher , 2012; König et al . , 2016 ) , share features with very young , highly plastic neurons which are still capable of remarkable structural changes ( i . e . newborn neurons; Bonfanti and Nacher , 2012; Brown et al . , 2003 ) . We suggest that cINs may serve as an important reservoir of undifferentiated cells in large-brained mammals . The relative occurrence of elements showing higher or lesser degrees of immaturity on the basis of their morphology and cell marker expression ( as previously described by Piumatti et al . , 2018; Rotheneichner et al . , 2018; Benedetti et al . , 2020 ) , further supports this view: type 2 cells ( the less immature elements; Piumatti et al . , 2018; Rotheneichner et al . , 2018 ) were abundant in mice ( animals with short lifespan and fast metabolism ) , whereas type 1 were more prevalent in naked mole rats ( neotenic mammals retaining features of immaturity for their extended lifespan; Penz et al . , 2015 ) ; all other species retain substantial amount of type 1 cells even at advanced ages . Type 1 cells , as highly immature neurons ( Piumatti et al . , 2018; Rotheneichner et al . , 2018; Benedetti et al . , 2020 ) , might retain a phenotype that permits a form of plasticity in large expanded ( relatively ‘stable’ ) neocortices in terms of disposable undifferentiated cells . In that sense , cINs should not be considered as an alternative to canonical neurogenesis , rather a parallel form of plasticity providing undifferentiated neurons , in the absence of cell division , in a region of the mammalian brain of utmost importance for cognition , not endowed with much capacity for neurogenesis in adulthood . According to a recent theory on the origin of the neocortex ( Aboitiz and Montiel , 2015 ) , neurogenesis persists in evolutionary ‘old parts’ of the mammalian brain linked to olfaction ( archicortex: olfactory bulb and hippocampus ) , which were of paramount importance in ancient mammals . These systems were subsequently replaced/integrated by the expansion of the isocortex as a ‘multimodal interface’ for behavioral navigation based on other sensory modalities ( vision and audition ) recruited into the expanding neocortex and contributing to multimodal association networks ( Aboitiz and Montiel , 2015 ) . As a result , larger mammalian brains are more composed of cortex , ranging from under 20% in relative volume in rodents to over 80% in humans ( Hofman , 1989 ) . In this evolutionary context , the cINs might represent an option for providing a reservoir of undifferentiated neurons in a brain structure not served by adult neurogenesis . It is poorly understood why these cells are restricted to layer II . Superficial layers ( II and III ) are involved in integrating corticocortical information and in associative learning , with respect to deep cortical layers mainly linked to subcortical structures . Accordingly , the proportion of cortex they occupy is largest in primate species and smallest in rodents , indicating difference in importance devoted to corticocortical connectivity across mammals ( Hutsler et al . , 2005 ) . In addition , neocortical histological organization develops in a sequence with pyramidal neurons of the deepest layers generated first and neurons exiting the stem cell pool later migrating to the more superficial layers ( McConnell , 1995 ) , a feature possibly linked to an extended maturational time of large brains . Finally , under the profile of their evolutionary origin , upper layers of the neocortex are thought to come through co-option of the olfactory cortex ( Luzzati , 2015 ) . On these bases , superficial layers might be more suitable to retain a reservoir of undifferentiated , plastic cells with respect to lower layers more specialized to host extracortical projection neurons . In conclusion , during evolution , the expanded neocortices of large-brained species might have adopted cINs as a reservoir of young cells compatible with their substantial stability and reduced capacity for neurogenesis , independent from singular functional specializations . The persistence of a significant population of immature cells could be part of a neocortical architecture shared by phylogenetically divergent species as a developmental correlate of brain enlargement . The immature neuron population revealed here might represent one of the anatomical substrates of the so called ‘brain reserve’ or ‘cognitive reserve’ which is thought to allow the maintenance of efficient cognitive functions throughout life and to exert a protective effect against aging ( Stern , 2017; La Rosa et al . , 2019 ) . Accordingly , the cIN population might be also important in the progressive maturation of cortical circuitries during postnatal and young ages . The occurrence and distribution of cINs in humans , as well as their possible modulation by physiological and/or environmental conditions in animal models or postmortem human brain tissue , are worthwhile of further investigation . The prevalent occurrence of such a reservoir in neocortices of large-brained mammals will represent a great challenge for future studies . Brains used in this study were collected from various institutions and tissue banks , all provided by the necessary authorizations ( see below and Supplementary file 1 ) . All experiments were conducted in accordance with current EU and Italian laws . Four prepuberal , four young adult , four middle age and four aged mice were analysed . Perfusion was performed under anesthesia ( i . p . injection of a mixture of ketamine , 100 mg/kg , Ketavet , Bayern , Leverkusen , Germany; xylazine , 5 mg/kg; Rompun , Bayer , Milan , Italy; authorization of the Italian Ministry of Health and the Bioethical Committee of the University of Turin; code 1112/2016-PR - courtesy of Annalisa Buffo ) and brains were postfixed for 4 hr . Four prepuberal naked mole rat brains were extracted a few minutes following euthanasia in accordance with Schedule 1 of the Animals ( Scientific Procedures ) Act 1986 , and fixed by immersion , while four young-adult and four middle age specimens were perfused . All brains were postfixed overnight . Four prepuberal and four young-adult female rabbits were used . Rabbits received one daily injection of BrdU ( Sigma; 40 mg/Kg ) for 5 consecutive days and then were killed 10 days after the last injection . Animals were deeply anesthetized ( ketamine 100 mg/kg - Ketavet , -and xylazine 33 mg/ kg body weight - Rompun ) and perfused with fixative ( Italian Ministry of Health , authorization n . 66/99-A ) . Tissues were postfixed for 6 hr . WE bats were captured in Nairobi , Kenya , and SC bats were captured in Kampala , Uganda . Permits and licenses were granted by the National Museums of Kenya and the Uganda National Council of Science and Technology ( No . 024/07/1 ) . Animals were trapped during night and kept in cages for 1–3 days before perfusion . They were deeply anaesthetized with sodium pentobarbital ( Nembutal , 60 mg/ml; 50 mg/kg ) and perfused . Brains were removed and postfixed for 2–18 days . The actual ages of the animals are not known , but they were aged as adults on the basis of the following criteria: the closure of the femoral and humeral epiphyseal plate , the body weight , the forearm length and sexual maturity ( evidence of lactation or pregnancy in female and testis size in male; Gatome et al . , 2010 ) . Eastern rock sengis were caught in Sherman life traps at the Goro Game Reserve , Limpopo Province , South Africa ( Permit 0089-CPM-401–00004 , CITES and Permit Management Office , Department of Environ- mental Affairs , Limpopo Province ) . Tissues were harvested from animals euthanized under projects in accord with the ethics guide-lines of South Africa ( University of Pretoria Clearance EC028-07 ) and the guidelines of the American Society of Mammalogists . Animals were trapped during night and , the next day , they were deeply anesthetized with pentobarbital ( 50 mg/kg ) , perfused and post-fixed for 24 hr ( Slomianka et al . , 2013 ) . Marmoset brains were extracted 1 hr after death and post-fixed for 3 months . The exact ages of the animals are unknown; they were aged as adults by experienced veterinarians , as described for bats . The foxes were euthanized with 5% sodium-thiopental , decapitated and perfused with fixative . Brains were dissected and postfixed in changes of fixative for 3–7 days . Experiments were conducted following the international guiding principles for biomedical research involving animals developed by the Council for International Organizations of Medical Sciences ( CIOMS ) and were also in compliance with the laws , regulations , and policies of the ‘Animal welfare assurance for humane care and use of laboratory animals , ’ permit number A5761-01 approved by the Office of Laboratory Animal Welfare ( OLAW ) of the National Institutes of Health , USA ( Huang et al . , 2015 ) . Four prepuberal and four young-adult sheep were perfused . The brains were dissected out , cut into blocks and post-fixed in the same fixative for 48 hr . Two years old animals received four intravenous injections of BrdU ( 1 injection/day , 20 mg/Kg in 0 . 9% saline; Sigma-Aldrich , France; Piumatti et al . , 2018; Brus et al . , 2013 ) . Three different survival times were analyzed: 1 , 2 and 4 months ( maturation time for neuroblasts in sheep is 1–4 months; Brus et al . , 2013 ) . Four middle age animal brains were collected 20 min after death , fixed and kept in fixative for 1 month . Four young-adult and four middle age cats , and two adult horse brains were extracted post-mortem ( the PMI was less than 1 hr for cats and between 2 and 20 min for horses ) , fixed and kept in the fixative solution for a 1 month ( cat ) and 3 months ( horse ) . Four young-adult and four aged chimpanzee brains from the National Chimpanzee Brain Resource ( www . chimpanzeebrain . org ) were used . Within 14 hr of each subject’s death ( body refrigerated soon after death ) , the brain was removed , immersed in 10% formalin and fixed for 10–14 days . The specimens were collected post-mortem from zoos and primate research centers , maintained in accordance with each institution’s animal care guidelines ( Schenker et al . , 2010 ) . The whole brain hemispheres were cut into coronal slabs ( 1–2 cm thick ) . The slabs were washed in a phosphate buffer ( PB ) 0 . 1 M solution , pH 7 . 4 , for 24–72 hr ( on the basis of brain size ) and then cryoprotected in sucrose solutions of gradually increasing concentration up to 30% in PB 0 . 1 M . Then they were frozen by immersion in liquid nitrogen-chilled isopentane at −80°C . Before sectioning , they were kept at −20°C for at least 5 hr ( time depending on the basis of brain size ) and then cut into 40 μm thick coronal sections using a cryostat or a sliding microtome . Free-floating sections were then collected and stored in cryoprotectant solution at −20°C until staining . Sections were used for histological staining procedures and immunocytochemistry . Histological analyses were performed on Toluidine blue - or cresyl violet - stained sections . For 3 , 3'-diaminobenzidine ( DAB ) immunohistochemistry , sections were rinsed in PBS 0 . 01 M , pH 7 . 4 . Antigen retrieval was performed using citric acid at 90°C for 5 min . After further washing in PBS 0 . 01 M , pH 7 . 4 , the sections were immersed in appropriate blocking solution ( 1–3% Bovine Serum Albumin , 2% Normal Horse Serum , 0 , 2–2%Triton X-100 in 0 . 01 M PBS , pH 7 . 4 ) for 90 min at RT . Following , sections were incubated with primary antibodies ( see below ) for 48 hr at 4°C . After washing in PBS 0 . 01 M , pH 7 . 4 , sections were incubated for 2 hr at RT with secondary antibodies ( Anti-goat , made in horse , 1:250 - BA-9500; Anti-goat made in rabbit , 1:250 – BA-5000; Anti-rabbit , made in horse , 1:250 – BA-1100; Anti-mouse made in horse , 1:250 – BA-2001; Vector Laboratories , Burlingame , CA 94010 ) . Then , sections were washed with PBS 0 . 01 M , pH 7 . 4 and incubated in avidin–biotin–peroxidase complex ( Vectastain ABC Elite kit; Vector Laboratories , Burlingame , CA 94010 ) for 1 hr at RT . The reaction was detected with DAB , as chromogen , in TRIS-HCl 50 mM , pH 7 . 6 , containing 0 . 025% hydrogen peroxide for few minutes and then washed in PBS 0 . 01 M , pH 7 . 4 . Sections were counterstained with Cresyl violet , mounted with DPX Mountant ( Sigma-Aldrich , 06522 ) and coverslipped . For immunofluorescence staining , sections were rinsed in PBS 0 . 01 M , pH 7 . 4 . Antigen retrieval was performed using citric acid at 90°C for 20 min . After further washes in PBS 0 . 01 M , pH 7 . 4 , sections were immersed in appropriate blocking solution ( 1–3% Bovine Serum Albumin , 2% Normal Donkey Serum , 1–2% Triton X-100 in 0 . 01M PBS , pH 7 . 4 ) for 90 min at RT . Then the sections were incubated for 48 hr at 4°C with primary antibodies ( see below ) , and subsequently with appropriate solutions of secondary antibodies for 4 hr at RT: Alexa 488-conjugated anti-mouse ( 1:400; Jackson ImmunoResearch , West Grove , PA - 715-545-150 ) , Alexa 488-conjugated anti-rat ( 1:400; Jackson ImmunoResearch , West Grove , PA - 712-546-153 ) , Alexa 488-conjugated anti-rabbit ( 1:400; Jackson ImmunoResearch , West Grove , PA - 711-545-152 ) , Cyanine 3 ( Cy3 ) -conjugated anti-goat ( 1:400; Jackson ImmunoResearch , West Grove , PA - 705-165-147 ) , Alexa 647-conjugated anti-mouse ( 1:400; Jackson ImmunoResearch , West Grove , PA - 715-605-151 ) . Immunostained sections were counterstained with 4' , 6-diamidino-2-phenylindole ( DAPI , 1:1000 , KPL , Gaithersburg , Maryland USA ) and mounted with MOWIOL 4–88 ( Calbiochem , Lajolla , CA ) . Primary antibodies and dilutions used for this study: goat anti-DCX ( 1:500–3500 , Santa Cruz Biotechnology , Santa Cruz , CA - sc-8066 ) , mouse anti-Ki-67 ( 1:500–1000 , BD Pharmigen - 550609 ) , rabbit anti-Ki-67 ( 1:600–1000 , Leica-Novocastra - NCLKi67p ) , rat anti-BrdU ( 1:300 , AbDSerotec , Kidlington , UK - OBT0030 ) , mouse anti-PCNA ( 1:30000 , Delta Biolabs , Gilroy , Calif - DB095 ) , mouse anti-PSA-NCAM ( 1:1400 , Millipore , Bellerica , MA - MAB5324 ) , mouse anti-NeuN ( 1:300 , Millipore , Bellerica , MA - MAB377 ) , rabbit anti-S100β ( 1:5000 , Swant , Swiss Antibodies , CH - 37A ) ; rabbit anti-Olig2 ( 1:1000 , Millipore , Bellerica , MA - AB9610 ) . The mammalian brains included in the current study differ in terms of brain size , gyrencephaly and overall neuroanatomical organization . In order to perform comparable analyses , four correspondent anterior-posterior , coronal brain levels ( L1-L4 ) were designated ( Figure 2B ) . To find the neuroanatomical landmarks of these brain levels in each species , the entire hemispheres were cut into 40 μm thick coronal sections . Then the corresponding levels were defined as coronal ( thick ) slices involving the same main brain structures: L1 , from anterior opening of the lateral ventricle ( or shift from lateral to olfactory ventricle in rabbit and sheep ) to L2; L2 , from anterior starting of internal capsule to L3; L3 , from anterior starting of the claustrum and/or amygdala to L4; L4 , from posterior closing of the lateral ventricle to an extension equivalent to that of L2 ( same number of 40 μm thick serial sections ) . For species in which neuroanatomy was previously described , existing atlases were used as a reference ( mouse: Allen Institute for Brain Science; marmoset , chimpanzee , rabbit , fox , cat: Comparative Mammalian Brain Collections; sheep: Michigan State University ) then matching the brain levels on our specimens; for the remaining species the procedure was performed by using our histologic specimens ( example given in Figure 2E ) . The paleocortex-neocortex ( allocortex-isocortex ) transition was easily identifiable on the cresyl violet-stained sections ( Figure 2F ) . Images were collected using a Nikon Eclipse 90i microscope ( Nikon , Melville , NY ) connected to a color CCD Camera , a Leica TCS SP5 , Leica Microsystems , Wetzlar , Germany and a Nikon Eclipse 90i confocal microscope ( Nikon , Melville , NY ) . Quantitative analyses were performed using Neurolucida software ( MicroBrightfield , Colchester , VT ) on DCX-DAB stained sections . The linear density of DCX+ cells present in layer II of the cerebral cortex was evaluated in 4 individuals of 10 species ( in each brain level , three sections were considered – one in the anterior , one in the central and one in the posterior part ) . In each section , the total perimeter of cortical layer II was traced and all DCX+ cells along its length were counted ( linear density = number of cells/mm ) . In addition to the linear density for the cortex , also separate densities for neocortex and paleocortex were evaluated . In the same sections , the morphology of cINs was evaluated and the number of type 1 and type 2 cells was counted using different markers selected from the ‘markers toolbar’ in Neurolucida software . The cell soma size ( diameter ) was obtained by evaluating the width orthogonal to main axis , measured in about 100 cells for each animal species using the Neurolucida ‘measure line' tool . In cat , sheep , rabbit , marmoset , NMR and mouse , the number of Ki-67+ nuclei ( PCNA in SC bat ) and the number of Ki67+/DCX+ double-stained cells were counted in an area corresponding to the neocortical upper layers ( I , II , III; Ki-67+ cells/mm2 ) . For each specimen , a cryostat section from each of the 4 brain levels was selected and , in each of them , three microscopic fields ( 40x magnification; 1 dorsal , 1 lateral , 1 medial along neocortical extension ) were analyzed . In rabbits , the same analysis was also performed for BrdU+ nuclei and BrdU+/DCX+ double-stained cells . The percentage of NeuN+/DCX+ and PSA-NCAM/DCX+ double-stained cells was calculated in cat , rabbit , marmoset and mouse by analyzing eight microscopic fields ( 40x magnification; 2 dorsal , 2 lateral , 2 ventral; 2 medial in the cortical perimeter ) from a central section in each of the four levels for each specimen . All images were processed using Adobe Photoshop CS4 ( Adobe Systems , San Jose , CA ) and ImageJ version 1 . 50b ( Wayne Rasband , Research Services Branch , National Institute of Mental Health , Bethesda , Maryland , USA ) . Adjustments to color , contrast , and brightness were made . All graphs and statistical analyses were performed using GraphPad Prism Software ( San Diego California , USA ) using different nonparametric tests: Mann-Whitney test , Kruskal-Wallis test with Dunn’s multiple comparison post test , Two-way ANOVA with Bonferroni post-hoc test and Spearman correlation coefficient r . p<0 . 05 was considered as statistically significant . Median was used as a central measure . Free software environment R was used to calculate and draw heatmaps and perform PCA . In particular for the heatmap , in order to highlight similarities and differences in the linear densities distributions of INs among the four neocortical layers across species , density values were transformed by subtracting the mean value of each species , and pheatmap package ( Kolde , 2019 ) was used with correlation as parameter value for row clustering distance and column as parameter value for scale function argument and PCA was calculated by prcomp package ( Vq , 2011 ) with center and scale parameters set to TRUE . The species median DCX+ neuron densities ( neocortex ) were used to perform an ancestral character state reconstruction of trait evolution mapped onto the phylogeny . This was implemented in Mesquite software , using a parsimony model . To determine the scaling relationships in our dataset we employed PGLS regression with a likelihood-fitted lambda transformation . The species median DCX+ neuron linear densities ( neocortex ) as the main variable of interest were used in these analyses . The PGLS was run against three different predictors - brain weight , layer II perimeter , and gyrification index . All data were log transformed prior to PGLS to fit power functions to linear regression , as is standard procedure . A phylogenetic tree of the species in the sample was downloaded from the TimeTree database ( Kumar et al . , 2017 ) . All regression plots are on a log scale and show the 95% confidence intervals .	To acquire new skills or recover after injuries , the mammalian brain relies on plasticity , the ability for the brain to change its architecture and its connections during the lifetime of an animal . Creating new nerve cells is one way to achieve plasticity , but this process is rarer in humans than it is in mammals with smaller brains . In particular , it is absent in the human cortex: this region is enlarged in species with large brains , where it carries out complex tasks such as learning and memory . Producing new cells in the cortex would threaten the stability of the structures that retain long-term memories . Another route to plasticity is to reshape the connections between existing , mature nerve cells . This process takes place in the human brain during childhood and adolescence , as some connections are strengthened and others pruned away . An alternative mechanism relies on keeping some nerve cells in an immature , ‘adolescent’ state . When needed , these nerve cells emerge from their state of arrested development and ‘grow up’ , connecting with the appropriate brain circuits . This mechanism does not involve producing new nerve cells , and so it would be suitable to maintain plasticity in the cortex . Consistent with this idea , in mice some dormant nerve cells are present in a small , primitive part of the cortex . La Rosa et al . therefore wanted to determine if the location and number of immature cells in the cortex differed between mammals , and if so , whether these differences depended on brain size . The study spanned 12 mammal species , from small-brained species like mice to larger-brained animals including sheep and non-human primates . Microscopy imaging was used to identify immature nerve cells in brain samples , which revealed that the cortex in larger-brained species contained more adolescent cells than its mouse counterpart . The difference was greatest in a region called the neocortex , which has evolved most recently . This area is most pronounced in primates – especially humans – where it carries out high-level cognitive tasks . These results identify immature nerve cells as a potential mechanism for plasticity in the cortex . La Rosa et al . hope that the work will inspire searches for similar reservoirs of young cells in humans , which could perhaps lead to new treatments for brain disorders like dementia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Phylogenetic variation in cortical layer II immature neuron reservoir of mammals
Nanobodies are single-domain antibodies of camelid origin . We generated nanobodies against the vertebrate nuclear pore complex ( NPC ) and used them in STORM imaging to locate individual NPC proteins with <2 nm epitope-label displacement . For this , we introduced cysteines at specific positions in the nanobody sequence and labeled the resulting proteins with fluorophore-maleimides . As nanobodies are normally stabilized by disulfide-bonded cysteines , this appears counterintuitive . Yet , our analysis showed that this caused no folding problems . Compared to traditional NHS ester-labeling of lysines , the cysteine-maleimide strategy resulted in far less background in fluorescence imaging , it better preserved epitope recognition and it is site-specific . We also devised a rapid epitope-mapping strategy , which relies on crosslinking mass spectrometry and the introduced ectopic cysteines . Finally , we used different anti-nucleoporin nanobodies to purify the major NPC building blocks – each in a single step , with native elution and , as demonstrated , in excellent quality for structural analysis by electron microscopy . The presented strategies are applicable to any nanobody and nanobody-target . Nanobodies represent antigen-binding domains of 'heavy-chain-only' camelid antibodies and are typically selected by phage display from an immune library ( Hamers-Casterman et al . , 1993; Arbabi Ghahroudi et al . , 1997; Muyldermans , 2013 ) . Their small size ( ~13 kDa ) , monoclonal nature and high specificity are ideal for applications like affinity purification or protein detection and localization ( Helma et al . , 2015 ) . Their utility as crystallization chaperones is also widely appreciated ( Pardon et al . , 2014; Desmyter et al . , 2015 ) . Nanobodies are commonly expressed in Escherichia coli and secreted into the oxidative periplasm , where their conserved internal disulfide bond can be formed ( Vincke et al . , 2012; Pardon et al . , 2014; Fridy et al . , 2014 ) . Periplasmic expression comes , however , with several drawbacks . For example , it often results in low final yield ( Baneyx and Mujacic , 2004 ) , probably due to saturation of the secretion machinery and aggregation of precursor proteins in the cytoplasm . The limited set of chaperones and high proteolytic activity in the periplasm also restrict the choices of fusion tags that can be used ( de Marco , 2009; Feilmeier et al . , 2000 ) . Furthermore , the purification of periplasmic proteins involves considerably more hands-on time than purification from the cytoplasm . In typical affinity chromatography applications , nanobodies are covalently attached to a resin , which later necessitates harsh conditions for the elution of bound target proteins ( Rothbauer et al . , 2008; Fridy et al . , 2014 ) . This is appropriate for an identification , but hardly for any further downstream structural or functional analysis of the purified target proteins . As a workaround , a native isolation of protein A-tagged protein complexes using a specific nanobody modified with a dithiothreitol ( DTT ) -cleavable crosslinker was recently reported ( Fridy et al . , 2015 ) . However , the achievable yield was modest , as most of the isolated complexes resisted release . Furthermore , the presence of any thiol-reducing agent ( like DTT or glutathion ) during binding is incompatible with this method . In traditional indirect immunofluorescence , epitopes are initially decorated with a primary antibody and detected with a fluorophore-labeled secondary one , each around 12–15 nm in size ( Harris et al . , 1998 ) . The effective displacement between label and epitope can reach up to 24–30 nm and thus significantly deteriorate the achievable precision and accuracy of protein localization by super-resolution fluorescence microscopy ( Hell , 2009; Huang et al . , 2009 ) . Nanobodies ( diameter: 4 nm ) are an ideal solution to this problem ( Ries et al . , 2012; Szymborska et al . , 2013 ) . This , however , requires a direct nanobody labeling . Ideally , labeling should be site-specific , so that the remaining small displacement between epitope and fluorescent dye can be predicted and corrected for in the measurements . So far , nanobodies were labeled at lysines by N-hydroxysuccinimide ( NHS ) ester fluorophores ( Ries et al . , 2012; Fridy et al . , 2014 ) , which is random and rarely quantitative . As we show below , it also deteriorates signal-to-background ratios or even completely abolishes epitope recognition . A workaround to this basic problem was the addition of a C-terminal oligo-lysine stretch to divert labeling from nanobody framework lysine residues ( described for the anti-GFP nanobody 'Enhancer' in Platonova et al . , 2015 ) . This , however , increases the epitope-label distance again . Furthermore , fluorescent labeling of nanobodies using Sortase A was presented ( Witte et al . , 2012 ) . This strategy is limited to the N- or C-terminus and uses modified fluorophores that are not readily available . Adding an extra C-terminal cysteine ( for subsequent maleimide modification ) to a periplasmically expressed nanobody was also not a satisfying solution , because it led to a severe reduction in yield and caused extensive dimerization ( Massa et al . , 2014 ) . Hence , we explored solutions to the above-described limitations of the current nanobody technology . We demonstrate functional cytoplasmic expression of nanobodies with protease-cleavable tags for native affinity purification and with engineered cysteines for site-specific fluorescent labeling . We chose the Xenopus nuclear pore complex ( NPC ) as a model target and developed a toolbox of high-affinity nanobodies against its major constituent proteins , nucleoporins ( Nups ) , which occur in large subcomplexes . Using specific nanobodies , we purified their target protein complexes from Xenopus egg extract in a single step with native elution based on proteolytic matrix-release . This allowed a direct analysis of nanobody-purified endogenous Nup complexes by negative stain electron microscopy . Labeling these anti-Nup nanobodies with NHS ester fluorescent dyes for imaging often produced non-functional reagent or significant background staining . We therefore implemented a simple and generally applicable strategy for obtaining site-specifically fluorophore-labeled nanobodies of superior imaging quality . It involves engineered cysteines at the nanobody surface , their modification with maleimide fluorophores , and leaves the internal framework cysteines fully intact . This strategy allowed super-resolution imaging of NPCs with a negligible label displacement and very low background . A novel strategy for rapid mapping of conformational nanobody epitopes via crosslinking mass spectrometry involving the engineered surface cysteines is also presented here . In order to provide new tools for studying NPCs , we generated nanobodies against constituents of the Xenopus NPC scaffold , namely Nup85 , Nup93 , and Nup155 , as well as against Nup98 and the Nup62•Nup58•Nup54 complex . The latter two species were included because their Phe-Gly ( FG ) -repeat domains form a permeability barrier within the central NPC channel ( Hülsmann et al . , 2012 ) . High-affinity nanobodies against all these targets were readily obtained from alpaca immune libraries by phage display . We noticed that these nanobodies could be produced in the cytoplasm of various E . coli strains , as recently reported for other nanobodies ( Olichon and Surrey , 2007; Zarschler et al . , 2013; Djender et al . , 2014 ) . We also observed that fusing nanobodies behind a His14-bdNEDD8 module ( Frey and Görlich , 2014 ) increased their yield dramatically ( Figure 1a ) . 10 . 7554/eLife . 11349 . 003Figure 1 . Affinity and thermostability of reduced and oxidized nanobodies . ( a ) Comparison of typical yields for the anti-Nup93 nanobody TP179 and the anti-Nup98 nanobody TP377 expressed either in the Escherichia coli BLR periplasm with a C-terminal His6-tag or in the oxidative cytoplasm of E . coli SHuffle with an N-terminal His14-bdNEDD8-tag . ( b ) Analysis of disulfide bond content using a maleimide shift assay . Anti-Nup93 nanobody TP179 and anti-Nup98 nanobody TP377 , expressed either in the oxidative periplasm of E . coli BLR , the oxidative cytosol of E . coli SHuffle or in the reductive cytoplasm of E . coli BLR , were subjected to modification with biotin-PEG23-maleimide in SDS–PAGE sample buffer ( -DTT ) and analyzed by non-reducing SDS–PAGE followed by Coomassie staining . ( c ) The redox state of the anti-Nup98 nanobody TP377 does not affect the affinity for its target . Biotinylated His14-Avi-bdSUMO-tagged Nup98716-866 was immobilized on Streptavidin agarose und used to bind the reduced GFP-tagged TP377 . Binding was in the absence or presence of an equimolar amount or fivefold excess of nanobody competitor , namely untagged TP377 produced either in the oxidative periplasm , in the mildly oxidative cytoplasm of E . coli SHuffle or in the reductive cytoplasm of BLR . Bound nanobodies were then eluted by proteolytic cleavage of the bdSUMO tag of Nup98 and analyzed by SDS–PAGE followed by Coomassie staining . Note that the oxidized , disulfide bond-stabilized nanobody ( produced in the periplasm ) behaved like the reduced variant ( produced in the E . coli BLR cytoplasm ) . Formation of the disulfide bond therefore does not seem to significantly contribute to the overall affinity . ( d ) Differential scanning fluorimetry ( thermofluor , Niesen et al . , 2007 ) analysis of nanobodies expressed in the oxidative periplasm ( red ) or the reductive cytosol ( blue ) of E . coli BLR . The anti-Nup93 and anti-Nup98 nanobodies were heated in the presence of Sypro Orange dye from 30 to 100°C and thermal unfolding curves were obtained . The melting temperature is derived from the inflection point of the curve . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 003 In order to test whether disulfide bond formation can occur in the cytoplasm and if this is important for nanobody function , we expressed an anti-Nup93 and an anti-Nup98 nanobody either in the periplasm , the reductive cytoplasm of E . coli BLR or in the cytoplasm of E . coli SHuffle . The latter strain contains a cytoplasmic disulfide isomerase and harbors mutations that render its cytoplasm ( moderately ) oxidative ( Lobstein et al . , 2012 ) . The obtained nanobody variants were then treated with biotin-PEG23-maleimide under denaturing conditions . Reduced nanobodies are thereby modified at their free cysteines , and the resulting size shift distinguishes them from disulfide-containing nanobodies ( Figure 1b ) . While periplasmic secretion resulted in fully oxidized nanobodies , only a fraction of the SHuffle-expressed nanobodies contained a disulfide bond . Cytoplasmic expression in E . coli BLR yielded completely reduced nanobodies . One could assume that the antigen affinity of nanobodies is negatively affected by a loss of their scaffold disulfide bond . A competition for antigen-binding revealed , however , no affinity difference between reduced and disulfide bond-containing anti-Nup98 nanobodies ( Figure 1c ) . As expected , we observed by differential scanning fluorimetry ( Niesen et al . , 2007 ) a decreased thermostability of fully reduced anti-Nup98 and anti-Nup93 nanobodies ( Figure 1d ) . Their melting temperatures of 47°C and 57°C are , however , still well above any reasonable incubation temperature for downstream applications . All nanobodies that we obtained via phage display against a wide range of antigens could be functionally produced in the E . coli cytoplasm . Only very few of those nanobodies contained a second pair of cysteines that can form an additional , solvent-exposed disulfide bond between the antigen-binding loops CDR II and CDR III which likely contributes to the overall affinity ( Govaert et al . , 2012 ) . However , most biochemical applications as well as imaging techniques like STORM require reducing conditions that disrupt accessible disulfide bonds , making such nanobodies a poor option anyway . Cytoplasmic expression of nanobodies provides a number of advantages . First , the yield often exceeds 100 mg per liter of culture and can be up to 30 times higher as compared to periplasmic expression ( Figure 1a ) . Second , it saves hands-on time , because a cumbersome preparation of a periplasmic fraction is bypassed , and third , a far broader range of fusion modules can be used . We exploited this for affinity purification of endogenous target protein complexes with nanobodies and native elution . For this strategy , we produced His14-Avi- ( GlySer ) 9-SUMOStar- ( GlySer ) 9-nanobody fusions and purified them by Ni2+ chelate affinity chromatography and imidazole elution ( Figure 2—figure supplement 1a ) . The Avi-tag can be biotinylated by cytoplasmic co-expression of the biotin ligase BirA in E . coli ( Schatz , 1993; Beckett et al . , 1999 ) or in vitro using the purified enzyme ( Fairhead and Howarth , 2015 ) . It then mediates binding of the purified nanobody to streptavidin magnetic beads . The interspersed long unfolded Gly-Ser spacers minimize steric hindrance effects . The SUMOStar module is an engineered SUMO variant that cannot be cleaved by endogenous eukaryotic desumoylases but by an engineered SUMOStar protease ( LifeSensors ) , ( Liu et al . , 2008 ) . In combination , these modules allow native elution of nanobody-bound target proteins or protein complexes by cleaving the tag with nanomolar concentrations of the SUMOStar protease . This strategy also provides a purer and more specific end product , because any protein species , which sticks non-specifically to the beads , will not be released . Thus , such highly specific protease elution makes the otherwise crucial control for matrix background-binding ( Marcon et al . , 2015; Mellacheruvu et al . , 2013 ) essentially dispensable . As a proof of principle , we purified five nucleoporin complexes from a Xenopus egg extract to near homogeneity ( Figure 2a and b ) . For each complex we achieved a ≈10 000-fold enrichment within a single native purification step and yields of around 50% . The anti-Nup85 nanobody retrieved the ≈750 kDa nine-membered Y-complex as well as Tpr and Elys as specific but sub-stoichiometric binding partners . We obtained substantial amounts of the complex , namely 50–100 µg from as little as 2 ml egg extract , which initially contained ≈150 µg or ≈100 nM of the complex ( Wühr et al . , 2014 ) . Post-elution with SDS sample buffer indicated a quantitative proteolytic release of the complex from the beads ( Figure 2—figure supplement 1b ) . 10 . 7554/eLife . 11349 . 004Figure 2 . Purification and native elution of NPC subcomplexes with specific nanobodies . ( a ) Schematic representation of the subcomplex organization and relative localization of Nups within an asymmetric unit of the eightfold rotational symmetric vertebrate NPC ( ONM/INM = outer and inner nuclear membrane ) . The nuclear and cytoplasmic rings of the structural NPC scaffold are mainly composed of the Nup107-Nup160 Y-shaped complex ( green ) . The central inner ring of the scaffold is composed of the Nup93 subcomplex ( blue ) . The scaffold is bound to the nuclear envelope via transmembrane Nups and further anchors FG-repeat nucleoporins ( e . g . Nup98 [red] and the Nup62•Nup58•Nup54 complex [brown] ) within the central channel , where they form the permeability barrier . Nups against which nanobodies were raised are highlighted in bold . ( b ) Native purification of major NPC scaffold subcomplexes and FG-repeat nucleoporins from Xenopus egg extract . Biotinylated His14-Avi- ( GlySer ) 9-SUMOStar- ( GlySer ) 9-tagged nanobodies were immobilized on magnetic Streptavidin beads and then incubated with Xenopus egg extract . After washing , nanobodies were gently eluted along with their bound target complexes by SUMOStar protease cleavage . One tenth of the eluates were analyzed by SDS–PAGE and Coomassie staining . All labeled bands were identified via mass spectrometry . The color code represents the subcomplex organization of the NPC as illustrated in ( a ) . A nanobody raised against Escherichia coli Maltose-binding protein ( MBP ) served as a negative control . NPC , nuclear pore complex . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 00410 . 7554/eLife . 11349 . 005Figure 2—figure supplement 1 . Optimization of native protein complex purification using nanobodies . ( a ) SDS–PAGE and Coomassie staining showing the expression of the anti-Nup98 nanobody TP377 carrying a protease-cleavable affinity tag ( His14-Avi- ( GlySer ) 9-SUMOStar- ( GlySer ) 9 ) in the Escherichia coli cytoplasm and its one-step purification using Ni2+ chelate affinity chromatography and imidazole elution . The Avi-tag mediates binding to Streptavidin after biotinylation by the biotin ligase BirA ( Beckett et al . , 1999; Schatz , 1993 ) . ( b ) Analysis of natively purified and remaining bead-bound material . Anti-Nup93 and anti-Nup85 nanobodies were used to purify their respective target complexes from crude Xenopus egg extract . After native elution with SUMOStar protease , the beads were heated in SDS–PAGE sample buffer containing 400 mM urea for 10 min at 97°C . Note that protease cleavage released the cognate complexes very efficiently and that the remaining bead-bound material essentially represents just the non-specific background , cleaved tags and leaked streptavidin . ( c ) Effect of a RanQ69L•GTP wash on FG-repeat Nup purification . FG repeat-bound nuclear transport receptor•cargo complexes were efficiently removed by washing the beads for 10 min at 4°C with 100 µl 1 µM RanQ69L5-180•GTP before elution with SUMOStar protease . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 005 The anti-Nup155 nanobody retrieved Nup155 as a single species . This might appear surprising as Nup155 is thought to contact Nup93 and Nup53/35 within the inner ring of the NPC scaffold ( Hawryluk-Gara et al . , 2005; Hawryluk-Gara et al . , 2008; Sachdev et al . , 2012 ) . We therefore assume that mitotic post-translational modifications transiently suppress interactions between these proteins . We also purified the Nup98•Gle2 and the Nup62•Nup58•Nup54 complex using anti-Nup98 and anti-Nup54 nanobodies , respectively . Here , we included a RanQ69L•GTP wash to release nuclear transport receptors , which otherwise would remain bound to the FG domains of Nup98 or the Nup62 complex ( Figure 2—figure supplement 1c ) . The anti-Nup93 nanobody purified the expected mixture of the two paralogous Nup93•Nup188 and Nup93•Nup205 complexes ( Theerthagiri et al . , 2010 ) , which are also a part of the structurally least understood NPC inner ring . In this case , we analyzed the natively eluted material straightaway by negative stain electron microscopy ( Figure 3a ) . Class averaging revealed characteristically curved α-solenoid fold-like particles , which are known to exhibit conformational flexibility ( Figure 3b ) . The obtained structures were very reminiscent of the hook and eye-shaped structures reported earlier for the Nup188 and Nup205 orthologues from Saccharomyces cerevisiae ( Amlacher et al . , 2011 ) and Myceliophthora thermophila ( Andersen et al . , 2013 ) . This suggests not only that the overall shape of the Nup93 complexes is conserved from fungi to vertebrates , but also that our single-step purification strategy for large protein complexes yields material of sufficient quality for a direct structural analysis . 10 . 7554/eLife . 11349 . 006Figure 3 . Structural analysis of natively purified Nup93 complexes . ( a ) Anti-Nup93 nanobody TP179-purified Nup93•Nup188 and Nup93•Nup205 complexes were subjected to the GraFix procedure ( Kastner et al . , 2008 ) and negative staining for analysis by electron microscopy . ( b ) Gallery of 12 selected class averages of Nup93•Nup188 and Nup93•Nup205 particles . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 006 In order to use anti-Nup nanobodies to image their targets within intact NPCs , we initially modified them with NHS ester fluorophores . We found , however , that such NHS-labeled nanobodies performed remarkably poorly , in particular when far-red fluorophores were used . As documented by the specific examples below , none of the NHS-labeled nanobodies had sufficient probe quality to allow acquisition of STORM images . We therefore explored alternative and more reliable nanobody-labeling strategies . One possibility was to label nanobodies at engineered ( and reduced ) cysteines with maleimides . This , however , posed the risk of modifying also the scaffold cysteines of the IgG-fold , which inevitably would cause an irreversible unfolding of the nanobodies . To address this issue , we incubated reduced nanobodies with biotin-PEG23-maleimide ( Figure 4 ) . After unfolding by urea , the scaffold cysteines became modified at either 37°C , 23°C , or 0°C . In native buffer , however , modification was quantitative only at 37°C , pointing to a transient exposure of the otherwise buried scaffold cysteines ( 'thermal breathing' ) . Importantly , they remained fully protected at 0°C , predicting that maleimide-labeling on ice would be fully selective for engineered surface cysteines . 10 . 7554/eLife . 11349 . 007Figure 4 . Maleimide modification of the internal cysteines of reduced nanobodies upon thermal unfolding . Indicated nanobodies , expressed in the reductive cytoplasm of Escherichia coli BLR , were incubated at the indicated temperatures in the presence or absence of a two-fold molar excess of biotin-PEG23-maleimide ( 1 . 45 kDa ) in buffer . The addition of 6 M urea served as a positive control for maleimide modification of the internal cysteines upon chemical unfolding . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 007 In order to better guide cysteine placement in the nanobody framework , we solved the crystal structure of the anti-Nup98 nanobody TP377 in complex with the globular Nup98 NPC anchor domain ( residues 716–866 ) at 1 . 9 Å resolution ( Figure 5a , Table 1 ) . TP377 contacts its target through all three CDR loops and does not block the absolute Nup98 C-terminus , which anchors Nup98 via Nup96 or Nup88 to the NPC scaffold ( Hodel et al . , 2002; Griffis et al . , 2003; Stuwe et al . , 2012 ) . The internal disulfide bond-forming cysteines Cys22 and Cys96 of TP377 are reduced in the crystal structure . 10 . 7554/eLife . 11349 . 008Figure 5 . Site-specific fluorescent labeling of nanobodies . ( a ) Crystal structure of the Nup98 NPC anchor domain ( Nup98716-866 , blue ) in complex with the anti-Nup98 nanobody TP377 ( beige ) . The three antigen-binding loops ( CDR I-III ) of TP377 are colored red . ( NT = N-terminus , CT = C-terminus ) ( b ) Tested positions of engineered cysteines ( yellow ) illustrated for nanobody TP377 . Antigen-binding loops are shown in red . ( c ) Quantitative labeling of TP377 with Alexa Fluor 488 maleimide . TP377 with cysteines at the indicated positions can be quantitatively labeled with Alexa Fluor 488 maleimide . Labeling introduces a size shift in SDS–PAGE . Detection was either by Coomassie staining or by in-gel fluorescence . ( 3xCys = NT-Cys + S7C + S71C ) ( d ) Digitonin-permeabilized Xenopus XL177 cells were incubated with 10 nM TP377 carrying a single Alexa Fluor 647 molecule at the indicated position . Cells were then washed , fixed , and counterstained with DAPI ( DNA ) . A characteristic nuclear rim stain indicates labeling of NPCs . Note that labeling of TP377 very close to its antigen-binding loops did not perturb binding . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 00810 . 7554/eLife . 11349 . 009Figure 5—figure supplement 1 . Expression and relative affinity of anti-Nup98 nanobody TP377 with engineered surface cysteines . ( a ) Amino acid sequence of anti-Nup98 nanobody TP377 illustrating the positions chosen for mutation to cysteine ( red ) . ( b ) SDS–PAGE and Coomassie staining showing the expression and purification of TP377 with three engineered cysteines ( NT-Cys , Ser7Cys , CT-Cys ) in the Escherichia coli cytoplasm . Single-step purification was performed using Ni2+ chelate affinity chromatography and cleavage using the bdNEDP1 protease . ( c ) Relative affinity of TP377 with different labeling ratio to Nup98716-866 . GFP-tagged TP377 was incubated with biotinylated His14-Avi-bdSUMO-tagged Nup98716-866 immobilized on Streptavidin agarose . For competition , unlabeled , 1x or 3x Alexa Fluor 488-labeled TP377 was added in equimolar amount or fivefold molar excess relative to GFP-TP377 . Bound nanobodies were eluted by bdSEN1P cleavage and analyzed by SDS–PAGE followed by Coomassie staining . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 00910 . 7554/eLife . 11349 . 010Table 1 . Crystallographic data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 010Nup98•Nb TP377 complexaData collectionSpace groupP41Cell dimensions a , b , c ( Å ) 66 . 59 , 66 . 59 , 87 . 90 α , β , γ ( ° ) 90 . 00 , 90 . 00 , 90 . 00Resolution ( Å ) 47 . 00-1 . 90 ( 1 . 95-1 . 90 ) bRsym or Rmerge0 . 128 ( >1 ) bI / σI27 . 7 ( 2 . 6 ) bCompleteness ( % ) 99 . 5 ( 98 . 7 ) bRedundancy27 . 4 ( 27 . 4 ) bRefinementResolution ( Å ) 47 . 00-1 . 90No . reflections Measured823105 Unique30218Rwork/Rfree0 . 167 / 0 . 196No . atoms Protein2176 Water145Wilson B-factor ( Å2 ) 27 . 4R . m . s . deviations Bond lengths ( Å ) 0 . 010 Bond angles ( ° ) 1 . 08Ramachandran statistics ( % ) Favored98 . 0 Allowed2 . 0 Outliers-aA single crystal was used for data collection . bValues in parentheses are for highest resolution shell . We next mutated solvent-exposed small residues ( Gly , Ser , and Ala ) at six alternative positions of the nanobody scaffold to cysteines ( Figure 5b; Figure 5—figure supplement 1a ) . We found that not only all individual mutants , but also nanobodies with up to three ectopic surface cysteines were well expressed and highly soluble in E . coli ( Figure 5—figure supplement 1b ) . Moreover , cysteines on all six positions on our model anti-Nup98 nanobody TP377 could be quantitatively labeled with maleimide fluorescent dyes ( Figure 5c ) . Even nanobodies carrying three fluorophores were readily obtained . Importantly , competitive binding assays indicated that the introduction of ectopic cysteines and their modification did not impair binding to the Nup98 target ( Figure 5—figure supplement 1c ) . Based on the crystal structure , we estimate that fluorophores attached via an N-terminal cysteine or A75C to anti-Nup98 nanobody TP377 can be as close as ~2 nm to the target Nup98 . In order to test nanobodies in imaging , we grew Xenopus laevis XL177 cells on coverslips , digitonin-permeabilized their plasma membranes , incubated them with low nanomolar concentrations ( 1–10 nM ) of labeled nanobody , and fixed them after several washing steps . In this workflow , even nanobodies with fixation-sensitive epitopes could bind their target . We first tested anti-Nup98 nanobody TP377 carrying a single Alexa Fluor 647 maleimide at the six alternative positions ( Figure 5d ) . In confocal laser scanning microscopy , all variants produced a very bright punctuate nuclear rim staining of XL177 cells , characteristic for NPCs , against a very low background . Combining minimal label displacement with ease of cloning , we routinely labeled our nanobodies via an N-terminal cysteine . This way , all chosen NPC targets ( Nup98 , Nup93 , Nup85 and Nup155 ) could be visualized with specific nanobodies carrying a single N-terminal Alexa Fluor 647 maleimide ( Figure 6a ) . Despite the presence of only one dye molecule per nanobody , we again obtained very bright nuclear rim stains with very low background . Staining of Nup155 required a prior permeabilization with Triton X-100 , probably because it is located in close proximity to the pore membrane and is likely buried by other NPC scaffold components ( Eisenhardt et al . , 2014; Mitchell et al . , 2010 ) . 10 . 7554/eLife . 11349 . 011Figure 6 . Immunofluorescence with site-specifically labeled anti-Nup nanobodies . ( a ) Xenopus XL177 cells were digitonin-permeabilized and stained with anti-Nup nanobodies carrying a single N-terminal Alexa Fluor 647 maleimide dye before fixation and DAPI staining . A characteristic nuclear rim stain indicates labeling of NPCs . A nanobody raised against Escherichia coli Maltose-binding protein ( MBP ) served as a negative control . ( b ) Labeling of the anti-GFP nanobody Enhancer with Alexa Fluor 647 NHS ester at lysines or at three engineered cysteines using Alexa Fluor 647 maleimide . Labeling introduces a size shift in SDS–PAGE . Detection was either by Coomassie staining or by in-gel fluorescence . ( c ) Staining of HeLa cells stably expressing GFP-tagged Nup153 with the anti-GFP nanobody labeled via NHS ester or maleimide Alexa Fluor 647 . The nanobody TP377 , raised against Xenopus ( x ) Nup98 , does not cross-react with human Nup98 and served as a negative control . The NHS-labeled GFP nanobody produced strong background-staining , while its maleimide-labeled version yielded bright nuclear rim stains . ( d ) Staining of XL177 cells with nanobodies labeled with Alexa Fluor 647 either at their internal lysine residues ( NHS ester dye ) or via engineered cysteines ( maleimide dye ) . Note that the widely used anti-GFP nanobody Enhancer produces significant background staining when labeled via lysines but not when using engineered cysteines and a maleimide dye . All nanobodies were used at a concentration of 10 nM and all images were obtained under identical settings . DOL , degree of labeling . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 011 For a direct comparison of the NHS chemistry for nanobody-labeling at lysines with maleimide-labeling at engineered surface cysteines , we chose Alexa Fluor 647 as a fluorophore and the widely used anti-GFP nanobody Enhancer ( Kirchhofer et al . , 2010 ) as an example ( Figure 6b ) . When a HeLa Nup153-GFP cell line was stained , we observed a brilliant NPC signal for the Alexa Fluor 647 maleimide-labeled 'Enhancer' , which perfectly coincided with the ( weaker ) GFP signal , and an extremely low background ( Figure 6c ) . In contrast , when this nanobody was labeled at lysines with Alexa Fluor 647 NHS ester , it produced strong nucleoplasmic and cytoplasmic background staining , which essentially obscured the specific signal . The degree of labeling ( DOL ) was the same for both variants . When the Alexa Fluor 647 NHS-labeled 'Enhancer' was applied to XL177 cells ( which lack a GFP-target ) , we again observed very strong background ( Figure 6d ) . In contrast , its Alexa Fluor 647 maleimide-labeled counterpart behaved like a perfect negative control . High background-staining was also observed with the Alexa Fluor 647 NHS-labeled anti-Nup98 nanobody TP377 . The anti-Nup93 nanobody TP179 contains a lysine in CDR II and even lost antigen-binding after NHS modification . In contrast , the Alexa Fluor 647 maleimide-labeled anti-Nup98 and anti-Nup93 nanobodies behaved as perfect imaging reagents and gave crisp NPC signals against very low backgrounds . This comparison indicated that modification of ( multiple ) framework lysines likely creates hydrophobic patches that favor unspecific binding and aggregation . This is certainly sequence context-dependent and milder when reducing the labeling density . However , we did not observe any such complications when labeling nanobodies via engineered cysteines . Due to a diameter well below the diffraction limit , NPCs have been studied by super-resolution microscopy using either indirect immunofluorescence ( Löschberger et al . , 2012; Göttfert et al . , 2013 ) or the anti-GFP nanobody ( Szymborska et al . , 2013 ) . Site-specific fluorescent labeling of nanobodies via cysteines now reliably yields 'renewable' high-quality imaging reagents that can bring fluorophores very close to their target . We therefore tested the performance of our anti-Nup nanobodies in STORM imaging ( Rust et al . , 2006 ) of XL177 cell NPCs ( Figure 7a–c ) . Interestingly , singly Alexa Fluor 647 maleimide-labeled anti-Nup nanobodies were sufficient to produce enough localizations to reconstruct very detailed views of individual NPCs , where multiple copies of the imaged target proteins ( Nup98 , Nup93 , and Nup155 ) appear arranged around the central NPC channel ( Figure 7c ) . A whole nucleus stained with the model anti-Nup98 nanobody TP377 is shown in Figure 7a and magnified views of the nuclear envelope stained with anti-Nup93 and anti-Nup155 nanobody are shown in Figure 7b . Even after applying higher concentrations ( ~100–300 nM ) of nanobody to saturate binding sites , we achieved very low background binding , indicating well-behaved imaging reagents . 10 . 7554/eLife . 11349 . 012Figure 7 . STORM imaging of nuclear pore complexes stained with site-specifically labeled anti-Nup nanobodies . ( a ) STORM image of an entire XL177 cell nucleus stained with anti-Nup98 nanobody TP377 carrying a single N-terminal Alexa Fluor 647 maleimide . ( b ) Close-up view of XL177 cell nuclear envelope regions stained with anti-Nup93 nanobody TP179 ( upper panel ) or an anti-Nup155 nanobody ( lower panel ) containing multiple NPCs . ( c ) STORM images of individual NPCs stained with indicated anti-Nup nanobodies . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 012 Site-specifically labeled nanobodies enabled us to visualize their targeted epitope with high precision . Mapping the corresponding 'visible' epitopes would therefore reveal surface areas of the target that are accessible in a cellular environment . The complementary 'invisible' epitopes on the other hand , would hint to regions that are buried in interaction interfaces . Epitope-mapping strategies based on binding assays to truncated or mutated antigens , co-crystallization or NMR observation of chemical shift perturbations are , however , not suited for high-throughput analysis or hardly applicable to conformational epitopes on protein complexes . We therefore considered crosslinking nanobodies to their target followed by sequencing of the crosslinked target peptide via mass spectrometry as a rapid epitope mapping strategy . Here , a crucial aspect is that a shorter crosslinker will provide a better spatial resolution , provided crosslinkable groups are in reach . As cysteines are by far the best crosslinkable groups , and because we had already placed cysteines at the nanobody surface in close proximity to bound targets , we assessed their suitability for epitope-mapping . As a proof of principle , we crosslinked two anti-Nup93 nanobodies , with or without an N-terminal cysteine , to Nup93 using either an NH2-to-NH2 ( Bis-NHS; BS3; 11 . 4Å ) or an SH-to-NH2 ( Mal-NHS; BMPS; 5 . 9 Å ) crosslinker ( Figure 8a ) . For both anti-Nup93 nanobodies ( TP179 and TP324 ) , exclusive amine-crosslinking was very inefficient and produced only few nanobody•Nup93 adducts that run at higher molecular weight in SDS–PAGE . However , combining the N-terminal cysteine on the nanobody with the ( far shorter ) heterobifunctional crosslinker , produced very prominent nanobody•Nup93 crosslinks . Their position was then clearly identifiable by LC-MS/MS ( Figure 8b–c , Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 11349 . 013Figure 8 . Rapid epitope mapping via crosslinking mass spectrometry . ( a ) Crosslinking of two different anti-Nup93 nanobodies ( TP179 and TP324 ) to Nup93 using amine-to-amine ( 'Bis-NHS'; BS3; 11 . 4 Å linker length ) or thiol-to-amine ( 'Mal-NHS'; BMPS; 5 . 9 Å linker length ) crosslinking reagents . The combination of the very short Mal-NHS crosslinker with an engineered cysteine close to the antigen-binding loops provided for both nanobodies by far the highest yield of crosslinked nanobody•Nup93 adduct . ( b ) List of identified crosslinked peptides involving Nup93 lysines and Cys-TP179 or Cys-TP324 . The crosslinked amino acids are highlighted in red ( see also Figure 8—figure supplement 1 ) . ( c ) Crosslinked lysines of Nup93 to the N-terminal cysteine on anti-Nup93 nanobodies TP179 ( red ) or TP324 ( blue ) are depicted on a structural model of Nup93168-end generated by I-TASSER ( Zhang , 2008 ) . Based on the orthologous yeast crystal structures ( Jeudy and Schwartz , 2007; Schrader et al . , 2008 ) , Nup93 is predicted to form a similar J-shaped structure ( color gradient: NT = N-terminus in blue to CT = C-terminus in orange ) . Whereas TP179 binds to the central portion , TP324 binds to the C-terminus of Nup93 . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 01310 . 7554/eLife . 11349 . 014Figure 8—figure supplement 1 . Representative MS/MS spectra of the crosslinked peptides derived from Nup93•nanobody complexes . The spectra with the best pLink score are shown for the crosslinks between: ( a ) TP179 Cys3 - Nup93 Lys607 , ( b ) TP179 Cys3 - Nup93 Lys612 , ( c ) TP324 Cys3 - Nup93 Lys762 , ( d ) TP324 Cys3 - Nup93 Lys765 , and ( e ) TP324 Cys3 - Nup93 Lys782 . The peaks of the b and y ions are labeled with their charge stages and m/z values . The b and y ions of the longer peptide in a crosslink pair are highlighted in magenta and red , respectively , and the b and y ions of the shorter peptide are highlighted in green and blue , respectively . Fragment ions with superscript 'x' represent those fragment ions with the other peptide crosslinked . DOI: http://dx . doi . org/10 . 7554/eLife . 11349 . 014 For a better visualization of the positions of the identified nanobody crosslinks we generated a structural model of Nup93168-end using I-TASSER ( Zhang , 2008 ) , based on structures of its yeast ortholog ( Jeudy and Schwartz , 2007; Schrader et al . , 2008 ) ( Figure 8c ) . We used the anti-Nup93 nanobody TP179 in STORM imaging of Nup93 within the NPC and could now map its accessible epitope . TP179 binds to the middle region of the J-shaped structure of Nup93 surrounding residues K607 and K612 ( Figure 8—figure supplement 1a–b ) , while TP324 has a C-terminal epitope surrounding lysines K762 , K765 , and K782 of Nup93 ( Figure 8—figure supplement 1c–e ) . The C-terminal region of Nup93 was previously shown to be essential for NPC assembly ( Sachdev et al . , 2012 ) . Accordingly , anti-Nup93 nanobody TP324 that targets the C-terminus of Nup93 does not stain intact NPCs ( data not shown ) , but rather represents a good candidate to selectively disrupt NPC assembly . We developed a well-characterized toolset of high-affinity nanobodies against the vertebrate NPC and established novel strategies to use these nanobodies to natively purify large NPC subcomplexes and to reliably label them with fluorophores for precise super-resolution localization . While these nanobodies will be very valuable to the nucleocytoplasmic transport field , we expect the presented strategies to be widely applicable to all nanobodies . Nanobodies against a single epitope of a larger protein complex now allow a native single-step purification of the entire complex , and thus a subsequent structural and functional analysis . This is certainly especially useful for complexes that are not directly accessible to recombinant production . Furthermore , nanobody-purified endogenous complexes can be used as antigens for another round of immunization , and binders to all complex components can then be selected from the successive nanobody library . Mapping epitopes via crosslinking mass spectrometry will become especially important when selecting nanobodies against such complex antigens ( like large subcomplexes , whole organelles or vesicles ) that cannot be made recombinantly . Combining these strategies therefore has the potential to significantly increase throughput in the selection and identification of renewable binders to eukaryotic proteomes ( Colwill et al . , 2011 ) . Finally , we introduced a method for a reliable fluorescent labeling of nanobodies using surface cysteines and maleimide chemistry . This way , we obtained well-behaved imaging reagents that can bring fluorescent dyes as close as 1–2 nm to their target . Maleimide-labeled nanobodies consistently recognized their antigens far better and produced less background than the corresponding NHS-modified variants . NHS esters have the additional disadvantage that they react not only with amino groups , but also rapidly hydrolyze in aqueous buffers . This makes it difficult to adjust labeling densities and requires adding them in substantial molar excess . In contrast , maleimide-labeling of exposed cysteines is quantitative even with just stoichiometric amounts of labeling reagent and thus far more economical . Site-specific and quantitative fluorescent labeling of nanobodies is going to be crucial for super-resolution microscopy aiming at a detailed structural analysis or determination of absolute protein copy numbers . It also allows predicting the effective label displacement , a fact that will be especially important when applying particle averaging techniques to localization microscopy data ( precision of <1 nm reported by Szymborska et al . , 2013 ) . Because of its well-defined dimension and symmetric structure , the NPC has become a benchmark for many new advancements of super-resolution microscopy ( Schermelleh et al . , 2008; Szymborska et al . , 2013; Göttfert et al . , 2013 ) . The anti-NPC nanobodies described here excelled in super-resolution imaging; they can be renewably produced in high yields and are therefore ideal labeling reagents for such benchmark studies . Two female alpacas , held at the Max Planck Institute for Biophysical Chemistry , were immunized with 0 . 5–1 . 0 mg protein or protein complex at 3–4 week intervals for 3–4 times . The antigens had been expressed recombinantly in E . coli , affinity-purified and mixed with a mild squalen/α-tocopherol/Tween-80-based adjuvant ( oil-in-water emulsion ) before immunization . In detail , we immunized one animal with xlNup93168-end , xtNup98716-866 , and the xlNup62342-547•Nup58267-490•Nup54146-535 complex ( Chug et al . , 2015 ) and another animal with full length xlNup155 and full length xlNup85 . Four days after the final boost , 100 ml of blood were collected from the immunized animal . Peripheral blood lymphocytes were isolated by density gradient centrifugation using Leucosep tubes ( Greiner Bio-One , Austria ) and total RNA was prepared according to Chomczynski and Sacchi ( Chomczynski and Sacchi , 2006 ) . For library generation , cDNA was generated from 30 µg of total RNA using the Superscript III kit ( Life Technologies ) with an IgG-CH2 domain specific primer , pCALL002 ( Conrath et al . , 2001 ) . For VHH domain amplification , a nested PCR was performed . The first PCR product was obtained using the primers AlpVh-L , AlpVHHR1 and AlpVHHR2 ( Maass et al . , 2007 ) , which anneal in the leader sequence and the VHH-specific hinge regions . The first PCR product served as a template for amplification with VHH framework 1 and framework 4-specific primers ( PT411: AATATAGGATCCCAAGTGCAGCTCGTRGAGTCTGG and 38: GGACTAGTGCGGCCGCTGGAGACGGTGACCTGGGT ) introducing BamHI and NotI restriction sites ( underlined ) , respectively . A previous study ( Rothbauer et al . , 2006 ) used NcoI , which according to our sequencing data very frequently cleaves within the CDR I-coding region , resulting in many truncated non-functional nanobody sequences . The BamHI and NotI digested VHH immune library was then cloned into a pHEN4-derived phagemid ( Arbabi Ghahroudi et al . , 1997 ) and used to transform E . coli TG1 ( Lucigen ) . A library of 2–3 x 108 individual transformants was infected with helper phage M13KO7 ( New England Biolabs ) and VHH-displaying bacteriophages were produced overnight while shaking at 37°C . Bacteriophages were purified from the culture supernatant by two successive precipitation steps with 4% PEG-8000 , 500 mM NaCl . The pellets were gently resuspended in 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl and the obtained phage stock solution used directly for selection . Panning was performed using recombinant antigens carrying an Avi-tag that was biotinylated in E . coli by co-expression of biotin ligase BirA ( Beckett et al . , 1999; Schatz , 1993 ) . For the first round of panning , biotinylated antigen was pre-immobilized on Dynabeads Streptavidin T1 ( Life Technologies ) . During later rounds , phages were incubated with biotinylated antigen in solution and then retrieved by adding magnetic beads . After thorough washing , bound phages were eluted and the obtained binders were characterized . Typically three rounds of panning with decreasing antigen concentration ( e . g . 100 nM , 20 nM , and 1 nM ) and increasingly thorough washing were performed . Nanobodies with protease-cleavable affinity tags or engineered cysteines were routinely expressed in the cytoplasm of E . coli BLR ( BL21 derivative; Novagen ) or E . coli SHuffle Express ( New England Biolabs ) . A 50 ml preculture ( Terrific Broth or 2YT medium containing 50 µg/ml Kanamycin ) was grown overnight at 28°C . The culture was then diluted with fresh medium to 250 ml . After 1 h of growth at 25°C , protein expression was induced for 3–5 h by adding 0 . 2 mM IPTG . After addition of 1 mM PMSF and 10 mM EDTA to the culture , bacteria were harvested by centrifugation , resuspended in lysis buffer ( 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl , 20 mM imidazole ) and then lysed by sonication . The lysate was cleared by ultracentrifugation for 1 . 5 h ( T647 . 5 rotor , Sorvall , 38 , 000 rpm ) at 4°C . For native affinity purification , nanobodies were fused to an N-terminal His14-Avi peptide ( GLNDIFEAQKIEWHE ) - ( GlySer ) 9-scSUMOStar- ( GlySer ) 9-tag and co-expressed with the biotin ligase BirA ( Beckett et al . , 1999; Schatz , 1993 ) in the presence of ~20 µg/ml biotin in the medium . Following lysis , nanobodies were purified by Ni2+ chelate affinity chromatography . After washing with lysis buffer , the bound protein was eluted with 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl , 500 mM imidazole . Alternatively , the purified enzyme BirA was added after binding the nanobody to a Ni2+ chelate affinity resin for on-column biotinylation in Bio-buffer ( 50 mM Tris/HCl pH 7 . 5 , 100 mM NaCl , 10 mM ATP , 12 . 5 mM MgCl2 , 250 µM biotin ) . For this , 1 µM BirA in twofold resin bed volume of Bio-buffer was incubated with resin under constant mixing for 2 h at room temperature . Nanobodies with engineered cysteines carried an N-terminal His14-bdNEDD8-tag and were affinity purified via Ni2+ chelate affinity chromatography . After washing , untagged nanobodies were eluted by cleavage with the bdNEDP1 protease ( Frey and Görlich , 2014 ) . Interphase low-speed supernatant ( LSS ) extract was prepared from Xenopus eggs essentially as described before ( Blow and Laskey , 1986 ) and stored at -80°C . LSS was thawed , diluted fourfold in S250 buffer ( 20 mM HEPES pH 7 . 5 , 90 mM KAc , 2 mM MgAc , 250 mM sucrose ) , supplemented with 5 mM ATP and 5 µg/ml Cytochalasin B and then centrifuged in Seton tubes ( SETON Scientific ) for 1 h at 235 , 000 g in a Sorvall Discovery M120 SE ultracentrifuge ( S52ST rotor ) . The lipid- and membrane-free high-speed supernatant ( HSS ) extract was retrieved by puncturing the side of the tube with a needle and served as starting material for affinity purifications . Biotinylated nanobodies were immobilized on magnetic Dynabeads MyOne Streptavidin T1 ( Life Technologies ) in S250 buffer for 30 min at 4°C . Remaining biotin-binding sites on Streptavidin were subsequently blocked by incubation with 50 µM Biotin-PEG-COOH ( Iris Biotech ) in S250 buffer for 15 min . The blocked beads were then added to Xenopus egg extract ( = HSS ) for 1 h at 4°C . Using a magnetic rack , the beads were separated from extract and washed twice in S250 buffer , followed by two washes in 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl , 0 . 05% Tween-20 . Nanobody•target protein complexes were then eluted by adding 0 . 5 µM SUMOStar protease ( Liu et al . , 2008 ) in 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl for 45 min at 4°C . Directly after elution , nanobody-purified Nup93•Nup188 and Nup93•Nup205 complexes were subjected to the GraFix protocol ( Kastner et al . , 2008 ) for complex stabilization . Briefly , ~200 pmoles of nanobody-purified complexes ( ~140 µl ) were loaded onto a 4 . 2 ml 5% – 20% ( w/v ) sucrose-gradient supplemented with 0 . 1% ( v/v ) glutaraldehyde in the 20% fraction . The gradient was run in a TH-660 ultra-centrifuge rotor ( Thermo Scientific; 34 , 000 rpm , 16 h , 4°C ) and then fractionated into 200 µl fractions . The chemically stabilized molecules from the peak fraction were adsorbed to a thin carbon film by surface flotation for 1 min and negatively stained in uranyl formate solution . Images were acquired at room temperature at a magnification of 117 , 333× on a 4k x 4k CCD camera ( TVIPS GmbH ) using twofold pixel binning ( 2 . 5 Å/pixel ) in a Philips CM200 FEG electron microscope ( Philips/FEI ) operated at 160 kV acceleration voltage . From the images , 8139 particles were selected ( Busche , 2013 ) and subjected to contrast transfer function correction ( Sander et al . , 2003 ) . Subsequently , an initial alignment-by-classification ( Dube et al . , 1993 ) step followed by iterative multi-reference alignment and multivariate statistical analysis were performed using IMAGIC ( van Heel et al . , 1996 ) , resulting in 2D class averages . Purified nanobodies with engineered cysteines were freshly reduced by adding 15 mM TCEP for 10 min on ice . Using PD-10 desalting columns ( GE Healthcare ) , the buffer was exchanged to Maleimide-labeling buffer ( 100 mM potassium phosphate pH 6 . 4 , 150 mM NaCl , 1 mM EDTA , 250 mM sucrose ) that had been vacuum degased and purged with argon . For a standard labeling reaction , 10 nmoles of nanobody ( concentration 75–150 µM ) were rapidly mixed with 12 nmoles of Alexa Fluor 647 C2 Maleimide ( Life Technologies ) ( from a 20 mM stock in DMF ) , neutralized to pH 7 . 5 with K2HPO4 and incubated for 1 . 5 h on ice . Free dye was separated from labeled nanobody by buffer exchange to Maleimide labeling buffer on PD10 desalting columns . Quantitative labeling was quality controlled by calculating the degree of labeling ( DOL ) , which defines the molar ratio of dye to protein , as well as by SDS–PAGE and Coomassie staining . In order to obtain nanobodies with three fluorophores , we recommend introducing cysteines at the N-terminus , Ser7 and Ala75 ( other amino acids can occur at these positions in different nanobodies ) of a given nanobody sequence to achieve the smallest possible label displacement . For easy cloning , three cysteines can also be introduced with primers in a single PCR reaction ( positions: N-terminus and Ser7 in the forward primer and at the C-terminus with the reverse primer ) . For Alexa Fluor 647 NHS-labeling , 10 nmoles of nanobody ( concentration 75–150 µM ) were incubated with an eightfold molar excess of dye ( 20 mM stock in DMF ) in 100 mM sodium bicarbonate pH 7 . 8 , 300 mM NaCl for 1 h at 23°C . Subsequently , the reaction was quenched and free dye was separated by buffer exchange to 50 mM Tris/HCl pH 7 . 5 , 300 mM NaCl , 250 mM sucrose on PD10 desalting columns . Xenopus laevis XL177 epithelial cells ( Miller and Daniel , 1977; Ellison et al . , 1985 ) were grown on coverslips at 27°C with 5% CO2 in Xenopus culture medium: ( 25% v/v water , 10% fetal bovine serum , 65% DMEM high glucose medium containing pyruvate and glutamine , and 50 U/ml penicillin + 50 µg/ml streptomycin ) . Alternatively Xenopus laevis A6 cells ( #ATCC CCL-102 TM ) can be used . Cells were pre-fixed for 30 s with 2 . 4% ( w/v ) paraformaldehyde in Transport buffer ( TRB ) ( 20 mM HEPES pH 7 . 5 , 5 mM MgAc , 110 mM KAc , 1 mM EGTA , 250 mM sucrose ) to prevent detachment of cells from the coverslips and briefly washed twice with TRB . The cells were then permeabilized for 8 min on ice with pre-chilled TRB containing 25 µg/ml Digitonin . Following two washes with TRB + 1% ( w/v ) Bovine Serum Albumin ( BSA ) for 5 min each , the cells were incubated with 1–10 nM of fluorescent nanobody for 15 min on ice . Subsequently , the cells were washed thrice for 5 min with TRB + 1% ( w/v ) BSA at room temperature and then fixed for 10 min with 3% ( w/v ) paraformaldehyde in TRB . The nuclear envelopes of the fixed cells were afterwards permeabilized with 0 . 3% Triton X-100 for 3 min , washed with 1xPBS and DNA was stained by addition of 2 µg/ml DAPI in 1xPBS for 10 min . The coverslips were mounted in SlowFade Gold or SlowFade Diamond Antifade Mountant ( Life Technologies ) and analyzed by confocal laser-scanning microscopy on a Leica SP5 microscope . In order to obtain the highest labeling efficiency , XL177 cells were stained with Alexa Fluor 647 maleimide-conjugated nanobodies initially after a short pre-fixation and digitonin permeabilization of the plasma membrane . The cells were subsequently fixed , the nuclear envelope was permeabilized with Triton X-100 and labeled nanobodies were added again . The optimal concentration of each nanobody for both steps was titrated before , using confocal microscopy . All STORM imaging experiments were carried out in MEA imaging buffer as previously described ( Dempsey et al . , 2011 ) . The buffer consisted of 50 mM Tris/HCl pH 8 . 0 , 10 mM NaCl , 10% glucose ( w/v ) , 10 mM β-mercaptoethylamine pH 8 . 5 ( Sigma , 30070 ) , and 1% of an enzymatic oxygen scavenger system stock solution , added to the buffer immediately prior to use . The oxygen scavenger stock solution was prepared by mixing glucose oxidase ( 10 mg , Sigma , G2133 ) with catalase ( 50 µl , 20 mg/ml−1 , Sigma , C30 ) in 1x PBS ( 200 µl ) , and centrifuging the mixture at 13 , 000 rpm for 1 min . STORM imaging measurements were performed using a custom-built STORM microscope , based on an inverted fluorescence microscope stand ( Olympus IX71 ) as previously described ( Dempsey et al . , 2011 ) . The microscope was fitted with a 100x oil-immersion objective lens ( Olympus UPLANSAPO , NA1 . 4 ) , which enabled efficient detection of single fluorophores . The objective lens was mounted on a piezo-positioner ( Piezo Jena ) , which enabled fine focus adjustment . A custom-built focus-lock system was used to maintain a stable focus during data acquisition . For STORM imaging , photo-switchable Alexa Fluor 647 was excited at 642 nm , and in some measurements , the sample was also exposed to 405 nm light to increase the activation rate of switching . A fiber laser ( MPB Communications , 2RU-VFL-P-1000-642 ) was used to generate 642 nm light . The laser illumination was configured such that the illumination angle could be varied between an epi-illumination geometry and a total internal reflection ( TIRF ) illumination mode . Typically , the sample was illuminated with oblique illumination ( not TIRF ) for reduced background signal . Fluorescence emission of Alexa Fluor 647 was detected using an EMCCD camera ( Andor Ixon DU860 ) . STORM data analysis was carried out using custom analysis software , as previously described ( Bates et al . , 2007 ) . The Xenopus Nup98716-866 NPC anchor domain and the anti-Nup98 nanobody TP377 were expressed with an N-terminal His14-bdSUMO-tag and purified using Ni2+ chelate affinity chromatography . Crystallization required an exchange of the surface-exposed cysteine 821 of Nup98 to serine . Highly pure untagged protein was cleaved off the column using 50 nM bdSEN1P protease ( Frey and Görlich , 2014 ) in 20 mM Tris , 20 mM NaCl . The complex was formed by incubating equimolar amounts of Nup98716-866 and TP377 o/n at 4°C and then subjected to anion exchange chromatography using a HiTrap Q HP 5 ml column ( GE Healthcare ) . The eluted complex was then further purified using gel filtration on a Hi-Load Superdex 75 16/60 column equilibrated in 20 mM Tris/HCl pH 7 . 5 , 50 mM NaCl . The complex was crystallized by the vapor diffusion method in sitting drops . 60 nl of a reservoir solution containing 45% ( w/v ) Pentaerythritol propoxylate ( 17/8 PO/OH; Jena Bioscience ) and 100 mM Tris pH 8 . 5 was mixed with 60 nl of the prepared protein complex solution concentrated to 25 mg/ml . Crystals grew within 1 day at 20°C and were flash-frozen in liquid nitrogen without additional cryo-protection . Diffraction data were collected at 100 K with a wavelength of 0 . 9787 Å on the beamline PXII at the Swiss Light Source ( SLS ) at the Paul Scherrer Institute , Switzerland . Crystals belonged to the space group P41 and diffracted to 1 . 9 Å ( see Table 1 ) . For structure determination , molecular replacement was performed in PHASER with a published nanobody structure ( PDB 4KRN; Schmitz et al . , 2013 ) as a search model . The resulting electron density map was used for automated model building in Phenix ( Adams et al . , 2010 ) . Anti-Nup93 nanobodies TP179 or TP324 and Nup93 ( ~20 µM each ) were incubated on ice for 30 min in Maleimide labeling buffer to allow complex formation . After adding 40 µM of crosslinking agent , the pH was increased to 7 . 5 and the reaction was continued for 1 h on ice . The following crosslinkers were used 'Mal-NHS' = BMPS ( 3-[Maleimido]propionic acid NHS ester , Iris Biotech , CAS #55750-62-4 ) and 'Bis-NHS' = BS3 ( Suberic acid bis[sulfo NHS ester] , Life Technologies , CAS #82436-77-9 ) . One-eighth of the reaction was loaded on a SDS–PAGE gel . The band corresponding to crosslinked products was excised and subjected to in-gel trypsin digestion as described ( Schmidt and Urlaub , 2009 ) . The peptide fragments were extracted in a solvent system containing 5% acetonitrile ( ACN ) , 0 . 1% formic acid ( FA ) to a final volume of 20–30 µl and submitted to liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) analysis . For LC-MS/MS analysis , 6 µl of the sample solution was injected into a nano-liquid chromatography system ( UltiMateTM 3000 RSLCnano system ) including a 3 cm × 150 µm inner diameter C18 trapping column in-line with a 30 cm × 75 µm inner diameter C18 analytical column ( both in-house packed with 1 . 9 µm C18 material , Dr . Maisch GmbH ) . Peptides were desalted on the trapping column for 3 min at a flow rate of 10 µl/min in 95% of mobile phase A ( 0 . 1% FA in H2O , v/v ) and 5% of mobile phase B ( 80% ACN and 0 . 05% FA in H2O , v/v ) , eluted from the trapping column , and separated on the analytical column using a 43 min linear gradient of 15–46% mobile phase B at a flow rate of 300 nl/min . Separated peptides were analyzed on-line with an Orbitrap Fusion mass spectrometer ( Thermo Scientific ) . The 20 most intense precursor ions with charge states 3–8 in the survey scan ( 380–1580 m/z scan range ) were isolated in the quadrupole mass filter ( isolation window 1 . 6 m/z ) and fragmented in the higher energy collisional dissociation ( HCD ) cell with 30% normalized collision energy . A dynamic exclusion of 20 s was used . Both the survey scan ( MS1 ) and the product ion scan ( MS2 ) were performed in the Orbitrap at 120 , 000 and 30 , 000 resolution , respectively . Spray voltage was set at 2 . 3 kV and 60% of S-lens RF level was used . Automatic gain control ( AGC ) targets were set at 5×105 and 5×104 for MS1 and MS2 , respectively . The raw data of LC-MS/MS analysis were converted to mascot generic format ( mgf ) files by Proteome Discoverer 2 . 0 . 0 . 802 software ( Thermo Scientific ) . The mgf files were searched against a FASTA database containing the sequences of the nanobody and Nup93 by pLink 1 . 22 software ( Yang et al . , 2012 ) using a target-decoy strategy . Database search parameters included mass accuracies of MS1 <10 ppm and MS2 <20 ppm , carbamidomethylation on cysteine and oxidation on methionine as variable modifications . The number of residues of each peptide on a crosslink pair was set between 4 and 100 . A maximum of two trypsin missed-cleavage sites were allowed . The results were obtained with 1% false discovery rate . The identified crosslinks were filtered with a threshold of at least two spectral counts and a pLink score < 10e-4 .	Antibodies not only protect humans and other animals against disease-causing bacteria and viruses . They can also be used as tools for medical diagnostics and basic research . Conventional antibodies consist of light and heavy protein chains , and both are required to bind to target molecules ( or antigens ) . Alpacas , llamas and camels , however , possess simpler antibodies that lack light chains and bind to antigens via a single protein domain . Such domains can be produced in "re-programmed" bacteria and are then called nanobodies . Compared to normal antibodies , nanobodies are 10-fold smaller , which is of great advantage in virtually all practical applications . Pleiner et al . made nanobodies against the nuclear pore complex ( or NPC for short ) – a nanoscopic machine for transporting large biological molecules in and out of the cell’s nucleus . These nanobodies can be linked to dyes called fluorophores and then used to stain NPCs so that they can be observed under a microscope . When fluorophores were attached , in the traditional way , via the amino acid lysine , all tested nanobodies performed poorly in fluorescence microscopy - pointing to a systematic problem . Pleiner et al . therefore explored an alternative , namely to label nanobodies via engineered cysteines . This was counterintuitive , because nanobodies contain already two other cysteines that must not be modified and that normally form a stabilizing “disulfide” bond . Pleiner et al . found , however , that the labeling reaction is absolutely specific for the engineered surface cysteines when it is performed at low temperature . This strategy consistently yielded imaging reagents that could effectively deliver fluorophores as close as 1-2 nanometers to their antigens . Nanobodies labeled in this way are therefore ideal to exploit the full potential of super-resolution microscopy . The engineered surface cysteines proved also useful as "position sensors" to report which region of an antigen is actually contacted by a given nanobody . Nanobodies are also used to purify protein complexes from crude cell extracts by a method called affinity chromatography . Previously , nanobodies were chemically attached to an insoluble matrix , and bound protein complexes were released under conditions that destroy interactions between proteins . Pleiner et al . now replaced the destructive step with a step that uses an enzyme to cut a bond and gently detach the nanobody ( along with any bound protein complex ) from the matrix . Bound protein complexes thus stay intact and can be studied further . In the future , this strategy can be applied to nanobodies that recognize tags commonly added to proteins ( i . e . GFP ) to isolate virtually any protein complex for functional assays or structural analyses .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology", "tools", "and", "resources" ]	2015	Nanobodies: site-specific labeling for super-resolution imaging, rapid epitope-mapping and native protein complex isolation
Transcriptional feedback loops are key to circadian clock function in many organisms . Current models of the Arabidopsis circadian network consist of several coupled feedback loops composed almost exclusively of transcriptional repressors . Indeed , a central regulatory mechanism is the repression of evening-phased clock genes via the binding of morning-phased Myb-like repressors to evening element ( EE ) promoter motifs . We now demonstrate that a related Myb-like protein , REVEILLE8 ( RVE8 ) , is a direct transcriptional activator of EE-containing clock and output genes . Loss of RVE8 and its close homologs causes a delay and reduction in levels of evening-phased clock gene transcripts and significant lengthening of clock pace . Our data suggest a substantially revised model of the circadian oscillator , with a clock-regulated activator essential both for clock progression and control of clock outputs . Further , our work suggests that the plant clock consists of a highly interconnected , complex regulatory network rather than of coupled morning and evening feedback loops . Circadian clocks are widespread in nature , presumably because they help diverse organisms prepare for predictable day/night cycles . Although specific components are not widely conserved , eukaryotic clocks are composed of interlocking negative transcriptional feedback loops ( Harmer , 2009 ) . In Arabidopsis , the first-identified clock genes function in a double negative feedback loop , with two morning-phased Myb-like transcription factors , CIRCADIAN CLOCK ASSOCIATED 1 ( CCA1 ) and LATE ELONGATED HYPOCOTYL ( LHY ) , repressing expression of an evening-phased pseudo-response regulator , TIMING OF CAB EXPRESSION 1 ( TOC1 or PRR1 ) , which in turn represses expression of CCA1 and LHY ( Schaffer et al . , 1998; Wang and Tobin , 1998; Strayer et al . , 2000; Alabadi et al . , 2001; Gendron et al . , 2012; Huang et al . , 2012; Pokhilko et al . , 2012 ) . CCA1 and LHY also promote the expression of PRR7 and 9 , two day-phased genes , and are in turn repressed by these PRRs and their homolog PRR5 , forming another negative feedback circuit ( Farre et al . , 2005; Nakamichi et al . , 2010 ) . Finally , TOC1 , GIGANTEA ( GI ) , and the evening complex components including LUX ARRHYTHMO ( LUX ) , EARLY FLOWERING 3 ( ELF3 ) and 4 ( ELF4 ) , act in double negative feedback loops with CCA1 , LHY , and PRR7 and 9 ( Fowler et al . , 1999; Park et al . , 1999; Dixon et al . , 2011; Helfer et al . , 2011; Nusinow et al . , 2011; Huang et al . , 2012; Pokhilko et al . , 2012 ) . Thus most characterized clock components repress expression of other clock components . A cis-regulatory element named the evening element ( EE ) [ ( A ) AAATATCT] has been found to be central to circadian clock function in plants . Most evening-phased central clock genes ( including TOC1 , PRR5 , GI , LUX and ELF4 ) contain the EE in their promoter regions ( Covington et al . , 2008; Harmer , 2009 ) and the two morning-phased components , CCA1 and LHY , bind directly to the EE to repress evening-phased clock gene expression ( Alabadi et al . , 2001 ) . The EE was first identified by its overrepresentation in the promoters of evening-phased genes ( Harmer et al . , 2000 ) and is sufficient to confer evening-phased expression on a reporter gene ( Harmer and Kay , 2005 ) . In addition to these two morning-phased transcriptional repressors that act via the EE , two pieces of evidence suggest that there is also a transcriptional activator ( s ) present in the afternoon that regulates the EE . First , if only repressors bind to the EE , loss of protein binding to the EE should result in constitutively high expression of EE-regulated genes; however , mutation of the EE causes decreased expression of an EE-regulated reporter gene ( Harmer and Kay , 2005 ) . Second , an afternoon/evening-phased activity that specifically binds the EE is present in plant extracts and persists in cca1 lhy mutants , consistent with the existence of a clock-regulated , afternoon-phased activator of the EE ( Harmer and Kay , 2005 ) . A clock-regulated activator of the EE might help to explain why evening-phased clock genes are expressed with a circadian rhythm in cca1 lhy plants rather than being arrhythmic ( Mizoguchi et al . , 2002 ) . A candidate activator of the EE is REVEILLE 8/ LHY-CCA1-LIKE 5 ( RVE8/LCL5 ) ( Farinas and Mas , 2011; Rawat et al . , 2011 ) . RVE8 has been shown to bind to the EE in vitro and in planta , and its protein levels display a circadian rhythm that peaks in the afternoon ( Gong et al . , 2008; Rawat et al . , 2011 ) . Furthermore , rve8 loss of function mutations cause a long circadian period ( Farinas and Mas , 2011; Rawat et al . , 2011 ) which is opposite to the phenotypes of cca1 or lhy loss of function mutants ( Green and Tobin , 1999; Mizoguchi et al . , 2002 ) . However , despite its ability to bind to the EE in the TOC1 and PRR5 promoters in planta , loss of RVE8 function does not significantly affect the transcript levels of these evening genes ( Farinas and Mas , 2011; Rawat et al . , 2011; Hsu and Harmer , 2012 ) , perhaps due to genetic redundancy or complex feedback regulation within the clock system . Here , we used an inducible RVE8 line and genome-wide expression profiling to identify hundreds of clock-regulated genes controlled by RVE8 . Experiments with an inhibitor of translation revealed that most evening-phased clock genes are directly induced by RVE8 . Consistent with RVE8 acting via the EE regulatory motif , we found that genes induced by RVE8 are enriched for the EE in their promoter regions . Furthermore , plants mutant for RVE8 and its two closest homologs , RVE4 and RVE6 , have lost the afternoon-phased EE-binding activity . Finally , rve4 rve6 rve8 triple mutants display an extremely long circadian period , with delayed and reduced expression of evening-phased clock genes . Together , these data suggest a considerably revised model of the plant clock , with an indispensable role for activators of transcription within the circadian regulatory network . Our work shows that rather than consisting of discrete , interlocked feedback loops , the plant circadian oscillator is more accurately described as a highly interconnected complex network . To identify RVE8 target genes , we generated a line with rapidly inducible RVE8 activity . A translational fusion between RVE8 and the glucocorticoid receptor ( GR ) , driven by the native RVE8 promoter , was introduced into rve8-1 plants . GR fusion proteins are held in the cytoplasm unless the synthetic ligand for GR , dexamethasone ( DEX ) , is applied , which allows the chimeric factor to move into the nucleus ( Picard et al . , 1988 ) . Both rve8-1 and rve8-1 RVE8::RVE8:GR plants have a long-period phenotype that is only rescued by DEX treatment of the rve8-1 RVE8::RVE8:GR line ( Figure 1A , B ) , demonstrating that the RVE8:GR fusion protein retains RVE8 function and acts in a drug-inducible manner . 10 . 7554/eLife . 00473 . 003Figure 1 . Activation of PRR5 by RVE8 induction is stronger in the afternoon . ( A ) and ( B ) Luciferase activity in mock ( A ) and DEX-treated ( B ) Col , rve8-1 and rve8-1 RVE8::RVE8:GR plants transgenic for the CCR2::LUC reporter . Plants were entrained in 12:12 light/dark ( LD ) cycles for 6 days and then sprayed with 30 µM DEX or 0 . 05% ethanol ( mock treatment ) plus luciferin before release to constant red light ( 30 µEi ) for imaging of bioluminescence . Mean + SEM from 17 to 25 plants are represented . ( C ) and ( D ) Transcript levels of PRR5 in response to induction of RVE8 activity in rve8-1 RVE8::RVE8:GR ( C ) or rve8-1 35S::RVE8:GR ( D ) at different time of day . 30 µM DEX or 0 . 05% ethanol ( mock ) was applied at the times indicated and the plants were harvested 2 hr later . Expression levels were quantified by qRT-PCR and normalized to PP2A . Mean ± SEM from three biological replicates are represented . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 003 We next examined the ability of DEX-inducible RVE8-GR to activate expression of a known RVE8 target , the evening-phased clock gene PRR5 ( Rawat et al . , 2011 ) . Since RVE8 protein levels are circadian-regulated , with peak protein abundance in the subjective afternoon ( Rawat et al . , 2011 ) , we tested the ability of RVE8 to activate PRR5 after DEX induction in the morning or afternoon . Induction of PRR5 by RVE8 is much stronger when RVE8 activity is induced in the afternoon ( Zeitgeber Time 6 [ZT6] , or 6 hr after lights on ) than when RVE8 is induced in the morning ( ZT0 ) ( Figure 1C ) . Similarly , although induction of constitutively expressed RVE8 ( 35S::RVE8:GR ) in the morning ( ZT0 ) is sufficient to induce PRR5 , this induction is much stronger when the DEX treatment is given in the afternoon ( ZT6 ) ( Figure 1D ) . These data indicate the ability of RVE8 to induce target genes is gated , with maximum activity in the afternoon . To globally identify RVE8 target genes , we induced RVE8 activity near the time of normal peak RVE8 protein accumulation ( Figure 2A ) and used RNA-seq analysis to characterize the transcriptome in response to RVE8 induction ( experimental design , Figure 2B; analysis summary , Supplementary file 1A , B ) . Verification of RNA-seq results using qRT-PCR showed excellent correlation between the two techniques , suggesting our RNA-seq results are reliable ( Figure 2—figure supplement 1 ) . Comparing mock- and DEX-treated RVE8:GR and rve8-1 plants , we found that 583 genes are specifically up- and 850 are down-regulated in response to RVE8 induction ( Figure 2C , D and Supplementary file 1C–F ) . Interestingly , a significantly higher proportion of both the up- and down-regulated RVE8 targets are clock-controlled ( Figure 2E , F , 64% and 62% , respectively ) than the one-third of the transcriptome expected by chance ( Covington et al . , 2008; Hsu and Harmer , 2012 ) . RVE8 thus preferentially regulates clock-controlled genes ( CCGs ) . 10 . 7554/eLife . 00473 . 004Figure 2 . Identification of RVE8 targets by RNA-seq . RNA-seq experimental design and data analysis . ( A ) Relative timing of RVE8 induction and RVE8 protein abundance during a day . Adapted from Rawat et al . ( 2011 ) . ( B ) Scheme of experimental design . ( C ) and ( D ) Weighted Venn diagrams of genes significantly responsive to RVE8 induction and/or DEX treatment . Genes up-regulated ( C ) or down-regulated ( D ) by RVE8 and/or DEX . Differentially expressed genes were identified using edgeR ( Robinson et al . , 2010 ) with an adjusted p-value <0 . 01 as the cutoff . Genes significantly different between ‘RVE8:GR + DEX’ and ‘RVE8:GR + mock’ or between ‘RVE8:GR + DEX’ and ‘rve8 + DEX’ are grouped into the RVE8-induced ( C ) or RVE8-repressed sets ( D ) shown in red circles . Genes significantly different between ‘rve8 + DEX’ and ‘rve8 + mock’ are grouped into the ‘DEX-induced’ ( C ) or ‘DEX-repressed’ ( D ) sets shown in blue circles . The genes uniquely induced or repressed by RVE8 ( the 583 and 850 genes shown in green areas in ( C ) and ( D ) , respectively ) were defined as RVE8-regulated and used for further analysis . ( E ) and ( F ) The relative proportion of clock-controlled genes ( CCGs ) and non-clock-controlled genes ( NCGs ) among RVE8 targets . RVE8-induced genes ( E ) ; RVE8-repressed genes ( F ) . ( G ) and ( H ) Circadian phase distributions of RVE8-regulated CCGs . CCGs up-regulated by RVE8 ( G ) ; CCGs down-regulated by RVE8 ( H ) . White box: subjective day; grey box: subjective night . X-axis , 0: subjective dawn , 12: subjective dusk . Phase estimates are from previously published data ( Hsu and Harmer , 2012 ) . See also Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 00410 . 7554/eLife . 00473 . 005Figure 2—figure supplement 1 . Expression levels as determined by RNA-seq and qRT-PCR are highly correlated . Expression levels of selected genes defined as RVE8-regulated in the RNA-seq experiment were examined using qRT-PCR . The logarithm of fold change values in the RNA-seq and the qRT-PCR data were plotted along with the linear regression line to examine the correlation relationship between the two methods . Pearson and Spearman correlation tests were performed in R . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 005 CCGs regulated by RVE8 are enriched for two complementary circadian phases , with the RVE8-induced genes enriched for an evening ( Figure 2G ) and the RVE8-repressed genes enriched for a morning phase ( Figure 2H ) . Many evening-phased oscillator genes are induced by RVE8 , including PRR5 , TOC1 , PRR3 , GI , LUX , and ELF4 ( Supplementary file 1G ) . In contrast , morning-phased oscillator genes including CCA1 , LHY , RVE8 itself , and a day-phased central clock gene , PRR9 , are found to be repressed by RVE8 ( Supplementary file 1G ) . Activation of evening-phased and repression of morning-phased central clock genes suggests that RVE8 acts as a key regulator within the central system . To identify possible in vivo RVE8 binding sites , we identified promoter motifs found more frequently than expected by chance among the CCGs up- or down-regulated in response to RVE8 induction . EE and EE-like sequences are significantly overrepresented in the RVE8-induced CCGs , both when compared to their frequency in all genes in the genome ( Supplementary file 2A ) and in all evening-phased CCGs ( Table 1A ) . This indicates that RVE8 preferentially regulates evening-phased genes containing an EE or EE-like promoter sequence . Since RVE8 directly binds to the EE in vitro and in vivo ( Rawat et al . , 2011 ) , this suggests that RVE8 may directly activate many evening genes via binding to the EE in their promoters . 10 . 7554/eLife . 00473 . 006Table 1 . Enrichment of EE , G-box-like and ME-like motifs in CCGs regulated by RVE8 compared to their occurrence in all CCGs previously defined as either evening-phased or morning-phased ( Hsu and Harmer , 2012 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 006 ( A ) Evening-phased genes ( CT 8 to CT 14 ) MotifSequenceCCGs ( 2709 genes ) RVE8-induced CCGs ( 278 genes ) pGenes with the motifCoverage ( % ) Genes with the motifCoverage ( % ) Short EEAAATATCT79429 . 315254 . 5<2 . 2 × 10−16*Long EEAAAATATCT44416 . 410437 . 52 . 06 × 10−15EE-likeAATATCT136050 . 219068 . 27 . 39 × 10−09** ( B ) Morning-phased genes ( CT 20 to CT 2 ) MotifSequenceCCGs ( 1572 genes ) RVE8-repressed CCGs ( 328 genes ) pGenes with the motifCoverage ( % ) Genes with the motifCoverage ( % ) G-box-likeBACGTRD118775 . 526681 . 00 . 0317ME-likeCCACA142990 . 930893 . 90 . 08297To determine whether the over-represented motifs found in RVE8 targets ( Supplementary file 2 ) are enriched when compared to the morning-phased and evening-phased CCG groups , the number of genes containing the motif in each phase group was compared to that in the up- or down-regulated RVE8 targets . Fisher's exact test was performed to determine if the ratios in both groups are significantly different ( p<0 . 05; p<0 . 01; *p<0 . 001 ) . Among CCGs repressed by RVE8 , we found motifs related to the G-box and morning element ( ME ) to be overrepresented when compared to all genes in the genome ( Supplementary file 2B ) . Since most RVE8-repressed genes are also morning-phased CCGs ( Figure 2H ) , we compared the frequency of these motifs between RVE8-repressed and all morning-phased CCGs . Unlike our results for the EE , the G-box and ME motifs are found at a similar rate in RVE8-repressed and in phase-matched CCGs ( Table 1B ) . The similar frequency of these two motifs in these two groups indicates that RVE8 activity is not preferentially correlated with the morning-phased related cis-regulatory elements . The preferential correlation of RVE8 activity with the EE , but not with the morning-associated motifs , suggests that RVE8 may directly activate evening-phased clock genes that then go on to repress morning-phased CCGs . To investigate whether RVE8 regulates morning and evening clock genes directly or indirectly , we induced RVE8 activity in the presence of cycloheximide ( CHX ) , a protein synthesis inhibitor , and then examined transcript levels of genes identified as RVE8-regulated in our RNA-seq experiment . Genes regulated by RVE8 both in the presence or absence of CHX would be considered direct targets while those only regulated by RVE8 in the absence of CHX would be considered indirect targets . CHX treatment increased the accumulation of transcripts regulated by the nonsense mediated mRNA decay ( NMD ) pathway ( Carter et al . , 1995; Arciga-Reyes et al . , 2006; Kurihara et al . , 2009 ) , suggesting that CHX treatment reduced or blocked translation as expected ( Figure 3—figure supplement 1A–C ) . Consistent with a role for RVE8 in activation of evening genes via direct binding to the EE , all of the EE-containing , evening-phased central clock and output genes examined are robustly induced by RVE8 even in the presence of CHX ( Figure 3A–F ) . In contrast , the RVE8-mediated repression of expression of all tested morning genes is reduced or abolished in the presence of CHX ( Figure 3—figure supplement 1D–G ) , suggesting that RVE8 represses these genes indirectly . In the case of PRR9 , induction of RVE8 in the presence of CHX actually causes increased expression levels rather than the decrease seen in the absence of CHX ( Figure 3—figure supplement 1E ) . RVE8-mediated activation of PRR9 is likely masked in the absence of CHX by the concomitant induction of strong repressors of PRR9 expression such as TOC1 and LUX ( Helfer et al . , 2011; Gendron et al . , 2012; Huang et al . , 2012 ) ( Figure 3C , E ) and is only revealed when the translation of these repressors is blocked . In summary , for all of the genes examined , we found that RVE8 directly activates evening-phased genes but indirectly represses the morning-phased genes . 10 . 7554/eLife . 00473 . 007Figure 3 . RVE8 activates evening genes directly . ( A ) – ( F ) Transcript levels of evening genes in response to RVE8 induction in the absence or presence of cycloheximide ( CHX ) . 7-day-old rve8-1 and rve8-1 RVE8::RVE8:GR plants were grown in light:dark ( LD ) cycles and mock- or DEX-treated in the absence or presence of CHX at ZT4 ( 4 hr after dawn ) and harvested at ZT8 ( 8 hr after dawn ) . ( A–E ) Evening-phased clock genes . ( F ) Evening-phased clock output gene . Transcript levels were determined by qRT-PCR and then normalized to PP2A . Mean ± SEM from three biological replicates are represented . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 00710 . 7554/eLife . 00473 . 008Figure 3—figure supplement 1 . Morning-phased genes are indirectly repressed in response to RVE8 induction . ( A ) – ( C ) Levels of transcripts controlled by nonsense-mediated mRNA decay ( NMD ) in response to cycloheximide ( CHX ) treatment . N . D . : not detectable . ( D ) – ( G ) Transcript levels of morning genes in response to RVE8 induction in the absence or presence of CHX . 7-day-old rve8-1 and rve8-1 RVE8::RVE8:GR plants were grown in light:dark ( LD ) cycles and mock- or DEX-treated in the absence or presence of CHX at ZT4 ( 4 hr after dawn ) and harvested at ZT8 ( 8 hr after dawn ) . ( D ) – ( F ) Morning-phased clock genes . ( G ) Morning-phased clock output gene . Transcript levels were determined by qRT-PCR and then normalized to PP2A . Mean ± SEM from three biological replicates are represented . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 008 Since our data suggest that RVE8 is primarily ( perhaps exclusively ) an activator of transcription , we examined the physiological functions of all RVE8-induced genes in order to identify clock output pathways that may be directly influenced by RVE8 . Functional classifications in which RVE8-induced genes are statistically overrepresented include regulation of the central oscillator ( Supplementary file 1H ) , as expected given the clock phenotype of rve8 plants ( Farinas and Mas , 2011; Rawat et al . , 2011; Hsu and Harmer , 2012 ) . In addition , genes acting in pathways related to responses to the environments ( including external stimulus , defense , temperature and stress ) , hormone regulation and metabolic processes are also enriched ( Supplementary file 1H ) . Together , these data suggest that RVE8 shapes the evening-phased expression of hundreds of genes , directly influencing a large number of circadian output pathways . Comparison of the phases of expression of RVE8-induced CCGs that have EE promoter motifs to the phases of all CCGs with EE sequences showed that the RVE8-regulated genes have a much narrower range of phases ( Figure 4A ) . Almost all RVE8-induced EE-containing genes have peak expression in the subjective evening . Interestingly , the mean peak phase for RVE8-regulated EE-containing CCGs is significantly earlier than that of all EE-containing CCGs , indicating that RVE8 regulates a subset of evening genes that have slightly earlier phase than average EE-containing evening genes ( Figure 4A ) . These data are consistent with the afternoon-phased RVE8 binding to the EE to induce expression of a subset of evening-phased genes . 10 . 7554/eLife . 00473 . 009Figure 4 . RVE8 functions through the EE . ( A ) Circadian phase distributions of all EE-containing CCGs and RVE8-induced EE-containing CCGs . The RVE8-induced EE-containing CCGs are enriched for an earlier phase than that of all EE-containing CCGs . The means of the phase distribution in these two groups ( 10 . 03 for RVE8-induced EE-containing CCGs; 10 . 75 for all EE-containing CCGs ) are significantly different ( p=0 . 007; Student's t-test ) . ( B ) Period of CCR2::LUC activity in rve4-1 , rve6-1 and rve8-1 single , double and triple mutants . Seedlings were grown in LD for 6 days and released to constant red plus blue light . Mean ± SEM from 34 to 50 plants . ( C ) Circadian rhythms are lengthened but still robust in rve4 rve6 rve8 mutants . Averaged bioluminescence of CCR2::LUC activity in Col , rve8-1 and rve4 rve6 rve8 triple mutants . Mean ± SEM from 20 to 25 plants . ( D ) An electrophoretic mobility shift ( EMSA ) assay with protein extracts made from Col and rve4 rve6 rve8 plants grown in LD for 11 days . Plants were harvested at the indicated times . A 50-fold molar excess of unlabeled EE ( WT competitor ) or mutated EE ( mutant competitor ) double-stranded DNA was added as indicated . Arrow: the predominant afternoon EE-binding activity , arrowhead: unbound probe . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 00910 . 7554/eLife . 00473 . 010Figure 4—figure supplement 1 . Characterization of RVE4 , RVE6 , and RVE8 mutant alleles . ( A ) – ( C ) RVE4 , RVE6 and RVE8 transcript levels in Col and the rve4 rve6 rve8 triple mutant . 7-day-old seedlings ( about 30 plants each ) were grown in 12:12 LD and harvested at ZT 0 and ZT 4 . RNA was isolated and qRT-PCR was performed . RVE4 and RVE8 transcripts are not detectable ( N . D . ) but ∼30% of normal RVE6 transcript levels are apparent in the triple mutant . Expression levels are normalized to PP2A . Mean ± SEM from three technical replicates are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 010 Loss of RVE8 function has neither a strong effect on clock function nor on expression levels of evening-phased genes ( Farinas and Mas , 2011; Rawat et al . , 2011; Hsu and Harmer , 2012 ) . This may be due to partial genetic redundancy , since there are four other close RVE8 homologs ( RVE3 , 4 , 5 , and 6 ) in the Arabidopsis genome ( Rawat et al . , 2009 ) and all of these proteins were found to bind to the EE ( Gong et al . , 2008; Rawat et al . , 2011 ) . To investigate whether RVE8 homologs play a partially redundant role with RVE8 in the circadian clock , we identified plants mutant for the two closest RVE8 homologs , RVE4 and RVE6 ( Figure 4—figure supplement 1A–C ) and examined clock function in single and higher order mutants . The pace of the clock in rve4-1 and rve6-1 single mutants is not significantly different from wild-type ( Figure 4B ) . However , combining loss-of-function RVE4 or RVE6 alleles with rve8-1 makes the period length much longer than rve8-1 alone , while the rve4 rve6 rve8 triple mutant has a period approximately 4 hr longer than wild-type ( Figure 4B ) . These data suggest that RVE4 , 6 and 8 play a partially redundant role in speeding up the pace of the clock . Despite the severe long period phenotype , the rve4 rve6 rve8 triple mutant displays robust circadian rhythms ( Figure 4C ) . We previously identified an afternoon-phased activity in plant extracts that specifically binds the EE , suggesting it might represent a cycling activator ( s ) for the EE ( Harmer and Kay , 2005 ) . Since we found RVE8 is an afternoon-phased activator of genes with EE in their promoters and that RVE4 and RVE6 play a partially redundant role with RVE8 in setting clock pace , we examined circadian-regulated EE-binding activity in the rve4 rve6 rve8 triple mutant plants in an in vitro EE-binding assay . As expected , extracts from wild-type plants have an afternoon-phased EE binding activity ( Figure 4D ) . Remarkably , this cycling EE-binding activity is abolished in the triple mutant ( Figure 4D ) . Given that we have previously found that RVE4 , RVE6 and RVE8 can all be affinity purified from plant extracts using EE sequences ( Rawat et al . , 2011 ) , this strongly suggests that RVE4 , 6 and 8 comprise a clock-regulated , evening-phased EE-binding activity . To further examine the functions of these RVEs ( RVE4 , 6 and 8 ) in plants , we examined the transcript profiles of genes we identified as RVE8 targets in the higher order rve mutants . In contrast to the rve8-1 single mutant , which has normal expression levels of evening genes ( Rawat et al . , 2011; Hsu and Harmer , 2012 ) , rve6 rve8 double and rve4 rve6 rve8 triple mutants grown in constant light ( LL ) display significantly reduced levels of PRR5 transcripts ( Figure 5—figure supplement 1A ) . Consistent with the progressively longer period in rve6 rve8 and rve4 rve6 rve8 mutants ( Figure 4B ) , these mutants also have a greater delay in onset of PRR5 transcript accumulation ( Figure 5—figure supplement 1A ) . We next examined expression levels of other clock genes in the triple mutant . The evening-phased genes , PRR5 and TOC1 , show a significant delay in onset of expression and reduced levels in the triple mutants compared to wild-type in both light/dark ( LD ) cycles ( Figure 5A , B ) and in constant light ( LL ) ( Figure 5E–F ) . Although the peak phase of PRR5 is not altered in the mutant in LD conditions ( Figure 5A ) , the delay in the timing of increasing PRR5 expression in rve4 rve6 rve8 in the afternoon suggests that this is due to complex regulation of PRR5 transcript levels by both light and the circadian clock . The morning-phased clock genes CCA1 and LHY do not show any obvious differences in expression levels in rve4 rve6 rve8 and wild-type plants during the day either when grown in LD ( Figure 5C , D ) or in LL ( Figure 5G , H ) . However , these two morning-phased genes display slightly reduced transcript levels in the late night when grown in LD ( ZT 21 ) ( Figure 5C , D ) . This might be explained either by the long period phenotype of the rve4 rve6 rve8 mutants or by elevated TOC1 levels at the end of the night ( Figure 5B ) since TOC1 is a repressor of CCA1 and LHY ( Gendron et al . , 2012; Huang et al . , 2012; Pokhilko et al . , 2012 ) . 10 . 7554/eLife . 00473 . 011Figure 5 . Expression of clock genes is altered in rve4 rve6 rve8 triple mutants . ( A ) , ( B ) , ( E ) , ( F ) , ( I ) , ( J ) , ( M ) and ( N ) Expression of evening genes in Col and rve4 rve6 rve8 . ( C ) , ( D ) , ( G ) , ( H ) , ( K ) , ( L ) , ( O ) and ( P ) Expression of morning genes in Col and rve4 rve6 rve8 . ( A ) – ( D ) transcript levels in diurnal cycles . Seedlings were grown in LD for 7 days . White box: day , grey box: night . Data in ( A–D ) are double plotted to facilitate comparisons . ( E ) – ( P ) Transcript levels in LL . ( E ) – ( H ) Gene expression plotted on a linear scale . ( I ) – ( L ) The data shown in ( E–H ) are plotted with a log10 scale on the y-axis to better visualize differences in trough levels between the two genotypes . Horizontal brackets highlight the phase delay between Col and rve4 rve6 rve8 mutants . ( M ) – ( P ) Transcript levels derived from a 1-hr resolution time course are presented with either every time point ( M and O ) or every third time point ( N and P ) displayed . Green arrows highlight the phase difference between Col and rve4 rve6 rve8 mutants at ZT30 . Transcript levels were determined by qRT-PCR and normalized to PP2A . Values represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 01110 . 7554/eLife . 00473 . 012Figure 5—figure supplement 1 . Clock gene expression in wild type and the rve4 rve6 rve8 mutant . ( A ) PRR5 expression in Col , rve6 rve8 and rve4 rve6 rve8 in LL . Seedlings were grown in LD for 7 days , released to constant light , and then harvested at the indicated times after the last dark-to-light transition . ( B ) and ( C ) Expression of GI and PRR9 in diurnal cycles . Seedlings were grown in LD for 7 days . White box: day , grey box: night . Values represent mean ± SEM . Data in ( B ) and ( C ) are double plotted to facilitate comparisons . ( D ) and ( E ) Transcript levels of GI and PRR9 in LL . Seedlings were entrained in 12:12 LD for 7 days and then released to constant light at time 0 . Samples were harvested at the times indicated . RNA was isolated and qRT-PCR was performed . Expression levels are normalized to PP2A . Data are presented as mean ± SEM from three technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 012 In addition to these expression level changes , de-synchronization between the evening and morning genes is observed in the rve4 rve6 rve8 triple mutants . At the end of the second subjective day in LL ( around ZT36 ) , the peak times of PRR5 and TOC1 transcript accumulation are delayed approximately 6 hr in the triple mutant relative to wild-type ( Figure 5E , F ) . In contrast , in these same samples , an approximately 3-hr phase delay is observed for trough levels of CCA1 and LHY in the mutant relative to controls ( Figure 5K , L ) . Similarly , at the third subjective morning ( around ZT48 ) , PRR5 and TOC1 trough levels display an approximately 9-hr phase delay in the triple mutant ( Figure 5I , J ) while an approximately 6-hr phase delay is observed in the peak expression levels for CCA1 and LHY at that time ( Figure 5G , H ) . This greater phase delay for evening compared to morning genes can be seen more clearly when plants grown in constant conditions are sampled at 1-hr intervals ( Figure 5M , O ) . In addition , a significant change in the waveform of the evening gene TOC1 ( Figure 5M ) but not the morning gene LHY ( Figure 5O ) is observed in the triple mutant in this high-resolution time course . Notably , the obvious change in the TOC1 waveform is lost when these same data are plotted at 3-hr resolution ( compare Figure 5N , M ) . Our data show that loss of multiple RVEs has an immediate effect on expression of evening genes and a delayed effect on morning genes , further supporting the main role of RVE8 as an activator of evening genes . Given the highly reticulated nature of the circadian network , altered expression of evening genes indirectly affects expression of morning genes . Similarly reduced and delayed expression was also observed for GI , an evening-phased EE-containing clock gene , in LD and in LL ( Figure 5—figure supplement 1B , D ) . Interestingly , the EE-containing day-phased gene , PRR9 also showed reduced levels in LD ( Figure 5—figure supplement 1C ) and on the third day in LL ( Figure 5—figure supplement 1E ) , consistent with PRR9 being directly activated by RVE8 as suggested by the induction experiments carried out in the presence of an inhibitor of translation ( Figure 3—figure supplement 1E ) . To further explore regulatory interactions between RVE8 and other clock components , we examined RVE8 expression in several clock mutants , including toc1-4 ( Hazen et al . , 2005a ) , lux-1 ( Hazen et al . , 2005b ) and CCA1-OX ( Wang and Tobin , 1998 ) . RVE8 expression is significantly reduced in all of these clock mutants in LD ( Figure 6A ) , indicating that TOC1 , LUX and CCA1 directly or indirectly regulate RVE8 expression . Since TOC1 , LUX , and CCA1 are thought to directly regulate expression of one or more of the PRR5 , 7 , and 9 genes ( Farre et al . , 2005; Helfer et al . , 2011; Huang et al . , 2012 ) , we hypothesized that reduced RVE8 expression in the toc1-4 , lux-1 and CCA1-OX mutants might be due to up-regulation of the PRRs . Indeed , we found that at dawn ( ZT0 ) , when RVE8 transcript levels normally peak ( Farinas and Mas , 2011; Rawat et al . , 2011 ) , multiple PRR genes are up-regulated in each of these mutants ( Figure 6B ) . These results are consistent with a model in which the PRRs directly repress RVE8 expression and other clock genes indirectly control RVE8 expression via regulation of PRR5 , 7 , and 9 ( Figure 6C ) . This model is supported by the increased RVE8 expression seen in prr5 prr7 prr9 mutants ( Rawat et al . , 2011 ) and the reported direct binding of PRR5 to the RVE8 promoter ( Nakamichi et al . , 2012 ) . Our findings that both PRR5 and PRR9 are directly activated by RVE8 ( Figure 3A; Figure 3—figure supplement 1E ) and that peak transcript levels of these genes are reduced in rve4 rve6 rve8 mutants ( Figure 5A , E; Figure 5—figure supplement 1C , E ) further support the model that the PRRs and RVEs regulate each other to form a negative transcriptional feedback loop . 10 . 7554/eLife . 00473 . 013Figure 6 . RVE8 expression is likely controlled by other clock genes through PRR5 , 7 and 9 . ( A ) RVE8 expression in Col , toc1-4 , lux-1 and CCA1-OX in LD . 7-day-old seedlings were collected at the times indicated and qRT-PCR was performed . Data are double-plotted to facilitate visualization . Values represent mean ± SEM . ( B ) Transcript levels of PRR5 , PRR7 and PRR9 at ZT 0 ( when RVE8 transcript levels normally peak ) in wild-type ( Col ) , toc1-4 , lux-1 , and CCA1-OX . Expression levels are normalized to PP2A . Data are represented as mean ± SEM from three technical replicates . ( C ) A proposed clock model integrating RVE8 as an activator of evening clock genes . The relative time of action of each component during diurnal cycles is shown from left to right . White box: day , grey box: night . REVEILLE/CCA1/LHY family proteins are shown in yellow; pseudo-response regulators are shown in blue; the evening complex components are shown in green . Clock components with one or more EE in their promoter regions are marked with red boxes . Red solid arrow: activation , red dashed arrow: activation only displayed in specific condition ( red arrows are based on the current study ) , black perpendicular bars: repression , black arrow: activation . In this study , we demonstrated that RVE8 directly activates multiple evening-phased clock and output genes and that RVE8 is regulated by TOC1 , LUX and CCA1 , likely indirectly through their control of PRR5 , 7 and 9 expression . For clarity , only transcriptional regulation is represented . DOI: http://dx . doi . org/10 . 7554/eLife . 00473 . 013 Circadian rhythms coordinate numerous physiological and behavioral events with the appropriate time of day , in large part through genome-wide circadian regulation of gene expression ( Lowrey and Takahashi , 2011; Farre and Weise , 2012 ) . Mechanisms governing the precise timing of circadian clock and output gene expression are therefore of great interest . It has previously been reported in both plants and mammals that central clock genes can directly regulate many output genes ( Gendron et al . , 2012; Menet et al . , 2012; Nakamichi et al . , 2012 ) . Our finding that hundreds of clock-regulated genes are induced or repressed upon induction of RVE8 suggests RVE8 is an important regulator of both the clock itself and output pathways . It seems likely that RVE8 is a direct activator of many evening-phased genes . EE sequences are significantly enriched among RVE8-induced targets relative to all evening-phased genes ( Table 1A ) . In addition , RVE8 binds to EE sequences in vivo and in vitro ( Gong et al . , 2008; Rawat et al . , 2011 ) and plants mutant for RVE8 and its close homologs RVE4 and RVE6 have lost an afternoon-phased EE binding activity ( Figure 4D ) . Furthermore , for all genes tested , activation of evening-phased genes by RVE8 does not require new protein synthesis ( Figure 3 ) . In contrast , genes repressed upon induction of RVE8 activity are primarily morning-phased and are not enriched for any promoter motif relative to all clock-regulated morning-phased genes ( Table 1B ) , suggesting RVE8 regulates these genes indirectly . In support of a largely indirect role for RVE8 in repression of gene expression , inhibition of translation reduced or eliminated decreases in gene expression upon RVE8 induction ( Figure 3—figure supplement 1D–G ) . Thus RVE8 is unique among Arabidopsis clock genes in that it acts primarily , and perhaps even exclusively , as an activator of gene expression . Unlike CCA1 and LHY , which were shown to have similar activity at different time of day in ethanol-inducible lines ( Knowles et al . , 2008 ) , we have found that RVE8 activity is gated with maximum activity in the afternoon ( Figure 1D ) . This discrepancy in gating regulation may explain why overexpression of CCA1 or LHY causes arrhythmicity ( Schaffer et al . , 1998; Wang and Tobin , 1998 ) while overexpression of RVE8 instead causes an advanced phase and short period phenotype ( Rawat et al . , 2011 ) . RVE8 transcript levels peak at dawn , but RVE8 protein levels peak in the subjective afternoon ( Rawat et al . , 2011 ) . Most RVE8-induced transcripts have a peak circadian phase between CT8 and CT12 ( Figure 2G ) , approximately 2–6 hr after the peak phase of RVE8 protein levels . This delay in RVE8 target gene transcript accumulation relative to RVE8 protein might be explained by antagonistic regulation of target genes by RVE8 and the cycling repressors CCA1 and LHY . CCA1 and LHY protein levels peak in the subjective morning ( Wang and Tobin , 1998; Kim et al . , 2003 ) , well before RVE8 . A mathematical model investigating the consequences of oppositely acting transcription factors on regulation of a common target gene predicted that when the phase of a cycling transcriptional repressor precedes that of a cycling transcriptional activator ( ‘repressor-precedes activator’ ) , the peak phase of expression of the output would occur after that of the activator ( Ueda et al . , 2005 ) . The genes both induced by RVE8 and containing EE motifs in their promoter regions are the most likely direct targets of RVE8 . Although most clock-controlled genes containing an EE have an evening phase , the EE-containing RVE8-induced genes are more specifically enriched for an early-evening phase ( Figure 4A ) , suggesting that RVE8 controls a subset of EE-containing genes . How these RVE8 targets are distinct from the rest of the EE-containing CCGs remains unclear . The clock may fine-tune expression of EE-containing genes through the action of multiple clock-controlled promoter motifs , generating the wide range of phases seen across all EE-containing genes ( Figure 4A ) . For example , it has been reported that a combination of morning- , day- , and night-phased DNA elements generates the day-phased expression of Cry1 in mammalian cells . In this case , the strength of night-phased repressors relative to the day-phased activators modulates the extent of phase delay ( Ukai-Tadenuma et al . , 2011 ) . The RVE8 homologs RVE4 , 5 and 6 have also been found associated with the EE in extracts made from plants harvested in the afternoon , suggesting that they might act in a similar manner to RVE8 ( Rawat et al . , 2011 ) . This possibility is supported by the further lengthening in circadian period seen in higher order mutants combining rve4 or rve6 with rve8 ( Figure 4B ) , suggesting these factors play partially redundant roles in speeding up the pace of the clock . The loss of afternoon-phased EE binding activity seen in the rve4 rve6 rve8 triple mutants but not in rve8 single mutants ( data not shown ) suggests these RVEs contribute to the activity of the clock-regulated afternoon-phased EE activator . Among the evening-phased central clock genes examined , all show significantly reduced and delayed expression in LD and in LL in rve4 rve6 rve8 ( Figure 5 and Figure 5—figure supplement 1 ) . The long period in rve4 rve6 rve8 mutants might in principle be due either to a decrease in peak levels or a delay in onset of expression of evening genes . However , consideration of the phenotypes of plants mutant for various evening-phased clock genes makes us favor the latter possibility . toc1 and prr5 mutants have short-period phenotypes ( Strayer et al . , 2000; Eriksson et al . , 2003; Yamamoto et al . , 2003 ) ; loss of GI causes a short period in most conditions ( Park et al . , 1999; Mizoguchi et al . , 2005; Martin-Tryon et al . , 2007 ) ; and lux and elf4 mutants are arrhythmic ( Doyle et al . , 2002; Onai and Ishiura , 2005; Hazen et al . , 2005b ) . Therefore , reduced expression of any of these EE-containing evening genes is unlikely to cause the long period phenotype displayed by rve4 rve6 rve8 . On the other hand , the delayed phase of expression of clock genes can first be observed in evening-phased genes and only later in morning-phased genes ( Figure 5 ) . This suggests that the long period seen in rve4 rve6 rve8 is mainly caused by delayed expression of evening genes , which then indirectly causes a delayed phase of expression of morning genes . In support of this idea , in RVE8-overexpressing plants ( which have a short-period phenotype ) , the peak phase of expression of TOC1 is clearly advanced soon after plants are released into free-run whereas phase advances are not seen for the morning-phased genes CCA1 and LHY until much later ( Rawat et al . , 2011 ) . Similarly , delays in the phase of post-transcriptional processes have previously been suggested to contribute to long-period phenotypes in animals ( Rothenfluh et al . , 2000; Syed et al . , 2011 ) . Most clock components in Arabidopsis are either regulated by the EE ( including most evening-phased genes and one day-phased gene , PRR9 ) or regulate the EE ( two morning-phased components , CCA1 and LHY , and the afternoon-phased activator , RVE8 ) ( Figure 6C ) . However , plants mutant for CCA1 and LHY , the sole previously defined circadian regulators of EE-containing clock genes , have persistent circadian rhythms , albeit with a short period ( Alabadi et al . , 2002; Mizoguchi et al . , 2002; Locke et al . , 2005 ) . Our discovery that RVE8 and its homologs are activators of the EE may explain the rhythmicity of cca1 lhy mutants . As modeled using Ueda et al's ‘repressor-precedes-activator’ formula ( Ueda et al . , 2005 ) , inhibition in the morning by CCA1 and LHY and activation by RVE8 in the afternoon would result in rhythmic expression of EE target genes with peak expression delayed relative to peak RVE8 protein levels . Reduction or loss of activity of the cycling repressor function ( CCA1/LHY ) would result in a phase advance , causing earlier expression of EE-containing target genes , but rhythms would persist due to clock-regulated RVE8 activity . Such a phase advance and consequent short-period phenotype is indeed observed in cca1 and lhy single and double mutants ( Green and Tobin , 1999; Mizoguchi et al . , 2002 ) . Interestingly , CCA1/LHY and RVE8 contain a similar Myb-like DNA binding domain and belong to the same family of transcription factors ( Rawat et al . , 2009 , 2011 ) . Even though they have distinct biochemical functions , with CCA1 and LHY serving as repressors and RVE8 as an activator of EE-containing genes , both CCA1/LHY and RVE8 are responsible for shaping the circadian pattern of expression of evening-phased genes . This joint regulation of common targets may explain why circadian rhythms persist upon mutation of the repressor Mybs or the activator Mybs alone . Current models of the plant clock suggest that it is composed of transcription factors that are primarily repressors of gene expression which interact to form interlocked morning and evening feedback loops ( Gendron et al . , 2012; Huang et al . , 2012; Pokhilko et al . , 2012 ) . However , our findings substantially revise this view . We have demonstrated that the RVEs are an integral part of the circadian oscillator but are primarily ( and perhaps exclusively ) activators of gene expression . In addition , our findings suggest that the view of the plant clock as constituted of coupled morning and evening transcriptional feedback loops is inadequate . RVE8 itself , with its morning-phased peak in transcript levels but afternoon-phased peak in protein levels ( Rawat et al . , 2011 ) , doesn't fit neatly into either the ‘morning’ or ‘evening’ category . Furthermore , the highly interconnected nature of the regulatory interactions underlying the plant clock ( Figure 6C ) make it virtually impossible to identify discrete regulatory feedback loops and suggest that the plant clock is best viewed as a highly interconnected , complex regulatory network . The RVE8::RVE8:GR construct was created using a PCR fusion-based approach ( Hobert , 2002 ) , placing a 2 . 5 kb genomic fragment of RVE8 ( containing ∼0 . 7 kb upstream of the translational start site ) and a 1 . 7 kb DNA fragment containing the GR coding sequence and OCS 3′ from pART7-GR ( donated by John Harada ) together . The PCR fusion product was then cloned into the NotI site in the binary vector pML-BART . The 35S::RVE8:GR construct was created by cloning RVE8 coding sequence into pART7-GR via XhoΙ and SmaΙ sites , and then subcloning into the NotI site in the binary vector pML-BART . The RVE8::RVE8:GR and 35S::RVE8:GR clones were transformed into rve8-1 CCR2::LUC+ via the floral dip method ( Zhang et al . , 2006 ) . Homozygous single-insertion site transformants were selected based on BASTA resistance in the T2 and T3 generations . T-DNA insertion mutants rve4-1 ( Salk_137617 ) and rve6-1 ( Salk_069978 ) ( Alonso et al . , 2003 ) were obtained from the Arabidopsis Biological Resources Center . Homozygous mutants were identified by PCR of genomic DNA using primers flanking the insertion site and complementary to the T-DNA left border ( primers are listed in Supplementary file 3 ) . rve4-1 and rve6-1 were crossed to rve8-1 CCR2::LUC+ to generate rve4 rve8 and rve6 rve8 double mutants and rve4 and rve6 single mutants , all carrying the CCR2::LUC+ reporter . The rve4 6 8 triple mutant was created by crossing rve4 rve8 CCR2::LUC+ and rve6 rve8 CCR2::LUC+ . lux-1 , toc1-4 and CCA1-OX were previously described ( Wang and Tobin , 1998; Hazen et al . , 2005a , 2005b ) . rve8-1 and rve8-1 RVE8::RVE8:GR seeds were sterilized and stratified on fine nylon mesh ( Small Parts , Logansport , IN; 100 µM 44% ) on Murashige and Skoog ( MS ) agar media containing 3% sucrose at 4°C in the dark for 2 days . The seedlings were grown under 12-hr light:12-hr dark condition with 50–60 µmol/m2/s white fluorescent light at 22°C for 7–8 days . At ZT4 ( 4 hr after lights on ) , the mesh and seedlings were transferred to liquid MS media containing 3% sucrose with 30 µM DEX ( Sigma D1881 , St . Louis , MO; 60 mM DEX stock solution was made in ethanol and stored at −20°C ) or 0 . 05% ethanol ( mock treatment ) . For cycloheximide treatment , 200 µM CHX ( Sigma C4859; stock solutions were 100 µg/µl in DMSO ) or 0 . 056% DMSO ( mock treatment ) was added at the time of DEX or ethanol mock treatment . After 2 or 4 hr incubation as indicated with gentle agitation , plants were quickly harvested , frozen in liquid nitrogen and stored in −80°C until processed . Total RNA from three biological replicates ( ∼30 plants each ) for each condition was isolated using Trizol ( Invitrogen , Grand Island , NY ) , treated with DNase ( Qiagen , Germantown , MD ) , and purified using the RNeasy MinElute Cleanup Kit ( Qiagen ) . The quality of the isolated total RNA was determined by NanoDrop ND 1000 ( NanoDrop Technologies , Wilmington , DE ) . Samples with both a 260 nm:280 nm ratio and a 260 nm:230 nm ratio between 2 and 2 . 3 were processed further . The RNA-seq libraries were prepared using a customized Illumina-based strand-specific multiplex library construction protocol modified from Wang et al . ( 2011 ) . Briefly , mRNA was isolated from 8 µg of total RNA using Dynabeads mRNA DIRECT Kit ( Invitrogen ) and fragmented to ∼200 nucleotide pieces . After the first strand cDNA synthesis was carried out using random primers , the second strand cDNA was synthesized using a special dNTP mix in which dTTP is replaced by dUTP . Following end-repair ( Y9140-LC-L; Enzymatics , Beverly , MA ) and addition of a dA to the 3′ end , both ends of cDNA were ligated with Y-shaped adaptors containing an index unique to each library . The second strand cDNA was then digested using Uracil DNA glycosylase ( Enzymatics ) . Primers partially complementary to the adaptor sequences were used to amplify the libraries for 12 PCR cycles using High-Fidelity Polymerase ( Phusion , Ipswich , MA ) . The libraries were further size-selected using a 1:1 volume of AMPure XP beads ( Beckman Coulter , Brea CA ) . The size and quality of resulting libraries were examined using a Bioanalyzer 2100 ( Agilent , Santa Clara , CA ) . The 12 libraries were then quantified by qPCR and equally pooled for 2 lanes of single end 50 bp sequencing in HiSeq 2000 machine ( Illumina , San Diego , CA ) . The adaptors containing index sequences and primers used for amplification are listed in Supplementary file 3 . The raw reads ( ∼310 . 3 million reads ) were initially subjected to quality filtering to remove low quality reads using the FASTX-toolkit ( Pearson et al . , 1997 ) with the following parameters ( −q 20 , minimum quality score to keep: 20; −p 85 , minimum percent of bases that must satisfy the quality score cut-off: 85 ) . A custom perl script was then used to remove Illumina adapter sequences from the resulting reads . The reads were then separated by their custom barcode sequences ( de-multiplexing ) using Fastx_barcode_splitter ( included in the FASTX toolkit ) allowing up to one mismatch per barcode . 16 to 22 million reads per libraries were obtained and aligned against the Arabidopsis cDNA representative_gene_model ( TAIR 10 ) using BWA ( Li and Durbin , 2009 ) and Samtools ( Li et al . , 2009 ) . ( The parameter used to map the reads for BWA was aln -l 20 . ) The resulting BAM files from the two lanes were merged using Samtools and then converted to SAM files . The reads from these SAM files were then separated based on their alignment to the forward or reverse strand . Only the reads mapped to the reverse strand were used to calculate the read counts using a custom R script , and these counts were then used in analysis of differential expression . edgeR was used to generate the pseudo-normalized counts for visualization and to carry out differential gene expression analysis ( Robinson et al . , 2010 ) using R 2 . 14 . 1 ( R Development Core Team , 2011 ) . Transcripts that have at least one count per million in at least three samples were considered expressed genes and kept for downstream analysis . Exact tests were performed using tagwise dispersion and the prior n was set to 6 . 25 . FDR 0 . 01 was used as a cut-off for differentially expressed genes . Genes significantly differentially expressed between the mock- and DEX-treated transgenic line ( rve8 RVE8::RVE8:GR ) , or between the DEX-treated rve8 RVE8::RVE8:GR and rve8 plants , were grouped into RVE8-induced or RVE8-repressed genes . Genes that are responsive to DEX treatment in rve8 mutant ( i . e . , the genes significantly differentially expressed between ‘rve8 + DEX’ and ‘rve8 + mock’ ) were removed from the RVE8-induced and RVE8-repressed gene lists . Only the genes uniquely induced or repressed by RVE8 ( i . e . , not showing the same trend in rve8 ) were used for further analysis . The significant gene sets ( both RVE-regulated or DEX-regulated ) are listed in Supplementary file 1C–F . Circadian phases of the 583 RVE8-induced genes and 850 RVE8-repressed targets were determined in a previous study using JTK_CYCLE ( Hsu and Harmer , 2012 ) . The 376 RVE8-induced cycling genes ( 64% of the induced genes ) and 525 RVE8-repressed cycling genes ( 62% of the repressed genes ) were subjected to phase and motif analysis . Distributions of the phases of the RVE8-induced and -repressed clock-regulated genes were plotted using the density function in R ( R Development Core Team , 2011 ) . Overrepresented motifs in the promoters of RVE8-regulated genes were identified using the SCOPE motif finder ( Carlson et al . , 2007 ) . Fixed regions of 1500 bp upstream of the translational start site ( corresponding to both strands ) of RVE8-regulated genes were used for computation of significance compared to all the genes in the genome . Significance is the negative logarithm of expectation . Significance greater than zero is statistically meaningful; the larger the significance value , the higher its statistical significance . Coverage indicates the percentage of genes that have at least one occurrence of the motif in question . The fractions of genes containing the top-scoring motifs among the evening-phased ( CT 8 to CT 14 ) and morning-phased ( CT 20 to CT 2 ) RVE8 targets were compared to the fractions found in all of the clock-regulated genes in the corresponding phase group . Fisher's exact test was performed in R to examine if the presence of these motifs in RVE8 targets is enriched compared to their presence in the evening- and morning-phased genes . For gene expression in diurnal cycles , around 30 seedlings per sample were grown under 12 hr white light ( 50–60 µmol/m2/s , generated using cool white fluorescent bulbs ) :12 hr dark at 22°C for 7 days and harvested at the times indicated . For gene expression in free-run , seedlings were released to constant white light after entrainment in diurnal cycles for 7 days , and harvested at the times indicated . RNA was isolated using Trizol ( Invitrogen ) and was then treated with DNase ( Qiagen ) . cDNA was synthesized using SuperScriptase ΙΙ ( Invitrogen ) following the manufacturer's protocol . qRT-PCR was performed as previously described ( Martin-Tryon et al . , 2007 ) . Three technical triplicates for each sample were run using iQ5 Real Time PCR machine ( Bio-Rad , Hercules , CA ) , and starting quantity was estimated from critical thresholds using the standard curve method . Data for each sample were normalized to the respective PROTEIN PHOSPHATASE 2A ( PP2A ) expression level . The primer sets for each transcript are listed in Supplementary file 3 . Luciferase imaging was performed and analyzed as previously described ( Martin-Tryon et al . , 2007 ) . Seedlings were entrained in 12 hr white light ( 50–60 µmol/m2 . /s; cool white fluorescent bulbs ) :12 hr dark at 22°C for 6 days before being released to constant red plus blue light ( 33µEi red light , 20µEi blue light ) for luciferase activity analysis using an ORCA ΙΙ ER ( Hamamatsu , Bridgewater , NJ ) CCD camera . Illumination was provided by monochromatic red and blue LED lights ( XtremeLux , Santa Clara , CA ) . Images were analyzed using MetaMorph ( Molecular Devices , Sunnyvale , CA ) and free-running periods were estimated using Fast Fourier Transform Non-Linear Least Squares ( Plautz et al . , 1997 ) . 11-day-old seedlings grown in 12 hr white light ( 50–60 µmol/m2/s; cool white fluorescent bulbs ) :12 hr dark cycle at 22°C were harvested at the times indicated . Plant whole-cell extracts were made and the electrophoretic mobility shift assay was performed as previously described ( Harmer and Kay , 2005 ) . Briefly around 1 . 5 g of tissue per sample was harvested , frozen in liquid nitrogen immediately and stored at −80°C until processed . The frozen tissue was ground to a fine powder , suspended in homogenization buffer ( 15 mM HEPES , pH 7 . 6 , 40 mM KCl , 5 mM MgCl2 , 1 mM DTT , 0 . 1 mM PMSF , and 1X complete protease inhibitor cocktail ) and ( NH4 ) 2SO4 was added to 0 . 4 M . The insoluble components were pelleted by ultracentrifugation and removed , then solid ( NH4 ) 2SO4 was added to the supernatant to ∼90% saturation . Proteins were pelleted by ultracentrifugation , resuspended in resuspension buffer ( 20 mM HEPES , pH 7 . 6 , 40 mM KCl , 0 . 1 mM EDTA , 10% glycerol , 1 mM DTT , 0 . 1 mM PMSF , and 1× complete protease inhibitor cocktail ) , and dialyzed using dialysis cartridges ( 7000 MWCO Slide-A-Lyzer; Pierce , Rockford , IL ) against dialysis buffer ( 20 mM HEPES , pH 7 . 2 , 40 mM KCl , 0 . 1 mM EDTA , 10% glycerol , 2 . 5 mM DTT , 0 . 1 mM PMSF ) . The dialyzed proteins were quantified , aliquoted and saved at −80°C until used . 15 µg of the dialyzed protein was incubated with 20 fmol of radiolabelled double-stranded DNA containing the EE and flanking sequences from the CCR2 promoter in reaction buffer ( 20 mM HEPES , pH 7 . 2 , 80 mM KCl , 0 . 1 mM EDTA , 10% glycerol , 2 . 5 mM DTT , 8 ng/µl poly [dI-dC] ) with or without the competitors as indicated for 15 min at room temperature . A 50-fold molar excess of unlabeled CCR2-EE ( WT competitor ) or mutated CCR2-EE ( mutated competitor ) DNA was added as indicated for binding-specificity control . The binding assays were resolved by electrophoresis on 5% non-denaturing polyacrylamide gels . The dried gel was imaged using a Storm PhosphorImager ( Molecular Dynamics , Sunnyvale , CA ) . The probe and competitor DNA sequences are listed in Supplementary file 3 .	We live in a world with a 24-hr cycle in which day follows night follows day with complete predictability . Life on earth has evolved to take advantage of this predictability by using circadian clocks to prepare for the coming of night ( or day ) , and plants are no exception . Even in constant darkness , characteristics such as leaf movements show a constant cycle of around 24 hr . Most circadian clocks rely on negative feedback loops involving various genes and proteins to keep track of time . In one of these feedback loops , certain genes—called morning-phased genes—are expressed as proteins during the day , and these proteins prevent other genes—called evening-phased genes—from producing proteins . As night approaches , however , a second feedback loop acts to stop the morning-phased genes being expressed , thus allowing the evening-phased genes to produce proteins . And as day approaches , expression of these genes is stopped and the whole cycle starts again . Many of the genes and proteins involved in the circadian system of Arabidopsis thaliana , a small flowering plant that is widely used as a model organism , have been identified , and its circadian clock was thought to rely almost entirely on proteins called repressors that block the transcription of genes . Now , Hsu et al . have shown that the Arabidopsis clock also involves proteins that increase the expression of certain genes at specific times of the day . Hsu et al . focused on the promoter regions of evening-phased genes: these regions are stretches of DNA that proteins called transcription factors bind to and either encourage the expression of a gene ( if the protein is a transcriptional activator ) or block its expression ( as a transcriptional repressor ) . In particular , they focused on a protein called RVE8 that is most strongly expressed in the afternoon and , based on previous research , is thought to activate the transcription of genes . Using genetically modified plants in which the gene for RVE8 can be turned on and off , they found that this protein led to increases in the expression of some genes , and reductions in the expression of others . Further analysis showed that RVE8 was able to activate the expression of evening-phased genes directly , without requiring that new proteins be made first . By contrast , morning-expressed genes were likely to be suppressed by RVE8 via an indirect mechanism that involved other proteins that had previously been activated by RVE8 . The expression of RVE8 itself is regulated by other clock genes and also by an undefined post-transcriptional process . Therefore rather than consisting of a morning feedback loop coupled to an evening feedback loop , with both loops being based on repressors , the plant clock is instead better viewed as a highly connected network of activators and repressors . Further research is clearly necessary to understand this unexpected complexity in the circadian clock of Arabidopsis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2013	Accurate timekeeping is controlled by a cycling activator in Arabidopsis
Dopamine neurons are thought to encode novelty in addition to reward prediction error ( the discrepancy between actual and predicted values ) . In this study , we compared dopamine activity across the striatum using fiber fluorometry in mice . During classical conditioning , we observed opposite dynamics in dopamine axon signals in the ventral striatum ( ‘VS dopamine’ ) and the posterior tail of the striatum ( ‘TS dopamine’ ) . TS dopamine showed strong excitation to novel cues , whereas VS dopamine showed no responses to novel cues until they had been paired with a reward . TS dopamine cue responses decreased over time , depending on what the cue predicted . Additionally , TS dopamine showed excitation to several types of stimuli including rewarding , aversive , and neutral stimuli whereas VS dopamine showed excitation only to reward or reward-predicting cues . Together , these results demonstrate that dopamine novelty signals are localized in TS along with general salience signals , while VS dopamine reliably encodes reward prediction error . Animals respond to new stimuli in a characteristic way across species , historically characterized as an ‘orienting reflex’ or a ‘what is it reflex’ ( Pavlov and Anrep , 1927; Sechenov , 1935; Sokolov , 1963 ) . Detection of novel stimuli is advantageous for survival because novel stimuli can signal potential rewards or potential threats . Orienting towards a novel stimulus and understanding it through exploration can allow future exploitation of potential rewards . In addition to behavioral advantages , novelty detection is fundamental for computation in our brain . For example , novelty detectors , or ‘novelty filters’ ( Kohonen and Oja , 1976; Marsland et al . , 2002 ) , can reduce the amount of total information so that we can focus on unexpected perceptions as inputs to pay attention to and to learn from . Indeed , behavioral studies have repeatedly shown that both humans and other animals have enhanced memory for novel items ( Kishiyama et al . , 2009; Restorff , 1933 ) . Physiologically , it is widely accepted that novelty responses are distributed over a network of many brain areas ( Courchesne et al . , 1975; Kishiyama et al . , 2009; Knight , 1996 ) . Among these , single unit recordings have shown that dopamine neurons in the midbrain increase their firing in response to the presentation of a novel stimulus in several species and behavioral paradigms ( Horvitz et al . , 1997; Ljungberg et al . , 1992; Schultz , 2015; Steinfels et al . , 1983 ) . As animals experience the repeated association of a stimulus and a reward , they learn to expect the reward when the stimulus is presented ( Pavlov and Anrep , 1927 ) . Dopamine neurons are thought to be the neural substrate underlying this type of learning because they signal reward prediction error: the difference between actual and expected reward values ( Bayer and Glimcher , 2005; Bromberg-Martin et al . , 2010; Clark et al . , 2012; Cohen et al . , 2012; Schultz et al . , 1997 ) . These neurons are thought to guide decision-making by broadcasting this information to many regions of the forebrain and reinforcing behaviors that lead to reward ( Barto et al . , 1999; Dayan and Niv , 2008; Haber , 2014; Montague et al . , 2004; Steinberg et al . , 2013 ) . Novelty responses in dopamine neurons ( Horvitz et al . , 1997; Schultz , 2015 ) were initially puzzling because animals cannot know whether a novel stimulus will reliably predict an outcome with a positive or negative value . One hypothesis was that dopamine neurons take an optimistic approach toward novel stimuli , assuming that they will predict a valuable outcome until proven wrong ( Hazy et al . , 2010; Kakade and Dayan , 2002 ) . This ‘optimistic initialization’ in dopamine neurons may have advantages . For example , the novelty responses in dopamine neurons may induce orienting behaviors towards novel stimuli , similar to dopamine responses to reward or reward-predicting cues that induce orienting behaviors ( Hazy et al . , 2010; Kakade and Dayan , 2002 ) . Further , dopamine novelty responses may allow computational exploration ( Dayan and Sejnowski , 1996 ) , or storage of the novel stimulus in working memory ( Braver and Cohen , 1999 ) , so that animals have a better chance to associate novel stimuli to potential rewards ( Hazy et al . , 2010; Kakade and Dayan , 2002 ) . However , these hypotheses do not necessarily fit with conflicting observations of animals’ behavioral responses to novel stimuli ( Gershman and Niv , 2015 ) . Indeed , depending on the experimental context , animals sometimes approach and sometimes avoid novel options compared to familiar ones ( Gershman and Niv , 2015 ) . One explanation for why some dopamine neurons respond to novel stimuli could be that some subpopulations of dopamine neurons are not strictly related to reward prediction error coding . Recent studies have shown that there is substantial diversity among dopamine neurons at the molecular level ( Grimm et al . , 2004; Lacey et al . , 1989; Lammel et al . , 2008; Roeper , 2013 ) as well as in their activity ( Brischoux et al . , 2009; Bromberg-Martin et al . , 2010 ) . For example , single unit recordings in monkeys showed that some dopamine neurons are inhibited by aversive outcomes and others are excited by them ( Matsumoto and Hikosaka , 2009 ) . This suggests that there are distinct types of dopamine neurons and that some do not encode pure value . Instead , the data suggest that some dopamine neurons encode value and others might encode ‘motivational salience’ ( the absolute value of ‘value’ ) . Recent anatomical studies have revealed that dopamine neurons with different projection targets are embedded in separate circuits . The entire set of inputs to dopaminergic nuclei includes a large number of regions ( Geisler et al . , 2007; Geisler and Zahm , 2005 ) . Neural circuit tracing using a modified rabies virus ( Wickersham et al . , 2007 ) enabled us to specifically label the monosynaptic inputs onto dopamine neurons , revealing that the ventral tegmental area ( VTA ) and the substantia nigra compacta ( SNc ) dopamine neurons receive slightly different inputs ( Watabe-Uchida et al . , 2012 ) . More recent studies have shown that dopamine neurons with different projection targets receive different inputs ( Beier et al . , 2015; Lerner et al . , 2015; Menegas et al . , 2015 ) . Specifically , we found that dopamine neurons projecting to the posterior ‘tail’ of the striatum ( TS ) have unique inputs compared to dopamine neurons projecting to many other brain regions , including the ventral striatum ( VS ) , suggesting that these neurons could have a distinct function ( Menegas et al . , 2015 ) . Based on our previous anatomical findings , in this study , we compared the dopamine axon activity in VS and TS while mice learned new odor-outcome associations ( we will call the bulk calcium signal that we observed from the axons of DAT+ midbrain dopamine neurons in the striatum ‘VS dopamine’ and ‘TS dopamine’ in the following sections ) . Our results revealed opposite dynamics for learning new cue-outcome associations in VS dopamine and TS dopamine . We observed a large response to novel cues in TS dopamine which subsequently decreased over the course of associative learning . On the other hand , we saw no response to novel cues in VS dopamine . Instead , VS dopamine gradually developed responses to reward-predicting cues during learning . These findings revealed that dopamine novelty coding is localized to the posterior part of the striatum , while VS dopamine faithfully encodes reward prediction error . Thus , novelty responses in dopamine may be better formalized separately from the reward prediction error ( RPE ) framework , rather than being included in the RPE framework . We used optical fiber fluorometry ( fiber photometry ) ( Kudo et al . , 1992 ) to record bulk calcium signals from the axons of midbrain dopamine neurons projecting to several regions of the striatum ( Kim et al . , 2016; Parker et al . , 2016 ) . We chose four regions: the ventral striatum ( VS ) , dorsomedial striatum ( DMS ) , dorsolateral striatum ( DLS ) , and the posterior tail of the striatum ( TS ) ( Figure 1 ) . To measure activity from dopamine axons in these regions , we infected midbrain dopamine neurons with a genetically encoded calcium indicator , GCaMP6m ( Akerboom et al . , 2012; Chen et al . , 2013 ) . To target dopamine neurons specifically , we injected a cre-dependent virus ( AAV-flex-GCaMP6m ) into both the VTA and SNc of transgenic mice expressing Cre recombinase under the control of dopamine transporter ( DAT-cre mice ) ( Bäckman et al . , 2006 ) crossed with reporter mice expressing red fluorescent protein ( tdTomato ) ( Jackson Lab ) . 10 . 7554/eLife . 21886 . 003Figure 1 . Recording dopamine activity across the striatum using fiber fluorometry . ( A ) Schematic of GCaMP virus injection and optic fiber implantation sites . Detailed schematic of recording setup is shown in Figure 1—figure supplement 1 . Sample raw data are shown in Figure 1—figure supplement 2 . ( B ) Distribution of optic fibers ( sagittal max-projection ) used for recording labeled red ( VS ) , orange ( DMS ) , blue ( DLS ) , and purple ( TS ) with dotted lines denoting ½ mm increments . Coronal sections are shown in Figure 1—figure supplement 3 . ( C ) Schematic of the basic trial structure . An odor cue ( CS ) ( 1 s duration ) is followed by an outcome ( US ) or no outcome after 1 s delay , followed by a random inter-trial interval ( ITI ) of 6–12 s . At a low frequency , unexpected outcomes are also delivered . ( D ) Licking in response to odors predicting reward ( blue ) or nothing ( grey ) . Odor onset is t = 0 and water delivery time is t = 2 , so anticipatory licking occurs between t = 0 and t = 2 ( quantified on the right ) . ( E ) An example of GCaMP virus infection . Green indicates AAV-flex-GCaMP6m infection ( top ) , red indicates genetically encoded tdTomato in DAT-cre-expressing neurons ( middle ) , and the bottom panel is an overlay of the two signals . Labeled axons in the striatum are shown in Figure 1—figure supplement 4 . ( F ) Example single trial responses to unpredicted water from GCaMP ( top ) and tdTomato ( middle ) from a single session in a mouse with a fiber implanted in VS . The average GCaMP signal across trials in that session are plotted in the bottom panel . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00310 . 7554/eLife . 21886 . 004Figure 1—figure supplement 1 . GCaMP6m recording and example traces . ( A ) Schematic of recording setup using 473 nm and 561 nm lasers to deliver light , and ultimately a 500 ± 20 nm bandpass filter to collect GCaMP signal and a 600 ± 20 nm bandpass filter to collect tdTomato signal . ( B ) Example raw voltage traces from each pre-amplifier are shown in green and red . The green trace corresponds to 480–520 nm light and the red trace corresponds to 580–620 nm light . Water delivery times are plotted ( blue ) , along with reward-predicting odor ( black ) and nothing-predicting odor ( grey ) delivery times . ( C ) An example of the individual ( grey ) and average ( blue ) unpredicted water responses from a single animal over the course of a month of recording every fourth day are plotted after calculating dF/F . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00410 . 7554/eLife . 21886 . 005Figure 1—figure supplement 2 . Example recording sessions . Examples of complete recording sessions from a VS-implanted fiber ( top ) and TS-implanted fiber ( bottom ) . The signal is the raw voltage continuously measured from the pre-amplifier . Scale bars indicate 1 volt and 1 min . In these sample recording sessions , the excitation laser was on continuously for ~25 min and then turned off ( traces dip sharply at the point of laser-off ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00510 . 7554/eLife . 21886 . 006Figure 1—figure supplement 3 . Distribution of recording fibers . Distribution of recording fibers in VS ( red ) , DMS ( orange ) , DLS ( blue ) , and TS ( purple ) . Each coronal image represents an optical slice that is 100 µm thick , and the fibers that fall within that range are plotted . After image registration , fiber positions were manually identified ( see Materials and methods , Figure 7—figure supplement 2 ) . Below , the summary panel from Figure 1B is duplicated . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00610 . 7554/eLife . 21886 . 007Figure 1—figure supplement 4 . Midbrain dopamine axon distribution in the striatum . The distribution of dopamine axons from anterior ( left ) to posterior ( right ) striatum . Red indicates genetically encoded tdTomato in the axons of DAT+ neurons and green indicates virally encoded GCaMP in the axons of DAT+ neurons . Top row: a mouse expressing tdTomato in dopamine neurons , with no GCaMP virus injection . Second row: GCaMP infection in the VTA only , resulting in stronger labeling of VS axons than TS axons . Third row: GCaMP infection in the SNC only , resulting in stronger labeling of TS axons than VS axons . Bottom row: GCaMP infection in the VTA and SNC , leading to labeling of dopamine axons throughout the striatum . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 007 We chronically implanted optic fibers into the striatum of these mice to deliver 473 nm and 561 nm light and collect GCaMP and tdTomato signals ( Chen et al . , 2012; Gunaydin et al . , 2014; Kim et al . , 2016 ) ( Figure 1A , Figure 1—figure supplement 1 ) . For these experiments , we continuously excited GCaMP with 473 nm light and continuously recorded GCaMP emission ( Figure 1—figure supplement 2 ) . We recorded from 65 fibers total , targeted to either VS ( n = 25 ) , DMS ( n = 8 ) , DLS ( n = 8 ) , or TS ( n = 24 ) ( Figure 1B , Figure 1—figure supplement 3 ) . In fixed tissue , we observed fluorescence from GCaMP6m+ axons primarily in the striatum of these mice ( Figure 1—figure supplement 4 ) . Mice were presented with odors paired with water delivery or no outcome ( Figure 1C ) . In some experiments , odors were paired with an aversive air puff or a mild tone . Infrequently , mice also received unpredicted water , air puff or tone ( ≤10% of trials ) . After training , mice licked with an increased frequency in response to the reward-predicting odor ( anticipating the reward ) , but not in response to odors that predicted no outcome ( Figure 1D ) , indicating that mice had learned an association between an odor and reward . We observed large responses to unpredicted reward in GCaMP , but not tdTomato , signals ( example traces shown in Figure 1E–F and Figure 1—figure supplement 1B ) and recorded from the same fibers over the course of several weeks with relative stability ( Figure 1—figure supplement 1C ) . We identified the fiber implant sites by clearing brains using CLARITY ( Chung and Deisseroth , 2013 ) , imaging them as intact volumes using a light sheet microscope ( Tomer et al . , 2014 ) , and aligning them to a single reference space ( Menegas et al . , 2015 ) . We categorized the location of fibers in the dorsal striatum into the DMS , DLS , or TS based on their medial-lateral and anterior-posterior positions ( see Materials and methods ) . VS fibers were spread throughout the core and lateral shell of the ventral striatum ( Figure 1—figure supplement 3 ) . TS fibers were located near the posterior end of the dorsal striatum ( Figure 1—figure supplement 3 ) . We will focus on VS and TS dopamine , because VS and TS dopamine displayed the most contrasting input patterns in our previous anatomical study ( Menegas et al . , 2015 ) . In order to examine dopamine activity in VS and TS during associative learning , we recorded both during the initial learning of new odor-outcome associations ( first time association , Figure 2 and Figure 3 ) and also during repeated learning where animals experienced new associations every day ( Figure 4 ) . 10 . 7554/eLife . 21886 . 008Figure 2 . Responses to novel odors in VS and TS dopamine . Comparison of VS and TS responses to novel odors in naïve animals . ( A ) Average response to the first presentation of a novel odor in VS dopamine , with SE bars . ( B ) Average responses over the course of the first 30 trials are shown in bins of 5 trials . ( C ) A heat map of responses to a novel odor over the course of a single session ( each row is one trial ) with yellow indicating an increase in signal and cyan indicating a decrease in signal . ( D–F ) TS dopamine responses to novel odors , plotted as in A–C . ( G ) Comparison of first-trial water responses in VS and TS ( left ) and first-trial responses to novel odors ( right ) . See Materials and methods . ( H ) Time course of responses to a novel odor in VS ( circles ) and TS ( squares ) over the course of 30 trials in bins of 5 trials . This data was analyzed based on odor decay rates to show that there was no large effect of odor decay in Figure 2—figure supplement 1 . Motion artifacts were examined in Figure 2—figure supplement 3 . GCaMP signal decay was measured in Figure 2—figure supplement 2 . Finally , response latencies are shown in Figure 2—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00810 . 7554/eLife . 21886 . 009Figure 2—figure supplement 1 . PID measurements of odor decay rates . ( A–B ) An example of PID measurements from an odor with a slow rate of decay: dimethoxybenzene . Plot on the left is a single trial PID measurement after 1:10 dilution , 1:20 dilution , or 1:100 dilutions . Plot on the right is the decay of the PID measurement over a session . ( C–D ) An example of PID measurements from an odor with a fast rate of decay: butenol . Plots are the same as above . ( E ) The average decay rate of slow decaying odors ( blue ) and fast decaying odors ( red ) over a session . ( F ) Average TS dopamine signal in response to novel odors with a slow decay rate ( blue ) or fast decay rate ( red ) . Data from 2 hr is plotted , separated based on odorant . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 00910 . 7554/eLife . 21886 . 010Figure 2—figure supplement 2 . GCaMP response decay within sessions . The average GCaMP responses to free water in VS ( A , D , G ) , free water in TS ( B , E , H ) , and familiar odors in TS ( C , F , I ) are plotted . Top row: average traces over the course of a session ( ~45 min ) with early responses plotted in blue and late responses plotted in red . Middle row: average peak responses plotted over the course of a session , with a linear fit of the data plotted in grey . Bottom row: heat maps of the average responses , with yellow indicating an increase in signal and cyan indicating a decrease in signal . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01010 . 7554/eLife . 21886 . 011Figure 2—figure supplement 3 . Animal body movement during trials . Heat maps from example sessions showing the total body movement following familiar odors predicting reward ( A ) , familiar odors predicting nothing ( B ) , and novel odors predicting reward ( C ) . The average trace from six animals is shown for each of these trial types ( D–F ) along with standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01110 . 7554/eLife . 21886 . 012Figure 2—figure supplement 4 . Latency of GCaMP responses to novel odors in TS . The first 5 TS dopamine GCaMP responses to a novel odor were compared across 12 naïve animals . ( A ) A histogram of the response latencies from each of the trials . The median response latency is 140 ms ( see Materials and methods ) . The bold line represents the cumulative probability that a response was observed at that time , in any trial . ( B ) An example trace from a single trial ( green ) as well as the average trace among all trials ( five trials per animal used , 12 animals total ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01210 . 7554/eLife . 21886 . 013Figure 3 . Opposite dynamics of VS and TS dopamine during initial learning of new odor-reward associations . Learning dynamics for VS dopamine ( A ) and TS dopamine ( B ) over the course of 3 weeks of training , as naïve animals learn an association between an odor and reward . Odor onset ( CS ) and water delivery time ( US ) are shown as dotted lines . Responses are compared on day 1 , day 7 , day 14 , and day 21 . The average traces are plotted in blue ( predicted reward ) and black ( unpredicted reward ) , with the standard error of the mean ( SEM ) . Individual animals’ responses can be found in Figure 3—figure supplement 1 . ( C ) Average licking in response to reward-predicting odor ( blue ) compared to average licking in response to unexpected reward ( black ) . ( D ) A quantification of the CS and US responses in VS from the above traces , over training compared to responses to unexpected water ( black ) . ( E ) A quantification of the CS and US responses in TS from the above traces , over training , compared to responses to unexpected water ( black ) . ( F ) The average number of anticipatory licks in the period between odor presentation and water delivery , compared over days of training . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01310 . 7554/eLife . 21886 . 014Figure 3—figure supplement 1 . Individual traces during initial learning of new odor-reward associations . Individual animals’ responses to ( A ) predicted water ( blue ) and ( B ) unpredicted water ( black ) over the course of learning . Average among animals is shown as a slightly darker trace . Each individual trace represents the average among trials for a single session , for that animal . The session day is indicated on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01410 . 7554/eLife . 21886 . 015Figure 4 . Opposite dynamics of VS and TS dopamine during repeated learning of new odor-reward associations . Responses to new cues predicting nothing ( A–C ) or water ( D–F ) in VS . Responses to new cues predicting nothing ( G–I ) or water ( J–L ) in TS . Responses to aversive air puffs are found in Figure 4—figure supplement 1 and quantified in Figure 4—figure supplement 2 . In the panels on the left , trials are color-coded such that red indicates the first trial and blue indicates the last trials of the session . Trials were quantified in bins of 5 trials . The middle panels show the average CS ( magenta ) and US ( green ) responses over the course of a session , again quantified in bins of 5 trials . The panels on the right are heat maps , where every line is a single trial . In these heat maps , yellow indicates an increase in signal and cyan indicates a decrease . Odor discrimination latency is quantified in Figure 4—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01510 . 7554/eLife . 21886 . 016Figure 4—figure supplement 1 . Opposite dynamics of VS and TS dopamine during repeated learning of new odor-puff associations . Responses to new cues predicting air puff in VS ( A–C ) . Responses to new cues predicting air puff in TS ( D–F ) . In the panels on the left , trials are color-coded such that red indicates the first trial and blue indicates the last trials of the session . Trials were quantified in bins of 5 trials . The middle panels show the average CS ( magenta ) and US ( green ) responses over the course of a session , again quantified in bins of 5 trials . The panels on the right are heat maps , where every line is a single trial . In these heat maps , yellow indicates an increase in signal and cyan indicates a decrease . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01610 . 7554/eLife . 21886 . 017Figure 4—figure supplement 2 . Dynamics of responses to puff-predicting odors . A comparison of the CS responses to novel air puff predicting odors and novel odors predicting nothing in VS dopamine ( left ) and TS dopamine ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01710 . 7554/eLife . 21886 . 018Figure 4—figure supplement 3 . Novel odor discrimination latency in TS dopamine . A quantification of the latency of discrimination between novel and familiar odors in the TS dopamine responses recorded from trained mice . ( A ) The average TS dopamine responses to novel ( red ) or familiar ( blue ) odors . The black trace is the difference between the familiar and novel odors – indicating the time course of discrimination . ( B ) A histogram of the latencies of discrimination . The median latency is 170 ms . The bold line represents the cumulative probability that discrimination had occurred by that time . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 018 For initial training , mice were first habituated in a recording set-up with head-fixed preparation for 2–3 days ( see Materials and methods ) . After this initial habituation , animals were presented with four randomly interleaved trial types: ( 1 ) unpredicted water , ( 2 ) odor predicting water followed by water delivery , ( 3 ) odor predicting water followed by no outcome ( omission ) , or ( 4 ) odor predicting no outcome . We first compared VS dopamine and TS dopamine responses to an odor predicting no outcome over the course of the session ( Figure 2 ) . At the time of novel odor presentation , TS dopamine was excited very strongly by new odors ( Figure 2D ) . By contrast , VS dopamine did not respond to novel odors – not even on the very first trial ( Figure 2A ) . We examined 13 animals with fibers implanted into VS and 12 animals with fibers implanted into TS . TS dopamine showed significant excitation above baseline following the presentation of novel odors ( p = 1 . 76x10−6 , t-test , n = 12 animals , Figure 2G ) , whereas responses to novel odors in VS dopamine were not significantly different from baseline ( p = 0 . 354 , t-test , n = 13 animals , Figure 2G ) . The responses to novel odors in TS dopamine decreased significantly over the course of the first 30 odor presentations ( p = 5 . 10x10−11 , repeated measures ANOVA , n = 12 animals , Figure 2E–F ) . The responses after five odor representations ( 6–10 trials ) were significantly smaller than the responses in the first five trials ( p = 1 . 45x10−4 , paired t-test , n = 12 animals ) . To determine whether this decrease in signal was caused by a decrease in effective odor concentration within sessions , we compared responses to fast-decaying and slow-decaying odors ( Figure 2—figure supplement 1 ) . TS dopamine responses to novel odors , both fast-decaying and slow-decaying odors , decreased over the course of a session ( Figure 2—figure supplement 1F ) , whereas TS dopamine responses to familiar odors did not change ( Figure 2—figure supplement 2C ) . In VS , no responses were seen over the course of the session ( Figure 2B–C ) . Both VS dopamine and TS dopamine showed excitation to water ( responses to the first presentation of unpredicted water are shown in Figure 2G ) . To determine whether the signals we observed could have been caused by a movement-related artifact , we used video analysis to quantify the total body movements of mice performing the task ( Figure 2—figure supplement 3 ) . We found that mice did not show major body movements in response to odors predicting no outcome ( Figure 2—figure supplement 3B ) or novel odors ( Figure 2—figure supplement 3C ) , although they performed a stereotypical approach behavior in response to odors predicting reward ( Figure 2—figure supplement 3A ) . To determine whether changes in signal intensity observed within a session were likely to have been related to bleaching , we compared responses to free water and familiar odors over the course of sessions ( Figure 2—figure supplement 2 ) . We found that both VS dopamine responses ( Figure 2—figure supplement 2A ) and TS dopamine responses ( Figure 2—figure supplement 2B ) to free water remained constant within sessions . Similarly , TS dopamine responses to familiar odors predicting no reward remained constant as well ( Figure 2—figure supplement 2C ) . On the first day of odor-outcome association learning , TS dopamine was strongly excited by new odors ( Figure 3 , Figure 3—figure supplement 1 ) . The responses to novel odors that predicted water were significantly larger than responses to unpredicted water itself ( p = 1 . 36x10−4 , paired t-test , n = 19 animals , Figure 3B ) . Dopamine responses to odors gradually decreased over 21 days ( p = 1 . 095x10−8 , n = 19 animals , repeated measures ANOVA ) . The responses to water-predicting odors in TS dopamine were significantly smaller on the seventh day of training than on the first day ( p = 1 . 93x10−5 , n = 19 animals , day 1 vs day 7 , Figure 3B ) . On the other hand , responses to predicted water did not change significantly ( p = 0 . 641 paired t-test , n = 19 animals , day 1 vs day 7 , Figure 3B ) . By contrast , VS dopamine did not respond to novel odors predicting water ( Figure 3 , Figure 3—figure supplement 1 ) . Instead , we observed an initially large excitation in response to water itself ( Figure 3A ) . Over the course of the first 7 days , responses to predicted water ( US responses ) significantly decreased ( p = 1 . 96x10−4 , paired t-test , n = 10 animals , Figure 3A ) . Notably , this was independent of any CS response developing . VS dopamine did not display significant responses to odor cues that predicted reward on the first day ( p = 0 . 099 , t-test , n = 10 animals , CS responses compared to baseline on day 1 ) or on day 7 ( p = 0 . 054 , t-test , n = 10 animals , CS responses compared to baseline on day 7 ) ( Figure 3D ) . In fact , responses to reward-predicting cues appeared only after 2 weeks of training ( Figure 3A ) . Of note , responses to unpredicted water remained constant over the course of learning ( Figure 3A–B ) . Anticipatory licking gradually increased in frequency over the course of learning ( p = 0 . 0052 , repeated measures ANOVA , n = 10 animals , Figure 3C ) . We tested whether repeated training affected the observed pattern for novel cues and reward signaling in VS and TS dopamine . We trained nine mice with VS fiber implants and 11 mice with TS fiber implants by introducing a new odor paired with a reward or no outcome every day for a week , and then measured dopamine activity while learning new odor-water or odor-nothing associations ( Figure 4 ) . We found that repeated training with odor-reward associations did not change responses to new odors in VS dopamine or TS dopamine . VS dopamine did not respond to new odors ( p = 0 . 8749 , t-test , n = 9 animals , trial one or p = 0 . 322 , t-test , n = 9 animals , trial 1–5 vs baseline , Figure 4B–E ) and TS dopamine strongly responded to new odors ( p = 0 . 0059 , t-test , n = 11 animals , trial one or p = 0 . 0027 , t-test , n = 11 animals , trial 1–5 vs baseline , Figure 4H–K ) . Indeed , TS dopamine showed excitation to 91% of new odor presentations ( response in trial one vs baseline ) . Dopamine axon signals in mice repeatedly trained on learning odor-outcome contingencies displayed the same trends in the dynamics , but at a much faster rate: within a single session ( Figure 4 ) rather than over the course of weeks ( Figure 3 ) . VS dopamine showed a decrease in US response followed by an increase in CS response , with no response to novel stimuli ( Figure 4A–F ) . TS dopamine decreased responses to either novel odor ( nothing-predicting or water-predicting ) ( Figure 4G–L ) . To better understand the dopamine response to novel odors , we also paired new odors with an aversive air puff in these well-trained mice ( Figure 4—figure supplement 1 ) . As in the cases of novel odors predicting water or nothing , VS dopamine showed no odor responses ( Figure 4—figure supplement 1A–C ) and TS dopamine responded strongly to the novel odor ( Figure 4—figure supplement 1D–F ) . Notably , the decrease in TS dopamine response to novel odors predicting air puff was much smaller than the decrease of TS dopamine response to novel odors predicting no outcome ( Figure 4—figure supplement 2 ) , indicating that the dynamics of the response depend on what the novel odor cue predicts . Anticipatory licking in response to the rewarded odor increased after a few trials ( p = 0 . 0387 , paired t-test , n = 20 animals , trial 1–5 water CS lick vs baseline ) ( Figure 5 , Figure 5—figure supplement 1 ) . The animals showed differences in anticipatory licking frequency depending on cues within 10 trials of training ( p = 0 . 000259 , paired t-test , n = 20 animals , trial 6–10 water CS lick vs nothing CS lick ) , indicating learning of the outcomes of the odor cue ( Figure 5C ) . VS dopamine did not show differences in responses to cues before 15 trials ( p = 0 . 7736 , paired t-test , n = 9 animals , water CS vs nothing CS trial 11–15 , Figure 5D ) , whereas responses to predicted water decreased quickly ( p = 0 . 020 , paired t-test , n = 9 animals , trial 1–5 vs 6–10 , Figure 5E ) . Plotting the CS and US responses as a function of anticipatory licks ( rather than time ) showed that mice behaviorally responded to reward-predicting odors faster than VS dopamine CS responses developed , while TS dopamine CS responses were present in all trials ( Figure 5—figure supplement 2 ) . TS dopamine decreased responses to cues depending on what the cue predicted within five trials ( p = 0 . 0213 , paired t-test , n = 11 animals , water CS vs nothing CS trial 1–5 , Figure 5D ) and the difference became smaller later in the session . 10 . 7554/eLife . 21886 . 019Figure 5 . Dynamics of anticipatory licking behaviors and VS and TS dopamine . Licking in response to new odors predicting reward ( A ) or no outcome ( B ) over the course of a session , in animals that have been trained with many new odor associations , as in Figure 4 ( see Materials and methods ) . Separate plots for VS-implanted mice and TS-implanted mice are shown in Figure 5—figure supplement 1 . ( C ) A quantification of the number of anticipatory licks elicited by each odor in VS-implanted animals ( left ) and TS-implanted animals ( right ) . The difference between licks following a rewarding odor and an unrewarding odor are shown as open circles . ( D ) A comparison of the CS responses to rewarding and unrewarding new odors in VS dopamine ( left ) and TS dopamine ( right ) . ( E ) A comparison of the US responses to either predicted water or predicted nothing in VS dopamine ( left ) and TS dopamine ( right ) . The relationship between GCaMP responses in VS and TS and anticipatory licking is shown in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 01910 . 7554/eLife . 21886 . 020Figure 5—figure supplement 1 . Comparison of licking in VS-implanted and TS-implanted animals . ( A ) Licking in response to new odors predicting reward ( left ) or no outcome ( right ) in animals with an optical fiber implanted in VS . ( B ) Licking in response to new odors predicting reward ( left ) or no outcome ( right ) in animals with an optical fiber implanted in TS . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 02010 . 7554/eLife . 21886 . 021Figure 5—figure supplement 2 . Relationship between CS/US responses and anticipatory licking . The relationship between VS dopamine signals ( A ) or TS dopamine signals ( B ) and anticipatory licking . The left panels are plots of CS ( cue ) responses and anticipatory licking . The panels on the right are plots of US ( water ) responses and anticipatory licking . Markers are colored based on the order of the trials , from blue ( first trial ) to red ( last trial ) . Trials are connected with thin black lines based on the order in the session . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 021 We examined the temporal dynamics of the responses to novel odors in TS dopamine . The median onset latency of responses to novel odors in TS dopamine was 140 ms ( Figure 2—figure supplement 4 ) and the median onset latency of discrimination between novel odors and familiar odors in TS dopamine was 170 ms ( Figure 4—figure supplement 3 ) . In summary , dopamine axon signals in VS and TS showed opposite initialization while learning stimulus-outcome relationships . Dopamine axon signals in TS showed strong excitation to novel cues that gradually decreased , whereas dopamine axon signals in VS did not respond to cues with an unknown outcome and instead gradually developed cue responses to odors reliably predicting reward . Further , TS dopamine quickly discriminated cues , resulting in differential decrease rates of responses to novel cues depending on the predicted outcome ( air puff , water , or nothing ) . In order to understand the relationship between novelty responses and value coding , we next examined responses to rewarding and non-rewarding stimuli in VS dopamine and TS dopamine . Mice were trained to associate odors with water or with no outcome . After several weeks of this training , in some sessions , trials with odors predicting either a mild tone ( 55 dB ) or an air puff were interleaved in addition to trials with water and trials with no outcome . We chose a very mild tone with a similar intensity to the background noise in the room to try to minimize the aversiveness of this stimulus . To estimate the aversiveness of auditory stimuli , we measured the behavioral responses to tones of different volumes in a different set of mice ( Figure 6—figure supplement 1 ) . We found that quiet tones did not cause freezing . When comparing the VS dopamine and TS dopamine responses to all stimuli , we observed that VS dopamine showed excitation only to reward and reward-predicting cues ( Figure 6A–C , Figure 6—figure supplement 2 ) , while TS dopamine was excited in response to a variety of stimuli including water , tone , air puff , odor cues predicting any of these outcomes , and also odor cues predicting no outcome ( Figure 6D–F , Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 21886 . 022Figure 6 . Responses to rewarding , aversive and neutral stimuli in VS and TS dopamine . Dopamine responses to water ( A ) , tone ( B ) , and air puff ( C ) in the ventral striatum and the posterior tail of the striatum ( D–F ) . Plots of average traces from each region contain dotted lines indicating odor ( CS ) and outcome ( US ) delivery times . ( A , D ) Responses to unpredicted reward ( cyan ) , predicted reward ( blue ) , omitted reward ( purple ) , and nothing odor ( grey ) are plotted in the left panels . For each trace , a quantification of the average peak response to the CS / US is shown on the right . ( B , E ) Responses to unpredicted tone ( dark green ) , predicted tone ( light green ) , omitted tone ( yellow ) , and nothing odor ( grey ) are plotted in the left panels . For each trace , a quantification of the average peak response to the CS / US is shown on the right . ( C , F ) Responses to unpredicted air puff ( red ) , predicted air puff ( orange ) , omitted air puff ( yellow ) , and nothing odor ( grey ) are plotted in the left panels . For each trace , a quantification of the average peak response to the CS / US is shown on the right . Data from individual sessions is shown in Figure 6—figure supplement 2 . Behavioral responses to the tone are shown in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 02210 . 7554/eLife . 21886 . 023Figure 6—figure supplement 1 . Behavioral quantification of tone responses . Total body movement in response to tones of different volumes . The tone used in Figure 6 was 55 dB , for comparison . ( A ) An example of single-trial responses to a quiet ( 60 dB ) tone . ( B ) An example of single-trial responses to a relatively loud ( 90 dB ) tone . ( C ) A plot of the average total body movement in response to tones of different volumes , with a lower number indicating a higher instance of freezing or remaining still . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 02310 . 7554/eLife . 21886 . 024Figure 6—figure supplement 2 . Individual session data . A heat map of the average responses from each session ( from all animals ) to odors predicting reward ( left ) or no outcome ( right ) . The top panels are responses in VS and the bottom panels are responses in TS . Each row is the average for a session . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 024 We next examined reward prediction error coding , which consists of three key characteristics: ( 1 ) larger responses to reward-predicting cues than unrewarded cues , ( 2 ) smaller responses to predicted rewards than unpredicted rewards , and ( 3 ) a decrease in activity following reward omission . With respect to reward prediction , both VS and TS dopamine had a larger excitatory response to reward-predicting cues than cues predicting nothing ( VS: p = 1 . 8x10−10 , TS: p = 6 . 5x10−6 , t-test , Figure 6A , D ) . Similarly , both VS and TS dopamine had a smaller response to a predicted reward than an unpredicted reward ( VS: p = 5 . 1x10−11 , TS: p = 1 . 0x10−4 , t-test , Figure 6A , D ) . However , there was a significant difference in the response to the omission of a predicted reward: whereas VS dopamine showed a dip below baseline following reward omission ( VS: p = 4 . 8x10−7 , t-test , Figure 6A , D ) , TS dopamine axon signal was still significantly higher than baseline following omission ( p = 2 . 2x10−8 , t-test , Figure 6A , D ) . In TS , reward prediction elicited sustained activity over the interval between odor presentation and reward onset , and reward delivery caused only a small increase above this level . In fact , although the average peak response was slightly higher in rewarded trials than unrewarded trials ( p = 0 . 0014 ) , the total response ( area under each curve ) after the outcome ( reward delivery or omission ) did not differ significantly ( p = 0 . 81 ) . Interestingly , we found that the signals observed in TS dopamine displayed components of prediction error in response to non-rewarding stimuli as well . For example , TS dopamine showed less excitation to predicted air puff or tone than unpredicted air puff or tone ( p = 0 . 00016 , p = 0 . 00074 , Figure 6E–F ) . Additionally , TS dopamine cue responses to air puff or tone predicting cues were larger than the responses to cues predicting no outcome ( air puff: p = 6 . 8x10−4 , tone: p = 3 . 4x10−4 , Figure 6E–F ) . Similar to water omission , the omission of an expected tone or expected air puff did not cause a dip or increase in signal ( Figure 6E–F ) . In summary , VS dopamine encodes RPE whereas TS dopamine responds to salient stimuli in general . TS dopamine encodes the prediction of salient stimuli , and decreases the responses to salient stimuli once they are predicted , which are characteristics of prediction error . Notably , however , we did not observe clear responses to the omission of expected salient stimuli . Finally , we examined the relationship between different responses in dopamine and location in the striatum more carefully . For this purpose , in addition to VS and TS , dopamine axon signals in more anterior parts of the dorsal striatum ( DMS and DLS ) ( Figure 1B ) were recorded . Signals in each animal were pooled across sessions and the average was compared in relation to the location of recording sites ( Figure 7 ) . We first examined whether responses to novel odors were localized within VS or TS . VS consists of multiple sub-nuclei ( Zahm and Brog , 1992 ) and it is suggested that there are functional differences between medial VS and lateral VS ( Ikemoto , 2007 ) . We did not observe systematic differences of novelty responses along dorsal-ventral or medial-lateral axis within VS ( Figure 7A , anterior ) , although our spatial resolution could not completely distinguish each sub-nucleus . We did not observe noticeable differences between novelty responses in different sub-regions of TS either ( Figure 7A , posterior ) . 10 . 7554/eLife . 21886 . 025Figure 7 . Maps of dopamine responses in VS and TS . The distribution of responses to novelty ( A ) , reward ( B ) , familiar odor predicting nothing ( C ) , air puff ( D ) , and reward omission ( E ) . In the left panels , a 3D view of the average response from each animal . Novelty responses are the first responses to a novel odor either in the naïve case ( i . e . Figure 2 ) or the trained case ( i . e . Figure 4 ) . Reward responses are the average response to unpredicted reward . Nothing responses are the response to cues predicting no outcome . Air puff responses are the average response to unpredicted air puff . Omission responses are the average response to the omission of expected water . In the middle panels , coronal max projections are shown from the 3D view . On the right , the correlation between signals from the fibers and their positions on the A-P axis is shown , along with a yellow line indicating the best fit . The plots of these responses are shown for VS , DMS , DLS , and TS in Figure 7—figure supplement 2 . Examples of the whole-brain images used to find recording sites are shown in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 02510 . 7554/eLife . 21886 . 026Figure 7—figure supplement 1 . Examples of light-sheet images of cleared brains used to determine fiber locations . Example autofluorescence images used to find the position of ( A ) 400 µm diameter or ( B ) 200 µm diameter optic fiber implants after clearing with CLARITY and imaging with a light sheet microscope . Panels on the left are horizontal optical slices , and the panels on the left are enlarged views of slices near the tip of each implanted fiber . Yellow arrows denote the position of the implant . Both example brains have implants in both VS and TS . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 02610 . 7554/eLife . 21886 . 027Figure 7—figure supplement 2 . Responses to rewarding and aversive stimuli in VS , DMS , DLS , and TS . A comparison of responses to water , water predicting cues , and water omission ( left panels ) with responses to air puff , airpuff predicting cues , and airpuff omission ( right panels ) , in ( A ) VS dopamine , ( B ) DMS dopamine , ( C ) DLS dopamine , and ( D ) TS dopamine . Averages among all sessions and all animals are plotted on the left in each panel , and a quantification of the peak responses to each stimulus / outcome are plotted on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 21886 . 027 We examined how novelty responses were localized in the striatum . The observed distribution supported the idea that dopamine novelty responses are localized in TS . Novelty responses were correlated with location along the anterior-posterior axis ( r = −0 . 92 , p = 1 . 68x10−9 , n = 40 animals , Pearson’s correlation ) . We next examined responses to cues predicting rewarding , neutral or aversive outcomes , and responses to the omission of reward . Responses for all these factors were correlated with the location along anterior-posterior axis to various degrees ( water: r = 0 . 61 , p = 5 . 76x10−5 , nothing: r = −0 . 16 , p = 2 . 61x10−7 , air puff: r = −0 . 76 , p = 4 . 58x10−8 , and omission: r = −0 . 38 , p = 1 . 68x10−9 , n = 40 animals ) . Finally , novelty responses were positively correlated with responses to nothing , air puff-predicting cues , and reward omission ( nothing: p = 8 . 76x10−7 , air puff: p = 1 . 01x10−14 , omission: p = 9 . 02x10−10 , n = 40 animals ) . By contrast , water responses were observed in both VS and TS , although the amplitudes of water responses were slightly anti-correlated with novelty responses ( p = 0 . 0138 , n = 40 animals ) . We observed that water responses were found in all parts of the striatum ( Figure 7—figure supplement 2 ) . Together , the differences in novelty responses and value coding between VS and TS dopamine suggested that novelty responses and value coding in dopamine axons are at least partially segregated in the striatum; lack of inhibitory responses to reward omission and excitatory responses to neutral or aversive stimulus are localized in TS and coincide with excitatory responses to novel stimuli . The novelty responses we observed in TS dopamine , and especially the lack of novelty responses in VS dopamine , stand in stark contrast to popular theories explaining how novelty responses could fit into the dopamine RPE framework ( Hazy et al . , 2010; Kakade and Dayan , 2002 ) . These theories proposed that dopamine neurons show ‘optimistic initialization’ to novel cues , promoting physical and/or computational exploration in search of potential rewards ( often referred to as a ‘novelty bonus’ ) . However , our results suggest that the ‘pure value’-coding VS dopamine axon signal does not include a novelty response , while the salience-coding TS dopamine axon signal does include a novelty response . Over the course of classical conditioning , dopamine axon signals in VS and TS showed different coding principles . For a cue with an unknown outcome , VS dopamine initialized with no value prediction and gradually accumulated evidence for value , whereas TS dopamine began with a large excitation to novel cues and gradually decreased its responses . These dynamics can be conceptualized such that CS-value in VS dopamine was initialized with no value , while CS-salience in TS dopamine was initialized with very high salience . The opposite dynamics in VS and TS dopamine during learning are reminiscent of two different views of associative learning: US associability and CS associability . Many models , including delta-rule , Rescorla-Wagner and temporal difference ( TD ) models ( Rescorla and Wagner , 1972; Sutton and Barto , 1998; Widrow and Hoff , 1960 ) , emphasize US associability; when an animal learns to associate a CS to a US , the nature of the US determines how well the animal can learn the association . In these models , US associability is determined by the prediction error of the US ( i . e . the discrepancy between prediction and reality ) . Once the prediction error decreases , the US loses its effectiveness for creating an association with a CS . By contrast , other models ( Mackintosh , 1975; Pearce and Hall , 1980 ) emphasize CS associability; when an animal learns to associate a CS to a US , the nature of the CS determines how well the animal can learn the association . In these models , CS associability alters attention to a stimulus , or the storage of a stimulus in working memory , to promote association with a US . Dopamine dynamics in VS are well suited to serve as US associability , the teaching signal of the Rescorla-Wagner model ( Rescorla and Wagner , 1972 ) . TS dopamine does not follow the dynamics that the Rescola-Wagner model predicts . During the learning of novel odor-outcome associations , rather than US responses , CS responses were predominant and then decreased with learning . The dynamics of TS dopamine suggest that they may serve as CS associability , providing the attention signals for learning . Novel stimuli may excite TS dopamine because of the unpredictability of the novel stimulus itself ( i . e . it occurs with no cue or context predicting it ) or because the outcome of these novel stimuli is unknown ( i . e . they could predict a positive outcome , a neutral outcome , or a negative outcome ) . These two phenomena have been implemented in the framework of CS associability . For example , Wagner ( Wagner , 1978 ) proposed that CS associability is correlated with the weakness of CS-context association ( or the ‘unpredictability’ of the CS in the context ) . On the other hand , Pearce and Hall ( Pearce and Hall , 1980 ) defined the CS associability as the unpredictability of the US ( by the CS ) in the previous trial . In either case , the novelty would promote learning because of high CS associability , although the latter model did not define CS associability on the very first trial . Our data could be explained with either framework , but one observed phenomenon prefers the latter model . The responses to a novel odor decreased more slowly when the odor was associated with some outcome ( water or air puff ) than when it was associated with no outcome . If novelty responses are determined solely by odor-context association , responses to both odors should decrease at the same speed . Furthermore , this would lead to an equal response to all familiar odors , which we did not observe . Importantly , previous learning models incorporated teaching signals in the Rescorla-Wagner model and attention signals by novelty responses in one dopamine RPE system , with novelty responses as an exception or a bonus to the system ( Hazy et al . , 2010; Kakade and Dayan , 2002 ) . Here , we propose that CS associability , or the ‘attention’ term in dopamine signals may not be an exception in RPE , but instead may be a separate system – localized in particular brain regions such as in TS . Thus , dopamine in TS signals novelty and salience , and dopamine in VS signals RPE , although both systems may co-exist in some brain areas . In general , one big limitation of the current reinforcement learning algorithms is the so-called ‘curse of dimensionality’ . As the number of stimuli in an environment becomes large as in most natural environments , it quickly becomes difficult to properly assign credit to relevant objects . Attention would be critical to reduce the amount of information for learning to a more realistic amount . Dopamine in TS may be specialized for this function . In line with this idea , a recent study suggested that putative dopamine neurons in lateral SNc in monkey represented ‘cognitive salience’ , which was correlated with working memory load ( Matsumoto and Takada , 2013 ) . Behaviorally , previous studies suggested that SNc is important for the acquisition of enhanced CS associability ( Lee et al . , 2008 , 2006 ) . Mechanistically , dopamine in the prefrontal cortex has been modeled to serve a ‘gating’ function to provide flexible updating in working memory ( Braver and Cohen , 1999 ) , and a similar mechanism may apply to dopamine in TS . Of note , different from adjacent striatal areas , TS , categorized in caudal extreme ( Hintiryan et al . , 2016 ) , does not receive projections from sensorimotor cortex , suggesting a functional distinction from other areas in the striatum . All in all , TS dopamine may function to ‘pre-process’ sensory inputs to weigh potentially important stimuli to make reinforcement learning more efficient in a complex environment . Thus , during the learning of an association between a stimulus and a reward , dopamine signals in VS and TS may cooperate . Salience prediction error in TS dopamine may serve as the CS associability of the stimulus , whereas value prediction error in VS dopamine may serve as the US associability of the reward . Alternatively , similar to value prediction error in VS dopamine during stimulus-reward association , salience prediction error in TS dopamine may reinforce stimulus-stimulus associations . In contrast to TS dopamine , VS dopamine activity appeared to faithfully signal RPE . Several theories have arisen to explain how the reward prediction error could be computed in the brain . Of these , popular models such as Houk’s model ( Houk and Adams , 1995 ) and temporal difference ( TD ) models ( Sutton , 1988; Sutton and Barto , 1998 ) assume that a single system controls both CS-related and US-related dopamine firing . Under these models , higher expectation causes CS responses to become larger and US responses to become smaller at the same time . However , we observed that changes in CS and US responses were not simultaneous . In VS , the decrease in response to predicted rewards ( US ) was faster , occurring over the course of a single session , whereas the increase in response to reward-predicting odors ( CS ) required weeks of training . Over-training accelerated the time course of these events , but not their temporal order . The development of CS responses in VS dopamine was also much slower than the associative learning observed at the behavioral level . By contrast , US responses decreased as anticipatory licking increased . These results demonstrate that US responses in VS dopamine are well suited as prediction error signals in the Rescorla-Wagner learning model , whereas CS responses are not time-locked to this learning . The time course of CS responses we observed in VS dopamine is not easily explained by simple TD models . In these models , during learning , RPE signals gradually transfer from the timing of the reward to the timing of the preceding stimulus ( Schultz et al . , 1997 ) . Although the step-wise transfer may explain a delay between the decrease of US responses and emergence of CS responses , such a gradual transfer has not been observed in single neuron recording of dopamine neurons ( Pan et al . , 2005 ) . With any learning rate longer than one trial , transferred signals may become temporally smeared until they become time-locked to the CS ( Pan et al . , 2005; Schultz et al . , 1997 ) , which may not be detected in recording of single units . This theory would predict that we would observe some increase in signal between the stimulus and outcome ( either smeared or sharp ) during learning because the monitoring of population activity likely provides more reliable detection of small signals . However , this type of increase was not apparent with our bulk recording method . Instead , we observed a gradual development of a CS response directly following cue presentation . It is possible that distinct mechanisms could cause VS dopamine excitation in response to a reward-predicting CS and VS dopamine suppression in response to a predicted US , as proposed in several previous models ( Brown et al . , 1999; O'Reilly et al . , 2007 ) . There are different types of novelty ( Schomaker and Meeter , 2015 ) . One example is spatial novelty , which could be signaled by different arrangements of objects/stimuli in the environment . This kind of environmental novelty is known to induce exploration of animals and accelerate learning in this environment ( Li et al . , 2003; Otmakhova et al . , 2013 ) . Another example is stimulus novelty , which is associated with objects/stimuli that animals have never encountered or do not remember . It has been reported that dopamine activity in VTA and dopamine in the ventral striatum increased in the former case , in novel environments ( Segovia et al . , 2010; Takeuchi et al . , 2016 ) . On the other hand , in the present study , we focused on the latter type of novelty . Dopamine axon responses to novel stimuli were localized in TS . One potential explanation is that depending on the training history and environments , a given type of novelty may cause the animals to expect potential rewards , resulting in the excitation of the value system . Interestingly , previous studies found that different brain areas are responsible for different kinds of novelty ( Schomaker and Meeter , 2015 ) . Further studies are needed to determine how dopamine in different striatal areas responds to different kinds of novelty in different training environments . In our previous study , dopamine neurons that project to TS were mainly observed in the lateral SNc of mice ( Menegas et al . , 2015 ) . A previous study proposed that putative dopamine neurons in the lateral SNc of monkeys encode ‘motivational salience’ , which is the absolute value of positive or negative ‘value’ ( Matsumoto and Hikosaka , 2009 ) . On the other hand , another study proposed that excitation of dopamine neurons in response to non-rewarding stimuli encodes the stimulus intensity , regardless of value ( Fiorillo et al . , 2013 ) . In the present study , the excitation of TS dopamine elicited by various neutral stimuli suggested that the responses in TS could be related to more general salience rather than motivational salience , although we cannot rule out the possibility that the tone and odor predicting nothing had positive or negative motivational values . On the other hand , the fact that signals encoded by TS dopamine are modulated by prediction suggests that they are not encoding pure physical salience ( i . e . stimulus intensity ) . Instead , TS dopamine appears to encode general stimulus salience prediction error , which includes prediction-dependent suppression and prediction . The novelty responses we observed may be the extreme case of salience prediction error , causing large excitation because of minimum prediction , rather than an exception . A recent study found that putative dopamine neurons which project to the tail of the caudate ( part of posterior striatum ) in monkey formed another group of dopamine neurons . These dopamine neurons did not respond to water reward but encoded ‘sustained values’ of visual cues , whereas putative dopamine neurons projecting to the anterior caudate encoded ‘updating values’ , when cue-outcome contingency was frequently changed ( Kim et al . , 2015 ) . However , our results indicate that the difference between VS and TS dopamine extends beyond their flexibility . The dynamics between them are different in nature , not only in learning speed . Most importantly , we found that TS dopamine did not encode ‘values’ . Where do salience signals come from ? How are salience signals regulated by novelty and experiences ? A map of monosynaptic inputs to TS-projecting dopamine neurons should provide critical information ( Menegas et al . , 2015 ) . Previous studies showed that various brain areas including olfactory and visual systems are modulated by experience ( Boehnke et al . , 2011; Kato et al . , 2012 ) . Whether dopamine neurons receive this processed information from sensory systems or whether more abstract information about salience and novelty is passed to dopamine neurons and sensory systems in parallel is an open question . Of note , behavioral responses to novel odors are very quick , within one respiration cycle in rats ( Wesson et al . , 2008 ) . The responses to novel odors in TS dopamine that we observed began within 200 ms , most likely within one respiration cycle , suggesting a potential contribution at the early stages of novelty . Optical fiber fluorometry ( fiber photometry ) was developed by Kudo et al . ( Kudo et al . , 1992 ) and has been applied in many studies to record the population activity of neurons from cell bodies , dendrites , or axons ( Adelsberger et al . , 2005; Davis and Schmidt , 2000; Murayama et al . , 2007 ) . In this study , we recorded the population activity of dopamine axons in the striatum using GCaMP6m ( Akerboom et al . , 2012; Chen et al . , 2013; Kim et al . , 2016; Parker et al . , 2016 ) . We should point out several limitations associated with the present technique . First , previous studies ( Fiorillo et al . , 2013; Schultz , 2015 ) proposed that there is a temporal separation of two signals ( stimulus intensity and value ) in single dopamine neurons . However , we may only be able to measure the sum of these signals because of the limited temporal resolution of our method ( population calcium imaging using GCaMP6m ) . Second , a recent study found that dopamine axons with distinct signals ( locomotion and reward ) coexist in the dorsal-most part of the dorsal striatum ( Howe and Dombeck , 2016 ) . Axons signaling different information might also co-exist in other areas of the striatum , and this could not be resolved with our bulk-imaging method ( because such signals would effectively be ‘averaged’ ) . Third , because the spatial resolution of z-axis is large with fluorometry ( ~500 µm ) , we have to be careful in interpreting the analysis of differences along dorsal-ventral axis . Dopamine axons passing through and below the ventral striatum to the cortex ( Aransay et al . , 2014 ) may have contributed to the signals in VS dopamine , although calcium transients in passing axons are smaller than in axon terminals and boutons ( Koester and Sakmann , 2000; Llano et al . , 1997 ) . Finally , the activity of axons of dopamine neurons may not directly correspond to amounts of dopamine release at synapses or spike activities in cell bodies . Dopamine neurons that project to the ventral striatum ( mainly medial shell ) are able to co-release glutamate ( Stuber et al . , 2010 ) . Neuronal activities can be modulated locally at axon terminals in the striatum by cholinergic neurons ( Threlfell et al . , 2012 ) . Most importantly , observed calcium transients may not reflect spike counts , because of autofluorescence , bleaching , motion artifacts and inevitable normalization . Although we only applied baseline normalization ( i . e . signals were subtracted with and then divided by the average signal in a 1 s period before CS in each trial ) in this study , additional methods using activity-independent wavelength of excitation ( Kudo et al . , 1992; Lerner et al . , 2015 ) or examination of emission spectrum ( Cui et al . , 2013 ) may improve fidelity , especially in freely moving animals . We found that dopamine responses to novel stimuli are more localized than previously believed . We propose to revise current RPE models so that novelty-driven and salience-driven attention is attributed to TS dopamine , rather than added to the RPE framework as a bonus ( Kakade and Dayan , 2002 ) . Thus , TS dopamine may be specialized for functions apart from value , such as attentional orientation ( Redgrave et al . , 1999 ) , working attention , and/or as a filter for learning ( Braver and Cohen , 1999; Dayan and Sejnowski , 1996; Matsumoto and Takada , 2013; Pearce and Hall , 1980 ) . Further , our proposal includes another important point: RPE is not contaminated or distorted in VS dopamine . VS dopamine purely signals RPE , increasing the validity of the original ideas regarding dopamine’s role in reinforcement learning ( Schultz et al . , 1997 ) . 85 male adult mice were used . These mice were the result of a cross between DAT ( Slc6a3 ) -Cre mice ( recombinase under the control of the dopamine transporter , DAT ) ( B6 . SJL-Slc6a3tm1 . 1 ( cre ) Bkmn/J , Jackson Laboratory; RRID:IMSR_JAX:006660 ) ( Bäckman et al . , 2006 ) and tdTomato mice such that they were heterozygous for DAT-Cre and also heterozygous for tdTomato ( Gt ( ROSA ) 26Sortm9 ( CAGtdTomato ) Hze , Jackson Laboratory ) . Animals were housed on a 12 hr dark/12 hr light cycle ( dark from 07:00 to 19:00 ) , one to a cage , and performed the task at the same time each day . All procedures were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Harvard Animal Care and Use Committee . To prepare animals for recording , we performed a single surgery with three key components: ( 1 ) AAV-FLEX-GCaMP virus infection into the midbrain , ( 2 ) head-plate installation , and ( 3 ) one or more optic fiber implants into the striatum . At the time of surgery , all mice were 2–3 months old . All surgeries were performed under aseptic conditions with animals anesthetized with isoflurane ( 1–2% at 0 . 5–1 . 0 l/min ) . Analgesia ( ketoprofen , 5 mg/kg , I . P . ; buprenorphine , 0 . 1 mg/kg , I . P . ) was administered for 3 days following each surgery . To express GCaMP specifically in dopamine neurons , we unilaterally injected 250 nl of AAV5-CAG-FLEX-GCaMP6m ( 1 × 1012 particles/ml , Penn Vector Core ) into both the VTA and SNc ( 500 nl total ) . To target the VTA , we injected virus at Bregma −3 . 0 , Lateral 0 . 6 , at all depths between 4 . 5 and 4 . 0 mm . To target SNc , we injected virus at Bregma −3 . 3 , Lateral 1 . 6 , at all depths between 4 . 0 and 3 . 5 mm . Virus injection lasted several minutes , and then the injection pipette was slowly removed over the course of several minutes to prevent damage to the tissue . So that mice could be head-fixed during recording , we installed a head-plate onto each mouse . To do this , we removed the skin above the surface of the brain , dried the skull using air , and glued the head-plate onto the top of the skull with C and B Metabond adhesive cement . We used circular head-plates to ensure that the skull above the striatum would be accessible for fiber implants . Finally , during the same surgery , we also implanted optic fibers into the VS , DMS , DLS , and TS ( 1–4 fibers per mouse ) . To do this , we first slowly lowered optical fibers ( either 200 µm or 400 µm diameter , Doric Lenses ) into the striatum . The coordinates we used for targeting were as follows: ( VS ) Bregma 1 . 0 , Lateral 1 . 25 , Depth 4 . 1 , ( DS ) Bregma 0 . 0 , Lateral 1 . 5 , Depth 2 . 25 , ( DLS ) Bregma −0 . 5 , Lateral 2 . 75 , Depth 2 . 5 , ( TS ) Bregma −2 . 0 , Lateral 3 . 25 , Depth 2 . 5 . Once fibers were lowered , we first attached them to the skull with UV-curing epoxy ( Thorlabs , NOA81 ) , and then a layer of black Ortho-Jet dental adhesive ( Lang Dental ) . After waiting fifteen minutes for this glue to dry , we applied a very small amount of rapid-curing epoxy ( Devcon , A00254 ) to attach the fiber cannulas even more firmly to the underlying glue and head-plate . After waiting fifteen minutes for the epoxy to dry , the surgery was complete . Fiber fluorometry ( photometry ) ( Kudo et al . , 1992 ) allows for recording of the activity of genetically defined neural populations in behaving mice by expressing a genetically encoded Ca2+ indicator , GCaMP6m ( Akerboom et al . , 2012; Chen et al . , 2013 ) and chronically implanting an optic fiber . The optic fiber ( 200 µm or 400 µm diameter , Doric Lenses ) allows chronic , stable , minimally disruptive access to deep brain regions and interfaces with a flexible patch cord ( Doric Lenses ) on the skull surface to simultaneously deliver excitation light ( 473 nm and 561 nm , Laserglow Technologies ) and collect GCaMP and tdTomato fluorescence emission ( see Figure 1—figure supplement 1 ) . Activity-dependent fluorescence emitted by cells in the vicinity of the implanted fiber’s tip was spectrally separated from the excitation light using a dichroic , passed through a single band filter , and focused onto a photodetector connected to a current preamplifier ( SR570 , Stanford Research Systems ) . To record Ca2+ transients from dopamine terminals , we injected a Cre-dependent adeno-associated virus ( AAV ) carrying the GCaMP6m gene into the VTA and SNc of transgenic DAT-Cre mice and implanted 200 µm or 400 µm diameter optic fibers in the striatum . During recording , optic fibers were connected ( 1–2 per recording session ) to patch cables which delivered excitation light ( 473 nm and 561 nm ) and collected all emitted light . The emitted light was subsequently split and filtered ( see Figure 1—figure supplement 1 ) and collected by a photodetector connected to a current preamplifier . This preamplifier output a voltage signal which was collected by a NIDAQ board . The NIDAQ board was connected to the same computer that was used to control odor , water , tone , and air puff delivery with Labview , so GCaMP and tdTomato signals could be readily aligned to task events such as odor delivery or reward delivery . After surgery , mice were given three weeks to recover and become habituated to the installed head-plate and implanted optic fibers . Additionally , this allowed time for viral expression . After this recovery period , mice were handled for 2–3 days and water deprived . Weight was maintained above 90% of baseline body weight . In the first 2–3 sessions , mice were head-fixed and given unexpected water at random intervals ( randomly drawn , between 1 and 20 s , with a mean of 10 s and a normal distribution ) . This allowed mice to become habituated to being head-fixed and allowed us to determine the appropriate laser power ( typically between 0 . 1 mW and 0 . 25 mW ) to record >5% dF/F free water responses ( typically between 10% and 50% ) . These sessions were important , so that recordings during odor-water association could begin from the very first odor presentation on the first day of data collection ( see Experimental Timeline ) . The volume of water was constant for all reward trials ( predicted or unpredicted ) in all conditions . Similarly , the same mild tone ( 15 kHz , 0 . 5 s , ~50 dB ) was used in all tone trials and the same intensity air puff was used in all air puff trials . Each behavioral trial began with an odor cue ( a puff of odor lasting 1 s ) , followed by a 1 s delay , and then an outcome ( either water , nothing , tone , or air puff ) . Odors were delivered using a custom olfactometer ( Uchida and Mainen , 2003 ) . Each odor was dissolved in mineral oil at 1:10 dilution . 30 µl of diluted odor was placed inside a filter-paper housing ( Thomas Scientific , Swedesboro , NJ ) . Example PID measurements are shown in Figure 2—figure supplement 1 . Odors were selected pseudorandomly for each animal . Odorized air was further diluted with filtered air by 1:14 to produce a 1500 ml/min total flow rate . A variable inter-trial interval of 6–12 s ( random ) was placed between trials . All trial types were randomized in all of the sessions . Each day , the mice did about 300 trials over the course of about an hour . On a recording day , they performed the same task , and we recorded for ~45 min , which is approximately 250 trials , with constant excitation from the laser and continuous recording . Recordings from the same fiber were interspersed with at least two days of no recording . On the first day of classical conditioning , odors were presented to mice for the first time , and either predicted no outcome or reward . We quantified the ‘novelty response’ as the response to the first odor presentation that the mouse experienced , which was associated with no outcome ( for n = 13 VS-implanted mice and n = 12 TS-implanted mice ) . For comparison , the response to the first unpredicted water presentation in those sessions was quantified as well . These ‘novelty responses’ were the first trials of the first day of classical conditioning ( Figure 2 ) , while the average responses for these sessions are reported as ‘Day 1’ and compared with later sessions in Figure 3 . Due to technical difficulties we encountered in recording the first response of a session , some of the first responses were not recorded . Therefore , the sample size ( number of animals ) is lower for first trial responses ( n = 12 mice for TS ) in Figure 2 than for average responses during ‘Day 1’ ( n = 19 mice for TS ) in Figure 3 . During classical conditioning , odor cues ( also called ‘conditioned stimuli’ or ‘CS’ ) were associated with either reward or no outcome . In the case of reward trials , water ( the ‘unconditioned stimulus’ or ‘US’ ) would follow odor presentation after 2 s , 90% of the time ( i . e . 10% omission ) . In ~10% of trials , unpredicted water was delivered without odor presentation . During training , GCaMP responses were recorded at time points ( Figure 3 ) rather than daily , to minimize bleaching or tissue damage . After 3 weeks of this classical conditioning training ( with one water predicting odor , one nothing-predicting odor , and occasional unpredicted water ) , mice were introduced to new odor-outcome association types . At this point in training , mice were also presented with odor-tone associations ( 20% of trials ) or odor-air puff associations ( 20% of trials ) , in addition to the two familiar odors associated with water and with no outcome , allowing us to multiplex data from learning onto this data regarding value or salience coding and prediction error coding . Unpredicted tone or air puff was also delivered in ~5% of trials . Data from these sessions was used in Figure 6 and Figure 7 , including data from DMS-implanted or DLS-implanted mice . Finally , a subset of these mice ( n = 11 ) were trained with two new odors each day ( one associated with water and one associated with no outcome ) , every day for a week , until mice began to discriminate between odors behaviorally within a few trials ( see Figure 5 ) rather than over the course of many days ( see Figure 3 ) . We referred to these mice as ‘overtrained mice’ . After this training , we recorded GCaMP and licking signals from these overtrained mice as they learned either new odor-water ( one third of sessions ) or new odor-nothing associations ( one third of sessions ) in Figure 4 . In one third of sessions , a new odor associated with air-puff was introduced in addition to the two familiar odors associated with water or with no outcome ( Figure 4—figure supplement 3 ) . We randomized whether the new odor of a session would predict water , nothing , or air puff to ensure that mice could not generalize that novel odors reliably predicted a particular outcome . Mice performed one session per day . GCaMP and tdTomato signals were collected as voltage measurements from the current preamplifiers using Labview ( Figure 1—figure supplement 1 ) . The 'dF/F' measurement was calculated by comparing the average signal in a 1 s period before each trial ( ‘F1’ ) with the signal at any given point during the trial ( ‘F2’ ) . The calculation for each point in the trial ( calculated in 1 ms bins ) was then simply dF/F = ( F2 – F1 ) / F1 . We used this measurement because it readily normalized signals ( i . e . in the case of low signal to noise ratio , the denominator would be larger ) . The average responses to a stimulus type within a session ( often ~50 trials per stimulus type ) were averaged , and these session averages were used as the data in each figure ( individual session averages can be found in Figure 3—figure supplement 1 and Figure 6—figure supplement 1 , and example individual single trial traces are shown in Figure 1 and Figure 1—figure supplement 1 ) . These session averages were compared across animals in two basic ways . ( 1 ) Traces were averaged and plotted ( as the average of all session averages ) along with the standard error ( the total number of sessions being the sample size ) as in Figure 5 , left panels . ( 2 ) Peak responses to cues/outcomes were quantified by finding the point with the maximum absolute value during 2 s following cue/outcome for each trial , then comparing the averages between sessions as in Figure 5 , right panels . Because traces were aligned using task events ( i . e . cue on time ) rather than behavioral events ( i . e . first inhalation ) , comparing peak responses ensured that signals , which were slightly offset in time relative to odor presentation , could be compared . To compute the latency of responses to novel cues , each trial was tested for difference from baseline in the first five novel odor trials using time bins of 50 ms . We called the ‘latency’ of the response ( in each trial ) the center of the first time bin where five consecutive time bins all showed significant difference from baseline . To compute the latency of novelty discrimination , the sessions were tested for significant difference between familiar and novel odors in the first five trials using time bin of 50 ms . We called the ‘latency’ of discrimination the center of the first time bin where five consecutive time bins all showed significance . While recording GCaMP signals , we also recorded licking . To measure licking , we used a detector that output a voltage based on the disruption of its infra-red light path . We set a threshold for signal corresponding to a ‘lick’ and then made the signal binary by finding each time point where the signal crossed the threshold so that it could easily be quantified . Our main quantification for licking was counting the number of ‘anticipatory licks’ , the licks following an odor ( CS ) and preceding the arrival of the outcome ( US ) . GCaMP responses and licking responses were collected through Labview during the training for offline analysis . Statistical analyses ( i . e . t-tests , ANOVA ) were run using Matlab ( Mathworks ) . All analyses considered a value of p≤0 . 05 significant , and exact p-values are reported in the text . To quantify body movement , we used a video camera to capture images of the mice while they performed the task and made a rough estimate of total body movement by subtracting each frame of the video from the last frame , using the ‘imabsdiff’ function in Matlab . We reported these values , which we took to be a proxy for body movement , as ‘arbitrary units’ or ‘a . u . ’ because they were measured in pixels . Brains were cleared as previously described ( Menegas et al . , 2015 ) at 37°C for 2 days , with a constant current of 1 . 2 amps . A Niagra 120 V ( Grey Beard Pumps #316 , Mt Holly Springs , PA , United States ) pump was used to circulate clearing solution . A Precision Adjustable 60 V/5A power supply ( Korad Technology #KA6005D , Shenzhen , China ) was used to provide current . A 5-gallon plastic container ( US Plastic #97 , 028 , Lima , Ohio , United States ) was used as a clearing solution reservoir and tubing was run though a second 5-gallon plastic container filled with water to cool the solution flowing through it . Chambers were constructed as previously described ( Chung and Deisseroth , 2013 ) using a Nalgene chamber ( Nalgene 2118–0002 , Rochester , NY , United States ) and platinum wire ( Sigma-Aldrich 267228 , St . Louis , MO , United States ) . Clearing was done in a room held at 37°C . Images were acquired with the Zeiss Z . 1 Light-sheet microscope ( Carl Zeiss , Jena , Germany ) . Brains were glued to the tip of a 1 ml syringe ( without needle ) such that the posterior tip of the cerebellum was in contact with the syringe . A 488 nm laser was used to excite GFP and a 647 nm laser was used to produce autofluorescence . Images were collected through a 5× objective with PCO-Edge scMOS 16 bit cameras ( PCO , Kelheim , Germany ) with 1920 × 1920 pixels . Each frame was 2000 × 2000 μm , so each pixel was roughly 1 . 04 μm . The step size between images was set to 5 . 25 μm , so the voxels were not isotropic . Brains were imaged horizontally from the dorsal side , and then rotated 180° for horizontal imaging from the ventral side . Each view was tiled with 7 × 6 tiles ( 14 , 000 × 12 , 000 μm ) and the two views were combined to create a continuous image . Autofluorescence images were subsequently downsized to 1400 × 1200 × 700 pixels for alignment to the reference space . In these downsized images , voxels have 10 μm spacing in all three dimensions . Brains were aligned to a previously described reference space comprised of the average of 25 brains ( Menegas et al . , 2015 ) . Alignment to this reference space was performed using Elastix ( Klein et al . , 2010 ) . We performed affine alignment followed by B-spline alignment based on mutual information , as previously proposed for human magnetic resonance imaging ( MRI ) image registration ( Metz et al . , 2011 ) . After alignment , fiber positions were manually determined by tracing the fiber paths to their termination points . After clearing and imaging each brain as a whole volume and aligning these images to determine the exact location of each fiber , we classified each fiber as either ( 1 ) an implant into VS , ( 2 ) an implant into DS , ( 3 ) an implant into TS , or ( 4 ) an incorrectly targeted fiber . Our analysis in the current paper focuses on comparing VS-implanted fibers to TS-implanted fibers . Most implants were successfully targeted to VS or TS . We classified implants as successful based on the following criteria . For VS: any fiber within the nucleus accumbens core or shell , between Bregma 2 . 0 and Bregma 0 . For DS: any fiber within the striatum anterior to Bregma 1 . 5 . For TS: any fiber within the striatum posterior to Bregma −1 . 5 . We discarded data from eight animals which had fibers incorrectly targeted . The fibers in these animals were often in areas of cortex directly adjacent to the intended recording site . We observed very little or no signal ( compared to our other recordings ) in these cases , likely due to the relatively sparse dopaminergic innervation of cortex relative to striatum in mouse .	New experiences trigger a variety of responses in animals . We are surprised by , move towards , and often explore new objects . But how does the brain control these responses ? Dopamine is a molecule that controls many processes in the brain and plays critical roles in various mental disorders , diseases that affect movement , and addiction . Rewarding experiences ( like a glass of cold water on a hot day ) can trigger dopamine neurons and studies have also shown that dopamine neurons respond to new experiences . This suggested that novelty may be rewarding in itself , or that novelty may signal the potential for future reward . On the other hand , it may be that different groups of dopamine neurons play different roles in responding to new or rewarding experiences . In 2015 , it was reported that dopamine neurons connected to the rear part of an area in the brain called the striatum receive signals from different parts of the brain than most other dopamine neurons . The dopamine neurons connected to this “tail” of the striatum preferentially received inputs from regions involved in arousal rather than reward , suggesting that they may have a unique role and transmit a different type of information . Now , Menegas et al . have shown that dopamine signals in different areas of the striatum separate reward from novelty and other signals in mice . The results demonstrate that new odors activate dopamine neurons projecting to the tail of the striatum , but that this activity fades as the novelty wears off ( as the mice learn to associate the odor with a particular outcome ) . By contrast , dopamine neurons projecting to the front of the striatum do not respond to novelty , but rather become more active as mice learn which odors accompany rewards ( only responding to odors that predict reward ) . The experiments also show that dopamine neurons in the tail of the striatum encode information about the importance of a stimulus . Together , these findings indicate that some of the roles dopamine plays in the brain may not be related to reward , but are instead linked to the novelty and importance of the stimulus . The next challenge will be to find out how the separate reward and novelty signals in dopamine neurons relate to the animals’ behavior . This may help us to better understand dopamine-related psychiatric conditions , such as depression and addiction .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Opposite initialization to novel cues in dopamine signaling in ventral and posterior striatum in mice
Neural patterning involves regionalised cell specification . Recent studies indicate that cell dynamics play instrumental roles in neural pattern refinement and progression , but the impact of cell behaviour and morphogenesis on neural specification is not understood . Here we combine 4D analysis of cell behaviours with dynamic quantification of proneural expression to uncover the construction of the zebrafish otic neurogenic domain . We identify pioneer cells expressing neurog1 outside the otic epithelium that migrate and ingress into the epithelialising placode to become the first otic neuronal progenitors . Subsequently , neighbouring cells express neurog1 inside the placode , and apical symmetric divisions amplify the specified pool . Interestingly , pioneer cells delaminate shortly after ingression . Ablation experiments reveal that pioneer cells promote neurog1 expression in other otic cells . Finally , ingression relies on the epithelialisation timing controlled by FGF activity . We propose a novel view for otic neurogenesis integrating cell dynamics whereby ingression of pioneer cells instructs neuronal specification . Neural specification relies on proneural genes , which are expressed in specific patterns and underlie the genesis , organisation and the function of the neurons that will subsequently differentiate ( Bertrand et al . , 2002; Huang et al . , 2014 ) . Many signals that pattern the nervous system have been identified . For example , gradients of Shh , BMP and Wnt establish thirteen different domains of neural progenitors in the mouse neural tube ( Ulloa and Briscoe , 2007 ) ; FGF8 and FGF3 control the site of retinogenesis initiation in chick and fish through regulation of ath5 expression ( Martinez-Morales et al . , 2005 ) ; and EGFR signalling determines the expression of a wave of l ( 1 ) sc in the Drosophila optic lobe ( Yasugi et al . , 2010 ) . Concomitant with cell specification , neural tissues undergo phases of morphogenesis and/or growth . Thus , the cells within a given domain are not static but perform complex cell behaviours . Recently , the contribution of such cell dynamics to neural patterning has been identified . In the neural tube , for instance , sharply bordered specification domains involve the sorting of cells along a rough Shh-dependent pattern ( Xiong et al . , 2013 ) . Additionally , differences in the rate of differentiation of cells ( which migrate out of the tissue ) between distinct domains of the neural tube help to establish the overall pattern during tissue growth ( Kicheva et al . , 2014 ) . Thus , dynamic spatial rearrangements of cells within a field that is being specified are integrated with patterning mechanisms of positional information by morphogens . In the inner ear , developmental defects in neurogenesis could result in congenital sensorineural hearing loss ( Manchaiah et al . , 2011 ) . Neurogenesis begins when an anterior neurogenic domain appears at the placode stage by the expression of the proneural gene neurog1 , which specifies neuronal precursors . The rest of the otic placode is non-neurogenic and generates non-neuronal cell types ( Ma et al . , 1998; Andermann et al . , 2002; Abello and Alsina , 2007; Radosevic et al . , 2011 ) . In the neurogenic domain , neurog1 induces neurod1 ( Ma et al . , 1996 , 1998 ) expression , which is required for delamination of neuroblasts from the epithelium ( Liu et al . , 2000 ) . Delaminated neuroblasts subsequently coalesce to form the statoacoustic ganglion ( SAG ) and differentiate into mature bipolar neurons ( Hemond and Morest , 1991; Haddon and Lewis , 1996 ) . The spatial restriction of the otic neurogenic domain relies on the integration of diffusible signals such as FGFs , SHH , Retinoic acid and Wnt ( reviewed in Raft and Groves , 20142015 ) as well as the function of transcription factors such as Tbx1 ( Radosevic et al . , 2011; Raft et al . , 2004 ) , Sox3 ( Abelló et al . , 2010 ) , Otx1 ( Maier and Whitfield , 2014 ) , Eya1 ( Friedman et al . , 2005 ) and Six1 ( Zou et al . , 2004 ) . In the inner ear , several FGFs ( Adamska et al . , 2001; Mansour et al . , 1993; Léger et al . , 2002; Alsina et al . , 2004; Vemaraju et al . , 2012; Alvarez et al . , 2003 ) , regulate the sequential steps of neurogenesis starting from the expression of neurog1 ( Vemaraju et al . , 2012; Léger et al . , 2002; Alsina et al . , 2004 ) and continuing to later events involving neuroblast expansion ( Vemaraju et al . , 2012 ) . Together with the regulation of spatial regionalisation , the number of neuronal progenitors produced depends on local cell–cell interactions mediated by the Notch pathway ( Adam et al . , 1998 ) . Remarkably , to date no studies have addressed how morphogenesis , cell behaviour and proneural dynamics impact otic neuronal specification . Here we use the zebrafish inner ear as a model to analyse the role of cell dynamics on neuronal specification . We identify pioneer cells that are specified outside the otic epithelium , ingress into the placode during epithelialisation and control local neuronal specification , suggesting an instructive role of these cells . Furthermore , we show that FGF signalling affects otic neurogenesis through the regulation of otic placode morphogenesis , influencing pioneer cell ingression . We have previously identified cell behaviours contributing to otic vesicle morphogenesis ( Hoijman et al . , 2015 ) and here we focused on the influence of cell dynamics in the establishment of the neurogenic domain . For this , we used a zebrafish BAC reporter line that expresses the fluorescent protein DsRed-Express ( DsRedE , a faster maturation version of DsRed [Bevis and Glick , 2002] ) under control of the neurog1 regulatory elements ( Drerup and Nechiporuk , 2013 ) . We imaged in 4D the otic development from stages of otic placode morphogenesis ( 15 hpf ) until neuroblast delamination is abundant and the central lumen is expanding ( 20 . 5 hpf , Figure 1A and B; Videos 1 and 2 ) . The overall pattern of DsRedE expression is highly consistent between embryos , being restricted to the most ventroanterolateral region of the placode until 19 hpf and expanding posteromedially at around 20 . 5 hpf ( Figure 1A and B; Videos 1 and 2 ) . This DsRedE expression pattern recapitulates the endogenous spatiotemporal pattern of neurog1 as analysed by in situ hybridisation ( ISH ) ( Radosevic et al . , 2014; Vemaraju et al . , 2012; Andermann et al . , 2002 ) . Moreover , DsRedE expressing cells delaminate ( Figure 3H; Videos 1 and 11 ) and are incorporated into the SAG ( Figure 1A and B; Video 3 ) , supporting the use of this line to analyse single cell dynamics of neuronal specification . 10 . 7554/eLife . 25543 . 003Figure 1 . Specification dynamics and morphogenesis of the otic neurogenic domain . ( A , B ) Selected frames of a video of an otic placode from a TgBAC ( neurog1:DsRedE ) n16 embryo shown in ( A ) 3D reconstructions ( dorsal view ) and ( B ) coronal ventral planes . Green in the right schemes shows the region imaged . Membranes are stained with memb-GFP . D:dorsal , V:ventral , A:anterior , P:posterior , M:medial and L:lateral . The asterisk indicates the region where the SAG is forming . Medial to the otic vesicle , DsRedE is also expressed in the neural tube . ( C ) Averagez-projection ( dorsal view ) of the inner ear at 17 hpf . Dashed line indicates the protuberance . ( D ) Scheme of the rectangular cuboid used for quantifications . Neurogenic region is shown in red . ( E , F , G ) Quantification of the number of cells ( E ) , the cellular density ( F ) and mitotic events ( G ) in the indicated regions at 19 hpf ( n = 11 ) ( E , F ) or between 14 and 18 . 5 hpf ( n = 2 ) ( G ) . Data are mean ± s . e . m . *p<0 . 0001 one sample t-test in ( E ) and unpaired t-test ( F ) . Scale bars , 20 µm . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00310 . 7554/eLife . 25543 . 004Video 1 . 4D imaging of otic neuronal specification . 3D reconstructed time-lapse of the otic vesicle from a TgBAC ( neurog1:DsRedE ) n16 embryo . Red: DsRedE fluorescence . Green: memb-GFP . Dorsal view . Time from the first frame is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00410 . 7554/eLife . 25543 . 005Video 2 . Specification dynamics visualized in individual cells . Selected coronal ventral planes from the z-stacks used for 3D reconstructions in Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00510 . 7554/eLife . 25543 . 006Video 3 . neurog1 expressing cells locate in the SAG after delamination . 3D reconstruction of the otic vesicle at 21 hpf . White arrow indicates the position of the SAG . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 006 We also analysed the cellular organisation of the neurogenic domain by performing a 3D morphometric analysis of this region . During the stages of neuronal specification , the shape of the otic vesicle is asymmetric , exhibiting a protuberance in the anterolateral region ( Figure 1C ) . To compare the properties of the neurogenic region with the rest of the otic vesicle , we built a rectangular cuboid with the vertices of the vesicle and divided it in eight regions of equal volume ( Figure 1D ) , in which we quantified the number of cells and the volume of tissue . By 19 hpf , the neurogenic domain region accumulated more cells ( 15 . 4 ± 0 . 4% of the total number of cells in the vesicle , 49 ± 3 cells of 311 ± 16 cells respectively ) than other regions ( mean non-neurogenic region: 12 . 0 ± 0 . 1% , 36 ± 2 cells , Figure 1E ) and presented higher cellular density ( Figure 1F; neurogenic region: 2 . 16 ± 0 . 03 nuclei/1 × 103 μm3 , mean non-neurogenic region: 1 . 60 ± 0 . 03 nuclei/1 × 103 μm3 ) . Quantification of all the mitotic events inside the vesicle between 14 and 18 . 5 hpf revealed that cell proliferation is also highly enriched in this region ( Figure 1G ) . While the increase in cell number in the neurogenic domain was moderate ( about 3% more cells than other regions ) , the enrichment in mitotic events led to about 41% of the total number of divisions to occur in this domain . Thus , in addition to a phase of transit-amplification of neuroblasts after delamination ( Vemaraju et al . , 2012 ) , neuronal progenitors also appear to multiply inside the otic vesicle . This analysis indicates that the neurogenic domain presents high cell number , high cell density and an increased proliferative activity . To analyse how the neurogenic domain is built , we decided to evaluate when and where cells of the neurogenic domain start to express neurog1 . We first aimed to capture the earliest specified cells . Epithelialisation of the otic placode progresses from 12 . 5 hpf until about 18 hpf ( Hoijman et al . , 2015 ) . While it has been reported that neurog1 expression in the otic placode begins at 15 hpf ( Radosevic et al . , 2014 ) , we found that already at 13 hpf there are rows of DsRedE expressing cells lateral to the neural tube and anterior to the epithelializing otic placode ( Figure 2A; Video 4 ) . These cells coincide with neurog1 expressing cells detected by ISH ( Figure 2—figure supplement 1A ) , and previously assumed to belong to the anterior lateral line placode ( Andermann et al . , 2002 ) . Unexpectedly , when we followed these cells we found that some of them migrate posteriorly and become incorporated into the anterolateral region of the otic epithelium , in a position corresponding to the neurogenic domain ( red brackets in Figure 2B; Video 5 ) . Therefore , these cells develop into otic and not lateral line cells . To confirm this cell ingression , we injected NLS-Eos mRNA at 1 cell stage to obtain a homogeneous nuclear staining with the photoconvertable protein throughout the embryo . At 13 hpf , we photoconverted Eos protein ( from green to red fluorescence ) in a group of nuclei anterior to the otic epithelium where the migrating cells are located . At 20 hpf , we detected photoconverted nuclei inside the vesicle ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 25543 . 007Figure 2 . Ingression of neurog1+ cells . ( A ) The otic epithelium and its anterior region at 13 hpf . Arrowheads highlight neurog1+ cells outside the otic epithelium . ( B ) Selected frames of a 3D reconstruction ( dorsal view ) of the otic placode following the movement of the anterior neurog1+ cells . Arrowheads at 14 . 5 hpf indicate neurog1+ cells before epithelialisation ( white: cells outside the placode , orange: ingressing cells ) . At 15 . 5 hpf red bracket identifies cells that will ingress ( shown at 17 hpf ) and blue bracket cells that will not ingress . In ( A ) and ( B ) the contrast of the red signal was increased to improve visualisation . ( C ) Selected planes of a 3D tracking of a single cell specifying during ingression ( white dot ) . At 108 min the cell is already epithelialised . Asterisk indicates the SAG . ( D–F ) 3D tracking of single cells during ingression . ( D ) 3D reconstruction ( dorsal view ) showing the initial position of the tracked cells ( white , pink and blue dots ) at 14 hpf . The violet dot indicates the posterior vertex of the placode . ( E ) 2D visualisation of the 3D tracks shown in ( D ) are displayed in a temporal color code . Each track was displaced in the y axis for better visualisation . The track of the posterior vertex of the placode is shown on the right ( see also Figure 2D ) . ( F ) Selected frames for the cell of the white track . At 150 min the cell is ingressing and completed at 240 min . At 300 min cytokinesis occurs . Membranes are stained with memb-mCherry . Embryos are Tg ( actb:H2B-venusFP ) . ( G ) Selected planes showing cell-membrane displacements during migration of the cell tracked in ( F ) . White arrowheads indicate protrusion of the cell front and orange arrowheads the position of the nucleus . ( H ) Schematic representation of the migration and ingression during epithelialisation ( see Figure 2—figure supplement 1 for further details ) . Blue line: laminin , green line: actin layer , red cells: neurog1+ cells , red arrows: migration of neurog1+ cells towards the otic placode . Scale bars , 20 µm . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00710 . 7554/eLife . 25543 . 008Figure 2—figure supplement 1 . Morphogenetic features related to ingression . ( A ) Whole mount ISH for neurog1 from 13 and 14 hpf Tg ( elA:GFP ) embryos . This transgenic line expresses GFP in rhombomeres 3 and 5 ( asterisks , at early stages rhombomere three express higher levels than rhombomere 5 ) , facilitating the spatial localization of the otic placode . An immunostaining for GFP was performed after the in situ hybridisation . Dotted lines highlight the limits of the hindbrain ( red ) and the forming otic placode ( white ) . White arrowheads indicate neurog1 expression . ( B ) Photoconversion of NLS-Eos stained nuclei at 13 hpf in a region anterior to the epithelium in TgBAC ( neurog1:DsRedE ) n16 embryos expressing memb-GFP . At 20 hpf photoconverted nuclei were observed in neurog1+ cells inside the vesicle ( arrowhead ) . ( C ) GFP reporting neurod1 expression in the non-ingressing pool of cells at 18 hpf ( in the SAG region ) from Tg ( neurod:GFP ) embryos ( blue bracket ) . Embryos are also TgBAC ( neurog1:DsRedE ) n16 and express memb-GFP . ( D ) Early stages of otic epithelialisation . Dashed line indicates the epithelialised part of the otic vesicle . Membranes are stained with memb-GFP . ( E ) Laminin staining at 14 and 22 hpf in transversal and coronal sections . Nuclei are counterstained with DAPI . White arrowheads indicate the forming otic placode . ( F ) 3D reconstruction ( dorsal view ) of an otic vesicle and its anterior region at 14 hpf from a Tg ( actb1:Lifect-GFP ) embryo . The white arrowheads indicate the actin layer that divides latero-medially the tissues lateral to the hindbrain in two regions ( white and yellow asterisks , see also reslice 1 ) . Reslices , built from the white bars 1 and 2 shown in the 3D reconstruction , show transversal sections anterior ( reslice 1 ) or at the position ( reslice 2 ) of the otic placode ( dashed line ) . h: hindbrain ( dotted line ) . ( G ) Cell ingression evaluated using NLS-Eos photoconversion at 13 hpf in neurog1hi1059 mutant embryos injected at 1 cell stage with memb-GFP and NLS-Eos mRNAs . White arrowheads indicate ingressed cells at 18 hpf . See also Figure 2H for a scheme of the morphological features described in this figure . Scale bars , 20 µm . Dotted lines in ( A–C ) outline the limits of the otic epithelium/vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00810 . 7554/eLife . 25543 . 009Video 4 . Early neurog1 expressing cells located anterior to the otic vesicle . 3D reconstruction of an otic vesicle and the anterior region at 13 hpf , showing the presence of DsRedE expressing cells ( white arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 00910 . 7554/eLife . 25543 . 010Video 5 . neurog1 expressing cells ingress in the otic epithelium . 3D reconstructed time-lapse showing the ingression of neurog1 expressing cells . Orange arrowheads indicate ingressing cells and white arrowheads cells that are outside the organ . Cells that will ingress are highlighted with a red bracket and the direction of movement by a red arrow . The group of neurog1 expressing cells that do not ingress is indicated by a blue bracket and arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 010 We also detected in the same anterior region a second pool of neurog1+ cells ( expressing also neurod1; Figure 2—figure supplement 1C ) that moves posteromedially without ingressing , remaining in the region of the SAG ( blue brackets in Figure 2B; Video 5 ) . The migrating cells are located laterally relative to a population of sparse cells from which they are segregated by an F-actin rich layer that runs anteroposteriorly until it reaches the placode ( Figure 2—figure supplement 1F and Figure 2H ) . These observations suggest that neurog1 expression is not sufficient for cell ingression . Additionally , neurog1 expression was not required for cell ingression , as some neurog1- cells ingress . Consistently , we detected cell ingression events in neurog1 mutant embryos ( neurog1hi1059 , Figure 2—figure supplement 1G ) . Interestingly , 3D tracking of individual cells of the ingressing pool revealed that some cells activate neurog1 expression while moving towards the epithelium and before their epithelialisation ( Figure 2C; Video 6 ) . Immediately after ingressing into the neurogenic domain , these cells divide and delaminate , thus undergoing a complete cycle of epithelialisation and de-epithelialisation in only a few hours . Analysis of the movement of these cells suggests that their migration is a directional process occurring in individual cells ( Figure 2D , E and F; Video 7; some cells of the same region migrate in other directions ) . We also observed that the leading front of cells periodically protrudes , followed by a rapid forward translocation of the nucleus ( Figure 2G; insets of Video 7 ) , as has been described during fibroblast migration ( Petrie and Yamada , 2015 ) . When tracking three neighbouring cells , we observed that while two of them ingress ( white and pink tracks ) , the third one ( blue track ) , which is initially positioned closer to the otic placode , divides during migration and the daughters do not ingress ( Figure 2D , E and F; Video 7 ) . These observations highlight that ingressing cells are interspersed with other cells that do not join the otic placode , and factors other than anteroposterior positional cues within the migrating population determine whether a cell will ingress or not into the otic placode . 10 . 7554/eLife . 25543 . 011Video 6 . 3D tracking of an individual cell during ingression , division and delamination . Coronal ventral planes from z-stacks selected to track an ingressing cell ( white dot ) . Note that it begins to express neurog1 before epithelialization . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 01110 . 7554/eLife . 25543 . 012Video 7 . 3D tracking of multiple cells during ingression . Initially , the position of three cells anterior to the otic epithelium is shown ( white , pink and blue dots ) . Tracking ( upper panels ) and 2D trajectory of each cell ( lower panel , yellow track shows the position of the posterior vertex of the placode ) are depicted . Insets highlight the mode of migration , with leading edge of the cell protruding ( white arrowheads ) before the forward displacement of the nucleus ( yellow arrowheads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 012 Particular morphogenetic features could facilitate the ingression of cells from the anterior region . As we previously reported , the otic placode is only epithelialised medially at these stages ( Hoijman et al . , 2015 ) . As epithelialisation progresses , at 14 hpf the posterior part of the placode is segregated from the surrounding cells , while the anterior region of the placode is not ( Figure 2—figure supplement 1D; Video 8 ) . Thus , the posterior part folds approximately 3 hr before the anterior one ( Figure 2—figure supplement 1D; Video 8 ) . During this period , and by the anterior unfolded region , migrating cells ingress into the otic epithelium . Moreover , the basal lamina at these early stages is only rudimentary and not continuous ( contrary to the one present at later stages surrounding the whole organ; Figure 2—figure supplement 1E ) . Therefore , the fact that the epithelium is still organizing could allow the migrating cells to ingress into the tissue before it is fully formed . 10 . 7554/eLife . 25543 . 013Video 8 . Detailed view of the morphogenesis of the otic placode . Time-lapse of memb-GFP expressing embryos showing the different stages of tissue epithelialisation . Note that the posterior region folds before the anterior one ( orange arrowhead highlights the unfolded anterior region ) . Lines indicate the epithelialised regions . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 013 In summary , our results show cells that are being specified outside the otic epithelium , migrate and ingress into the prospective neurogenic domain , constituting the earliest neuronal specified cells of the organ . We next evaluated if , in addition to ingressing cells , other cells start to express neurog1 within the neurogenic domain . We visualised the activation of neurog1 expression inside the otic vesicle in real-time ( Figure 3A; Video 9 ) , a process that we refer to as ‘local specification’ . Dynamic quantification of DsRedE fluorescence levels in individual cells ( Fcell ) indicated that the rate of increase in the signal is variable among cells ( Figure 3B , mean rate of increase ranging between 0 . 15 and 0 . 54 a . u . /min , n = 11 cells ) . However , we found that when the signal reaches a critical level ( between 45 . 5 and 52 . 5 a . u . in Figure 3B , gray region with red dots ) , cells begin to delaminate ( visualised by the movement of the cell body to the basal domain of the epithelium ) . This suggests that cells delaminate relative to neurog1 levels and not to the time elapsed since they initiated neurog1 expression ( Figure 3B and C ) . 10 . 7554/eLife . 25543 . 014Figure 3 . Local specification and divisions of neurog1 expressing cells . ( A ) Selected planes showing DsRedE expression dynamics in locally specified cells ( white and blue dots ) from TgBAC ( neurog1:DsRedE ) n16 embryos expressing memb-GFP . Asterisk indicates the SAG . The embryo is 16 . 5 hpf at the beginning of the time-lapse . ( B ) Quantification of DsRedE fluorescence over time for 11 cells locally inducing neurog1 . Red dots indicate beginning of delamination . The gray region highlights the interval of fluorescence levels at which all cells delaminate . ( C ) Box plot made from the quantifications shown in ( B ) , illustrating that at the moment of delamination , the time elapsed from the initiation of neurog1 expression is highly variable , while the expression levels are not . The value for each cell was normalized by the mean of the cell group . ( D , E ) neurog1+ mitotic cells ( white dots ) contacting ( D ) or not ( E ) the central lumen ( dashed line ) . 19 ( D ) and 17 ( E ) hpf embryos are shown . ( F ) Pard3-GFP localisation in the central lumen and the anterolateral region ( white arrowhead ) . Membranes are stained with memb-mCherry . ( G , H ) Divisions ( white dots ) located in the lumen ( G ) or the apical scaffold ( H , z-projection ) . 20 ( G ) and 18 hpf ( H ) embryos are shown . ( I ) Selected planes from a 3D time-lapse of a neurog1+ mitosis . White and blue dots track the daughter cells . Dashed lines indicate the approximated limit of the vesicle . Selected planes for each daughter cell are shown from 60 min onwards . At 129 min cells are delaminated . Asterisk indicates the SAG . The embryo is 18 hpf at the beginning of the time-lapse . ( J ) Reslice of a frame at 98 min from the video shown in ( H ) showing the z proximity between the tracked daughter cells during delamination ( the red signal was removed for better visualisation ) . Scale bars , 20 µm . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 01410 . 7554/eLife . 25543 . 015Figure 3—figure supplement 1 . Cell division can precede neurog1 expression . 3D tracking of a neurog1− cell ( white dot ) that divides and subsequently their daughters express DsRedE and delaminate . Dotted lines outline the limits of the otic vesicle . Asterisk indicates the SAG . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 01510 . 7554/eLife . 25543 . 016Video 9 . Real-time activation of neurog1 expression in local specified cells . Coronal ventral planes from z-stacks selected to follow the beginning of DsRedE expression in two individual cells that are being specified locally ( white and blue dots ) . Insets show higher magnification images . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 016 As we mentioned above , higher mitotic events occur in the neurogenic domain . Therefore , division could also contribute to the domain by adding neurog1 expressing cells ( neurog1+ cells ) to the domain . To address this , we performed a 4D analysis of cell divisions and found that every cell divides only once in the 7 hr period analysed ( n = 27/27 ) . Mitotic cells are found either contacting the central lumen ( Figure 3D ) or not ( peripheral divisions ) ( Figure 3E ) . Interestingly , these latter cells are apposed to an accumulation of the apical determinant Pard3 that forms a scaffold perpendicular to the central luminal surface of the vesicle , running from the lumen to the periphery ( Figure 3F; Video 10 ) . Thus , similar to the apical mitosis occurring in the central lumen , peripheral divisions are also in contact with an apical surface ( Figure 3G and H ) . 10 . 7554/eLife . 25543 . 017Video 10 . Apical scaffold formation dynamics . 3D reconstructed time-lapse of Pard3-GFP ( gray ) localization during otic morphogenesis ( dorsal view ) . Pard3-GFP in the otic vesicle ( green arrows ) or in the superficial external superficial ( orange arrows ) is shown . The anterolateral apical scaffold forms early during placode development and is transitory . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 017 In neurogenic tissues , either asymmetric ( daughter cells become one progenitor and one neuron ) or symmetric ( both daughter cells with the same fate ) divisions can occur ( Taverna et al . , 2014; Chenn and McConnell , 1995; Das and Storey , 2012 ) . This depends on factors such as the apicobasal position of the dividing cell and the orientation of the mitotic spindle ( Das and Storey , 2012 ) . Our dynamic analysis of neurog1 activation allowed us to assess the modes of divisions within the otic neurogenic domain . We observed that all divisions in the neurogenic domain have the cleavage plane perpendicular to the apical surface regardless of their position in the epithelium or their neurog1 expression ( Figure 3G and H ) . When analysing the fate of the daughter cells after division , we found all were symmetric ( 27/27 ) : both daughter cells delaminate after division ( 20/27 delaminate during the timeframe analysed , 7/27 are positioned to delaminate at the end of the acquisition ) . However , division can occur either before ( 13/25 ) or after ( 12/25 ) the induction of neurog1 expression . Interestingly , daughter cells from mitoses of a neurog1+ cell with high levels of DsRedE expression ( neurog1+Hi cell ) rapidly delaminate , remaining in close contact as they move to the periphery of the tissue ( Figure 3I and J; Video 11 ) . On the other hand , daughter cells from mitosis of cells not expressing neurog1 ( neurog1− ) , or only at low levels ( neurog1+Low ) , remain in the epithelium after division , where they increase the DsRedE signal over a variable period of time ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 25543 . 018Video 11 . Coordinated and quick delamination after division of neurog1 expressing cells . Coordinated delamination: in the upper panel , coronal planes tracking an individual cell before division ( white dot ) and their daughters after division and until delamination ( white and blue dots ) are shown . In the lower panel , 2D movement of the tracked cells is shown . Note the coordinated behaviour of daughter cells moving in close contact to the periphery of the tissue and delaminating simultaneously . Quick delamination after division: tracking of other cell including sagittal planes in the lower panel . Only one daughter is tracked ( white dot ) . White lines indicate the limits of the vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 018 In summary , divisions in the neurogenic domain are symmetric and apical . Furthermore , there is not a preferential sequence of events concerning neurog1 activation and division . Taken together , our analysis of the origin of neurog1+ cells revealed that they are added to the neurogenic domain by three different mechanisms: cell ingression , local expression and cell division . The incorporation of the ingressing cells and their rapid exit from the otic vesicle led us to wonder about their role in the establishment of the neurogenic domain . These early-specified cells might contribute to the neurogenic domain by their inclusion as specified cells and/or play additional roles . To address this question , we decided to eliminate these cells during their migration , before they reach the otic epithelium . For this , we identified the stream of migrating cells by their DsRedE signal ( Figure 4A ) , laser-ablated them unilaterally at 12 . 5 hpf ( Figure 4B ) , and examined the effects on neuronal specification in 3D in the otic vesicle at 18 . 5 hpf ( Figure 4C–H; Video 12 ) , before delamination becomes significant . Neurog1 expression was analysed by quantification of the Fcell in all cells belonging to the neurogenic domain ( Figure 4C and D ) . Ablation of a limited number of cells ( 2–3 cells per laser pulse; see Material and methods for more details ) led to a decrease in the global level of DsRedE expression ( calculated as the sum of the Fcell for all neurog1+ cells ) in the vesicle of the ablated side as compared to the contralateral vesicle on the non-ablated side of the embryo ( Figure 4C and E; non-ablated side: 1492 ± 58 , ablated side: 454 ± 44 a . u ) . Applying an increased number of laser pulses ablated more cells , which seems to lead to a more severe specification phenotype ( compare embryos 1 and 2 from Figure 4C , which received 1 and 3 laser pulses respectively ) , despite the overall morphology of the neurogenic domain being unaffected . Analysis of both neurog1 expression in the otic epithelium at 21 hpf and the phenotype of the SAG at 42 hpf confirms that the effect of ablation persists and , thus , does not appear to represent a delay in neuronal specification ( Figure 4—figure supplement 2A , B and C; Video 12 ) . The effect of ablation is specific to otic neurog1 expression , since DsRedE expression in the neural tube was not affected ( Figure 4—figure supplement 2D ) . Moreover , we observed a phenotype only after ablating anterior future ingressing cells: ablation of neurog1+ cells in another location ( posterior to the placode at 13 hpf , Figure 4—figure supplement 1B ) or developmental stage ( anterior to the vesicle at 19 hpf , Figure 4—figure supplement 1C ) did not affect neurog1 expression in the otic vesicle . 10 . 7554/eLife . 25543 . 019Figure 4 . Ingressing cells instruct local neuronal specification . ( A , B ) Laser ablation of neurog1+ cells before ingression . Two different embryos are shown . Images of the otic epithelium and its anterior region at 12 . 5 hpf just before ( A ) and after ( B ) laser-ablation . White arrowheads indicate neurog1+ cells . Blue arrowheads localise the ablated region . Embryo 1 only received one laser pulse and embryo 2 three laser pulses ( only two are visible in this plane ) . The contrast of the red signal was increased to improve visualisation . ( C–H ) neurog1 expression pattern inside the vesicle after ablation . ( C ) Average z-projections of embryos shown in ( A , B ) 5 hr after ablation ( 18 . 5 hpf ) . The ablated side and their contralateral non-ablated side of the same embryo are shown . ( D ) Quantification of Fcell in each neurog1+ cell of the vesicles shown in ( C ) . Each dot indicates one cell . Green lines indicate the mean of each condition . The number of neurog1+ cells in each vesicle is: embryo 1 , non-ablated side: 24 , ablated side: 8; embryo 2 , non-ablated side: 25 , ablated side: 2 . ( E–H ) Parameters of neuronal specification at the single cell level are shown: global level of DsRed expression ( E ) Nneurog1+ ( F ) , F¯ cell ( G ) , and Nneurog1+Hi ( H ) . Data are mean ± s . e . m . ( n = 6 ) . t-test p<0 . 0001 , p<0 . 0005 , p<0 . 05 . ( I ) Scheme with of different explanations of how early ablation of ingressing cells influences F¯ cell inside the vesicle at later stages . ( i ) In absence of cell ablation the neurogenic domain is composed by ingressing and local specified cells , with a characteristic value for F¯ cell . ( ii ) If the distribution of cells with high and low fluorescence levels is equal between the ingressing and the local specified cells , ablation of ingressing cells does not change the F¯ cell . Thus , this possibility does not explain the observed decrease in F¯ cell after ablation . ( iii ) If the neurog1+Hi cells are mainly ingressing cells , ablation of these cells reduces the F¯ cell . However , Figure 4J shows that neurog1+Hi cells are mainly resident cells of the epithelium . ( iv ) If an instruction from the ingressing cells to the local specified cells is present , ablation of the ingressing cells decreases the F¯ cell . The intensity of red depicts the DsRedE level of expression in each cell . ( J ) Dots show the location at 13 . 5 hpf of backtracked cells corresponding to neurog1+Hi cells at 19 hpf in a non-ablated embryo . Pink dot: neurog1+ ingressed cell . White dots: neurog1- cells . The 3D reconstruction of the placode shown is representative of two different analysed embryos . All embryos are TgBAC ( neurog1:DsRedE ) n16 and membranes are stained with memb-GFP . Scale bars , 20 µm . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 01910 . 7554/eLife . 25543 . 020Figure 4—figure supplement 1 . Calibration and specificity of ablation experiments . ( A ) Calibration of cell ablations . A laser pulse ( as described in Materials and methods ) was applied to embryos expressing H2B-mCherry in a mosaic manner lateral to the neural tube . In example 1 , two nuclei were stained in the imaged region before ablation ( white arrowheads ) . After the laser pulse , a red ablation bubble was observed as consequence of the death of the two stained cells ( blue arrowhead ) . In example 2 , the nuclei of neighbouring cells ( numbered from 1 to 5 ) are surrounding two target cells ( white arrowheads ) . Imaging after ablation indicated that the targeted cells died , but the neighbouring cells remained healthy and only slightly displaced in space . In example 3 , a similar behavior as in example two can be observed , but the intact neighbouring cells are in close contact with the dead cells , highlighting the fact that ablation is highly specific and restricted to the targeted cells ( white arrowheads , see also Video 14 ) . ( B ) and ( C ) Ablation at a posterior region or a late developmental stage . On the left , laser ablation of neurog1+ cells located posterior to the otic epithelium at 13 hpf ( B ) or anterior to the otic vesicle at 19 hpf ( C ) . White arrowheads indicate neurog1+ cells . Blue arrowheads localise the ablated region . The embryos received one laser pulse . On the right , z-projection images of neurog1 expression pattern inside vesicles at 20 ( B ) or ( 22 ) hpf from the ablated and contralateral non-ablated sides of the embryo are shown . Quantifications of the F¯cell and the Nneurog1+ are shown as the fold change of ablated/non-ablated sidesx100 ( n = 5 in ( B ) and n = 7 in ( C ) ) . Data are mean ± s . e . m . Scale bars , 20 µm . Dotted lines outline the limits of the otic epithelium/vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 02010 . 7554/eLife . 25543 . 021Figure 4—figure supplement 2 . Late neurogenic phenotypes after ablation and specification analysis of non-proliferative otic placodes . ( A ) Z-projection images of the embryos shown in Figure 4 , A and B , 8 hr after ablation ( 21 hpf ) . The ablated side and their contralateral non-ablated side of the same embryo are shown ( images are representative of the phenotypes observed in 4 embryos at this stage , see also Video 12 ) . Asterisk indicates the SAG . ( B ) Quantification of the mean DsRedE fluorescence in each neurog1+ cell of the vesicles shown in ( A ) . Each dot indicates one cell . Green lines indicate the mean of each condition . ( C ) Z-projection images of otic vesicles and the SAG at 42 hpf from an neurog1-DsRedE embryo ablated at 13 hpf ( the ablated side and their contralateral non-ablated side of the same embryo are shown ) . Note the reduction in size of the SAG in the ablated size of the embryos . Images are representative of 3 embryos analysed . ( D ) Quantification of the F¯cell in a region of the neural tube adjacent to the otic vesicle 5 hr after ablation ( 18 hpf ) . Data are mean ± s . e . m . ( 70 cells were counted in each region , n = 3 ) . ( E ) Z-projection images of neurog1 expression pattern inside the vesicle at 20 hpf in DMSO and AH treated embryos . On the right , quantifications of the F¯cell and the Nneurog1+ are shown as the fold change of the AH group respect to the DMSO group ( n = 14 for DMSO and n = 12 for AH ) . Data are mean ± s . e . m . t-test , *p<0 . 0001 . Scale bars , 20 µm . Asterisk indicates the SAG . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 02110 . 7554/eLife . 25543 . 022Video 12 . Ablation of pioneer cells before ingression affects neurog1 expression in the neurogenic domain at later stages . 3D reconstruction: DsRedE signal in the neurogenic domain ( red ) of otic vesicles at 21 hpf corresponding to the previously ablated and contralateral non-ablated sides of the same embryo . A single plane of the memb-GFP signal from each vesicle is shown for better 3D orientation ( green ) . The DsRedE fluorescence coming from cells outside the otic vesicle was removed with FIJI to improve the visualisation of the phenotype inside the vesicle . z-stack: sequence of coronal planes from dorsal to ventral of neurog1 expression in the otic vesicle at 21 hpf in ablated and contralateral non-ablated sides of the embryo . The DsRedE expression levels can be visualised in single cells ( quantifications of specification phenotypes were performed on this type of z-stacks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 022 When comparing the number of neurog1+ cells ( Nneurog1+ ) , we also found a reduction in the ablated side vesicle compared to the control vesicle ( Figure 4F; non-ablated side: 23 . 8 ± 1 . 4 cells , ablated side: 10 . 0 ± 0 . 8 cells ) . This result could be partially explained by the failure of the ablated cells to ingress into the forming neurogenic domain . These results also indicate that when ablating the cells that will be part of the neurogenic domain , the cells now located in the same position do not change their fate and become neural specified , as expected if cell identity would be dictated by cell position . Interestingly , the number of cells eliminated by ablation ( and the ones produced by their divisions ) would be too small to account for the large decrease in the number of neurog1+ cells in the vesicles of the ablated side ( Figure 4F ) . This suggests that ingressing cells play an instructive role on the specification of other cells of the neurogenic domain ( i . e . local specification ) . To shed light on this possibility , we calculated the mean value for Fcell ( F¯ cell ) in vesicles from each experimental condition . This parameter was also reduced by the ablation ( Figure 4G; non-ablated side: 60 . 1 ± 2 . 5 , ablated side: 43 . 6 ± 4 . 8 a . u . ) , suggesting that the global reduction in fluorescence was not only caused by a decrease in the number of neurog1+ cells ( Figure 4Iii ) , but that the neurog1 transcriptional activity inside these cells was also reduced . Accordingly , the number of neurog1+Hi cells ( Nneurog1+Hi ) was also significantly lowered by ablation ( Figure 4H; non-ablated side: 6 . 0 ± 0 . 6 , ablated side: 1 . 0 ± 0 . 4 cells ) . However , it is possible that the neurog1+Hi cells at the time point analysed are mainly ingressed cells , and thus by eliminating them , we decreased the F¯cell in each vesicle by a relative increase in neurog1+Low cells ( Figure 4Iiii , see figure legend for detailed explanation of the scheme ) . We discarded this possibility by backtracking cells identified as neurog1+Hi at 19 hpf from non-ablated embryos , and observing that most of them are neurog1- cells at 13 hpf positioned inside the epithelising placode before ingression takes place , therefore belonging to the pool of cells specified locally ( Figure 4I and J ) . Given that both the number and expression levels of neurog1+ cells were reduced by ablation , it is possible that a cell community effect takes place , in which the presence of more neurog1+ cells favours higher expression levels in the pool of progenitors being specified . However , the effect of cell ablation was not recapitulated when proliferation was blocked by incubation with aphidicolin and hydroxyurea ( AH ) ( Hoijman et al . , 2015 ) . This treatment decreased the number of neurog1+ cells at 20 hpf ( fold change AH/DMSO: 49 , 6 ± 6 . 3% , Figure 4—figure supplement 2E ) but the mean levels of neurog1 expression were not affected ( fold change AH/DMSO: 110 ± 11% , Figure 4—figure supplement 2E ) . This result suggests that cell number and expression levels are not necessarily linked during otic neurog1 expression and highlights the specific relevance of the ingressing cells in promoting the transcription of the neurog1 gene . Altogether , these results indicate that these cells act as pioneer neurogenic cells , contributing to the neurogenic domain both through their incorporation as neurog1+ cells and by promoting neurog1 expression non-autonomously in other cells of the domain . To understand how the specification processes identified above are promoted , we decided to explore the role of FGF signalling , a pathway reported to control both neurog1 expression in the vesicle and the number of neurons in the SAG ( Wang et al . , 2015; Vemaraju et al . , 2012 ) . To this aim , neurog1:DsRedE embryos were incubated with the FGFRs inhibitor SU5402 from 11 hpf until 19 hpf , beginning the treatment after placode induction and before otic morphogenesis starts ( Figure 5A and B ) . Analysis of neuronal specification indicated that SU5402 treatment reduced the global level of DsRedE expression ( Figure 5C ) , in agreement with the previous ISH analysis of neurog1 expression ( Vemaraju et al . , 2012; Léger et al . , 2002 ) . This reduction was caused not only by a decreased mean level of neurog1 expression in each cell ( Figure 5B and C ) , but also by a reduction in the number of neurog1+ cells ( Figure 5C , and particularly in the neurog1+Hi cells ) . To confirm that the FGF pathway is mediating the mentioned phenotype , we crossed a transgenic line expressing a dominant negative isoform of the FGF receptor 1 fused to GFP under the control of a heat-shock ( hs ) promoter ( hsp70:dnfgfr1-EGFP ) ( Norton et al . , 2005 ) with the TgBAC ( neurog1:DsRedE ) nl6 line . Inducing transgene expression at 10 hpf phenocopied at 20 hpf the effect on otic neurog1 expression observed in SU5402 treated embryos ( Figure 5D and E ) . 10 . 7554/eLife . 25543 . 023Figure 5 . FGF control of neuronal specification . ( A–C ) neurog1 expression pattern inside the vesicle in embryos incubated in DMSO or SU5402 . ( A ) Images of otic vesicles at 19 hpf incubated from 11 hpf in DMSO or SU5402 ( ventral planes ) . ( B ) Quantification of Fcell for cells of vesicles from the groups shown in ( A ) . Each dot indicates one cell . Green lines indicate the mean of each condition . n = 5 for DMSO and n = 6 for SU5402 . ( C ) Parameters of neuronal specification at the single cell level for the data shown in ( B ) : global level of DsRed expression , F¯cell , Nneurog1+ and Nneurog1+Hi are shown as fold change of SU5402/DMSOx100 . ( D , E ) neurog1 expression pattern inside the vesicle from neurog1:DsRedE;hsp70:dnfgfr1-EGFP/+ or neurog1:DsRedE embryos heat-shocked at 10 hpf . ( D ) Z-projections of otic vesicles at 20 hpf . ( E ) Parameters of neuronal specification are shown: global level of DsRed expression , F¯cell , Nneurog1+ and Nneurog1+Hi ( n = 8 ) . ( F ) Photoconversion at 13 hpf of NLS-Eos stained nuclei in a region anterior to the otic epithelium . Embryos expressed memb-GFP and were treated with DMSO or SU5402 from 11 hpf ( z-projections ) . At 18 hpf , photoconverted nuclei is observed inside the vesicle of the DMSO treated embryo . High magnification in the right ( dotted square , Scale bar 10 µm ) . Yellow dotted lines indicate the limits of the otic epithelium . ( G ) Quantification of the number of photoconverted nuclei inside the vesicle ( n = 6 for DMSO and n = 7 for SU5402 ) . ( H , I ) Photoconversion experiments as in ( F , G ) but on hsp70:dnfgfr1-EGFP/+ and sibling embryos heat-shocked at 10 hpf . ( H ) Z-projections of the photoconversion and cell ingression . ( I ) Quantification of the number of photoconverted nuclei inside the vesicle ( n = 7 for siblings and n = 6 for hsp70:dnfgfr1-EGFP/+ ) . ( J ) Selected images from a time-lapse of hsp70:dnfgfr1-EGFP/+ embryos heat-shocked at 10 hpf . Note that as early as 14 hpf the anterior part of the otic tissue is already folding , at 15 hpf the process is advanced ( red arrowhead ) , and at 15 . 5 hpf the anterior and posterior regions seem to be symmetrically folded ( see also Video 13 ) . ( K ) Laminin immunostainings at 16 hpf in hsp70:dnfgfr1-EGFP/+ and sibling embryos heat-shocked at 10 hpf . The nuclei were counterstained with DAPI . High magnification in the right ( dotted square , Scale bar 10 µm ) . The images are representative of 6 embryos analysed . Note the formation of a continuous layer of laminin in some regions ( white arrowheads ) . ( L ) Scheme of cell dynamics playing a role in neuronal patterning of the inner ear . FGF signalling delays anterior tissue folding allowing the ingression of pioneer neurog1+ cells in the prospective neurogenic domain of the otic epithelium . These pioneer cells promote neurog1 expression in other cells of the neurogenic domain . In addition , neurog1+ cells divide symmetrically and delaminate . Data are mean ± s . e . m . t-test p<0 . 001 , p<0 . 005 , p<0 . 01 , p<0 . 05 . Scale bars , 20 µm . White dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 02310 . 7554/eLife . 25543 . 024Figure 5—figure supplement 1 . Analysis of cell division controlled by SU5402 and neurog1 expression in FGF10a mutant embryos . ( A ) pH3 immunostainings at 16 hpf in otic vesicles from DMSO and SU5402 treated embryos . Two embryos in each experimental group are shown . The nuclei were counterstained with DAPI . ( B ) Quantification of the pH3 immunostainings ( n = 12 ) . ( C ) and ( D ) neurog1 expression pattern inside the vesicle in neurog1-DsRedE;fgf10a-/- or neurog1-DsRedE; sibling embryos . ( C ) Z-projection images of vesicles at 20 hpf . ( D ) Quantifications of the F¯ cell and the Nneurog1+ are shown ( n = 10 for siblings and n = 5 for fgf10-/- ) . ( E ) 3D Tracking of photoconverted NLS-Eos nuclei in hsp70:dnfgfr1-EGFP/+ induced embryos . Z-projections of resliced sagittal sections are shown . Arrowheads indicate examples of tracked nuclei ( each color correspond to a different cell ) . The cells indicated with white and pink arrowheads in the latter panels were not identified in the two first time points , due to their lateral movement out and in of the video during the posterior migration . ( F and G ) Tissue folding analysis ( G ) or cell ingression analysis by NLS-Eos photoconversion ( F ) during placode formation in FGF3 overexpressing embryos ( heat-shock at 11 hpf of the Tg ( hsp70:fgf3 ) line ) . In ( F ) the posterior region remains unfolded at late stages ( 20 hpf ) . Arrowhead: a deformation of the lumen is observed in the posterior region of the vesicle , which is found only anteriorly in wild type embryos associated with the unfolded tissue . Data are mean ± s . e . m . Scale bars , 20 µm . Dotted lines outline the limits of the otic vesicle . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 024 We realised that the phenotypes produced by blocking FGF signalling are similar to those resulting from cell ablation . Furthermore , given that FGF blockade strongly reduces the number of SAG neurons when it is performed early during otic development ( Wang et al . , 2015 ) , we hypothesise that FGF signalling might control the early cell ingression event . We tested this idea by blocking the FGF signalling from 11 hpf onwards ( both using SU5402 or the hsp70:dnfgfr1-EGFP transgene ) , photoconverting NLS-Eos in cells located anterior to the otic epithelium at 13 hpf ( Figure 5F and H , left panels ) and , subsequently , quantifying the number of photoconverted nuclei inside the otic vesicle at 18 hpf ( Figure 5F and H ( right panels ) , G and I ) . As shown in Figure 5G and I , SU5402 treatment or DNFGFR1-EGFP induction significantly reduce the number of ingressed cells ( DMSO: 4 . 7 ± 1 . 1 cells , SU5402: 1 . 0 ± 0 . 4 cells; heat-shocked siblings 5 , 3 ± 0 . 4 cells; heat-shocked hsp70:dnfgfr1-EGFP/+: 0 . 2 ± 0 . 2 cells ) . These results suggest that the FGF pathway contributes to neuronal specification in the otic vesicle by promoting the ingression of the pioneer cells into the neurogenic domain . To gain insights into how the FGF pathway influences cell ingression , we performed time-lapse imaging during otic placode morphogenesis in embryos expressing DNFGFR1-EGFP . Tracking of photoconverted cells in these embryos showed that they still move towards the otic epithelium but remain outside ( Figure 5—figure supplement 1E ) . Interestingly , in these embryos the anterior region of the epithelium folds at an earlier stage in development than in control embryos ( Figure 5J; Video 13 ) , becoming synchronous with folding of the posterior region ( and not asynchronously as in the wild type embryos , Figure 2—figure supplement 1D; Video 8 ) . Additionally , the otic basal lamina also formed earlier in DNFGFR1-EGFP expressing embryos than in siblings ( Figure 5K ) . Conversely , overexpression of FGF3 by heat-shocking a hsp70:fgf3 line did not affect the anterior events ( folding and cell ingression , Figure 5—figure supplement 1F and G ) suggesting that endogenous anterior FGF levels are sufficient to mediate these processes . However , this manipulation led to a delay in folding of the posterior part of the epithelium , ( a region where endogenous FGFs are not acting ) , supporting the notion that FGFs regulate otic epithelialisation . Altogether , these results suggest that endogenous FGF activity delays the final steps of anterior otic placode morphogenesis , providing time for cell ingression before the epithelial barriers appear . 10 . 7554/eLife . 25543 . 025Video 13 . Synchronous folding of the anterior and posterior regions of the otic placode in dnfgfr1-EGPF expressing embryos . Time-lapse during placode morphogenesis in Tg ( dnfgfr1-EGFP ) embryos heat-shocked at 10 hpf . Lines indicate the epithelial folding . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 025 Although important in other contexts , the control of proliferation does not seem to play a central role in the FGF signalling effect on otic specification , as blocking FGF did not modify the number of otic cells positive for phospho-Histone 3 ( pH3+ cells , Figure 5—figure supplement 1 , A and B ) . Moreover , not only does the FGF pathway control the number of neurog1+ cells but also the mean levels of neurog1 expression ( as we show above with the AH experiments , both parameters were not coupled ) . The otic neurogenic domain emerges in a defined ventroanterolateral position due to the dialogue of several signalling pathways that regionalise the otic placode ( Maier et al . , 2014; Fekete and Wu , 2002; Abello and Alsina , 2007; Raft and Groves , 20142015 ) . In light of this , within the otic placode the fate of each cell would be dictated by its position in the tissue ( Bok et al . , 2007 , 2005; Brigande et al . , 2000; Whitfield and Hammond , 2007 ) upon the influence of the extrinsic signals . However , we observe that some ingressing cells are specified prior to their incorporation to the anterolateral domain of the otic epithelium . Moreover , when ingressing cells are laser ablated , the cells in the otic vesicle located in the position of the ingressed cells ( i . e . receiving the same putative diffusing morphogens ) do not seem to adopt a neurogenic fate . This suggests that secreted factors establish a region competent for neurogenic specification , to which the ingressing cells ( and probably other mechanisms ) provide instructive signals to induce neurog1 expression . In agreement with this possibility , Tbx1 , the main transcription factor involved in otic neurogenic regionalisation , is a repressor of neurog1 expression . Tbx1 is excluded from the anterior part of the vesicle , making the region competent to be induced by neurogenic signals ( Bok et al . , 2011; Radosevic et al . , 2011; Raft et al . , 2004 ) . Thus , in addition to the reported role of cell movements on the spatial delimitation of different domains of the neural tube ( Xiong et al . , 2013; Kicheva et al . , 2014 ) , we propose that coordination between cell movement and cell communication contributes to the neuronal pattern of the otic vesicle . In embryos mutant for FGF3 , FGF8 and FGF10 , and embryos in which FGF signalling has been temporally blocked , distinct phases of otic neural development are impaired ( Wright and Mansour , 2003; Zelarayan et al . , 2007; Pirvola et al . , 2000; Léger et al . , 2002; Vemaraju et al . , 2012; Alsina et al . , 2004; Alvarez et al . , 2003 ) . Our work indicates that FGF signalling promotes ingression of pioneer cells into the neurogenic domain , suggesting that some of the previously reported effects on neurog1 expression could be due to this novel role . Additionally , FGF signalling is known to control cell behaviour in other organs , such as epithelialisation and cell migration during kidney tubulogenesis and lateral line development ( Atsuta and Takahashi , 2015; Aman and Piotrowski , 2008 ) . Particularly in the inner ear , FGF signalling controls epithelial invagination during otic morphogenesis in the chick ( Sai and Ladher , 2008 ) . We have identified a role of this pathway in zebrafish otic morphogenesis , delaying tissue folding during epithelialisation , and thus influencing neurogenesis . Additionally , it is possible that the FGF pathway also impinges on cell migration . The candidate ligands for the FGF effects on morphogenesis might be FGF8 and FGF3 coming from the hindbrain ( Maves et al . , 2002 ) and FGF3 from the endoderm and mesoderm ( McCarroll and Nechiporuk , 2013 ) . FGF10a is also expressed at these stages in the region where the pioneer cells are migrating ( McCarroll and Nechiporuk , 2013 ) . However , neurog1 expression is normal in otic vesicles of FGF10a mutant embryos ( Figure 5—figure supplement 1C and D ) , indicating that this ligand is most probably not involved in these processes . A question that emerges from our analysis is how ingressing cells regulate neurog1 expression in their neurogenic domain neighbours . The Notch pathway could participate in this process . However , since Notch activation reduces the number of specified neuronal cells via lateral inhibition ( Haddon et al . , 1998; Abelló et al . , 2007 ) and ingression enhances it , the instructive signal should inhibit Notch activity in the resident cells of the vesicle . Given that inhibition of cell ingression reduced not only the number of neurog1+ cells but also the mean expression levels , the mechanism for instruction seems to rely on the activation of the neurog1 promoter more than in stimulation of proliferation . This hypothesis is supported by the fact that: ( a ) FGF pathway blockade reduced both the number of neurog1+ cells and the mean neurog1 expression levels without affecting proliferation , and ( b ) AH inhibition of proliferation did not affect the mean levels of neurog1 expression . Our 4D analysis allowed us to address the mode of division in the otic neurogenic domain for first time . We found that in all cases including both neurog1− and neurog1+ cells , both daughter cells acquire a neuronal fate . During the time frame analysed , no divisions were found where one daughter cell remained as a neurog1- progenitor while the other activated the proneural expression , as has been described in the neural tube ( Wilcock et al . , 2007; Das and Storey , 2012; Taverna et al . , 2014 ) . We cannot exclude , however , that asymmetric divisions occur at later times or at very low frequency . Studies of fixed chick otic vesicles described the presence of mitosis in the basal side of the epithelium in addition to the luminal ones ( Alvarez et al . , 1989 ) . Such mitoses were termed ‘basal divisions’ similar to the ones taking place in the retina in which mitotic cells are no longer polarized apically and in contact with the ventricular membrane ( Weber et al . , 2014 ) . In our study , we also observed non-luminal mitoses , but our data show that these divisions remain in contact with a Pard3 scaffold and therefore still keep their apical polarity . Neural specification usually occurs in epithelialised tissues . However , we observed activation of neurog1 expression in pioneer cells before epithelialisation , suggesting that stable cell-cell contacts would be dispensable to initiate proneural expression . Similarly , in mouse neurog2 is expressed in migrating sensory neuron precursors ( Marmigère and Ernfors , 2007 ) , although its expression begins before exiting the epithelium and migration ( Zirlinger et al . , 2002 ) . We were able to visualise the transit of an otic neuronal progenitor from neurog1 expression to delamination . Analysis of neurog1 expression levels suggests that delamination occurs once a given threshold of proneural expression is reached; probably associated to neurod1 induction . The otic placode and other cranial placodes originate from a large common pre-placodal region ( PPR ) adjacent to the neural plate ( Bailey and Streit , 20052006 ) . Precursors from the PPR segregate and coalesce into individual cranial placodes , which progressively acquire specific identities ( Breau and Schneider-Maunoury , 2014; Streit , 2002; Bhat and Riley , 2011; Saint-Jeannet and Moody , 2014; McCarroll et al . , 2012 ) . Our data revealed that otic neurog1 is expressed before of what it was conceived and outside the epithelium by a group of cells that ingress during morphogenesis . This suggests that neural specification might precede the acquisition of a defined placodal identity . Thus , we propose that some PPR precursors might already be committed to a neural fate and that their subsequent allocation into the placodes ( by random or directed movements ) provides them one or another placodal identity . Further work in this direction might shed light into this hypothesis . In conclusion , our study reveals that cell movements underlie an instruction essential for otic neuronal specification , a crucial step in neurogenesis . Unravelling the complex mechanisms that determine the number of neurons incorporated in a forming ganglion may provide insights leading to a better understanding of the anomalies associated with auditory neuropathies . The following zebrafish lines were used in this study: AB wild-type , TgBAC ( neurog1:DsRedE ) nl6 ( Drerup and Nechiporuk , 2013 ) , Tg ( neurod:GFP ) ( Obholzer et al . , 2008 ) , Tg ( actb1:Lifeact-GFP ) ( Behrndt et al . , 2014 ) Tg ( Xla . Eef1a1:H2B-Venus ) ( Recher et al . , 2013 ) , Tg ( hsp70:dnfgfr1-EGFP ) pd1 ( Lee et al . , 2005 ) , Tg ( elA:GFP ) ( Labalette et al . , 2011 ) , neurog1hi1059 ( Golling et al . , 2002 ) , Tg ( hsp70:fgf3 ) ( Hammond and Whitfield , 2011 ) , and a cross between the TgBAC ( neurog1:DsRedE ) nl6 and the mutant fgf10a+/− ( Norton et al . , 2005 ) . They were maintained and bred according to standard procedures ( Westerfield , 1993 ) at the aquatic facility of the Parc de Recerca Biomèdica de Barcelona ( PRBB ) . All experiments conform to the guidelines from the European Community Directive and the Spanish legislation for the experimental use of animals . Live embryos were embedded in low melting point agarose at 1% in embryo medium including tricaine ( 150 mg l−1 ) for dorsal confocal imaging using a 20x ( 0 . 8 NA ) glycerol-immersion lens . Imaging was done using a SP5 Leica confocal microscope in a chamber heated at 28 . 5°C . 20 to 80 µm thick z-stacks spanning a portion or the entire otic vesicle ( a z-plane imaged every 0 . 5–2 µm ) were taken every 1 to 3 min for 2–12 hr . Raw data were processed , analysed and quantified with FIJI software ( Schindelin et al . , 2012 ) . For visualisation purposes , the images were despeckled . For quantifications of neurog1 expression , images were not modified . Videos were assembled selecting a plane from every z-stack at every time point to better visualise the phenotype ( or track a cell ) or shown as 3D reconstructions . A representative video from at least three different embryos is shown . Images in figures are either shown as confocal coronal sections , 3D reconstructions or average z-projections . To track the trajectory of individual cells , 3D videos were analysed using the MtrackJ , Manual tracking plugins of ImageJ ( Meijering et al . , 2012 ) , and temporal colour code applied to generate a single image of the tracks . To perform quantifications in different regions of the otic vesicle , we live imaged a z-stack and built a rectangular cuboid defined by external vertices of the otic vesicle . The cuboid was divided in eight equally sized regions , and quantifications were performed inside each region . Before quantification , the z-stacks were aligned in 3D to correct for variability in orientations during mounting to guarantee the coronal sectioning of the vesicle . For volume calculation , the x-y area of the tissue in each plane of the z-stack was measured and then multiplied by the z spacing every plane ( the volume of the lumen was subtracted ) . The number of cells in each region was determined manually by counting H2B-mCherry stained nuclei on z-stacks , using the Cell counter plugging of ImageJ . 3D visualisation of Lyn-GFP plasma membrane staining helped the identification of each single cell . To quantify the number of cell divisions in the otic epithelium in a period of time , high temporal resolution videos ( 1 min frequency ) in 3D of H2B-GFP stained nuclei were analysed manually to detect every chromosome segregation event . The number of divisions in each region of the vesicle was determined building a cuboid as described above for each time point . To ablate a group of cells , a two-photon laser beam ( 890 nm ) from a Leica SP5 microscope was applied over one side of the embryos mounted in agarose ( the contralateral side was maintained intact as a control ) . We used embryos with mosaic H2B-mCherry nuclear staining ( mRNA injected at 16 cell stage ) to calibrate the settings of the microscope required to ablate 2–3 cells in each ablation pulse ( Figure 4—figure supplement 1A; Video 14 ) . Each pulse consisted in approximately 5 s of 30% laser power applied in a ROI of about 70 µm2 imaged with a 20x air objective and a digital zoom of 64x . In neurog1-DsRedE embryos , the cells to ablate were identified by single photon confocal imaging recognizing the DsRedE fluorescence in cells anterior ( or posterior ) to the otic placode/vesicle . Right after ablation , imaging of the vesicle was performed to confirm the damage caused ( dead cells were clearly visualised ) . Sequential pulses at different locations were applied to ablate an increased number of cells . No damage outside the ablated region was observed . Ablated embryos were maintained mounted at 28°C until the moment in which specification analysis was performed ( see below ) . 10 . 7554/eLife . 25543 . 026Video 14 . Calibration of cell ablation . 3D reconstruction of z-stacks acquired before and after ablation of embryos expressing H2B-mCherry in some cells adjacent to the neural tube . The neighbouring cells remain undamaged after ablation of the targeted cells ( white arrowhead ) . The damage is indicated by the blue arrowhead . The embryos also express globally memb-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 25543 . 026 To detect ingression of cells into the epithelium , photoconversion of NLS-Eos expressing nuclei was performed with UV light ( λ = 405 nm , using a 20x objective in a Leica SP5 system ) on 13 hpf mounted embryos . A 3D ROI of about 1 × 105 μm3 located 25 µm apart from the anterior limit of the epithelialising placode was photoconverted . Photoconversion was checked by confocal imaging right after UV illumination . The number of photoconverted cells was quantified using the Cell counter plugin from FIJI ( DMSO = 58 ± 9 cells; SU5402 = 59 ± 7 cells , n = 8 ) . The embryos were then removed from the agarose and incubated in embryo medium until 20 hpf to check for cell ingression by 3D imaging . When blockade of FGFR was performed , the embryos were dechorionated at 11 hpf , incubated with SU5402 or DMSO in embryo medium until 13 hpf , mounted in agarose including SU5402 or DMSO , photoconverted , imaged , unmounted , and incubated in presence of the drugs in solution until 19 hpf . In some cases , the TgBAC ( neurog1:DsRedE ) nl6 , the neurog1hi1059 ( embryos genotyped by PCR after imaging ) , Tg ( hsp70:fgf3 ) , or Tg ( hsp70:dnfgfr1-EGFP ) lines were used . In the latter case , time-lapses at 5 min resolution time were performed to track photoconverted nuclei over time . To analyse specification phenotypes z-stacks were acquired with fixed settings ( laser power and detector gain ) between different experimental groups ( or vesicles in the case of ablations ) . The settings were adjusted to detect a range of increased or decreased fluorescence levels without saturation or lack of signal . DsRedE fluorescence was quantified in single slices using imageJ . A small region of a few pixels was created and a mean fluorescence level in each cell ( Fcell ) was calculated by averaging three quantifications in different x , y and z positions of the cytosol ( the background was deducted from each measurement ) . To consider a cell positive for DsRedE expression , a threshold was defined empirically for each set of experiments , as the minimum level at which DsRedE expression in different z slices is unambigously detected ( to avoid mistakes produced by fluorescence coming from cells located at other z positions ) . We then calculated the mean Fcell in each vesicle ( F¯ cell ) , the number of neurog1+ positive cells , and the global level of DsRed expression as the sum of the Fcell for all the neurog1+ cells in a vesicle . neurog1+Hi cells were defined as the ones that have fluorescent level higher than 1 . 5x F¯ cell of the control ( DMSO or non-ablated side ) vesicles . Dynamic quantifications were performed by sequentially measuring fluorescence at consecutive times of a video in the same cell . The mean rate of increase in fluorescence was calculated as ΔFΔt . The same single cell fluorescence quantifications were performed in the neuroepithelial cells of the hindbrain , in a region adjacent to the otic vesicle . To label cellular and subcellular structures , mRNA encoding for the following fusion proteins were injected at 1 cell stage after being synthesised with the SP6 mMessenger mMachine kit ( Ambion ) : H2B-mCherry , H2B-GFP or NLS-Eos ( 100–150 pg ) ( Sapede et al . , 2012 ) , Pard3-GFP ( 50–75 pg ) ( Buckley et al . , 2013 ) , Lyn-EGFP ( memb-GFP 100–150 pg ) , membrane-mCherry ( 100–150 pg ) . For the specification analysis , TgBAC ( neurog1:DsRedE ) nl6 dechorionated embryos were treated with SU5402 25 µm ( Merk Millipore 572630 ) , aphidicolin 300 µM ( Merck ) in combination with hydroxyurea 100 mM ( Sigma ) , or DMSO ( Sigma ) added to the embryo medium . For determination of the number of pH3+ cells , DMSO or SU5402 treated embryos from 13 to 16 hpf were fixed and processed for the immunostainings . The heat shock was performed by incubating 10 hpf embryos in preheated water at 39° during 30 min . Fluorescence from DNFGFR1-EGFP was detectable from about one hour after initiation of the shock . Induced embryos were selected at 12 hpf . For photoconversion or laminin immunostaining , EGFP- embryos were used as controls . For DsRedE expression analysis in which a membrane staining is relevant , neurog1:DsRedE embryos injected with memb-GFP at 1 cell-stage were heat shocked and used as controls . For experiments using the fgf10a+/-; neurog1:DsRedE line , the embryos were mounted and imaged at 20 hpf for DsRedE expression analysis , recovered from the agarose , and incubated until 5 dpf , when the fgf10a-/- mutants embryos were identified by the absence of pectoral fins . For immunostaining , dechorionated zebrafish embryos were fixed in 4% PFA overnight at 4°C and immunostaining was performed either on whole-mount or cryostat sections . Embryos for sections were cryoprotected in 15% sucrose and embedded in 7 . 5% gelatine/15% sucrose . Blocks were frozen in 2-Methylbutane ( Sigma ) for tissue preservation and cryosectioned at 14 µm on a Leica CM 1950 cryostat . After washing in 0 . 1% PBT , and blocking in 0 . 1% PBT , 2% Bovine Serum Albumin ( BSA ) , and 10% normal goat serum ( NGS ) for 1 hr at RT , embryos were incubated overnight at 4°C in blocking solution with the appropriate primary antibodies: rabbit anti-Laminin ( Sigma , 1:200 ) , rabbit anti-pH3 ( Abcam , 1:200 ) . After extensive washing in 0 . 1% PBT , donkey anti-rabbit Alexa-488 ( Thermo fisher scientific A21206; 1:400 ) was incubated overnight at 4°C in blocking solution . Sections were counterstained with 1 µg/ml DAPI , mounted in Mowiol ( Sigma-Aldrich ) and imaged in a Leica SP5 confocal microscope . Synthesis of antisense RNA and whole-mount in situ hybridisation were performed as previously described ( Thisse et al . , 2004 ) to generate a probe against neurog1 ( Itoh and Chitnis , 2001 ) . Dechorionated Tg ( elA:GFP ) ( which express GFP in rhombomeres 3 and 5 ) zebrafish embryos were fixed in 4% paraformaldehyde ( PFA ) overnight at 4°C and dehydrated in methanol series , rehydrated again and permeabilized with 10 mg/ml proteinase K ( Sigma ) at RT for 5–10 min depending on their stage . Digoxigenin-labeled probe was hybridised overnight at 70°C , detected using anti-digoxigenin-AP antibody at 1∶2000 dilution ( Roche ) and developed with NBT/BCIP ( Roche ) . After the ISH , an immunostaining for the GFP expressed from the transgene was performed ( primary antibody: rabbit anti-GFP ( Torrey Pinnes; 1:400 ) , secondary antibody: anti-rabbit Alexa-488 ( Thermo fisher scientific A21206; 1:400 ) ) . Embryos were post-fixed overnight in 4% PFA and used for imaging mounted in 100% glycerol . All statistical comparisons are indicated in figure legends including one sample and unpaired t-test performed using GraphPad . The box plot was generated in excel .	The inner ear is responsible for our senses of hearing and balance , and is made up of a series of fluid-filled cavities . Sounds , and movements of the head , cause the fluid within these cavities to move . This activates neurons that line the cavities , causing them to increase their firing rates and pass on information about the sounds or head movements to the brain . Damage to these neurons can result in deafness or vertigo . But where do the neurons themselves come from ? It is generally assumed that all inner ear neurons develop inside an area of the embryo called the inner ear epithelium . Cells in this region are thought to switch on a gene called neurog1 , triggering a series of changes that turn them into inner ear neurons . However , using advanced microscopy techniques in zebrafish embryos , Hoijman , Fargas et al . now show that this is not the whole story . While zebrafish do not have external ears , they do possess fluid-filled structures for balance and hearing that are similar to those of other vertebrates . Zebrafish embryos are also transparent , which means that activation of genes can be visualized directly . By imaging zebrafish embryos in real time , Hoijman , Fargas et al . show that the first cells to switch on neurog1 do so outside the inner ear epithelium . These pioneer cells then migrate into the inner ear epithelium and switch on neurog1 in their new neighbors . A substance called fibroblast growth factor tells the inner ear epithelium to let the pioneers enter , and thereby controls the final number of inner ear neurons . The work of Hoijman , Fargas et al . reveals how coordinated activation of genes and movement of cells gives rise to inner ear neurons . This should provide insights into the mechanisms that generate other types of sensory tissue . In the long term , the advances made in this study may lead to new strategies for repairing damaged sensory nerves .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2017	Pioneer neurog1 expressing cells ingress into the otic epithelium and instruct neuronal specification
Giant viruses are ecologically important players in aquatic ecosystems that have challenged concepts of what constitutes a virus . Herein , we present the giant Bodo saltans virus ( BsV ) , the first characterized representative of the most abundant group of giant viruses in ocean metagenomes , and the first isolate of a klosneuvirus , a subgroup of the Mimiviridae proposed from metagenomic data . BsV infects an ecologically important microzooplankton , the kinetoplastid Bodo saltans . Its 1 . 39 Mb genome encodes 1227 predicted ORFs , including a complex replication machinery . Yet , much of its translational apparatus has been lost , including all tRNAs . Essential genes are invaded by homing endonuclease-encoding self-splicing introns that may defend against competing viruses . Putative anti-host factors show extensive gene duplication via a genomic accordion indicating an ongoing evolutionary arms race and highlighting the rapid evolution and genomic plasticity that has led to genome gigantism and the enigma that is giant viruses . Viruses are the most abundant biological entities on the planet and there are typically millions of virus particles in each milliliter of marine or fresh waters that are estimated to kill about 20% of the living biomass each day in surface marine waters ( Suttle , 2007 ) . This has major consequences for global nutrient and carbon cycles , as well as for controlling the composition of the planktonic communities that are the base of aquatic foodwebs . Although the vast majority of aquatic viruses are less than 100 nm in diameter and primarily infect prokaryotes , it is increasingly clear that a subset of the viruses in aquatic ecosystems are comparative Leviathans that have been colloquially classified as giant viruses . The first isolated giant virus in the family that later became known as the Mimiviridae , infects a marine heterotrophic flagellate that was initially identified as Bodo sp . ( Garza and Suttle , 1995 ) , and later shown to be Cafeteria roenbergensis ( Fischer et al . , 2010 ) . Subsequently , the isolation and sequencing of mimivirus , a giant virus infecting Acanthamoeba polyphaga ( La Scola et al . , 2003; Raoult et al . , 2004 ) , transformed our appreciation of the biological and evolutionary novelty of giant viruses . This led to an explosion in the isolation of different groups of giant viruses infecting Acanthamoeba spp . including members of the genera Pandoravirus , Pithovirus , Mollivirus , Mimivirus and Marseillevirus ( Boughalmi et al . , 2013; Colson et al . , 2013; Legendre et al . , 2014; Legendre et al . , 2015; Philippe et al . , 2013 ) . Although each of these isolates expanded our understanding of the evolutionary history and biological complexity of giant viruses , all are pathogens of Acanthamoeba spp . , a widespread taxon that is representative of a single evolutionary branch of eukaryotes , and which is not a major component in the planktonic communities that dominate the world’s oceans and large lakes . As knowledge of mimiviruses infecting Acanthamoeba spp . has expanded , it has become evident based on analysis of metagenomic data that giant viruses and their relatives are widespread and abundant in aquatic systems ( Hingamp et al . , 2013; Mozar and Claverie , 2014; Schulz et al . , 2017 ) . However , except for Cafeteria roenbergensis Virus ( CroV ) that infects a microzooplankton ( Fischer et al . , 2010 ) , and the smaller phytoplankton-infecting viruses Phaeocystis globosa virus PgV-16T ( Santini et al . , 2013 ) , Chrysochromulina Ericina Virus ( Gallot-Lavallée et al . , 2017 ) , and Aureococcus anophagefferens virus ( Moniruzzaman et al . , 2014 ) , the only members of the Mimiviridae that have been isolated and characterized infect Acanthamoeba spp . Motivated by the lack of ecologically relevant giant-virus isolates , we isolated and screened representative microzooplankton in order to isolate new giant-viruses that can serve as model systems for exploring their biology and function in aquatic ecosystems . Herein , we present Bodo saltans virus ( BsV ) , a giant virus that infects the ecologically important kinetoplastid microzooplankter Bodo saltans , a member of the phylum Euglenazoa within the supergroup Excavata . This group of protists is well represented by bodonids in freshwater environments and by diplonemids in the oceans ( Flegontova et al . , 2016; Simpson et al . , 2006 ) . Kinetoplastids are remarkable for their highly unusual RNA editing and having a single large mitochondrion , the kinetoplast , that contains circular concatenated DNA ( kDNA ) that comprises up to 25% of the total genomic content of the cell ( Shapiro and Englund , 1995; Simpson et al . , 2006 ) , and are well known as causative agents of disease in humans ( e . g . Leishmaniasis and sleeping sickness ) and livestock ( Jackson et al . , 2016; Mukherjee et al . , 2015 ) . At 1 . 39 MB , BsV has one of the largest described complete genome of a cultured strain within the giant virus family Mimiviridae . Based on a recruitment analysis of metagenomic reads , BsV is representative of the most abundant group within the Mimiviridae in the ocean and is the only isolate of the klosneuviruses , a group only known from metagenomic data ( Schulz et al . , 2017 ) . The BsV genome exhibits evidence of significant genome rearrangements and recent adaptations to its host . In an effort to isolate giant viruses that infect ecologically relevant organisms , we isolated protistan microzooplankton from a variety of habitats and screened them against their associated virus assemblages . One such screen using water collected from a temperate eutrophic pond in southern British Columbia , Canada , yielded a giant virus that we have classified as Bodo saltans virus , Strain NG1 ( BsV-NG1 ) that infects an isolate of the widely occurring kinetoplastid , Bodo saltans ( Strain NG , CCCM6296 ) . The addition of BsV to a culture of Bodo saltans ( ~2 . 5 × 105 cells ml−1 ) at a virus particle to cell ratio of two , measured by flow cytometry , resulted in free virus particles 18 hr later . Viral concentrations peaked at 2 . 5 × 107 particles ml−1 , while host cell density dropped to 25% of uninfected control cultures ( Figure 1 ) . The closely related strain Bodo saltans HFCC12 could not be infected by BsV-NG1 , suggesting strain specificity . Transmission electron microscopy ( TEM ) revealed that BsV is an icosahedral particle approximately 300 nm in diameter ( Figure 2A ) . The particle consists of at least six layers akin to observations of Acanthamoeba polyphaga mimivirus ( ApMV ) ( Mutsafi et al . , 2013 ) . The DNA-containing core of the virion was surrounded by a core wall and an inner membrane , and a putative membrane sitting under a double-capsid layer ( Figure 2A ) . A halo of approximately 25 nm surrounds the virion . A possible stargate-like structure , as observed in ApMV , is associated with a depression of the virus core Figure 2—figure supplement 1B , D ) , which presumably releases the core from the capsid during infection ( Klose et al . , 2010; Mutsafi et al . , 2014 ) . The healthy Bodo saltans cell presents intricate intracellular structures , including the characteristic kinetoplast and a pronounced cytostome and cytopharynx ( Figure 2B , Figure 2—figure supplement 1A ) . In infected cells , virus factories were always observed in the cell’s posterior and particles always matured toward the posterior cell pole in a more spatially organized way compared to other Mimiviridae ( Figure 2C , Figure 2—figure supplement 1C ) ( Mutsafi et al . , 2010 ) . As infection progressed , the Golgi apparatus disappeared and the nucleus degraded , as evidenced by the loss of the nucleolus and heterochromatin ( Figure 2B , C ) ; yet , the kinetoplast remained intact , as indicated by the persistence of the characteristic kDNA structure ( Figure 2B , C , Figure 2—figure supplement 1A , C ) . Virus factories were first observed at 6 hr post-infection ( p . i . ) as electron-dense diffuse areas in the cytoplasm . By 12 h p . i . , the virus factory had expanded significantly and reached a maximum size of about one-third of the host cell , taking up most of the cytoplasm . The first capsid structures appeared at this time . At 18 h p . i . , the first mature virus particles were observed , coinciding with the first free virus particles observed by flow cytometry ( Figure 1 ) . By 24 h p . i . , most infected cells were at the late stage of infection with mature virus factories ( Figure 2C , D ) . During virus replication , membrane vesicles were recruited through the virus factory where capsid proteins accumulated and disrupted the vesicles ( Figure 2D ) ( Mutsafi et al . , 2013 ) . The vesicle/capsid structures accumulated in the periphery of the virus factory where the capsid was formed ( Figure 2C , D ) . Once the capsid was completed , the viral genome was packaged into the capsid at the vortex opposite to the putative stargate structure ( Figure 2C , D ) . The internal structures of the virus particle matured in the cell periphery and accumulated below the host cytoplasmic membrane where they often remained for an extended period of time ( Figure 2D , Figure 2—figure supplement 1D ) . Besides being released during cell lysis , mature virus particles were observed budding in vesicles from the host membrane , reminiscent of a mechanism described for Marseillevirus ( Figure 2—figure supplement 1D ) ( Arantes et al . , 2016 ) . Combined PacBio RSII and Illumina MiSeq sequencing resulted in the assembly of a 1 , 385 , 869 bp linear double-stranded DNA genome ( accession number MF782455 ) , making the BsV genome one of the largest complete viral genomes described to date , surpassing those of mimiviruses infecting Acanthamoeba spp . The GC content is 25 . 3% ( Figure 3 ) and , as reported for other giant viruses ( Raoult et al . , 2004 ) , much lower than the ~50% observed for Bodo spp . ( Jackson et al . , 2016 ) ; this suggests the absence of large scale horizontal gene transfer with the host in recent evolutionary history . The genome encodes 1227 predicted open-reading frames ( ORFs ) with a coding density of 85% , with the ORFs distributed roughly equally between the two strands consistent with the constant GC-skew ( Figure 3 ) . Unlike ApMV , BsV does not display a central peak in GC skew and therefore does not have an organized bacterial like origin of replication ( Raoult et al . , 2004 ) . The genomic periphery has a slightly skewed GC ratio due to the tandem orientation of repeated ORFs . Codon preference is highly biased toward A/T-rich codons and the amino acids Lysine , Asparagine , Isoleucine , and Leucine ( 10 , 9 . 8 , 9 . 6 , 8% ) , which are preferentially encoded by A/T only triplets . The translation of the predicted ORFs resulted in proteins ranging from 43 to 4840 aa in length with an average length of 320 aa . Promotor analysis revealed a highly conserved early promotor motif ‘AAAAATTGA’ that is identical to that found in mimiviruses and CroV ( Fischer et al . , 2010; Priet et al . , 2015 ) . A poorly conserved late promotor motif ‘TGCG’ surrounded by AT-rich regions was also observed . ORFs are followed by palindromic sequences , suggesting a hairpin-based transcription termination mechanism similar to ApMV ( Byrne et al . , 2009 ) . Several non-coding stretches rich in repetitive sequences were observed , but no function could be attributed to them . Based on a BLASTp analysis , 40% of ORFs had no significant similarity to any other sequences and remained ORFans ( Figure 4B ) . Most proteins ( 27% ) matched sequences from eukaryotes; two % of these matched best to sequences from isolates of B . saltans . The next largest fraction ( 22% ) were most similar to viruses in the nucleocytoplasmic large DNA viruses ( NCLDV ) group , while the remaining ORFs were most similar to bacterial ( 9% ) or archaeal ( 1% ) sequences . In gene cluster analysis , only 45% of protein-coding gene clusters are shared with related viruses such as CroV and klosneuviruses , highlighting the low number of conserved core genes amongst these viruses ( Figure 4—figure supplement 1 ) . Essential genes for replication , translation , DNA replication and virion structure are located in the central part of the genome , while the periphery is occupied by duplicated genes , including 148 copies of ankyrin-repeat-containing proteins ( Figure 3 ) . While no function could be attributed to 54% of ORFs , the largest identifiable fraction of annotations are involved in DNA replication and repair ( Figure 4A ) . Coding sequences for proteins associated with all classes of DNA repair mechanisms were identified including DNA mismatch repair ( MutS and Uvr helicase/DDEDDh 3'−5' exonucleases ) , nucleotide excision repair ( family-2 AP endonucleases ) , damaged-base excision ( uracil-DNA glycosylase and formamidopyrimidine-DNA glycosylase ) and photoreactivation ( deoxyribodipyrimidine photolyase ) . The repair pathways are completed by DNA polymerase family X and NAD-dependent DNA ligase . Sequences were also found that code for proteins involved in DNA replication , including several primases , helicases , and an intein-containing family-B DNA polymerase , as well as replication factors A and C , a chromosome segregation ATPase , and topoisomerases 1 ( two subunits ) and 2 . Sequences associated with proteins mediating recombination were also identified including endonucleases and resolvases , as well as the aforementioned DNA repair machinery . There were 47 sequences identified that matched enzymes involved in protein and signal modification , with the majority being serine/threonine kinases/phosphatases . These are potentially involved in host cell takeover . The genome of BsV is rich in coding sequences involved in transcription . An early transcription factor putatively recognizing the highly conserved AAAAATTGA motif and a late transcription factor putatively targeting TGCG were identified , whereas the target sequence of a third transcription factor is unknown . Further , a TATA-binding protein , a transcription initiation factor ( TFIIIB ) and a transcription elongation factor ( TFIIS ) were identified that should aid transcription . As well , RNA polymerase subunits a , b , c , e , f , g and I were identified and are assisted by DNA topoisomerases Type 2 and 1B . BsV encodes amRNA specific RNase III , a poly A polymerase , several 5’ capping enzymes and methyl transferases . Transcription is presumably terminated in a manner similar to that described in ApMV , since hairpin structures were detected in the 3’ UTR of most transcripts ( Priet et al . , 2015 ) . They are probably recognized and processed by the viral encoded RNase III in a manner similar to ApMV ( Byrne et al . , 2009 ) . After hairpin loop cleavage , the poly-A tail is added by the virally encoded poly-A polymerase . The 5’ capping is accomplished by the virus-encoded mRNA capping enzyme , as well as several cap-specific methyltransferases . The extensive cap modification suggests that BsV is independent of the trans-splicing of splice-leader mRNA containing cap structures found in kinetoplastids ( Stuart et al . , 1997 ) . BsV also encodes several enzymes associated with nucleic-acid transport and metabolism , including several AT-specific nucleic-acid synthesis pathway components . For instance , adenylosuccinate , thymidylate and pseudouridine synthetases and kinases , as well as ribonucleoside-diphosphate reductase were evident . Other ORFs were associated with nucleotide salvaging pathways , including nucleoside kinases , phosphoribosyl transferases , and cytidine and deoxycytidylate deaminase . A mitochondrial carrier protein was identified that , similar to ApMV , likely provides dATP and dTTP directly from the kinetoplast to the virus factory , as evident from electron microscopic observations ( Monné et al . , 2007 ) . Several genes were identified that are involved in membrane trafficking . A system based on soluble N-ethylmaleimide-sensitive factor ( NSF ) attachment proteins ( SNAPs ) and the SNAP receptors ( SNAREs ) appears to have been acquired from the host by horizontal gene transfer in the recent evolutionary past . In combination with several NSF homologues , including the vesicular-fusion ATPases that also seems to have been acquired from the host . Other proteins putatively involved in membrane trafficking are rab-domain containing proteins , ras-like GTPases , and kinesin motor proteins . The BsV genome encodes four major capsid proteins . One of these proteins contains several large insertions between conserved domains shared among all four capsid proteins , and with 4194 aa boasts a size almost seven times that of its paralogs . This enlarged version of the major capsid protein might be responsible for creating the halo around the virus particles observed by TEM , by producing shortened fibers similar to those observed in ApMV ( Figure 2A ) ( Xiao et al . , 2009 ) . Further , the genome contains two core proteins , several chaperones and glycosylation enzymes suggesting that proteins are highly modified before being incorporated into the virus particle . There were numerous ORFs that were similar to genes encoding metabolic proteins , like enzymes putatively involved in carbohydrate metabolism . However , no one continuous metabolic pathway could be assembled and therefore these enzymes likely complement host pathways . BsV also encodes coenzyme synthetases such as CoA and NADH and to meet the demand for amino acids that are rare in the host , BsV encodes the key steps in the synthesis pathways of glutamine , histidine , isoleucine , and asparagine . Another group of genes putatively mediate competitive interactions , either directly with the host , or with other viruses or intracellular pathogens . These include genes involved in the production of several toxins such as a VIP2-like protein as well as antitoxins containing BRO domains . Further , a partial bleomycin detox pathway was found , as well as multidrug export pumps and partial restriction modification systems . While BsV encodes a complex translation machinery , it differs markedly from those described in other members of the Mimiviridae and shows the largest turnover of these genes resulting in a net contraction ( Figure 4C ) . Eukaryotic translation initiation factors include the commonly seen eIF-2a , eIF-2b , eIF-2g , eIF-4A-III and eIF-4E , as well as several pseudogenes related to eIFs . Eukaryotic elongation factor 1 is also present as is eukaryotic peptide chain release factor subunit 1 . Notable is the absence of eIF-1; instead , BsV encodes a version of IF-2 that appears to have been acquired from the host and is functionally analogous to eIF-1 in kinetoplastids . The most striking difference to other NCLDVs is the absence of tRNAs . Uniquely among NCLDVs , BsV encodes several tRNA repair genes . These genes include putative RtcB-like RNA-splicing ligase , putative CAA-nucleotidyltransferase , tRNA 2'-phosphotransferase/Ap4A_hydrolase , putative methyltransferase , a TRM13-like protein , pseudouridine synthase and tRNA ligase/uridine kinase . Most of these genes appear to have been recently acquired from the host ( Figure 4—figure supplement 2 ) . Other translation modification enzymes found in BsV and other NCLDVs include tRNA ( Ile ) -lysidine synthase , tRNA pseudouridine synthase B and tRNA 2'-phosphotransferase . Similar to the tRNAs , there are few aminoacyl-tRNA synthetases ( aaRS ) in BsV . Three of the recognizable aaRS are pseudogenes and show signs of recent nonsense mutations or ORF disruptions by genome rearrangements ( aspRS , glnRS , and asnRS ) . The only complete aaRS proteins are isoleucine-tRNA synthetase , found in all members of the Mimivridae , and a phenylalanyl-tRNA synthetase . Genes in the genomic periphery have undergone massive duplication , with 148 copies of ankyrin repeat proteins , mostly present in directional tandem orientation ( Figure 3 ) . These sequences are quite variable and encode between 4 and 17 ankyrin-repeat domains . There is evidence of very recent sequence duplication resulting in direct or inverted repeat regions that contain complete ankyrin-repeat coding sequences and further expand the repeat clusters ( Figure 3—figure supplement 1A ) . Interestingly , the 5’ coding region of many ankyrin-repeat containing protein ORFs contain fragments of catalytic domains of essential viral genes such as DNA polymerases or the MutS repair protein ( Figure 3—figure supplement 1C ) . In contrast to described giant virus genomic mobilomes consisting of virophages and transpovirons , the BsV genome is dominated by inteins , autocatalytic proteinases , and self-splicing group 1 introns ( Desnues et al . , 2012; Fischer and Suttle , 2011; Santini et al . , 2013; La Scola et al . , 2008 ) . These mobile elements spread by targeting the DNA coding regions of essential genes for virus replication by deploying unrelated homing endonucleases encoded by internal ORFs nested within the elements ( red ORFs in Figure 5 ) . Inteins that are closely related to those in Mimiviridae and Phycodnaviridae are found in the BsV DNA polymerase family B gene , while other unrelated inteins are found in the DNA-dependent RNA polymerase subunits A and B genes ( polr2a and polr2b: Figure 5 ) . The inteins in the RNA polymerase genes seem to be devoid of an active homing endonuclease , and are therefore fixed , suggesting an evolutionary ancient invasion . The intein in DNA pol B may be an exception , as an HNH endonuclease is located in close genomic proximity and could promote homing in a trans-acting fashion . The group 1 self-splicing introns seem to have independently invaded the RNA polymerase subunit 1 and 2 genes , since these introns carry different homing endonucleases ( HNH and GIY-YIG type ) and their ribozymes differ in secondary structure ( Figure 5 , Figure 5—figure supplement 1B ) . Subsequent to the initial integration of introns containing endonucleases ( e . g . polr2a-i1b in Figure 5 ) , these homing endonucleases seeded ‘offspring’ introns within the same gene ( e . g . polr2a-i1a and –i1c in Figure 5 ) . These secondary introns show conserved secondary RNA structure , but lack the homing endonuclease of their parental intron . Therefore , the secondary introns probably rely on the trans-homing of their parental introns’ endonucleases . The highly conserved sequence for some of the offspring introns ( 94 . 4% sequence identity for polr2a-i1a and -i1c , Figure 5—figure supplement 1A ) suggests that these have spread relatively recently , while other introns that only show conservation in their secondary structure probably represent older invasions . Besides proliferating introns , the BsV genome is also home to two distinct actively proliferating transposon classes . Phylogenetic analysis of BsV places it within the Mimiviridae . Whole genome analysis based on gene cluster presence/absence of NCLDVs resulted in BsV clustering within the large mimiviruses ( recently proposed ‘Megavirinae’ ) and separate from the small mimiviruses ( ‘Mesomimivirinae’ ) ( Figure 6A ) ( Gallot-Lavallée et al . , 2017 ) . BsV is closest affiliated with the genomes of the Klosneuviruses , assembled from metagenomic data , and to a lesser degree with CroV . Phylogenetic analysis of five concatenated highly conserved NCLDV core genes reproduced this pattern within the Mimiviridae ( Figure 6B ) . Within the ‘Megavirinae’ , three clades emerged , the Acanthamoeba-infecting Mimiviruses , the metagenomic klosneuviruses ( Cato- , Hoko- , Klosneu- , and Indivirus ) with BsV , and CroV as the sole member of an outgroup ( Figure 6B ) . Phylogenies of the individual genes placed BsV within the klosneuviruses in three of five cases ( Figure 6—figure supplements 1 , 2 ) . When metagenomic reads of NCLDV DNA polymerase B sequences from the TARA oceans project ( http://www . igs . cnrs-mrs . fr/TaraOceans/ ) were mapped to a maximum likelihood tree of DNA polymerase B sequences fromthe Mimiviridae , it was apparent that the ‘Megavirinae’ were dominated by klosneuviruses , with BsV as their only isolate ( Figure 6C ) . Thus , BsV is representative of the largest group of identifiable icosahedral giant viruses in the oceans with CroV being the sole representative of the second most abundant clade . AaV: Aureococcus anophagefferens virus; AcV: Anomala cumrea entomopoxvirus; AMaV: Acanthamoeba castellanii mamavirus; AMgV: Moumouvirus goulette; AMoV: Acanthamoeba polyphaga moumouvirus; ApMV: Acanthamoeba polyphaga mimivirus; ASFV: African swine fever virus BA71V; AtcV: Acanthocystis turfacea chlorella virus 1; BpV: Bathycoccus sp . RCC1105 Virus; BsV: Bodo saltans virus NG1; CatV: Catovirus; CeV: Chrysochromulina ericina virus 1B; CroV: Cafeteria roenbergensis virus BV-PW1; EhV: Emiliania huxleyi virus 86; FauV: Faustovirus E12; HokV: Hokovirus; HvV: Heliothis virescens ascovirus 3e; IiV: Invertebrate iridescent virus; IndV: Indivirus; ISKV: Infectious spleen and kidney necrosis virus; KloV: Klosneuvirus; LauV: Lausannevirus; MarV: Marseillevirus T19; MoVs: Mollivirus sibericum; MpVS: Micromoas pusillae Virus SP-1; MsV: Melanoplus sanguinipess entomopoxvirus; MVc: Megavirus chiliensis; MyxV: Myxoma virus; OLV1: Organic Lake Phycodnavirus 1; OLV2: Organic Lake Phycodnavirus 2; OtV: Ostreococcus tauri virus 1; PbCV: Paramecium bursaria chlorella virus 1; PgV: Phaeocystis globosa virus 16T; PiVs: Pithovirus sibericum P1084-T; PoV: Pyramimonas orientalis virus; PpV: Phaeocystis pouchetii virus; PVd: Pandoravirus dulcis; PVs: Pandoravirus salinus; SfV: Spodoptera frugiperda ascovirus 1a; SGV: Singapore grouper iridovirus; TnV: Trichoplusia ni ascovirus 2 c; VacV: Vaccinia virus; WiV: Wiseana iridescent virus; YLV1: Yellowstone lake phycodnavirus 1 Particle structure , functional features , like the transcription machinery , and phylogenetic analysis firmly place BsV within the Mimiviridae , making it one of the largest completely sequenced genome of the family . BsV groups with the klosneuviruses and is separate from CroV and the Acanthamoeba infecting mimiviruses , and falls within the proposed ‘Megavirinae’ ( Figure 6 ) . A separate subfamily was proposed for the metagenomic klosneuviruses , the ‘Klosneuvirinae’ , which would make BsV as the first isolate and the type species of this subfamily ( Schulz et al . , 2017 ) . The high representation of the klosneu- and BsV-like viruses in metagenomic reads from the TARA oceans survey is consistent with BsV being representative of the most abundant group of Mimiviridae , and possibly all icosahedral giant viruses in the oceans ( Hingamp et al . , 2013 ) . The initial detection of the klosneuviruses in low complexity fresh water metagenomes further supports the global prevalence of this group . The SNAP/SNARE membrane fusion system found in BsV appears to have been recently acquired from the bodonid host via horizontal gene transfer . This system could mediate membrane fusion in a pH-dependent manner ( Itakura et al . , 2012 ) . Accordingly , we propose a phagocytosis-based infection strategy for BsV: As described for ApMV , BsV is ingested through the cytostome and is phagocytosed in the cytopharynx before being transported in a phagosome toward the posterior of the cell ( Mutsafi et al . , 2010 ) ; here the viral SNAP/SNARE interacts with the host counterparts to initiate the fusion of the inner virus membrane with the phagolysosomal membrane upon phagolysosome acidification , releasing the viral genome into the cytoplasm . This scenario is supported by the localization of the virus factory at the posterior of the cell and virus particle structure ( Figure 2A , C and Figure 2—figure supplement 1B–C ) . According to this hypothesis , SNAP/SNARE proteins must be present in membranes of the mature virus particles and only get exposed after the stargate opens . The SNAP/SNARE system might also be involved in recruiting membrane vesicles from host organelles to the virus factory during maturation of the virus particle as has been described for pox viruses ( Figure 2D ) ( Laidlaw et al . , 1998 ) . As a representative of environmentally highly abundant viruses , BsV might regularly experience competition for host resources . The putative toxin-antitoxin systems observed in BsV might be involved in competing with other parasites of viral or prokaryotic nature for these resources , by inhibiting their metabolism or damaging their genome as proposed for ApMV ( Boyer et al . , 2011 ) . Most remarkable , however , are the site-specific homing endonucleases encoded by the self-splicing group 1 introns and inteins that have invaded several genes essential for BsV replication . These invasions also seem to be part of the competitive arsenal of BsV , fending off related virus strains competing for abundant and common hosts such as bodonids . During superinfection of two related viruses , having selfish elements encoding homing endonucleases targeting essential genes might be a competitive advantage . As the two competing virus factories are established in the cytoplasm , the endonucleases encoded within the intron or intein cleave the unoccupied locus in the genome of the intron/intein-free virus . The intron or intein containing virus’ genome stays intact , since the target sequences of the endonucleases within its genes are masked by the insertion of the intron or intein . Thus , the intron/intein-containing virus is reducing the ability of the competing virus to replicate ( Figure 7 ) . A similar mechanism has been described for competing phages in which an intron-encoded or derived homing endonuclease mediates marker exclusion during superinfection , causing selective sweeps of genes in the vicinity of the endonuclease through the phage population ( Belle et al . , 2002; Goodrich-Blair and Shub , 1996; Kutter et al . , 1995 ) . More credence to this hypothesis is given by the RNA polymerase sequences encoded by the proposed catovirus and klosneuvirus ( Schulz et al . , 2017 ) . These genes ( polr2a and polr2b ) are fragmented in a manner similar to that observed in BsV , and also appear to encode homing endonucleases between gene fragments , suggesting the presence of self-splicing introns is common in the relatives of BsV such as the klosneuviruses . Since the hosts for klosneuviruses are unknown , it is possible that they compete with BsV for the same hosts , or at least experience similar competition . The presence of non-fixed inteins in other giant viruses hints to past invasions and selection for inteins in a manner resembling the hypothesis proposed here for introns ( Culley et al . , 2009; Gallot-Lavallée et al . , 2017 ) . The retention of fixed inteins in BsV and other giant viruses suggests that there are still viruses in the environment that encode the relevant endonucleases that apply selective pressure to retain the inteins . A similar situation might explain the presence of an intein in the DNA polymerase of Pandoravirus salinus , but not in P . dulcis . Thus , pandoraviruses may be an excellent model to experimentally explore the proposed mechanism of intron homing endonuclease mediated competition . The absence of tRNAs in the BsV genome is remarkable since tRNAs are found in all complete genomes of giant viruses and even in many moderately sized NCLDV genomes ( Figure 4C , Figure 4—figure supplement 3 ) . This might be an adaptation to the unusual RNA modification found in kinetoplastids that also encompasses tRNA editing ( Alfonzo and Lukeš , 2011; Stuart et al . , 1997 ) . Similarly , Trypanosoma mitochondrial tRNAs are exclusively nuclear encoded ( Hancock and Hajduk , 1990 ) . BsV likely cannot replicate this unusual editing , and thus relies on using host tRNAs . Hence , BsV encodes tRNA repair genes to compensate for the lack of tRNA synthesis and to maintain the available tRNA pool in the host cell . Most of these genes appear to have been recently acquired from the host ( Figure 4C , Figure 4—figure supplement 2 ) . Like the tRNAs , most virus-encoded aminoacyl-tRNA synthetases might not recognize the highly modified tRNAs present in the host and are therefore degrading in the absence of positive selective pressure ( Figure 4C , Figure 4—figure supplements 2 , 3 ) . The presence of three recognizable pseudogenes of aminoacyl-tRNA synthetases is especially remarkable in this context and argues for an evolutionary recent degradation ( Figure 4—figure supplement 2 ) . This turnover in translational machinery components in BsV and the klosneuviruses , combined with the apparent diverse origin of these genes , suggests that the translation machinery found in giant viruses is the result of rapid adaptation by gene acquisition via horizontal gene transfer , as has been recently proposed by Schulz et al . ( 2017 ) . BsV demonstrates that such genes can be readily purged from the virus genome if they are not required in a new host . Thus , BsV provides further evidence that the translation machinery encoded by NCLDVs is a homoplasic trait and need not be ancient in origin . The 148 copies of ankyrin-repeat domain proteins in the genomic periphery of BsV are telltale signs of an expanded genomic accordion ( Figure 3 ) ( Elde et al . , 2012 ) . The observation of almost identical sequences in the very periphery of the genome is consistent with the genomic accordion hypothesis , in which the most recent duplications are closest to the genome ends ( Figure 3 , Figure 3—figure supplement 1A , B ) . The genomic recombinations causing the gene duplications can also lead to the disruption of coding sequences that might explain the comparatively low coding density of BsV . Proteins with ankyrin-repeat domains are multifunctional attachment proteins that in pox viruses determine host range by inhibiting host innate immune system functions ( Camus-Bouclainville et al . , 2004 ) . Further ankyrin-repeat domain proteins are used by bacterial intracellular pathogens like Legionella to manipulate eukaryotic host cells ( Pan et al . , 2008 ) . The presence of fragments of the catalytic domains of essential viral genes in many ankyrin-repeat containing genes is of further importance ( Figure 3—figure supplement 1C ) . This suggests a decoy defense mechanism , where these fusion proteins mimic the targets of host antiviral defense systems disrupting essential viral functions . By acting as decoy targets , they immobilize the proposed host factors upon binding via their ankyrin-repeat domains similar to what occurs in vaccinia virus ( Elde et al . , 2009 ) . The immobilized host factors might even be degraded in a ubiquitin-dependent manner reminiscent of the situation in pox viruses as suggested by the presence of several ubiquitin conjugating enzymes encoded in the BsV genome ( Sonnberg et al . , 2008 ) . An ankyrin-repeat-based defense system might explain the observation of cells surviving or avoiding infection that can persist in the presence of the virus ( Figure 1C ) . Alternatively , the protein-protein interaction mode of ankyrin repeat proteins might aid attachment and induction of phagocytosis as the bodonid host cells have changing surface antigens ( Jackson et al . , 2016 ) . Whatever the true function of the ankyrin-repeat proteins might be , they clearly highlight the importance of the genomic accordion in giant virus genome evolution driven by evolutionary arms races and complement previous observations of a contracting genomic accordion in ApMV ( Boyer et al . , 2011 ) . Bodo saltans virus ( BsV ) has the one of the largest sequenced genome of the Mimiviridae and is representative of the most abundant members of this family in aquatic ecosystems . BsV is also the first described DNA virus that infects kinetoplasts , or any member of the supergroup Excavata , a major evolutionary lineage of eukaryotes , and is the first isolate of a subfamily within the Mimiviridae that was based only on metagenomic data . BsV highlights the genomic plasticity of giant viruses via the genomic accordion , which allows for large-scale genome expansions and contractions via non-homologous recombination . The recent duplications in BsV demonstrate genome expansion in action and exemplifies the mechanisms leading to genome gigantism in the Mimiviridae . Further , the function of the expanding genes suggests that strong evolutionary pressure is placed on these viruses by a virus-host arms race that has driven genomic expansion . Genomic plasticity is further apparent in the translational machinery , which shows signs of recent gene loss and rapid adaptation to its bodonid host . This emphasizes that the translational machinery of giant viruses is an acquired homoplasic trait not derived from a common ancestor . An invasion of selfish elements in essential genes suggests interference competition among related viruses for shared hosts . Bodo saltans virus provides significant new insights into giant viruses and their biology . Virus concentrates were collected from 11 fresh water locations in southern British Columbia , Canada ( 49°49'4"N , 123° 7'46"W; 49°42'5"N , 123° 8'47"W; 49°37'34"N , 123°12'27"W; 49° 6'12"N , 122° 4'38"W; 49° 5'22"N , 122° 7'1"W; 49°18'10"N , 122°42'9"W; 49° 8'27"N; 123° 3'16"W; 49°13'21"N , 123°12'43"W; 49°13'13"N , 123°12'41"W; 49°14'52"N , 123°13'59"W; 49°15'58"N , 123°15'34"W ) . To concentrate giant viruses , 20 l water samples were prefiltered with a GF-A filter ( Millipore , Bedford , MA; nominal pore size 1 . 1 um ) over a 0 . 8 um PES membrane ( Sterlitech , Kent , WA ) . Filtrates from all locations were pooled and were concentrated using a 30 kDa MW cut-off tangential flow filtration cartridge ( Millipore , Bedford , MA ) ( Suttle et al . , 1991 ) . Bodo saltans strain NG , the host of BsV , was isolated from a water sample collected near the sediment surface of the pond in Nitobe Memorial Garden , The University of British Columbia , Canada ( 49°15'58"N , 123°15'34"W ) . Clonal cultures were obtained by end-point dilution , and maintained in modified DY-V artificial fresh water media with yeast extract and a wheat grain ( Andersen et al . , 2005 ) . The identity of the host organism was established by 18S sequencing and the strain was deposited at the Canadian Center for the Culture of Microorganisms ( http://www3 . botany . ubc . ca/cccm/ ) reference number CCCM 6296 ( von der Heyden and Cavalier-Smith , 2005 ) . Bodo saltans NG cultures were inoculated at approximately 2 × 105 cells/ml with the pooled giant virus concentrate from all 11 locations . Cell numbers were determined by flow cytometry and compared to a medium only mock-infected control culture ( LysoTracker Green ( Molecular Probes ) vs . FSC on FACScalibur ( Becton-Dickinson , Franklin Lakes , New Jersey , USA ) ) ( Rose et al . , 2004 ) . After a lytic event was observed , the lysate was filtered through a 0 . 8 um PES membrane ( Sterlitech ) to remove host cells . The lytic agent was propagated and a monoclonal stock was created by three consecutive end point dilutions . The concentrations of the lytic agent were screened by flow cytometry using SYBR Green ( Invitrogen Carlsbad , CA ) nucleic acid stain after 2% glutaraldehyde fixation ( vs SSC ) ( Brussaard , 2004 ) . Cell numbers represented in Figure 1C are supplied in ‘Source_data_Fig1C_raw’ . The similarity to the flow cytometry profile of Cafeteria roenbergensis virus suggested that the lytic agent was in deed a giant virus . For Illumina sequencing , exponentially growing B . saltans cultures were infected at a concentration of approximately 5 × 105 cells ml−1 with BsV lysate ( 107 VLP ml−1 ) at a multiplicity of infection ( MOI ) of ~0 . 5 . After 4 days , when host cell densities had dropped below 30% , cultures were centrifuged in a Sorvall SLC-6000 for 20 min , 5000 rpm , 4°C to remove remaining host cells and the supernatant was consecutively subjected to tangential flow filtration with at 30 kDa cut-off ( Vivaflow PES ) and concentrated approximately 100x . Viral concentrates were subjected to ultracentrifugation at 28 , 000 rpm , 15°C for 8 hr in a Ti90 fixed angle rotor ( Beckman-Coulter , Brea , CA ) . Pellets were resuspended and virions lysed using laurosyl acid and proteinase K subjected to pulsed-field gel electrophoresis on a CHEF II pulse field gel electrophoresis aperture ( BioRad ) for 25 hr at 14°C in a 0 . 8% LMP agarose gel with 60–180S switchtimes and 16 . 170 ramping factor in 0 . 5 TBE under 5 . 5 V/cm and 120° . Genomic DNA was visualized under UV light after 30 min SYBR gold ( Invitrogen Carlsbad , CA ) staining . The dominant PFGE band belonging to genomic BsV DNA ( 1 . 35 Mb ) was cut and DNA was extracted using a GELase kit ( Illumina , San Diego ) and ethanol purified according to manufacturer’s protocol . Libraries were prepared using the Illumina Nextera XT kit ( Illumina , San Diego , CA ) as per manufacturer’s recommendation and library quality and quantity were checked by Bioanalyzer 2100 with the HS DNA kit ( Agilent Technology ) . 300 bp paired-end sequencing was performed on an Illumina MiSeq platform by UCLA's Genoseq center ( Los Angeles , CA ) to a nominal sequencing depth of 800x . Sequence quality was examined using FastQC ( RRID:SCR_014583 , http://www . bioinformatics . bbsrc . ac . uk/projects/fastqc/ ) and sequence reads were quality trimmed ( parameters: minlen = 50 qtrim = rl trimq = 15 ktrim = r k = 21 mink = 8 ref=$adapters hdist = 2 hdist2 = 1 tbo = t tpe = t ) and cleared of human ( parameters: minid = 0 . 95 maxindel = 3 bwr = 0 . 16 bw = 12 quickmatch fast minhits = 2 qtrim = lr trimq = 10 ) and PhiX ( parameters: k = 31 hdist = 1 ) sequences against the whole respective genomes using BBMap v35 ( http://sourceforge . net/projects/bbmap/ ) . For PacBio sequencing , BsV was concentrated using precentrifugation and TFF analogously to the Illumina sequencing step . Next , the concentrate was further concentrated by sedimenting it onto a 40% Optiprep 50 mM Tris-Cl , pH 8 . 0 , 2 mM MgCl2 cushion for 30 min at 28 , 000 rpm , 15°C in a SW40Ti rotor in an ultracentrifuge ( Beckman-Coulter , Brea , CA ) . An Optiprep ( Sigma ) gradient was created by underlaying a 10% Optiprep solution in 50 mM Tris-Cl , pH 8 . 0 , 2 mM MgCl2 with a 30% solution followed by a 50% solution and was equilibration over night at 4°C . One ml of viral concentrate from the 40% cushion was added atop the gradient and the concentrate was fractionated by ultracentrifugation in an SW40 rotor for 4 hr at 25 , 000 rpm and 18°C . The viral fraction was extracted from the gradient with a syringe and washed twice with 50 mM Tris-Cl , pH 8 . 0 , 2 mM MgCl2 followed by centrifugation in an SW40 rotor for 20 min at 7200 rpm and 18°C and were finally collected by centrifugation in an SW40 rotor for 30 min at 7800 rpm and 18°C . Purity of the concentrate was verified by flow cytometry ( SYBR Green ( Invitrogen Carlsbad , CA ) vs SSC on a FACScalibur ( Becton-Dickinson , Franklin Lakes , NJ ) . High-molecular-weight genomic DNA was extracted using phenol-chloroform-chloroform extraction . Length and purity were confirmed by gel electrophoresis and Bioanalyzer 2100 wit the HS DNA kit ( Agilent Technology ) . PacBio RSII 20 kb sequencing was performed by the sequencing center of the University of Delaware . Reads were assembled using PacBio HGAP3 software with 20 kb seed reads and resulted in a single viral contig of 1 , 384 , 624 bp , 286 . 1x coverage , 99 . 99% called bases and a consensus concordance of 99 . 9551% ( Chin et al . , 2013 ) . Cleaned up Illumina reads were mapped to the PacBio contig to confirm the PacBio assembly as well as extending the contig’s 5’ end by 1245 bp to a total viral genome length of 1 , 385 , 869 bp . Open-reading frames were predicted using GLIMMER ( RRID:SCR_011931 , Delcher et al . , 2007 ) with a custom start codon frequency of ATG , GTG , TTG , ATA , ATT at 0 . 8 , 0 . 05 , 0 . 05 , 0 . 05 , 0 . 05 as well as stop codons TAG , TGA , TAA , minimum length 100 bp , max overlap 25 , max threshold 30 . Promoter motives were analyzed by screening the 100 bp upstream region of CDS using MEME ( RRID:SCR_001783 , Bailey et al . , 2009 ) . tRNAs were predicted with tRNAscanSE ( RRID:SCR_010835 , Lowe and Eddy , 1997 ) . Group 1 introns were predicted by disruptions in coding sequences and secondary RNA structure was predicted using S-fold ( Ding et al . , 2004 ) . Intron splicing was confirmed using RT-PCR and Sanger sequencing with gene-specific primers designed to span the predicted splice sited predicted by S-fold . Functional analysis of CDS was performed after translation with BLASTp against the nr database with an e-value threshold of 10−5 as well as rps conserved domain search against CDD v3 . 15 . Coding sequences were manually assigned to functional classes based on predicted gene function using Geneious R9 ( Kearse et al . , 2012 ) . The annotated genome of BsV-NG1 was deposited in GenBank under the accession number MF782455 . The Bodo saltans NG 18S sequence was deposited in GenBank under the accession number MF962814 . Whole genome content phylogeny was performed by OrthoMCL ( RRID:SCR_007839 , Li et al . , 2003 ) . Available whole genome sequences of NCLDV from NCBI were downloaded . We first performed gene clustering using OrthoMCL ( 42 ) with standard parameters ( Blast E-value cutoff = 10−5 and mcl inflation factor = 1 . 5 ) on all protein-coding genes of length ≥100 aa . This resulted in the definition of 3001 distinct clusters . Gene clusters are available in source file ‘Source_data_Fig4_S3_all-Mimi_COGs’ , ‘Source_data_Fig4C_translation-Mimi_COGs’ , ‘Source_data_Fig6A-Fig4C_orthoMCL_groups’ , and ‘Source_data_Fig6A_gene-clusters’ . We computed a presence/absence matrix based on the genes clusters and calculated a distance matrix using the according to Yutin et al . ( 2009 ) ( Yutin et al . , 2009 ) . The R script is provided in ‘Source_data_Fig6A_script’ . Gene clusters were used to infer ancestral gene content by posterior probabilities in a phylogenetic birth-and-death model in COUNT ( Csurös , 2010 ) The COUNT file is available under Source_data_Fig4c_count-session . Additional gene substitutions were added to the model where phylogenetic analysis of individual genes strongly suggested accordingly ( see next section ) . The Alignments and phylogenetic tees can be found as ‘Source_data_Fig4_S1-2_XX_YY’ , where ‘XX’ stands for the respective gene and ‘YY’ stands for ‘aln’ or ‘tree’ designates an alignment or tree file . Alignments of aa sequences were performed in Geneious R9 ( RRID:SCR_010519 ) using MUSCLE with default parameters ( RRID:SCR_011812 , Edgar , 2004 ) . Proteins used for the concatenated NCVOG tree were DNA polymerase elongation subunit family B ( NCVOG0038 ) , D5-like helicase-primase ( NCVOG0023 ) , packaging ATPase ( NCVOG0249 ) , Poxvirus Late Transcription Factor VLTF3-like ( NCVOG0262 ) , and DNA or RNA helicases of superfamily II ( NCVOG0076 ) ( Yutin et al . , 2014 ) . Residues not present in at least 2/3 of the sequences were trimmed and ProtTest 3 . 2 was used for amino-acid substitution model selection ( RRID:SCR_014628 , Darriba et al . , 2011 ) . The resulting alignment can be found under ‘Source_data_Fig6B_ncvog_cat’ as well as ‘Source_data_Fig6-S_ncvogXXXX’ representing alignments for the individual gene trees , where ‘XXXX’ represents the NCVOG number . Maximum likelihood trees were constructed with RAxML rapid bootstrapping and ML search with 1000 Bootstraps utilizing the best fitting substitution matrixes determined by prottest ( RRID:SCR_006086 , Stamatakis , 2014 ) . Maximum likelihood trees of translational genes , based on alignments by Schulz et al . ( 2017 ) where available , were constructed using PhyML ( RRID:SCR_014629 , Guindon et al . , 2010; Schulz et al . , 2017 ) . Phylogenetic trees were computed with PhyloBayes-7MPI 1 . 4 f in two Markov Chain Monte Carlo chains under the CAT-GTR model for 10 , 000 to 25 , 000 generations . The consensus tree was based on both chains , removing the first 1000 generations . Convergence was confirmed with bpcomp and tracecomp ( RRID:SCR_006402 , Lartillot et al . , 2013 ) . Trees were visualized in Figtree ( A . Rambaut - http://tree . bio . ed . ac . uk/software/figtree/ ) . Translated environmental assemblies identified by Hingamp et al . as representing NCLDV DNA polymerase B family genes ( http://www . igs . cnrs-mrs . fr/TaraOceans/ ) were mapped to a Mimiviridae DNA polymerase B family reference tree created as described above with pplacer ( Hingamp et al . , 2013; RRID:SCR_004737 , Matsen et al . , 2010 ) . Environmental reads were aligned to the reference alignment using clustalw and were mapped under Bayesian setting . The fat tree was visualized with Archaeopteryx ( https://sites . google . com/site/cmzmasek/home/software/archaeopteryx ) . Of the 401 input sequences , 256 mapped within the Mimiviridae and are displayed in the figure .	In oceans , rivers and lakes , there are about a million viruses in every milliliter of water . Most of these viruses are tiny , often 10 or 100 times smaller than bacteria . However , a few reach a similar size and complexity to bacteria , and so stand out as relative giants . Relative to other viruses , Giant Viruses have much more DNA in their genome , which in turn provides the genetic template to produce the proteins that allow viruses to reproduce largely independently of its host . Typically , more than half of the genes encoded by Giant Viruses have no evident similarity to genes from other viruses or cellular life . Sequencing DNA from ocean water suggests that Giant Viruses are abundant and ecologically important; yet , few have been isolated from the microbes that they infect . Without being able to study Giant Viruses in the laboratory , little can be known about their biology , the way they infect their hosts , and their broader influence on aquatic life . Deeg et al . have now isolated and characterized the giant Bodo saltans virus ( BsV ) , a Giant Virus that infects an ecologically important microbe commonly found in aquatic environments . Sequencing the genome of BsV revealed many previously unknown genes , as well as several unusual features . For example , the genome contains movable genetic elements that might help to fend off other giant viruses by cutting their genomes . In addition , the set of genes used by BsV to translate mRNA templates into proteins differs from those found in other giant viruses , implying that they are not derived from a more complex common ancestor . The size of the genome appears to have grown rapidly by the duplication of genes at the end of the genome – a feature known as a genomic accordion . The identity of the duplicated genes suggests that there is an evolutionary arms race with its host that forces genome expansion . Further studies of the BsV genome could help researchers to understand the origin of gigantism in the genomes of giant viruses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2018	The kinetoplastid-infecting Bodo saltans virus (BsV), a window into the most abundant giant viruses in the sea
Cerebellar granule cell progenitors ( GCP ) proliferate extensively in the external granule layer ( EGL ) of the developing cerebellum prior to differentiating and migrating . Mechanisms that regulate the appropriate timing of cell cycle withdrawal of these neuronal progenitors during brain development are not well defined . The p75 neurotrophin receptor ( p75NTR ) is highly expressed in the proliferating GCPs , but is downregulated once the cells leave the cell cycle . This receptor has primarily been characterized as a death receptor for its ability to induce neuronal apoptosis following injury . Here we demonstrate a novel function for p75NTR in regulating proper cell cycle exit of neuronal progenitors in the developing rat and mouse EGL , which is stimulated by proNT3 . In the absence of p75NTR , GCPs continue to proliferate beyond their normal period , resulting in a larger cerebellum that persists into adulthood , with consequent motor deficits . During development of the central nervous system ( CNS ) proliferation , migration and differentiation of neuronal progenitors , and the precise transition among these processes , is critical for normal development . In the early stages of cerebellar formation , granule cell progenitors ( GCPs ) originate from the rhombic lip of the fourth ventricle and migrate to the anlage of the developing cerebellum forming the external granule layer ( EGL ) ( Altman and Bayer , 1978 ) . In this layer , the progenitor cells proliferate extensively , and subsequently withdraw from the cell cycle and migrate toward the inner granule layer ( IGL ) to form the adult structure of the cerebellum . The precise control of the transition from proliferation to differentiation is key for regulating the final size of the cerebellum . The expansion of granule cell progenitors in the EGL is largely driven by sonic hedgehog ( Shh ) ( Dahmane et al . , 1999; Wallace , 1999; Wechsler-Reya and Scott , 1999 ) . Disturbances in hedgehog signaling can lead to medulloblastoma , the most abundant type of pediatric tumor ( Goodrich et al . , 1997; Zhao et al . , 2015 ) . Several mitogenic ligands for GCPs have been identified in addition to sonic hedgehog , including insulin-like growth factor 2 ( IGF2 ) ( Hartmann et al . , 2005 ) and Notch2 ( Hartmann et al . , 2005 ) , however , little is known about factors that signal withdrawal from the cell cycle and initiation of differentiation . Since GCPs start migrating to the IGL as early as postnatal day ( P ) 4 and continue until P17-P20 , this is a progressive process with waves of cells that exit the cell cycle and start to migrate , while others stay in the EGL and continue to proliferate ( Hatten et al . , 1997 ) , raising the question of what regulates the withdrawal of these progenitors from the cell cycle in the continued presence of Shh . Neurotrophins are a family of growth factors , comprised of NGF , BDNF , NT-3 and NT-4 , that regulate many aspects of neuronal development and function . Neurotrophins are synthesized as precursor proteins of approximately 35 kDa that can either be cleaved to generate the mature factor , or secreted as the proneurotrophin ( Lee et al . , 2001 ) . Mature neurotrophins bind preferentially to Trk receptors to regulate neuronal survival and differentiation , while proneurotrophins preferentially interact with the p75 neurotrophin receptor ( p75NTR ) in a complex with a member of the sortilin family ( Nykjaer et al . , 2004 ) . In the cerebellum , BDNF has been shown to regulate survival and migration of GCPs via the TrkB receptor ( Schwartz et al . , 1997; Borghesani et al . , 2002 ) . Consistent with this role , TrkB is highly expressed throughout the EGL and IGL . In contrast , the p75NTR receptor is highly expressed in the EGL , but not the IGL ( Carter et al . , 2003 ) . The p75NTR can mediate many different functions depending on the cell context by binding to distinct co-receptors and recruiting specific intracellular binding proteins ( Barker , 2004; Charalampopoulos et al . , 2012 ) . The most well-characterized function for p75NTR is in promoting neuronal apoptosis following brain injury ( Friedman , 2010; Ibáñez and Simi , 2012 ) . In the present work , we demonstrate a novel role for p75NTR in regulating proliferation of neuronal progenitor cells in the cerebellum . As with other cellular functions , the regulation of proliferation by p75NTR is dependent on cellular context , this receptor has been shown to promote cell cycle entry in developing retinal and cortical neurons ( Morillo et al . , 2012; López-Sánchez and Frade , 2013; Frade and Ovejero-Benito , 2015 ) and cell cycle exit of a variety of tumor cells ( Jin et al . , 2007 ) and glial cells ( Cragnolini et al . , 2009; 2012 ) . However , a role for p75NTR in regulating cell cycle exit has not been previously demonstrated in neuronal progenitors in the developing brain . We show that the absence of p75NTR led to a delay in cell cycle exit of GCPs , indicating that p75NTR is necessary for timely withdrawal from the cell cycle . The lack of p75NTR was sufficient to increase the size of the cerebellum , a difference that persisted into adulthood , compromising the normal motor/balance function of these animals . In addition , we investigated the p75NTR ligands expressed in the cerebellum that elicited cell cycle arrest of GCPs , and demonstrate a specific role for proNT3 in blocking sonic hedgehog-induced proliferation . Our results suggest that proNT-3 stimulation of p75NTR regulates the timing of cessation of proliferation of GCPs during cerebellar development , and reveal a novel role for p75NTR in developing neuronal progenitors . During development p75NTR is highly expressed in the EGL throughout the cerebellum ( Figure 1a , b ) . As Purkinje cells develop , they also express p75NTR , such that at P21 when the EGL is nearly gone , p75NTR remains expressed in Purkinje cells ( Figure 1a , b ) . A Western blot of cerebellum lysates from different postnatal ages shows peak expression of p75NTR at P7 ( Figure 1c ) . Although the major function for p75NTR that has been defined in the CNS is promoting neuronal death , especially after injury , few apoptotic cells were detected in the cerebellum in vivo in either WT of p75NTR-/- ( Ngfr-/- ) mice ( Figure 1—figure supplement 1 ) , consistent with previous studies ( Carter et al . , 2003 ) . 10 . 7554/eLife . 16654 . 003Figure 1 . Development of p75NTR in the rat cerebellum . ( a ) Low magnification images of sagittal sections through the entire cerebellum from postnatal day ( P ) 2 through P21 showing the abundant immunolabeling for p75NTR in the EGL , which decreases by P21 . Size bar indicates either 200 µm or 400 µm , as indicated . ( b ) High magnification images of lobe 6 showing p75NTR labeling in the outer EGL . Arrows indicate the inner EGL where the neurons lack p75NTR . Size bar indicates 10 µm and is the same for all the images in B . ( c ) Western blot of cerebellum lysates from the indicated ages . Tissue was lysed with RIPA buffer containing protease inhibitors and 20 µg of total protein was separated on a 10% gel and probed for p75NTR . The gel was re-probed for actin as a loading control and is representative of 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 00310 . 7554/eLife . 16654 . 004Figure 1—figure supplement 1 . No differences in TUNEL labeling in the EGL between Ngfr-/- and WT mice . Sections from WT or Ngfr-/- mouse P7 cerebellum were processed for TUNEL labeling . Few labeled cells were detected and no difference between genotypes was observed . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 004 During early postnatal development , when most of the cells in the EGL are actively proliferating , nearly all the GCPs are positive for p75NTR , but as development proceeds and the cells start to differentiate and migrate , two sub-populations can be distinguished in the EGL ( Figures 1b , 2 ) . The external EGL ( eEGL ) where cells express p75NTR and proliferation markers such as Ki67 ( Figure 2a ) and the internal EGL ( iEGL ) where p75NTR expression is downregulated , the cells stop proliferating and express doublecortin ( DCX ) as they start the migratory process toward the IGL ( Figure 2b ) . This clear boundary between proliferating cells expressing p75NTR and migrating cells that lack p75NTR expression suggests that this receptor may be involved in regulating the transition of GCPs from a proliferating to a migrating population . 10 . 7554/eLife . 16654 . 005Figure 2 . Expression of p75NTR , Ki67 , and DCX in the P7 rat cerebellum . ( a ) Confocal image of p75NTR and the proliferation marker Ki67 showing colocalization in the external EGL . Note that p75NTR is also expressed in developing Purkinje cells . ( b ) p75NTR is downregulated in the inner EGL when DCX is expressed . eEGL – external External Granule Layer , iEGL – inner External Granule Layer , PCL – Purkinje Cell Layer . Size bar is 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 005 BDNF is well known to promote survival and migration of granule cells via the TrkB receptor ( Borghesani et al . , 2002; Segal et al . , 1992; Courtney et al . , 1997 ) . Consistent with its role in mediating survival and migration of GCPs , TrkB is highly expressed throughout the EGL , not just in the proliferative external zone where p75NTR is expressed , and is also present throughout the IGL ( Klein et al . , 1990; Minichiello and Klein , 1996 ) . The differential expression of p75NTR and TrkB suggest that these receptors mediate distinct functions in the developing cerebellum . The abundant expression of p75NTR in the EGL , colocalizing with proliferation markers , suggested a possible role in cell cycle regulation of GCPs . To assess this , EdU was injected at P5 , P7 , P10 , and P14 . In wild type mice the age of maximal GCP proliferation in the EGL is at P5 , compared to P7 in the rat . Low magnification images show the overview of the cerebellum with similar EdU incorporation in WT and Ngfr-/- mice at P5 , and the progressive decrease in EGL labeling in the WT beginning at P7 , starting with the anterior lobes . EdU labeling remained high in the Ngfr-/- mice at P7 and P10 and only began to decrease at P14 ( Figure 3a ) . GCPs in lobe 6 were the last to leave the cell cycle , and incorporation of EdU into cells in the EGL of this region was quantified . EdU labeling remained elevated in the Ngfr-/- mice compared with WT , with the difference most apparent at P10 and P14 ( Figure 3b , c ) . To confirm these results , we compared expression of Ki67 in the EGL between wild type and Ngfr-/- mice at different developmental ages . At early postnatal ages ( P2-P5 ) , during maximal proliferation , there was no difference in the number of cells that expressed Ki67 between WT and Ngfr-/- mice ( Figure 3d , e ) . However , at P7 in the EGL of WT mice , as cells began to withdraw from the cell cycle , there was a consequent reduction in the number of Ki67-positive cells ( Figure 3e ) . In contrast , this reduction was not observed in the Ngfr-/- mice , where the number of proliferating cells remained high . As with EdU incorporation , this difference persisted and was even more evident at P10 and P14 , when few proliferating GCPs remained in the EGL of WT mice . Cerebellar tissue taken from WT and Ngfr-/- mice at different developmental ages also showed a clear difference in the level of cyclin E1 at P10 and P14 ( Figure 3f ) . These data indicated that the absence of p75NTR resulted in continued GCP proliferation in the EGL , and support a potential role for p75NTR in regulating the proper timing of cell cycle exit of GCPs . 10 . 7554/eLife . 16654 . 006Figure 3 . Cell cycle withdrawal of GCPs in the EGL is delayed in the Ngfr-/- mice compared to wild type mice . ( a ) Low magnification images of the entire cerebellum showing incorporation of EdU at postnatal ages from P5 to P14 . GCPs in WT mice began to decrease EdU incorporation at P7 , which continued decreasing at P10 and P14 as the progenitors left the cell cycle . Mice lacking p75NTR continued to incorporate EdU at high levels at P7 and P10 , and only began to decrease proliferation at P14 . Size bar is 200 µm . ( b ) High magnification of EdU labeling in lobe 6 showing continued EdU incorporation Ngfr-/- mice compared to WT from P7 through P14 . Nuclei labeled with Draq5 are shown in gray . Size bar indicates 10 µm . ( c ) Quantification of EdU labeling of WT and Ngfr-/- mice . EdU-labeled cells were counted across 150 µm in the EGL of lobe 6b and are graphed relative to the total number of cells labeled with Draq5 . At least three brains per genotype were analyzed at each age . significantly different from WT at p=0 . 0001 . ( d ) Developmental expression of Ki67 from P2 through P14 , confirming the increased expression of proliferation markers in the Ngfr-/- mice compared to WT from P7 thorough P14 . Size bar indicates 10 µm . ( e ) Quantification of cells expressing Ki67 in lobe 6b . Labeled cells were counted across 150 µm in the EGL of lobe 6b and are graphed relative to the total number of cells labeled with Draq5 . At least three brains per genotype were analyzed at each age . significantly different from WT at p=0 . 0001 . Data in graphs in c and e are expressed as mean +/- SEM of at least three independent experiments . Asterisks indicate significance by ANOVA with Tukey’s posthoc analysis , p values indicated below each graph . ( f ) Western blot showing the comparison of cyclin E1 expression in cerebellar lysates from WT or Ngfr-/- mice at the indicated postnatal ages . Differences in cyclin E levels between WT and knockout mice are evident at P7 , P10 and P14 . Blot is representative of three independent experiments . ( g ) Progressive increase in area of the cerebellum in Ngfr-/- mice compared to WT at P5 , P7 and P10 . P7 Ngfr-/- significantly different from P7 WT at 0 . 038 by t-test P10 Ngfr-/-significantly different from P10 WT at p=0 . 0174 by t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 00610 . 7554/eLife . 16654 . 007Figure 3—source data 1 . Mean number of labeled cells of 3 slides for each animal , and statistical analysis for the graphs shown in 3c , 3e , and 3g . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 007 To determine whether the continued proliferation of GCPs in the absence of p75NTR resulted in a larger cerebellum , we compared the area of the cerebellum of WT and Ngfr-/- mice in mid-sagittal sections . At P5 no difference in size was observed , but at P7 and P10 the size of the cerebellum of Ngfr-/- animals was larger compared to WT animals ( Figure 3g ) . This progressive difference in size corresponds to the timing of the differences we observed in expression of proliferation markers , becoming apparent when the GCPs of WT mice began to withdraw from the cell cycle at P7 , while the GCPs of the Ngfr-/- mice continued to proliferate . p75NTR can respond to many different ligands by associating with distinct co-receptors . When it forms a complex with Trk receptors , p75NTR can facilitate responses to mature neurotrophins , however in association with a member of the sortilin family p75NTR binds proneurotrophins . To identify which ligand ( s ) might act via p75NTR to promote cell cycle exit of GCPs , granule cells from P7 rat cerebella were cultured and exposed to Shh with or without the different mature or proneurotrophins in the presence of BrdU to evaluate effects on proliferation . Shh is a well-established mitogen for GCPs , and induced an increase in BrdU incorporation in GCP cultures , as expected . Incubation of cultured GCPs with any of the mature neurotrophins , NGF , BDNF , NT3 , or NT4 had no effect on either the basal level of proliferation or Shh-induced BrdU incorporation , even at a dose of 100 ng/ml at which they can bind p75NTR ( Dechant et al . , 1994 ) ( Figure 4 ) . However , when the proneurotrophins were tested , proNT3 completely prevented the Shh-induced increase in BrdU incorporation ( Figure 5a , b ) . This inhibition was reversed by anti-proNT3 ( Figure 5c ) . Since activation of p75NTR by proneurotrophins is known to promote apoptosis in neurons ( Lee et al . , 2001; Volosin et al . , 2008 ) , TUNEL assay and cleaved caspase 3 staining were performed on cultured GCPs treated with proNT3 . No significant difference was observed when cultured neurons were incubated with or without proNT-3 ( Figure 5 —figure supplement 1 ) , suggesting that the difference in BrdU incorporation was due to cell cycle arrest and not apoptosis of GCPs . Neither proNGF nor proBDNF had any effect on BrdU incorporation induced by Shh ( Figure 5a ) , suggesting that this was a specific effect of proNT-3 . Both proNGF and proBDNF , as well as proNT3 , induced death of cultured hippocampal neurons at the doses tested , confirming that they were functional ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 16654 . 008Figure 4 . Mature neurotrophins have no effect on proliferation of cultured P7 rat GCPs in the absence or presence of Shh . GCPs from P7 rat were cultured with BrdU for 48 hr in the absence or presence of Shh with or without the different mature neurotrophins . BrdU labeling was analyzed by in-cell Western on the LiCor Odyssey . ( a ) GCPs with or without 20 ng/ml or 100 ng/ml of NGF . ( b ) GCPs with or without 20 ng/ml or 100 ng/ml of BDNF . ( c ) GCPs with or without 20 ng/ml or 100 ng/ml of NT3 . ( d ) GCPs with or without 20 ng/ml or 100 ng/ml of NT4 . Data are expressed as mean +/- SEM from three independent experiments . Asterisk indicates that Shh-treated cells increased BrdU incorporation compared to controls by ANOVA with Tukey’s posthoc analysis , p values are indicated below each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 00810 . 7554/eLife . 16654 . 009Figure 4—source data 1 . Mean values for each experiment and statistical analysis for all graphs . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 00910 . 7554/eLife . 16654 . 010Figure 5 . ProNT3 , but not proNGF or proBDNF , prevented Shh-induced proliferation of cultured GCPs . ( a ) GCPs were cultured from P7 rat cerebella with BrdU in the absence or presence of Shh with or without 4 or 8 ng/ml of proNGF , proBDNF or proNT3 for 48 hr . Cells were analyzed by in-cell Western for BrdU incorporation . ( b ) P7 rat GCPs were cultured for 48 hr with BrdU in the absence or presence of Shh , proNT3 , or Shh + proNT3 . Cells were fixed and immunostained for BrdU , and the number of labeled cells was counted , shown in the graph . ( c ) P7 rat GCPs were cultured without or with Shh , Shh+proNT3 , or Shh+proNT3+anti-proNT3 for 48 hr and analyzed by in-cell Western for BrdU incorporation . Data in the graphs are expressed as mean values +/- SEM from at least 3 independent experiments . Asterisks indicate significance by ANOVA with Tukey’s posthoc analysis , with the indicated p value below each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01010 . 7554/eLife . 16654 . 011Figure 5—source data 1 . Mean values for each experiment and statistical analysis for all graphs in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01110 . 7554/eLife . 16654 . 012Figure 5—figure supplement 1 . ProNT3 does not induce apoptosis of cerebellar neurons . Cerebellar neurons were cultured from P7 rat and treated overnight with proNT3 . Cells were labeled with TUNEL and Dapi , and expressed as the ratio of TUNEL-positive cells to total cell number . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01210 . 7554/eLife . 16654 . 013Figure 5—figure supplement 2 . Proneurotrophins induced death of hippocampal neurons . Hippocampal neurons were cultured from E18 rat embryos and treated overnight with 4 ng/ml of the indicated proneurotrophin , eliciting loss of approximately 30% of the cells , consistent with previous results ( Friedman , 2000 ) . After removal of the medium , cultured cells were lysed and intact nuclei were counted using a hemacytometer . Nuclei of dead cells either disintegrate , or if in the process of dying , appear pyknotic and irregularly shaped . In contrast , nuclei of healthy cells are phase bright and have clearly defined limiting membranes . Cell counts were performed in triplicate wells from 3 independent experiments . Data are expressed as the mean percentage of total surviving cells +/- SEM . Asterisks indicate values different from control at p<0 . 05 by ANOVA with Tukey’s posthoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 013 To confirm that the inhibition of Shh-induced BrdU incorporation by proNT3 was mediated through p75NTR , GCPs were cultured from WT or Ngfr-/- mice . In the cells lacking p75NTR , proNT-3 was unable to reduce the level of BrdU incorporation induced by Shh , demonstrating that proNT-3 required p75NTR to promote cell cycle exit of GCPs ( Figure 6a ) . In addition , we assessed which member of the sortilin family could function as a co-receptor for proNT3 in regulating GCP proliferation . Only SorCS2 was detected in the EGL ( Figure 6b ) . To assess whether SorCS2 was the functional co-receptor for proNT3 , cultured GCPs treated with Shh+proNT3 were exposed to anti-SorCS2 , which blocked the ability of proNT3 to decrease BrdU incorporation ( Figure 6c ) . These data indicate that SorCS2 , as well as p75NTR , was required for the anti-proliferative actions of proNT3 . 10 . 7554/eLife . 16654 . 014Figure 6 . ProNT3 requires p75NTR and SorCS2 to block Shh-induced GCP proliferation . ( a ) GCPs from wild type or Ngfr-/- mice were cultured with Shh , proNT3 or Shh+proNT3 for 48 hr in the presence of BrdU and analyzed by in-cell Western analysis of anti-BrdU . ( b ) Immunostaining for SorCS2 demonstrates expression of this co-receptor in the EGL , shown for P7 , P10 and P14 . Size bar indicates 10 µm . ( c ) GCPs from P7 rat were cultured with Shh , proNT3 and anti-SorCS2 . Anti-SorCS2 reversed the effects of proNT3 on Shh-induced BrdU incorporation , but had no effect by itself or with either proNT3 or Shh alone . Data in the graphs are expressed as mean values +/- SEM from at least 3 independent experiments . Asterisks indicate significantly different from control by ANOVA with Tukey’s posthoc analysis , with p=0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01410 . 7554/eLife . 16654 . 015Figure 6—source data 1 . Mean values for each experiment and statistical analysis for graphs in 6A and 6c . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 015 ProNT3 was the only factor of all the mature and proneurotrophins that prevented Shh-induced proliferation of GCPs , therefore we investigated whether proNT3 was expressed in the developing cerebellum . Immunostaining with an antibody to the pro domain revealed expression of proNT3 in Purkinje cells ( Figure 7a ) , appropriately localized to affect GCPs in the internal EGL where they stop proliferating prior to migrating to the IGL . 10 . 7554/eLife . 16654 . 016Figure 7 . Expression and secretion of proNT3 in cerebellum . ( a ) Immunostaining of P7 and P14 rat cerebellum with an antibody to the pro domain of proNT3 shows the presence of proNT3 in Purkinje cells , labeled with calbindin . Note the abundant proNT3 labeling in the dendrites , especially apparent at P14 . Nuclei labeled with Draq5 are shown in gray in the merged image . ( b ) Cultures of P7 rat cerebellum were treated with or without 25 mM KCl to depolarize the cells , and the media was analyzed by immunoprecipitation for NT3 followed by Western blot for proNT3 , demonstrating that proNT3 can be secreted from cerebellar cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01610 . 7554/eLife . 16654 . 017Figure 7—figure supplement 1 . Validation of the proNT3 antibody . Total brain lysates from Ntf3+/+ , Ntf3+/- , and Ntf3-/- mice were immunoprecipitated with goat anti-proNT3 or goat control IgG , and probed with anti-proNT3 on a Western blot . ProNT3 produced from HEK cells was run for comparison . The proNT3 band is seen only in Ntf3+/+ and Ntf3+/- lysates IP’d with anti-proNT3 . ( b ) Sections from P0 WT but not Ntf3-/- mice show immunostaining for proNT3 in cingulate cortex , as seen for NT3 mRNA by in situ hybridization ( Friedman et al . , 1991 ) and NT3-LacZ reporter expression ( Vigers et al . , 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 017 Since proNT3 is the precursor to mature NT3 , which is highly expressed in Purkinje cells ( Zhou and Rush , 1994; Friedman et al . , 1998 ) , it was critical to ascertain whether uncleaved proNT3 could be secreted in the cerebellum . Cultures were prepared from P7 cerebellum and media was collected from cells depolarized with 25 mM KCl or undepolarized , and analyzed by immunoprecipitation followed by Western blot . ProNT3 was detected in the media with or with KCl treatment , indicating that cerebellar neurons can secrete proNT3 , and the release was not dependent on activity ( Figure 7b ) . To validate the specificity of the anti-proNT3 antibody , immunoprecipitation and immunostaining were performed on brain tissue from newborn Ntf3-/- mice and compared to tissue from Ntf3+/+ and Ntf3+/- mice ( Figure 7 —figure supplement 1 ) . Shh signaling requires HDAC1 activity to maintain GCPs in a proliferating and undifferentiated state ( Canettieri et al . , 2010 ) . As GCP proliferation decreased starting at P7 in WT mice , the level of HDAC1 was reduced in the EGL . However , the GCPs in the Ngfr-/- mice continued proliferating , maintaining high levels of HDAC1 , even at P14 ( Figure 8a ) . Not only were there more GCPs expressing HDAC1 , consistent with more proliferating cells in the EGL of Ngfr-/- mice , the intensity of HDAC1 expression was also higher in the knockout mice ( Figure 8b , c ) . To determine whether stimulation of p75NTR could regulate levels of HDAC1 , cultured cerebellar granule cells were treated with Shh with or without proNT3 . Shh induced HDAC1 , as expected , however treatment with proNT3 eliminated the Shh-mediated induction of HDAC1 within one hour of treatment ( Figure 8d , e ) , indicating that activation of p75NTR was able to block a critical component of Shh signaling . HDAC1-mediated deacetylation of Gli2 is necessary for this transcription factor to translocate to the nucleus and induce Gli1 mRNA ( Canettieri et al . , 2010 ) , which leads to induction of additional genes needed to promote proliferation . Since proNT3 decreased levels of HDAC1 , we investigated whether proNT3 also prevented induction of Gli1 mRNA by Shh . Cultured cerebellar neurons treated with Shh with or without proNT3 for 24 hr were analyzed by qPCR for Gli1 mRNA , and showed that proNT3 reduced Gli1 mRNA induction by Shh ( Figure 8f ) . 10 . 7554/eLife . 16654 . 018Figure 8 . HDAC1 expression and regulation by Shh and proNT3 . ( a ) Immunostaining for HDAC1 in WT and Ngfr-/- mice at P14 . Size bar is 20 µm . ( b ) High magnification images of HDAC1 staining in the EGL in WT and Ngfr-/- mice at P14 . Size bar is 10 µm . ( c ) Quantification of fluorescence intensity ( mean gray value ) of HDAC1 staining lobe 6b of the EGL at P14 . indicates significance at p=0 . 0024 by student’s t-test , 6 brains of each genotype were analyzed . ( d ) Western blot of cultured GCPs from P7 rat cerebellum treated as indicated and probed for HDAC1 . ( e ) Quantification of Western blots from 3 independent experiments showing that treatment with Shh increased HDAC1 expression , which was reduced by proNT3 within 1 hr . ProNT3 alone had no effect on HDAC1 expression . * indicates significantly different from control at p=0 . 0034 . ( f ) Regulation of Gli1 mRNA by Shh and Shh+proNT3 . ** indicates significantly different from control , * indicates significantly different from Shh alone , p=0 . 004 by ANOVA with Tukey’s posthoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 01810 . 7554/eLife . 16654 . 019Figure 8—source data 1 . Mean values for each experiment and statistical analysis for all graphs in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 019 p75NTR is expressed in many brain regions during development , and even in the cerebellum this receptor is expressed in Purkinje cells as well as the GCPs in the EGL ( Figure 1 ) . To determine the specific role of p75NTR in the EGL , we generated mice that lack p75NTR expression in the EGL by mating floxed Ngfr mice ( Ngfr fl/fl ) ( Bogenmann et al . , 2011 ) with Atoh1 ( Math1 ) -CRE mice ( Jackson labs ) . These mice retained p75NTR expression in Purkinje cells and the meninges but lacked p75NTR in the EGL ( Figure 9a ) . Analysis of Ki67 expression in these animals demonstrated that the absence of p75NTR specifically from the GCP population was sufficient to increase the number of proliferating GCPs at P7 , P10 , and P14 , indicating a delay in cell cycle withdrawal in the EGL , similar to what we observed in the global Ngfr-/- mice ( Figure 9b ) . In addition to Ki67 expression , we also analyzed EdU incorporation and similarly observed higher level of incorporation at P7 , P10 , and P14 in the Ngfrfl/fl:Atoh1-Cre mice compared to floxed mice without Cre recombinase , which were identical to WT ( Figure 9c ) . 10 . 7554/eLife . 16654 . 020Figure 9 . Specific deletion of p75NTR from the EGL elicits increased GCP proliferation . ( a ) Ngfrfl/fl mice crossed with the Atoh1-Cre show lack of p75NTR specifically in the EGL while retaining p75NTR expression in Purkinje cells and meninges . Size bar indicates 20 µm . ( b ) Both Ngfr-/- and Ngfrfl/fl-Atoh1-Cre mice show increased Ki67 labeling in the EGL at P7 , P10 and P14 . Total nuclei labeled with Draq5 are shown in gray . Labeled cells were counted across 150 µm in the EGL of lobe 6b and are graphed relative to the total number of cells labeled with Draq5 . ( c ) EdU labeling showing increased incorporation at P7 , P10 , and P14 in both Ngfr-/- and Ngfrfl/fl-Atoh1-Cre mice . Total nuclei labeled with Draq5 are shown in gray . Labeled cells were counted across 150 µm in the EGL of lobe 6b and are graphed relative to the total number of cells labeled with Draq5 . At least three mice per genotype were analyzed at each age . Size bars in B and C indicate 10 µm . Data in all graphs are expressed as mean +/- SEM of at least three independent experiments . Asterisks indicate significantly different from WT by ANOVA with Tukey’s posthoc analysis with the p values below each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 02010 . 7554/eLife . 16654 . 021Figure 9—source data 1 . Mean number of labeled cells of 3 slides for each animal , and statistical analysis for all graphs in Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 021 Since we observed continued proliferation of GCPs during the development of the cerebellum when p75NTR was removed either globally or specifically from the EGL , with an increase in cerebellar size ( Figure 3g ) , we sought to determine whether there was any persistent consequence for cerebellar size and function in adult animals . The increased cerebellar size in both the global and EGL-specific ( Ngfrfl/fl:Atoh1-Cre ) p75NTR-/-mice persisted into adulthood ( Figure 10a ) . To further elucidate a possible functional consequence of these developmental changes , we analyzed the motor/balance coordination in these animals using the rotarod test . Animals were trained 5 times at slow speed ( 10 rpm ) for 60 s with a 60 s recovery break between each trial . Animals were tested the next day with the same time interval for test/recovery but with increasing speeds of 5 rpm per trial until 40 rpm was reached . As shown in Figure 10b , the global Ngfr-/- mice performed poorly even at low speed . The latency to fall was significantly faster compared to the WT mice . However , the Ngfrfl/fl:Atoh1-Cre mice that specifically lacked p75NTR in the developing EGL also showed significant motor deficits compared to WT mice beginning at 20 rpm , indicating that the absence of this receptor specifically from the EGL during development caused a persistent deficit in motor function in the adult . 10 . 7554/eLife . 16654 . 022Figure 10 . The absence of p75NTR during EGL development has persistent effects in the adult . ( a ) Cerebellar size of both Ngfr-/- and Ngfrfl/fl-Atoh1-Cre was increased in the adult compared to WT mice . * indicates significance at p=0 . 0003 . ( b ) Motor balance on the rotarod was impaired in both Ngfr-/- and Ngfr fl/fl-Atoh1-Cre . At least eleven mice per genotype were tested . Asterisk indicates significantly different from WT , # indicates Ngfr -/- mice performed significantly worse than Ngfrfl/fl-Atoh1-Cre , p=0 . 0001 by ANOVA with Tukey’s posthoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 02210 . 7554/eLife . 16654 . 023Figure 10—source data 1 . Mean values for each experiment and statistical analysis for all graphs in Figure 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 16654 . 023 Sonic hedgehog ( Shh ) is a highly efficacious mitogen for GCPs and is vital for their clonal expansion within the EGL ( Dahmane et al . , 1999; Wechsler-Reya and Scott , 1999 ) . Shh continues to be expressed during cerebellar maturation , which raises the question of how some GCPs in the EGL stop responding to Shh and begin to migrate to the IGL while others keep proliferating . Previous studies have shown that P5-6 is the time of maximal GCP proliferation in mice ( Roussel and Hatten , 2011 ) , and our results show that this coincides with the peak of p75NTR expression in the EGL . At this stage , almost all cells in the EGL were positive for both p75NTR and proliferation markers such as Ki67 and incorporation of EdU . After P7 , two sublayers became apparent in the EGL . In the external zone , GCPs continued to proliferate and express p75NTR , while cells in the internal layer of the EGL no longer expressed p75NTR or proliferation markers as they began to differentiate and migrate tangentially in the medial-lateral plane ( Legué et al . , 2015 ) and radially towards the IGL ( Wang and Zoghbi , 2001 ) . In WT mice the number of proliferating GCPs in the EGL began to decrease at P7 , and continued to decrease until few proliferating cells remained in the EGL at P14 . In contrast , in both the global Ngfr-/- mice and the mice specifically lacking p75NTR in the EGL ( Ngfrfl/fl:Atoh1-Cre mice ) , the number of proliferating , Ki67-positive cells remained elevated at P7 , P10 , and P14 . However , the granule cells eventually ceased proliferating , and no tumors formed in these mice . These results suggest that there is a delay in cell cycle exit of GCPs lacking p75NTR , sufficient to cause an enlarged cerebellum that persisted into adulthood , and defects in motor coordination in these animals . Since no tumors formed in these mice and the GCPs in the Ngfr-/- mice eventually stopped proliferating , it is likely that other anti-mitotic factors caused the cells to eventually leave the cell cycle . Several factors have been identified that promote cell cycle exit , including BMP2 and 4 ( Rios et al . , 2004 ) , Wnt3 ( Anne et al . , 2013 ) , and PACAP ( Nicot et al . , 2002 ) , each acting by a distinct mechanism to block Shh activation of the Gli transcription factors . However , despite the presence of these other factors , the specific absence of p75NTR caused the GCPs to continue proliferating beyond the usual developmental period , indicating that this receptor is important for the timing of cell cycle withdrawal . A question that remains to be addressed is whether the different anti-proliferative factors have redundant effects to ensure normal development of the cerebellum , or if each factor is necessary for maturation of different subpopulations of granule cells , all of them necessary for proper function . We observed that the expression of p75NTR persisted the longest in lobes VIb , VII , and VIII , consistent with a previous study ( Carter et al . , 2003 ) and in the same regions where we detected proliferation persisting the longest in the EGL , however the delay in cell cycle exit of GCPs in both the global Ngfr-/- and the Ngfrfl/fl:Atoh-Cre mice was evident throughout the cerebellum . The p75NTR has previously been shown to regulate cell cycle of Schwann cells by recruiting an adapter protein , SC1 , which transcriptionally represses cyclin E ( Chittka et al . , 2004 ) . We have previously shown that p75NTR can also repress cyclin E in proliferating astrocytes ( Cragnolini et al . , 2012 ) , suggesting that cyclin E may be a target of p75NTR anti-proliferative signaling . Interestingly , we detected a clear difference in cyclin E1 expression in the cerebellum of Ngfr-/- mice compared to WT , which was particularly evident at P10 and P14 , the ages showing the greatest differences in GCP proliferation . Shh is expressed by Purkinje cells and regulates the proliferation of the GCPs ( Dahmane et al . , 1999 ) . Shh signals by regulating the Gli transcription factors , specifically activating Gli2 , which then targets Gli1 for transcriptional induction ( Blaess et al . , 2006; Hui and Angers , 2011 ) . Gli1 and Gli2 are acetylated proteins , and their transcriptional activation requires deacetylation by HDAC1 ( Canettieri et al . , 2010 ) , which is highly expressed in proliferating GCPs ( Yoo et al . , 2013 ) . HDAC1 is also required for deacetylation of histones , a crucial developmental regulatory event leading to condensation of chromatin and transcriptional repression , maintaining the cells in an immature , undifferentiated state ( Dovey et al . , 2010 ) . Previous studies have shown that HDAC activity is necessary for Shh-induced proliferation of cultured GCPs ( Lee et al . , 2013 ) . Coincident with the continued proliferation of GCPs in the Ngfr-/- mice , we observed that HDAC1 expression remained elevated in the EGL compared to the WT mice , where HDAC1 decreased as the cells ceased proliferating . We demonstrated that Shh induced HDAC1 in cultured GCPs and that this induction was reversed by proNT3 . ProNT3 also attenuated the induction of Gli1 mRNA by Shh , suggesting a mechanism by which p75NTR signaling may interfere with the Shh pathway by inhibiting HDAC1-mediated deacetylation of Gli2 to block Shh signaling , thereby promoting withdrawal from the cell cycle . Overexpression of Shh has previously been shown to result in an enlarged cerebellum due to overproduction of granule cells ( Corrales et al . , 2004 ) . We have similarly observed an enlarged cerebellum in mice lacking p75NTR in the EGL , suggesting that removing this block of Shh signaling has a similar consequence as overexpression of Shh , resulting in the production of excess granule cells and an enlarged cerebellum . In this study we demonstrated for the first time that proNT3 is a negative regulator of GCP proliferation , acting via p75NTR . NT3 expression has previously been detected in the developing cerebellum , however the mRNA was localized by in situ hybridization to granule cells ( Lindholm et al . , 1993 ) and the protein has been localized to Purkinje cells ( Friedman et al . , 1998; Zhou and Rush , 1994 ) . In this study we demonstrate that proNT3 protein is also present in Purkinje cells , and can be secreted as the uncleaved proneurotrophin . Mature neurotrophins had no effect on GCP proliferation . However , since all the proneurotrophins can activate p75NTR we expected that any proneurotrophin would reduce Shh-induced proliferation of GCPs . Surprisingly , proNGF and proBDNF had no effect on GCP proliferation , although they effectively promoted death of hippocampal neurons . Thus , proNT-3 proved to be the only neurotrophin to promote cell cycle exit in this neuronal population , suggesting that there is specificity among the different p75NTR ligands for their ability to influence distinct cellular functions . The basis for this specificity is unknown , however p75NTR has been shown to bind many different ligands to trigger distinct responses . Even the same ligand binding to p75NTR on the same cells may trigger different responses depending on the state of the cell . For example , NGF treatment of cultured hippocampal neurons initially promotes neurite outgrowth via p75NTR ( Brann et al . , 1999 ) but later promotes p75NTR-dependent neuronal death ( Friedman , 2000 ) . Additionally , how the pro domains of proneurotrophins influence p75NTR activity is not understood . Recent studies showed that the pro domain of proBDNF induced LTD in hippocampal slices via a p75NTR-dependent mechanism , but the proNGF pro domain failed to do so ( Mizui et al . , 2015 ) . Additionally , the BDNF pro domain induced growth cone retraction , which required p75NTR even though the pro domain does not directly bind p75NTR , but rather binds to the co-receptor of the sortilin family , in this case SorCS2 ( Anastasia et al . , 2013 ) . Interestingly , the pro domain of proBDNF has a naturally occurring Val66Met polymorphism . Although both the Val66 and Met66 pro domains bound the SorCS2 receptor , only the Val66 pro domain induced growth cone collapse , the Met66 pro domain did not ( Anastasia et al . , 2013 ) , indicating that slight differences in the pro domain can lead to differences in functional outcome . Our current study also identified SorCS2 as the co-receptor for proNT3 actions on the GCP population , suggesting that the distinct pro domains of proneurotrophins differentially bind to the SorCS2 co-receptor and influence how p75NTR interacts with specific signaling pathways to regulate cellular responses , which may account for the specificity of proNT3 in regulating GCP cell cycle exit . A previous study had suggested that p75NTR may mediate some of the effects of BDNF on cerebellar patterning and foliation . However , using the global Ngfr-/- mice this observation may have been largely due to effects in Purkinje cells , which also express p75NTR , and knockout of p75NTR led to a worsening of Purkinje cell morphology seen with reduced BDNF levels ( Carter et al . , 2003 ) . The expression of p75NTR in Purkinje cells is not likely to play a role in cell cycle regulation since these cells are already postmitotic , an additional indication that the function of p75NTR depends on the cell context . That study also showed that p75NTR did not appear to play a role in apoptosis in the developing cerebellum , consistent with our findings . p75NTR is highly expressed throughout the brain during development . Even within the cerebellum this receptor is found in Purkinje cells as well as in GCPs . Moreover , p75NTR is developmentally expressed in many neuronal populations involved in motor function , such as spinal motor neurons ( Ernfors et al . , 1989 ) , striatum , and cortex ( Yan and Johnson , 1988 ) . Not surprisingly , therefore , the global Ngfr-/- mice performed poorly on the rotarod test . To ascertain whether the changes in the granule cell population resulting from the specific lack of p75NTR in the EGL had any consequence for motor function , floxed Ngfr mice were mated with the Atoh1-Cre mice to eliminate p75NTR from the EGL while retaining the receptor in Purkinje cells and other neuronal populations . These mice also demonstrated a significant delay in GCP cell cycle exit , a larger cerebellum , and a deficit in motor/balance function on the rotarod test . Although the deficit was not as severe as that seen with the global Ngfr knockout mice , these data indicate that the EGL-specific lack of p75NTR during development was sufficient to cause persistent loss of function into adulthood . Thus , the delayed withdrawal from the cell cycle resulting in expanded proliferation of GCPs are likely to have altered the ratio of granule cells to other neuronal populations , impacting the development of appropriate circuitry for motor function . In summary , we demonstrate a novel function for p75NTR in regulating the timing of cell cycle withdrawal in granule neuron progenitors in the developing cerebellum . ProNT3 specifically antagonized Shh-induced proliferation of GCPs , decreasing the level of HDAC1 and induction of Gli1 mRNA , indicating a potential mechanism for interfering with Shh signaling and facilitating exit of these progenitors from the cell cycle prior to migrating and differentiating . Since precise regulation of these events is critical for normal development , the continued proliferation of GCPs in the absence of p75NTR led to increased cerebellar size that persisted into adulthood , with deficits in motor behavior . All animal studies were conducted using the National Institutes of Health guidelines for the ethical treatment of animals with approval of the Rutgers Animal Care and Facilities Committee . Cerebella were removed under sterile conditions from P7 pups after euthanizing with CO2 . Meninges and small blood vessels were removed under a dissecting microscope . Tissue was minced and dissociated using the papain dissociation kit ( Worthington LK003150 ) . Dissociated neurons were plated onto 24 well plates ( 1 × 105 cells in 300 µl of serum free media ) , 48 well plates ( 1 × 105 cells in 100 µl of serum free media ) or 6 well plates ( 1 × 106 cells per well in 1 ml of serum free media ) coated with poly-D-lysine ( 0 . 1 mg/ml ) . Serum free medium consisted of 1:1 MEM and F12 , with glucose ( 6 mg/ml ) , insulin ( 2 . 5 mg/ml ) , putrescine ( 60 µM ) , progesterone ( 20 nM ) , transferrin ( 100 µg/ml ) , selenium ( 30 nM ) , penicillin ( 0 . 5 U/ml ) and streptomycin ( 0 . 5 µg/ml ) . To assay for proNT3 secretion , media was collected from cultures , filtered through 0 . 22 µm syringe filter , and immunoprecipitated with 2 µg/ml of anti-NT-3 ( R&D AF267 , RRID:AB_2154250 ) at 4°C , and probed on a Western blot with anti-proNT-3 ( R&D AF3056 , RRID:AB_2154250 , 1:500 ) . Cells were cultured as described above and treated with 2–4 ng/ml of proNT-3 for 48 hr . Cells were then fixed with 4% paraformaldehyde ( PFA ) /PBS for 15 min and permeabilized with 0 . 5% Triton × 100 for 20 min . TUNEL assay was performed following the manufacturers specification ( Promega G3250 ) . 10 pictures per coverslip were taken with a Nikon Eclipse TE200 microscope . Number of DAPI and TUNEL-positive cells were quantified using Image J Version 1 . 49 . Cerebellar neurons were cultured as above for 24 hr in the absence or presence of Shh or Shh+proNT3 . RNA was isolated using Trizol ( Ambion ) , and analyzed by quantitative real-time PCR using a Roche 480 II Light Cycler and the Roche Light Cycler sybr green kit . Primers for Gli1 were: forward - GCTGTCGGAAGTCCTATTCAC , reverse - GCCTTCCTGCTCACACATATAA , and for GAPDH: forward - CACCGACCTTCACCATCTTGT , reverse – TTCTTGTGCAGTGCCAGCC . Tissue or cells were washed with ice-cooled PBS and homogenized using 1% NP40 , 1% triton , 10% glycerol in TBS buffer ( 50 mM Tris , pH 7 . 6 , 150 mM NaCl ) with protease inhibitor cocktail ( Roche Products , 11 836 153 001 ) . Proteins were quantified using the Bradford assay ( Bio-Rad 500–0006 ) and equal amounts of protein were run on SDS gels and transferred to nitrocellulose membrane . To ensure equal protein levels , blots were stained with Ponceau prior to incubation with antibody . The blots were then rinsed and blocked in 5% nonfat dried skim milk in TBS-T for 2 hr at RT . Blots were incubated with primary antibodies diluted 1:1000 in 1% BSA in TBS buffer overnight at 4°C . The blots were washed with TBS-T 3 × 15 min each and incubated with Licor secondary antibody for 1 hr at RT . All secondary antibodies were diluted 1:10 , 000 . Membranes were washed 3 × 15 min each in TBS-T . The membranes were analyzed using Licor Odyssey infrared imaging system ( LICOR Bioscience , Lincoln , NE ) . To confirm equal protein levels , blots were reprobed with actin . All analyses were performed at least three times in independent experiments . For detection of secreted proNT3 , an equal volume of media was immunoprecipitated with anti-NT3 and probed for proNT3 or NT3 . For validation of the proNT3 antibody ( R&D AF3056 , RRID:AB_2154250 ) , equal protein from brain lysates of WT , heterozygous , or Ntf3-/- mouse tissue from P0 pups was immunoprecipitated with anti-proNT3 and probed with anti-proNT3 ( Figure 7—figure supplement 1 ) . Animals were deeply anesthetized with ketamine/xylazine and perfused with 4% PFA/PBS . Brains were removed and postfixed in 4% PFA/PBS overnight at 4°C , then cryopreserved with 30% sucrose . Sections ( 20 µm ) were cut using a Leica cryostat , and mounted onto charged slides . Sections were permeabilized with 0 . 5% triton in PBS for 10 min and blocked with 1% BSA and 5% donkey serum in PBS for 1 hr at room temperature . Primary and secondary antibodies were prepared in 1% BSA . Sections were incubated with primary antibodies overnight at 4°C in a humidified chamber . Antibodies used were: Ki67 ( Abcam 15580 , RRID:AB_443209 , 1/500 ) , anti-p75 ( R&D AF367 , RRID:AB_2152638 , 1/500 ) , anti-p75 ( Millipore MAB365 , RRID:AB_2152788 , 1/1000 ) , anti-proNT-3 ( R&D AF3056 , RRID:AB_2154250 , 1/200 ) , anti-NT-3 ( R&D AF-267-NA , RRID:AB_354434 , 1/200 ) , anti-BrdU ( Millipore 05–633 , RRID:AB_309861 , 1/50 ) . All secondary antibodies were diluted 1:1000 , and incubated for 1 hr at RT . Nuclei were labeled with 1 µg/ml Dapi/PBS for 1 min at RT or 1 mM Draq5 ( BioStatus DR-50200 ) for 30 min and mounted using Prolong Gold ( Life Technologies P36931 ) . Controls for immunostaining included incubation with secondary antibodies in the absence of primary antibodies . The proNT3 antibody was tested for specificity on sections from Ntf3-/- mice compared to WT from P0 pups , since the Ntf3 -/- mice die after birth . For in vivo analysis of proliferation , 40 µl of 10 µM solution of Edu/PBS was injected i . p . 2 hr before perfusing the animals for immunocytochemical analysis . After cryostat sectioning , EdU was developed following the manufacturers protocol ( Molecular Probes C10337 ) . Ten pictures per coverslip were obtained on a Zeiss LSM 510meta microscope with LSM acquisition software . Draq5 and EdU positive cells were counted using Image J v1 . 49 . Mice with homozygous floxed alleles of p75NTR ( Ngfrfl/fl ) were mated with Atoh1-Cre mice ( Jackson Labs ) , and pups were obtained postnatally at different developmental ages . The genotype of Ngfrfl/fl; Atoh1-Cre animals were confirmed by PCR , and the absence of p75NTR in the EGL was confirmed by immunostaining . Balance and coordination were tested using the rotarod assay ( Korpi et al . , 1999 ) . Mice were placed on the rod for training in 5 sessions of 60 s each at low speed ( 10 rpm ) , and were tested the following day for 5 sessions at 60 s each with increasing speed in each subsequent session until 40 rpm . After 5 min of rest , the testing protocol was repeated . The number of seconds the mouse stayed on the rod before falling was scored . At least twelve mice per genotype were tested . HEK 293 cells were grown to 90% confluence in a 10 cm plate and split into 5 plates . The cells were allowed to attach for 4 hr in the incubator . Media was replaced with 5 ml of Optimem ( Gibco 31985–070 ) and cells were returned to the incubator for 20 min . 0 . 5–1 µg of total DNA and 20 µl of lipofectamine ( Thermo Fischer Scientific 18324–0204–020 ) were mixed in 1 ml Optimem media and left at RT for 20 min . Media was replaced with 4 ml fresh Optimem and the DNA/Lipofectamine was added to the cells overnight . The next day media was replaced with DMEM + penicillin ( 5 U/ml ) and streptomycin ( 5 µg/ml ) . After 48 hr the media was collected , aliquoted and stored at −80°C . The concentration of proNT-3 was estimated by Western blot analysis run with a known concentration of NT-3 . Experimental groups were compared using either student’s t-test or ANOVA followed by Tukey’s posthoc analysis , as appropriate , p<0 . 05 was considered significant . The specific statistical analysis is indicated in each figure legend .	Many proteins control how cells are organised in your body . For example , a group of proteins called the neurotrophins help to control the life and death of brain cells . After a brain injury , neurotrophins can cause nerve cells to die by working together with another protein called p75NTR . The p75NTR protein is also found in another type of brain cell called granule progenitor cells . These cells exist before birth and divide to make new cells that form a part of the brain called the cerebellum , which controls how your body moves . Granule progenitor cells do not typically die , so it is not known what p75NTR does in these cells . Zanin et al . investigated the role that p75NTR plays in the formation of the cerebellum in mice . The experiments showed that p75NTR controls when granule progenitor cells stop producing new brain cells . The progenitor cells of mutant mice that lack this protein produced too many new cells , which resulted in these mice having larger cerebellums and being less able to control their movements . Further experiments showed that p75NTR interacts with a neurotrophin called proNT3 in the cerebellum , which is able to stop granule progenitor cells dividing even in the presence of other proteins that encourage cells to divide . An important challenge for the future is to work out why the mice lacking p75NTR are less able to control their movements .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2016	Proneurotrophin-3 promotes cell cycle withdrawal of developing cerebellar granule cell progenitors via the p75 neurotrophin receptor
How transcription factor dimerization impacts DNA-binding specificity is poorly understood . Guided by protein dimerization properties , we examined DNA binding specificities of 270 human bZIP pairs . DNA interactomes of 80 heterodimers and 22 homodimers revealed that 72% of heterodimer motifs correspond to conjoined half-sites preferred by partnering monomers . Remarkably , the remaining motifs are composed of variably-spaced half-sites ( 12% ) or ‘emergent’ sites ( 16% ) that cannot be readily inferred from half-site preferences of partnering monomers . These binding sites were biochemically validated by EMSA-FRET analysis and validated in vivo by ChIP-seq data from human cell lines . Focusing on ATF3 , we observed distinct cognate site preferences conferred by different bZIP partners , and demonstrated that genome-wide binding of ATF3 is best explained by considering many dimers in which it participates . Importantly , our compendium of bZIP-DNA interactomes predicted bZIP binding to 156 disease associated SNPs , of which only 20 were previously annotated with known bZIP motifs . Multiple sequence-specific transcription factors ( TFs ) converge at enhancers and promoters to control the expression of genes ( Ptashne and Gann , 2002 ) . Such TF assemblages permit integration of multiple cellular signals to regulate targeted gene networks ( Ciofani et al . , 2012; Kittler et al . , 2013; Xie et al . , 2013 ) . The combinatorial use of a limited number of TFs provides the means to finely control the complex cellular transcriptome . The ability of a given TF to interact with different partners expands the DNA targeting repertoire . It also enhance specificity by focusing combinations of factors to a more specific set of regulatory sites across the genome . Different partners also alter the regulatory potential of a given TF , at times completely altering its regulatory output such that the factor switches from an activator to a repressor of transcription depending on the binding partners with which it associates ( Ptashne and Gann , 2002 ) . The bZIP class of human TFs is well suited to play a role in signal integration and combinatorial transcriptional control ( Lamb and McKnight , 1991; Miller , 2009; Tsukada et al . , 2011 ) . bZIPs bind DNA as either homo- or heterodimers; the discovery in 1988 that JUN and FOS could bind to DNA as a heterodimer immediately suggested the potential for combinatorial regulation by this family ( Franza et al . , 1988; Lamb and McKnight , 1991 ) . Interestingly , the human bZIP network displays greater ability to form heterodimers compared to simpler eukaryotes , suggesting that more complex combinatorial regulation may contribute to organismal complexity ( Reinke et al . , 2013 ) . bZIP proteins also interact with other classes of TFs to stabilize higher order oligomers at enhancers ( Jain et al . , 1992; Murphy et al . , 2013; Thanos and Maniatis , 1995 ) . Their role in nucleating such multi-factor complexes is supported by the observation that certain bZIP dimers such as AP-1 ( FOS•JUN ) and CEBPA can function as ‘pioneer’ factors that bind inaccessible chromatin and enable the assembly of other TFs at regulatory sites ( Biddie et al . , 2011; Collins et al . , 2014 ) . Notably , the choice of dimerizing partner not only impacts DNA recognition properties but can also influence regulatory function of a given bZIP . For example , ATF3 homodimer acts as a repressor , whereas ATF3•JUN activates transcription ( Hsu et al . , 1992 ) . As a class , bZIPs regulate diverse biological phenomena ranging from response to stress at the cellular level , organ development at the tissue level and viral defense , circadian patterning , memory formation , and ageing at an organismal level ( Costa et al . , 2003; Herdegen and Waetzig , 2001; Jung and Kwak , 2010; Male et al . , 2012 ) . Given their central role in various processes , mutations in bZIP proteins are implicated in the etiology of diseases ranging from cancer and diabetes to neuronal malfunction , developmental defects and behavioral dysfunction ( Lopez-Bergami et al . , 2010; Tsukada et al . , 2011 ) . Fifty-three bZIPs encoded by the human genome can be grouped into 21 families , and as homodimers they are known to bind at least six distinct classes of DNA motifs , including sites labeled as TRE , CRE , CRE-L , CAAT , PAR , and MARE ( Figure 1 ) ( Deppmann et al . , 2006; Jolma et al . , 2013 ) . In 1991 , Hai and Curran ( 1991 ) showed that some heterodimers have DNA-binding specificities that are distinct from those of each partnering bZIP . For example , JUN•ATF2 heterodimer binds to a cognate site in the ENK2 promoter that is not bound by either JUN•JUN or ATF2•ATF2 homodimers . However , the past 20 years have provided only a handful of additional examples of how bZIP heterodimerization influences DNA-binding specificity ( Cohen et al . , 2015; Jolma et al . , 2015; Vinson et al . , 1993; Yamamoto et al . , 2006 ) . Central questions about bZIP transcription factors remain unanswered: What is the influence of protein dimerization on DNA binding ? Does DNA stabilize dimer formation ? Which protein dimers can bind DNA ? Which sequences do they bind ? And , how do bZIP-binding sites contribute to cellular function and the etiology of various diseases ? 10 . 7554/eLife . 19272 . 003Figure 1 . Overview of human bZIP homodimer and heterodimer DNA-binding specificities . ( A ) Summary of SELEX-seq results categorized by protein-protein interaction ( PPI ) affinity ( Reinke et al . , 2013 ) . Specificity profiles were classified as resulting in a motif arising from DNA binding by either a homodimer ( brown ) or a heterodimer ( dark brown ) , or not resulting in a motif ( white ) . Some profiles could not be unambiguously assigned to a homo vs . heterodimer ( light brown ) . ( B ) Pairwise comparisons of the DNA-binding preferences of 102 bZIP dimers ( 22 homodimers and 80 heterodimers ) using the z-scores for 1222 unique 10 bp sequences corresponding to the 50 top ranked sequences for each dimer . Throughout the paper , the biotinylated bZIP is listed first when describing a heterodimer . ( C ) Representative motifs bound by bZIP homodimers and heterodimers reported in this study . Heterodimer motifs were grouped as Conjoined , Variable spacer , and Emergent . The color code defined here for half sites ( colored arrows above motifs ) is used throughout the figures . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 003 10 . 7554/eLife . 19272 . 004Figure 1—source data 1 . Data for Figure 1C . Pairwise comparison ( Pearson's correlation ) of the DNA-binding preferences of 102 bZIP dimers using the CSI intensity for 1222 10 bp sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 004 10 . 7554/eLife . 19272 . 005Figure 1—figure supplement 1 . Cognate site identification by SELEX-sequencing . ( A ) In CSI by SELEX-seq , a DNA library with a randomized 20 bp region is incubated with a bZIP pair in which one bZIP partner ( light grey ) was biotinylated and the other partner ( light grey ) was labeled with fluorescein ( blue star ) . bZIP partners were mixed in 3:1 molar ratios with the biotinylated partner at the lower concentration . Affinity purification using the less abundant biotinylated partner enriched for heterotypic dimers . ( B ) Reproducibility of CSI by SELEX-Seq . Scatter plots of CSI intensities ( z-scores ) for all 10-mers for replicate samples and ( C ) reciprocal samples . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 00510 . 7554/eLife . 19272 . 006Figure 1—figure supplement 2 . ATF3 CSI Intensity ( z-score ) correlates with equilibrium association constant . ( A ) DNA sequence of oligonucleotides used for determining binding constants . ( B ) Correlation between CSI intensity ( z-score ) and association constant ( Ka ) for ATF3 . Binding constants were measured by EMSA . Error bars are ± S . D . of at least duplicate measurements . ( C ) Representative autoradiographs of EMSA experiments from which binding constants were calculated using non-linear regression . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 00610 . 7554/eLife . 19272 . 007Figure 1—figure supplement 3 . Pairwise comparison of bZIP homodimers reported in this study and bZIP dimers reported by Jolma et al . ( Jolma et al . , 2013 ) . ( A ) Hierarchical clustering was performed using the CSI intensities ( z-scores ) of 871 unique 10 bp sequences corresponding to the 50 top ranked sequences identified from each dimer . Corresponding bZIP pairs are highlighted in matching color . ( B ) Scatter plots comparing CSI intensity ( z-score ) for all 10-mers of bZIP dimers from this study with bZIP dimers previously reported by Jolma et al . ( 2013 ) . ( top left ) BATF3 vs . BATF3; ( top right ) CEBPG vs . CEBPG; ( bottom left ) ATF4 vs . ATF4; ( bottom right ) ATF4 vs . ATF4•CEBPG . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 00710 . 7554/eLife . 19272 . 008Figure 1—figure supplement 4 . bZIP heterodimer specificity . Pearson’s correlations ( r ) of all 10-mers between replicate experiments of bZIP dimers ( top ) , and correlations between a bZIP heterodimer and the bZIP homodimer that was used to pull-down the heterodimer . The average ( ± standard deviation ) Pearson’s correlation ( r ) for eight replicate samples was 0 . 8 ± 0 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 00810 . 7554/eLife . 19272 . 009Figure 1—figure supplement 5 . DNA sequence preferences for FOS•CEBPE , FOS•CEBPG , FOSL1•CEBPE , and FOSL1•CEBPG . Left , PPI affinity for the corresponding heterodimer is shown . Middle , MEME motifs are represented as DNA logos . Right , 2-dimensional scatter plots comparing the CSI intensities for all 10-mers . CRE/CAAT ( TGACGTAA ) sites are colored red , and TRE/CAAT ( TGAGCAA ) sites are colored orange . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 009 Resolving these issues requires systematic examination of the DNA-binding specificities of bZIP homo- and heterodimers . Fifty-three human bZIP proteins can potentially form as many as 1431 distinct dimers . Quantitative experiments using fluorescence resonance energy transfer ( FRET ) in solution indicated that ~ 30% of all possible bZIP dimers form in the absence of DNA . Most bZIPs can form dimers with different partners , potentially greatly expanding the repertoire of cognate sites that might be targeted by different heterodimers ( Reinke et al . , 2013 ) . We used this protein-protein interaction ( PPI ) dataset to prioritize 270 bZIP dimers for bZIP-DNA interactome studies and to apply FRET-based methods to distinguish DNA-bound heterodimers from homodimers . Insights that emerged from our compendium of bZIP-DNA interactomes include: ( i ) identification of new bZIP cognate sites , ( ii ) evidence for three classes of heterodimer-binding sites ( conjoined half-sites , variably-spaced half-sites , and unpredicted emergent cognate sites ) , ( iii ) ability of individual bZIP heterodimers to target a range of binding sites , ( iv ) evidence for varying heterodimer selectivity between distinct sequences currently classified as a single consensus motif , ( v ) improved ability to account for in vivo genome-wide occupancy of heterodimers , and ( vi ) identification of bZIP cognate sites at 156 SNPs linked to human diseases and quantitative traits . DNA sequence preferences of bZIP heterodimers reported here serve as a valuable resource for many purposes including , but not limited to , evaluating potential bZIP dimer binding at genomic binding sites , providing hypotheses about mechanisms underlying the etiology of disease-linked SNPs , and predicting binding specificities of heterodimeric bZIPs from other species . To elucidate DNA sequence-recognition properties of TFs that form obligate dimers , we examined 270 pairs of purified human bZIP proteins . These pairs were composed from 36 bZIP proteins representing 21 bZIP families and encompassing the diversity of all 53 bZIPs encoded in the human genome ( Supplementary file 1A ) . Given the 666 potential dimeric pairs that can be formed with 36 bZIPs , we used biophysically measured PPIs to prioritize the dimers that were examined ( Reinke et al . , 2013 ) . We selected 126 pairs ( 97 hetero- and 29 homodimer ) that form stable dimers with PPI dissociation constants ( Kd ) less than 1 µM at 21°C in the absence of DNA . In addition , we tested 144 ( 137 hetero- and seven homo- ) dimer combinations that do not stably associate in solution in the absence of DNA ( Kd >1 µM at 21°C ) . For most TFs , including the bZIP class , the DNA specificity of the isolated DNA binding domain ( DBD ) is typically indistinguishable from the full-length factor ( Jolma et al . , 2013 ) . Therefore , we focused our efforts on the bZIP domain , which comprises the basic region that binds DNA and the leucine zipper dimerization module that forms a coiled-coil . Recombinant proteins , overexpressed in bacteria , were purified to homogeneity . Two versions of each protein were made , one conjugated to biotin at the carboxyl terminal and the other without . This set of highly purified DNA binding proteins enabled examination of the innate DNA-binding sequence specificity of 36 representative human bZIPs . Individual bZIP partners were mixed in 3:1 molar ratios with the biotinylated partner at the lower concentration; affinity purification of the protein-DNA complexes using the less abundant biotinylated partner enriched for heterodimers . To additionally favor examination of the heterodimer , whenever possible , the interaction partner with the weaker homodimer was biotinylated and used for isolating protein-DNA complexes . Each protein dimer is denoted by a dot between each monomer – for example , JUN•ATF3 . The DNA binding sites are indicated either by a specific sequence or their classical designations , such as CRE or CAAT , or by half-sites connected by a hyphen , such as CRE-CAAT . DNA-binding specificity of the pairs was queried using systematic evolution of ligands by exponential enrichment coupled to deep sequencing ( SELEX-seq ) ( Figure 1—figure supplement 1 ) ( Jolma et al . , 2010; Slattery et al . , 2011; Tietjen et al . , 2011; Zhao et al . , 2009; Zykovich et al . , 2009 ) . In our cognate site identification ( CSI ) effort using SELEX-seq , a DNA library spanning the entire sequence space of a 20-mer ( 1012 different sequence permutations ) was independently incubated with each of the 270 different bZIP pairs ( 234 hetero- and 36 homodimers ) , and protein-bound DNA sequences were enriched , amplified , and re-probed for an additional two cycles to further enrich cognate sites over non-cognate sites that comprise the majority of the starting library . The starting DNA library as well as selectively enriched sequences were barcoded and sequenced using massively parallel DNA sequencing methods . The CSI intensity , corresponding to the z-score for the enrichment of a sequence , was computed for each 10-mer as described in the methods . Repeated experiments demonstrated an average Pearson’s correlation r = 0 . 8 ± 0 . 1 between CSI intensities from replicates ( Figure 1—figure supplement 1 ) . The CSI intensity ( z-score ) correlates with the binding affinity for a particular sequence ( Figure 1—figure supplement 2 ) ( Carlson et al . , 2010; Puckett et al . , 2007; Tietjen et al . , 2011 ) . In three cases , reciprocal biotinylation of each partner was performed to ensure that the choice of partner did not skew the results ( Pearson’s correlation for comparing experiments was r = 0 . 89–0 . 98; Figure 1—figure supplement 1C ) . Among the 270 pairs tested were 12 homodimers that had been previously examined by other groups ( Jolma et al . , 2013 ) . Overall , we found excellent agreement between the cognate sites identified using highly purified bZIP modules in our study versus full-length proteins in unpurified cell lysates in other studies , with only a few inconsistent examples that can be seen in Figure 1—figure supplement 3A ( e . g . MAFB , NFE2 , ATF4 ) . Interestingly , we found that the previously reported DNA specificity of ATF4 has a higher correlation with the specificity of ATF4•CEBPG heterodimer identified in this report than with the specificity of the ATF4 homodimer ( Figure 1—figure supplement 3B ) , suggesting that CEBPG possibly formed a complex with ATF4 in the cell lysates used in the prior study ( Jolma et al . , 2013 ) . This observation highlights the advantage of using highly purified proteins over cell lysates and validates our focus on the bZIP domain to capture the innate specificity of this class of transcription factors that bind DNA as obligate dimers . Overall , 30 out of the 36 bZIP proteins tested in this study enriched specific DNA sequences as part of at least one dimer . bZIPs that are not known to bind DNA as homodimers did not yield cognate sites in our studies ( e . g . JUNB and FOS ) ( Deng and Karin , 1993; Hai and Curran , 1991 ) . 73 of 126 ( 58% ) bZIP pairs that dimerize in the absence of DNA yielded specific cognate sites ( Figure 1A ) . Surprisingly , 29 of the 144 ( 20% ) bZIP pairs that do not stably associate in the absence of DNA ( Kd >1 µM at 21°C ) yielded evidence of sequence-specific binding to DNA , indicating that PPIs were stabilized by binding to specific DNA sites ( Figure 1A; Supplementary file 1C ) . This finding has important implications , given that the majority of the potential bZIP PPI space consists of protein pairs that do not associate strongly in the absence of DNA ( Reinke et al . , 2013 ) . For 184 bZIP-DNA interactomes that showed evidence for an enriched motif , we computationally parsed and retained datasets that could be attributed with high confidence to 80 heterodimers and 22 homodimers ( Materials and methods ) . We assigned a specificity profile to a heterodimer when it bound sequences that were significantly different ( t-test p<0 . 05 ) from the sequences preferred by the homodimer of the biotinylated bZIP ( e . g . ATF4 vs . ATF4•CEBPA , r = 0 . 1; Figure 1—figure supplement 4 ) or when the biotinylated bZIP did not bind specific DNA sequences as a homodimer ( e . g . FOS and JUNB ) . Of the 22 homodimers , comprehensive DNA-binding specificity is reported for the first time for human ATF2 , ATF3 , ATF6 , ATF6B , CEBPA , CREB1 , FOSL1 , JUN , MAFB , and NFE2L1 . Hierarchical clustering of the 102 bZIP-DNA interactomes readily identified six previously known classes of bZIP-binding sites ( TRE , CAAT , PAR , MARE , CRE , and CRE-L ) ( Figure 1B–C ) . Notably , several bZIP homodimers ( ATF6 , ATF6B , CREB3L1 , and JUN ) enriched more than one motif ( Supplementary file 2 ) . Examining the cognate sites bound by heterodimers highlighted the ability of some heterodimers to bind homodimer motifs as well as a range of other heterodimer-specific motifs . Such binding to multiple motifs is reminiscent of previous studies that reported bZIP dimers binding to different sites with different affinities ( Badis et al . , 2009; Kim and Struhl , 1995; König and Richmond , 1993 ) . Interestingly , several heterodimers that bind classic bZIP homodimer motifs such as TRE , CRE-L , or CRE displayed clear differences in their preference for a subset of sequences categorized under a single consensus motif ( e . g . compare motifs 9 and 10 in Figure 1 with the CRE-L site ) . This was also true for different homodimers ( e . g . compare CRE-L binding profiles for ATF6 , CREB3 , CREB3L1 , and XBP1 in Supplementary file 2 ) . Thus , the binding data reported here reveal a sequence sub-structure to classic consensus motifs . Moreover , the sub-structure highlights differences in DNA-binding specificity between closely related dimers . Three classes of bZIP heterodimer motifs were identified and are illustrated in Figure 1C: ‘Conjoined’ sites for which half-sites preferred by each contributing monomer are juxtaposed ( such as the CRE-CAAT site represented by motif 1 , or the MARE-CRE site of motif 7 ) , ‘Variably-spaced’ sites for which half-sites overlap ( as is the case in motifs 2 and 4 ) , and ‘Emergent’ sites for which binding preferences could not have been readily inferred based on the half-site preferences of each partner ( motifs 3 , 5 , 8 , 9 , and 10 ) . In other words , an emergent site arises as a consequence of heterodimer formation and is not simply comprised of the conjoined or variably-spaced half-sites preferred by each monomer . An elegant study of Hox-Exd heterodimers identified ‘latent’ sites that were preferred by different Hox factors when they bound DNA in conjunction with Exd ( Slattery et al . , 2011 ) . Preferences for different sequences at the interface of half-sites or sequences flanking the half sites were observed for different classes of Hox-Exd heterodimers . In our studies , we observed a change in half-site preference of certain bZIPs when they bound DNA as heterodimers . In some instances , homodimers bound with low affinity to sites that emerged as high-affinity sites in the context of a heterodimer , whereas in other cases entirely new site preferences emerged . We classified such newly acquired binding preferences as emergent sites because they are not readily inferred from the binding preferences of homodimers . While a large fraction of heterodimers bind conjoined sites , it was surprising to find that closely related heterodimers such as FOS•CEBPG and FOS•CEBPE preferred different arrangements of half-sites , with the former heterodimer preferring the 8 bp conjoined CRE-CAAT site ( motif 1 5’TGACGCAA3’ ) and the latter preferring the 7 bp variably-spaced TRE-CAAT site ( motif 4 5’TGAGCAA3’ ) . Figure 1—figure supplement 5 highlights the unexpectedly poor correlation between the binding preferences of these two heterodimers and between the binding preferences of FOSL1•CEBPG and FOSL1•CEBPE . Similarly , other heterodimers bound both conjoined and variably spaced motifs ( see JUNB•ATF3 and MAFB•ATF5 in Supplementary file 2 ) ; however , the preference for one arrangement over the other was not amenable to predictions based on the binding preferences of each contributing partner of the heterodimer . Emergent sites pose a particular challenge for current models of DNA binding site predictions that are based on protein homology ( Weirauch et al . , 2014 ) . Emergent cognate sites for heterodimers can be subdivided into two categories: ( i ) ‘gain-of-specificity’ motifs that display a change in half-site preferences for a bZIP or ( ii ) motifs that display a ‘loss-of-specificity’ for one half-site . An example of the first category includes a switch in the half-site preferences of BATF family members , from a CRE half-site ( 5′TGAC3′ ) that is preferred in homodimers to a CRE-L ( 5′CCAC3′ ) half-site that is preferred by many BATF-containing heterodimers ( compare motifs 3 and 8–10 , and see examples in Supplementary file 2 such as BATF2•ATF3 , BATF2•JUN , BATF3•ATF3 , BATF3•ATF4 ) . An example of the second category is DDIT3•CEBPG binding to 5′ATTGCA3′ ( motif 5 ) ( Ubeda et al . , 1996 ) , with heterodimers displaying no apparent requirement for one half-site . Overall , for the 80 bZIP heterodimers with binding motifs reported here , 72% of the motifs can be classified as conjoined , 16% as emergent , and 12% as variably-spaced . Nine out of the 80 heterodimers ( 11% ) enriched two motifs ( Supplementary file 2 ) . For example , BATF•CEBPG enriched both CRE-CAAT and CRE-L-CAAT motifs . To examine the full specificity spectrum of individual bZIP dimers , we displayed DNA binding data as specificity and energy landscapes ( SELs ) ( Carlson et al . , 2010; Tietjen et al . , 2011 ) . In a SEL , all possible sequences of a given length are arranged within concentric circles based on their homology to a seed motif . The seed motif is often derived from position weight matrices ( PWMs ) of the most enriched sequences ( Figure 2A ) . The innermost circle contains all sequences that have an exact sequence match to the seed motif ( 0-mismatch ) . As each enriched sequence placed in this ring is an exact match to the seed motif , the source of varying CSI intensity ( z-score ) is the contribution of the sequences flanking the seed motif . The 1-mismatch ring contains all sequences that differ from the seed motif at any one position , or a Hamming distance of one . The subsequent rings , going outwards , display sequences with increasing number of mismatches to the seed motif . The height and color of each point represents the CSI intensity for the corresponding sequence . As noted above , CSI intensity correlates with binding affinity where measured ( Figure 1—figure supplement 2 ) ( Carlson et al . , 2010; Hauschild et al . , 2009; Puckett et al . , 2007; Tietjen et al . , 2011; Warren et al . , 2006 ) . Although there are far more low-affinity sequences than enriched sequences ( as depicted by the illustrative histogram in Figure 2A ) , the moderate-to-low affinity sites ( low CSI intensity ) are often overlooked by motif searching algorithms . Such sequences readily emerge in SEL display of the entire binding data ( Carlson et al . , 2010; Tietjen et al . , 2011 ) . In Figure 2A , we illustrate how SELs are built and we note that a SEL can be constructed using any sequence as a seed motif . The choice of a different seed motif simply alters the placement of the sequences on the landscape without changing the underlying binding preferences of a protein for a given sequence . 10 . 7554/eLife . 19272 . 010Figure 2 . Specificity and energy landscapes ( SELs ) and motifs for bZIP heterodimers . ( A ) SEL displays CSI intensities for all sequence permutations of a given binding site size ( k-mers ) . Sequences are organized with respect to any selected seed motif; however , a k-mer representing PWM-derived motif is typically used . CSI intensities correlate with equilibrium binding affinities . As an example , the arrangement of 6-mer sequences for a simplified 4-mer seed motif is shown . The innermost circle displays the intensities for all sequences that have an exact match to the seed motif ( 0-mismatch ring ) . In this ring , sequences are arranged in a clockwise manner with sequences that include residues 5′ of the seed motif at the start , sequences with residues that flank both 5′ and 3′ ends in the middle , and 3′ flanking sequences at the end ( context ) . The subsequent 1-mismatch ring contains the sequences that differ at one position from the seed . The sequences are organized clockwise starting with mismatches at the first position and ending with mismatches at the last position of the motif . Within each sector , the mismatches at a given position ( indicated by x ) are organized in alphabetical order ( A , C , G , and T ) . The 2-mismatch ring contains all permutations with two positional differences with the seed , similarly ordered . ( B ) Left , SEL for JUN•ATF3 heterodimer using CRE ( 5′TGACGTCA3′ ) as the seed motif . By displaying the 10 bp sequence space , preferred sequences become apparent . Peaks corresponding to emergent and variably-spaced sites are identified by arrows . Right , SEL displaying 12 bp sequences for ATF4•CEBPG heterodimer using CRE-CAAT ( 5′ATGACGCAAT3′ ) as the seed motif . ( C ) Heatmap of the relative CSI intensities of 102 bZIP dimers ( columns ) for the 10 sites highlighted in Figure 2B as well as constituent half-sites of the six classic bZIP motifs ( rows ) . Displayed is the maximum CSI intensity of all the 10 mers matching the site . bZIP dimers are listed in the same order as in Figure 1B . ATF3 , ATF4 , CEBPG , and JUN homodimers are marked by asterisks . While bZIPs do not bind as monomers to half-sites , the occurrence of bZIP half-sites within motifs is displayed in the second set of rows to enable comparison between the half-site preferences versus the CSI intensity for motifs that display these half-sites in different combinations or in different contexts . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 010 10 . 7554/eLife . 19272 . 011Figure 2—source data 1 . Data for Figure 2C . Relative CSI intensity for 102 bZIP dimers for different DNA-binding sites and half-sites . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 011 SEL plots of 102 bZIP homo- and heterodimers reveal that the impact of flanking sequence context and the range of different cognate sites bound by most bZIPs is far richer than might be inferred from motifs represented as PWMs ( Supplementary file 2 ) . In Figure 2B , the SELs of JUN•ATF3 and ATF4•CEBPG illustrate the broader insights that emerged from examining specificity profiles of these two heterodimers . JUN•ATF3 binds a CRE site composed of conjoined half-sites for JUN and ATF3 . Visualizing the entire JUN•ATF3-DNA interactome via a SEL shows that the binding of JUN•ATF3 heterodimer to CRE is significantly influenced by the sequence context that flanks the motif ( see affinity variations in the 0-mismatch ring ) . Additionally , the 3-mismatch ring of the SEL identifies several high-intensity peaks corresponding to additional cognate sites . As indicated , one of these is a variably spaced site , and another is an emergent site 5′TGACGCAT3′ . Thus , the SEL highlights that this single heterodimer binds multiple classes of cognate sites . On the other hand , the SEL for ATF4•CEBPG shows that the seed motif 5′ATGCGCAAT3′ bound by this heterodimer is relatively insensitive to context effects ( 0-mismatch ring ) . The 1-mismatch ring indicates that both half-sites are not equally tolerant of mismatches , with mismatches in the 5′TGA3′ core of the CRE site dramatically reducing binding , whereas the 5′CAA3′ site is tolerant of deviations at the first position of the half-site but sensitive to deviations in the 5′AA3′ positions . Similar insights can be obtained from the SELs for each of the102 bZIP dimers that are reported in Supplementary file 2 . Our compendium of SEL plots greatly extend previous reports that bZIP dimers bind a range of sequences with different degrees of affinity ( Badis et al . , 2009; Kim and Struhl , 1995; König and Richmond , 1993 ) . To examine whether the set of sequences that are pointed out in SELs of JUN•ATF3 and ATF4•CEBPG from Figure 2B are bound by homodimers or any bZIP in our compendium , we displayed the relative preferences of each dimer for this set of binding sites in a heatmap ( Figure 2C ) . Each column displays the relative preference of each of the 102 bZIP dimers for different sequences , including half-sites of all six classical homodimer motifs . An examination of row 3 , which displays preferences of all 102 dimers for the emergent site 5’TGACGCAT3’ , indicates that this site is highly preferred by JUN•ATF3 and to some extent by JUN•CEBPG but not by homodimers formed by ATF3 , CEBPG , or JUN ( denoted by asterisks ) . While not as exclusive as the JUN•ATF3 emergent site , the conjoined CRE-CAAT site is primarily targeted by heterodimers formed by the CEBP family of bZIPs . The heatmap does indicate that this heterodimer-preferred site permits low-affinity binding by the CEBPG homodimer . Interestingly , data in row 8 reveals that substituting 5’CAAT3’ with 5’TAAT3’ in the CRE-CAAT site perturbs binding by CEBP heterodimers in a non-uniform manner , unmasking hidden differential sequence preferences of related heterodimers that are opaque to current models that use protein homology to predict cognate site preferences . The C-to-T substitution also expands the repertoire of bZIPs that bind this mutated site . DBP , HLF , and NFIL3 as homo- and hetero- dimers display an unmistakable affinity for this modified CRE-CAAT site that recreates the PAR half-site that is a target of this set of bZIPs . On the other hand , a different substitution at the same position ( 5’GAAT3’ ) dramatically reduces the binding of all bZIPs to this version of the CRE-CAAT site ( row 9 ) . Furthermore , the importance of the sequences flanking a binding site ( rows 4 and 5 ) or the contribution of each half-site to the binding of bZIPs ( rows 8–10 ) is also made evident by the heatmap . In essence , SEL plots alongside comparative heatmaps of affinities of proteins for a range of cognate sites bring a new appreciation for diversity of DNA sequences that can be targeted by a given factor . To validate the ability of heterodimers to bind cognate sites identified by CSI analysis , we used an electrophoretic mobility shift assay ( EMSA ) in which a FRET signal distinguished homo- vs . heterodimers in protein-DNA complexes ( Figure 3A; Materials and methods ) ( Reinke et al . , 2013 ) . We first used EMSA-FRET to assay bZIP dimers formed by mixing fluorescein and TAMRA labeled versions of 16 proteins drawn from different bZIP families . For 15 homodimers for which we could detect binding to DNA , the mixed-dye homodimer could be easily distinguished from both of the single-dye homodimers , as shown for CEBPG and ATF3 homodimers binding to CAAT and CRE sites , respectively ( Figure 3A ) . This assay was then used to demonstrate that the ATF3•CEBPG heterodimer bound the conjoined CRE-CAAT site better than either parental homodimer ( Figure 3A ) . Furthermore , swapping fluorophores did not alter the binding properties of the resulting heterodimer ( last panel of Figure 3A ) . DNA fluorescence coincides with the protein FRET signals , confirming that protein-DNA complexes were being observed in the EMSA gels ( Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 19272 . 012Figure 3 . Influence of bZIP protein dimerization on DNA binding . ( A ) EMSA-FRET assay used to quantify bZIP heterodimers and homodimers binding to DNA . Fluorescein and TAMRA are depicted as blue and green stars , respectively . In the EMSA gel , homodimers give rise to pseudo-colored blue ( fluorescein ) or green ( TAMRA ) signals , whereas heterodimers give a FRET signal that is pseudo-colored red . ( B ) EMSA-FRET results for bZIP dimers binding to selected heterodimer-specific emergent sites ( brown ) and conjoined half-sites ( blue ) . Bar graphs show the percent of the indicated DNA oligomer bound by each dimer . The PPI strength of each dimer is indicated with gray-scale circles sized according to the Kd for a given protein-protein interaction . Homodimers are marked with an asterisk ( * ) . ( C ) EMSA-FRET results for bZIP dimers tested for binding to DNA sites composed of conjoined half-sites . Left , dimers tested against two different sites composed of conjoined half-sites . Right , dimers tested against a single site . Data are displayed as in B . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 01210 . 7554/eLife . 19272 . 013Figure 3—figure supplement 1 . Influence of bZIP protein dimerization on DNA binding . ( A ) Detecting heterodimer DNA complexes using an EMSA-FRET assay . Top , Fluorescein signal in blue , TAMRA signal in green , and FRET signal in red . Bottom , TYE 665 labeled DNA site . ( B ) Three examples to explain the notation used in part C summarize data for DNA binding by homodimers and heterodimers composed of ( left ) ATF3 and DBP , ( middle ) ATF3 and CEBPA and ( right ) BATF3 and JUN . Within each example , rows indicate different bZIP dimers . The top row describes the homodimer formed by the first-mentioned bZIP , the bottom row is for the other homodimer , and the middle row contains data for the heterodimer . Within each example , each column represents binding to a different DNA site composed of two half-sites . DNA-binding affinity is indicated using a green-scale heatmap with key indicating % binding at far right . The color of the cell border indicates strength of the protein-protein interaction as measured previously by FRET , indicated by yellow-scale heatmap at right . ATF3•DBP example: top row is ATF3 binding to CRE-PAR , middle row is ATF3 • DBP heterodimer binding to CRE-PAR , bottom is DBP homodimer binding to CRE-PAR . ATF3•CEBPA example: Top row is ATF3 homodimer , middle row is ATF3 • CEBPA heterodimer , and bottom row is CEBPA homodimer . Binding is to CRE-CAAT in left column and TRE-CAAT in right column . BATF3•JUN example: Top row is BATF3 homodimer , middle row is BATF3 • JUN heterodimer and bottom row is JUN homodimer . Binding is to CRE-CRE in left column and CRE-CREA in right column . ( C ) Complete set of EMSA-FRET data . Examples in B are included in this grid and other cells can be interpreted analogously . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 01310 . 7554/eLife . 19272 . 014Figure 3—figure supplement 2 . Heterospecific binding of DNA . Top , DNA sequences composed of optimal half sites . Bottom , comparison of an optimal DNA site to a heterodimer-specific non-optimal DNA site . DNA sequences for EMSA-FRET experiments are reported in Supplementary file 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 014 We used this EMSA-FRET assay to quantify the DNA binding of 83 bZIP homodimers and heterodimers comprised of 16 proteins . Each heterodimer was systematically examined with DNA sites that were constructed by conjoining the preferred half-site ( s ) for each bZIP . Figure 3B shows EMSA-FRET data for six heterodimers and corresponding homodimers binding to heterodimer-specific sites ( three conjoined sites and three emergent sites ) . For these sites , the CSI intensity for the heterodimers is higher than the scores for either of the two contributing homodimers . The EMSA-FRET data demonstrate clearly that neither the JUN nor the ATF3 homodimers associate with the emergent site identified for JUN•ATF3 ( Figure 3B and Figure 3—figure supplement 2 ) . Similarly , emergent sites identified for ATF4•CEBPA and ATF4•JUN , and several conjoined sites such as TRE-CAAT for ATF3•CEBPA , CRE-L-CAAT for BATF3•CEBPA , and CRE-CRE-L for BATF3•JUN , were validated by EMSA-FRET as bona fide heterodimer-specific cognate sites that show weaker binding , or no binding , by the contributing homodimers ( Figure 3B ) . EMSA-FRET data also validate the ability of BATF3 to bind emergent CRE-L half-sites as a heterodimer ( in addition to the CRE site preferred by the homodimer ) . The complete EMSA-FRET data are presented in a more compact format in Figure 3—figure supplement 1 . A striking result of our CSI analysis is that conjoined half-sites form a substantive fraction ( ~70% ) of the cognate sites bound by heterodimers . To determine how frequently DNA half-sites derived from homodimer-binding data , when presented as conjoined sites , would bind the corresponding heterodimers , we tested DNA binding by EMSA-FRET for stably interacting bZIP heterodimers ( PPI: Kd <1 µM at 21°C ) , and the corresponding homodimers ( Figure 3C ) . Consistent with CSI analysis , 52 out of 56 bZIP pairs that form stable heterodimers bound the DNA site made by conjoining the half-site preferred by each monomer . Specific binding to conjoined sites was also detected for 6 out of 27 ( 22% ) pairs that do not stably associate in the absence of DNA ( PPI: Kd >1 µM at 21°C ) . This fraction is similar to the 20% of bZIP pairs ( 29 out of 144 ) that showed sequence-specific DNA binding in SELEX-seq experiments despite their apparent inability to dimerize in the absence of DNA . Activating Transcription Factor 3 ( ATF3 ) is a member of the CREB/ATF family . Initially identified as a suppressor of inflammation and the adaptive immune response in resting cells , ATF3 is now associated with numerous diseases including a variety of aggressive and widely occurring cancers ( Tanaka et al . , 2011; Thompson et al . , 2009; Yin et al . , 2008 ) . ATF3 is able to interact with a large variety of TFs to function as a regulatory hub of cellular adaptive response ( Gilchrist et al . , 2006; Hai et al . , 1999 , 2010 ) . As a homodimer , ATF binds to CRE sites and represses a wide array of genes ( Hai et al . , 1999 , 2010 ) . However , as a heterodimer with JUN or JUND , ATF3 activates transcription of targeted genes ( Chu et al . , 1994; Filén et al . , 2010; Hsu et al . , 1992 ) . To test the hypothesis that heterodimerization with other bZIPs might alter DNA-binding specificity and possibly genomic targets , we analyzed SELEX-seq for 20 different ATF3 heterodimers spanning the full range of PPI affinities . DNA-binding specificities could be assigned with high confidence to nine heterodimers that displayed a range of DNA sequence preferences , including affinity for the CRE site preferred by the homodimer ( Figures 4A and B ) . Importantly , distinct DNA-binding preferences among ATF3•CEBP and ATF3•BATF heterodimers and their corresponding homodimers were detected . The motifs enriched by the ATF3 homo- and heterodimers can be described in five broad categories: CRE , TRE , CRE-CAAT , CRE-L , and the emergent 5′TGACGCAT3′ site ( Figure 4B ) . Scatter plots illustrate instances where the CSI intensities of ATF3 heterodimers differ markedly from those of the parent homodimers ( Figure 4C and D ) . For example , as evident from high CSI intensities , CRE-CAAT sites ( red ) are preferably bound by ATF3•CEBPG as compared to ATF3 or CEBPG ( Figure 4C , top panel ) . Similarly , scores for TRE ( green ) and 5′TGACGCA3′ ( black ) are higher for JUN•ATF3 than for JUN or ATF3 ( Figure 4C , middle panel ) . BATF3•ATF3 ( Figure 4C , bottom panel ) and BATF2•ATF3 ( Figure 4D , top panel ) enrich CRE-L sites ( blue ) , further supporting that CRE-L is an emergent site for BATF family heterodimers ( also with JUN in Figure 3C ) . Figure 4D further highlights the differences between CRE and TRE binding by ATF3 in its homo versus heterodimer state . An important and recurring observation is that several ATF3 heterodimers ( BATF3•ATF3 , JUN•ATF3 , and JUNB•ATF3 ) can associate with the CRE site that is bound by the ATF3 homodimer ( Figure 4 ) . 10 . 7554/eLife . 19272 . 015Figure 4 . ATF3 heterodimers bind a range of distinct cognate sites . ( A ) Hierarchical clustering of pairwise comparisons of DNA-binding specificity ( 10-mers ) for ATF3 homodimer and 9 ATF3-containing heterodimers . ( B ) DNA logos showing the MEME motifs derived from the top 1000 12-mer sequences for ATF3 homodimer and ATF3-containing heterodimers . Grey-scale circles next to dimer names indicate PPI strength using the scale from Figure 3 . ( C ) 3-dimensional and ( D ) 2-dimensional scatter plots comparing the DNA-binding specificities of bZIP homodimers vs . ATF3-containing heterodimers . Scatter plots of quantile-normalized CSI intensities ( z-scores ) of ATF3 dimers for 80 , 000 10-mers are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 015 To determine the extent to which cognate sites identified by SELEX-seq can explain genome-wide occupancy in cells , we examined ChIP-seq data for ATF3 in four different human cell lines . H1 human embryonic stem cells , HEPG2 liver-derived hepatocellular carcinoma cells , and K562 erythroblastoma cells have been examined comprehensively ( Encode , 2011 ) . The fourth cell line , GBM1 from Glioblastoma multiforme , is an aggressive brain cancer , wherein ATF3 is a tumor suppressor and its loss of function is indicative of high-grade cancer and poor prognosis ( Gargiulo et al . , 2013 ) . As a first step , we identified ATF3 ChIP-seq peaks and examined the overlap between the genomic sites occupied by ATF3 in all four lines . Only a small number of sites ( 119 ) were common between the four cell lines , although the number increased to 1602 genomic loci if only the ENCODE cells lines ( H1 , K562 , and HEPG2 ) were examined ( Figure 5A ) . This is a minor fraction of the over 10 , 000 peaks identified in K562 and about a third of the 4808 ATF3-bound sites in H1 cells . 10 . 7554/eLife . 19272 . 016Figure 5 . ATF3 binds to different genomic regions using diverse motifs . ( A ) Venn diagram of the numbers of ATF3-bound regions determined by ChIP-seq in different cell lines . ( B ) Heatmap of the False Positive Rate ( FPR ) -cutoffs at which ATF3 ChIP-seq peaks ( rows ) are detected as positive for ATF3 or ATF3-dimer binding . Peaks were scored using CSI intensities of the ATF3 homodimer or ATF3-containing heterodimers ( columns ) in H1hESCs , K562 , and HEPG2 cells , and clustered by FPR-cutoffs across all dimers . ( C ) Same as ( B ) for the glioblastoma multiforme ( GBM1 ) cell line . Highlighted clusters ( blue and green ) contain DNA motifs preferred by different ATF3 dimers and are enriched with different Gene Ontology Biological Process terms . False Discovery Rates ( q-values ) for each GO term are shown . See Supplementary file 1H . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 01610 . 7554/eLife . 19272 . 017Figure 5—figure supplement 1 . ROC curves . ( A ) Area Under the Receiver Operating Characteristic curve ( AUC-ROC ) values for the intersection of ChIP-seq peaks determined using in vitro specificity profiles of the corresponding bZIP heterodimer , as described in Materials and methods . x-axis: False-Positive Rate; y-axis: True-Positive Rate ( TPR ) . ChIP-seq peaks from specified cell lines were downloaded from the ENCODE project . ( B ) ROC curves and AUC values for ChIP-seq peaks ( all peaks ) determined using DNA-binding specificity profiles of the corresponding bZIP homodimer . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 017 ATF3 ChIP-seq peaks likely include both homodimer and heterodimer bound regions . To assess how well the in vitro discovered cognate sites explain bound sites in a cellular context , we used area under the curve-receiver operating characteristic ( AUC-ROC ) values , plotting the true-positive rate ( TPR ) versus false-positive rate ( FPR ) for peak detection ( Materials and methods ) . ATF3 homodimer sites spanning the entire spectrum of CSI intensities ( z-scores ) yielded 0 . 67–0 . 77 AUC values ( Supplementary file 1E and Figure 5—figure supplement 1 ) . Using the AUC-ROC approach , we examined the ability of CSI profiles of nine different ATF3 heterodimers as well as the ATF3 homodimer to identify ATF3 ChIP-seq peaks that might represent heterodimer-bound regions . We used published RNA-seq datasets to verify the expression of the bZIP genes used for the ATF3 heterodimer analysis ( Supplementary file 1F ) ( Encode , 2011; Gargiulo et al . , 2013 ) . Each of the 10 CSI datasets captures a large but varying fraction of the ATF3 peaks and , intriguingly , these data reveal that different ATF3 heterodimers perform better in different cell lines ( Supplementary file 1E ) . For example , JUN•ATF3 gives 0 . 85 AUC in the Glioblastoma line , whereas BATF3•ATF3 better explains the ChIP-seq peaks in H1 and HepG2 cells with AUC of 0 . 69 and 0 . 70 , respectively . While AUC-ROC curves are not robust to subtle changes , the differences we observe may reflect underlying cell line specific differences in the abundance and regulatory roles of different ATF3 heterodimers . The underlying epigenetic landscapes would further exacerbate these differences . Nevertheless , when considered together , the ATF3 homodimer combined with nine different heterodimers can account for a much larger fraction of ChIP-seq peaks than can the homodimer alone . For example , at an FPR-cutoff of 0 . 10 , in the Glioblastoma cell line , the ATF3 homodimer classified just 39% of the ATF3 ChIP-seq peaks as positive , whereas 85% of the peaks are classified positive by at least one of the ATF3-containing dimers at FPR 0 . 10 . Similar analysis for other cell lines and at different FPR cutoffs is reported in Supplementary file 1G . Given the cell-type-specific differences in genomic sites occupied by ATF3 , we scored the ATF3-bound loci for each cell line using the CSI data for 10 ATF3 dimers . Peaks were then clustered based on the FPR-cutoffs for each bound region ( Figure 5B–C and Materials and methods ) . All four cell lines show clear clusters of sites where one or more heterodimer detects a peak at a lower FPR compared to the ATF3 homodimer . Several such clusters are apparent for heterodimers with CEBP or BATF family members . A striking result that emerged from the analysis of the GBM1 cell line is that multiple ATF3-bound genomic loci were better described by ATF3-heterodimers than the homodimer . For GBM1 , we further examined two clusters of ChIP peaks for which heterodimers scored better than the ATF3 homodimer ( Figure 5C , blue and green clusters in the dendrogram ) . In the blue cluster , de novo motif discovery revealed enrichment of a CRE-CAAT motif , which is the motif with maximal CSI intensities for CEBP•ATF3 dimers . De novo motif search of ChIP-seq peaks in the green cluster identified the TRE motif , which is the top ranked motif for ATF3 heterodimers formed with JUNB , JUN , FOS , and FOSL1 , all of which are expressed in GBM1 cells ( Supplementary file 1F ) . This is in contrast to the CRE motif preferred by the ATF3 homodimer . Gene ontology functional annotations of genes linked to the CRE-CAAT ( blue ) and TRE ( green ) clusters also differ substantially ( Figure 5C and Supplementary file 1H ) . CRE-CAAT sites preferred by ATF3•CEBP heterodimers ( blue cluster ) enriched for gene ontology ( GO ) terms related to immune response and JAK-STAT signaling , whereas TRE sites ( green cluster ) enriched for GO terms associated to nutrient sensing , PDGF signaling , and cell junction regulation . This observation lends support to the notion that heterodimers drive cell-type and signal-specific gene networks . Sharpening our focus to a subset of genomic loci that are co-occupied by ATF3 and another bZIP permitted us to examine whether heterodimer cognate sites were evident at co-occupied genomic loci . In Tier 1 ENCODE cell lines such as H1 and K562 , occupancy of multiple TFs has been charted across the genome ( Dunham et al . , 2012 ) . We first examined loci co-occupied by ATF3 and CEBPB or JUN . In H1 embryonic stem cells , we identified a region on chromosome I that shows overlapping ChIP peaks for ATF3 and CEBPB ( Figure 6A , top panel ) . This locus is also resistant to DNAse I , suggesting that ATF3 and CEBPB are binding to a seemingly inaccessible part of the genome . Plotting CSI intensities for a given TF across the genome generates CSI-Genomescapes ( Figure 6A–B , bottom panels; Materials and methods ) . CSI-Genomescapes in the co-occupied region identified a high-intensity site for the ATF3•CEBPA heterodimer , whereas no high-intensity sequences were found for ATF3 or CEBPA homodimers ( Figure 6A ) . CEBPA is the closest homolog to CEBPB for which CSI data were obtained . Similar CSI-Genomescape analysis of a locus with overlapping ATF3 and JUN peaks readily identified the JUN•ATF3 emergent site ( 5′TGACGCAT3′ ) . This site is within DNase I accessible euchromatin , and CSI-Genomescapes provide scant support for either JUN or ATF3 homodimer binding to this site ( Figure 6B ) . 10 . 7554/eLife . 19272 . 018Figure 6 . bZIP heterodimer DNA sites are bound in vivo . ( A ) ChIP-seq traces for ATF3 ( blue ) and CEBPB ( orange ) and DNase I hypersensitivity ( black ) trace for in H1 human embryonic stem cells . Below , CSI-Genomescape for bound genomic regions for ATF3 and CEBPA homodimers and ATF3•CEPBA heterodimer . CEBPA and CEBPB share 76% identity over their bZIP domain . ( B ) ChIP-seq traces for ATF3 ( blue ) and JUN ( green ) and DNAse I hypersensitivity trace ( black ) in K562 cells . Below , CSI-Genomescape for bound genomic region for ATF3 and JUN homodimers , and for JUN•ATF3 heterodimer . ( C ) Venn diagram of bound regions ( ChIP-seq peaks ) for ATF3 and CEBPB in H1hESC and for ( D ) ATF3 and JUN in K562 cells . ( E ) Violin plots of CSI-seq scores for the ChIP-seq peaks derived from the intersection of ATF3 and CEBPB ChIP peaks ( 1018 overlapping peaks ) in H1 stem cells using in vitro data for ATF3 , CEBPA , CEBPB ( from Jolma et al . ) ( Jolma et al . , 2013 ) , CEBPE , CEBPG homodimers and ATF3•CEBPA , ATF3•CEBPE , and ATF3•CEBPG heterodimers . CSI intensities were quantile normalized . ( F ) Violin plots of CSI-seq scores for the ChIP-seq peaks derived from the intersection of ATF3 and JUN ChIP peaks ( left , 6539 overlapping peaks ) in K562 cells , left . Violin plots for the subset of overlapping peaks of ATF3 and JUN containing a match for the heterodimer-specific site TGACGCAT ( 39 peaks ) , right . Peaks were scored using ATF3 and JUN homodimers , and JUN•ATF3 heterodimers . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 01810 . 7554/eLife . 19272 . 019Figure 6—figure supplement 1 . CSI intensities for bound genomic regions . Violin plots of CSI intensities ( z-scores ) for ( A ) Negative regions were taken from ±5 kb from the center of each ATF3 and CEBPB overlapping ChIP peaks in H1 cells . ( B ) Negative regions were taken from ±5 kb from the center of each ATF3 and JUN overlapping ChIP peaks in K562 cells . ( C ) ATF3 ChIP-seq peaks after removing peaks that overlap with CEBPB in H1 cells . ( D ) CEBPB ChIP-seq peaks after removing peaks that overlap with ATF3 in H1 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 019 Next , we identified genomic loci that are co-occupied by ATF3 and CEBPB in H1 ( 1018 overlapping ChIP peaks ) or ATF3 and JUN in K562 cells ( 6539 overlapping ChIP peaks; Figure 6C–D ) . We then used CSI data of different homo- and heterodimers to assign CSI scores within these co-occupied regions . Violin plots clearly demonstrate that regions co-occupied by ATF3 and CEBPB have higher CSI intensities when scored with ATF3•CEBP heterodimers than when scored with ATF3 and CEBP homodimers ( Figure 6E and Figure 6—figure supplement 1 ) . In contrast , for loci co-occupied by JUN and ATF3 , violin plots indicate that cognate sites for JUN•ATF3 heterodimer perform only marginally better at explaining the genomic binding data than sites preferred by JUN or ATF3 homodimers ( Figure 6F , left panel ) . This observation is consistent with the ability of the JUN•ATF3 heterodimer to bind consensus CRE sites that are also bound by each contributing homodimer . The perceptibly higher CSI intensity when using JUN•ATF3 cognate sites might arise from heterodimer-preferred TRE sites or heterodimer-specific emergent sites . To examine this possibility , we utilized CSI-Genomescapes to score all co-occupied regions that include emergent heterodimer-specific 5′TGACGCAT3′ sites ( 39 sites ) . When this subset of genomic regions was examined with homodimer CSI data , the violin plots reveal the inability of ATF3 homodimer cognate sites to account for the ChIP signals , whereas JUN homodimers account for some of the JUN occupancy at those regions ( Figure 6F , right panel ) . In contrast , the ATF3•JUN heterodimer cognate sites showed the highest scores for the emergent site . Armed with 102 CSI profiles of bZIP dimers , we scrutinized 5076 non-coding single-nucleotide polymorphisms ( SNPs ) that are associated with diseases and quantitative traits ( Maurano et al . , 2012 ) . We reasoned that non-coding SNPs that are not assigned to known TF cognate sites might be explained with our compendium of new bZIP-DNA interactomes . As a first step , we calibrated our CSI data by examining SNPs that are known to alter binding by CREB1 and CEBPA ( Figure 7A top panel and Figure 7—figure supplement 1A ) . The minor allele of rs10993994 in the promoter of the MSMB gene has been associated with prostate cancer and it creates a cognate site that is bound by CREB1 ( Lou et al . , 2009 ) . Similarly , the minor allele of rs12740374 has been associated with myocardial infarction , aberrant plasma levels of low-density lipoprotein cholesterol ( LDL-C ) , and enhanced expression of SORT1 gene in the liver ( Musunuru et al . , 2010 ) . Biochemical studies have demonstrated that the G-to-T change generates an optimal CAAT site that is bound by CEBPA . We applied CSI-Genomescape analysis to both SNPs . In both cases , the minor allele has a higher CSI intensity than the corresponding major allele , suggesting that the minor alleles of these SNPs create CEBPA- and CREB1-binding sites ( Figure 7A and Figure 7—figure supplement 1 ) . The CSI-Genomescape for rs7631605 site is particularly interesting because it predicts disruption of the emergent site 5′TGACGCAT3′ ( Figure 7A middle panel ) . This allele is associated to Alzheimer’s disease and mild cognitive impairment ( MCI ) and elevated levels of phosphorylated Tau-181P ( Han et al . , 2010 ) . Additionally , CSI-Genomescape predicts that rs1869901 , a variant associated with schizophrenia , impacts binding of FOS•CEBPE by altering a TRE-CAAT site ( Figure 7A bottom panel ) . 10 . 7554/eLife . 19272 . 020Figure 7 . bZIP heterodimers and human diseases and traits . ( A ) CSI-Genomescape predicts increased binding by CREB1 to the alternate allele of rs10993994 and decreased binding to alternate alleles of rs7631605 and rs1869901 by JUN•ATF3 and FOS•CEBPE heterodimers , respectively . ( B ) Scatterplot of FOS•JUN predicted CSI intensities for reference and alternative alleles of 5076 autosomal SNPs linked to human diseases and quantitative traits identified in genome-wide association studies . SNPs and disease/traits classifications are from Maurano et al . ( Maurano et al . , 2012 ) . ( C ) ( left ) Number of SNPs predicted to increase or decrease bZIP binding by twofold at different stringency levels determined by noise factor F ( see Materials and methods ) . The F values at which a twofold difference in CSI score is predicted for rs12740374 ( # ) and rs10993994 ( * ) are indicated in red . ( right ) Distribution of predicted fold changes in bZIP binding for GWAS SNPs using CSI Intensities , using F = 25 . Dashed lines mark a twofold change . Red lines indicate the predicted change in binding of CREB1 and CEBPA to rs10993994 ( * ) and rs12740374 ( # ) . ( D ) Predicted fold-change in CSI score of sequences centered at SNPs linked to disease or quantitative traits . A total of 156 SNPs have a predicted increase ( red ) or decrease ( blue ) of ≥2 fold in CSI score for at least one bZIP dimer , when F = 25 ( Materials and methods and Supplementary file 1I ) . Fold-changes are relative to the reference genome hg19 . Rows ( SNPs ) are organized by class of disease/trait . Columns ( bZIP dimers ) are clustered by DNA specificity as in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 02010 . 7554/eLife . 19272 . 021Figure 7—figure supplement 1 . Genomescapes , transcription factor binding , and chromatin environment for selected SNPs . ( A ) Left , CSI Genomescape and right , UCSC genome browser screen shots of the genomic and chromatin context of SNPs rs12740374 and rs10993994 . ( B ) UCSC genome browser screen shots of the genomic and chromatin context of SNPs rs3758354 and rs17293632 . UCSC genome browser tracks for ChIP-seq peaks for selected bZIPs in ENCODE cell lines , ChIP-seq signal for histone 3 lysine 27 acetylation ( H3K27Ac marks ) , and DNAse I hypersensitive regions . DOI: http://dx . doi . org/10 . 7554/eLife . 19272 . 021 A scatterplot of CSI intensity scores for FOS•JUN ( AP-1 ) for reference ( hg19 ) or alternate alleles reveals SNPs that create or disrupt binding sites ( Figure 7B ) . The plot shows that nearly all the 5076 SNPs are near the origin and do not lead to large differences in CSI scores for the FOS•JUN heterodimer . However , a striking example of predicted increase in binding is rs3758354 , a SNP associated with schizophrenia , depression , and bipolar disorder ( Huang et al . , 2010 ) . In contrast , a decrease in FOS•JUN heterodimer binding is predicted for rs17293632 , a variant linked to Crohn’s autoimmune disorder ( Franke et al . , 2010 ) . ChIP-seq studies in several cell lines examined by the ENCODE consortium have shown binding by FOS and JUN to both loci , providing support that these sites are accessed by bZIP proteins in a cellular context ( Figure 7—figure supplement 1B ) . Extending beyond AP-1 , we used 102 bZIP CSI profiles to score both alleles of the 5076 SNPs and calculated a predicted fold-change in CSI intensity , which correlates with binding affinity ( Figure 1—figure supplement 2 ) ( Carlson et al . , 2010; Puckett et al . , 2007 ) . Similar correlations also hold true for other high-throughput platforms ( Berger et al . , 2006; Fordyce et al . , 2010; Slattery et al . , 2011 ) . We added a noise factor to our scoring function to make the fold-change predictions less sensitive to low CSI intensities ( Figure 7C; Materials and methods ) . A total of 156 SNPs yielded a greater than twofold difference in CSI intensity between the reference and alternate alleles ( Figure 7C–D ) . Displaying the predicted increase ( blue ) or decrease ( red ) in binding by 102 bZIP dimers at 156 SNPs reveals minor alleles that are targeted by unique heterodimers as well as mutations that have wide-ranging impacts on multiple bZIP dimers . For example , rs10994336 is predicted to increase CSI intensity by at least twofold for 44 out of 102 bZIP pairs reported here . We also report that 80% of the identified changes impact bZIP heterodimers . In the richly annotated RegulomeDB database that ties SNP impact to occurrence of TF-binding sites , only 20 of 156 SNPs are currently annotated with a bZIP motif ( Boyle et al . , 2012 ) . It is particularly important to note that many of the SNPs in the database are annotated with PWMs derived from bZIP homodimers , whereas our CSI intensity fold-change predictions for 22 homo- and 80 bZIP heterodimers make use of the entire bZIP-DNA interactomes ( all 10-mers ) . The clusters in Figure 7D also point to potential roles of bZIP proteins in less understood diseases and provide new hypotheses for the etiology of such diseases and traits . Transcription factors rarely function alone , different TFs are activated by different cellular stimuli , and specific combinations of TFs converge at specific genomic loci to regulate expression of genes ( Ptashne and Gann , 2002 ) . Such combinatorial control provides the means to integrate multiple signals and tune the expression of specific genes or sculpt genome-wide transcriptomes in a nuanced manner . The ability of different TFs to form hetero-oligomers via PPI and protein-DNA interaction is an essential feature of this process . While most eukaryotic TFs can bind DNA as monomers , the bZIP class of TFs only binds DNA as homo- or heterodimers . The ability of bZIPs to form heterodimers appears to increase with increasing evolutionary complexity , with human bZIPs displaying more intricate heterodimerization networks than C . elegans and D . melanogaster , which in turn , exhibit more complex dimerization networks than S . cerevisiae ( Reinke et al . , 2013 ) . Comprehensive PPI analysis has shown that 36 human bZIP monomers can form nearly 217 heterodimers , greatly expanding the repertoire of factors that can potentially bind DNA ( Reinke et al . , 2013 ) . We demonstrate that this diversity of dimers expands the DNA sequence space that can be targeted by bZIPs . Our study further reveals that nearly 20% of the non-interacting bZIP pairs examined can be induced to dimerize at cognate DNA sites , providing yet greater diversity from a modest number of contributing monomers . Given the large repertoire of human bZIP heterodimers , this family of TFs is particularly amenable to effect combinatorial control . Indeed , this potential was recognized long ago ( Bohmann et al . , 1987; Franza et al . , 1988; Lamb and McKnight , 1991 ) . An ever-increasing body of evidence now implicates bZIPs in numerous aspects of cellular and organismal function . Given their importance , a systematic study of DNA binding by bZIP heterodimers is clearly essential to understanding their functions . However , despite large surveys charting the TF-DNA interactomes ( Badis et al . , 2009 , 2008; Berger et al . , 2008; Carlson et al . , 2010; Fordyce et al . , 2010; Franco-Zorrilla et al . , 2014; Grove et al . , 2009; Jolma et al . , 2010 , 2013; Kamesh et al . , 2015; Nitta et al . , 2015; Noyes et al . , 2008; Siggers et al . , 2012; Wei et al . , 2010; Weirauch et al . , 2014 ) , bZIP dimers were under-scrutinized with only a handful of heterodimers reported thus far ( Cohen et al . , 2015; Jolma et al . , 2015; Mann et al . , 2013 ) . Thus , it was quite unclear prior to this work how dimerization between different bZIP partners would impact DNA recognition . The DNA-binding profiles for 80 heterodimers , which we report alongside equivalent data for 22 homodimers , is the largest bZIP heterodimer DNA-binding data reported to date and provides unprecedented insight into the impact of heterodimer formation . Guided by PPI data , we examined the DNA-binding specificities of 126 stable dimers and 144 bZIP dimers that display no dimerization even at 1 µM . These 270 bZIP pairs represent a wide survey of the 666 potential pairs that can be formed by 36 monomers . The bZIP-DNA interactomes and specificity landscapes that emerged revealed three classes of cognate sites and several heterodimers displayed an ability to interact with more than one class of binding site . Of the three classes , conjoined half-sites were the most abundant , with nearly 72% of all heterodimers displaying some affinity for such sites . The second class contained variably-spaced half-sites , often overlapping by a single nucleotide . The final class , comprising 16% of the sites , was the least expected ‘emergent’ class of binding sites , where new non-obvious preferences for half-sites were revealed . Emergent sites targeted by heterodimers fall into ‘loss of specificity’ or ‘gain of specificity’ categories , as defined above . EMSA-FRET analyses not only quantified the relative affinities of hetero- and homodimers for these sites but also revealed the widespread ability of heterodimers to associate with cognate sites that have typically assumed to be bound by homodimers . More closely examining the emergent site targeted by ATF3•JUN , we find its occurrence at multiple locations across the human genome and , more importantly , several of these sites are co-occupied by ATF3 and JUN in vivo . Further emphasizing a physiological role for these non-obvious binding sites , a SNP that disrupts this site is linked to neurological diseases ( Han et al . , 2010 ) . The high granularity of our CSI data also revealed that sequences flanking well-studied homodimer motifs , such as CRE , can impart sub-structure to the motif that is recognized and preferentially bound by different bZIP pairs . Access to such nuanced specificity preferences allows better annotation of genome-wide binding data for bZIPs for which specificity profiles and high-quality ChIP data exist . This is particularly relevant because it is not uncommon for ChIP or genomic DNase I footprinting experiments to identify TF-bound regions that lack matches to the consensus motifs for a given TF . Our results suggest that a fraction of such in vivo occupied regions likely contain heterodimer binding sites . Another important insight from our comparative analysis of genome-wide binding profiles across four cell types is that a given heterodimer associates with distinct set of genomic loci in each cell type . The results suggest that underlying chromatin and epigenetic landscapes in different cell types may contribute significantly to the sites that are accessed by bZIP dimers . In this context , the ability of ATF3•CEBPB to bind a cognate site within seemingly closed chromatin is consistent with the ability of bZIPs such as CEBPB and FOS•JUN to function as ‘pioneer’ factors that first associate with closed chromatin and enable binding of additional TFs to yield transcriptionally active euchromatin ( Biddie et al . , 2011; Garber et al . , 2012 ) . Whether the ability to bind just one half site is important , or whether DNA-templated dimerization of bZIPs confers any added ability to bind an otherwise inaccessible enhancer in closed chromatin , remains to be determined . Finally , the specificity and binding energy profiles of 102 bZIP dimers enables a more nuanced examination of SNPs that have been linked by genome-wide association studies to various diseases and quantitative traits . The vast majority of SNPs associated with diseases occur in non-coding regions of the genome and most are not readily annotated by the available TF-DNA interactomes perhaps in part because the focus has been on obtaining consensus motifs of monomeric or homodimeric TFs . Rather than consensus motifs , the use of the full spectrum of binding specificity may enable more accurate mapping of TF-binding sites onto SNPs that are linked to diseases and phenotypic traits . Our compendium of CSI profiles accurately predicted creation of known bZIP cognate sites by previously validated SNPs . Of the 156 SNPs predicted by our CSI profiles to impact bZIP binding , nearly 77% were mapped to bZIP heterodimers , highlighting the importance of determining protein-DNA interactomes for heterodimer TFs . Nearly 64% created bZIP binding sites and were ‘gain of function’ changes relative to the human reference genome . These results are consistent with the 10-fold greater abundance of bZIP heterodimers over homodimers and the observation that aberrant stimulation of gene networks is arguably a greater contributor to disease etiology ( Bell et al . , 2015; Lee and Young , 2013; Mansour et al . , 2014 ) . SNPs that disrupt binding also contribute to disease , an example of this form of regulatory perturbation being the loss of the emergent JUN•ATF3-binding site that is associated with Alzheimer’s and other neurological , cognitive and behavioral disorders . Our bZIP-DNA interactomes identify 156 SNPs that potentially impact 646 bZIP binding events , any one of these could potentially contribute to the associated ailments . Not only do our data help better annotate the genome , they also serve as an invaluable resource to generate hypotheses on how genetic variants may contribute to the etiology of a range of diseases . The recent emergence of powerful high-throughput platforms for mapping protein-DNA interactomes has brought the goal of comprehensively mapping the binding specificities of all individual human TFs within reach ( Carlson et al . , 2010; Jolma et al . , 2013; Stormo and Zhao , 2010; Weirauch et al . , 2014 ) . However , it is clear from our work as well as the recent work of others that the binding of TFs to each other , and/or to adjacent DNA sites , can influence binding specificity profiles in important ways ( Ansari and Peterson-Kaufman , 2011; Garvie et al . , 2001; Grove et al . , 2009; Jolma et al . , 2015; Mann et al . , 2013; Siggers et al . , 2012; Slattery et al . , 2011 ) . Our study heralds the important next wave of specificity mapping , in which the field will tackle the effects of higher order interactions and begin to relate these to the transcriptional control of key biological processes . Human bZIP proteins containing the basic-region and coiled-coiled domains with an N-terminal 6x His tag and a C-terminal intein-chitin binding domain were expressed as described previously ( Reinke et al . , 2013 ) . Sequences are in Supplementary file 1A . Briefly , Escherichia coli RP3098 cells transformed with bZIP clones were grown in 0 . 5 L LB cultures at 37°C to OD600 = 0 . 4–0 . 8 . Expression was induced with the addition of 0 . 5 mM IPTG ( Isopropyl β-D-thiogalactopyranoside ) and cultures incubated for 3–4 hr at which point cells were pelleted . Cells pellets were resuspended in 20 mM HEPES pH 8 . 0 , 500 mM NaCl , 2 mM EDTA ( ethylenediaminetetraacetic acid ) , 1 M guanidine-HCl , 0 . 2 mM PMSF ( phenylmethylsulfonyl fluoride ) , and 0 . 1% Trition X-100 ) . Cells were then sonicated and the lysate poured over a column of 1 ml chitin beads to bind the protein ( NEB , Ipswich , MA ) . The column was then washed and equilibrated in EPL buffer ( 50 mM HEPES pH 8 . 0 , 500 mM NaCl , 200 mM MESNA ( 2-mercaptoethanesulfonic acid ) , 1 M guanidine-HCl ) . The bZIP domain was then cleaved from the intein and labeled with biotin on the C-terminus by incubation for at least 16 hr in 1 ml EPL buffer containing 1 mg/ml cysteine-lysine-biotin peptide ( CELTEK Peptides , Nashville , TN ) . The cleaved and biotin-labeled proteins were then eluted from the column using EPL buffer without MESNA and then diluted fivefold into denaturing buffer ( 6 M guanidine-HCl , 5 mM imidazole , 0 . 5 M NaCl , 20 mM TRIS , 1 mM ( DTT ) Dithiothreitol , pH 7 . 9 ) and bound to a column containing 1 ML Ni-NTA beads ( QIAGEN , Hilden , Germany ) . Columns were washed and proteins eluted with 60% ACN ( Acetonitrile ) /0 . 1% TFA ( Trifluoroacetic acid ) . The labeled proteins were then lyophilized , resuspended , and desalted using spin-columns ( Bio-Rad , Hercules , CA ) . Proteins were stored in 10 mM potassium phosphate pH 4 . 5 at −80°C . Peptide concentrations were determined by measuring absorbance at 280 nM in 6 M guanidine-HCl/100 mM sodium phosphate pH 7 . 4 . The fluorescein and TAMRA-labeled proteins used in gel-shift assays were generated as described previously ( Reinke et al . , 2013 ) . Cognate-binding sites for bZIP homo- and heterodimers were determined by SELEX-seq ( Jolma et al . , 2010; Tietjen et al . , 2011; Zhao et al . , 2009; Zykovich et al . , 2009 ) . A DNA library ( Integrated DNA Technologies , Inc . ) with a central randomized 20 bp region ( 1012 possible sequences ) , flanked by constant sequences used for amplification was used ( Supplementary file 1B ) . In vitro selections were performed as follows . For bZIP homodimers , purified , C-terminal biotinylated-bZIP proteins ( 50 nM ) were added to 100 nM of DNA library ( Binding buffer: 1x PBS ( 10 mM PO43- , 137 mM NaCl , 2 . 7 mM KCl ) , pH 7 . 6 , 2 . 5 mM DTT , 50 ng/µl poly dI-dC , 0 . 1% BSA ) and incubated at room temperature for 1 hr . The DNA library concentration and volume ( 20 µl ) were such that there was a high probability of sampling at least one copy of every 20-mer sequence ( 1012 permutations ) . bZIP-DNA complexes were enriched with streptavidin-coated magnetic beads ( Dynabeads , Invitrogen , Carlsbad , CA ) following the manufacturer’s protocol . After pull-down , three quick washes with 100 µl ice-cold binding buffer were performed to remove unbound DNA . Beads were resuspended in a PCR master mix ( EconoTaq PLUS 2X Master Mix , Lucigen ) and the DNA was amplified for 15 cycles . Amplified DNA was column purified ( QIAGEN ) , quantified by absorbance at 260 nm , and used for subsequent binding rounds . Three rounds of selection were performed . For bZIP heterodimers , one bZIP partner had a C-terminal biotin tag . bZIP-DNA complexes were pulled down with streptavidin-coated magnetic beads . Several steps were followed to decrease DNA binding by competing homodimers: ( 1 ) a 1:3 molar ratio with an excess of the non-biotinylated bZIP was used to shift the thermodynamic equilibrium from the biotin-labeled homodimer; ( 2 ) the biotin-bZIP used for pull-down was chosen as the more weakly interacting homodimer of the two interaction partners . As a convention , when naming bZIP heterodimers , the bZIP-biotin is listed first , unless otherwise stated . After three rounds of selection , an additional PCR was done to incorporate Illumina sequencing adapters and a unique 6 bp barcode for multiplexing . The starting library ( Round 0 ) was also barcoded . Up to 180 samples were combined and sequenced in a single Illumina GAIIx or HiSeq2000 lane . Illumina sequencing yielded ~180 million reads per lane . Reads were de-multiplexed by requiring an exact match to the 6 bp barcode and truncated to include only the 20 bp derived from the random portion of the library . On average , we obtained 850 , 000 reads per barcode . The occurrence of every k-mer ( lengths 8 through 14 bp ) was counted using a sliding window of size k . To correct for biases in our starting DNA library , we took the ratio of the counts of every k-mer to the expected number of counts in the starting library . The starting library was modeled using a fifth-order Markov Model derived from the sequencing reads corresponding to the starting library ( Round 0 ) ( Slattery et al . , 2011 ) . We then calculated a CSI intensity ( z-score = ( x – µ ) / σ ) for each k-mer , using the distribution of k-mer enrichment values for that dimer . The most enriched 10 , 12 , and 14 bp subsequences were used to derive PWM motifs using MEME . Samples that failed to enrich specific sequences relative to the starting library ( Round 0 ) or that only enriched low-complexity sequences were not included in further analysis . Data files for 20 bp reads and normalized 10 bp sequences are available at https://ansarilab . biochem . wisc . edu/computation . html . Previously reported bZIP-DNA interaction data were downloaded from study PRJEB3289 in the European Nucleotide Archive ( http://www . ebi . ac . uk/ena/data/view/PRJEB3289 ) ( Jolma et al . , 2013 ) . 20 bp reads for bZIP proteins and their corresponding 20 bp DNA library ( round 0 ) were analyzed as described previously . Binding profiles were defined for each bZIP pair using the CSI intensities ( z-scores ) of 1222 unique 10-mer sequences . This set of 10-mers is composed of the 50 highest-scoring sequences for each dimer . Unsupervised hierarchical clustering of pair-wise binding profile similarities , assessed by Pearson’s correlation coefficient r , was done using R . Dendrograms and heatmaps were generated using the heatmap . 2 function in the gplots R-package . Heterodimers were labeled as such if the bZIP-DNA complex was pulled-down by a biotinylated bZIP that does not binds DNA as a homodimer in our experimental conditions . If the bZIP used for pull-down of the bZIP heterodimer also bound DNA as a homodimer , the observed DNA specificity was assigned to the heterodimer only if the heterodimer specificity landscape was different ( t-test p<0 . 05 ) from the homodimer specificity , assessed by correlation scores ( Figure 1—figure supplement 4 ) . PWMs were derived from the 1000 most enriched 12-mer sequences ( ranked by z-score ) for each bZIP pair , using the MEME ( Bailey and Elkan , 1994 ) . The most enriched 14-mer sequences were used for MAF dimers . MEME was run with following parameters: -dna -mod anr -nmotifs 10 -minw 8 -maxw 18 -time 7200 -maxsize 60000 –revcomp . SELs display high-throughput protein-DNA ( or protein-RNA ) binding data for both array and sequencing methods ( Campbell et al . , 2012; Carlson et al . , 2010; Tietjen et al . , 2011 ) . The organization of data in SEL is detailed in Figure 2A . The SELs shown in this work were generated from 10- , 12- , or 14-mer intensity files . Seed motifs were derived from PWM-derived DNA logos or from the highest intensity k-mer , and are shown on top of each SEL . The length of the seed motifs has to be smaller than the k-mer length of the CSI intensity file . The software to generate SELs is provided as Source Code Files ( SEL_10MER and SEL12MER_14MER ) . An electrophoretic mobility shift assay ( EMSA ) with fluorescence resonance energy transfer ( FRET ) readout was used to validate bZIP heterodimer binding to DNA . The assay relies on the ability to observe FRET between two fluorophores , TAMRA and fluorescein , as well as to detect each fluorophore in the absence of FRET ( Figure 3A ) . The assay also measures DNA fluorescence to ensure that protein-DNA complexes are being examined . Two versions of each bZIP were made , one conjugated to TAMRA and the other to fluorescein . We observed that the fluorophores reproducibly retard ( TAMRA ) or increase ( fluorescein ) the mobility of the bZIP protein that they are attached to and thus assist in resolving each heterodimer with respect to the two homodimers formed by contributing partners . The sequences of all the DNA sites used are listed in Supplementary file 1D . Each site was flanked by six constant nucleotides on each side ( GAGTCC-site-CCGTAG ) . Oligos modified on the 5′ end with the dye TYE 665 ( IDT , Coralville , IA ) were annealed with an unlabeled reverse-complement oligo . Binding reactions contained 50 nM of each fluorescein- and TAMRA-labeled proteins , 10 nM annealed dye-labeled DNA in 20 µl of binding buffer ( 50 mM potassium phosphate pH 7 . 4 , 150 mM KCl , 0 . 1% BSA , 0 . 1% Tween-20 , 5 ng/µl poly ( dI-dC ) , 0 . 5 mM TCEP ) . Samples were mixed , incubated at 37°C for 30 min , and then at 21°C for 30 min . NOVEX 6% DNA retardation gels were loaded with 16 µl of each sample ( Life Technologies , Carlsbad , CA ) and run at 300V for 20–22 min at 22–25°C . Gels were then imaged using a Typhoon 9500 scanner ( GE Healthcare Bio-Sciences Corp . , Piscataway , NJ ) with separate channels for fluorescein , TAMRA , TYE 665 , and FRET . Bleed through between channels was corrected using the spectral-unmixing plugin in ImageJ ( http://rsb . info . nih . gov/ij/ ) . The amount of DNA bound for the homodimers was calculated by quantifying the DNA signal that corresponded to all three bound species ( fluorescein homodimer , TAMRA homodimer , and mixed-dye homodimer ) . For the heterodimers , the amount of DNA bound was calculated by quantifying the DNA signal that corresponded to the mixed dye heterodimer . The amount of bound DNA was divided by the amount of unbound DNA run without protein added . For each heterodimer , the interaction was measured twice , with the fluorescein and TAMRA dye on different proteins , and the average of the two measurements is reported . ChIP-seq peaks from the ENCODE project used in this work were downloaded from ftp://hgdownload . cse . ucsc . edu/goldenPath/hg19/encodeDCC/wgEncodeAwgTfbsUniform/ ( Dunham et al . , 2012 ) . Overlapping genomic regions of ChIP-seq peaks were determined and extracted using bedops ( Neph et al . , 2012 ) . For ATF3 ChIP-seq in GBM1 cells , aligned reads ( . bam file ) were downloaded from GEO ( GSE33912 ) . ATF3 peaks were called using the MACS tool ( Zhang et al . , 2008 ) in the Galaxy ( Goecks et al . , 2010 ) platform using default parameters . Overlapping ATF3-bound regions between different cell lines ( Figure 5 ) were determined using the ChIPpeakAnno R-package ( Zhu et al . , 2010 ) . A CSI Genomescape is a plot generated by assigning in vitro CSI intensities ( z-scores ) to genomic regions . To generate the CSI Genomescapes in Figures 6 and 7 , a 10 bp sliding window was used to score reported ChIP-seq peaks using quantile-normalized CSI intensities for different bZIP dimers as follows: Given a bZIP pair and a ChIP-seq peak , the peak was assigned the maximum CSI intensity for any 10-mer within the reported peak . CSI Genomescapes of ChIP-seq data sets were then used to generate Receiver Operating Characteristic ( ROC ) curves to reflect how well the in vitro binding data for different bZIPs explains the ChIP-seq data . In this analysis , ChIP-seq peaks were used as a true positive set , whereas two regions of equal length ±5 kbp from the center of each peak ( that did not overlap another ChIP-seq peak ) were chosen to make the true negative set . The fraction of regions in the positive vs . negative sets with scores above a varying CSI intensity cutoff were plotted to generate ROC curves ( True Positive Rate vs . False Positive Rate ) . ATF3-bound regions ( ChIP-seq peaks ) were scored with the CSI intensities for ATF3 homodimer or for ATF3-containing heterodimers to generate the areas under the curves . Heatmaps and clustergrams in Figure 5 were made by hierarchical clustering of the lowest FPR-cutoff values at which peaks were detected as positives using the CSI intensities of the ATF3 containing dimers . ROC curves and heatmaps were generated in MATLAB . Motif finding within ChIP-seq peaks was done with MEME-ChIP with default settings ( Machanick and Bailey , 2011 ) . Enrichment of functional annotations of genomic regions was done with Genomic Regions Enrichment of Annotations Tool ( GREAT ) with default settings ( McLean et al . , 2010 ) . Gene Ontology annotations that are significantly enriched ( FDR < 0 . 05 ) by both binomial and hypergeometric test are shown . The False Discovery Rate ( q-value ) is corrected for multiple hypothesis tests . SNPs linked to diseases or quantitative traits by GWAS were obtained from the Supplemental Table S2 from Maurano et al . , which reports human SNPs associated to diseases and quantitative traits ( Maurano et al . , 2012 ) . For each SNP , we considered 21 bp region centered on the SNP ( 10 bp on each side ) and assigned a score using the CSI intensity data all 10-mers . We scored both alleles using a 10 bp sliding window and assigning the highest CSI intensity ( z-score ) in the 21 bp fragment; each 21 bp region was scored with twelve 10 bp windows . We calculated a predicted fold-difference in CSI intensity between a given SNP and its reference allele ( hg19 ) using the following formula: ( CSI Intensity for alternate allele − Minimum CSI Intensity+A ) ( CSI Intensity for reference allele ( hg19 ) − Minimum CSI Intensity+A ) where the A = ( Maximum CSI Intensity – Minimum CSI Intensity ) * F , Minimum CSI Intensity = minimum CSI Intensity ( z-score ) among the scored SNPs , Maximum CSI Intensity = maximum CSI Intensity ( z-score ) among the scored SNPs . And F is a noise factor which was varied from 1% to 90% , from lower to higher stringency in estimating the predicted difference in CSI intensity . We added a noise factor ( F ) to the formula to make the fold-change prediction less sensitive to low CSI scores and decrease the number of false-positives predictions .	Most cells in our body contain the same DNA blueprint , which encodes all the genes needed for every process in the body . However , only certain genes need to be active in a particular cell at any given time . Proteins known as transcription factors control the activity of genes by binding to DNA near the start of the genes and switching genes on or off as required . Often transcription factors work together to regulate specific genes in response to signals from other cells or the environment . Failure to control the activity of genes can give rise to cancer , diabetes and a wide array of other diseases . The bZIP family of transcription factors regulates the activities of many genes . These transcription factors work in pairs to bind a specific DNA site . They either partner with an identical molecule or with a different bZIP transcription factor . Different combinations of bZIP pairs prefer to bind different stretches of DNA . Except for a few examples , it is not yet understood how bZIP pairs work together to find the right target DNA . Rodriguez-Martinez , Reinke , Bhimsaria et al . have identified all of the DNA sites that 102 pairs of human bZIP transcription factors can bind to . The experiments show that over two thirds of the bZIP pairs bind to DNA sequences that each individual partner prefers . However , in many cases , the choice of a partner can change the DNA sequence that the pair targets in a manner that could not have been predicted based on the preferences of each partner alone . This suggests that , by pairing up , bZIP transcription factors are able to change their preferences for which location they target in the DNA . The experiments also show that many of the 102 pairs could bind to more than one type of site . Thus , the ability of bZIP proteins to interact with different partners greatly expands the locations on genomic DNA from which they can regulate the activity of different genes . DNA sequences vary between different individuals and some variants can predispose individuals to certain diseases . Rodriguez-Martinez et al . found that bZIP pairs can bind to over a hundred DNA variants that are associated with disease . The next challenge is to find out how specific variations in DNA can lead to the formation or elimination of bZIP binding sites that cause disease . In the future , DNA editing methods may make it possible to specifically fix such changes in our genomes to reduce the risk of disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "computational", "and", "systems", "biology" ]	2017	Combinatorial bZIP dimers display complex DNA-binding specificity landscapes
The transcription factor SOX2 is central in establishing and maintaining pluripotency . The processes that modulate SOX2 activity to promote pluripotency are not well understood . Here , we show SOX2 is O-GlcNAc modified in its transactivation domain during reprogramming and in mouse embryonic stem cells ( mESCs ) . Upon induction of differentiation SOX2 O-GlcNAcylation at serine 248 is decreased . Replacing wild type with an O-GlcNAc-deficient SOX2 ( S248A ) increases reprogramming efficiency . ESCs with O-GlcNAc-deficient SOX2 exhibit alterations in gene expression . This change correlates with altered protein-protein interactions and genomic occupancy of the O-GlcNAc-deficient SOX2 compared to wild type . In addition , SOX2 O-GlcNAcylation impairs the SOX2-PARP1 interaction , which has been shown to regulate ESC self-renewal . These findings show that SOX2 activity is modulated by O-GlcNAc , and provide a novel regulatory mechanism for this crucial pluripotency transcription factor . SOX2 ( sex determining region Y-box 2 ) is a transcription factor necessary for embryonic stem cell ( ESC ) self-renewal ( Arnold et al . , 2011; Masui et al . , 2007 ) . Precise control of SOX2 is critical for ESC maintenance , since increased or decreased expression of SOX2 interferes with self-renewal and pluripotency ( Kopp et al . , 2008; Masui et al . , 2007 ) . Post-translational modifications ( PTMs ) of SOX2 may play a role in its regulation , as SOX2 is reported to be phosphorylated , methylated , ubiquitinylated , SUMOylated , acetylated , and PARylated ( Baltus et al . , 2009; Brumbaugh et al . , 2012; Fang et al . , 2014; Gao et al . , 2009; Lai et al . , 2012; Swaney et al . , 2009; Tahmasebi et al . , 2013; Tsuruzoe et al . , 2006; Van Hoof et al . , 2009; Zhao et al . , 2011 ) . We have previously shown SOX2 is O-linked N-acetlyglucosamine ( O-GlcNAc ) modified in mouse ESCs ( mESCs ) ( Myers et al . , 2011 ) . O-GlcNAcylation is the dynamic and regulatory mono-glycosylation of nucleocytosolic proteins catalyzed by a single O-GlcNAc transferase ( OGT ) and removed by a single hydrolase ( OGA/MGEA5/NCOAT ) . O-GlcNAc signaling is essential for embryo viability ( O'Donnell et al . , 2004; Shafi et al . , 2000; Yang et al . , 2012 ) and mESC self-renewal ( Jang et al . , 2012 ) . While OGT is critical for mESC maintenance , the protein- and site-specific functions of O-GlcNAcylation in mESCs have not been fully elucidated . Here , we show that O-GlcNAcylation of SOX2 at serine 248 ( S248 ) is dynamically regulated in mESCs . Upon differentiation , O-GlcNAc occupancy is reduced and SOX2 is predominantly unmodified at this site . Replacement of wild type SOX2 ( SOX2WT ) with an O-GlcNAc-deficient mutant SOX2 ( SOX2S248A ) results in increased reprogramming efficiency . mESCs with SOX2S248A as their sole source of SOX2 have increased expression of genes associated with pluripotency and exhibit a decreased requirement for OCT4 . SOX2S248A exhibits altered genomic occupancy and differential association with transcriptional regulatory complexes . O-GlcNAcylation directly inhibits the SOX2 and PARP1 interaction , which plays a regulatory role in mESC pluripotency ( Gao et al . , 2009; Lai et al . , 2012 ) . This study implicates O-GlcNAc modification in coordinating genomic occupancy and protein-protein interactions of SOX2 in ESCs , and provides molecular insight into how this broadly expressed transcription factor is regulated to promote the pluripotency-specific expression program . Previously , we reported in mESCs SOX2 was O-GlcNAcylated in the transactivation domain ( TAD ) ( Myers et al . , 2011 ) , a region described to possess several other PTMs ( Brumbaugh et al . , 2012; Swaney et al . , 2009; Tahmasebi et al . , 2013; Tsuruzoe et al . , 2006; Van Hoof et al . , 2009 ) . To investigate whether PTMs in the TAD are subject to developmental regulation , we analyzed SOX2 during the initial stages of differentiation . Knock-in FLAG/HA tagged SOX2 mESCs ( KI cells; Lai et al . , 2012 ) were cultured in media containing MEK and GSK3b inhibitors ( 2i ) and leukemia inhibitory factor ( LIF ) , or were induced to differentiate by removing LIF and adding retinoic acid ( RA ) . Liquid chromatography coupled with tandem mass spectrometric ( LC-MS/MS ) analysis of SOX2 in self-renewal conditions revealed four main populations of the TAD peptide containing serine 248 ( hereinafter referred to as TAD peptide ) : unmodified , O-GlcNAcylated at one of two sites ( S248 or T258 ) , phosphorylated at S253 , and doubly modified with O-GlcNAcylation at S248 and phosphorylation at S253 , ( Figure 1A—figure supplements 1–5 ) . Removal of LIF and addition of RA for 48 hr resulted in a marked decrease in the O-GlcNAc occupancy of the TAD peptide with no change in the phosphorylation stoichiometry ( Figure 1B–C ) . These data indicate O-GlcNAcylation of SOX2 S248 is responsive to differentiation cues . 10 . 7554/eLife . 10647 . 003Figure 1 . SOX2 O-GlcNAc levels change during differentiation . ( A ) Diagram of SOX2 ( bottom , with TAD and high mobility group DNA binding domain [HMG] indicated ) , the TAD peptide sequence ( middle; amino acid numbering from the Uniprot accession number P48432 ) , and the PTM isoforms identified on the TAD peptide ( top , grey and white rectangles , g indicates O-GlcNAc and p indicates phosphate ) . Mass spectra can be seen in Figure 1—figure supplement 1 . ( B ) and ( C ) Extracted ion chromatographs ( XICs ) of SOX2 TAD peptide PTM states from ( B ) undifferentiated KI SOX2 mESCs ( 2i+L ) or ( C ) differentiated KI SOX2 mESCs ( RA 48 hr ) . Traces for each PTM isoform are colored differently , key provided in the inset in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00310 . 7554/eLife . 10647 . 004Figure 1—figure supplement 1 . ETD-MS/MS spectra of SOX2 unmodified TAD peptide described in Figure 1A . Proton transfer species are not labeled , ppm , parts per million; b . p . , base peak . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00410 . 7554/eLife . 10647 . 005Figure 1—figure supplement 2 . ETD-MS/MS spectra of SOX2 GlcNAc-S248 TAD peptide described in Figure 1A . Proton transfer species are not labeled , ppm , parts per million; b . p . , base peak . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00510 . 7554/eLife . 10647 . 006Figure 1—figure supplement 3 . ETD-MS/MS spectra of SOX2 GlcNAc-T258 TAD peptide described in Figure 1A . Proton transfer species are not labeled , ppm , parts per million; b . p . , base peak . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00610 . 7554/eLife . 10647 . 007Figure 1—figure supplement 4 . ETD-MS/MS spectra of SOX2 phospho-S253 TAD peptide described in Figure 1A . Proton transfer species are not labeled , ppm , parts per million; b . p . , base peak . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00710 . 7554/eLife . 10647 . 008Figure 1—figure supplement 5 . ETD-MS/MS spectra of SOX2 co-modified GlcNAc-S248/phospho-S253 TAD peptide described in Figure 1A . Proton transfer species are not labeled , ppm , parts per million; b . p . , base peak , † indicates co-isolating contamination peak . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00810 . 7554/eLife . 10647 . 009Figure 1—figure supplement 6 . O-GlcNAcylation of OCT4 at T228 is undetectable in mESCs . OCT4 was FLAG-affinity purified from ZHBTc4 F-Oct4 ESCs ( the same cell line employed in [Jang et al . , 2012] ) , which express FLAG-tagged OCT4 . Peptides containing T228 , the residue reported to be O-GlcNAcylated ( Jang et al . , 2012 ) , were analyzed by LC-MS/MS . ( A ) XICs ( 10 ppm ) for predicted m/z of an unmodified tryptic peptide containing T228 ( top , black panel ) and the proposed O-GlcNAc modified peptide ( bottom , red panel ) . XICs show the signal for the unmodified peptide , but no signal for the O-GlcNAc modified version . ( B ) XICs ( 10 ppm ) for m/z of predicted O-GlcNAc modified LysC-derived peptides containing T228 is not detectable ( two charge states , with and without Met oxidation ) . ( C ) Zoom of MS1 scans from Figure supplement 6B where potential signal would exists . No signal with the appropriate charge state was detected for any LysC derived peptide indicating all signal in B is background . ( D ) Western blot analysis of purified FLAG-tagged OCT4 from ZHBT c4 F-OCT4 mESCs failed to detect O-GlcNAcylation . ( E ) Western blot analysis of 3xF-SOX2 purified from OSFLAG-WTKM transduced MEFs detects SOX2 O-GlcNAcylation , which can be blocked by the addition of 100 mM free GlcNAc to the primary antibody incubation . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 00910 . 7554/eLife . 10647 . 010Figure 1—figure supplement 7 . Incorrect assignment of GluC digest OCT4 peptide mass spectrum containing T228 . ( A ) Modified mass spectrum from Jang et al . identifying OCT4 T228 as O-GlcNAc modified ( Figure 3D; Cell Stem Cell 17 May 2012 [doi: 10 . 1016/j . stem . 2012 . 03 . 001] ) . Problematic issues are 1 ) discrepancy with well-established fragmentation patterns for O-linked glycopeptides; 2 ) confusing labeling of peaks ( original black arrows , which obscured peak identifications , were changed to green for this figure ) ; 3 ) incorrect peptide sequence . GluC digestion should yield ICKSETLVQARKRKRTSIE , where the initial Ile is missing from the Jang et al . assignment; 4 ) only OCT4 peptide sequences were searched increasing false positive rates; and 5 ) the precursor mass and charge state were not reported making the data uninterpretable . ( B ) Ion trap CAD mass spectrum of synthetic glycopeptide of the reported amino acid sequence shows the original spectrum in Jang et al . was misinterpreted . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 010 In mESCs SOX2 heterodimerizes with OCT4 , which is also reported to be O-GlcNAcylated in this cell type ( Jang et al . , 2012 ) . Thus , it is possible that OGT targets both these transcription factors when they are complexed together , prompting us to query OCT4 O-GlcNAcylation . However , we were unable to detect OCT4 O-GlcNAcylation in mESCs ( Figure 1—figure supplements 6–7 ) . To query whether O-GlcNAcylation at S248 is present in other contexts , we examined the PTM profile of the SOX2 TAD peptide during somatic cell reprogramming . We used four-factor retroviral reprogramming ( Oct4 , Sox2 , Klf4 and Myc; OSKM ) in which SOX2 contained a triple FLAG tag ( OSFLAG-WTKM ) . LC-MS/MS analysis of purified SOX2 six days after transduction of mouse embryonic fibroblasts ( MEFs ) showed S248 is O-GlcNAcylated ( Figure 2A ) . In addition , mutation of S248 to alanine ( S248A ) resulted in loss of O-GlcNAcylation without affecting the other PTMs of the TAD peptide ( Figure 2B–C ) . These results demonstrate S248 is O-GlcNAc modified during somatic cell reprogramming and suggest a connection between this SOX2 PTM and pluripotency . 10 . 7554/eLife . 10647 . 011Figure 2 . O-GlcNAc-deficient SOX2 , SOX2S248A , increases somatic cell reprogramming efficiency . ( A ) Diagram of SOX2 and the PTMs identified from MEFs transduced with OSFLAG-WTKM , labeled as described in Figure 1A . Spectra can be found at tinyurl . com/iPSC-3xF-SOX2-ETD and tinyurl . com/iPSC-3xF-SOX2-HCD . ( B ) XICs of 3xF-SOX2WT TAD peptide PTM states from OSFLAG-WTKM-transduced MEFs . ( C ) XICs of 3xF-SOX2S248A TAD peptide PTM states from OSFLAG-S248AKM-transduced MEFs . Color key the same as in ( B ) . ( D ) Number of GFP+ colonies from 1000 Nanog-Gfp MEFs transduced with OSWTKM or OSS248AKM and cultured on SNL feeders for 18 or 20 days ( n=7 +/- S . E . M . ) . ( E ) Chimeric mouse derived from iPSCs obtained from transducing Nanog-Gfp MEFs with OSS248AKM and his black offspring , demonstrating germline transmission . ( F ) Western blots against FLAG , SOX2 , OGT and TUBULIN for the first six days of reprogramming with either OSFLAG-WTKM or OSFLAG-S248AKM . “Endo” refers to the apparent molecular weight at which the endogenous SOX2 would be expected , “3xF” refers the the FLAG tagged version from the viral transduction . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01110 . 7554/eLife . 10647 . 012Figure 2—figure supplement 1 . Immunofluorescence staining against FLAG in MEFs six days after transduction with either OSFLAG-WTKM or OSFLAG-S248AKM shows similar nucleocytoplasmic distribution . E14 mESCs are used as a staining negative control . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01210 . 7554/eLife . 10647 . 013Figure 2—figure supplement 2 . SOX2S248D also increases somatic cell reprogramming efficiency . Relative increase , compared to OSWTKM , in number of GFP+ colonies from 1000 Nanog-Gfp MEFs that were infected with OSS248AKM or the phosphomimetic OSS248DKM and cultured on SNL feeders for 20 days after infection ( n=7 for OSWTKM and OSS248AKM , two for OSS248DKM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 013 To determine whether the S248A mutation impacted induced pluripotent stem cell ( iPSC ) colony formation , we used somatic cell reprogramming of Nanog-Gfp reporter MEFs ( Takahashi and Yamanaka , 2006 ) . Nanog-Gfp MEFs transduced with OSS248AKM produced significantly more GFP+ iPSC colonies compared to OSWTKM ( Figure 2D ) . iPSCs generated with OSS248AKM exhibited standard colony morphology and contributed to chimeric mice capable of germ line transmission ( Figure 2E ) , indicating these OSS248AKM iPSCs exhibit the features of normal iPSCs . By Western blot and immunostaining of MEFs transduced with OSFLAG-WTKM or OSFLAG-S248AKM showed equal levels of exogenous SOX2 for the first six days ofof reprogramming ( Figure 2F and Figure 2—figure supplement 1 ) , indicating comparable expression of WT and S248A triple FLAG tagged SOX2 . OGT levels were also similar for the first six days of reprogramming between OSFLAG-WTKM and OSFLAG-S248AKM transduced MEFs ( Figure 2F ) . These results indicate that SOX2S248A is more efficient than wild type SOX2 at inducing pluripotency and suggest O-GlcNAcylation at S248 inhibits SOX2 activity . The homologous SOX2 residue has been reported to be phosphorylated in human ESCs ( Swaney et al . , 2009 ) . While our lab and others were unable to detect this phosphorylation in mESCs ( Brumbaugh et al . , 2012 ) or during murine reprogramming , the S248A mutation could potentially remove a regulatory phosphorylation site . Therefore , we performed reprogramming experiments using the phospho-mimetic SOX2 mutant , S248D . This mutation also increased reprogramming efficiency ( Figure 2—figure supplement 2 ) , suggesting it is the removal of an O-GlcNAcylation site , and not of a phosphorylation site , which mediates the effect on SOX2 activity . Since the reprogramming results suggested the S248A mutation increased SOX2 activity , we examined whether this mutant SOX2 supported mESC self-renewal . We generated mESC lines that express either a FLAG-tagged wild-type Sox2 transgene ( fSOX2-Tg cells ) or an S248A transgene ( fS248A-Tg cells ) ( Figure 3A ) . We introduced the transgenes into 2TS22C mESCs , in which endogenous Sox2 is removed and a doxycycline repressible SOX2 cDNA transgene supports self-renewal ( Masui et al . , 2007 ) ( Figure 3—figure supplement 1 ) . Under doxycycline repression , the sole source of SOX2 in these transgenic lines is the FLAG-tagged wild-type or S248A mutant SOX2 ( Figure 3B ) . SOX2 levels in fSOX2-Tg and fS248A-Tg mESCs are comparable to SOX2 levels in the 2TS22C parental cell line and nucleo-cytoplasmic distribution was not altered by the mutation ( Figure 3C ) . OCT4 and NANOG abundance and distribution were comparable between fSOX2-Tg and fS248A-Tg mESCs ( Figure 3C ) , arguing that there is no gross effect on these pluripotency transcription factors . 10 . 7554/eLife . 10647 . 014Figure 3 . SOX2S248A can replace wild type SOX2 in mESCs . ( A ) Characterization of fSOX2-Tg and fS248A-Tg mESCs . fSOX2-Tg and fS248A-Tg mESCs exhibit AP staining , a marker of pluripotency , similar to parental 2TS22C cells . ( B ) Western blot analysis of SOX2 and FLAG in 2TS22C , fSOX2-Tg and fS248A-Tg mESCs . TUBULIN ( TUB ) is used as a loading control . “3xFLAG” and “untagged” refer to expected molecular weights of SOX2 with the 3xFLAG tag or no tag , respectively . ( C ) Immunofluorescence staining for NANOG , SOX2 , FLAG and OCT4 in wild type E14 , parental 2TS22C , fSOX2-Tg , and fS248A-Tg mESCs . Antibody staining is green , nuclear stain with DAPI is blue . ( D ) and ( E ) XICs of the TAD peptides of SOX2 immunopurified from fSOX2-Tg ( D ) and fS248A-Tg ( E ) mESCs . Insets: pie charts showing the mean percentage of each PTM form to total TAD peptide signal ( n=3 ) . The doubly phosphorylated TAD peptide is below the limit of quantitation for both cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01410 . 7554/eLife . 10647 . 015Figure 3—figure supplement 1 . Diagram of creation of fSOX2-Tg or fS248A-Tg lines . ( A ) Derivation of 2TS22C mESCs , which are deleted for endogenous copies of Sox2 and express tetracycline-off ( tet-off ) transgenic Sox2 , summarized from ( Masui et al . , 2007 ) . ( B ) 2TS22C cells were transfected with the constructs pCAG-3xF-SOX2WT-IRES-puroR ( upper ) or pCAG-3xF-SOX2S248A-IRES-puroR ( lower ) , in which Sox2 expression is driven by the CAG promoter . After 24 hr , doxycycline was added to cultures to repress the tet-regulated copy of SOX2 expressed by the parental line 2TS22C . 48 hr after transfection , puromycin was added to cultures to select for stable integrants . After two weeks , colonies with ESC-like colony morphology were expanded and characterized for SOX2 expression . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01510 . 7554/eLife . 10647 . 016Figure 3—figure supplement 2 . Diagram of SOX2 and the PTMs identified from fSOX2-Tg cells , labeled as described in Figure 1A . Spectra can be found at tinyurl . com/3xF-SOX2-ETD and tinyurl . com/3xF-SOX2-HCD . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01610 . 7554/eLife . 10647 . 017Figure 3—figure supplement 3 . Label-free MS1 analysis of synthetic SOX2 TAD peptides . The WT unmodified ( grey ) , WT O-GlcNAc modified ( red ) and S248A unmodified peptides ( blue ) were loaded at known concentrations and the MS1 peak area was plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 017 LC-MS/MS analysis of immunopurified SOX2 from fSOX2-Tg mESCs identified nine PTM forms of the SOX2 TAD peptide ( Figure 3—figure supplement 2 ) . LC-MS analysis of the TAD peptide precursor masses from fSOX2-Tg mESCs showed unmodified and singly O-GlcNAcylated were the most abundant forms of the SOX2 TAD peptide ( 33 . 2 and 44 . 1% of total TAD , respectively ) ( Figure 3D ) . LC-MS analysis confirmed the loss of S248 O-GlcNAcylation in fS248A-Tg mESCs ( Figure 3E ) . Analysis of synthetic SOX2 TAD peptides showed chromatographic separation of PTM or mutant isoforms , and lack of electrospray ionization suppression , validating our label free quantitation approach ( Figure 3—figure supplement 3 ) . In addition , the TAD peptide in fS248A-Tg mESCs showed increased phosphorylation at S253 , from 10 . 9 to 18 . 7% of total TAD , suggesting cross talk between phosphorylation and O-GlcNAcylation . To determine if the S248A mutation altered global transcript levels we used microarrays to compare the gene expression profiles of fSOX2-Tg and fS248A-Tg mESCs . Significant changes in mRNA levels were observed , with 320 genes up regulated and 344 genes down regulated in fS248A-Tg cells ( Figure 4A ) and gene set enrichment analysis of differentially expressed genes did not show significant enrichment of any pathways . Several genes that promote pluripotency and self-renewal were upregulated , while several genes associated with differentiation were down-regulated in fS248A-Tg cells compared to WT . RT-qPCR confirmed the differential expression of these pluripotency or differentiation genes ( Figure 4B ) . These data suggest the S248A mutation alters the balance between self-renewal and differentiation gene expression in mESCs . 10 . 7554/eLife . 10647 . 018Figure 4 . fS248A-Tg mESCs show altered gene expression and decreased dependence on OCT4 . ( A ) Volcano plot of global changes in gene expression between fSOX2-Tg and fS248A-Tg cells . Red indicates genes with increased or decreased expression ( fold change cutoff 1 . 5 and paired t-test p<0 . 05 ) ( Supplementary file 1a ) . ( B ) RT-qPCR of select genes differentially expressed between fSOX2-Tg and fS248A-Tg cells ( * indicates p<0 . 05 , n=3 , +/- S . E . M . ) . ( C ) fSOX2-Tg or fS248A-Tg cells were depleted of OCT4 using siRNA pools ( esiRNAs ) and Western blot analysis of OCT4 and TUBULIN were performed . ( D ) and ( E ) , ( D ) AP staining and ( E ) quantitation of fold change in AP staining three days after OCT4 or GFP depletion in fSOX2-Tg and fS248A-Tg cells . Additional example fields of view for relative quantitation can be seen in Figure 4—figure supplement 1 . F , RT-qPCR analysis of Oct4 and Nanog mRNA levels in fSOX2-Tg or fS248A-Tg cells depleted of OCT4 compared to the control knockdown of GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 01810 . 7554/eLife . 10647 . 019Figure 4—figure supplement 1 . AP activity staining of fSOX2-Tg and fS248A-Tg cells three days after Gfp or Oct4 knockdown . Four fields of view , shown here , were used for relative AP stain quantitation for a representative replicate . Mock knockdown ( Gfp ) was used as the baseline for AP staining . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 019 The altered gene expression profile of fS248A-Tg cells suggested this mutation may promote self-renewal at the expense of differentiation . Therefore , we examined the effects of OCT4 depletion , which causes mESCs to differentiate ( Figure 4C ) ( Hough et al . , 2006 ) . While fSOX2-Tg mESCs exhibited altered cell and colony morphology ( Figure 4—figure supplement 1 ) , decreased AP staining ( Figure 4D ) , and decreased expression of Nanog ( Figure 4E ) , fS248A-Tg mESCs were relatively unaffected by OCT4 depletion . These data indicate that fS248A-Tg mESCs can maintain key features of pluripotency when OCT4 levels are reduced , and are consistent with a role for the O-GlcNAc modification inhibiting SOX2 activity . To examine whether the altered gene expression associated with the S248A mutation was accompanied by changes in SOX2 genomic occupancy , we performed FLAG chromatin immunoprecipitation followed by next generation sequencing ( ChIP-seq ) to compare SOX2 genomic distribution in fSOX2-Tg and fS248A-Tg mESCs ( Figure 5A ) . SOX2 distribution exhibited considerable overlap , with 4 , 191 sites bound in both lines ( Figure 5B ) . The mutant form of SOX2 occupied 1000 sites not bound by the wild type form ( Figure 5A ) . De novo motif analysis identified the SOX2 binding motif in fS248A-Tg specific peaks ( Figure 5C ) . In mESCs , SOX2 and OCT4 heterodimerize and co-occupy a substantial portion of their target regulatory sequences ( Boyer et al . , 2005 ) . De novo motif analysis of SOX2 peaks shared between fSOX2-Tg and fS248A-Tg mESCs identified the OCT4:SOX2 motif ( Figure 5D ) , which was present in 2335 of the shared peaks . The OCT4:SOX2 motif was not identified in any of the fS248A-Tg-specific peaks ( Figure 5E ) . These data indicate the S248A mutation alters SOX2 genomic distribution , increasing its ability to associate with SOX2 binding sites that would not ordinarily be bound by wild type SOX2 in mESCs . 10 . 7554/eLife . 10647 . 020Figure 5 . S248A mutation alters genome-wide distribution of SOX2 . ( A ) Representative UCSC genome browser tracks of FLAG ChIP-seq in fSOX2-Tg ( blue ) and fS248A-Tg ( red ) cells . Examples of fS248A-Tg specific peaks ( Pou5f1 , Esrrb ) and shared peaks ( Abca4 , Sox2 ) are shown for 2 biological replicates ( 2 technical replicates were performed for each biological replicate , Spearman correlations for technical replicates are 1 , for biological replicates 0 . 45 for fSOX2-Tg and 0 . 55 for fS248A-Tg ) . Each track is 15 kb . Green arrows indicate fS248A-Tg specific peaks . For Sox2 track , the region shown is not encompassed in the deletion removing endogenous Sox2 . ( B ) Overlap ( purple ) in called peaks from anti-FLAG ChIP-seq in fSOX2-Tg ( blue ) and fS248A-Tg ( red ) mESCs . ( C ) De novo SOX2 motif identified in shared ChIP-seq peaks between fSOX2-Tg and fS248A-Tg cells ( top ) compared to the canonical SOX2 motif [Jaspar M01271] ( bottom ) . ( D ) OCT4:SOX2 motif identified in peaks shared between fSOX2-Tg and fS248A-Tg cells using de novo motif analysis ( top ) compared to the canonical OCT4:SOX2 motif [Jaspar MA0142 . 1] ( bottom ) . ( E ) Proportion of peaks containing a motif matching the OCT4:SOX2 de novo motif in shared peaks ( left ) and fS248A-Tg specific peaks ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 020 S248 lies in the TAD of SOX2 , a region responsible for interactions with transcriptional regulatory machinery ( Ambrosetti et al . , 2000; Nowling et al . , 2000; Yuan et al . , 1995 ) . Therefore , we tested whether the S248A mutation altered SOX2 centered protein-protein interactions ( PPIs ) . We performed affinity purifications against FLAG from nuclear extracts of fSOX2-Tg , fS248A-Tg or an equivalent mESC line expressing transgenic HA-tagged SOX2 ( haSOX2-Tg , Figure 6—figure supplement 1 ) and used quantitative LC-MS to identify proteins that co-purified with FLAG in each cell type . We identified 329 proteins enriched in both fSOX2-Tg and fS248A-Tg , but not haSOX2-Tg FLAG IPs . Many of these interactors exist in complexes involved in histone modification , DNA damage repair , or nucleosome remodeling ( Figure 6A ) . Several SOX2 interactors have been previously described ( Cox et al . , 2013; Engelen et al . , 2011; Gao et al . , 2012 ) , indicating fSOX2-Tg and fS248A-Tg cells recapitulate some known SOX2 interactions ( Supplementary file 1b ) . 10 . 7554/eLife . 10647 . 021Figure 6 . O-GlcNAcylation of SOX2 at S248 alters protein-protein interactions . ( A ) Interaction diagram of a subset of SOX2 interactors that exhibit differential association with 3xF-SOX2S248A relative to 3xF- SOX2WT . Color of circles indicates with which SOX2 proteoform a protein preferentially interacts . Interaction diagram based on high confidence , experimental interactions identified by STRING . ( B ) , Anti-FLAG IP-WB for SOX2 , PARP1 , GATAD2B , and SMARCA4 in fSOX2-Tg and fS248A-Tg cells . ( C ) Heatmap of median enrichment values of NuRD subunits that preferentially associate with 3xF-SOX2WT or 3xF-SOX2S248A as determined by quantitative mass spectrometry ( n=3 ) . ( D ) Western blot analysis of in vitro interaction between SOX2 +/- O-GlcNAcylation and PARP1 . Bio-SOX2 and His-OGT were incubated with and without UDP-GlcNAc , Bio-SOX2 purified away from OGT and UDP-GlcNAc using streptavidin beads and incubated with GST-PARP1 . Western blots examine proteins associated with streptavidin beads . Comparable amounts of input and pull down were loaded for all blots , except O-GlcNAc , in which more material was loaded in the pull down lanes . WB , Western blot; GST , glutathione S-transferase tag; Bio , biotinylated Bio tag; His , polyhistidine tag . ( E ) Flow chart outlining scheme for D . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 02110 . 7554/eLife . 10647 . 022Figure 6—figure supplement 1 . Creation of haSOX2-Tg , where HA-tagged SOX2 is the sole source of SOX2 . Western blots compare parental line to haSOX2-Tg ESCs . Below table shows MS analysis identifies GlcNAc-S248 in these cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 02210 . 7554/eLife . 10647 . 023Figure 6—figure supplement 2 . O-GlcNAc site mapping by ETD-MS/MS of recombinant Bio-tagged human SOX2 incubated with recombinant human OGT and UDP-GlcNAc . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 023 We next examined whether any co-purifying proteins were enriched in either the fSOX2-Tg or fS248A-Tg co-IPs , by plotting the enrichment ratios between S248A and wild type . 22 of the interacting proteins were enriched at least four-fold in the co-IP with fSOX2-Tg ( z-score > 1 . 5 ) and 60 were enriched in the fS248A-Tg co-IP ( Supplementary file 1b ) . Co-IP followed by Western blotting corroborated the IP-MS data , showing preferential enrichment of PARP1 and GATAD2B with mutant and wild type SOX2 , respectively , while SMARCA4 was associated equally with both forms of SOX2 ( Figure 6B ) . Examination of the protein complexes enriched by either wild type or S248A SOX2 showed a subset of components behaved discordantly with the rest of the complex subunits . For example , MBD3 and MTA3 , both of which can be a part of the NuRD complex , were consistently enriched in fS248A-Tg co-IPs while other NuRD components were enriched with fSOX2-Tg ( Figure 6B ) . To more thoroughly investigate the subunit distribution of a subset of the NuRD complex , we used a targeted proteomic approach based on interacting proteins from an MBD3 co-IP experiment . We performed anti-FLAG affinity purifications in FLAG tagged MBD3 mESCs ( Yildirim et al . , 2011 ) followed by LC-MS to generate a representative set of NuRD complex peptides . The top two , best scoring , unique peptides for each NuRD component were used to determine the relative enrichment of these proteins from fSOX2-Tg and fS248A-Tg co-IPs ( Supplementary file 1c ) . Targeted analysis showed the majority of the NuRD complex preferentially associated with SOX2WT , while MBD3 and MTA3 components prefer SOX2S248A ( Figure 6C ) . These results suggest that the S248A mutation can affect the stoichiometry of subunits in complexes that associate with SOX2 . The altered PPIs with SOX2S248A may occur as a direct result of the lack of O-GlcNAcylation of S248 . To test whether the O-GlcNAcylation of SOX2 was directly responsible for alterations in a PPI , we used recombinant proteins to assess the effect of this PTM on the SOX2-PARP1 interaction . O-GlcNAcylation of Bio-tagged , recombinant human SOX2 ( Bio-SOX2 , 96% identical to mouse ) by recombinant human OGT ( His-OGT , 99% identical to mouse ) , which depended on the sugar donor UDP-GlcNAc , was detected by Western blotting ( Figure 6D ) and specificity confirmed by mass spec ( Figure 6—figure supplement 2 ) . Bio-SOX2 was bound to streptavidin magnetic beads to remove OGT and UDP-GlcNAc . Beads bound by O-GlcNAcylated or unmodified SOX2 were incubated with GST-tagged , recombinant human PARP1 ( GST-PARP1 , 91% identical to mouse ) ( Figure 6E ) . Pull down efficiency of Bio-SOX2 was not affected by O-GlcNAcylation and His-OGT was not detected in pull downs , indicating any potential SOX2:OGT interaction was not stable under our wash conditions . Unmodified Bio-SOX2 pulled down GST-PARP1 , indicating the interaction between mouse SOX2 and PARP1 can be recapitulated by their conserved human homologues . Pulldown efficiency of GST-PARP1 by glycosylated Bio-SOX2 was diminished compared to that of unmodified Bio-SOX2 ( Figure 6D ) . Together , these data demonstrate SOX2 O-GlcNAcylation directly alters its interaction with a transcriptional regulatory protein involved in maintaining the balance of self-renewal and differentiation ( Figure 7 ) . 10 . 7554/eLife . 10647 . 024Figure 7 . Model for the role of O-GlcNAcylation in regulation of SOX2 in mESCs . ( A ) O-GlcNAc ( sugar moiety ) affects the affinity of SOX2 ( red ) for interacting proteins ( ovals ) . Some proteins ( blue shapes ) exhibit greater affinity for unmodified SOX2 , while others exhibit lower affinity ( orange shapes ) . In addition , O-GlcNAcylation affects SOX2 binding to a subset of target DNA sequences . ( B ) As a result of altered genomic distribution and protein-protein interactions when SOX2 cannot be O-GlcNAcylated ( SOX2S248A ) , pluripotency gene expression is promoted at the expense of differentiation . DOI: http://dx . doi . org/10 . 7554/eLife . 10647 . 024 Depletion of OGT , the sole enzyme that mediates intracellular O-GlcNAcylation , disrupts mESC self-renewal ( O'Donnell et al . , 2004; Shafi et al . , 2000 ) , prompting us to identify OGT targets to elucidate link between O-GlcNAc and self-renewal . Using an unbiased strategy for enrichment of native O-GlcNAcylated nuclear peptides , we previously identified SOX2 S248 as an OGT substrate ( Myers et al . , 2011 ) . Here , we find that S248 is O-GlcNAcylated during somatic cell reprogramming and that mutation of this residue to alanine increases reprogramming efficiency . We also find mESCs expressing SOX2S248A exhibit changes in transcription consistent with increased expression of pluripotency promoting genes at the expense of differentiation promoting genes . Together , these analyses from both mESCs and during somatic cell reprogramming reveal the S248A mutation promotes SOX2 activity , which suggests SOX2 O-GlcNAcylation is inhibitory during maintenance and establishment of pluripotency . Our data indicate that S248 O-GlcNAcylation is regulated by developmental signaling molecules , since removing LIF and adding RA to trigger differentiation resulted in a substantial decrease in this PTM . This decrease in S248 O-GlcNAcylation appears contradictory to the finding that the S248A mutation , which eliminates O-GlcNAcylation , promotes mESC self-renewal . However , the effects of this , or any , PTM are likely to be context specific , and determined by the transcription factors and signaling molecules present in each cell type . GlcNAc-S248 may inhibit SOX2 activity in mESCs , where the decrease in this PTM may alter SOX2 activity upon differentiation , such that SOX2 functions appropriately for the changing cellular context . As this work shows , use of methods that allow analysis of SOX2 PTM-specific PPIs and genomic occupancy may be crucial to understand how combinations of PTMs are used to regulate SOX2 activity in response to developmental cues . In addition to changes in gene expression , the S248A mutation altered SOX2 genomic distribution . As well as occupying the same sites as SOX2WT , SOX2S248A was found at an additional 1000 sites . The majority of these sites contained a predicted SOX2 binding motif , indicating the mutation allows SOX2S248A to occupy sites that SOX2WT is unable to access in mESCs . This result suggests O-GlcNAcylation can regulate the affinity of SOX2 for its target sites . Since the mutation lies in the TAD , but affects the activity of the high mobility group DNA binding region , the mutation or loss of O-GlcNAcylation could affect secondary protein structure and/or PPIs . Single molecule imaging of SOX2 with a deleted TAD showed altered DNA occupancy and higher site-specific residence time , supporting the idea that TAD mutations affect SOX2 DNA binding ( Chen et al . , 2014 ) . Upon OCT4 knockdown , fS248A-Tg mESCs did not exhibit as dramatic a change in colony morphology as fSOX2-Tg mESCs . In addition , the new sites of genomic occupancy seen in fS248A-Tg mESCs did not have nearby predicted OCT4 motifs . Together these results suggest the O-GlcNAc-deficient SOX2 may exhibit an altered reliance on OCT4 for binding and regulation of gene expression . Further studies aimed at querying the effects of the S248A mutation on genome-wide OCT4 distribution should provide insight into whether this SOX2 PTM alters OCT4 association with target sequences , and may address if O-GlcNAc impacts OCT4/SOX2 heterodimerization . Our proteomic analyses indicated there are substantial differences in SOX2-centered PPIs in fS248A-Tg mESCs . Many of the proteins that exhibited differential interaction between wild type and mutant SOX2 are components of complexes implicated in chromatin regulation . These altered associations may underlie the transcriptional and genomic occupancy changes seen in the mutant mESCs . In addition , components of the PARP-XRCC and the DNA mismatch repair ( MMR ) complexes were enriched with SOX2S248A , suggesting the possibility that these complexes may function in transcriptional regulation in addition to DNA repair . Consistent with this hypothesis , DNA damage complexes promote mESC self-renewal and iPSC generation ( Fong et al . , 2011 ) . Our PARP1 results are consistent with previous reports describing its interaction with SOX2 in mESCs ( Gao et al . , 2009; Lai et al . , 2012 ) . In both these studies , differentiation of mESCs promotes the PARP1-SOX2 interaction . We found PARP1 interaction with SOX2 is disrupted by O-GlcNAc and decreased GlcNAc-S248 in differentiating ESCs , consistent with a model in which the developmentally regulated decrease in S248 O-GlcNAcylation promotes PARP1 interaction . During differentiation , the SOX2-PARP1 interaction inhibits SOX2 binding to enhancers of genes that are necessary for self-renewal ( Lai et al . , 2012 ) . We find that SOX2S248A exhibits increased association with PARP1 without decreased occupancy of pluripotency genes under self-renewal conditions . This result indicates the SOX2-PARP1 interaction alone does not inhibit SOX2 binding to pluripotency gene targets and suggests additional developmentally regulated alterations in SOX2 PTMs or interaction partners contribute to control of SOX2 occupancy . The increase in SOX2 S253 phosphorylation in fS248A-Tg mESCs suggests there is potential for crosstalk between O-GlcNAcylation and phosphorylation in the TAD peptide in mESCs . In general , phosphorylation and O-GlcNAcylation both occur in structurally flexible regions of proteins , although there is no correlation or anti-correlation in the linear proximity of one PTM to another ( Trinidad et al . , 2012 ) . It is unlikely phospho-S253 has a substantial impact on pluripotency gene expression , as S253 phosphorylation is dispensable for mESC self-renewal ( Ouyang et al . , 2015 ) . However , because the TAD of SOX2 is an unstructured domain ( Reményi et al . , 2003 ) , and is O-GlcNAcylated and phosphorylated , the molecular basis for the potential crosstalk may be specific to a different cellular context . O-GlcNAc signaling is essential for pluripotency . However , reliable O-GlcNAc site identification , as well as investigation into this PTM’s function , is in their infancy . This study shows that although global O-GlcNAcylation is necessary for mESC self-renewal , a key pillar of pluripotency can be inhibited by O-GlcNAc modification . This work provides a new mechanism for the regulation of SOX2 through O-GlcNAcylation , and illustrates the role of this PTM in pluripotency and self-renewal is more complex than previously appreciated . 2TS22C mESCs ( accession number AES0125 ) , the parental cell line for derivation of tagged SOX2 lines ( Masui et al . , 2007 ) , were obtained through Riken BioResource Center . Identity of this line , which contains a SOX2 transgene under control of a tet-repressible promoter , was authenticated by culturing cells with and without doxycycline and examining SOX2 expression by Western blotting . To create the fSOX2-Tg ESC line , 4 ug of the plasmid CAG-3xF-Sox2 was transfected with Lipofectamine 2000 ( Thermo Fisher [Invitrogen] , Waltham , MA ) into a 6-well plate containing 2TS22C cells . 2TS22C mouse ESCs express SOX2 under control of a tetracycline-repressible system . Twenty-four hours after transfection 1 ug/mL doxycycline was added to silence expression of the TetO Sox2 . Forty-eight hours after transfection , 7 ug/mL puromycin was added , to select for the integration and expression of CAG-3xF-Sox2 . After about two weeks , colonies exhibiting the typical ESC morphology were expanded and tested via western blot , morphology and alkaline phosphatase staining ( Clontech , Mountain View , CA ) . The same strategy was used to generate the fS248A-Tg and haSOX2-Tg ESC lines . Mouse ESC lines were routinely passaged by standard methods in ESC media ( KO-DMEM , 10% FBS , 2 mM glutamine , 1X non-essential amino acids , 1x pencillin/streptomycin , 0 . 1 mM b-mercaptoethanol and recombinant leukemia inhibitory factor ) . 2TS22C , fSOX2-Tg , fS248A-Tg , and haSOX2-Tg ESCs were cultured in ESC media with 1 ug/mL doxycycline hyclate ( Sigma ) . KI mESCs ( Lai et al . , 2012 ) were cultured in N2B27 plus 2i ( Thermo Fisher [Life Technologies] , Waltham , MA ) and LIF or 500 uM retinoic acid ( Sigma-Aldrich , St . Louis , MO ) . Mycoplasma testing was carried out every two months until the cells were found to be negative for several successive tests . CAG-3xF-Sox2 was constructed by inserting mouse Sox2 cDNA into pCMV-3xFLAG 7 . 1 ( Sigma-Aldrich ) and then subcloning 3xF-Sox2 via InFusion cloning ( Clonetech ) into CAG-HA-Sox2-IP ( Addgene plasmid 13459 ) , replacing the HA tag with the triply FLAG tagged . S248A mutations in CAG-3xF-Sox2 were performed with Quickchange site directed mutagenesis ( Agilent Technologies [Stratagene] , Santa Clara , CA ) ( Supp Table S4 ) . pMXs-mouse Sox2 wild type or S248A , along with pMXs-Oct4 , Klf4 and c-Myc were used ( Okita et al . , 2007; Takahashi and Yamanaka , 2006 ) . The pMXs-3xF-Sox2 was created by cloning Sox2 into pCMV-3xFLAG 7 . 1 ( Sigma-Aldrich ) and then subcloning 3xF-Sox2 via InFusion cloning ( Clonetech ) into pMXs . Antibodies were purchased from: Abcam , United Kingdom: SOX2 ab75179 ( immunofluorescence [IF] and in vitro assay WB , TUBULIN GTU-88 ab11316 ( Western blot [WB] ) , HA ab13834 ( WB , IP ) , OCT4 ab19857 ( IF ) Reprocell USA INC , Boston , MA: NANOG , RCAB002P-F ( IF ) SIGMA-Aldrich: OGT DM-17 ( WB ) , FLAG A8592 ( WB ) , FLAG F1804 ( IF and ChIP ) Santa Cruz Biotechnology Inc , Santa Cruz , CA: OGT SC32921 ( in vitro assay WB ) , OCT4 SC8628x ( WB ) Bethyl Laboratories , Inc , Montgomery , TX: SMARCA4 A300-813A ( WB ) , PARP1 A301-375A ( WB ) , GATAD2B 301-281A ( WB ) Pierce/ThermoFisher: O-GlcNAc MA1072 ( WB ) ActiveMotif , Carlsbad , CA: PARP1 39 , 559 ( in vitro assay WB ) Bio-Rad Laboratories , Inc , Hercules , CA: goat anti-rabbit HRP conjugate 172–1019 and goat anti-mouse HRP conjugate 172–1011 ( WB ) Jackson Laboratory , Bar Harbor , ME: Alexa Fluor 488-donkey anti-rabbit IgG 711-545-152 ( IF ) Alkaline phosphatase activity staining was performed according to manufacturer's instructions ( Stemgent , Cambridge , MA , 00–0055 ) . KI mESCs , fSOX2-Tg or fS248A-Tg cells were expanded to one to three 15 cm2 dishes depending on the experiment . Cells were harvested by trypsinization , washed once with cold PBS and frozen in liquid nitrogen . Whole cell pellets were lysed in RIPA buffer without SDS , containing 500 nM Thiamet G ( Caymen Chemicals , Ann Arbor , MI ) , 1X HALT protease and phosphatase inhibitors ( Thermo Fisher[Pierce] ) , 2 mM TCEP ( Sigma ) and 20 mM N-ethylmaleimide ( Sigma-Aldrich ) and sonicated ( with a probe sonicator on methanol ice for three rounds of pulses , 3 s on , 2 off , 10 s total , at 35% ) . Anti-FLAG-based purifications were performed with anti-FLAG M2 Dynabeads ( Sigma-Aldrich , M8823 ) . Whole cell lysates were incubated with M2 beads at room temperature for 75 min , washed once with lysis buffer and three times with 25 mM ammonium bicarbonate ( ABC ) with 150 mM NaCl . Proteins were eluted with 100 mM glycine pH 4 for five minutes . Western blot and SDS-PAGE analysis was used to assess purification efficiency . For HA-SOX2 purification , anti-HA antibodies were coupled to aldehyde-coated magnetic beads ( BioClone , San Diego , CA ) via reductive amination in 20 mM bicine pH 7 . 8 and purified as above . Two 10 cm2 dishes of CD1 MEFs infected with OSFLAG-WTKM or OSFLAG-S248AKM for six days were analyzed the same way . Silver or Coomassie stained SDS-PAGE gel bands were excised and digested in-gel with sequence grade trypsin ( Roche , Switzerland ) . After 5% formic acid/50% acetonitrile extraction , peptides were dried by vacuum centrifugation , gel particulates were removed via C18 Zip Tips ( MERCK Millipore , Billercia , MA ) , dried , resuspended in 0 . 1% formic acid and analyzed by LC-MS/MS . Chromatography was performed on a Nanoacquity HPLC ( Waters , Milford , MA ) at 400 nl/min with a BEH130 C18 2 . 1x150 mm column ( Waters ) . A 90- or 120-minute gradient from 98% solvent A ( 0 . 1% formic acid ) to 22% solvent B ( 0 . 1% formic acid in acetonitrile ) was used . Peptides were analyzed by an LTQ-Orbitrap Velos mass spectrometer ( Thermo Scientific , Waltham , MA ) . After the survey scan of m/z 400–1 , 600 was measured in the Orbitrap at 30 , 000 resolution , the top three multiply charged ions were selected for both HCD and ETD . Automatic gain control for MS/MS was set to 2000 . Normalized collision energy for HCD was set at 35 while the ETD activation time was charge state dependent , based on 100 ms for doubly charge precursors . Supplemental activation was implemented for ETD reactions . Dynamic exclusion of precursor selection was set for 25 s . WT GlcNAc-S248 , WT unmodified and the S248A SOX2 TAD were synthesized by New England Peptide and analyzed as described above . Fragment mass spectra were converted into peaklists using the in-house software PAVA . HCD and ETD data were searched separately using ProteinProspector version 5 . 10 . 0 against the UniProt database with a concatenated database . Only mouse and human genomes were used for the database searching . Precursor tolerance was set to 10 ppm , whereas fragment mass error tolerance was set to 0 . 6 Da for ETD and 20 ppm for HCD . N-terminal acetylation , methionine oxidation , loss of N-terminal methionine and glutamate conversion to pyroglutamate were allowed as variable modifications . For ETD data , HexNAc modifications to serine and threonine residues and phosphorylation to serine/threonine/tyrosine was allowed as variable mass modifications . For HCD , phosphorylation was searched the same way though HexNAc was considered as a neutral loss . Methylation ( mono , di- and tri- ) of K and R , monomethylation of D , E and H ( artifact from MeOH fixing PAGE gels ) , acetylation of K and R , and ADP-ribosylation to C , E , K , N , S , and R were searched separately . SLIP scoring was used to distinguish possible positional isomers of HexNAc and/or phosphopeptides ( Baker et al . , 2011 ) . Relative abundances of each modified or unmodified peptide were calculated using the ICIS area calculated from XICs in Xcalibur ( Thermo Scientific ) at a 10 ppm mass tolerance . Six to ten 10 cm2 plates worth of haSOX2-Tg , fSOX2-Tg or fS248A-Tg cells were harvested for nuclear extract preparations in biological duplicate . Nuclear extracts were as previously described with minor modifications ( Dignam et al . , 1983 ) . Buffers A , C and D were supplemented with 2 µM Thiamet G , 2 µM PUGNAc ( Tocris Bioscience , United Kingdom ) and 1X HALT protease and phosphatase inhibitors and instead of dialysis of extracts , two volumes of buffer D were used to dilute salt concentration . Seven µL of M2 beads per 10 cm2 plate were used per co-IP and samples were nutated at 4oC for two hours . Beads were washed once with Buffer D plus inhibitors , then twice with 50 mM ABC with 150 mM NaCl . Each wash was only as long as it took to transfer the beads to a new , cold tube and place on the magnetic rack . Beads were then resuspended in 50 µL 100 mM ABC with 500 ng trypsin and shaken at 37oC for one hour . Supernatant was transferred to a new tube , the beads were washed once with 50 µL ABC and combined to digest overnight . Digestions were desalted with one or two Zip Tips , depending on size of experiment , and dried via vacuum centrifugation . IP-WB experiments were performed similarly except proteins were eluted with 2X SDS-PAGE loading buffer without reducing agent . Chromatography was performed on a Nanoacquity HPLC ( Waters ) at 400 nl/min with an EASY-spray 15 cm x 75 µm ID , PepMap C18 , 3 µm column ( Thermo Scientific ) . A 90-minute gradient from 98% solvent A ( 0 . 1% formic acid ) to 22% solvent B ( 0 . 1% formic acid in acetonitrile ) was used . Peptides were analyzed on a Q-Exactive Plus mass spectrometer ( Thermo Scientific ) . After the survey scan of m/z 400–1 , 600 was measured in the Orbitrap at 70 , 000 resolution , the top ten multiply charged ions were selected for HCD and measured at 17 , 500 resolution . Normalized collision energy for HCD was set at 35 , Dynamic exclusion of precursor selection was set for 25 s . The label-free quantiation ( LFQ ) feature of MaxQuant ( 1 . 5 . 1 . 0 ) was used to quantify protein signals for proteins identified in the co-IP experiments . For SOX2 , only non-TAD peptides were used for protein level quantitation . Proteins were determined to be SOX2 interactors by taking the average ratio of the LFQ intensity of the protein of interest ( POI ) from the FLAG-tagged mESC lines over the HA-tagged mEScs . This average ratio was log2-transformed , and was normalized by the global median . If the POI was two standard deviations from the mean it was considered specific to SOX2 . Most proteins discussed were manually verified using Skyline . To determine POIs that differentially interact with wild type or S248A SOX2 , we took the ratio of the log2-transformed ratio of FLAG/HA for S248A over the wild type . After normalization by the global median , POIs were considered to be differential interactors if they had a z-score of greater than 1 . 5 . All of these peptides were manually verified using Skyline . For targeted analyses , FLAG-tagged MBD3 mESCs , a generous gift from the Fazzio laboratory ( Yildirim et al . , 2011 ) were analyzed as described above . From these analyses the top two most intense , non-homologous peptides identified from NuRD subunits were used to monitor their co-purification with SOX2 isoforms in fSOX2-Tg and fS248A-Tg cells in three separate , biological replicates . XICs were extracted manually using Xcalibur software . Relative enrichment of proteins co-purified with either SOX2 form was determined as described above . Peptides used for this analysis are listed in Supplementary Table 1b . Samples for co-IP Western blot corroboration of LC-MS data were performed as described above in biological duplicate . Antibodies used are described above . MEFs were derived from wild type CD1 mice or the Nanog-Gfp-IRES-Puror mice ( Okita et al . , 2007 ) and cultured in MEF media ( KO-DMEM , 10% FBS , 2 mM glutamine , 1X non-essential amino acids , 1X pencillin/streptomycin , and 0 . 1 mM β-mercaptoethanol ) . pMXs vectors containing Oct4 , c-Myc , Klf4 , eGfp , dsRed , or wild type or S248A Sox2 , with or without FLAG , were transfected with Fugene 6 ( Promega , Madison , WI , E2691 ) into PlatE cells . Twenty-four hours after transfection , the media was changed . The next day , the retroviral supernatant was collected from transfected PlatE cells , filtered through 0 . 4 um filters and combined with each other at equal ratios . Polybrene ( Merck Millipore , TR-1003-G ) was added to a final concentration of 4 ug/ml . The virus-containing media was added to Nanog-Gfp or wild type MEFs that were passaged less than five times . Media was replaced the next day and every other until six days after transduction . At day six , MEFs were trypsinized and either prepared for experiments or 1000 cells were plated onto γ-irradiated SNL feeders . These 1000 MEFs were cultured in ESC media until GFP+ colonies were counted at day 20 ( Nanog-Gfp MEFs ) . Microinjection of iPSCs to generate chimera mice was conducted at Cornell University Stem Cell and transgenic core facility . iPSCs were grown on mouse embryonic fibroblasts ( produced at the Cornell stem cell core ) and mitotically inactivated by irradiation ( 3000 Rads ) . To produce donor embryos , wild type albino mice of the strain http://jaxmice . jax . org/strain/000058 . html were mated , embryos were flushed from the uterus at day 3 . 5 , and the iPSCs were injected into the blastocyst of each embryo ( 15–30 cells per embryo ) . Injected embryos were then transferred to 2 . 5-day pseudo pregnant recipient animals and pup chimaerism was determined by coat color . Chimeras were mated to age-matched wild type animals of the same albino strain used for embryo donors . iPSC contribution to the germline was determined by coat color of the resultant pups . Total RNA was extracted with Trizol ( Thermo Fisher [Invitrogen] ) according to manufacturer’s instructions . Arraystar Inc , Rockville , MD ( http://www . arraystar . com ) prepped and hybridized the samples , and performed the data analysis . For RT-qPCR , 1 µg total RNA was reverse transcribed to cDNA with iScript ( Bio-Rad Laboratories ) , diluted 1:20 or 1:50 , depending on the abundance of the transcript , and 4 µL was used . Quantitative PCR was performed on a CFX Connect Real-time PCR detection system ( Bio-Rad laboratories ) with SensiFast SYBR Lo-ROX PCR master mix ( Bioline , Taunton , MA , BIO-94020 ) . Fold enrichment was determined by 2- ( ΔCq ) method ( ΔCq= Cq ( gene ) -Cq ( GusB ) ) . Primers are listed in Supplementary file 1d . The recombinant poly-His tagged human OGT ( His-OGT ) expression plasmid was a generous gift from Suzanne Walker . His-OGT was expressed as previously described ( Gross et al . , 2005 ) . His-OGT was purified by Ni-NTA agarose resin ( Qiagen , Germany ) , eluted and buffer exchanged into 50 mM Tris-HCl , pH 7 . 8 , 300 mM NaCl . To generate a biotinylated SOX2 ( Bio-SOX2 ) , the human SOX2 cDNA was cloned into the expression vector , pGV358avi , as a fusion construct linked to an N-terminal Avi-tag and a C-terminal intein-chitin binding domain ( Redding et al . , 2015 ) . Bio-SOX2 was produced in E . coli BL21 ( DE3 ) in media supplemented with 200 nM biotin . The protein was purified from a clarified lysate by passage over a chitin column ( New England Biolabs ( NEB ) , Ipswich , MA ) and incubated overnight in buffer containing 20 mM Tris-HCl , pH 8 . 5 , 500 mM NaCl , 1 mM EDTA , 10% glycerol and 50 mM DTT . Bio-SOX2 was then eluted from the chitin column and dialyzed into storage buffer ( 20 mM Tris-HCl , pH 8 . 5 , 500 mM NaCl , 1 mM EDTA , 10% glycerol and 1 mM DTT ) for long term storage at -80oC . Equimolar ratios of Bio-SOX2 and His-OGT were incubated in an O-GlcNAc assay buffer ( 50 mM Tris pH 7 . 4 , 12 . 5 mM MgCl2 , 2% glycerol , 0 . 2 mM PMSF , 1 mM DTT ) for one hour at 37oC with or without the sugar donor , UDP-GlcNAc ( 100 µM ) . After one hour , 50 µl of streptavidin magnetic beads ( Thermo Fisher , 11205D ) were resuspended in PBS/BSA ( 0 . 1% ) buffer were added and the reactions were incubated for 45 min at 4oC . Bio-SOX2 bound beads were washed twice with PBS/BSA , once with low salt buffer ( PBS/BSA+150 mM NaCl ) and once with high salt buffer ( PBS/BSA+300 mM NaCl ) to eliminate His-OGT and unincorporated UDP-GlcNAc . Finally , the beads were resuspended in 50 µl of PBS/BSA buffer and incubated at 4oC for 1 hr with recombinant GST-PARP1 protein ( 0 . 14 µM ) ( Sigma-Aldrich , SRP0192 ) . The beads were washed three times in PBS/BSA buffer and eluted in 1X SDS-PAGE loading buffer . Pull down and input samples were resolved on a 10% SDS-PAGE gel and Western blotted for SOX2 , ( Abcam , ab75179 ) , OGT ( SantaCruz Biotechnology Inc , SC 32921 ) O-GlcNAc ( Thermo Fisher [Pierce , MA1072 ) and PARP1 ( Active Motif , AM 39559 ) . siRNA pools ( esiRNAs ) were created by in vitro cleavage of double-stranded RNA ( Fazzio et al . , 2008 ) . In vitro transcription templates for double stranded RNAs were generated using the primers acquired from the Riddle database ( Kittler et al . , 2007 ) . Transfections were performed using Lipofectamine RNAiMax according to manufacturer's instructions . The cells from ∼80% confluent 10 cm2 dishes were crosslinked with 1% formaldehyde for 10 min at room temperature . After quenching with 125 mM glycine for 5 min , the cells were washed twice with ice cold PBS . Cell pellets were resuspended in 1ml lysis buffer 1 ( 50 mM HEPES-KOH , pH 7 . 6 , 140 mM NaCl , 1 mM EDTA , 10% ( v/v ) Glycerol , 0 . 5% NP-40 , 0 . 25% Triton X-100 , complete protease inhibitor cocktail ( PIC ) ( Roche , 11697498001 ) for 10 min at 4oC , followed by centrifugation at 1 , 350 x g for 5 min at 4oC . Discard the supernatant and resuspend the pellet in 1 ml of lysis buffer 2 ( 10 mM Tris-HCl pH 8 . 0 , 200 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , PIC ) for 10 min in at 4oC , followed by centrifugation at 1 , 350 x g for 5 min at 4 oC . Finally resuspend the cell pellet in 700 µl of lysis buffer 3 ( 10 mM Tris-HCl pH 8 . 0 , 100 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 0 . 1% sodium deoxycholate , 0 . 5% N-Lauroylsarcosine , 1X PIC ) and sonicate to desired length using Bioruptor ( UCD-200; Diagenode , Denville , NJ ) for 30 min [10 min x 3 times ( each cycle consists of 30 s on , 30 s off ) at high settings] . Clarify the lysates by centrifugation at 10 , 000 rpm in a microcentrifuge for 10 min at 4oC . Transfer the sonicated chromatin to a new tube and store at -80oC or use it immediately for chromatin immunoprecipitation ( ChIP ) . We used two sets of chromatin ( long and short ) for ChIP and library preparation . Long ChIP samples were generated using sonicated chromatin consisting of DNA fragments ranging from sizes 1 kb–500 bp . Short ChIP samples were generated using sonicated chromatin consisting of DNA fragments ranging from sizes 500–200 bp . First , FLAG-coupled protein G beads were generated by incubating 10 ug of FLAG antibody ( Sigma-Aldrich , F1804 ) with 50 μl of prewashed protein G magnetic beads ( NEB S1430S ) in PBS/BSA ( 5 mg/ml ) buffer at 4oC , overnight . Chromatin Immunoprecipitation is performed by incubating these FLAG-coupled protein G beads with the sonicated chromatin in ChIP buffer ( 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 2 mM EDTA , 1% Triton X-100 ) at 4oC , overnight . 10% of the total chromatin used for the ChIP is set aside as input sample and stored at -20°C until further use . The magnetic beads were washed twice with ChIP buffer , once with ChIP buffer containing 500 mM NaCl , four times with RIPA buffer ( 10 mM Tris-HCl pH 8 . 0 , 0 . 25 M LiCl , 1 mM EDTA , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate ) , and once with TE buffer ( 10 mM Tris-HCl , pH 8 . 0 , 1 mM EDTA ) . Finally elute DNA from the beads by adding 100 µl of elution buffer ( 20 mM Tris-HCl pH 8 . 0 , 100 mM NaCl , 20 mM EDTA , 1% SDS ) twice and incubating for 15 min at 65°C . The eluted DNA and the input samples were then reverse crosslinked at 65oC for overnight , followed by RNase A ( 0 . 2 mg/ml ) digestion at 37oC for 2 h and Proteinase K ( 0 . 2 mg/ml ) ( NEB , P8102S ) digestion at 55oC for 1 hr . The ChIP and input DNA were recovered by phenol-chloroform extraction and ethanol precipitation . ChIP samples were end-repaired , A-tailed and adaptor ligated using barcode adaptors . Briefly , DNA was end-repaired using a combination of T4 DNA polymerase , E . coli DNA Pol I large fragment ( Klenow polymerase ) and T4 polynucleotide kinase . The blunt , phosphorylated ends were treated with Klenow fragment ( exo minus ) and dATP to yield a protruding 3- 'A' base for ligation of barcoded adapters which have a single 'T' base overhang at the 3’ end . DNA purification on Qiagen mini elute columns was performed following each enzyme reactions . The adaptor ligated material was then PCR amplified with Phusion polymerase using 16 cycles of PCR before size selection of 200–350 bp fragments on a 2% agarose gel . The ChIP-seq library samples were purified using Qiagen gel extraction kit , and its concentration was determined on Agilent Bioanalyzer using High Sensitivity DNA chip ( Agilent Technologies , Santa Clara , CA ) . Libraries with different barcodes were multiplexed together at equimolar concentrations and single-end sequencing ( 50 bp ) was performed at Center for Advanced Technology , genomics core facility , UCSF . Each lane of the HiSeq 2000 ( Illumina , San Diego , CA ) had five libraries ( four ChIP DNA and one input DNA ) multiplexed with barcodes added to the 5' end of the sequence . Reads were identified and then mapped to the mm9 assembly of the mouse genome using the Bowtie aligner ( Langmead and Salzberg , 2012 ) . Normalized and background-corrected measures of ChIP signal were created by randomly choosing 10 million unique tags from each dataset , calculating the tag density within 75 bp of each 20 bp bin of the mm9 assembly mouse genome , and then subtracting the matched 10-million-tag-normalized input tag density from each dataset . UCSC genome browser was used to visualize the ChIPseq peaks . All ChIP-seq peaks present in both the long and short biological replicates were used for motif analysis . Motifs were identified by running MEME ( Bailey and Elkan , 1994 ) on fS248A-Tg mESC peaks ( -dna -mod zoops -nmotifs 3 -minw 6 -maxw 30 -time 6058 -revcomp -maxsize 1000000 ) . De novo motifs were matched to known motifs using Tomtom ( Gupta et al . , 2007 ) . The ChIP-seq and microarray data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE69594 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE 69594 ) . Cells were stained for AP activity as previously described above . NIS-Elements Basic Research software automatically acquired the darkness of the staining . Four fields of view for each experiment was taken and averaged to plot the change in darkness between gene of interest knockdown and Gfp .	Embryos develop from stem cells , which have the ability to mature into any type of cell in the body . The activity of proteins called transcription factors determines whether a stem cell will become a specialized cell type or remain in an immature “pluripotent” state that has the potential to become any cell type . These transcription factors bind to the cell’s DNA to regulate the activity of target genes . SOX2 is a transcription factor that helps to maintain embryonic stem cells in a pluripotent state . In 2011 , a group of researchers showed that a specific sugar molecule was added to SOX2 in mouse embryonic stem cells , in a process called O-GlcNAcylation . Now , Myers , Peddada et al . – including the researchers who performed the 2011 study – have studied the effects of this SOX2 modification in more detail . Transcription factors have two major activities – they bind to DNA and recruit other proteins that can turn target genes on or off . Myers , Peddada et al . found that , in pluripotent stem cells , a complex pattern of O-GlcNAcylation is present on SOX2 in a region that is responsible for recruiting other proteins . In addition , SOX2 O-GlcNAcylation decreases when stem cells are directed to become a new cell type . Further experiments investigated gene activity in stem cells that contained a mutant form of SOX2 that cannot be O-GlcNAc modified . In these cells , genes that help to maintain the cell in a pluripotent state were more active than in normal cells . The mutant form of SOX2 was altered in its ability to bind DNA and to associate with proteins that control gene activity . Myers , Peddada et al . ’s findings raise several questions . Does O-GlcNAcylation control the activity of SOX2 in other cell types , such as neurons and cancer cells , in which this modification can be detected on SOX2 ? Why does a modification on the portion of the SOX2 that is thought to interact with other proteins affect SOX2 DNA binding activity ? Finally , understanding how O-GlcNAcylation is employed to regulate SOX2 activity in response to developmental cues remains a major challenge .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "biochemistry", "and", "chemical", "biology" ]	2016	SOX2 O-GlcNAcylation alters its protein-protein interactions and genomic occupancy to modulate gene expression in pluripotent cells
Transcriptional gene silencing ( TGS ) can serve as an innate immunity against invading DNA viruses throughout Eukaryotes . Geminivirus code for TrAP protein to suppress the TGS pathway . Here , we identified an Arabidopsis H3K9me2 histone methyltransferase , Su ( var ) 3-9 homolog 4/Kryptonite ( SUVH4/KYP ) , as a bona fide cellular target of TrAP . TrAP interacts with the catalytic domain of KYP and inhibits its activity in vitro . TrAP elicits developmental anomalies phenocopying several TGS mutants , reduces the repressive H3K9me2 mark and CHH DNA methylation , and reactivates numerous endogenous KYP-repressed loci in vivo . Moreover , KYP binds to the viral chromatin and controls its methylation to combat virus infection . Notably , kyp mutants support systemic infection of TrAP-deficient Geminivirus . We conclude that TrAP attenuates the TGS of the viral chromatin by inhibiting KYP activity to evade host surveillance . These findings provide new insight on the molecular arms race between host antiviral defense and virus counter defense at an epigenetic level . RNA silencing is a host defense mechanism to combat invading nucleic acids . One type of RNA silencing is referred to as post-transcriptional gene silencing ( PTGS ) . In PTGS , double-stranded RNAs ( dsRNAs ) are processed by Dicer-like ribonucleases into small-interfering RNAs ( siRNAs ) . Mature siRNAs are incorporated into an Argonaute ( AGO ) -centered RNA-induced silencing complex ( RISC ) to regulate expression of target genes through RNA cleavage or translational repression . PTGS has evolved as a universal defense response toward all viruses because dsRNAs can result from intermediates in RNA virus replication , highly structured RNA virus genomes , or from viral transcripts . To evade this surveillance mechanism , virtually all plant viruses are known to encode suppressor proteins that are able to block different key steps of the PTGS pathway ( Ding and Voinnet , 2007 ) . While the host/virus battle at the PTGS level has been well appreciated , virus suppression at a transcriptional gene silencing ( TGS ) level is poorly understood . In eukaryotes , the nuclear DNA is wrapped onto histone octamers to constitute chromatin . The chromatin undergoes various DNA and histone methylations , and these modifications have variable effects on gene expression depending on the precise residues , contexts , and modification complexity . Histone methylation takes place on lysine and arginine residues of the amino-terminal tails ( Kouzarides , 2007; Greer and Shi , 2012 ) . The prevailing dogma is that histone 3 lysine 4 tri-methylation ( H3K4me3 ) is mostly associated with transcriptionally active euchromatin , while H3K9me2 and H3K27me3 are repressive marks ( Deal and Henikoff , 2011; Feng and Jacobsen , 2011 ) . Histone methylation is catalyzed by SET domain containing methyltransferases , specifically , H3K9me2 is deposited by Arabidopsis Su ( var ) 3-9 homolog 4 , Kryptonite ( KYP ) ( Du et al . , 2014a ) , and its paralogs ( SUVH5 , 6 ) , while H3K27 methylation is carried out by the Polycomb repressive complex 2 ( PRC2 ) , which includes Curly Leaf ( CLF ) ( Liu et al . , 2010; Zheng and Chen , 2011 ) . Local H3K9me2 and H3K27me3 can spread over wide regions to elicit heterochromatin configuration . In animals , the propagation of histone methylation entails co-repressor heterochromatin protein 1 ( HP1 ) , whereas in plants , KYP acts synergistically with DNA methyltransferases ( i . e . , Chromomethylase 3 [CMT3] ) to constitute a mutually reinforcing cycle of DNA and histone methylation to secure TGS ( Du et al . , 2012 , 2014a ) . Histone methylation not only regulates endogenous gene expression but also invasive DNAs such as transposons and viruses ( Narasipura et al . , 2014 ) . Plant DNA viruses , exemplified by Geminivirus , form minichromosomes in the host ( Hanley-Bowdoin et al . , 2013 ) . Both Geminivirus DNA and associated histones are methylated in infected cells , whereas viral methylation is reduced in methylation-deficient hosts , methylation-compromised Arabidopsis mutants are hypersusceptiple to Geminivirus infection and show exacerbated disease symptoms ( Raja et al . , 2008 ) . Thus , plants appear to employ methylation of viral chromatin to limit viral replication and transcription ( Aregger et al . , 2012; Pumplin and Voinnet , 2013 ) . On the other hand , Geminiviruses encode a multi-functional protein called transcriptional activation protein ( TrAP/AL2/AC2 ) that counters the epigenetic defense ( Raja et al . , 2008; Buchmann et al . , 2009 ) . It has been shown that TrAP inhibits adenosine kinase ( ADK ) ( Wang et al . , 2005 ) . ADK catalyzes the synthesis of 5′ AMP from adenosine and ATP , a process that promotes the regeneration of S-adenosyl-methionine ( SAM ) , the major methyl donor in the cell ( Moffatt et al . , 2002; Buchmann et al . , 2009 ) . Consequently , the TrAP-mediated inhibition of ADK activity likely impedes downstream trans-methylation events , including viral chromatin methylation in the nucleus ( Bisaro , 2006; Buchmann et al . , 2009 ) . In parallel , some Geminivirus encode a TrAP positional homolog , named C2 , that is able to stabilize SAM decarboxylase 1 to downregulate the methyl group metabolism ( Zhang et al . , 2011 ) . It seems that interfering with the methyl cycle is a common suppression mechanism for Geminivirus-encoded TrAP/AL2/C2 proteins . In addition , C2 also subverts the activity of COP9 signalosome to inhibit jasmonate signaling ( Lozano-Durán et al . , 2011 ) , suggesting its multiple functions in viral counter-defense . Here , we investigated the suppression mechanism of TrAP proteins , encoded by two Geminivirus members , Tomato Golden Mosaic Virus ( TGMV ) and Cabbage Leaf Curl Virus ( CaLCuV ) . We found that constitutive expression of TGMV-TrAP in Arabidopsis thaliana caused morphological abnormalities that mimic loss-of-function mutants of numerous TGS components including lhp1 ( like-heterochromatin1 ) and clf . Microarray analyses of TrAP transgenic plants and lhp1 mutants revealed a substantial overlap in reprogrammed host genes at a genome-wide level . Through biochemical screening , we identified KYP as the bona fide target of TrAP . We demonstrated in vitro that TrAP binds to the catalytic domain of KYP and inhibits its enzymatic activity; while in vivo , TrAP decreases the repressive H3K9me2 marks and H3K9me2-dependent CHH methylation in gene-rich regions . We also found that KYP directly associates with the Geminivirus minichromosome and deposits H3K9me2 marks on viral chromatin . In addition , kyp mutants but not wild-type plants sustain low systemic infection of CaLCuV lacking TrAP protein . Taken together , we propose that KYP-catalyzed H3K9me2 is a line of the innate immunity against invading DNA pathogens , and Geminivirus TrAP functions to inactivate KYP to counter host defense . Thus , this study provides new insight into the host–virus interaction at the TGS level . To study the suppression mechanism of TrAP , we generated 235 Arabidopsis transgenic lines overexpressing full-length TGMV TrAP , with or without Flag-Myc4 , 3HA , or CFP epitopes . These transgenic plants were confirmed by northern ( data not shown ) or western blot assays ( Figure 1A and Figure 3—figure supplement 1A ) . Importantly , the majority of the transgenic lines exhibited developmental abnormalities consisting of short statues , strongly upward curled cotyledons and true leaves ( Figure 1B ) . Moreover , these overexpressing lines exhibited early flowering compared to wild-type ( WT ) plants . These phenotypes were morphologically distinct from loss-of-function mutants of ADK1 ADK2 ( Weretilnyk et al . , 2001; Moffatt et al . , 2002 ) , SnRK1 ( Shen et al . , 2009 , 2014 ) , PEAPOD2 ( Lacatus and Sunter , 2009 ) , and rgsCaM ( Chung et al . , 2014 ) , a calmodulin-like protein , which are also targets or partners of TrAP . This result indicated that TrAP exerts some novel cellular function ( s ) . 10 . 7554/eLife . 06671 . 003Figure 1 . TrAP caused developmental abnormalities in Arabidopsis but not through miRNA pathway . ( A ) Western blot analysis of 35S-TrAP-3HA and 35S-TrAP-CFP transgenic lines . Arrows indicate the locations of the tagged TrAP proteins; * Cross-reaction band serves a loading control . ( B ) Morphological defects of Arabidopsis transgenic plants expressing 35S-TGMV TrAP . Photographs were taken of 10-day seedlings . ( C ) RNA blot analysis of PHB and AGO1 transcripts in the TrAP overexpression transgenic plants using gene-specific random-labeled probes . 25S rRNA is a loading control . ( D ) sRNA blot analysis of miRNA and miRNA* in the TrAP overexpression transgenic plants . Total RNA was prepared from a pool of T2 transgenic plants ( n > 50 for each line ) . sRNA blots were probed using 5′ end 32P-labeled oligonucleotide probes complementary to the indicated miRNA or miRNA* . 5S rRNA and tRNA are a loading control . All the samples were run in the same gel; but the lane order of miRNAs was rearranged to match that of miRNAs . ( E ) sRNA blot analysis of miR165 loading into Arabidopsis RNA-induced silencing complexes ( RISCs ) . RNA was extracted from flowers or Flag-AGO1 immunoprecipitates of transgenic plants harboring 35S-TrAP or 35S-TrAP-3HA in ago1-36; PAGO1-Flag-AGO1 background and ago1-36; PAGO1-Flag-AGO1 control plants ( Baumberger and Baulcombe , 2005 ) . Top panel , total RNA; middle panel , each lane contained sRNA associated with Flag-AGO1 immunoprecipitated from 0 . 4 g of flowers . Bottom panel , the input and immunoprecipitate of Flag-AGO1 were analyzed by Western blot assays in the same samples for sRNA blots . A cross-reacting band ( ) was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 00310 . 7554/eLife . 06671 . 004Figure 1—figure supplement 1 . sRNA blot analysis of additional miRNAs and siRNA in the TrAP overexpression transgenic plants . Total RNA was prepared from a pool of T2 transgenic plants ( n > 50 for each line ) . sRNA blots were probed using 5′ end 32P-labeled oligonucleotide probes complementary to the indicated miRNAs or siRNA . U6 serves as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 004 Developmental anomalies of transgenic plants expressing viral suppressors have been associated with interruption of the miRNA pathway . To test this , we compared expression levels of several miRNAs and their targets between Col-0 wild-type and TGMV TrAP transgenic plants . Plants expressing Cucumber mosaic virus-encoded 2b suppressor ( Zhang et al . , 2006 ) and ago1-27 , a hypomorphic allele of ago1 ( Morel et al . , 2002 ) , were used as controls . We observed that the accumulation of miR165 and miR168 and their targets , PHB and AGO1 transcripts , in the 35S-TGMV TrAP lines was comparable to the amount in wild-type plants ( Figure 1C , D ) . We further confirmed that loading of miRNAs into AGO1-centered RISCs was not affected by TrAP ( Figure 1E ) . The same results were obtained with miR167 , miR159 , and ta-siRNA480/255 and their corresponding targets ( Figure 1—figure supplement 1 ) . Thus , unlike most of previously reported viral suppressors , TrAP does not act on the miRNA pathway . To study how TrAP altered plant development , we mined publicly available databases and literature for the molecular and morphological phenotype of 35S-TrAP lines . We found that 35S-TrAP transgenic lines phenocopied several mutants in the epigenetic pathway including lhp1 ( Kotake et al . , 2003 ) ( Figure 2A ) and clf mutants ( Chanvivattana et al . , 2004 ) , with respect to the early flowering and upward curling of leaves . CLF belongs to PRC2 , a complex that catalyzes the deposition of H3K27me3 marks . LHP1 ( Nakahigashi et al . , 2005 ) , on the other hand , associates to silent genes in euchromatin and directs the spreading of the silent status to adjacent loci ( Turck et al . , 2007; Zhang et al . , 2007a , 2007b; Farrona et al . , 2008; Zheng and Chen , 2011 ) . Thus , coordinate activities of CLF and LHP1 result in chromatin methylation and transcriptional repression ( Farrona et al . , 2008 ) . 10 . 7554/eLife . 06671 . 005Figure 2 . TrAP is genetically involved in the TGS pathway . ( A ) 35S-TrAP transgenic plants phenocopied lhp1 mutants . Photographs were taken of 15-day seedlings . ( B , C ) Microarray results were validated by qRT-PCR analysis . Only 12 randomly selected loci were shown . ( D ) Gene ontology analysis of the TrAP-regulated DEGs . The numbers adjacent to the pies represent the ratio of genes in each category over the total DEGs . ( E ) Genome-wide overlapping of the genes regulated by TrAP and loss-of-function lhp1 . White and black numbers correspond to upregulated and downregulated genes , respectively . Maroon number indicates the genes that are differentially deregulated in both genotypes . ( F ) Heatmap of the commonly deregulated genes in the 35S-TrAP and lhp1 lines . The typical gene-ontology categories are shown on top . ( G , H ) Microarray and RNA blot analyses of epigenetically regulated flowering genes in the TrAP transgenic lines and lhp1 mutants . Cyt450 is a control . ( I ) qRT-PCR analysis of TEs in heterochromatic regions in the lhp1 mutant and TrAP transgenic lines . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 005 We examined global expression profiles of 7-day-old 35S-TrAP transgenic plants compared to Col-0 wild-type using an Affymetrix ATH1 GeneChip and identified 586 genes that are differentially expressed in the 35S-TrAP transgenic plants ( q < 0 . 005 ) . Of these , 295 transcripts were elevated whereas 291 were reduced ( Figure 2E ) . We performed real-time PCR and RNA blot assays to validate the microarray results for the differentially expressed genes ( DEGs ) . Among 25 genes randomly tested , we confirmed the ATH1 results for 24 , indicating that the microarray results were reliable ( Figure 2B , C; data not shown ) . Gene-Ontology ( GO ) analysis placed the DEGs into seven functional categories ( Figure 2D; Supplementary file 1 ) : hormone response ( 86 genes ) , stress response ( 94 genes ) , development regulation ( 50 genes ) , transcriptional regulation ( 20 genes ) , RNA metabolism ( 13 genes ) , post-translational modification ( 36 genes ) , and general metabolism ( 182 genes ) , plus a set of 105 un-annotated genes ( Figure 2D; Supplementary file 1 ) . Then , we compared the DEG profiles of the TrAP overexpression lines and loss-of-function lhp1 mutant ( Figure 2E , F; Supplementary file 2 ) . Transcriptome analysis revealed that out of 295 genes significantly upregulated in the 35S-TrAP transgenic lines , 120 ( 40 . 7% ) were also upregulated in lhp1 mutant . This is significantly greater than 1 . 28% expected by chance ( p < 2 . 2e−16 , Pearson's Chi-squared test ) . Interestingly , among the co-upregulated genes are a group of flowering-stimulated transcriptional factors including the key flowering-time integrator , FT , and 12 other genes such as TFL1 , AGL5 , and AGL9 ( Farrona et al . , 2008 ) ( Figure 2F–H; Supplementary file 2; Supplementary file 8 ) clustered in the transcriptional regulation category . Importantly , all these genes are regulated through epigenetic pathways and account for the early flowering phenotypes of lhp1 mutant and possibly of TrAP transgenic plants as well ( Gan et al . , 2013 ) . Other highly represented categories included 31 genes involved in aging and 116 genes engaged in stress responses ( Supplementary file 2 ) . Notably , the stress-responsive genes included genes specific to biotic stress such as PR4 , WRKY18 , FLS2 , and PDF1 . 2; additionally , genes related to chemical stress , such as PTR3 and TAT3 , were also identified . Thus , constitutive expression of TrAP could trigger plant senescence and innate defense pathways , and this activation is potentially through interference with the LHP1-related epigenetic silencing . Similarly , out of 291 genes significantly downregulated in the 35S-TrAP transgenic lines , 137 ( 47 . 1% ) were also repressed in lhp1 mutant . This is significantly greater than 1 . 25% expected by chance ( p < 2 . 2e−16 , Pearson's Chi-squared test ) ( Supplementary file 2 ) . Genes related to auxin response were of special interest . Of the 32 DEGs involved in the auxin pathway , 29 genes were downregulated in the 35S-TrAP transgenic plants , classified as small auxin upregulated mRNAs ( SAURs ) . These results suggested a possible hyposensitivity to auxin in lhp1 mutants and TrAP transgenic plants , which could explain the smaller statues of both genotypes . Consistent with this hypothesis , numerous auxin-repressed loci including PS2 and aging genes like TET9 , SAG13 , and SRG1 were upregulated in both lines . Concomitantly , six genes related to cell growth and five genes engaged in cell wall loosening were also repressed . Significant genome-wide overlap of 35S-TrAP and lhp1 loss-of-function-responsive genes suggested that TrAP might be genetically involved in the LHP1-related TGS pathway . Since LHP1 is believed to reside in euchromatic regions , we wondered whether TrAP also deregulates expression of heterochromatic loci . To this end , we selected numerous transposable elements ( TEs ) that were not recovered from the microarray assays and assessed them directly by qRT-PCR . Excitingly , most of the tested transposons were transcriptionally active ( Figure 2I ) , further suggesting that TrAP indeed inhibits the TGS pathway , in both euchromatic and heterochromatic regions . Given that TrAP transgenic plants phenocopied several TGS mutants and displayed transcriptional activation of heterochromatic loci , we hypothesized that TrAP epistatically regulates a TGS integrator ( s ) , indirectly leading to deregulation of the epigenetic marks . Analysis of the microarray data challenged this possibility as no significant changes in the transcripts of any canonical TGS components were revealed ( Supplementary file 3 ) . An alternative hypothesis was that TrAP directly interferes with the function of a TGS component ( s ) . To test this in an unbiased manner , we used luciferase complementation imaging ( LCI ) assay to screen 34 TGS-related proteins and some other cellular factors for TrAP interaction ( Zhang et al . , 2011 ) ( Supplementary file 4 ) . In the LCI experiments , the N- and C-terminal parts of firefly luciferase ( NLuc and CLuc ) are fused to different test proteins to be transiently expressed in Nicotiana benthamiana ( N . benthamiana ) . When NLuc and CLuc are brought together through interaction of the test proteins , catalytic activity is restored and recorded through CCD camera . In our LCI screening , we recovered LHP1 and KYP , a SUVH-type H3K9me2 methyltransferase , suggesting that TrAP is physically close to LHP1 and/or KYP proteins in vivo ( Figure 3B ) . Next , we carried out confocal microscopy imaging assays . TrAP co-localized with both LHP1 and KYP in scattered but not yet clearly defined nuclear foci , whereas co-expression of TrAP-CFP and other YFP-tagged proteins in N . benthamiana cells did not show such patterns ( Figure 3C ) . These observations further suggested that TrAP was in the same complexes as LHP1 or KYP . To further examine if TrAP interacted with these proteins , we conducted co-immunoprecipitation ( Co-IP ) assays ( Figure 3D ) . Interestingly , we validated the TrAP-KYP interaction ( Figure 3D ) but not TrAP-LHP1 ( data not shown ) , indicating that the LCI signal resulting from the TrAP-LHP1 combination likely involved additional cofactors between TrAP and LHP1 in vivo ( Figure 3B ) . Alternatively , TrAP-LHP1 interaction might be transient or unstable in our stringent co-IP condition . We observed that expression of CLuc-HA3-KYP in N . benthamiana yielded truncated proteins of various lengths that accumulated to comparable levels as the full-length protein; only the full-length KYP showed specific interaction with TrAP , implying the KYP C-terminal domain as the interaction interface with TrAP ( Figure 3D ) . We further confirmed the in vivo TrAP-KYP interaction by Förster resonance energy transfer-acceptor bleaching ( FRET-AB ) , using TrAP-CFP as a donor and YFP-KYP as an acceptor . Shortly , FRET involves the energy transfer from an excited donor to an adjacent acceptor when the fluorophores are less than 10 nm apart . If the fluorophores are coupled , the excited donor leads to acceptor emission; bleaching the acceptor allows the emission of the donor to be measured ( Figure 3E ) . Consecutive cycles of YFP-KYP bleach/recovery correlated with release/quench on the TrAP-CFP signal ( Figure 3E , F ) ; the positive FRET-AB ( CaLCuV-TrAP 2 . 699 ± 0 . 23 , TGMV-TrAP 3 . 8228 ± 0 . 58 , p < 0 . 05 ) corroborated TrAP-CFP/YFP-KYP interaction ( Figure 3G ) . To examine whether TrAP interacted with KYP under physiological conditions , we isolated TrAP complexes through two-step immunoprecipitation ( Zhu et al . , 2013 ) from stable transgenic plants expressing Flag-Myc4-TrAP under the inducible promoter ( XVE ) ( Zuo et al . , 2000 ) followed by mass spectrometry analysis . A total of 624 peptides representing 288 unique sequences were recovered from the TrAP sample; of those , 31 unique peptides matched specifically to KYP/SUVH4 and were not found in control immunoprecipitates using Col-0 plants ( Figure 3—figure supplement 1 ) . Together , all these assays clearly indicated that TrAP and KYP interact in vivo . 10 . 7554/eLife . 06671 . 006Figure 3 . TrAP interacts with KYP in vivo . ( A ) Schematic representation of the luciferase complementation imaging assay shows the different combinations of infiltrated constructs that were fused either to N-terminal ( NLuc ) and C-terminal ( CLuc ) regions of luciferase . ( B ) Screening of host factors targeted by TrAP . The infiltration positions of the constructs ( red arrows ) and luminescence signal resulting from the protein–protein interaction in a leaf are shown . FTSZ and LOM2 serve as negative controls . ( C ) Confocal imaging assays show the co-localization of TrAP-CFP with YFP-KYP and YFP-LHP1 in the nuclei in Nicotiana benthamiana . FTSZ serves as a negative control . ( D ) Specific interaction between KYP and TrAP was confirmed in N . benthamiana by co-immunoprecipitation ( Co-IP ) . Constructs harboring 35S-Myc-TrAP-nLuc and cLuc-HA-KYP were co-infiltrated in N . benthamiana leaves . IP was conducted using anti-Myc antibody . Western blot analyses were done with the crude extract ( input ) and the IP products using anti-Myc or -HA antibodies . Truncated versions ( red arrows ) serve as an internal control . ( E ) Exemplary imaging of FRET assays of TrAP-CFP and YFP-KYP co-expressed in a nucleus . The nucleus is irradiated with 458 nm laser to excite the CFP fluorophore . Three regions were selected for the assay: #1 , autofluorescence control , #2 fluorophore decay control , and #3 , FRET-acceptor bleaching test . Regions #1 and #3 were treated with pulses of 514 nm laser to bleach the YFP fluorophore . The CFP signal is then visible in region #3 when the emission of CFP is dequenched . ( F ) Quantification of the signals from each fluorophore observed during FRET-AB experiment in E . ( G ) FRET is positive for YFP-KYP paired with the CFP-tagged TrAPs from either CaLCuV or TGMV . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 00610 . 7554/eLife . 06671 . 007Figure 3—figure supplement 1 . Mass spectrometry analyses confirmed endogenous KYP as a bona fide TrAP-interacting partner . ( A ) β-estradiol Inducible expression of Flag-4Myc-TrAP protein in Arabidopsis transgenic lines ( B ) 31 peptides ( green ) uniquely match to the KYP sequence . ( C ) Description of recovered peptides matching KYP/SUVH4 . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 007 We performed in vitro pull-down assays to examine whether TrAP interacts directly with KYP . We found that maltose-binding protein ( MBP ) -KYP , but not MBP and other MBP-tagged control proteins , was able to pull down GST-tagged TGMV- and CaLCuV-encoded TrAP proteins ( Figure 4A , B ) . This interaction was specific as MBP-KYP was unable to pull down GST protein alone . Notably , we did not observe direct LHP1-TrAP interaction in the parallel pull-down experiments ( Figure 4—figure supplement 1 ) ; and this result was consistent with the in vivo Co-IP experiments . From N- to C- termini , KYP contains SRA , PreSET , SET , and PostSET domains . To further define the protein domain ( s ) responsible for the specific interaction , we generated five truncations of KYP ( Figure 4C ) . Pull-down assays showed that the SET domain interacted with TrAP proteins from both TGMV and CaLCuV at a higher affinity , whereas SRA domain might interact with CaLCuV TrAP at a reduced affinity in vitro ( Figure 4D–F ) . These results were consistent with the results of the in vivo Co-IP experiments in which only full-length KYP , but not C-terminal-truncated versions , could recover TrAP ( Figure 3D ) . Recent structural analysis on KYP revealed that SET and pre-SET domains constitute two modules: one forms a narrow pocket harboring the H3 tail ( 1–15aa ) , whereas the other binds the SAM cofactor , together with the post-SET domain ( Du et al . , 2014a ) . Because the post-SET domain does not seem to contribute to the KYP-TrAP interaction , TrAP could potentially occupy the histone-binding cavity ( Figure 4F ) . 10 . 7554/eLife . 06671 . 008Figure 4 . TrAP interacts directly with KYP through the SET domain . ( A , B ) In vitro pull-down assays showed that KYP specifically interacted with TGMV TrAP ( A ) and CaLCuV TrAP ( B ) . Left panel , Coomassie brilliant blue R250 staining of the proteins shows their mobility . Right panel , output of in vitro pull-down assays . All His-MBP-tagged bait proteins and His-GST-tagged prey proteins were purified from Escherichia coli using Ni-NTA columns . In all assays , 2 . 5 μg of prey proteins were pulled down with the indicated bait proteins ( 2 . 5 μg each ) using amylose resin . The recovered MBP-tagged bait proteins were monitored by Coomassie brilliant blue R250 staining . All the experiments were done at the same time , and the samples were run in the same gels . The spacers in the images indicate digital rearrangements of the pictures . The output of the GST-tagged prey proteins was analyzed by western blot using a monoclonal anti-GST antibody . ( C ) Schematic diagram of full-length and truncated forms of KYP . The numbers on the left refer to the amino acid residues in KYP protein . Locations of SRA , pre-SET , SET , and post-SET domains are shown . ( D ) In vitro pull-down assays of truncated KYP proteins and TrAPs . The experiments were done as in ( A , B ) . The GST negative control was loaded in the lanes marked with ( * ) . ( E ) Summary of interaction between the truncated KYP proteins and TrAP encoded by TGMV and CaLCuV . ( F ) Model of possible KYP-TrAP interaction based on the experimental results from panels D and E . KYP structure was generated in Chimera from PDB: 4QEN data set ( Du et al . , 2014b ) , domains are color coded and indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 00810 . 7554/eLife . 06671 . 009Figure 4—figure supplement 1 . TrAP does not interact with LHP1 in vitro . In vitro pull down assays of GST-TrAP by MBP-LHP1 is shown . Left panel , Coomassie brilliant blue R250 staining of the proteins shows their mobility . Right panel , output of in vitro pull-down assays . The recovered MBP-tagged bait proteins were monitored by Coomassie brilliant blue R250 staining . The output of the GST-tagged prey proteins was analyzed by western blot using a monoclonal anti-GST antibody . TrAP dimerization is shown as positive control . 2 . 5 μg of prey proteins were pulled down with the indicated bait proteins ( 2 . 5 μg each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 00910 . 7554/eLife . 06671 . 010Figure 4—figure supplement 2 . TrAP directly interacts with KYP paralogs SUVH2 , 5 , and 6 . In vitro pull down assays . Left panel , coomassie brilliant blue R250 staining of the proteins shows their mobility . Right panel , output of in vitro pull-down assays . The recovered MBP-tagged bait proteins were monitored by Coomassie brilliant blue R250 staining . The output of the GST-tagged prey proteins was analyzed by western blot using a monoclonal anti-GST antibody . SUVH4/KYP was used as a positive control . 2 . 5 μg of prey proteins were pulled down with the indicated bait proteins ( 2 . 5 μg each ) . All the experiments were performed simultaneously and run in two separate gels . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 010 Since the SET domain is well conserved among histone methyltransferases ( HMTase ) ( Liu et al . , 2010 ) , we next investigated whether TrAP interacts with KYP paralogs . In vitro pull-down assays showed that TrAP indeed interacted with numerous tested HMTases ( SUVH2 , 5 , and 6 ) ( Figure 4—figure supplement 2 ) . Notably , loss-of-function mutants of SUVH2 display early-flowering phenotype , suggesting that TrAP might target this protein in vivo as well . The specific interaction of TrAP with the catalytic SET domain of the SUVHs prompted the question of whether TrAP affects KYP activity . To test this , we set up an in vitro reconstitution of H3K9 methylation using His-GST-KYP purified from Escherichia coli as the enzyme source , recombinant histone 3 as the substrate , and methyl-14C-SAM as the methyl donor ( Figure 5A ) . Under our experimental conditions , 1 µM GST-KYP methylated 3 µM of histone 3 in less than 5 min at 37°C , as detected by saturation of the radioactive signal ( Figure 5B ) . Excitingly , incubation of His-GST-KYP with His-MBP-TrAP from either TGMV or CaLCuV reduced the initial velocity of KYP transmethylation activity in a dose-dependent fashion , whereas His-MBP alone did not affect KYP catalysis ( Figure 5B ) . Quantification of signal intensity revealed that the TrAP-KYP molar ratio of 2 was enough to cause approximately 50% inhibition of KYP activity ( half maximal inhibitory concentration [IC50] ) ( Figure 5C ) . Thus , our results indicated that TrAP potently inhibited the catalytic function of the HMTase . 10 . 7554/eLife . 06671 . 011Figure 5 . TrAP inhibited HMTase activity of KYP in vitro . ( A ) Coomassie staining of purified proteins uses for the assays . ( B ) In vitro HMTase reconstitution assays with different molar ratio of MBP and MBP-TrAP proteins ( 0–10× ) relative to GST-KYP . The recombinant KYP was incubated without ( buffer only ) or with the indicated proteins before addition of histone 3 and C14-SAM . The reactions were done in a 6-min time course; aliquots were resolved in 18% SDS PAGE and stained with Coomassie blue R250 to show histone 3 input ( top panels ) . The dried gels were auto-radiographed to detect 14C-methylated histone 3 . ( C ) . Plotting of KYP initial velocity vs TrAP/KYP molar ratio . The initial velocity was calculated from the slope of the linear range in the signal vs time plot for each assay , and then the values were normalized using the non-inhibitor control as a standard of 1 to obtain the relative initial velocity with standard deviation ( SD ) from at least three biological repeats . The relative initial velocity is plotted as a function of inhibitor:enzyme molar ratio in a logarithmic scale of base 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 011 Given that in vivo TrAP genetically interferes with the TGS pathway , and in vitro it physically interacts with HMTases ( KYP , SUVH2 , 5 , 6 ) and inhibits the activity of KYP , we wondered whether TrAP alters KYP function in vivo . To address this question , we conducted chromatin immunoprecipitation ( ChIP ) analyses of H3K4me3 , H3K9me2 , and K3K27me3 marks ( Figure 6; Figure 6—figure supplement 1; Figure 6—figure supplement 2 ) on numerous KYP-regulated TEs ( Figure 6A ) in TrAP overexpression plants as compared to Col-0 , ddm1 , kyp , and lhp1 control plants . Remarkably , all tested loci in TrAP transgenic lines displayed consistent reduction of H3K9me2 and H3K27me3 , whereas changes of H3K4me3 were variable . This molecular phenotype mimicked kyp mutant at all tested loci , but not lhp1 mutant . This result indicated inhibitory effect of TrAP on KYP activity on the TEs in vivo . 10 . 7554/eLife . 06671 . 012Figure 6 . ChIP-qPCR analyses of H3K4me3 , H3K9me2 , and H3K27me3 in TrAP-regulated loci in vivo . ( A ) TrAP-activated transposons in heterochromatic regions contained reduced H3K9me2 and H3K27me3 but did not show consistent variation in H3K4me3 marks . ( B ) TrAP-upregulated flowering genes showed consistently reduced H3K9me2 and H3K27me3 marks compared to wild-type Col-0 . ( C ) TrAP-downregulated genes displayed variable changes of H3K9me2 and H3K27me3 marks and no obvious changes of H3K4me3 mark . ( D ) Tubulin ( TUB8 ) was used as internal control for all the ChIP experiments; the percentage enrichment vs input is shown . ChIP assays were conducted on 11-day-old seedlings using antibodies specific for H3K9me2 ( Abcam , Cat# ab1220 ) , H3K27me3 ( Millipore , Cat# 07-449 ) , and H3K4me3 ( Millipore , Cat# 04-745 ) . Enrichment of H3K4me3 and H3K27me3 in each locus is normalized to that of TUB8; H3K9me2 enrichment is plotted as percentage of input . The standard deviation ( SD ) was calculated from at least three biological repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 01210 . 7554/eLife . 06671 . 013Figure 6—figure supplement 1 . Western blot analysis to show specificity of antibodies used for ChIP assays in the study . Crude extract ( A ) and isolated nuclei ( B ) were probed with antibodies against histone 3 , H3K9me2 ( Abcam Cat# ab1220 ) , H3K27me3 ( Millipore Cat# 07-449 ) and H3K4me3 ( Millipore Cat# 04-745 ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 01310 . 7554/eLife . 06671 . 014Figure 6—figure supplement 2 . ChIP-PCR assays for selected flowering genes and heterochromatic loci confirm ChIP-qPCR . ( A ) ChIP-PCR analysis of various histone 3 modifications in flowering genes in different genetic backgrounds . ( B ) ChIP-PCR analysis of various histone 3 modifications in TEs in different genetic backgrounds . ChIP assays were conducted on 9-day-old seedlings using antibodies specific for H3K9me2 ( Abcam Cat# ab1220 ) , H3K27me3 ( Millipore Cat# 07-449 ) , and H3K4me3 ( Millipore Cat# 04-745 ) . The PCRs were done with 22 cycles for the input samples and with 30 cycles after ChIP . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 014 Since TrAP is a transcriptional activator protein , we next asked whether the increased transcription of the TrAP-responsive protein-coding genes is associated with changes in the histone methylation status . First , we screened numerous loci that showed transcriptional deregulation in the TrAP overexpression lines ( Figure 2B–D , G , H ) for the presence of various histone modifications . We identified a dozen loci in which H3K9me2 marks are easily detected in wild-type plants . We then conducted ChIP-qPCR for these loci in the TrAP transgenic plants . As expected , H3K9me2 and H3K27me3 were reduced in most of the tested loci in ddm1 , kyp and lhp1 mutants , whereas H3K4me3 was enriched . These results are consistent with the generally antagonistic roles of H3K4me3 and H3K9me2 modifications . Specifically , six out of seven TrAP deregulated loci including the flowering-promoting genes displayed greater than twofold reduction in H3K9me2 and H3K27me3 in the TrAP overexpression plants compared to wild-type Col-0 , while changes in H3K4me3 were inconsistent . This scenario was similar to that observed for TEs ( Figure 6B , C ) . Collectively , the ChIP assays indicated that TrAP interferes with the epigenetic pathways through reducing repressive H3K9me2 marks in vivo . KYP , SUVH5 , and SUVH6 are required for maintenance of non-CG ( CHG and CHH ) methylation in Arabidopsis ( Stroud et al . , 2013; Stroud et al . , 2014 ) . We predicted that inhibition of KYP function by TrAP might indirectly cause reduction in non-CG DNA methylation . To test this , we conducted genome-wide bisulfite sequencing with 11-day-old seedlings of TrAP transgenic plants , kyp mutant , and Col-0 . Consistent with previous studies , kyp mutant showed genome wide loss of methylation in CHG ( ∼42 . 1% ) and CHH ( ∼21 . 7% ) , but not in CG ( ∼2% ) contexts when compared to Col-0 . To our surprise , TrAP transgenic plants only exhibited decrease in methylation of CHH ( ∼11% ) but not CHG ( Figure 7A ) . 10 . 7554/eLife . 06671 . 015Figure 7 . TrAP reduces CHH DNA methylation in vivo . ( A ) Genome-wide heatmap of DNA methylation levels in Col-0 , kyp mutant , and TrAP transgenic plants ( Left ) . Sequence context of all cytosine , CG , CHG , and CHH methylation was depicted as black , yellow , blue , and red , respectively . The percentage of cytosine methylation is shown for each genotype ( Right ) . ( B ) Average distribution of context-specific DNA methylation in Gene- and TE-rich regions in Col-0 ( red ) , kyp mutants ( green ) , and TrAP transgenic plants ( blue ) . ( C ) . Overlap of CHH DMRs between TrAP transgenic plants and kyp mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 01510 . 7554/eLife . 06671 . 016Figure 7—figure supplement 1 . Gene ontology of CHH hypomethylated genes in TrAP transgenic plants and kyp mutant . The genes associated to the CHH hypomethylated DMRs in both TrAP transgenic and kyp mutant plants underwent Gene Ontology analysis using AgriGo tool with TAIR10 as reference . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 016 To further analyze the effect of TrAP on DNA methylation , we identified differentially methylated regions ( DMRs ) by scanning the genomes in 200 bp tiles and comparing the level of methylation among kyp and TrAP overexpression plants with Col-0 ( ‘Materials and methods’ ) ( Supplementary files 5 , 6 , 7 ) . Given that non-CG methylation is highly co-localized and predominant in TE-rich heterochromatic regions ( Stroud et al . , 2013 , 2014; Shen et al . , 2014; Dubin et al . , 2015; Yang et al . , 2015 ) , we separated the DMRs into gene- and TE-rich regions ( GRR and TERR respectively ) ; When compared to Col-0 , TrAP transgenic plants displayed loss of CHH methylation at GRR but not in TERR , nor in CG or CHG contexts ( Figure 7B ) . Remarkably , of the 3442 and 1784 GRR hypomethylated DMRs identified in kyp mutant and TrAP transgenic plants , 1642 were shared . To better understand the effect of TrAP expression on CHH DNA methylation , the GRR DRMs were further separated into promoter , terminator , UTR , intronic , and coding regions . We found that TrAP hypomethylated DMRs in coding sequences , UTRs , and introns overlapped almost completely with the kyp mutant ( 96 . 7% , 96% , and 95 . 4% respectively ) ( Figure 7C ) . Notably , gene ontology analysis of the overlapped genes pointed to response genes , specifically in the protein kinase category ( Figure 7—figure supplement 1; Supplementary file 7 ) . This substantial overlap explains about one-third of the total CHH DMRs in kyp mutant , and one half of kyp DMRs in genic regions ( Figure 7C ) . Together , these results suggest an inhibitory effect of TrAP in KYP-dependent CHH DNA methylation . In animals , H3K9 methylation promotes TGS and latency of integrated viruses . In plants , Geminivirus constitutes into a minichromosome that also undergoes epigenetic regulation . The specific TrAP-KYP interaction , inhibition of KYP activity in vitro , and reduction of H3K9me2 and CHH methylation in vivo suggest that KYP is a major factor in combating viruses . Previous studies showed that kyp mutants are hypersusceptible to Geminivirus infection ( Hanley-Bowdoin et al . , 2013 ) ; these experiments were reproducible in our hands ( Figure 8A–D ) . Moreover , the hypersusceptibility of kyp mutant to the virus could be rescued by the wild-type KYP transgene under the control of both the native and the constitutive 35S promoters ( Figure 8A–D ) ; further validating a role for KYP in regulation of viral infection . In line with the phenotypic complementation , wild-type KYP almost completely rescued H3K9me2 defects of the TEs in the kyp mutant ( Figure 8E , lanes 1–4 ) . Interestingly , although the accumulation of KYP transcripts was substantially increased when transcribed from the 35S promoter when compared to the native promoter , the steady-state protein level was only twofold to threefold higher in the 35S-Flag-4Myc-KYP than in the PKYP-Flag-4Myc-KYP transgenic plants ( data not shown ) . These observations suggest a possible homoeostatic regulation of this critical TGS component . 10 . 7554/eLife . 06671 . 017Figure 8 . KYP methylates Geminivirus chromatin as a host defense . ( A ) Western blot analysis of kyp complementation lines expressing PKYP- or 35S-Flag-4Myc-KYP using anti-myc antibody . * , a cross-reaction band serves as a loading control . ( B ) Representative CaLCuV symptoms with different severities . ( C ) Time course of CaLCuV symptom development in kyp mutant and the complementation lines . The mean values were calculated with SD from at least three experiments ( >30 plants/line ) . ( D ) CaLCuV symptom severity in Col-0 , kyp mutant , and the complementation lines . The mean values were calculated with SD from at least three experiments ( >30 plants/line ) . ( E ) ChIP-PCR of H3K9me2 marks in heterochromatic loci in kyp mutant and complementation lines inoculated with mock or CaLCuV . Note: CaLCuV infection largely removed H3K9me2 marks from heterochromatic loci . ( F ) Schematic linearized representation of the regions of viral genome A selected for ChIP assays . ( G ) ChIP-qPCR assays showed KYP-dependent enrichment of H3K9me2 in several tested loci in the viral chromatin . The relative value of histone methylation in each sample was normalized to that of wild-type control where the signal was arbitrarily assigned a value of 1 with standard deviation ( SD ) from at least three biological repeats . Note: the region defined by # Primer 7 serves as a negative control . ( H ) Western blot analysis to detect Flag-Myc4-KYP in the ChIP ( IP ) samples using anti-myc antibody . ( I ) ChIP-PCR assays showed that KYP binds to the viral minichromosome . The ChIP assays were done with a monoclonal anti-Flag antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 01710 . 7554/eLife . 06671 . 018Figure 8—figure supplement 1 . Virus chromatin contains H3K9me2 marks . ChIP-PCR assays of H3K9me2 marks on the CaLCuV DNA A . ChIP assays were conducted on 9-day-old seedlings using antibodies specific for H3K9me2 ( Abcam Cat# ab1220 ) , H3K27me3 ( Millipore Cat# 07-449 ) , and H3K4me3 ( Millipore Cat# 04-745 ) . The PCRs were done with 22 cycles for the input samples and with 28 cycles after ChIP . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 018 Geminivirus minichromosome harbors H3K9me2 ( Figure 8G; Figure 8—figure supplement 1 ) ( Hanley-Bowdoin et al . , 2013 ) . Interestingly , H3K9me2 marks were dramatically decreased on several , but not all tested loci in the viral genome in kyp mutant , and the methylation marks were further restored by the KYP transgene , suggesting that viral chromatin H3K9me2 is catalyzed by KYP . Notably , the H3K9me2 amount was substantially enhanced in the locus defined by Primer #9 in the transgenic KYP overexpression plants ( Figure 8G ) . It is noteworthy that this locus harbors the promoter element for TrAP itself ( Shung and Sunter , 2009 ) . This result suggests that the tight control of TrAP expression likely determines the balance between the host and virus interactions . If so , the result might also explain the relatively milder viral symptoms of 35S-Flag-Myc4-KYP lines compared to WT control or PKYP-Flag-Myc4-KYP complementation lines ( Figure 8A–D ) . To further test whether KYP methylated the viral chromatin , we performed ChIP assays using monoclonal anti-Flag antibody to pull down Flag-Myc4-KYP bound chromatin ( Figure 8H ) . Excitingly , KYP was found at all tested loci in the viral chromatin ( Figure 8I ) . Notably , regions delineated by the primer #7 showed KYP-independent H3K9me2 marks but were still immunoprecipitated in the KYP-chromatin complex . A possible explanation is incomplete chromatin shearing of the small viral mini-chromosome under the conditions used in this experiment , which were standardized for host chromatin ChIP . Alternatively , this might suggest the presence of additional epigenetic regulation that masks KYP activity on this locus . This notwithstanding , our results indicate that KYP directly deposits the H3K9me2 mark on the Geminivirus minichromosome to reinforce the silent status of the virus . Since constitutive expression of TrAP reduced H3K9me2 in vivo , we wondered if Geminivirus infection could also decrease the repressive marks in the host . To this end , we tested the KYP-controlled endogenous transposons for H3K9me2 accumulation . CaLCuV-infected plants showed 60–100% H3K9me2 loss in the tested loci compared to the amount in the mock-inoculated plants ( Figure 8E ) , this is reminiscent of the molecular phenotype of TrAP transgenic plants , which showed lower enrichment of H3K9me2 mark at the studied loci ( Figure 6C ) . These results indicated that Geminivirus infection largely removed the repressive H3K9me2 marks in the transposons , and that the removal resulted at least in part from TrAP function . Given that TrAP is known to activate the expression of viral genes in the minichromosome and that endogenous transposons can serve as a proxy for viral genomes , it is conceivable that virus-encoded TrAP acts to suppress KYP in order to prevent the deposition of H3K9me2-repressive marks in the epigenome , to activate the expression of viral genes . Previous studies show that TrAP is required for the accumulation of the virus infective form , single-stranded ( ss ) DNA ( Hayes and Buck , 1989 ) . TrAP is indispensable for systemic infection of Begomoviruses because it activates the expression of the viral ssDNA binding proteins ( nuclear shuttle protein and coat protein ) , which are essential for releasing the virus from the nucleus and for cell-to-cell spreading ( Sunter and Bisaro , 1992 ) . If TrAP-mediated transcriptional activation and accumulation of ssDNA result from inhibition of KYP and correspondingly heterochromatin formation , then infectivity of CaLCuV lacking TrAP should be impaired in wild-type but recovered in kyp plants . To test this hypothesis , we engineered a CaLCuV variant without functional TrAP protein ( CaLCuV Δtrap ) by changing a single nucleotide ( T to A ) that produces an amber mutation and introduces an XbaI restriction site after the sixth codon in the TrAP gene ( Figure 9A ) . Then , we assessed systemic infection of CaLCuV and CaLCuV Δtrap on wild-type and kyp plants ( Figure 9B ) . Consistent with the hypersusceptible phenotype of kyp mutants to CaLCuV ( Figure 8C , D ) , these plants accumulated significantly higher titers of CaLCuV relative to wild-type control as both the replicative intermediate , open circle ( OC ) , and the infective particle , SS DNA ( Figure 9C ) . Consistent with previous reports , the wild-type plants did not show any symptoms of infection ( Figure 9E ) or systemic accumulation of CaLCuV Δtrap , as evidenced by Southern blot analyses and PCR ( Figure 9C ) . Particularly , PCR amplification followed by XbaI digestion of the viral TrAP region in wild-type plants infected with CaLCuV Δtrap showed loss of the XbaI restriction site ( Figure 9—figure supplement 1 ) . Excitingly , the viral DNA was detected , in low amount but reproducibly , in kyp mutants . Further PCR amplification of the TrAP gene followed by XbaI digestion , confirmed that the detected viral DNAs in kyp mutants indeed came from CaLCuV Δtrap and not from a reversion of the mutation in the virus genome . These results show that kyp plants sustained systemic infection of CaLCuV Δtrap . Noticeably , the titers of accumulated CaLCuV Δtrap were much lower than the ones of CaLCuV , suggesting either redundant activity of KYP paralogs in host defense or necessity of additional functions of TrAP besides inhibition of KYP activity for the virus to achieve efficient infection . 10 . 7554/eLife . 06671 . 019Figure 9 . Infectivity of CaLCuV lacking functional TrAP protein . ( A ) Sequence alignment of CaLCuV Δtrap and CaLCuV sequences . The translated amino acids are shown for each sequence , and the XbaI restriction site resulting from the T to A point mutation is highlighted . ( B ) Schematics of the systemic infection experiment . Plants with eight true leaves ( depicted in gray ) were inoculated with the begomovirus , and 18 days post inoculation nine to 11 newly emerged , not inoculated , rosetta leaves ( depicted in bright green ) were collected to test for virus systemic infection . ( C ) Southern blot analysis of viruses in non-inoculated leaves of infected plants . Ethidium bromide staining of total genomic DNA serves as a loading control ( top panels ) . Southern blots were probed against CaLCuV DNA A common region ( CR ) ( bottom panels ) ; the viral populations are indicated as the replicative intermediate open circle ( OC ) , linear ( L ) , super coiled ( SC ) and the infective particle ss DNA . ( D ) Genotypic confirmation of the systemically amplified viruses . Top panels show PCR amplification of a TrAP-containing region; bottom panels show EcoRI and XbaI digestions of PCR products to examine the presence of the amber mutation in the gene . ( E ) Exemplary phenotypes of wild-type and kyp mutant plants inoculated with mock , CaLCuV and CaLCuV Δtrap . ( F ) Model of TrAP suppression of KYP activity to prevent epigenetic silencing of the viral chromatin . Geminivirus genome is packed on histone octamers to form a minichromosome . The minichromosome undergoes extensive H3K9me2 modification deposited by host-encoded KYP , and this modification could be further reinforced by DNA methylation , leading to formation of viral heterochromatin . As a counter-defense strategy , Geminivirus-encoded TrAP protein inhibits KYP activity to maintain the euchromatic status of the minichromosome to allow active replication and transcription of viral genes and correspondingly to escape host surveillance . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 01910 . 7554/eLife . 06671 . 020Figure 9—figure supplement 1 . CaLCuV lacking functional TrAP protein cannot cause systemic infection in wild-type plants . The region corresponding to TrAP gene was amplified by PCR using the primers AL3_cterm_F and AL1_cterm_R ( Supplementary file 8 ) from total DNA extracted from plants infected with CaLCuV or CaLCuV Δtrap . 5 μl of the PCR products were run , and 3 μl were used for digestion with EcoRI or XbaI to examine the presence of the amber mutation in all the amplified products . DOI: http://dx . doi . org/10 . 7554/eLife . 06671 . 020 TrAP was among the first viral suppressors identified to interfere with the TGS pathway . The prevailing model is that TrAP lowers the reservoir of transferable methyl groups by targeting ADK , a key enzyme in the SAM pathway . Here , we propose a novel model in which TrAP regulates TGS by directly targeting KYP ( Figure 9F ) . Several pieces of evidence supported our notion: ( 1 ) TrAP genetically interfered with the TGS pathway ( Figure 2 ) ; ( 2 ) TrAP directly interacted with KYP in vivo and also with other HMTases in vitro ( Figure 3; Figure 4 ) ; ( 3 ) TrAP inhibited the catalytic activity of KYP in vitro ( Figure 5 ) ; ( 4 ) TrAP reduced the repressive H3K9me2 marks in vivo ( Figure 6 ) , and correspondingly , reactivated numerous loci that are otherwise repressed by KYP ( Figure 2 ) ; ( 5 ) TrAP decreased CHH methylation in gene-rich regions that are also regulated by KYP ( Figure 7 ) ; ( 6 ) methylation of viral chromatin entailed KYP ( Figure 8 ) ; ( 7 ) KYP bound the viral chromatin ( Figure 8 ) ; and ( 8 ) kyp mutants but not wild-type plants sustain low systemic infection of CaLCuV lacking TrAP protein ( Figure 9 ) . To our best understanding , this is the first evidence that a viral protein directly suppresses HMTases in the host TGS machinery . Given that TrAP protein could concurrently limit the upstream supply of the methyl groups and directly inhibit downstream enzymatic activity of KYP , Geminiviruses appear to have evolved sophisticated strategies to cripple the host TGS pathway . What would be the biological advantages of blocking the TGS pathways ? In eukaryotes , chromatin appears to be a critical battleground for virus–host interaction . Animals use histone modifications to reinforce the latency of integrated viruses ( du Chene et al . , 2007; Narasipura et al . , 2014; Park et al . , 2014 ) . Plant DNA viruses including Geminiviruses and pararetroviruses replicate as nuclear minichromosomes or episomes . Clearly , TGS functions as an immune system to control virus replication and the expression of viral genes , in a similar fashion as the repression of endogenous TEs and transposon remnants . In previous studies and here in our experiments , kyp mutants displayed hypersusceptibility to Geminivirus infection , which could be rescued by exogenous wild-type KYP gene . KYP directly acts on viral chromatin and deposits the repressive H3K9me2 mark on the viral chromatin . Moreover , viruses lacking the essential TrAP protein can , although inefficiently , cause systemic infection in KYP-deficient hosts . All these facts point out the unambiguous role of KYP-mediated TGS in defense against viral infection . On the other hand , TrAP functions to inhibit KYP catalytic activity , reducing the repressive H3K9me2 mark , to activate transcription of viral genes . Our KYP reconstitution assays show that TrAP:KYP molar ratio of around 2 is enough to cause ∼50% inhibition of KYP activity , indicating that TrAP is a potent inhibitor of HMTase activity . Thus , direct inhibition of KYP represents a novel counter-defense mechanism for virus survival in the hosts . This mechanism could account for the long-documented essential role of TrAP in expression of viral genes including the coat protein and the nuclear shuttle protein ( Sunter and Bisaro , 1992; Yang et al . , 2007; Shen et al . , 2009; Hanley-Bowdoin et al . , 2013 ) . In plants , H3K9me2 marks often correlate with non-CG DNA methylation , and in particular with CHG modification . The tight coordination results from a self-enforcing loop consisting of KYP and CMT3 ( Du et al . , 2014a ) . Briefly , KYP methylates H3K9 to generate the binding sites for CMT3 , which further methylates CHG DNA to create more binding sites for KYP . Consequently , the crosstalk between DNA and histone methylation ascertains the silent status of heterochromatin . We hypothesized that TrAP targets KYP to reduce H3K9me2 marks , and this inhibition would further decrease DNA methylation both on the host and on the viral genome . Here , we found that ectopic expression of TrAP had no effect on CHG but only on CHH methylation , at sites that are up to 96 . 7% overlapped with the hypo-methylated regions in kyp mutant ( Figure 7C ) . In plants , sequence contexts of CHG and CHH methylation are largely overlapped throughout the genome and are maintained to a limited extent by all non-CG methyltransferases ( Stroud et al . , 2014 ) . One outstanding question would be how could TrAP differentially alter methylation of CHH rather than CHG , given that TrAP targets KYP ? Methylation in the CHH context is catalyzed by CMT2 and DRM1/2 in Arabidopsis , and the sites regulated through the two sets of enzymes are mostly non-overlapping . Thus , the functions of CMT2 and DRM1/2 are mostly non-redundant at CHH sites ( Stroud et al . , 2013 , 2014 ) . DRM1/2 largely catalyze CHH methylation at the TE-rich regions ( Stroud et al . , 2014 ) . In our study , we did not observe any effect of TrAP on CHH methylation in the TEs , suggesting that TrAP might not ( or not sufficiently ) interrupt the RdDM pathway that entails DRM1/2 and 24-nt siRNAs ( Stroud et al . , 2014 ) . Importantly , it has been recently discovered that bulk CHH methylation is maintained by CMT2 ( Shen et al . , 2014; Stroud et al . , 2014; Dubin et al . , 2015 ) and that its activity is largely dependent on H3K9me2 . In this scenario , CMT2 recognizes methylated H3K9 but preferentially binds to di-methylated over mono-methylated histone tails . This preference towards H3K9me2 is not observed in the CHG methyltransferase CMT3 , which can equally bind to all forms of H3K9 methylation ( Stroud et al . , 2014 ) . We envision that methylation at CHH sites could be more sensitive to changes in H3K9me2 than CHG , since CMT3 could still maintain CHG DNA methylation in the presence of H3K9me1 resulting from residual activity of HMTase , as would be the case when KYP is inhibited by TrAP . Alternatively , CHH methylation might play an important but yet unappreciated regulatory role in host defense genes ( Figure 7—figure supplement 1 ) . If so , preferentially targeting these loci by TrAP protein might represent a new counter-defense mechanism . In the host , Geminivirus DNAs also undergo extensive methylation modification ( Raja et al . , 2008 ) . In our study , TrAP reduces CHH methylation in the host and possibly in the viral genome , too . Remarkably , the Geminivirus replicase , Rep/AC1/C1 and the embedded protein AC4/C4 , downregulate the expression of host DNA methyltransferases MET1 and CMT3 ( Pumplin and Voinnet , 2013; Rodriguez-Negrete et al . , 2013 ) ; hence , it interrupts the reinforcing loop of histone and DNA methylation . Consistent to this study , cmt3 mutants exhibit hypersusceptibility to viral infection ( Raja et al . , 2008 ) . The fact that Rep/C4 repress CMT3 expression in the host is in perfect alignment with our ChIP assay results on CaLCuV-infected plants , where the loss of H3K9me2 is even more severe than the observed in TrAP transgenic plants . Thus , interruption of the compelling feedback loop of histone and DNA methylation represents an important strategy to sustain transcription of viral chromatin . Together , it seems that the synergistic inhibition of histone and DNA methyltransferases by Geminivirus proteins evolves as a powerful tactic to win the arms race between host and pathogen . Our in vitro assays clearly demonstrated that TrAP predominantly binds to the catalytic domain of KYP and inhibited its enzymatic activity . Whether TrAP might alter KYP conformation or block the accessibility of substrates to the active sites upon interaction awaits future structural analysis . We note , however , that Arabidopsis has 49 SET domain containing proteins of which 31 are considered to have HMTase activity ( Liu et al . , 2010 ) . Arising from this fact is whether TrAP specifically targets KYP or promiscuously acts on additional HMTases . Although TrAP overexpression plants display molecular features in common with kyp mutant ( i . e . , reduced H3K9me2 levels ) , and an early flowering phenotype similar to loss-of-function mutant of the kyp paralog , SUVH2 , the transgenic plants are phenotypically different from kyp single mutant . This could be due to functional redundancy between KYP and its paralogs in vivo; alas , the morphological phenotype of the higher-order mutants has not been fully documented . We also noticed that TrAP-overexpression plants are morphologically similar to lhp1 , which is characteristic of the H3K27me3 pathway . Moreover , the substantial overlapping of TrAP-responsive genes with lhp1-regulated genes strongly suggested that TrAP might target the LHP1-related H3K27me3 pathway ( Zheng and Chen , 2011 ) . Indeed , most of our tested loci in the host genome exhibited decreased H3K27me3 levels , consistent with the fact that ∼50% genes were co-regulated by TrAP and LHP1 in a genome-wide scale . Interestingly , recent ChIP–chip studies revealed that H3K27me3 and LHP1-bound sites are predominantly distributed in the euchromatic regions ( Turck et al . , 2007; Zhang et al . , 2007a , 2007b ) . This distribution is not correlated with KYP-dependent H3K9me2 marks that are highly enriched at pericentromeric regions as large and uninterrupted heterochromatic blocks ( Liu et al . , 2010; Black et al . , 2012; Du et al . , 2014a; Liu et al . , 2014 ) . H3K9me2 can also occur in euchromatic regions but rather exist as small heterochromatin patches ( Zheng and Chen , 2011 ) . In our current study , we did not further examine whether TrAP physically targets H3K27me3 HMTases in the LHP1 pathway . But it is plausible that TrAP inhibits both KYP and LHP1 pathways . Notably , we did not observe the repression of transcriptionally active histone methylation marks , such as H3K4me3 and H3K36me3 . This observation suggests that TrAP does not target the HMTases that are writers for the active marks . How can TrAP distinguish KYP from those active writers remains unclear , but we propose that structural components other than the SET domain might contribute to the recognition and affinity of TrAP binding to HMTases ( Figure 4 ) . Alternatively , additional cellular factors might also contribute to the specificity . The direct consequence of TrAP-dependent inhibition of KYP activity is to activate viral transcription and replication . Because KYP is a key effector of TGS in the host and regulates a broad array of endogenous genes , interference with this core component would reprogram the expression profile of the host genome and thus trigger a series of downstream cascade signaling events that impact the balance of host/virus interaction . As an example , TrAP suppresses auxin and cell growth , whether this change might constitute a defense mechanism for the host benefit or create a favorable cellular niche for virus propagation remains for further investigation . In conclusion , our results support the notion that Geminivirus-encoded TrAP protein interferes with the TGS pathway and abrogates epigenetic silencing by direct interaction with KYP and inhibition of its transmethylation activity . Thus , Geminvirus TrAP functions clearly different from most of previously characterized viral suppressors , which target various steps of the PTGS pathway ( Ding and Voinnet , 2007 ) . Together with previous studies ( Raja et al . , 2008 ) , we provide evidence that KYP evolves as a critical immune system to control invading nucleic acids in plants; this is reminiscent of the roles of human SUV39H1 in maintaining the latency of HIV ( du Chene et al . , 2007 ) , Epsteinn-Barr virus ( Imai et al . , 2014 ) , and some other mammalian viruses . Thus , sequestering or interfering with this core component is an effective strategy for Geminivirus to block TGS and to subvert host defense; we expect this strategy to be used by other suppressors in plants and mammals . Given that many viral suppressors interrupt different steps in the PTGS pathways to efficiently combat host surveillance , it would not be surprising that additional key cellular factors in the TGS pathway , besides HMTase , were readily targeted by invading viruses in eukaryotes . Importantly , it is believed that compounds or drugs that alter chromatin methylation might ultimately be the most effective means of combating disease . Our discovery that TrAP inhibits a histone-modifying enzyme also offers a new natural strategy to develop epigenetic-targeted drugs to cure human diseases that arise from epigenetic dysfunction ( Højfeldt et al . , 2013 ) or to engineer new biotechnological products to improve agricultural productivity . All the plant constructs were made using the Gateway system ( Invitrogen , Carlsbad , CA , United States ) ( Zhang et al . , 2005 ) . The destination vectors ( containing the destination cassette–DC- ) pHyg-DC-CFP , pBA-DC-CFP , pBA-DC-3HA , pBA-DC , pBA-Flag-4Myc-DC , and pER10-YFP-DC ( Zhang et al . , 2006b ) were used for transient expression in N . benthamiana and for stable Arabidopsis thaliana transformation . The vector pER10 for betaestradiol-inducible expression under the XVE promoter ( Zuo et al . , 2000 ) was modified to obtain the destination vector pER10cLUC-3HA-DC . For this , the pER10 vector was linearized with XhoI , filled in with Klenow fragment; and further digested with PacI . In parallel , the cLuc-3HA-DC insert was obtained by linearizing the pCambia1300-cLuc-3HA-DC ( Zhang et al . , 2011 ) plasmid treated with SacI/Klenow treatment and further digested with PacI . The vector and insert were ligated and transformed into E . coli DB3 . 1 and finally confirmed by sequencing . This vector was used together with pCambia Myc-DC-nLUC ( Zhang et al . , 2011 ) for transient expression in N . benthamiana . The cDNA or DNA fragments were cloned into pENTR/D vectors , confirmed by sequencing , and then transferred to the appropriate destination vectors by recombination using the LR Clonase ( Invitrogen ) . The primers for the cloning are listed in the Supplementary file 8 . To drive KYP expression from its native promoter , we amplified a 2 . 7-Kb genomic region immediately upstream of the KYP start codon with the primers PKYP EcoRV for and PKYP BamHI rev . The binary vector pBA002a Flag-4Myc-KYP was obtained by the Gateway system using LR clonase . The resultant plasmid and PCR product harboring KYP promoter were digested with EcoRV/BamHI and ligated with T4 DNA ligase . The plasmid was confirmed by sequencing using the primer PKYP seq For . For expression of recombinant proteins , the cDNA or DNA fragments were cloned into pMCSG9 or pMCSG10 vectors to produce His-MBP- or His-GST-tagged proteins respectively , by ligation independent cloning ( Eschenfeldt et al . , 2009 ) using primers that include 18 nt identical to the ends of the linearized pMCSG vector ( Supplementary file 8 ) . The pMCSG plasmids were linearized with the blunt end restriction enzyme SspI , and the sticky ends were generated by T4 DNA Polymerase ( NEB , Ipswich , MA , United States ) supplemented with 2 . 5 mM dGTP . In parallel , the complementary sticky ends in the PCR products were generated supplementing the T4 DNA Polymerase ( NEB ) with 2 . 5 mM dCTP . The mixture of treated plasmid:PCR ( 3:4 , respectively ) was incubated on ice for 30 min to allow annealing of the complementary free strands and transformed in E . coli DH5α . The plasmids were further confirmed by sequencing and transformed into E . coli BL21 Rossetta DE3 for expression . To engineer the CaLCuV Δtrap DNA A infective clone the pNSB1090 plasmid was subjected to site directed mutagenesis by amplification with the primers CaLCuV_AL2_null_XbaI_for and CaLCuV_AL2_null_XbaI_rev ( Supplementary file 8 ) , digested overnight with DpnI , cleaned with the QIAquick PCR Purification Kit ( Qiagen , Germantown , MD , United States ) , and amplified in E . coli DH5α . The plasmid was confirmed by sequencing and transformed in Agrobacterium tumefasciens ABI for the virus infection assays . A . thaliana ( Col-0 ) plants were transformed with binary vectors by the floral-dip method ( Clough and Bent , 1998; Zhang et al . , 2006a ) . The transgenic seeds were selected on standard MS medium ( Murashige and Skoog , 1962 ) containing the appropriate selective agents: 10 mg/l glufosinate ammonium ( Sigma-Aldrich , St . Louis , MO , United States ) or 25 mg/l hygromycin ( Sigma ) , together with 100 mg/l carbenicillin ( Sigma ) . kyp mutant ( SALK_130630C ) was obtained from the Arabidopsis Biological Resource Center ( ABRC ) and confirmed by genotyping and qRT-PCR . For in vitro pull-down or HMTase reconstitution assays , the recombinant proteins were purified following either one or two-step affinity purification procedure . His-MBP-tagged proteins were first purified through Immobilized Metal Affinity Chromatography ( IMAC ) using the Ni-NTA resin ( Qiagen ) , followed by amylose resin ( NEB ) according to manufacturers' protocols . His-GST-tagged proteins were initially purified through sepharose glutathione column ( GE Healthcare , United Kingdom ) followed by IMAC . Specifically for HMTase assays , His-MBP-CaLCuV_TrAP , His-MBP-TGMV_TrAP , and His-MBP were prepared in lysis buffer ( 50 mM Tris-HCl pH 9 , 300 mM NaCl , 10 mM 2-mercaptoethanol , 2 mM PMSF ) , incubated with the Ni-NTA resin at 4°C for 1 hr , eluted with 300 mM imidazole and immediately incubated with the amylose resin ( NEB ) at 4°C for 1 hr . The proteins were eluted with 10 mM maltose and the elute was further separated by size exclusion chromatography ( SEC ) in column buffer ( 20 mM Tris-HCl pH 9 , 100 mM NaCl ) ; the fractions containing the target protein were pulled together , concentrated to 100 µM , aliquoted and stored at −80°C until usage . His-GST-KYP was prepared in PBS buffer ( 140 mM NaCl , 2 . 7 mM KCl , 10 mM Na2HPO4 , 1 . 8 mM KH2HPO4 , pH 7 . 3 , 10 mM 2-mercaptoethanol ) incubated with the sepharose glutathione 1 hr at 4°C , eluted with elution buffer ( 50 mM Tris-HCl pH 9 , 200 mM NaCl , 10 mM reduced glutathione , 10 mM 2-mercaptoethanol ) . The elute was further purified through the Ni-NTA column and finally through SEC in column buffer ( 20 mM Tris-HCl pH 9 , 100 mM NaCl ) ; the fractions containing the target protein were pulled together , concentrated to 25 μM and aliquoted for usage . For in vitro pull down assay , both the prey ( His-GST-CaLCuV-TrAP , His-GST-TGMV-TrAP , and His-GST ) and the bait ( His-MBP-Bait ) proteins were purified by IMAC using the lysis buffer ( 50 mM Tris-HCl pH 8 , 300 mM NaCl , 20 mM imidazole , 2 mM PMSF ) , incubated with the Ni-NTA resin at 4°C for 1 hr , eluted with 300 mM imidazole and immediately dialyzed in storage buffer ( 20 mM Tris-HCl pH 8 , 150 mM NaCl , 2 mM 2-mercaptoethanol , 50% Glycerol ) at 4°C overnight . In vitro pull-down assays and in vivo Co-IP were done as described ( Zhang et al . , 2005 ) . Briefly , 2 . 5 μg of 6His-GST-tagged prey proteins were pre-absorbed to 50 µl of the amylose resin ( NEB ) for 1 hr at 4°C in 1 ml of binding buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 2% glycerol , 0 . 6% Triton X-100 , 0 . 5 mM 2-mercaptoethanol , 2 mM PMSF ) . The proteins were recovered by ultracentrifugation at 12 , 000×g for 2 min , transferred to a second tube containing 2 . 5 µg of the MBP-tagged bait protein , and incubated at room temperature for 2 hr . The protein complexes were harvested by adding 50 µl amylose resin beads , followed by 2 hr incubation at room temperature , and cleaned with six vigorous washes with buffer . The pulled-down proteins were resolved by SDS-PAGE and the preys were detected by western blot using anti-GST antibody . For Co-IP experiments , N . benthamiana leaves were collected 2 days after agroinfiltration , ground in liquid nitrogen and stored at −80°C until use . For the assay , total proteins were extracted from 0 . 4 g of ground powder in 1 . 2 ml ( 3 vol ) of IP buffer ( 40 mM Tris-HCl pH 7 . 5 , 300 mM NaCl , 5 mM MgCl2 , 2 mM EDTA , 4 mM DTT , 0 . 5% Triton X-100 , 1 mM PMSF , 5% glycerol , 1 pellet/25 ml Complete EDTA-free protease inhibitor [Roche , Indianapolis , IN , United States] ) ; then , the soluble proteins were cleared twice by ultracentrifugation at 20 , 000 × rcf for 15 min at 4°C . The protein complexes were immunoprecipitated with 15 µl Anti-c-Myc-agarose affinity gel ( Sigma–Aldrich #A7470 ) at 4°C for 2 hr , the unspecific-bound proteins were removed by four consecutive washes with the IP buffer with 10 min incubation each at 4°C . The protein complexes were eluted in 200 μl of elution buffer ( 5 mM EDTA , 200 mM NH4OH ) for 20 min . The supernatant was collected , frozen in liquid nitrogen and dried using the Savant SpeedVac concentrator; finally , the sample was solubilized in 50 μl of 2× SDS-loading buffer for western blot analyses . 9-day-old wild-type control and Arabidopsis transgenic plants expressing XVE-Flag-Myc4-TrAP were induced for 16 hr with 25 µM ß-estradiol in liquid MS media , ground in liquid nitrogen and stored at −80°C until use . For the assay , total proteins were extracted from 10 g of ground powder in 40 ml ( 4 vol ) of IP buffer ( 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 4 mM MgCl2 , 50 μM ZnCl2 , 0 . 1% Triton X-100 , 1 mM PMSF , 1% glycerol , 1 pellet/17 ml Complete EDTA-free protease inhibitor [Roche] , 15 μM MG132 ) ; then , the soluble proteins were cleared twice by ultracentrifugation at 20 , 000 × rcf for 15 min at 4°C . The protein complexes were first immunoprecipitated using 500 μl of Anti-FLAG M2 magnetic beads ( Sigma-Aldrich , Cat# M8823 ) and incubated in slow rotation for 2 hr at 4°C , the nonspecific-bound proteins were removed by three consecutive washes with 15 ml of IP buffer for 10 min incubation each at 4°C . The protein complexes were then eluted by competition with 100 mg/ml FLAG peptide and subsequently immunoprecipitated with 100 µl Anti-c-Myc-agarose affinity gel ( Sigma–Aldrich #A7470 ) at 4°C for 1 . 5 hr , the nonspecific-bound proteins were removed by five consecutive washes with the IP buffer with 5 min incubation each at 4°C . The protein complexes were eluted in 200 μl of elution buffer ( 5 mM EDTA , 200 mM NH4OH ) for 20 min . The supernatant was collected , frozen in liquid nitrogen and dried using the Savant SpeedVac concentrator; finally , the sample was solubilized in 30 μl of 2× SDS-loading buffer and run in 10% SDS-PAGE . The samples were run to one-third of the gel , stained with Coomassie blue and collected by excising the whole lane for mass spectrometry analysis in the Taplin Mass Spectrometry Facility at Harvard Medical School . The plant material was lysed in CTAB buffer ( 100 mM Tris HCl pH 8 . 0 , 20 mM EDTA pH 8 . 0 , 1 . 4 M NaCl , 2% CTAB , 2% ß-mercaptoethanol ) ; then total DNA was extracted with phenol:chlorophorm:isoamyl alcohol ( 25:24:1 ) and precipitated with 2-propanol . The DNA was treated with RNase A and further purified with phenol:chlorophorm:isoamyl alcohol ( 25:24:1 ) and precipitated with ethanol , then dissolved in ultrapure water . The specified amount of DNA was separated by electrophoresis in 0 . 8% agarose , transferred overnight by capillarity to a Hybond-N membrane ( GE Healthcare ) , and probed with 32P-labeled probe targeting the CR region of CaLCuV DNA A . The probe was obtained by PCR using the primers CR_F and CR_R ( Supplementary file 8 ) and labeled using the Rediprime II DNA Labeling System ( GE Healthcare ) following the manufacturer's instructions . Total RNA was extracted using Trizol reagent from either adult rosette leaves or 2-week-old seedlings of independent transgenic lines , the RNA blots were then performed as previously described ( Zhang et al . , 2006b ) . Immunoprecipitation of Flag-AGO1-associated small RNAs were performed as described ( Zhang et al . , 2006b ) . RNA was recovered with Trizol reagent from the immunoprecipitates , separated in 8 M urea , 15% polyacrylamide gels and subjected to RNA blot analysis of low-molecular-weight RNAs . The LCI was performed on 4-week-old N . benthamiana leaves infiltrated with various combinations of A . tumefaciens GV3101 harboring pCambia Myc-TrAP-nLUC or pCambia Myc-nLUC and A . tumefaciens ABI carrying pER10cLUC-3HA or pER10cLUC-3HA-candidate proteins . The agrobacteria containing the pER10 plasmids were incubated with 25 µM beta-estradiol for 3 hr prior infiltration , and all the cultures were adjusted to OD600 = 0 . 8 . The transfected leaves were assayed 2 days after agroinfiltration by adding the substrate ( 10 mM luciferin ) . The sprayed leaves were incubated in total darkness for 5 min and photographed using an electronmultiplying charge-coupled device ( EMCCD ) camera , Cascade II 512 , from Photomerics ( Roper Scientific , Trenton , NJ , United States ) . The images were processed with WinView32 Ver 2 . 5 . 19 . 7 ( Roper Scientific ) . Leaves of 4-week-old tobacco plants ( N . benthamiana ) were agroinfiltrated with syringe without needle as previously described ( Zhang et al . , 2005 ) with A . tumefasciens ABI carrying pBA-TrAP-CFP and pER10-YFP-Test protein . The agrobacteria containing the pER10 plasmids were incubated with 25 µM beta-estradiol for 3 hr prior infiltration , and all the cultures were adjusted to OD600 = 0 . 8 . The plants were maintained for 2 days at 24°C ( 16 hr light/8 hr dark ) . The co-localization was evaluated using a Nikon inverted microscope Eclipse Ti-E ( Nikon , Japan ) . CFP signal was measured by excitation with Shutter 10-3 filter 3 ( CFPHQ [Ex] ) , and emission was detected at 485 nm; YFP used Shutter 10-3 filter 4 ( YFPHQ [Ex] ) and emission was detected at 540 nm . The images were processed using NIS-Elements-AR 4 . 30 . 01 ( Nikon ) and Adobe Photoshop software . FRET-AB experiments were performed on N . benthamiana epidermal cells of 4-week-old leaves agroinfiltrated with a 1:1 mixture of pBA-TrAP-CFP and pER10-YFP-KYP to a final OD600 = 0 . 8 . YFP and CFP signals were captured with a Zeiss LSM 710 confocal Microscope ( Zeiss , Germany ) . FRET was determined by the acceptor photobleaching method ( Kenworthy , 2001; Daelemans et al . , 2004 ) . First , to define the base line , the signal intensities of a pre-photobleach CFP ( donor ) and YFP ( acceptor ) were acquired by exciting with the 458 and 514-nm laser lines , respectively . Then , three regions of interest in the cell were selected: #1 , Autofluorescence control; #2 , non-photobleaching control; and #3 , FRET-AB region . The CFP donor was excited with the 458 nm laser line for all FRET experiments; the emission of both CFP and YFP was recorded at 485 nm and 540 nm . Regions #1 and #3 were rendered free of YFP by consecutive cycles of bleaching recovery with the 514-nm laser line until no recovery of YFP was detected . The CFP and YFP signals were monitored throughout the experiment . After correction for background with control region #1 and for photobleaching of the donor because of imaging with control region #2 , the FRET efficiencies ( E ) in the region #3 was calculated from the CFP signal using FRETEff=1−DpreDpost , where D is the mean intensity of the donor CFP in the area where the acceptor was bleached , before ( Dpre ) and after ( Dpost ) acceptor bleaching . The FRET efficiency is considered positive when Dpost > Dpre . The image and statistical analyses were performed with the FRET module for the ZEN software ( Zeiss ) . The average FRET efficiency and its standard deviation were calculated from the FRET efficiencies of each individual cell in 27–30 cells per experiment . The standard Student's t-test was used to determine the statistical significance of the results . In vitro HMTase reactions were modified from ( Rea et al . , 2000; Tachibana et al . , 2001 ) as follows: 20 µl of reaction mixture containing 3 . 3 µM Histone 3 . 2 ( NEB ) , 1 µM His-GST-KYP , and 50 nCi of S-adenosyl-[methyl-14C]-L-methionine in HMTase buffer ( 50 mM Tris-HCl pH 9 , 10 mM MgCl2 , 1 mM ß-mercaptoethanol , 250 mM sucrose ) was incubated for 0–10 min at 37°C . The reaction products were separated by 18% SDS-polyacrylamide gel electrophoresis and visualized by Coomassie Brilliant Blue R-250 staining; then , the gels were fixed 1 hr in fixing solution ( 25% Ethanol , 2% Glycerol ) and scintillated for 30 min in 1 M sodium salicylate . Gels were dried 2 hr at 80°C . The 14C signal was detected by fluorography using in a preflashed Classic autoradiography film blue sensitive; Filters Kodak Wratten No . 22 and No . 96 were used together for preflashing . Preflashed film was exposed 5–7 days at −80°C . The film was developed using a Kodak M35A X-OMAT Processor , and the results were digitalized in a Chemi-Doc XRS System and analyzed with the Image Lab Software ( Bio-Rad , Hercules , CA , United States ) . To assess the impact of TrAP on KYP activity , His-MBP-TrAP or His-MBP were pre-incubated with His-GST-KYP in different molar ratios , ranging from 0 to 10 , for 1 hr at room temperature , then the assays were proceeded as described above . The experiments were performed 3–5 times for statistical analysis . The analysis of histone modifications was performed as described ( Saleh et al . , 2008 ) . Two grams of 9-day-old seedlings were crosslinked with 1% formaldehyde for 10 min by vacuum infiltration at 4°C; the reaction was stopped with 2 M Glycine to a final concentration of 100 mM at room temperature . Plants were rinsed 5 times with ice cold with water , flash-frozen in liquid nitrogen , and ground with mortar and pestle . The powder was suspended in 6 vol ( 12 ml ) of nuclei isolation buffer ( 15 mM PIPES-KOH pH 6 . 8 , 0 . 25 M sucrose , 0 . 9% Triton X-200 , 5 mM MgCl2 , 60 mM KCl , 15 mM NaCl , 1 mM CaCl2 , 1 mM PMSF , 1 pellet/50 ml Complete EDTA-free Protease inhibitor [Roche] ) , filtered through two layers of Miracloth and centrifuged at 11 , 000 × rcf for 10 min in 4°C . After discarding the supernatant , the pellet was resuspended in 1 ml of Nuclei lysis buffer ( 50 mM HEPES pH 7 . 5 , 1 mM EDTA pH 8 . 0 , 150 mM NaCl , 1% SDS , 0 . 1% Sodium Deoxycholate , 1% Triton X-100 , 1 pellet/50 ml Complete EDTA-free Protease inhibitor [Roche] ) ; the samples were sonicated in 10 cycles 30 s ON and 90 s OFF , using the Bioruptor ( Diagenode , Belgium ) at the highest power in 4°Cs . The sonicated samples were centrifuged for 10 min at 21 , 000 × rcf in 4°C . 100 μl of the clarified chromatin was diluted 10-fold with Nuclei lysis buffer without SDS for each assay . The immunoprecipitation was accomplished by the addition of 40 µl Protein A Dynabeads ( Invitrogen ) and 3 μl of the pertinent antibody , followed by 6 hr incubation at 4°C on mild rotation . The beads-conjugated complexes were washed with 1 ml of Low salt buffer ( 20 mM Tris-HCl pH 8 . 0 , 2 mM EDTA pH 8 . 0 , 150 mM NaCl , 0 . 5% Triton X-100 , 0 . 2% SDS ) , followed by 1 ml of high salt buffer ( 20 mM Tris-HCl pH 8 . 0 , 2 mM EDTA pH 8 . 0 , 500 mM NaCl , 0 . 5% Triton X-100 , 0 . 2% SDS ) , then with 1 ml of LiCl buffer ( 10 mM Tris-HCl pH 8 . 0 , 1 mM EDTA pH 8 . 0 , 250 mM LiCl , 1% sodium deoxycholate , 1% NP-40 ) , and finally twice with 1 ml of TE ( 10 mM Tris-HCl , 1 mM EDTA , pH 8 . 0 ) by incubating 5 min at 4°C in between washes . The samples were eluted twice at room temperature with 250 μl of elution buffer ( 100 mM NaHCO3 , 0 . 5% SDS ) , for 15 and 30 min , respectively . The samples were decrosslinked with 100 mM NaCl at 65°C overnight , followed by Proteinase K treatment for 90 min at 45°C . The DNA was purified by Phenol:Chloroform:IsoamylAlcohol 25:24:1 , and precipitated in 100% Ethanol at −80°C . The antibodies used are mono-clonal anti-H3K9me2 ( Abcam , #Ab1220 ) ; monoclonal anti-H3K4me3 ( Millipore , cat #04-745 ) ; mono-clonal anti-H3K27me3 ( Millipore , cat #07-449 ) . The immunoprecipitation of Flag-4Myc-KYP-Chromatin complexes was done as in ( Wierzbickiet al . , 2008 ) , using Anti-FLAG M2 magnetic beads ( Sigma-Aldrich , Cat# M8823 ) . Two grams of rosette leaves 1–12 of mock or CaLCuV inoculated plants at 18 dpi were crosslinked with 1% formaldehyde for 25 min by vacuum infiltration at 4°C; the reaction was stopped with 2 M Glycine to a final concentration of 100 mM . Plants were rinsed five times with ice cold with water , flash-frozen in liquid nitrogen , and ground with mortar and pestle . The powder was suspended in 5 vol ( 10 ml ) of Honda Buffer ( 20 mM HEPES-KOH pH 7 . 4 , 0 . 44 M sucrose , 1 . 25% ficoll , 2 . 5% Dextran T40 , 10 mM MgCl2 , 0 . 5% Triton X-100 , 5 mM DTT , 2 mM PMSF , 1 pellet/25 ml Complete EDTA-free Protease inhibitor [Roche] ) , filtered through two layers of Miracloth and centrifuged at 2000 × rcf for 15 min in 4°C . After discarding the supernatant , the nuclear precipitates were washed three times with 1 ml of Honda buffer; subsequently , the pellet was suspended in 300 μl of Nuclei lysis buffer ( 50 mM Tris-HCl pH 8 . 0 , 10 mM EDTA pH 8 . 0 , 1% SDS , 2 mM PMSF , 1 pellet/25 ml Complete EDTA-free Protease inhibitor [Roche] ) and sonicated in ten cycles 30 s ON and 90 s OFF , using the Bioruptor ( Diagenode ) at the highest power in 4°Cs . The sonicated samples were centrifuged for 10 min at 21000 × rcf in 4°C . 100 μl of the clarified chromatin was diluted 10-fold with ChIP dilution buffer ( 16 . 7 mM Tris-HCl pH 8 . 0 , 1 . 2 mM EDTA pH 8 . 0 , 167 mM NaCl , 1 . 1% Triton X-100 , 1 pellet/25 ml Complete EDTA-free Protease inhibitor [Roche] ) per ChIP . The immunoprecipitation was accomplished by the addition of 40 µl of Anti-FLAG M2 magnetic beads ( Sigma-Aldrich , Cat# M8823 ) , followed by 2 hr incubation at 4°C on mild rotation . The beads were washed five times with 1 ml of Washing buffer ( 20 mM Tris-HCl pH 8 . 0 , 2 mM EDTA pH 8 . 0 , 150 mM NaCl , 1% Triton X-100 , 1% SDS , 2 mM PMSF , 1 pellet/25 ml Complete EDTA-free Protease inhibitor [Roche] ) incubating 5 min at 4°C in between; then , two more washes with 1 ml TE buffer incubating 5 min at 4°C . Finally , the samples were eluted twice at room temperature with 125 μl of Elution buffer ( 100 mM NaHCO3 , 0 . 5% SDS ) , for 15 and 30 min , respectively . The samples were decrosslinked and the DNA extracted as above . Microarray analyses using the Affymetrix ATH1 platform were performed with two biological replicates using wild-type plants , 35S-TrAP transgenic plants , and lhp1 mutants . Seedlings were grown on MS medium with 1% sucrose for 7 days . One mg of total RNA was used for reverse transcription using MessageAmp II aRNA kits ( Ambion ) and 15 mg of labeled cRNA for hybridization . GeneChip hybridization and scanning were performed at the Genomics Resource Center , Rockefeller University , New York . Statistical analysis of microarray data was performed using R software . Initially the microarray plates were tested for quality by an M plot and the data normalized by the RMA method from the Affy package . Subsequently , the distribution of the samples was assessed with scatter plots and the normalized data sets were evaluated with the Moderate t-test from R package limma for p-value computation . Then , the eBays method was used to compute moderated t-statistics and log-odds of differential expression by empirical Bayes shrinkage of the standard errors towards a common value . The moderated t-statistic ( t ) is the logFC to its standard error . In our DEG results our thresholds are p-value < 0 . 05 and logFC >1 ( up-regulated ) or logFC < −1 ( down-regulated ) . The False Discovery Rate was approximated from the eBayes adjusted p-value . The significance of the overlapping data sets was calculated through Pearson's Chi-squared test with 1° of freedom . Expression levels of the tested genes were examined by quantitative RT-PCR . Total RNAs were prepared from 9 days-old seedlings and treated with DNase before being subjected to cDNA synthesis using Superscript III reverse transcriptase ( Invitrogen ) primed by random primers . The EF1a gene ( Williams et al . , 2005 ) was included as an internal control for normalization . The enrichment levels of specific genes after ChIP assay were also tested by quantitative PCR . Primers are listed in Supplementary file 8 . The quantitative PCRs were performed in 384-well plates with an ABI7900HT real-time PCR system using the SYBR Green I master mix ( Applied Biosystems , Waltham , MA , United States ) in a volume of 10 μl . PCR conditions were as follows: 50°C for 2 min , 95°C for 10 min , 45 cycles of 96°C for 10 s followed by 60°C for 1 min . Three biological repeats were performed , and the reactions were performed in triplicate for each run . The comparative CT method was used to evaluate the relative quantities of each amplified product in the samples . The threshold cycle ( CT ) was automatically determined for each reaction by the system according to the default parameters . The specificity of the PCR was determined by dissociation curve analysis of the amplified products using the standard method installed in the system . Approximately 500 ng of genomic DNA were used to generate libraries as described ( Feng et al . , 2011 ) using premethylated adapters ( NEXTFlex Bisulfite-Seq Adapters #511911 , Bioo Scientific , Austin , TX , United States ) . The adaptor-ligated fragments were purified by QIAQuick column ( Qiagen ) and bisulfite converted using the EpiTect Kit ( Qiagen , Germany ) following the manufacturer's instructions . The converted DNA was later enriched by 15 cycles of PCR using the Pfu Turbo Cx Polymerase ( Agilent , Santa Clara , CA , United States ) , using the specific primers provided by Bios Scientific for enrichment . The library was finally purified with Agencourt AMPure XP beads ( Beckman Coulter , Pasadena , CA , United States ) according to manufacturer's instruction . The libraries were single-end sequenced using HiSeq High Output with read length of 50 bp . Base calling and sequence cleaning was performed with the standard Illumina software , then the clean reads were mapped to the Arabidopsis genome ( Version: TAIR10 ) using Bismark v0 . 14 . 3 ( Krueger and Andrews , 2011 ) with default parameters; the PCR duplicates were removed , and the methylation information was obtained with bismark with cutoff 3 . The DMRs were obtained using swDMR ( Wang et al . , 2015 ) with window 200 , step size 100 , ( left 1000 , right 1000 ) , the samples were compared using the Kruskal–Wallis analysis of variance with p-value < 0 . 01 . The DMRs were then annotated using BEDtools ( Quinlan et al . , 2010 ) . 3-week-old Col-0 wild-type , kyp mutant and Flag-4Myc-KYP complemented were infected by agroinfiltration of the CaLCuV infective clones of DNAA and DNAB pNSB1090 and pNSB1091 ( Arguello-Astorga et al . , 2007 ) ; the progression of the disease was evaluated daily in terms of time of symptom development and the severity of the symptoms observed . The assays were replicated 3 times , 36 plants of each genotype were used per assay , grown in short day conditions ( 8 hr light/16 hr dark ) . To assess systemic infection , we conducted virus infection assays using CaLCuV or CaLCuV Δtrap on thirty wild-type and thirty kyp mutant plants grown in short day condition ( 16 hr dark/8 hr light ) at eight true leaves developmental stage , then we harvested nine to eleven newly emerged rosette leaves of each individual 18 days after inoculation ( Figure 9B ) .	Many viruses can infect plants and cause diseases that can reduce crop yields . The Geminiviruses are a family of plant viruses that are transmitted by insects and infect tomato , cabbage , and many other crop plants . These viruses hijack the plant cells that they infect and force the plant cells to make viral proteins using instructions provided by the genes in the virus' own DNA . To make proteins , DNA is first copied into molecules of messenger ribonucleic acid ( or mRNA ) in a process called transcription . However , plants can defend themselves by blocking the transcription of viral DNA through ‘transcriptional gene silencing’ . In plant cells , DNA is packaged around proteins called histones to form a structure called chromatin . Small chemical tags attached to the histones can alter the structure of chromatin to regulate the activity of the genes encoded within it . For example , ‘methyl’ tags added to certain histones can block transcription and lower the activity of a gene . DNA from viruses can also associate with histones inside plant cells meaning that transcriptional gene silencing can take place by the addition of these methyl tags . Many Geminiviruses produce a protein called TrAP , which can activate transcription , but it is not clear how this works . Castillo-González et al . studied the TrAP proteins from two different Geminiviruses that can infect crop plants . The commonly used model plant , Arabidopsis thaliana , was genetically engineered to produce high levels of these TrAP proteins . These ‘transgenic’ plants did not develop properly: they grew more slowly , had abnormal leaves , and flowered earlier . Furthermore , hundreds of plant genes were more active than usual in the transgenic plants , which suggests that TrAP inhibits transcriptional gene silencing . Further experiments showed that TrAP directly binds to a plant enzyme called KYP—which normally deposits methyl groups on chromatin and prevents it from working . This reduces the number of methyl groups that are attached to histones associated with both viral and plant chromatin , which results in the activation of genes that would normally be switched off . Castillo-González et al . 's findings show how Geminiviruses can stop transcriptional gene silencing of chromatin that contains virus DNA to evade the host plant's defenses . The next challenge is to understand how TrAP inhibits KYP , which may present new ways to genetically engineer plants to become resistant to infection by viruses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "plant", "biology" ]	2015	Geminivirus-encoded TrAP suppressor inhibits the histone methyltransferase SUVH4/KYP to counter host defense
During transcription initiation , RNA polymerase ( RNAP ) binds to promoter DNA , unwinds promoter DNA to form an RNAP-promoter open complex ( RPo ) containing a single-stranded ‘transcription bubble , ’ and selects a transcription start site ( TSS ) . TSS selection occurs at different positions within the promoter region , depending on promoter sequence and initiating-substrate concentration . Variability in TSS selection has been proposed to involve DNA ‘scrunching’ and ‘anti-scrunching , ’ the hallmarks of which are: ( i ) forward and reverse movement of the RNAP leading edge , but not trailing edge , relative to DNA , and ( ii ) expansion and contraction of the transcription bubble . Here , using in vitro and in vivo protein-DNA photocrosslinking and single-molecule nanomanipulation , we show bacterial TSS selection exhibits both hallmarks of scrunching and anti-scrunching , and we define energetics of scrunching and anti-scrunching . The results establish the mechanism of TSS selection by bacterial RNAP and suggest a general mechanism for TSS selection by bacterial , archaeal , and eukaryotic RNAP . During transcription initiation , RNA polymerase ( RNAP ) and one or more transcription initiation factor bind to promoter DNA through sequence-specific interactions with core promoter elements , unwind a turn of promoter DNA to form an RNAP-promoter open complex ( RPo ) containing an unwound ‘transcription bubble , ’ and select a transcription start site ( TSS ) . The distance between core promoter elements and the TSS can vary . TSS selection is a multi-factor process , in which the outcome reflects the contributions of promoter sequence and reaction conditions . TSS selection by bacterial RNAP and the bacterial transcription initiation factor σ involves four promoter-sequence determinants: ( i ) distance relative to the promoter −10 element ( preference for TSS selection at the position 7 bp downstream of the promoter −10 element; Aoyama and Takanami , 1985; Sørensen et al . , 1993; Jeong and Kang , 1994; Liu and Turnbough , 1994; Walker and Osuna , 2002; Lewis and Adhya , 2004; Vvedenskaya et al . , 2015; Winkelman et al . , 2016a; Winkelman et al . , 2016b ) ; ( ii ) identities of the template-strand nucleotide at the TSS and the template-strand nucleotide immediately upstream of the TSS ( strong preference for a template-strand pyrimidine at the TSS and preference for a template-strand purine immediately upstream of the TSS; Aoyama and Takanami , 1985; Sørensen et al . , 1993; Jeong and Kang , 1994; Liu and Turnbough , 1994; Walker and Osuna , 2002; Lewis and Adhya , 2004; Vvedenskaya et al . , 2015; Winkelman et al . , 2016a; Winkelman et al . , 2016b ) ; ( iii ) the promoter ‘core recognition element , ’ a segment of nontemplate-strand sequence spanning the TSS that interacts sequence-specifically with RNAP ( preference for nontemplate-strand G immediately downstream of the TSS; Vvedenskaya et al . , 2016 ) , and ( iv ) the promoter ‘discriminator element , ’ a nontemplate-strand sequence immediately downstream of the promoter −10 element that interacts sequence-specifically with σ ( preference for TSS selection at upstream positions for purine-rich discriminator sequences , and preference for TSS selection at downstream positions for pyrimidine-rich discriminator sequences; Winkelman et al . , 2016a , 2016b ) . In addition to these four promoter-sequence determinants , the concentrations of initiating NTPs ( Sørensen et al . , 1993; Liu and Turnbough , 1994; Walker and Osuna , 2002; Vvedenskaya et al . , 2015; Wilson et al . , 1992; Qi and Turnbough , 1995; Tu and Turnbough , 1997; Walker et al . , 2004; Turnbough , 2008; Turnbough and Switzer , 2008 ) and DNA topology ( Vvedenskaya et al . , 2015 ) also influence TSS selection . It has been hypothesized that variability in the distance between core promoter elements and the TSS is accommodated by DNA ‘scrunching’ and ‘anti-scrunching , ’ the defining hallmarks of which are: ( i ) forward and reverse movements of the RNAP leading edge , but not the RNAP trailing edge , relative to DNA , and ( ii ) expansion and contraction of the transcription bubble ( Vvedenskaya et al . , 2015; Winkelman et al . , 2016a , 2016b; Vvedenskaya et al . , 2016; Robb et al . , 2013 ) . In previous work , we showed that TSS selection exhibits the first hallmark of scrunching in vitro ( Winkelman et al . , 2016a ) . Here , we show that TSS selection also exhibits the first hallmark of scrunching in vivo , show that TSS selection exhibits the second hallmark of scrunching and anti-scrunching , and define the energetics of scrunching and anti-scrunching . In our prior work , we demonstrated that bacterial TSS selection in vitro exhibits the first hallmark of scrunching by defining , simultaneously , the TSS , the RNAP leading-edge position , and RNAP trailing-edge position for transcription complexes formed on a library of 106 promoter sequences ( Winkelman et al . , 2016a ) . We used RNA-seq to define the TSS , and we used unnatural-amino-acid-mutagenesis , incorporating the photoactivatable amino acid p-benzoyl-L-phenylalanine ( Bpa ) , and protein-DNA photocrosslinking to define RNAP-leading-edge and trailing-edge positions ( Winkelman et al . , 2016a ) . The results showed that the discriminator element ( Haugen et al . , 2006; Feklistov et al . , 2006 ) influences TSS selection and does so through effects on sequence-specific σ-DNA interaction that select between two alternative paths of the DNA nontemplate strand ( Winkelman et al . , 2016a ) . The results further showed that , as the TSS changes for different discriminator sequences , the RNAP-leading-edge position changes , but the RNAP-trailing-edge position does not change ( Winkelman et al . , 2016a ) . For example , replacing a GGG discriminator by a CCT discriminator causes a 2 bp downstream change in TSS ( from the position 7 bp downstream of the −10 element to the position 9 bp downstream of the −10 element , due to differences in sequence-specific σ-DNA interaction that result in different paths of the DNA nontemplate strand ) , causes a 2 bp downstream change in RNAP leading-edge position , but does not cause a change in RNAP trailing-edge position ( Figure 1A ) . Here , to determine whether bacterial TSS selection in vivo also exhibits the first hallmark of scrunching , we adapted the above unnatural-amino-acid-mutagenesis and protein-DNA-photocrosslinking procedures to define RNAP leading-edge and trailing-edge positions in TSS selection in living cells ( Figure 1 , Figure 1—figure supplements 1–2 ) . We developed approaches to assemble , trap , and UV-irradiate RPo formed by a Bpa-labeled RNAP derivative in living cells , to extract crosslinked material from cells , and and to map crosslinks at single-nucleotide resolution ( Figure 1—figure supplement 1 ) . In order to assemble , trap , and UV-irradiate RPo in living cells , despite the presence of high concentrations of initiating substrates that rapidly convert RPo into transcribing complexes , we used a mutationally inactivated RNAP derivative , β’D460A , that lacks a residue required for binding of the RNAP-active-center catalytic metal ion and initiating substrates ( Zaychikov et al . , 1996 ) ( Figure 1—figure supplements 1–2 ) . Control experiments confirm that , in vitro , in both the absence and presence of initiating substrates , the mutationally inactivated RNAP derivative remains trapped in RPo , exhibiting the same pattern of leading-edge and trailing-edge crosslinks as for wild-type RNAP in the absence of initiating substrates ( Figure 1—figure supplement 2 ) . In order to introduce Bpa at the leading-edge and trailing-edge of RPo in living cells , we co-produced , in Escherichia coli , a Bpa-labeled , decahistidine-tagged , mutationally inactivated RNAP derivative in the presence of unlabeled , untagged , wild-type RNAP , using a three-plasmid system comprising ( i ) a plasmid carrying a gene for RNAP β’ subunit that contained a nonsense codon at the site for incorporation of Bpa , the βʹD460A mutation , and a decahistidine coding sequence; ( ii ) a plasmid carrying genes for an engineered Bpa-specific nonsense-suppressor tRNA and an engineered Bpa-specific aminoacyl-tRNA synthase ( Chin et al . , 2002 ) ; and ( iii ) a plasmid containing a promoter of interest ( Figure 1—figure supplement 1A ) . ( Using this merodiploid system , with both a plasmid-borne mutant gene for βʹ subunit and a chromosomal wild-type gene for βʹ subunit , enabled analysis of the mutationally inactivated RNAP derivative without loss of viability . ) In order to perform RNAP-DNA crosslinking and to map resulting RNAP-DNA crosslinks , we then grew cells in medium containing Bpa , UV-irradiated cells , lysed cells , purified crosslinked material using immobilized metal-ion-affinity chromatography targeting the decahistidine tag on the Bpa-labeled , decahistidine-tagged , mutationally inactivated RNAP derivative , and mapped crosslinks using primer extension ( Figure 1—figure supplement 1B ) . The results showed an exact correspondence of crosslinking patterns in vitro and in vivo ( Figure 1B , ‘in vitro’ vs . ‘in vivo’ lanes ) . The RNAP leading edge crosslinked 2 bp further downstream on CCT than on GGG , whereas the RNAP trailing edge crosslinked at the same positions on CCT and GGG ( Figure 1B ) . We conclude that TSS selection in vivo shows the first hallmark of scrunching . The results in Figure 1 establish that TSS selection in vitro and in vivo exhibits the first hallmark of scrunching . However , definitive demonstration that TSS selection involves scrunching also requires demonstration of the second hallmark of scrunching: that is , changes in transcription-bubble size . To determine whether bacterial TSS selection exhibits the second hallmark of scrunching we used a magnetic-tweezers single-molecule DNA-nanomanipulation assay that enables detection of RNAP-dependent DNA unwinding with near-single-base-pair spatial resolution and sub-second temporal resolution ( Revyakin et al . , 2004 , 2005 , 2006 ) to assess whether TSS selection correlates with transcription-bubble size for GGG and CCT promoters ( Figure 2 ) . The results indicate that transition amplitudes for RNAP-dependent DNA unwinding upon formation of RPo with CCT are larger than those for formation of RPo with GGG , on both positively and negatively supercoiled templates ( Figure 2B , left and center ) . Transition-amplitude histograms confirm that transition amplitudes with CCT are larger than with GGG , on both positively and negatively supercoiled templates ( Figure 2B , right ) . By combining the results with positively and negatively supercoiled templates to deconvolve effects of RNAP-dependent DNA unwinding and RNAP-dependent compaction ( Revyakin et al . , 2004 , 2005 , 2006 ) , we find a 2 bp difference in RNAP-dependent DNA unwinding for CCT vs . GGG ( Figure 2C ) , corresponding exactly to the 2 bp difference in TSS selection ( Figure 1B ) . We conclude that TSS selection shows the second hallmark of scrunching . According to the hypothesis that TSS selection involves scrunching or anti-scrunching , TSS selection at the most frequently observed , modal TSS position ( 7 bp downstream of −10 element for majority of discriminator sequences , including GGG ) involves neither scrunching nor anti-scrunching , TSS selection downstream of the modal position involves scrunching ( transcription-bubble expansion ) , and TSS selection upstream of the modal position involves anti-scrunching ( Vvedenskaya et al . , 2015; Winkelman et al . , 2016a , 2016b; Vvedenskaya et al . , 2016; Robb et al . , 2013 ) . The results in Figures 1–2 apply to the modal TSS position and a TSS position 2 bp downstream of the modal TSS position . To generalize and extend the results to a range of different TSS positions , including a position upstream of the modal TSS position expected to involve anti-scrunching , we exploited the ability of oligoribonucleotide primers ( ‘nanoRNAs’; Goldman et al . , 2011 ) to program TSS selection ( Figure 3A , Figure 3—figure supplement 1 ) . We analyzed a consensus bacterial promoter , lacCONS , and used four ribotrinucleotide primers , UGG , GGA , GAA , and AAU , to program TSS selection at positions 6 , 7 , 8 , and 9 bp downstream of the −10 element ( Figure 3A ) . Experiments analogous to those in Figure 1 show a one-for-one , bp-for-bp correlation between primer-programmed changes in TSS and changes in RNAP-leading-edge position . The leading-edge crosslink positions with the four primers differed in single-nucleotide increments , but the trailing-edge crosslink positions were the same ( Figure 3B ) . With the primer GGA , which programs TSS selection at the modal position ( 7 bp downstream of −10 element for this discriminator sequence ) , the leading-edge crosslinks were exactly as in experiments with no primer ( Figure 3—figure supplement 1 ) . With primers GAA and AAU , which program TSS selection 1 and 2 bp downstream ( positions 8 and 9 ) , leading-edge crosslinks were 1 and 2 bp downstream of crosslinks with GGA ( Figure 3B ) . With primer UGG , which programs TSS selection 1 bp upstream ( position 6 ) , leading-edge crosslinks were 1 bp upstream of crosslinks with GGA ( Figure 3B ) . The results show that successive single-base-pair changes in TSS selection are matched by successive single-base-pair changes in RNAP leading-edge position . We conclude that the first hallmark of scrunching is observed for a full range of TSS positions , including , importantly , a position upstream of the modal TSS expected to involve anti-scrunching . We next used magnetic-tweezers single-molecule DNA-nanomanipulation to analyze primer-programmed TSS selection . To enable single-base-pair resolution , we reduced the DNA-tether length from 2 . 0 kb to 1 . 3 kb , thereby reducing noise due to compliance ( Figure 4—figure supplement 1; see Revyakin et al . , 2005 ) . The resulting transition amplitudes , transition-amplitude histograms , and RNAP-dependent DNA unwinding values for TSS selection with saturating concentrations of the four primers show a one-for-one , base-pair-for-base-pair correlation between primer-programmed changes in TSS and changes in RNAP-dependent DNA unwinding ( Figure 4 ) . With primer GGA , which programs TSS selection at the modal position ( 7 bp downstream of the −10 element for this discriminator sequence ) , DNA unwinding was exactly as in experiments with no primer ( Figure 4—figure supplement 2 ) . With primers GAA and AAU , which program TSS selection 1 and 2 bp further downstream ( positions 8 and 9 ) , DNA unwinding was ~1 and ~2 bp greater than with GGA ( Figure 4 ) . With primer UGG , which programs TSS selection 1 bp upstream , DNA unwinding was ~1 bp less than that in experiments with GGA ( Figure 4 ) . The results show that successive single-base-pair changes in TSS selection are matched by successive single-base-pair changes in DNA unwinding for a full range of TSS positions including , importantly , a position upstream of the modal TSS expected to involve anti-scrunching . Taken together , the results of protein-DNA photocrosslinking ( Figure 3 , Figure 3—figure supplement 1 ) and DNA-nanomanipulation ( Figure 4 , Figure 4—figure supplement 2 ) demonstrate , definitively , the scrunching/anti-scrunching hypothesis for TSS selection . To quantify the energetic costs of scrunching and anti-scrunching , we measured primer-concentration dependences of lifetimes of unwound states ( Figures 5–6 , Figure 6—figure supplement 1 ) . For each primer , increasing the primer concentration increases the lifetime of the unwound state ( tunwound ) , as expected for coupled equilibria of promoter unwinding , promoter scrunching , and primer binding ( Figures 5–6 ) . The results in Figure 6D show that the slopes of plots of mean tunwound ( t¯unwound ) vs . primer concentration differ for different primers . Fitting the results to the equation describing the coupled equilibria ( Figure 6C ) yields values of KNpNpN , ΔGNpNpN , Kscrunch , and ΔGscrunch for the four primers ( Figure 6E , Figure 6—figure supplement 1 ) . The results indicate that scrunching by 1 bp requires 0 . 7 kcal/mol , scrunching by 2 bp requires 1 . 7 kcal/mol , and anti-scrunching by 1 bp requires 1 . 8 kcal/mol ( Figure 6E , Figure 6—figure supplement 1 ) . The results provide the first experimental determination of the energetic costs of scrunching and anti-scrunching in any context . We hypothesize that energetic costs on the same scale , ~0 . 7–1 . 8 kcal/mol per scrunched bp , also apply in the structurally and mechanistically related scrunching that occurs during initial transcription by RNAP ( Revyakin et al . , 2006; Kapanidis et al . , 2006 ) . We note that , according to this hypothesis , the scrunching by ~10 bp that occurs during initial transcription ( Revyakin et al . , 2006 ) results in an increase in the state energy of the transcription initiation complex by a total of ~7–18 kcal/mol ( ~10 x ~0 . 7–1 . 8 kcal/mol ) . This is an increase in state energy potentially sufficient to yield a ‘stressed intermediate’ ( Revyakin et al . , 2006; Straney and Crothers , 1987 ) having scrunching-dependent ‘stress’ comparable to the free energies of RNAP-promoter and RNAP-initiation-factor interactions that anchor RNAP at a promoter ( ~7–9 kcal/mol for sequence-specific component of RNAP-promoter interaction and ~13 kcal/mol for RNAP-initiation-factor interaction; 24 ) and therefore is an increase in state energy potentially sufficient to pay energetic costs of breaking RNAP-promoter and RNAP-initiation-factor interactions in the transition from transcription initiation to transcription elongation . The ΔGscrunch values for bacterial TSS selection obtained in this work account for the range of TSS positions and the relative utilization of different TSS positions in bacterial transcription initiation . The ΔGscrunch values for TSS selection at positions 6 , 7 , 8 , and 9 of a promoter where the modal TSS is position 7 ( 0–1 . 8 kcal/mol ) all are less than or comparable to 3kBT ( ~2 kcal/mol ) , where kB is the Bolztmann constant and T is temperature in °K , indicating that TSS selection at these positions requires no energy beyond energy available in the thermal bath . Indeed , the probabilities of TSS selection at positions 6 , 7 , 8 , and 9 as observed in a comprehensive analysis of TSS-region sequences ( 8% , 55% , 29% , 7%; Vvedenskaya et al . , 2015 ) can be predicted from the Boltzmann-distribution probabilities for the ΔGscrunch values for TSS selection at these positions ( 3% , 70% , 23% , 4%; Figure 6F ) . The finding that values of ΔGscrunch for scrunching and anti-scrunching in TSS selection are ~1 kcal mol−1 bp−1 and ~2 kcal mol−1 bp−1 , respectively , implies that TSS selection at positions >2 bp downstream or >1 bp upstream of the modal position would exceed the energy fluctuations available to 99% of molecules at 20–37°C , and therefore explains the observation that TSS selection >2 bp downstream or >1 bp upstream of the modal position occurs rarely ( Vvedenskaya et al . , 2015 ) . TSS selection by archaeal RNAP , eukaryotic RNAP I , eukaryotic RNAP II from most species , and eukaryotic RNAP III involves the same range of TSS positions as TSS selection by bacterial RNAP ( positions ± 2 bp from the modal TSS; Learned and Tjian , 1982; Samuels et al . , 1984; Thomm and Wich , 1988; Reiter et al . , 1990; Fruscoloni et al . , 1995; Zecherle et al . , 1996 ) . We propose that TSS selection by all of these enzymes is mediated by scrunching and anti-scrunching driven by energy available in the thermal bath . In contrast , TSS selection by S . cerevisiae RNAP II involves a range of TSS positions of 10 s to 100 s of bp ( long-range TSS scanning; Giardina and Lis , 1993; Kuehner and Brow , 2006 ) . We propose that TSS scanning by S . cerevisiae RNAP II also is mediated by scrunching and anti-scrunching , but , in this case , involves not only energy from the thermal bath , but also energy from the ATPase activity of RNAP II transcription factor TFIIH ( Sainsbury et al . , 2015 ) . This proposal could account for the ATP-dependent , TFIIH-dependent cycles of DNA compaction and de-compaction of 10 s to 100 s of bp observed in single-molecule optical-tweezer analyses of TSS scanning by S . cerevisiae RNAP II ( Fazal et al . , 2015 ) . Wild-type E . coli RNAP core enzyme used in transcription experiments was prepared from E . coli strain NiCo21 ( DE3 ) ( New England BioLabs ) transformed with plasmid pIA900 ( Svetlov and Artsimovitch , 2015 ) as described ( Winkelman et al . , 2015 ) . Wild-type RNAP for single-molecule DNA-nanomanipulation experiments was prepared from E . coli strain BL21 ( DE3 ) ( New England Biolabs ) transformed with plasmid pVS10 ( Artsimovitch et al . , 2003 ) as described ( Artsimovitch et al . , 2003 ) . Bpa-containing RNAP core-enzyme derivatives for in vitro protein-DNA photocrosslinking ( βʹR1148Bpa for analysis of RNAP leading-edge positions; βʹT48Bpa for analysis of RNAP trailing-edge positions ) were prepared from E . coli strain NiCo21 ( DE3 ) ( New England BioLabs ) co-transformed with plasmid pEVOL-pBpF ( Chin et al . , 2002; Addgene ) and plasmid pIA900-βʹR1148Bpa ( Winkelman et al . , 2015 ) or pIA900-βʹT48Bpa ( Winkelman et al . , 2015 ) , as in Winkelman et al . ( 2015 ) . Bpa-containing , mutationally inactivated , RNAP core-enzyme derivatives for in vitro and in vivo protein-DNA photocrosslinking ( βʹR1148Bpa;βʹD460A for analysis of RNAP leading-edge positions; βʹT48Bpa;βʹD460A for analysis of RNAP trailing-edge positions ) were prepared from E . coli strain NiCo21 ( DE3 ) ( New England BioLabs ) co-transformed with plasmid pEVOL-pBpF ( Chin et al . , 2002; Addgene ) and plasmid pIA900-βʹR1148Bpa;βʹD460A or pIA900-βʹT48Bpa;βʹD460A [constructed from pIA900-βʹR1148-Bpa ( Winkelman et al . , 2015 ) and pIA900-βʹT48-Bpa ( Winkelman et al . , 2015 ) by use of site-directed mutagenesis with primer ‘JW30’ , as in Winkelman et al . ( 2015 ) ] . σ70 was prepared from E . coli strain BL21 ( DE3 ) ( New England Biolabs ) transformed with plasmid pσ70-His ( Marr and Roberts , 1997 ) as described ( Marr and Roberts , 1997 ) . To form RNAP holoenzyme , 1 μM RNAP core enzyme and 5 μM σ70 in 10 mM Tris-Cl , pH 8 . 0 , 100 mM KCl , 10 mM MgCl2 , 0 . 1 mM EDTA , 1 mM DTT , and 50% glycerol were incubated 30 min at 25°C . Oligodeoxyribonucleotides ( desalted ) were purchased from IDT ( sequences in Supplementary file 1 ) . Oligoribonucleotides ( HPLC-purified ) were purchased from Trilink Biotechnologies . Experiments in Figure 1B and Figure 1—figure supplement 2B were performed using reaction mixtures ( 60 μl ) containing 20 nM RNAP holoenzyme derivative , 4 nM plasmid pCDF-CP-lacCONS-GGG or pCDF-CP-lacCONS-CCT , carrying derivatives of the lacCONS promoter ( Mukhopadhyay et al . , 2001 ) having a GGG or CCT discriminator [prepared by inserting a synthetic 248 bp DNA fragment ( 5'- GAAGCCCTGCATTAGGGGTACCCTAGAGCCTGACCGGCATTATAGCCCCAGCGGCGGATCCCTGCGGGTCGACAAGCTTGAATAGCCATCCCAATCGAACAGGCCTGCTGGTAATCGCAGGCCTTTTTATTTGGATGGAGCTCTGAGAGTCTTCGGTGTATGGGTTTTGCGGTGGAAACACAGAAAAAAGCCCGCACCTGACAGTGCGGGCTTTTTTTTTCGACCAAAGGGACGACCGGGTCGTTGGT- 3' ) between positions 3601 and 460 of pCDF-1b ( EMD-Millipore ) , yielding plasmid pCDF-CP , followed by ligating a 200 bp BglI-digested DNA fragment carrying the lacCONS promoter with GGG or CCT discriminator ( 5'-GTTCAGAGTTCTACAGTCCGACGATCGCGGATGCTTGACAGAGTGAGCGCAACGCAATAACAGTCATCTAGATAGAACTTTAGGCACCCCAGGCTTGACACTTTATGCTTCGGCTCGTATAATGGGGATGCATGTGAGCGGATAACAATGCGGTTAGGCTTAGAGCGCTTAGTCGATGCTGGAATTCTCGGGTGCCAAGG−3' or 5'-GTTCAGAGTTCTACAGTCCGACGATCGCGGATGCTTGACAGAGTGAGCGCAACGCAATAACAGTCATCTAGATAGAACTTTAGGCACCCCAGGCTTGACACTTTATGCTTCGGCTCGTATAATCCTGATGCATGTGAGCGGATAACAATGCGGTTAGGCTTAGAGCGCTTAGTCGATGCTGGAATTCTCGGGTGCCAAGG−3'; −35 and −10 elements underlined; discriminator in bold ) with BglI-digested plasmid pCDF-CP] , 0 or 1 mM ATP , 0 or 1 mM CTP , 0 or 1 mM GTP , and 0 or 1 mM UTP in 60 µl 10 mM Tris-Cl , pH 8 . 0 , 70 mM NaCl , 10 mM MgCl2 , and 0 . 1 mg/ml bovine serum albumin . After 20 min at 37°C , reactions were terminated by addition of 100 µl 10 mM EDTA pH 8 . 0 and 1 mg/ml glycogen . Samples were extracted with acid phenol:chloroform ( Sambrook and Russell , 2001 ) ( Ambion ) , and RNA products were recovered by ethanol precipitation ( Sambrook and Russell , 2001 ) and re-suspended in 6 . 5 µl water . The RNA products were analyzed by primer extension to define TSS positions . Primer-extension was performed by combining 6 . 5 µl RNA products in water , 1 µl 1 µM 32P-5'-end-labeled primer ‘s128a’ [Supplementary file 1; 200 Bq/fmol; prepared using [γ32P]-ATP ( PerkinElmer ) and T4 polynucleotide kinase ( New England Biolabs ) ; procedures as in Sambrook and Russell , 2001] , and 1 µl 10x avian myelobastosis virus ( AMV ) reverse transcriptase buffer ( New England BioLabs ) heating 10 min at 90°C , cooling to 40°C at 0 . 1 °C/s , and incubating 15 min at 40°C; adding 0 . 5 μl 10 mM dNTP mix ( 2 . 5 mM dATP , 2 . 5 mM dGTP , 2 . 5 mM , dCTP , and 2 . 5 mM dTTP; New England Biolabs ) and 1 μl 10 U/μl AMV reverse transcriptase ( New England BioLabs ) ; and incubating 1 hr at 50°C . Primer-extension reactions were terminated by heating 20 min at 85°C; 10 μl 1x TBE ( Sambrook and Russell , 2001 ) , 8 M urea , 0 . 025% xylene cyanol , and 0 . 025% bromophenol blue was added; and samples were analyzed by electrophoresis on 8 M urea , 1X TBE polyacrylamide gels UreaGel System; National Diagnostics ) ( procedures as in Sambrook and Russell , 2001 ) , followed by storage-phosphor imaging ( Typhoon 9400 variable-mode imager; GE Life Science ) . TSS positions were determined by comparison to products of a DNA-nucleotide sequencing reaction obtained using a PCR-generated DNA fragment containing positions −129 to +71 of the lacCONS-GGG promoter and primer ‘s128a’ ( Thermo Sequenase Cycle Sequencing Kit; Affymetrix; methods as per manufacturer ) . Experiments in Figure 3B , were performed analogously , but using a 1 . 3 kb DNA fragment carrying positions −687 to +644 of the lacCONS promoter ( Mukhopadhyay et al . , 2001 ) prepared by PCR amplification of plasmid pUC18-T20C2-lacCONS [prepared by replacing the SbfI-XbaI segment of plasmid pUC18 ( Thermo Scientific ) with a 2 . 0 kb SbfI-XbaI DNA fragment obtained by PCR amplification of Thermus aquaticus rpoC gene with primers Taq_rpoC_F and Taq_rpoC_R ( Supplementary file 1 ) and digestion with XbaI and SbfI-HF ( New England BioLabs ) , yielding plasmid pUC18-T20C2 , followed by inserting a synthetic 117 bp DNA fragment carrying the lacCONS promoter ( 5'-CGGATGCTTGACAGAGTGAGCGCAACGCAATAACAGTCATCTAGATAGAACTTTAGGCACCCCAGGCTTGACACTTTATGCTTCGGCTCGTATAATGTGTGGAATTGTGAGCGGATA-3'; −35 and −10 elements underlined; discriminator in bold ) into the KpnI site of plasmid pUC18-T20C2] with primers ‘LY10’ and ‘LY11’ ( Supplementary file 1 ) , and performing experiments in the presence of 0 or 1 mM UGG , GGA , GAA , or AAU . E . coli strain NiCo21 ( DE3 ) ( New England BioLabs ) transformed with plasmid pCDF-CP-lacCONS-GGG or pCDF-CP-lacCONS-CCT was plated on LB agar ( Sambrook and Russell , 2001 ) containing 50 μg/ml spectinomycin and 50 μg/ml streptomycin , single colonies were inoculated into 25 ml LB broth ( Sambrook and Russell , 2001 ) containing 50 µg/ml spectinomycin and 50 µg/ml streptomycin in 125 ml Bellco flasks , and cultures were shaken ( 220 rpm ) at 37°C . When cell densities reached OD600 = 0 . 6 , 2 ml aliquots were centrifuged 2 min at 4°C at 23 , 000xg , and resulting cell pellets were frozen at −80°C . Cell pellets were thawed in 1 ml TRI Reagent ( Molecular Research Center ) at 25°C for 5 min , completely re-suspended by pipetting up and down , incubated 10 min at 70°C , and centrifuged 2 min at 25°C at 23 , 000 x g . Supernatants were transferred to fresh 1 . 7 ml microfuge tubes , 200 µl chloroform ( Ambion ) was added , vortexed , and samples were centrifuged 1 min at 25°C at 23 , 000 x g . Aqueous phases were transferred to a fresh tube and nucleic acids were extracted with acid phenol:chloroform ( Sambrook and Russell , 2001 ) . Nucleic acids were recovered by ethanol precipitation ( Sambrook and Russell , 2001 ) , and re-suspended in 20 μl 10 mM Tris-Cl , pH 8 . 0 . Primer extension was performed as described in the preceding section . Experiments in Figure 1B were performed using reaction mixtures ( 10 μl ) containing 50 nM Bpa-containing RNAP holoenzyme derivative βʹR1148Bpa ( for analysis of RNAP leading-edge positions ) or βʹT48Bpa ( for analysis of RNAP trailing-edge positions ) and 4 nM plasmid pCDF-CP-lacCONS-GGG or plasmid pCDF-CP-lacCONS-CCT in 10 mM Tris-Cl , pH 8 . 0 , 70 mM NaCl , 10 mM MgCl2 , and 0 . 1 mg/ml bovine serum albumin . Reaction mixtures were incubated 5 min at 37°C , UV-irradiated 5 min at 25°C in a Rayonet RPR-100 photochemical reactor equipped with 16 × 350 nm tubes ( Southern New England Ultraviolet ) , and resulting protein-DNA crosslinks were mapped using primer extension . Primer-extension reactions ( 12 . 5 µl ) were performed by combining 2 µl aliquot of crosslinking reaction , 1 µl 1 µM 32P-5'-end-labeled primer ‘s128a’ ( for analysis of leading-edge positions ) or primer ‘JW85’ ( for analysis of trailing-edge positions ) [Supplementary file 1; 200 Bq/fmol; prepared using [γ32P]-ATP ( PerkinElmer ) and T4 polynucleotide kinase ( New England Biolabs ) ; procedures as in Sambrook and Russell , 2001 , 1 μl 10X dNTPs ( 2 . 5 mM dATP , 2 . 5 mM dCTP , 2 . 5 mM dGTP , 2 . 5 mM TTP , 0 . 5 μl 5 U/μl Taq DNA polymerase ( New England BioLabs ) , 5 μl 5 M betaine , 0 . 625 μl 100% dimethyl sulfoxide , and 1 . 25 µl 10x Taq DNA polymerase buffer ( New England BioLabs ) ; and cycling 16–40 times through 30 s at 95°C , 30 s at 53°C , and 30 s at 72°C . Primer-extension reactions were terminated , and primer-extension products were analyzed as in the preceding section . Experiments in Figure 1—figure supplement 2B were performed analogously , but using Bpa-containing , mutationally inactivated , RNAP derivatives βʹR1148Bpa; β'D460A ( for analysis of RNAP leading-edge positions ) and βʹT48Bpa; β'D460A ( for analysis of RNAP trailing-edge positions ) Experiments in Figure 3B and Figure 3—figure supplement 1B were performed analogously , but using reaction mixtures also containing 0 or 1 mM of ribotrinucleotide primers UGG , GGA , GAA , or AAU , and using 32P-5'-end-labeled primers ‘JW62’ and ‘JW61’ ( Supplementary file 1 ) in primer-extension reactions . Experiments in Figure 1B were performed using a three-plasmid merodiploid system that enabled production of a Bpa-containing , mutationally inactivated , decahistidine-tagged RNAP holoenzyme derivative in vivo and enabled trapping of RPo consisting of the Bpa-containing , mutationally inactivated , decahistidine-tagged RNAP holoenzyme derivative and a lacCONS promoter with GGG or CCT discriminator in vivo , and UV-irradiation of cells ( Figure 1—figure supplement 1 ) . E . coli strain NiCo21 ( DE3 ) ( New England BioLabs ) transformed sequentially with ( i ) plasmid pCDF-CP-lacCONS-GGG or plasmid pCDF-CP-lacCONS-CCT , ( ii ) plasmid pIA900-βʹT48Bpa; βʹD460A or plasmid pIA900-βʹR1148Bpa; βʹD460A , and ( iii ) plasmid pEVOL-pBpF ( Chin et al . , 2002; Addgene ) was plated to yield a confluent lawn on LB agar ( Sambrook and Russell , 2001 ) containing 100 μg/ml carbenicillin , 50 μg/ml spectinomycin , 50 μg/ml streptomycin , and 25 μg/ml chloramphenicol; cells were scraped from the plate and used to inoculate 250 ml LB broth ( as described above ) containing 1 mM Bpa ( Bachem ) , 100 μg/ml carbenicillin , 50 μg/ml spectinomycin , 50 μg/ml streptomycin , and 25 μg/ml chloramphenicol in a 1000 ml flask ( Bellco ) to yield OD600 = 0 . 3; the culture was shaken ( 220 rpm ) 1 hr at 37°C in the dark , isopropyl-β-D-thiogalactoside was added to 1 mM; and the culture was further shaken ( 220 rpm ) 3 hr at 37°C in the dark . Aliquots ( 7 ml ) were transferred to 13 mm x 100 mm borosilicate glass test tubes ( VWR ) , UV-irradiated 20 min at 25°C in a Rayonet RPR-100 photochemical reactor equipped with 16 × 350 nm tubes ( Southern New England Ultraviolet ) , harvested by centrifuging 15 min at 4°C at 3000xg , and cell pellets were frozen at −20°C . Cell pellets were thawed 30 min at 4°C , re-suspended in 40 ml 50 mM Na2HPO4 pH 8 . 0 , 1 . 4 M NaCl , 20 mM imidazole , 14 mM β-mercaptoethanol , 0 . 1% Tween20 , and 5% ethanol containing 2 mg egg white lysozyme . Cells were lysed by sonication 5 min at 4°C . , cell lysates were centrifuged 40 min at 4°C at 23 , 000xg , and supernatants were added to 1 ml Ni-NTA-agarose ( Qiagen , Germantown , MD ) in 1 ml 50 mM Na2HPO4 , pH 8 . 0 , 1 . 4 M NaCl , 20 mM imidazole , 0 . 1% Tween-20 , 5 mM β-mercaptoethanol , and 5% ethanol , and incubated 30 min at 4°C with gentle rocking . The Ni-NTA-agarose was loaded into a 15 ml polyprep column ( BioRad ) , the resulting column was washed with 10 ml of 50 mM Na2HPO4 , pH 8 . 0 , 300 mM NaCl , 20 mM imidazole , 0 . 1% Tween-20 , 5 mM β-mercaptoethanol , and 5% ethanol and eluted with 3 ml of the same buffer containing 300 mM imidazole . The eluate was concentrated to 0 . 2 ml using an 1000 MWCO Amicon Ultra-4 centrifugal filter ( EMD Millipore ) ; the buffer was exchanged to 0 . 2 ml 20 mM Tris-Cl , pH 8 . 0 , 200 mM KCl , 20 mM MgCl2 , 0 . 2 mM EDTA , and 1 mM DTT using the 1000 MWCO Amicon Ultra-4 centrifugal filter ( EMD Millipore ) ; 0 . 2 ml glycerol was added; and the sample was stored at −20°C . Protein-DNA crosslinks were mapped by denaturation followed by primer extension . Denaturation was performed by combining 25 μl crosslinked RNAP-DNA , 25 μl water , 15 μl 5 M NaCl , and 6 μl 100 µg/ml heparin; heating 5 min at 95°C; cooling on ice . Denatured crosslinked RNAP-DNA was purified by adding 20 µl MagneHis Ni-particles ( Promega ) equilibrated and suspended in 10 mM Tris-Cl , pH 8 . 0 , 1 . 2 M NaCl , 10 mM MgCl2 , 10 μg/ml heparin , and 0 . 1 mg/ml bovine serum albumin; washing once with 50 µl 10 mM Tris-Cl , pH 8 . 0 , 1 . 2 M NaCl , 10 mM MgCl2 , 10 μg/ml heparin , and 0 . 1 mg/ml bovine serum albumin; washed twice with 50 µl 1x Taq DNA polymerase buffer ( New England BioLabs ) ; and resuspended in 10 µl 1x Taq DNA polymerase buffer . Primer extension was performed using 2 μl aliquots of purified denatured crosslinked RNAP-DNA , using procedures essentially as described above for experiments in Figure 1B . 2 . 0 kb DNA fragments carrying single centrally located lacCONS-GGG , lacCONS-CCT , or lacCONS promoters were prepared by digesting plasmid pUC18-T20C2-lacCONS-GGG or plasmid pUC18-T20C2-lacCONS-CCT [prepared by inserting a synthetic 80 bp DNA fragment carrying a derivative of the lacCONS promoter ( Mukhopadhyay et al . , 2001 ) having a GGG or CCT discriminator ( 5'-CATCTAGATCACATTTTAGGCACCCCAGGCTTGACACTTTATGCTTCGGCTCGTATAATGGGGATGCATGTGAGCGGATA-3' or 5'-CATCTAGATCACATTTTAGGCACCCCAGGCTTGACACTTTATGCTTCGGCTCGTATAATCCTGATGCATGTGAGCGGATA −3'; −35 and −10 elements underlined; discriminator in bold ) into the KpnI site of plasmid pUC18-T20C2] or plasmid pUC18-T20C2-lacCONS with XbaI and SbfI-HF ( New England BioLabs ) , followed by agarose gel electrophoresis . 1 . 3 kb DNA fragments carrying a single centrally located lacCONS promoter were prepared by PCR amplification of plasmid pUC18-T20C2-lacCONS , using primers ‘LY10’ and ‘LY11’ ( Supplementary file 1 ) , followed by treatment with Dam methyltransferase ( New England BioLabs ) , digestion with XbaI and SbfI-HF ( New England BioLabs ) , and agarose gel electrophoresis . DNA constructs for magnetic-tweezers single-molecule DNA-nanomanipulation were prepared from the above 2 . 0 kb and 1 . 3 kb DNA fragments by ligating , at the XbaI end , a 1 . 0 kb DNA fragment bearing multiple biotin residues on both strands [prepared by PCR amplification of plasmid pARTaqRPOC-lacCONS using primers ‘XbaRPOC4050’ and ‘RPOC3140’ Supplementary file 1 ) and conditions as described ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 Experiments were performed essentially as described ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 Experiments in Figure 2 ( experiments addressing TSS selection for promoters with GGG or CCT discriminator sequence ) , were performed using standard reactions containing mechanically extended , torsionally constrained , 2 . 0 kb DNA molecule carrying GGG or CCT promoter ( extension force = 0 . 3 pN; superhelical density = 0 . 021 for experiments with positively supercoiled DNA; superhelical density = −0 . 021 for experiments with negatively supercoiled DNA ) and RNAP holoenzyme ( 10 nM for experiments with positively supercoiled DNA; 0 . 5 nM for experiments with negatively supercoiled DNA ) in 25 mM Na-HEPES , pH 7 . 9 , 75 mM NaCl , 10 mM MgCl2 , 1 mM dithiothreitol , 0 . 1% Tween-20 , 0 . 1 mg/ml bovine serum albumin ) at 30°C . Data from each of three single DNA molecules were pooled [differences in plectoneme size ( at superhelical density ± 0 . 021 ) and Δl¯obs ≤ 5%] . Experiments in Figure 4—figure supplement 1 ( experiments demonstrating that reduction in DNA-fragment length from 2 . 0 to 1 . 3 kb enables single-bp resolution ) were performed using standard reactions containing mechanically extended , torsionally constrained , 2 . 0 kb or 1 . 3 kb DNA molecule carrying lacCONS promoter ( extension force = 0 . 3 pN; initial superhelical density = 0 . 021 or 0 . 024 for experiments with 2 . 0 kb or 1 . 3 kb positively supercoiled DNA; superhelical density = −0 . 021 or −0 . 024 for experiments with 2 . 0 kb or 1 . 3 kb negatively supercoiled DNA ) and RNAP holoenzyme ( 10 nM for experiments with positively supercoiled DNA; 0 . 5 nM for experiments with negatively supercoiled DNA ) in the buffer of the preceding paragraph at 30°C . For each DNA-fragment length , data were collected on one single DNA molecule . Experiments in Figure 4 and Figure 4—figure supplement 2 ( experiments addressing primer-programmed TSS selection with primers UGG , GGA , GAA , AAU ) were performed using standard reactions containing mechanically extended , torsionally constrained 1 . 3 kb DNA molecule carrying lacCONS promoter ( extension force = 0 . 3 pN; initial superhelical density = 0 . 024 for experiments with positively supercoiled DNA; superhelical density = −0 . 024 for experiments with negatively supercoiled DNA ) and RNAP holoenzyme ( 10 nM for experiments with positively supercoiled DNA; 0 . 5 nM for experiments with negatively supercoiled DNA ) in the buffer of the preceding paragraph at 30°C . Primers UGG , GGA , GAA , and AAU were present at 0 or 1 mM , 0 or 1 µM , 0 or 2 . 5 µM , and 0 or 1 mM , respectively . For experiments with positively supercoiled DNA , data from each of seven single DNA molecules were normalized based on Δl¯obs , pos in absence of primer and pooled; for experiments with negatively supercoiled DNA , data from each of two single DNA molecules were normalized based on Δl¯obs , neg in absence of primer and pooled . Experiments in Figures 5–6 and Figure 6—figure supplement 1 ( experiments addressing primer-concentration dependences of tunwound in primer-programmed TSS selection ) were performed using standard reactions containing mechanically extended , torsionally constrained , 2 . 0 kb DNA molecule carrying lacCONS promoter ( extension force = 0 . 3 pN; initial superhelical density = 0 . 021 ) and RNAP holoenzyme ( 10 nM ) in the buffer of the preceding paragraph at 30°C . Each titration consisted of recordings in absence of primer followed by recordings in presence of primer at increasing concentrations . ( 0 , 0 . 50 , 0 . 90 , 1 . 3 , 250 , 500 , 750 , and 1000 μM for UGG; 0 , 0 . 25 , 0 . 50 , and 1 . 0 µM for GGA; 0 , 0 . 50 , 1 . 0 , and 1 . 5 µM for GAA; 0 , 0 . 40 , 0 . 80 , 1 . 2 , 200 , 400 , 600 , and 800 μM for AAU ) . For each titration , data were collected on one single DNA molecule . For experiments with negatively supercoiled DNA , for which t¯unwound >> 1 h ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 Raw time traces were analyzed to yield DNA extension ( l ) as described ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 Changes in l attributable to DNA unwinding ( Δlu ) and changes in l attributable to DNA compaction ( Δlc ) were calculated as: Δlu = ( Δlobs , neg + Δlobs , pos ) /2 , and Δlc = ( Δlobs , pos - Δlobs , neg ) /2 , where Δlobs , pos and Δlobs , neg are observed changes in l in experiments with positively supercoiled DNA and negatively supercoiled DNA , as described ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 Lifetimes of unwound states ( tunwound ) were extracted from single-molecule traces as described ( Revyakin et al . , 2004 , 2005 , 2006 , Revyakin et al . , 2003 For experiments in absence of primer ( Qi and Turnbough , 1995; Figure 6—figure supplement 1 ) :R+P⇌KBRPc⇌k−2k2RPo where R , P , RPc , and RPo denote RNAP holoenzyme , promoter , RNAP-promoter closed complex , and RNAP-promoter open complex; andt¯unwound=1k−2 For experiments in presence of primer GGA , which programs TSS selection at modal position and therefore does not require scrunching or anti-scrunching for TSS selection ( Figure 6—figure supplement 1 ) :R+P+NpNpN⇌KBRPc+NpNpN⇌k−2k2RPo+NpNpN⇌KNpNpNRPo:NpNpN where NpNpN denotes primer; andt¯unwound=KNpNpNk−2∗[NpNpN]+1k−2 For experiments in presence of primer UGG , or GAA , or AAU , which program TSS selection at positions different form modal TSS , and therefore require scrunching or anti-scrunching for TSS selection ( Figures 4–6 and Figure 6—figure supplement 1 ) :R+P+NpNpN⇌KBRPc+NpNpN⇌k−2k2RPo+NpNpN⇌KscrunchRPo′+NpNpN⇌KNpNpNRPo′:NpNpN where RPo' denotes a scrunched or anti-scrunched RPo; andt¯unwound=KNpNpN∗Kscrunchk−2∗[NpNpN]+1+Kscrunchk−2 Kscrunch , KNpNpN , ΔGNpNpN , and ΔGscrunch were obtained by fitting slopes and y-intercepts of linear-regression fits of plots of t¯unwound vs . primer concentration ( Figure 6D and Figure 6—figure supplement 1 ) to the equation of Figure 6C , stipulating Kscrunch = 1 and ΔGscrunch = 0 for primer GGA , which programs TSS selection at modal position and therefore does not require scrunching or anti-scrunching for TSS selection . Data in Figure 2 are means ± SEM of at least 70 technical replicates of each of three biological replicates ( three single DNA molecules ) for positively supercoiled DNA and at least 50 technical replicates of each of three biological replicates ( three single DNA molecules ) for negatively supercoiled DNA . Data in Figure 4—figure supplement 1A–B are means ± SEM of at least 100 technical replicates for a single DNA molecule ( positively supercoiled DNA ) or at least 70 technical replicates for a single DNA molecule ( negatively supercoiled DNA ) . Data in Figure 4—figure supplement 1C are means ± SEM of randomly selected subsets of n = 30 . Similar results were obtained for ten different randomly selected subsets of n = 30 . Data in Figure 4B–C and Figure 4—figure supplement 2 are means ± SEM of at least 40 technical replicates for each of seven biological replicates ( seven single DNA molecules ) for positively supercoiled DNA and at least 50 technical replicates for each of two biological replicates ( two single DNA molecules ) for negatively supercoiled DNA . Data in Figures 5–6 and Figure 6—figure supplement 1 are means ± SEM of at least 150 technical replicates for one single DNA molecule for each of the four primers .	Genes store the information needed to build and repair cells . This information is written in a chemical code within the structure of DNA molecules . To make use of the information , cells copy sections of a gene into a DNA-like molecule called RNA . An enzyme called RNA polymerase makes RNA molecules from DNA templates in a process called transcription . RNA polymerase can only make RNA by attaching to DNA and separating the two strands of the DNA double helix . This creates a short region of single-stranded DNA known as a “transcription bubble” . RNA polymerase can start transcription at different distances from the sites where it initially attaches to DNA , depending on the DNA sequence and the cell’s environment . It had not been known how RNA polymerase selects different transcription start sites in different cases . One hypothesis had been that differences in the size of the transcription bubble – the amount of unwound single-stranded DNA – could be responsible for differences in transcription start sites . For example , RNA polymerase could increase the size of the bubble through a process called “DNA scrunching” , in which RNA polymerase pulls in and unwinds extra DNA from further along the gene . Yu , Winkelman et al . looked for indicators of DNA scrunching to see whether it contributes to the selection of transcription start sites . By mapping the positions of the two edges of RNA polymerase relative to DNA , they saw that RNA polymerase pulls in extra DNA when selecting a transcription start site further from its initial attachment site . Next , by measuring the amount of DNA unwinding , they saw that RNA polymerase unwinds extra DNA when it selects a transcription start site further from its initial attachment site . This was the case for both RNA polymerase in a test tube and RNA polymerase in living bacterial cells . The results showed that DNA scrunching accounts for known patterns of selection of transcription start sites . The findings hint at a common theory for the selection of transcription start sites across all life by DNA scrunching . Understanding these basic principles of biology reveals more about how cells work and how cells adapt to changing conditions . The experimental methods developed for mapping the positions of proteins on DNA and for measuring DNA unwinding will help scientists to learn more about other aspects of how DNA is stored , copied , read , and controlled .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2017	The mechanism of variability in transcription start site selection
Most age-related human diseases are accompanied by a decline in cellular organelle integrity , including impaired lysosomal proteostasis and defective mitochondrial oxidative phosphorylation . An open question , however , is the degree to which inherited variation in or near genes encoding each organelle contributes to age-related disease pathogenesis . Here , we evaluate if genetic loci encoding organelle proteomes confer greater-than-expected age-related disease risk . As mitochondrial dysfunction is a ‘hallmark’ of aging , we begin by assessing nuclear and mitochondrial DNA loci near genes encoding the mitochondrial proteome and surprisingly observe a lack of enrichment across 24 age-related traits . Within nine other organelles , we find no enrichment with one exception: the nucleus , where enrichment emanates from nuclear transcription factors . In agreement , we find that genes encoding several organelles tend to be ‘haplosufficient , ’ while we observe strong purifying selection against heterozygous protein-truncating variants impacting the nucleus . Our work identifies common variation near transcription factors as having outsize influence on age-related trait risk , motivating future efforts to determine if and how this inherited variation then contributes to observed age-related organelle deterioration . The global burden of age-related diseases such as type 2 diabetes ( T2D ) , Parkinson’s disease ( PD ) , and cardiovascular disease ( CVD ) has been steadily rising due in part to a progressively aging population . These diseases are often highly heritable: for example , narrow-sense heritabilities were recently estimated as 56% for T2D , 46% for general hypertension , and 41% for atherosclerosis ( Wang et al . , 2017 ) . Genome-wide association studies ( GWAS ) have led to the discovery of thousands of robust associations with common genetic variants ( Claussnitzer et al . , 2020 ) , implicating a complex genetic architecture as underlying much of the heritable risk . These loci hold the potential to reveal underlying mechanisms of disease and spotlight targetable pathways . Aging has been associated with dysfunction in many cellular organelles ( López-Otín et al . , 2013 ) . Dysregulation of autophagic proteostasis , for which the lysosome is central , has been implicated in myriad age-related disorders including neurodegeneration , heart disease , and aging itself ( Mizushima et al . , 2008 ) , and mouse models deficient for autophagy in the central nervous system show neurodegeneration ( Hara et al . , 2006; Komatsu et al . , 2006 ) . Endoplasmic reticular ( ER ) stress has been invoked as central to metabolic syndrome and insulin resistance in T2D ( Ozcan et al . , 2004 ) . Disruption in the nucleus through increased gene regulatory noise from epigenetic alterations ( López-Otín et al . , 2013 ) and elevated nuclear envelope 'leakiness' ( D'Angelo et al . , 2009 ) has been implicated in aging . Dysfunction in the mitochondria has even been invoked as a ‘hallmark’ of aging ( López-Otín et al . , 2013 ) and has been observed in many common age-associated diseases ( Lane et al . , 2015; Petersen et al . , 2004; Mootha et al . , 2003; Schapira et al . , 1990; Bender et al . , 2006; Wanagat et al . , 2001; Ashar et al . , 2017 ) . In particular , deficits in mitochondrial oxidative phosphorylation ( OXPHOS ) have been documented in aging and age-related diseases as evidenced by in vivo ( Estrada et al . , 2012 ) P-NMR measures ( Petersen et al . , 2004; Fleischman et al . , 2010 ) , enzymatic activity ( Mootha et al . , 2003; Schapira et al . , 1990; Fannin et al . , 1999; Trounce et al . , 1989; Kelley et al . , 2002; Patti et al . , 2003; Stump et al . , 2003 ) in biopsy material , accumulation of somatic mitochondrial DNA ( mtDNA ) mutations ( Bender et al . , 2006; Wanagat et al . , 2001; Taylor et al . , 2003 ) , and a decline in mtDNA copy number ( mtCN ) ( Ashar et al . , 2017 ) . Given that a decline in organelle function is observed in age-related disease , a natural question is whether inherited variation in loci encoding organelles is enriched for age-related disease risk . Although it has long been known that recessive mutations leading to defects within many cellular organelles can lead to inherited syndromes ( e . g . mutations in >300 nuclear DNA ( nucDNA ) -encoded mitochondrial genes lead to inborn mitochondrial disease; Frazier et al . , 2019 ) , it is unknown how this extends to common disease . In the present study , we use a human genetics approach to assess common variation in loci relevant to the function of ten cellular organelles . We begin with a deliberate focus on mitochondria given the depth of literature linking it to age-related disease , interrogating both nucDNA and mtDNA loci that contribute to the organelle’s proteome . This genetic approach is supported by the observation that heritability estimates of measures of mitochondrial function are substantial ( 33–65%; Curran et al . , 2007; Xing et al . , 2008 ) . We then extend our analyses to nine additional organelles . To our surprise , we find no evidence of enrichment for genome-wide association signal in or near mitochondrial genes across any of our analyses . Further , of 10 tested organelles , only the nucleus shows enrichment among many age-associated traits , with the signal emanating primarily from the transcription factors ( TFs ) . Further analysis shows that genes encoding the mitochondrial proteome tend to be tolerant to heterozygous predicted loss-of-function ( pLoF ) variation and thus are surprisingly ‘haplosufficient’ – that is , show little fitness cost with heterozygous pLoF . In contrast , nuclear TFs are especially sensitive to gene dosage and are often ‘haploinsufficient , ’ showing substantial purifying selection against heterozygous pLoF . Thus , our work highlights inherited variation influencing gene-regulatory pathways , rather than organelle physiology , in the inherited risk of common age-associated diseases . To systematically define age-related diseases , we turned to recently published epidemiological data from the United Kingdom ( U . K . ) ( Kuan et al . , 2019 ) in order to match U . K . Biobank ( UKB ) ( Sudlow et al . , 2015 ) cohort . We prioritized traits whose prevalence increased as a function of age ( Materials and methods ) and were represented in UKB ( https://github . com/Nealelab/UK_Biobank_GWAS ) and/or had available published GWAS meta-analyses ( Teslovich et al . , 2010; Ehret et al . , 2011; Manning et al . , 2012; Morris et al . , 2012; Schunkert et al . , 2011; Estrada et al . , 2012; Christophersen et al . , 2017; Pattaro et al . , 2016; Nalls et al . , 2019; Lambert et al . , 2013; Figure 1A , Appendix 1 ) . We used SNP-heritability estimates from stratified linkage disequilibrium score regression ( S-LDSC , https://github . com/bulik/ldsc ) ( Finucane et al . , 2015 ) to ensure that our selected traits were sufficiently heritable ( Supplementary file 1 , Materials and methods , Appendix 1 ) , observing heritabilities across UKB and meta-analysis traits as high as 0 . 28 ( bone mineral density ) , all with heritability Z-score > 4 . We then computed pairwise genetic and phenotypic correlations between the age-associated traits to compare their respective genetic architectures and phenotypic relationships ( Figure 1B , Materials and methods ) . In general , genetic correlations were greater in magnitude than respective phenotypic correlations , potentially as GWAS are less sensitive to purely non-genetic factors that may influence phenotypes ( e . g . measurement error ) . As expected we find a highly correlated module of primarily cardiometabolic traits with high density lipoprotein ( HDL ) showing anti-correlation ( Bulik-Sullivan et al . , 2015 ) . Interestingly , several other traits ( gastroesophageal reflux disease ( GERD ) , osteoarthritis ) showed moderate genetic correlation to the cardiometabolic trait cluster while atrial fibrillation , for which T2D and CVD are risk factors ( Wasmer et al . , 2017 ) , showed phenotypic , but not genetic , correlation . Our final set of prioritized , age-associated traits included 24 genetically diverse , heritable phenotypes ( Supplementary file 1 ) . Of these , 11 traits were sufficiently heritable only in UKB , three were sufficiently heritable only among non-UKB meta-analyses , and 10 were well-powered in both UKB and an independent cohort . To test if age-related trait heritability was enriched among mitochondria-relevant loci , we began by simply asking if ~1100 nucDNA genes encoding the mitochondrial proteome from the MitoCarta2 . 0 inventory ( Calvo et al . , 2016 ) were found near lead SNPs for our selected traits represented in the NHGRI-EBI GWAS Catalog ( https://www . ebi . ac . uk/gwas/ ) ( MacArthur et al . , 2017 ) more frequently than expectation ( Materials and methods , Appendix 1 ) . To our surprise , no traits showed a statistically significant enrichment of mitochondrial genes ( Figure 2—figure supplement 1A ) ; in fact , six traits showed a statistically significant depletion . Even more strikingly , MitoCarta genes tended to be nominally enriched in fewer traits than the average randomly selected sample of protein-coding genes ( Figure 2—figure supplement 1B , empirical p = 0 . 014 ) . This lack of enrichment was observed more broadly across virtually all traits represented in the GWAS Catalog ( Figure 2—figure supplement 1C ) . We also examined specific transcriptional regulators of mitochondrial biogenesis ( TFAM , GABPA , GABPB1 , ESRRA , YY1 , NRF1 , PPARGC1A , PPARGC1B ) and found very little evidence supporting a role for these genes in modifying risk for the age-related GWAS Catalog phenotypes ( Appendix 1 ) . To investigate further , we turned to U . K . Biobank ( UKB ) . We compiled and tested loci encoding the mitochondrial proteome ( Figure 2A ) with which we interrogated the association between common mitochondrial variation and common disease . First , we considered all common variants in or near nucDNA MitoCarta genes , as well as two subsets of MitoCarta: mitochondrial Mendelian disease genes ( Frazier et al . , 2019 ) and nucDNA-encoded OXPHOS genes . Second , we obtained and tested mtDNA genotypes at up to 213 loci after quality control ( Materials and methods ) from 360 , 662 individuals for associations with age-related traits . First , we used S-LDSC ( Finucane et al . , 2015; Finucane et al . , 2018 ) and MAGMA ( https://ctg . cncr . nl/software/magma ) ( de Leeuw et al . , 2015 ) , two robust methods that can be used to assess gene-based heritability enrichment accounting for LD and several confounders , to test if there was any evidence of heritability enrichment among MitoCarta genes ( Materials and methods ) . We found no evidence of enrichment near nucDNA MitoCarta genes for any trait tested in UKB using S-LDSC ( Figure 2B , Figure 2—figure supplement 2A ) , consistent with our results from the GWAS Catalog . We replicated this lack of enrichment using MAGMA at two different window sizes ( Figure 2—figure supplement 2C , Figure 2—figure supplement 2E; all q > 0 . 1 ) . Given the lack of enrichment among the MitoCarta genes , we wanted to ( 1 ) verify that our selected methods could detect previously reported enrichments and ( 2 ) confirm that common variation in or near MitoCarta genes could lead to expression-level perturbations . We first successfully replicated previously reported enrichment among tissue-specific genes for key traits using both S-LDSC ( Figure 2—figure supplement 3 , Figure 2—figure supplement 4 ) and MAGMA ( Figure 2—figure supplement 5 , Figure 2—figure supplement 6 , Appendix 1 , Materials and methods ) . We next confirmed that we had sufficient power using both S-LDSC and MAGMA to detect physiologically relevant enrichment effect sizes among MitoCarta genes ( Figure 2—figure supplement 7 , Materials and methods , Appendix 1 ) . We finally examined the landscape of cis-expression QTLs ( eQTLs ) for these genes and found that almost all MitoCarta genes have cis-eQTLs in at least one tissue and often have cis-eQTLs in more tissues than most protein-coding genes ( Figure 2—figure supplement 8 , Materials and methods , Appendix 1 ) . Hence , our selected methods could detect physiologically relevant heritability enrichments among our selected traits at gene-set sizes comparable to that of MitoCarta , and common variants in or near MitoCarta genes exerted cis-control on gene expression . Next , we considered mtDNA loci genotyped in UKB , obtaining calls for up to 213 common variants passing quality control across 360 , 662 individuals ( Materials and methods , Appendix 1 ) . We found no significant associations on the mtDNA for any of the 21 age-related traits available in UKB using linear or logistic regression ( Materials and methods , Figure 2C , Figure 2—figure supplement 9; Source data 2 ) . As a control and to validate our approach , we also performed mtDNA-GWAS for specific traits with previously reported associations . A recent analysis of ~147 , 437 individuals in BioBank Japan revealed four distinct traits with significant mtDNA associations ( Yamamoto et al . , 2020 ) . Of these , creatinine and aspartate aminotransferase ( AST ) had sufficiently large sample sizes in UKB . We observed a large number of associations throughout the mtDNA for both traits ( p < 1 . 15 * 10-5 , Figure 2—figure supplement 9E ) . Thus , our mtDNA association method was able to replicate robust mtDNA associations among well-powered traits . We sought to replicate our negative results in an independent cohort . We turned to published GWAS meta-analyses ( Teslovich et al . , 2010; Ehret et al . , 2011; Manning et al . , 2012; Morris et al . , 2012; Schunkert et al . , 2011; Estrada et al . , 2012; Christophersen et al . , 2017; Pattaro et al . , 2016; Nalls et al . , 2019; Lambert et al . , 2013; Supplementary file 1 ) and successfully replicated the lack of enrichment for MitoCarta genes across all 10 traits with an available independent cohort GWAS using S-LDSC ( Figure 2D , Figure 2—figure supplement 2B ) and MAGMA ( Figure 2—figure supplement 2D , Appendix 1; all q > 0 . 1 ) . Importantly , while we were unable to pursue analyses for PD and Alzheimer’s disease in UKB due to limited case counts , we tested MitoCarta genes among well-powered meta-analyses for these disorders ( Appendix 1 ) and observed no enrichment ( Figure 2D; all q > 0 . 1 ) . In summary , we tested ( 1 ) nucDNA loci near genes that encode the mitochondrial proteome in the GWAS Catalog , UKB , and GWAS meta-analyses , ( 2 ) transcriptional regulators of mitochondrial biogenesis in the GWAS Catalog , and ( 3 ) mtDNA variants in UKB . We found no convincing evidence of heritability enrichment for common age-associated diseases near these mitochondrial loci . We next asked whether heritability for age-related diseases and traits clusters among loci associated with any cellular organelle . We used the COMPARTMENTS database ( https://compartments . jensenlab . org ) to define gene-sets corresponding to the proteomes of nine additional organelles ( Binder et al . , 2014 ) besides mitochondria ( Materials and methods ) . We used S-LDSC to produce heritability estimates for these categories in the UKB age-related disease traits , finding evidence of heritability enrichment in many traits for genes comprising the nuclear proteome ( Figure 3A , Materials and methods ) . No other tested organelles showed evidence of heritability enrichment . Variation in or near genes comprising the nuclear proteome explained over 50% of disease heritability on average despite representing only ~35% of tested SNPs ( Figure 3—figure supplement 1 , Appendix 1 ) . We successfully replicated this pattern of heritability enrichment among organelles using MAGMA in UKB at two window sizes ( Figure 3—figure supplement 2A , Figure 3—figure supplement 2B ) , again finding enrichment only among genes related to the nucleus . With over 6000 genes comprising the nuclear proteome , we considered largely disjoint subsets of the organelle’s proteome to trace the source of the enrichment signal ( The Gene Ontology Consortium et al . , 2019; Ashburner et al . , 2000; Lambert et al . , 2018; Figure 3B , Materials and methods , Appendix 1 ) . We found significant heritability enrichment within the set of 1804 genes whose protein products are annotated to localize to the chromosome itself ( q < 0 . 1 for nine traits , Figure 3C , Figure 3—figure supplement 3A ) . Further partitioning revealed that much of this signal is attributable to the subset classified as TFs ( Lambert et al . , 2018 ) ( 1523 genes , q < 0 . 1 for 10 traits , Figure 3D , Figure 3—figure supplement 3B ) . We replicated these results using MAGMA in UKB at two window sizes ( Figure 3—figure supplement 2 ) , and also replicated enrichments among TFs in several ( but not all ) corresponding meta-analyses ( Figure 3—figure supplement 4 ) despite reduced power ( Figure 2—figure supplement 7H ) . We generated functional subdivisions of the TFs ( Materials and methods , Appendix 1 ) , finding that the non-zinc finger TFs showed enrichment for a highly similar set of traits to those enriched for the whole set of TFs ( Figure 3—figure supplement 5D , Figure 3—figure supplement 6B , Figure 3—figure supplement 7B , Figure 3—figure supplement 8B ) . Interestingly , the KRAB domain-containing zinc fingers ( KRAB ZFs ) ( Kapopoulou et al . , 2016 ) , which are recently evolved ( Figure 3—figure supplement 5H ) , were largely devoid of enrichment even compared to non-KRAB ZFs ( Figure 3—figure supplement 5E , Figure 3—figure supplement 6C , Figure 3—figure supplement 7C , Figure 3—figure supplement 8C ) . Thus , we find that variation within or near non-KRAB domain-containing TF genes has an outsize influence on age-associated disease heritability . We next turned to recently published GWAS assessing parental lifespan ( Timmers et al . , 2019 ) and ‘healthspan’ via first morbidity hazard ( Zenin et al . , 2019 ) . Both traits showed highly significant heritability via S-LDSC ( h2 ( s . e . ) = 0 . 0265 ( 0 . 0019 ) and 0 . 0348 ( 0 . 003 ) respectively , Materials and methods ) . Enrichment analysis of organelles among these traits revealed a significant enrichment for the nucleus for parental lifespan ( p = 0 . 0003 ) using MAGMA ( Figure 4 ) . While we observed only a nominally ‘suggestive’ enrichment for the nucleus for healthspan ( p = 0 . 058 ) , S-LDSC showed significant nuclear heritability enrichment ( p = 0 . 0016 , Figure 4—figure supplement 1 ) . Analysis of spatial subsets of the nuclear proteome showed significant enrichment for TFs and proteins localizing to the chromosome in both aging phenotypes using MAGMA ( Figure 4 ) and for healthspan using S-LDSC ( Figure 4—figure supplement 1 ) . In light of observing heritability enrichment only among nuclear transcription factors , we wanted to determine if the fitness cost of pLoF variation in genes across cellular organelles mirrored our results . Mitochondria-localizing genes and TFs play a central role in numerous Mendelian diseases ( Frazier et al . , 2019; Jimenez-Sanchez et al . , 2001; Worman and Courvalin , 2002; Cleaver , 1994 ) , so we initially hypothesized that genes belonging to either category would be under significant purifying selection ( i . e . , constraint ) . We obtained constraint metrics from gnomAD ( https://gnomad . broadinstitute . org ) ( Karczewski et al . , 2020 ) as the LoF observed/expected fraction ( LOEUF ) . In agreement with our GWAS enrichment results , we observed that the mitochondrion on average is one of the least constrained organelles we tested , in stark contrast to the nucleus ( Figure 5A ) . In fact , the nucleus was second only to the set of 'haploinsufficient' genes ( defined based on curated human clinical genetic data; Karczewski et al . , 2020 , Materials and methods ) in the proportion of its genes in the most constrained decile , while the mitochondrion lay on the opposite end of the spectrum ( Figure 5B ) . Interestingly , even the Mendelian mitochondrial disease genes had a high tolerance to pLoF variation on average in comparison to TFs ( Figure 5C ) . Even across different categories of TFs , we observed that highly constrained TF subsets tend to show GWAS enrichment ( Figure 5-Figure supplement 1 , Figure 3-Figure supplement 5E ) relative to unconstrained subsets for our tested traits . Indeed , explicit inclusion of LOEUF as a covariate in the enrichment analysis model ( Materials and methods ) reduced the significance of ( but did not eliminate ) the enrichment seen for the TFs ( Figure 5-Figure supplement 2B , Figure 5-Figure supplement 2E , Figure 5-Figure supplement 2F ) . Thus , while disruption in both mitochondrial genes and TFs can produce rare disease , the fitness cost of heterozygous variation in mitochondrial genes appears to be far lower than that among TFs . This dichotomy reflects the contrasting enrichment results between mitochondrial genes and TFs and supports the importance of gene regulation as it relates to evolutionary conservation . Heritability point estimates and standard errors for age-related traits are listed in Supplementary file 1 . Genetic and phenotypic correlation point estimates and standard errors/p-values plotted in Figure 1B are available in Figure 1—source data 1 . Summary statistics from mtDNA-GWAS ( plotted in Figure 2 and Figure 2—figure supplement 9 ) are available in Source data 2 . All gene-based enrichment analysis p-values and point estimates are available in Source data 1 and Source data 3 . Period prevalence data for diseases in the UK can be obtained from Kuan et al . , 2019 . Gene-sets can be found using COMPARTMENTS ( https://compartments . jensenlab . org ) , MitoCarta 2 . 0 ( https://www . broadinstitute . org/files/shared/metabolism/mitocarta/human . mitocarta2 . 0 . html ) , Lambert et al . , 2018 ( DOI: 10 . 1016/j . cell . 2018 . 01 . 029 ) , Frazier et al . , 2019 ( DOI: 10 . 1074/jbc . R117 . 809194 ) , Finucane et al . , 2018 ( https://alkesgroup . broadinstitute . org/LDSCORE/ ) , Kapopoulou et al . , 2016 ( DOI: 10 . 1111/evo . 12819 ) , and the MacArthur laboratory ( https://github . com/macarthur-lab/gene_lists , copy archived at swh:1:rev:fcc849637bd71e683bffc618e1a48081a8df08f8 ) , Minikel , 2021 . Gene age estimates were obtained from Litman and Stein , 2019 ( DOI: 10 . 1053/j . seminoncol . 2018 . 11 . 002 ) . GWAS catalog annotations can be obtained from: https://www . ebi . ac . uk/gwas . Heritability estimates across UKB can be obtained at: https://nealelab . github . io/UKBB_ldsc/ . UKB summary statistics can be obtained from Neale lab GWAS round 2: https://github . com/Nealelab/UK_Biobank_GWAS , ( copy archived at swh:1:rev:dc7b7b590413ec96a45a64f7213f50a3a0606198 ) , Howrigan , 2021 . Annotations for the Baseline v1 . 1 and BaselineLD v2 . 2 models as well as other relevant reference data , including the 1000G EUR reference panel , can be obtained from https://alkesgroup . broadinstitute . org/LDSCORE/ . eQTL and expression data in human tissues can be obtained from GTEx: https://www . gtexportal . org . Constraint estimates can be found via gnomAD: https://gnomad . broadinstitute . org . See citations for publicly available GWAS meta-analysis summary statistics ( Teslovich et al . , 2010; Ehret et al . , 2011; Timmers et al . , 2019; Zenin et al . , 2019; Manning et al . , 2012; Morris et al . , 2012; Schunkert et al . , 2011; Estrada et al . , 2012; Christophersen et al . , 2017; Pattaro et al . , 2016; Nalls et al . , 2019; Lambert et al . , 2013 ) . Sex-standardized period prevalence of over 300 diseases was obtained from an extensive survey of the National Health Service in the UK as reported previously ( Kuan et al . , 2019 ) . To select high prevalence late-onset diseases , we ranked diseases with a median onset over 50 years of age by the sum of the period prevalence of all age categories above 50 . We selected the top 30 diseases using this metric and manually mapped these traits to similar or equivalent phenotypes with publicly available summary statistics from UKB and/or well-powered meta-analyses ( e . g . Parkinson’s Disease and Alzheimer’s Disease for dementia ) resulting in 24 traits with data available in UKB ( RRID:SCR_012815 ) , meta-analyses , or both ( Supplementary file 1 ) . We manually mapped selected age-related diseases and traits to corresponding phenotypes in UKB . In parallel , we searched the literature to identify well-powered EUR-predominant GWAS ( referred to as meta-analyses ) that ( 1 ) used primarily non-targeted arrays , ( 2 ) had publicly available full summary statistics , and ( 3 ) did not enroll individuals from UKB to serve as independent replication ( Appendix 1 ) . We produced heritability estimates using stratified linkage-disequilibrium score regression ( S-LDSC , https://github . com/bulik/ldsc ) ( Finucane et al . , 2015 ) atop the BaselineLD v2 . 2 model using reference LD scores computed from 1000G EUR ( https://alkesgroup . broadinstitute . org/LDSCORE/ ) . We computed the heritability Z-score , a statistic that captures sample size , polygenicity , and heritability ( Finucane et al . , 2015 ) , and included only traits with heritability Z-score > 4 ( Appendix 1 ) for further analysis . Pairwise genetic correlations , rg , were computed using linkage-disequilibrium score correlation ( Bulik-Sullivan et al . , 2015 ) on all selected age-related traits with heritability Z-score > 4 . We used UKB summary statistics ( https://github . com/Nealelab/UK_Biobank_GWAS ) for all sufficiently powered traits; summary statistics from meta-analyses were used for eGFR ( Pattaro et al . , 2016 ) , Alzheimer’s Disease ( Lambert et al . , 2013 ) , and Parkinson’s Disease ( Nalls et al . , 2019 ) as these traits showed heritability Z-score > 4 within meta-analyses but not in UKB ( Supplementary file 1 ) . p-Values for genetic correlation represented deviation from the null hypothesis rg=0 . Traits were ordered by their contribution to the first eigenvector of the absolute value of the correlation matrix , with point estimates and standard errors available in Source data 1 . Bonferroni correction was applied producing a p-value cutoff of 0 . 05/242+212=1 . 0310-4 , accounting for both genotypic and phenotypic correlation hypothesis tests . Pairwise phenotypic correlations , rp , were computed for all 21 traits with well-powered individual level data available in UKB ( Supplementary file 1 ) . Pearson correlation was computed between continuous traits via cor . test in R ( RRID:SCR_001905 ) with a two-sided alternative . Tetrachoric correlation was used to compute correlations between binary traits and biserial correlation was used for correlations between binary and continuous traits , using the polychor and polyserial functions of the polycor package in R using the two-step approximation , respectively . These approaches model a latent normally distributed variable underlying binary traits . p-Values were computed using a normal approximation using standard error estimates from polycor . Point estimates and standard errors are available in Figure 1—source data 1 . We mapped variants in the GWAS Catalog ( RRID:SCR_012745 ) ( obtained on September 5th , 2019 , https://www . ebi . ac . uk/gwas/ ) meeting genome-wide significance ( p < 5e-8 ) to genes using provided annotations , producing a set of trait-associated genes for each trait . We manually selected phenotypes represented in the GWAS Catalog matching our set of age-associated traits with > 30 trait-associated genes . For each trait , we computed the proportion of trait-associated genes that were mitochondria-localizing ( defined via MitoCarta2 . 0; Calvo et al . , 2016 , RRID:SCR_018165 ) and tested for enrichment or depletion relative to overall genome background using two-sided Fisher’s exact tests . We corrected for multiple hypothesis tests with the Benjamini-Hochberg ( BH ) procedure at FDR q-value < 0 . 1 . We also computed the test statistic Ngenrich , defined as the number of age-associated traits showing a nominal ( not necessarily statistically significant ) enrichment for a given gene-set g , for the MitoCarta genes . We then generated an empirical null distribution for Ngenrich . We drew 1000 random samples of protein-coding genes , where each sample contained the same number of genes as the set of mitochondria-localizing genes and computed Ngenrich for each of these gene-sets ( Figure 2—figure supplement 1B ) . The one-sided p-value , defined as Pr⁡Ngenrich≤x under the null , was subsequently obtained . We expanded our enrichment/depletion analysis to all 332 traits in the GWAS Catalog with over 30 trait-associated genes; for enrichment or depletion testing , we used two-sided Fisher’s exact tests and corrected for multiple hypothesis testing with the BH procedure at FDR q-value < 0 . 1 . UKB summary statistics previously formatted for use with LDSC and filtered to HapMap3 ( HM3 ) ( RRID:SCR_004563 ) SNPs ( https://github . com/Nealelab/UKBB_ldsc ) were used for analysis with S-LDSC . For analysis with MAGMA v1 . 07b ( de Leeuw et al . , 2015 ) , we included variants from the full Neale Lab UKB Round 2 GWAS summary statistics ( https://github . com/Nealelab/UK_Biobank_GWAS ) with INFO > 0 . 8 and MAF > 0 . 01 , and excluded any variants flagged as low confidence ( a heuristic defined by MAF < 0 . 001 or expected case MAC < 25 ) . Summary statistics obtained from publicly available GWAS meta-analyses ( Teslovich et al . , 2010; Ehret et al . , 2011; Manning et al . , 2012; Morris et al . , 2012; Schunkert et al . , 2011; Estrada et al . , 2012; Christophersen et al . , 2017; Pattaro et al . , 2016; Nalls et al . , 2019; Lambert et al . , 2013 ) were reported in varied formats . We manually verified the genome build upon which each meta-analysis reported results and ensured that all sets of summary statistics contained columns listing p-value , variant rsID , genome-build specific coordinates , and if available , variant-specific sample size ( Supplementary file 1 ) . If variant coordinates or rsID were not provided , the relevant columns were obtained from dbSNP ( RRID:SCR_002338 ) database version 130 ( for hg18 ) or 146 ( for hg19 ) . We used the summary statistic munging script provided with S-LDSC ( https://github . com/bulik/ldsc ) to generate summary statistics compatible with S-LDSC , restricting to HM3 SNPs as these tend to be best behaved for analysis with LDSC . For use of meta-analyses with MAGMA ( de Leeuw et al . , 2015 ) , we restricted analysis to variants with INFO > 0 . 8 and MAF > 0 . 01 if such information was provided . To account for the multiple hypothesis tests performed throughout this study for age-related traits , we obtained p-value thresholds via the BH procedure at FDR < 0 . 1 for all gene-sets assessed for a given method and cohort type ( where the two cohort types were UKB and meta-analysis ) . The BH procedure at FDR < 0 . 1 was also applied to our analyses of parental lifespan and healthspan . We extensively use S-LDSC and MAGMA to perform gene-set enrichment analyses among GWAS summary statistics . To test enrichment with S-LDSC , SNPs were mapped to each gene with a 100 kb symmetric window as recommended ( Finucane et al . , 2018 ) and LD scores were computed using the 1000G EUR reference panel ( RRID:SCR_006828 ) ( https://alkesgroup . broadinstitute . org/LDSCORE/ ) and subsequently restricted to the HM3 SNPs . We used S-LDSC to test for heritability enrichment controlling for 53 annotations including coding regions , enhancer regions , 5’ and 3’ UTRs , and others as previously described ( Finucane et al . , 2015 ) ( baseline v1 . 1 , referred to as baseline model hereafter ) . We also used MAGMA with both 5 kb up , 1 . 5 kb down and 100 kb symmetric windows to test for enrichment . MAGMA gene-level analysis was performed with the 1000G EUR LD reference panel to account for LD structure , and gene-set analysis was performed including covariates for gene length , variant density , inverse minor allele count ( MAC ) , as well as log-transformed versions of these covariates . Statistical tests for both S-LDSC and MAGMA were one-sided , considering enrichment only . For both methods , we included the relevant superset of genes as a control to ensure that our analysis was competitive ( Appendix 1 ) . We refer to this approach as the ‘usual approach . ’ All enrichment effect size estimates and p-values are available in Source data 1 and Source data 3 . We obtained the set of nuclear-encoded mitochondria-localizing genes using MitoCarta2 . 0 ( Calvo et al . , 2016 ) and used the literature to obtain the subset of MitoCarta genes involved in inherited mitochondrial disease ( Frazier et al . , 2019 ) as well as those producing components of oxidative phosphorylation ( OXPHOS ) complexes . We used both S-LDSC and MAGMA to test for enrichment in the usual way ( Materials and methods ) controlling for the set of protein-coding genes to ensure a competitive analysis ( Appendix 1 ) . We also tested mitochondria-localizing genes for enrichment in meta-analyses using S-LDSC and MAGMA with the same parameters as for UKB traits ( Appendix 1 ) . To obtain the set of genes most expressed in a given tissue versus others , we obtained t-statistics computed from GTEx ( RRID:SCR_013042 ) v6 gene-level transcript-per-million ( TPM ) data corrected for age and sex as published previously ( Finucane et al . , 2018 ) . For each tissue , we selected the top 2485 genes ( 10% ) with the highest t-statistics for tissue-specific expression , producing tissue-expressed gene-sets . We selected nine tissues based on expectation of enrichment for our tested traits in UKB ( e . g . liver for LDL levels , esophageal mucosa for GERD ) . We used both S-LDSC and MAGMA to test for enrichment in the usual way ( Materials and methods ) controlling for the set of tissue-expressed genes to ensure a competitive analysis ( Appendix 1 ) . Tissue-expressed gene-set analyses were performed on meta-analyses with S-LDSC and MAGMA on the same tissues using the same parameters as used in UKB . To test for the effects of gene-set size on power , we selected 10 positive control tissue-trait pairs based on ( 1 ) the presence of tissue enrichment in UKB with S-LDSC and MAGMA and ( 2 ) if the observed enrichment was biologically plausible . The pairs tested were liver-HDL , liver-LDL , liver-TG , liver-cholesterol , pancreas-glucose , pancreas-T2D , atrial appendage-atrial fibrillation , sigmoid colon-diverticular disease , coronary artery-myocardial infarction , and visceral adipose-HDL . We then , in brief , used an empirical sampling-based approach , generating random subsamples of a selected set of tissue-expressed gene-sets at four different gene-set sizes ( 1523 , 1105 , 800 , and 350 genes ) , defining power as the proportion of trials showing a significant enrichment ( Appendix 1 ) . We used the same sub-sampled gene-sets for enrichment analysis using both S-LDSC and MAGMA in the usual way ( Materials and methods ) controlling for the set of tissue-expressed genes to ensure a competitive analysis ( Appendix 1 ) . We used the same gene-sets among the subset of the positive control traits that showed enrichment in the corresponding meta-analysis to verify power for the meta-analyses ( Appendix 1 ) . We obtained the set of eGenes from GTEx ( RRID:SCR_013042 ) v8 across 49 tissues ( https://www . gtexportal . org ) , filtering to only include cis-eQTLs with q-value < 0 . 05 . To determine how the landscape of cis-eQTLs for MitoCarta genes compared to other protein-coding genes , we regressed the number of tissues with a detected cis-eQTL for a given gene x , NxeQTL , onto an indicator for membership in a given organellar proteome ( Ixorganelle ) , controlling for gene length , log gene length , breadth of expression ( τx ) , and the number of tissues with detected expression > 5 TPM ( Nxexpress , Appendix 1 ) . To quantify breadth of expression , we obtained median-per-tissue GTEx v8 TPM expression values and computed τ ( Yanai et al . , 2005 ) after removing lowly expressed genes with maximal cross-tissue TPM < 1 , defined as:τx=∑i=1n ( 1−x^i ) n−1wherex^i=ximax1≤i≤nxiwhere xi is the expression of gene x in tissue i with n tissues . τ ranges from 0 to 1 , with lower τ indicating broadly expressed genes and higher τ indicating more tissue specific expression patterns . Because GTEx sampled multiple tissue subtypes ( e . g . brain sub-regions ) that show correlated expression profiles ( Melé et al . , 2015 ) which bias τx , NxeQTL , and Nxexpress upward , for each broader tissue class ( brain , heart , artery , esophagus , skin , cervix , colon , adipose ) , we selected a single representative tissue when computing these quantities ( Figure 3—figure supplement 5B , Appendix 1 ) . We used LD scores computed from the 1000G EUR reference panel . The model , fit via ordinary least squares for each tested organelle , was:NxeQTL∼Ixorganelle+Nxexpress+τx+log⁡ ( genelength ) +genelength We obtained mtDNA genotype data on 265 variants as obtained on the UK Biobank Axiom array and the UK BiLEVE array from the full UKB release ( RRID:SCR_012815 ) ( Sudlow et al . , 2015 ) . To perform variant QC , we used evoker-lite ( RRID:SCR_009145 ) ( Morris et al . , 2010 ) to generate fluorescence cluster plots per-variant and per-batch and manually inspected the results , removing 19 variants due to cluster plot abnormalities ( Supplementary file 2a , Appendix 1 ) . We additionally removed any variants with heterozygous calls , within-array-type call rate < 0 . 95 , and with less than 20 individuals with an alternate genotype . For case-control traits , we removed any phenotype-variant pair with an expected case count of alternate genotype individuals of less than 20 , resulting in a maximum of 213 variants tested per trait ( Appendix 1 ) . To perform sample QC , we restricted samples to the same samples from which UKB summary statistics were generated ( https://github . com/Nealelab/UK_Biobank_GWAS ) , namely unrelated individuals seven standard deviations away from the first 6 European sample selection PCs with self-reported white-British , Irish , or White ethnicity and no evidence of sex chromosome aneuploidy . We additionally removed any samples with within-array-type mitochondrial variant call rate < 0 . 95 , resulting in 360 , 662 unrelated samples of EUR ancestry . We generated the LD matrix for mitochondrial DNA variants using Hail v0 . 2 . 51 ( https://hail . is ) pairwise for all 213 variants tested across all post-QC samples . We ran mtDNA-GWAS for all 21 UKB age-related phenotypes as well as creatinine and AST using Hail v0 . 2 . 51 via linear regression controlling for the first 20 PCs of the nuclear genotype matrix , sex , age , age2 , sexage , and sex*age2 as performed for the UKB GWAS ( https://github . com/Nealelab/UK_Biobank_GWAS ) . We also used Hail to run Firth logistic regression with the same covariates for case/control traits . As we observed that some mitochondrial DNA variants were specific to array type , we also ran linear regression including array type as a covariate; we did not perform logistic regression with array type as a covariate due to convergence issues from complete separation of variants assessed only on a single array type . We defined mtDNA-wide significance using a Bonferroni correction by p=0 . 054337≈1 . 15e-5 . COMPARTMENTS ( RRID:SCR_015561 ) ( https://compartments . jensenlab . org ) ( Binder et al . , 2014 ) is a resource integrating several lines of evidence for protein localization predictions including annotations , text-mining , sequence predictions , and experimental data from the Human Protein Atlas . We used this resource to obtain the degree of evidence ( a number ranging from 0 to 5 ) linking each gene to localization to one of 12 organelles: nucleus , cytosol , cytoskeleton , peroxisome , lysosome , endoplasmic reticulum , Golgi apparatus , plasma membrane , endosome , extracellular space , mitochondrion , and proteasome . To avoid noisy localization assignments due to weak text mining and prediction evidence , we only considered localization assignments with a score > 2 as described previously ( Binder et al . , 2014 ) . We subsequently assigned compartment ( s ) to each gene by selecting the compartment ( s ) with the maximal score within each gene . We only included compartments containing over 240 genes due to limited power at smaller gene-set sizes and used MitoCarta2 . 0 ( Calvo et al . , 2016 ) to obtain a higher confidence set of genes localizing to the mitochondrion , resulting in gene-sets representing the proteomes of 10 organelles . S-LDSC and MAGMA were used to test for enrichment across the UKB age-related traits for these gene-sets in the usual way , controlling for the set of protein-coding genes . S-LDSC was also used to obtain estimates of the percentage of heritability explained by each organelle gene-set . To produce interpretable sub-divisions of the nucleus , we used Gene Ontology ( GO ) ( RRID:SCR_017505 ) ( The Gene Ontology Consortium et al . , 2019; Ashburner et al . , 2000 ) to identify terms listed as children of the nucleus cellular component ( GO:0005634 ) . We used Ensembl ( RRID:SCR_002344 ) version 99 ( Yates et al . , 2020 ) to obtain a first pass set of genes annotated to each sub-compartment of the nucleus ( or its children ) . After manual review of sub-compartments with > 90 genes , we selected nucleoplasm ( GO:0005654 ) , nuclear chromosome ( GO:0000228 ) , nucleolus ( GO:0005730 ) , nuclear envelope ( GO:0005635 ) , splicosomal complex ( GO:0005681 ) , nuclear DNA-directed RNA polymerase complex ( GO:0055029 ) , and nuclear pore ( GO:0005643 ) . We excluded terms listed as ‘part’ due to poor interpretability and manually excluded similar terms ( e . g . nuclear lumen vs nucleoplasm ) . To generate a high confidence set of genes localizing to each of these selected sub-compartments , we then turned to the COMPARTMENTS resource which assigns localization confidence scores for each protein to GO cellular component terms . We assigned members of the nuclear proteome to these selected nuclear sub-compartments using same the approach outlined for the organelle analysis ( Materials and methods ) . After filtering our selected sub-compartments to those containing > 240 genes , we obtained four categories: nucleoplasm , nuclear chromosome , nucleolus , and nuclear envelope . The nuclear chromosome annotation was largely overlapping with a manually curated high-quality list of TFs ( Lambert et al . , 2018 ) however was not exhaustive; as such , we merged these lists to generate the chromosome and TF category . To improve interpretability , we removed genes from nucleoplasm that were also assigned to another nuclear sub-compartment , constructed a list of other nucleus-localizing proteins not captured in these four sub-compartments , and included only genes annotated as localizing to the nucleus ( Materials and methods ) . S-LDSC and MAGMA were used to test for enrichment across the UKB age-related traits for these gene-sets in the usual way while controlling for the set of protein-coding genes ( Materials and methods ) . We used a published , curated , high-quality list of TFs ( Lambert et al . , 2018 ) to partition the Chromosome and TF category into TFs and other chromosomal proteins . To determine which TFs are broadly expressed versus tissue specific , we computed τ per TF across all selected tissues after removing lowly expressed genes with maximal cross-tissue TPM < 1 ( Materials and methods , Appendix 1 ) . The threshold for tissue-specific genes was set at τ≥0 . 76 based on the location of the central nadir of the resultant bimodal distribution ( Figure 3—figure supplement 5A ) . To identify terciles of TFs by age , we obtained relative gene age assignments for each gene previously generated by obtaining the modal earliest ortholog level across several databases mapped to 19 ordered phylostrata ( Litman and Stein , 2019 ) . DNA-binding domain ( DBD ) annotations for the TFs were obtained from previous manual curation efforts ( Lambert et al . , 2018 ) . S-LDSC and MAGMA were used to test for enrichment across the UKB age-related traits for these gene-sets in the usual way while controlling for the set of protein-coding genes ( Materials and methods ) . We also tested TFs for enrichment in meta-analyses using S-LDSC and MAGMA with the same parameters as for UKB traits ( Appendix 1 ) . We obtained gene-level gnomAD ( RRID:SCR_014964 ) v2 . 1 . 1 constraint tables ( https://gnomad . broadinstitute . org ) , haploinsufficient genes , and olfactory receptors ( Karczewski et al . , 2020 ) ( https://github . com/macarthur-lab/gene_lists ) . Constraint values as loss-of-function observed/expected fraction ( LOEUF ) were mapped to genes within organelle , sub-mitochondrial , sub-nuclear , and TF binding domain gene-sets . To test for enrichment with constraint as a covariate , we used MAGMA with UKB age-related traits . We mapped variants to genes and performed the gene-level analysis as done previously for the mitochondria-localizing gene and organelle analysis . We included LOEUF and log LOEUF as covariates for the gene-set analysis in addition to the default covariates ( gene length , SNP density , inverse MAC , as well as the respective log-transformed versions ) via the –condition-residualize flag .	Getting older increases our risk of experiencing a wide range of diseases , such as diabetes , heart disease and neurodegenerative disease . The genetic variations that we inherit from our parents play a major role in predicting this risk . However , the biological networks involved in this process are extremely complex and remain challenging to decipher . Prior studies have suggested that specialised structures inside our body’s cells , called organelles , may have an important role to play in aging . Organelles represent self-contained biological factories inside each cell , designed to perform specific tasks . Examples include the nucleus , which harbours most of the cell’s genetic material , and mitochondria , which help provide cells with energy . Organelles tend to deteriorate and become dysfunctional with age , and mitochondria in particular are badly affected by the ageing process . A decline in organelle activity has been thought to explain ageing and the development of age-related diseases . However , this has never been systematically tested on a large scale at the inherited genetic level . Gupta et al . assessed whether common inherited genetic variation in genes associated with ten different organelles could affect the risk of age-related disease , using a database of DNA samples from more than 300 , 000 individuals . They considered 24 diseases and traits that become more common with advanced age . Gupta et al . discovered that inherited variants in or near genes associated with the nucleus were consistently linked to age-related disease risks . Most of this signal arose from genes encoding the nuclear transcription factors , proteins that help to control the rate at which genes are expressed . However , variants in genes associated with other organelles , including mitochondria , did not appear to be linked to age-related diseases . This research suggests that inherited variation in transcription factors in the nucleus could act as genetic levers that increase the risk of common , age-related diseases . It also suggests that common genetic variation in other cellular organelles may not be as heavily involved in the development of such diseases . Such insights into the cellular structures and biological pathways involved in ageing and age-related disease also establish new targets for drugs to prevent or treat disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2021	Human genetic analyses of organelles highlight the nucleus in age-related trait heritability
The fibroblast growth factor FGF21 was labeled with molecularly defined gold nanoparticles ( AuNPs ) , applied to human adipocytes , and imaged by cryo-electron tomography ( cryo-ET ) . Most AuNPs were in pairs about 80 Å apart , on the outer cell surface . Pairs of AuNPs were also abundant inside the cells in clathrin-coated vesicles and endosomes . AuNPs were present but no longer paired in multivesicular bodies . FGF21 could thus be tracked along the endocytotic pathway . The methods developed here to visualize signaling coupled to endocytosis can be applied to a wide variety of cargo and may be extended to studies of other intracellular transactions . Imaging of cell structure has been performed using fluorescence light microscopy at modest resolution on living cells in real time , and using electron microscopy at higher resolution on fixed , embedded , sectioned material . The power of fluorescence light microscopy has been extended by super-resolution techniques ( Baddeley and Bewersdorf , 2018 ) , while advances in cryo-electron microscopy ( cryo-EM ) have yielded structures of purified proteins at near atomic resolution ( Peplow , 2017 ) , and have enhanced tomography of intact cells ( Oikonomou and Jensen , 2017 ) . Cryo-ET provides an opportunity to study proteins as they interact with a myriad of other factors ( Beck and Baumeister , 2016; Irobalieva et al . , 2016 ) , often lost during protein purification . Very large multi-protein assemblies , such as ribosomes and chemoreceptor arrays , scatter electrons strongly enough that they can be recognized in electron micrographs of frozen hydrated specimens ( Briegel and Jensen , 2017 ) . Our approach , employing AuNP conjugates , enables the identification and image processing of most molecules and molecular assemblies , which are too small to be detected against the background of scattering from the cellular milieu . To that end , we have developed defined heavy atom clusters , targeted to individual molecules ( Azubel and Kornberg , 2016 ) . We report here on the application of such clusters to the fibroblast growth factor FGF21 in human primary adipocytes . FGFs are essential in cell biology , either by their participation in cell proliferation , cell survival and cell motility ( paracrine FGFs ) , or by their connection to metabolic processes ( endocrine FGFs ) . These diverse activities share a common first step: binding of FGFs to cell membrane receptors . There are four genes for FGF receptors ( FGFRs ) , which produce seven alternatively spliced variants . Paracrine and endocrine FGFs , totaling 15 and three secreted proteins , respectively , compete for binding to these seven FGFRs ( Ornitz and Itoh , 2015 ) . Binding requires co-factors: paracrine FGFs are assisted by heparan sulfate , and endocrine FGFs by either αKlotho or βKlotho ( Kilkenny and Rocheleau , 2016 ) . Binding leads to FGFR dimerization and activation of FGFR tyrosine kinase activity , which triggers RAS-MAPK , PI3K-AKT , and PLCγ1 signaling cascades ( Ornitz and Itoh , 2015 ) . Whereas signaling is commonly thought to occur at the cell surface , it continues in endosomal locations ( Jean et al . , 2010 ) ( Haugsten et al . , 2011 ) . Moreover , signaling cascades are interrupted when endocytosis is inhibited ( Yaqoob et al . , 2014 ) . Endocytosis modulates signaling , as the specific endocytic pathway ( Mayor and Pagano , 2007 ) determines whether the receptor is recycled to the cell surface or destined for degradation ( Haugsten et al . , 2005 ) . Signaling must therefore be studied in the context of membrane internalization and vesicle trafficking . A fundamental question regarding the activation of the signaling cascade is the stochiometry of the ternary complex ( FGF-receptor-cofactor ) . Competing models have been proposed ( Goetz and Mohammadi , 2013; Yie et al . , 2012 ) ( Pomin , 2016 ) ( Kilkenny and Rocheleau , 2016 ) . The crystal structure of FGF2-FGFR1c ( extracellular domains D2-D3 ) and heparan sulfate showed a 2:2:2 ternary complex ( Schlessinger et al . , 2000 ) . αKlotho and βKlotho differ significantly in both size and shape from heparan sulfate , and also compete with some paracrine FGFs for the same regions to bind receptors ( Goetz and Mohammadi , 2013 ) . Thus a different mode of binding that could lead to a different stochiometry for endocrine ternary complexes could not be ruled out . Indeed , subsequent studies of FGF21-FGFR1c-βKlotho have favored a 1:2:1 model ( Ming et al . , 2012 ) . Most recently , the crystal structure of a 1:1:1 complex of membrane proximal portion of extracellular FGFR1c , soluble αKlotho , and FGF23 was described , and dimerization of the αKlotho complex was observed in the presence of heparan ( Chen et al . , 2018 ) . The extracellular domain of βKlotho bound to the C-terminus of FGF21 was also determined by X-ray crystallography , revealing a 1:1 complex , suggested to lead to an overall 2:2:2 complex ( Lee et al . , 2018 ) . We focus here on the FGF21-FGFR1c-βKlotho ternary complex . In recent years , FGF21 has emerged as a potential candidate for treatment of obesity and type II diabetes ( Kharitonenkov and DiMarchi , 2015 ) . Pleiotropy of FGF21 includes effects on glucose and lipid metabolism in adipocyte tissue ( Degirolamo et al . , 2016 ) . FGF21 signals through FGFR1c , FGFR2c and FGFR3c , provided that βKlotho is accessible ( Kilkenny and Rocheleau , 2016 ) . Both FGFR1c and βKlotho are endogenously expressed in adipocyte tissue . The pathway of FGF21-FGFR1c-βKlotho complex internalization remains an open question . Evidence for both clathrin-dependent ( Jean et al . ) and clathrin-independent ( Haugsten et al . , 2011 ) pathways , for different combinations of FGF and FGFR , has been presented . Regarding the FGF21-FGFR1c-βKlotho complex , dynamin-dependent endocytosis has been suggested ( Yaqoob et al . , 2014 ) . However , dynamin has been found associated to both clathrin-dependent and clathrin-independent endocytosis ( Mayor and Pagano , 2007 ) . With the use of gold-labeled FGF21 ( AuNP-FGF21 ) and cryo-ET , we captured different states of activation , internalization , and traffic of the FGF21-FGFR1c-βKlotho ternary complex , from binding and complex formation at the cell surface , to coated pits , to coated vesicles , to endosomes , and finally , to multivesicular bodies , in which the complexes were disrupted . These observations are clearly indicative of clathrin-dependent endocytosis . Finally , subtomogram averaging and helical reconstruction revealed structures of other important components , including putative AAA+ ATPases , actin filaments , and microtubules , giving a three-dimensional picture of the entire pathway . A 144-gold atom nanoparticle ( AuNP ) was conjugated with an FGF21 variant bearing a surface-exposed cysteine residue ( Xu et al . , 2013 ) , as described ( Azubel and Kornberg , 2016 ) . Interaction in ternary complexes was assessed using membrane preparations from three cell sources: parental CHO cells , in which neither FGFR1c nor βKlotho are expressed; transformed CHO cells overexpressing βKlotho and FGFR1c or only βKlotho; and human primary adipocytes , in which βKlotho and FGFR1c are endogenously expressed . Vesicles were treated at 4°C with either AuNP-FGF21 or a gold-labeled single chain antibody fragment ( AuNP-scFv ) that binds βKlotho , and washed to remove unbound gold conjugate . Grids for cryo-EM were prepared by plunge-freezing . Micrographs of vesicles from parental CHO cells membrane preparations treated with AuNP-FGF21 showed no associated AuNPs , whereas micrographs of vesicles from primary adipocytes membrane preparations treated with AuNP-FGF21 showed pairs of AuNPs ( Figure 1—figure supplement 1 ) . AuNPs were distinguishable from other particles because of an effect of the contrast transfer function , producing a bright halo around the strongly scattering gold core ( Figure 1—figure supplement 2 ) . Pairing of particles cannot be determined from 2D images alone , as two particles in close proximity in the x-y plane may be far apart in z . Tilt series were therefore collected for membrane preparations from CHO cells overexpressing βKlotho and FGFR1c treated with AuNP-FGF21 , followed by tomographic reconstruction , showing that 85% of AuNPs were in true pairs ( Figure 1—figure supplement 3a–g ) , indicative of two copies of FGF21 in the receptor complex . Treatment of these CHO membrane vesicles with AuNP-scFv against βKlotho also resulted in a high percentage of pairs of particles ( Figure 1—figure supplement 3h ) , indicative of an overall 2:2:2 stoichometry for the receptor complex . When membrane preparations from CHO cells expressing only βKlotho were treated with AuNP-scFv against βKlotho , pairs of particles were not observed ( Figure 1—figure supplement 3j ) , showing that βKlotho did not dimerize on its own . When the same vesicles were treated with AuNP-FGF21 , however , pairs of particles were again observed ( Figure 1—figure supplement 3i ) . Either two molecules of FGF21 bind to one βKlotho , or FGF21 induces dimerization of overexpressed βKlotho , even in the absence of receptor . Because overexpression of FGFR1c and βKlotho may lead to receptor auto-activation ( Sørensen et al . , 2006 ) , and with a view to studies on intact cells ( see below ) , we repeated the analysis with AuNP-FGF21 on membrane preparations from human adipocytes . As before , tilt series were collected , followed by tomographic reconstruction , revealing 89% of AuNPs in true pairs , with an average separation ( center-to-center distance ) of 80 ± 15 Å ( Figure 1 ) . With use of the AuNPs to improve the alignment of the tilt series ( Figure 1—figure supplement 4 ) , protein densities on both inner and outer surfaces of the membrane were revealed ( Figure 1a and Figure 1—figure supplement 4c ) . A key requirement for extension of the analysis to intact cells is sufficient thinness of the cells for cryo-EM . CHO cells were not well suited in this regard , but cytoplasmic regions of adipocyte cells grown on Holey-Carbon Au mesh grids were as thin as 200–300 nm near the cell periphery ( Figure 2 and Figure 2—figure supplement 1 ) . As in the case of vesicles from CHO and adipocyte cells membrane preparations , most AuNP particles were in pairs ( 88% ) on the adipocyte cell surface ( Figure 3 ) . AuNP pairs showed a tendency to cluster , consistent with previous reports of clustering of FGF receptors from immunofluorescence studies with anti-FGFR antibodies ( Gao et al . , 2015 ) . AuNP pairs were found in areas surrounding filipodia and , most notably , above invaginations of the cell surface membrane with clathrin nets beneath ( Figures 2 and 3 ) . The occurrence of most AuNP-FGF21 in pairs pertains to the stoichiometry of the ternary complex . Our findings are suggestive of the occurrence of 2:2:2 FGF21-FGFR1c-βKlotho complexes in vivo . A number of familiar structures were visible in the tomograms of adipocyte cells ( Figures 2 and 3 ) : membranes ( both cell surface and vesicular ) , clathrin nets , actin filaments , microtubules , and hexameric rings . The resolution of the tomograms was sufficient to distinguish intercalating legs of neighboring clathrin triskelions ( Fotin et al . , 2004 ) ( Figure 4a ) . Actin filaments and microtubules were confirmed by helical reconstruction and docking high-resolution structures into the reconstructions ( Figure 4b ) . Hexameric rings , averaged from subtomograms , corresponded in outline and dimensions to the p97 AAA+ ATPase , although NSF and Vps4p , with similar structures , could not be excluded ( Figure 4c ) . When grids were exposed to the AuNP-FGF21 conjugate for 1 h at 4°C and transferred to 22°C before freezing , AuNP pairs were observed in clathrin-coated vesicles , about 100 nm in diameter ( Figure 2 , Figure 3 and Figure 3—figure supplement 1 ) , similar in size to clathrin-coated vesicles isolated from cells , but larger and less regular in shape than vesicles assembled in vitro ( Kirchhausen et al . , 2014 ) . After 1 h at 37°C , AuNP pairs were observed in endosomes ( Figure 3 and Figure 3—figure supplement 1d ) . Not only were almost all AuNPs paired in both clathrin-coated vesicles and endosomes ( 89% and 88% , respectively ) , but they were also invariably adjacent to the inner membrane surface , pointing to persistence of the FGF21-FGFR1c-βKlotho complex . Finally , after overnight incubation at 37°C , AuNPs were observed in multivesicular bodies ( MVBs ) . Among 44 AuNPs observed inside five MVBs in different cells , no two AuNPs were closer than 250 Å to one another . AuNPs in MVBs were not only unpaired but also unassociated with the vesicle membranes , indicating the disruption of the FGF21-FGFR1c-βKlotho complex in MVBs ( Figure 3 and Figure 3—figure supplement 1f ) . Our findings demonstrate a clathrin-dependent pathway , and point to accessory factors in the process . Thus , clathrin pits were seen to be associated with abundant actin filaments , including y-shaped filaments ( Figure 2 and Figure 2—figure supplement 2 ) , and with hexameric rings ( Figures 2 and 3 ) . Clathrin nets were clearly resolved in 11 tomograms coming from nine different cells . In all cases the nets were surrounded by y-shaped actin . In 10 of the 11 tomograms , at least one hexameric ring was found within 50 nm of the net , and hexameric rings were observed in all cases if the search was expanded to 75 nm from the net . The number of hexameric rings within 75 nm varied among nets from one to 21 . The association of hexameric rings with clathrin nets was supported by the orientation of the rings . The bottom surface of the rings was larger than the top surface ( Figure 4c ) and the bottom was always oriented toward clathrin ( Figure 5 ) . Our findings are in keeping with the literature regarding the role of actin filaments and of y-shaped filaments in clathrin-mediated endocytosis ( Kaksonen et al . , 2006 ) , and also in keeping with the literature regarding p97-clathrin interaction and the involvement of p97 in endosomal sorting ( Meyer et al . , 2012 ) . Our findings go further , showing persistence of the FGF21-FGFR1c- βKlotho complex in endosomes , and disruption of the complex in MVBs . The example of an MVB shown here lies in proximity to a long microtubule ( Figure 3 and Figure 3—figure supplement 1f ) . As some FGFs travel all the way to the nucleus ( Sørensen et al . , 2006 ) , and membrane vesicles are transported along microtubules , the MVB may be involved in transport of FGF to the nucleus . Our results from imaging AuNPs in human adipocytes by cryo-ET are of both mechanistic and methodological significance . They contribute to the emerging picture of the FGF signalling mechanism and trafficking inside cells . They show that two copies of FGF21 are present in the FGF21-FGFR1c-βKlotho ternary complex in cells , and that two copies of βKlotho are present as well , pointing to an overall 2:2:2 stochiometry . Second , FGF21-FGFR1c-βKlotho complexes undergo clathrin-dependent endocytosis . Information from multiple tomograms shows that the ternary complexes undergo clathrin-dependent endocytosis and gives a three-dimensional picture of the entire pathway . The same approach can be applied to other FGFs . Further study of both endocrine and paracrine FGFs would shed light on the complex regulation of FGFRs-induced signaling cascades . With regard to methodological significance , our findings extend previous investigations by EM tomography of plastic-embedded sections and by cryo-EM of protein-receptor complexes in liposomes , performed with the use of commercial gold nanoparticle preparations ( He et al . , 2008; He et al . , 2009 ) . In the future , AuNPs of different sizes ( Azubel et al . , 2014; Azubel et al . , 2017 ) conjugated with different antibodies may be used to track multiple components of a receptor complex at the same time . The approach may be used not only for tracking a variety of cargos but also , by the introduction of AuNP-scFv conjugates in cells , for studies of other intracellular transactions . E38C-FGF21 ( Xu et al . , 2013 ) and a single chain antibody fragment ( scFv ) against βKlotho were conjugated with 3MBA-Au144 nanoparticles ( NPs ) ( Azubel et al . , 2017 ) as described ( Azubel and Kornberg , 2016 ) with minor modifications . Briefly , 200 μM E38C-FGF21 or 34 μM anti-βKlotho scFv were reduced with 1 mM TCEP for 1 h at 37°C . Reduced E38C-FGF21 was incubated on ice for 15 min , and reduced anti-βKlotho scFv was incubated for 45 min at 37°C , in the presence of twofold excess of 3MBA-Au144 NPs in both cases . Conjugates were passivated by treatment with 2 . 5 mM glutathione ( GSH ) for 30 min on ice ( AuE38C-FGF21 ) or 45 min at 37°C ( anti-βKlotho scFv ) . Passivated conjugates were run in a 10% glycerol , 12% polyacrylamide gel in Tris-borate-EDTA buffer at 150 V . The gel band corresponding to the conjugate was excised , and crushed and soaked overnight in PBS . AM-1/D Chinese Hamster Ovary ( CHO ) cells stably expressing both human βKlotho and human FGFR1c ( Amgen proprietary cell line derived from CHO cells previously characterized ( Hecht et al . , 2012; Shi et al . , 2018 ) ) were suspended in 50 ml buffer containing 10 mM HEPES pH 7 . 5 , 100 mM NaCl , 1 mM EDTA , and one tablet protease inhibitor ( Roche ) . Cells were lysed by Dounce Homogenization ( 30 strokes on ice ) , followed by a spin at 1000 rpm for 10 min . Supernatant was transferred to a 50 ml centrifuge tube and volume was brought up to 40 ml before centrifugation at 16 , 000 rpm for 30 min . The pellet was resuspended in 1 ml buffer ( 10 mM HEPES pH 7 . 5 , 100 mM NaCl , 1 mM EDTA ) . 10 µg of anti-βKlotho were added followed by incubation at room temperature for 2–3 h . 100 µl 50% slurry protein A beads were added and sample was rotated for 1 h at room temperature . Beads were let to settle down and washed with 10 mM HEPES pH 7 . 5 , 100 mM NaCl , 1 mM EDTA twice . 10 µl Caspase three were added and the sample was incubated overnight at 4°C . 1 ml buffer ( 10 mM HEPES pH 7 . 5 , 100 mM NaCl , 1 mM EDTA ) was added and the sample was transferred to a centrifugation tube for a 30 min spin at 16 , 000 rpm . The pellet was washed twice , resuspended in 40 µl buffer ( 10 mM HEPES pH 7 . 5 , 100 mM NaCl , 1 mM EDTA ) and stored at −80°C . 7-day differentiated human adipocyte cells were suspended in 50 ml of PBS buffer containing one tablet protease inhibitor ( Roche ) . Cells were lysed by Dounce Homogenization ( 30 strokes on ice ) , followed by centrifugation at 1000 rpm for 10 min . Supernatant was transferred to a 50 ml centrifuge tube and volume was brought up to 40 ml before centrifugation at 16 , 000 rpm for 30 min . The pellet was then resuspended in 40 µl PBS and stored at −80°C . Membrane preparations ( ~5 mg/ml ) from 12 different experiments were incubated with either AuE38C-FGF21 ( 0 . 03 mg/ml ) or anti-βKlotho scFv ( 0 . 03 mg/ml ) on ice for 30 min . The sample was centrifugated and washed with 1X PBS three times , or until the supernatant was colorless . 2 . 5 µl resuspended membranes were mixed with 0 . 5 µl 10 nm BSA Gold Tracer ( EMS , Haltfield , PA , USA ) before applying to glow discharged 200 mesh copper R2/2 Quantifoil grids ( Quantifoil Micro Tools GmbH , Jena , Germany ) . Blotting and plunge-freezing into liquid ethane ( at −178°C ) were performed with a Leica EM GP ( Leica Microsystems , Wetzlar , Germany ) set to 5 s pre-blotting time , 6 s blotting time , no post-blotting time , 22°C and 90% humidity . One vial of Cryoperserved Human Subcutaneous Preadipocyte cells ( Zen Bio , NC , USA ) was thawed by immersing in a 37°C water bath and gently shaking . Cells were transferred to a 50 ml tube containing 9 ml of pre-warmed Subcutaneous Preadipocyte Growth Medium ( PM-1 ) ( Zen Bio , NC , USA ) . Cells were centrifugated for 3 min at 1200 rpm . Medium was aspirated , and cells were resuspended in 5 ml PM-1 and transferred to a 75 cm2 flask containing 10 ml of pre-warmed PM-1 . Cells were grown in an incubator at 37°C in the presence of 5% CO2 , for 24 h , or until they were confluent . PM-1 was aspirated and 15 ml of Adipocyte Differentiation Medium ( DM-2 ) ( Zen Bio , NC , USA ) was added . Differentiation proceeded for 5–7 d in an incubator at 37°C in the presence of 5% CO2 . Medium was aspirated and cells were washed with 10 ml pre-warmed 1X PBS , before adding 3 ml pre-warmed CellStripper ( Corning , VA , USA ) . The flask was put back into a 37°C incubator for 5–10 min , or until the cells lifted off the plate . Cells were washed off with 7 ml of 1X PBS , collected in a 50 ml tube , and centrifugated for 3 min at 1200 rpm . Cells were resuspended in DM-2 at a density of ~105 cells/ml and plated in six-well plates , containing three to four pre-treated 10 nm BSA Gold Tracer ( EMS , Haltfield , PA , USA ) and fibronectin-coated 200 mesh gold R2/2 London finder Quantifoil grids ( Quantifoil Micro Tools GmbH , Jena , Germany ) per well . After overnight incubation at 37°C in the presence of 5% CO2 , the grids were placed upside down in a nine-well Teflon plate containing 30 µl drops of 35 µM AuE38C-FGF21 , incubated on ice , or at room temperature , or 37°C for 1 h , or at 37°C overnight , and washed with 1X PBS . Grids were mounted onto Leica EM GP ( Leica Microsystems , Wetzlar , Germany ) so grids could be blotted from the reverse side . Before blotting and plunge-freezing , 3 µl of 10 nm BSA Gold Tracer ( EMS , Haltfield , PA , USA ) were added . Blotting and plunge-freezing into liquid ethane ( at −180°C ) were performed with a Leica EM GP ( Leica Microsystems , Wetzlar , Germany ) set to 2 s pre-blotting time , 4 s blotting time , no post-blotting time , 22°C and 95% humidity . Cells grown on grids from more than 20 experiments were taken for cryo-ET data collection . Tilt series were collected either on a FEI ( Eindhoven , The Netherlands ) Tecnai F20 FEG transmission electron microscope operating at 200 kV , or on a FEI ( Eindhoven , The Netherlands ) F30 G2 Polara FEG transmission electron microscope operating at 300 kV and equipped with an energy filter ( slit width 20 eV for higher magnifications; Gatan , Inc . ) . Images were recorded using a 4k × 4k K2 Summit direct detector ( Gatan , Inc . ) operating in the electron counting mode . Tilt series were recorded using SerialEM ( Mastronarde , 2005 ) software at magnifications with corresponding pixel sizes ranging from 1 . 28 to 2 . 42 Å . Either a bidirectional or a dose-symmetric tilt schemes ( Hagen et al . , 2017 ) were implemented from −60° to +60° with an increment of 2° at 2–6 µm underfocus , and total dose around 120 e-/Å2 . Tilt-series were aligned and processed with the IMOD software package ( Kremer et al . , 1996 ) . After binning the aligned tilt series by threefold , reconstructions into 3D tomograms were done with back projection , which helps to unequivocally identify Au nanoparticles , and with SIRT ( Simultaneous Iterative Reconstruction Technique ) for increased contrast . Subtomogram 3D-averaging and helical reconstruction were performed using PEET software package ( Heumann et al . , 2011 ) . Initial segmentation was done with IMOD software package ( Kremer et al . , 1996 ) and Chimera software package ( Pettersen et al . , 2004 ) was used for visualization and docking of pdb structures into density maps .	Following a molecule’s movement around a cell is a bit like looking for a needle in a haystack . Cells contain thousands of different components that can be difficult to distinguish between when viewed using a microscope . It helps to have a method to tag the molecule of interest to make it more easily visible . Electron microscopes can capture images that reveal much finer details than traditional light microscopes . To create an electron microscope image , a high-powered beam of electrons strikes the molecules in the sample being studied . Heavier atoms scatter electrons more strongly than lighter atoms , thus , fewer electrons reach the detector and the atoms appear darker in the images . Gold atoms are heavier than the atoms that make up biological molecules ( mostly carbon , nitrogen and oxygen ) . ‘Tagging’ molecules that you want to study using clusters of gold atoms would therefore help to highlight them inside cells . Azubel et al . have now developed a method to attach gold nanoparticles to small molecules , and used the technique to track the movement of a protein called fibroblast growth factor 21 ( FGF21 ) in human fat cells . It had previously been discovered that rats fed a high fat diet live longer and do not gain weight when treated with FGF21 . Understanding how FGF21 works could therefore help researchers to develop new treatments for obesity and type II diabetes . Azubel et al . captured many electron microscope images of cells containing tagged FGF21 proteins . This revealed that two copies of the protein work together . First , each copy of FGF21 attaches to a receptor on the surface of the cell . The two FGF21-receptor pairs bind together to form part of a larger ‘complex’ . The complex is engulfed by part of the nearby cell membrane , which pinches off from the rest of the membrane to form a compartment known as a vesicle . The FGF21-receptor complex stays bound together as the vesicle travels along the cell’s internal skeleton . Eventually , portions of the vesicle’s membrane ‘bud’ to form a new compartment called a multivesicular body . At this point , the FGF21 proteins and the receptors separate from each other . Future work could build on these results in an effort to improve how we treat obesity and type II diabetes . The gold nanoparticle tracking technique developed by Azubel et al . could also be used to track other proteins using electron microscopy . This opens the way to determining the structures that proteins form when they are inside cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "cell", "biology", "tools", "and", "resources" ]	2019	FGF21 trafficking in intact human cells revealed by cryo-electron tomography with gold nanoparticles
The mutualistic endosymbiosis between cnidarians and dinoflagellates is mediated by complex inter-partner signaling events , where the host cnidarian innate immune system plays a crucial role in recognition and regulation of symbionts . To date , little is known about the diversity of thrombospondin-type-1 repeat ( TSR ) domain proteins in basal metazoans or their potential role in regulation of cnidarian-dinoflagellate mutualisms . We reveal a large and diverse repertoire of TSR proteins in seven anthozoan species , and show that in the model sea anemone Aiptasia pallida the TSR domain promotes colonization of the host by the symbiotic dinoflagellate Symbiodinium minutum . Blocking TSR domains led to decreased colonization success , while adding exogenous TSRs resulted in a ‘super colonization’ . Furthermore , gene expression of TSR proteins was highest at early time-points during symbiosis establishment . Our work characterizes the diversity of cnidarian TSR proteins and provides evidence that these proteins play an important role in the establishment of cnidarian-dinoflagellate symbiosis . Host-microbe interactions , both beneficial and detrimental , are ancient and ubiquitous , and are mediated by a myriad of molecular and cellular signalling events between the partners . Hosts are under selective pressures to develop recognition mechanisms that tolerate beneficial symbionts and destroy negative invaders , while microbes evolve to successfully invade and either benefit or exploit their hosts ( Eberl , 2010; Bosch and McFall-Ngai , 2011 ) . Cnidarian-dinoflagellate mutualisms , such as those that form coral reefs , are one such host-microbe interaction for which we are just beginning to uncover the molecular conversations between partners that result in the establishment and maintenance of a healthy partnership ( Davy et al . , 2012 ) . Most cnidarian-dinoflagellate partnerships are established anew with each cnidarian host generation . The photosynthetic dinoflagellates ( Symbiodinium spp . ) are taken from the environment into host gastrodermal cells via phagocytosis and , instead of being digested , the symbionts persist and colonize the host . Discovery-based , high-throughput ‘omics' techniques have previously been employed to uncover candidate genes and pathways that could play a role in inter-partner recognition and regulation processes in cnidarian-dinoflagellate symbioses ( Meyer and Weis , 2012; Mohamed et al . , 2016 ) . Two such transcriptomic studies comparing expression patterns of symbiotic and aposymbiotic individuals of the sea anemone species Anthopleura elegantissima and Aiptasia pallida ( Rodriguez-Lanetty et al . , 2006; Lehnert et al . , 2014 ) , started us down a path to an in-depth examination of thrombospondin-type-1-repeat ( TSR ) -domain-containing proteins ( hereafter referred to as TSR proteins ) in both partners of the symbiosis . Both studies found significant upregulation of a homologue to a scavenger receptor type B1 ( SRB1 ) in symbiotic anemones . The structure and diversity of SRB1s have now been characterized in a variety of cnidarians , including A . elegantissima and A . pallida ( Neubauer et al . , 2016 ) . SRB1s function in innate immunity in metazoans in a variety of ways , including , in mammals , activation of the tolerogenic , immunosuppressive transforming growth factor beta ( TGFβ ) pathway ( Asch et al . , 1992; Masli et al . , 2006; Yang et al . , 2007 ) . When the TSR domains of the extracellular matrix glycoprotein thrombospondin bind to CD36 , latent TGFβ is converted to its active form , which in turn launches tolerogenic pathways downstream . Subsequent studies in another sea anemone model system , A . pallida , demonstrated a role for TGFβ in the regulation of cnidarian-dinoflagellate symbioses ( Detournay et al . , 2012 ) . This warranted further examination of genes related to TGFβ pathway activation and turned our focus to thrombospondins . Our initial search for thrombospondin and other TSR protein homologues revealed a rich literature on thrombospondin-related anonymous proteins ( TRAPs ) that play important roles in apicomplexan endoparasites , such as when Plasmodium attaches to and invades mammalian host cells ( Kappe et al . , 1999; Vaughan et al . , 2008; Morahan et al . , 2009 ) . Specifically , the WSPCSVTCG motif ( Figure 1 ) within the TRAP TSR binds sulfated glycoconjugates on host cells ( Morahan et al . , 2009 ) . This piqued our interest in TSRs even more , because apicomplexans and dinoflagellates are sister taxa within the alveolates ( Burki et al . , 2008; Adl et al . , 2012 ) . There is therefore the potential for homologous strategies of symbiont invasion and persistence in hosts that spurred our interest in a deeper investigation of TSR homologues within Symbiodinium , as well as within host cnidarians . 10 . 7554/eLife . 24494 . 003Figure 1 . Schematic representation of human thrombospondin 1 protein . The three TSR ( Thrombospondin Structural homology Repeat ) domains are depicted by three red diamonds . The amino acid sequence of the second TSR sequence is shown with six conserved cysteines in red . Known binding motifs and capabilities of the human thrombospondin TSR domain two are listed and depicted in boxes . ( Redrawn from Zhang and Lawler , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 003 The large TSR protein superfamily includes mammalian thrombospondins ( depicted in Figure 1 ) , and many proteins in metazoans and other eukaryotes ( Adams and Tucker , 2000; Tucker , 2004 ) . The superfamily is composed of secreted and transmembrane proteins with a large array of functions involving protein-protein and other steric interactions . TSR superfamily members are diverse , suggesting that the highly-conserved TSR domain has been duplicated and shuffled numerous times among superfamily members . For example , 41 human genes contain one or more TSR domain copies ( Silverstein , 2002 ) , while there are 27 and 14 TSR superfamily members in C . elegans and Drosophila , respectively ( Tan et al . , 2002 ) . The TSR domain consists of approximately 60 amino acids ( Figure 1 ) , with several highly conserved motifs and five or six conserved cysteine residues that participate in disulfide bridge formation and domain folding ( Adams and Lawler , 2011 ) . Thrombospondins were originally characterized in mammals . They are extracellular , multi-domain , calcium-binding glycoproteins that play pleiotropic tissue-specific roles involving interactions with cell surfaces , cytokines and the extracellular matrix ( Adams and Lawler , 2004 ) . Protein-protein interactions involving the TSR domain , including binding to SRB1/CD 36 ( see Figure 1 ) , are central to thrombospondin protein function . A systematic search for TSR proteins across the Cnidaria has not been conducted to date . However , a study of vertebrate thrombospondin protein homologues in Nematostella vectensis found that , although most of the multi-domain architecture is present , crucially , the three TSR domains are missing ( thus adding a confusing naming problem to the categorization of these genes ) ( Bentley and Adams , 2010; Tucker et al . , 2013 ) . There is , however , growing evidence that cnidarians possess numerous genes that contain TSR domains . Two rhamnospondin genes with eight TSR domains were identified in the colonial hydroid Hydractinia symbiolongicarpus that are expressed in the hypostome of feeding polyps and were proposed to function in microbe binding ( Schwarz et al . , 2007 ) . A study in Hydra oligactis also demonstrated high expression of several genes for TSR proteins in the hypostome and proposed potential functions in nerve net development or defense ( Hamaguchi-Hamada et al . , 2016 ) . Within anthozoans , several TSR proteins were identified in two species of corals , Acropora palmata and Montastraea faveolata ( Schwarz et al . , 2008 ) , and in a study identifying candidate symbiosis genes across ten cnidarian species ( Meyer and Weis , 2012 ) . Therefore , while a number of studies have focused on characterization and localization of cnidarian TSR proteins , their proposed functions have not yet been investigated . The aim of this study was to characterize and compare the TSR protein repertoire of seven cnidarian species ( six symbiotic , one non-symbiotic ) and two symbiotic dinoflagellate species , to identify putative ligands for SRB1/CD36 in host sequence resources and TRAP-like proteins in the Symbiodinium genome . Using six anthozoan genomic and transcriptomic resources , we compared vertebrate TSR proteins of known function with the cnidarian TSR repertoire . We investigated the presence of known binding motifs and their conservation within the cnidarian TSR domains . In addition , we explored the function of TSR proteins in cnidarian-dinoflagellate symbiosis , using the sea anemone A . pallida , a globally-adopted model system for the study of this symbiosis ( Weis et al . , 2008; Goldstein and King , 2016 ) . We tested the hypothesis that TSR proteins are involved in symbiont colonization of the host during onset of symbiosis , and whether the proteins of interest are of host or symbiont origin . Functional studies were performed in which TSR-domain function was blocked , or exogenous TSRs were added to determine the effect on colonization levels at the onset of symbiosis . Overall , we describe a diverse TSR protein repertoire in anthozoans that contains homologues to known vertebrate proteins in addition to novel domain combinations . In addition , we provide functional evidence for the importance of host-derived TSR proteins in the establishment of the cnidarian-dinoflagellate symbiosis . The overall numbers of TSR proteins identified from the four genomes , N . vectensis , A . pallida , A . digitifera , and S . pistillata were much higher than those identified from transcriptomes . Searches revealed a rich and diverse repertoire of TSR proteins within the seven anthozoan species , when compared to mammalian TSR superfamily members of known function; the largest groups identified were the ADAMTS metalloproteases and the properdin-like TSR-only proteins ( Figure 2 ) . Putative thrombospondins with similar domain structure to human thrombospondins 3 , 4 and 5 were identified in all species . None of the cnidarian resources searched contained a thrombospondin-like protein with TSR domains . Large numbers of TSR-only proteins were identified in comparison to those known in mammals , where complement factor properdin is the only example of a protein containing only TSR repeats aside from a signal sequence . TSR protein sequences containing novel protein domain architecture were also identified , including those with astacin metalloproteases , von Willebrand factors ( VWAs ) , trypsin , Stichodactyla helianthus K+ channel toxin ( ShK ) domains and immunoglobulin domains ( Figure 2 ) . 10 . 7554/eLife . 24494 . 004Figure 2 . Domain architecture of cnidarian TSR super-family proteins compared to known vertebrate TSR-domain-containing proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 00410 . 7554/eLife . 24494 . 005Figure 2—figure supplement 1 . The TSR domain is very well conserved from cnidarians up to humans , with binding motifs for glycosaminoglycans ( GAGs ) and the type B scavenger receptors , CD36/SRB1 . All three-dimensional folding sites are present as described by Tan et al . ( 2002 ) for the crystal structure of human TSP1 TSR2 . Six conserved cysteine residues are highlighted in yellow and form three disulfide bridges ( C1–C5 , C2–C6 and C3–C4 ) . Three conserved tryptophan residues are shown in blue boxes and mark the ‘WXXW’ protein-binding motif . Amino acids that form the R layers are marked with purple boxes , and pairings forming 3 R layers are as follows: R3-R4 , R2-R5 and R1-R6 . The Β strands are annotated at the bottom in blue strands A , B and C . Please refer to Tan et al . ( 2002 ) for a more detailed explanation of the three-dimensional folding . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 005 TSR domains taken from a selection of identified cnidarian TSR proteins , show very strong amino acid sequence homology to the second TSR repeat in the human thrombospondin 1 protein ( Figure 2—figure supplement 1 ) . Features contributing to the three-dimensional folded protein described from the crystal structure of the TSR repeat of human thrombospondin 1 ( Tan et al . , 2002 ) are present in the cnidarian TSRs , including: ( 1 ) six cysteine residues , shown to form disulfide bridges; ( 2 ) three tryptophan residues forming the WXXWXXW motif which participates in protein and glycosaminoglycan binding sites ( GAG binding ) ; and ( 3 ) polar residues ( such as arginine , lysine and glutamine ) present in the RXRXR motif , forming salt bridges with other polar residues that aid in folding . In addition , all sequences contain the CSVTCG and GVQTRXR motifs , which bind SRB1/CD36 ( Zhang and Lawler , 2007 ) . Searches of the S . minutum genome identified 175 contigs containing TSR domains , however none of the predicted proteins contained VWA domains ( Figure 3 ) . TSRs were alone or in repeats of up to 16 . In contrast , most apicomplexan TSR protein sequences possess one or more VWA domains and all have a C-terminal transmembrane domain . Searches of the S . microadriaticum genome revealed similar results and included proteins containing only the TSR domains in repeats up to 20 . An alignment of TSR domains , including those from apicomplexan TRAP proteins , human thrombospondins 1 and 2 , S . minutum , S . microadriaticum and two cnidarian TSR proteins is shown in Figure 3—figure supplement 1 . S . minutum TSRs have five or six cysteines , a variation that is consistent with apicomplexan TRAP proteins ( Morahan et al . , 2009 ) . The CD36/SRB1 but not the GAG-binding sites are well conserved in S . minutum sequences . In contrast , S . microadriaticum TSR domains contain six cysteines and are more similar to human and cnidarian TSRs than apicomplexan TSR domains . 10 . 7554/eLife . 24494 . 006Figure 3 . Schematic representation of members of the TSR gene family in dinoflagellates and apicomplexan parasites . TSRs from the dinoflagellates Symbiodinium minutum and S . microadriaticum are shown in green . Apicomplexan TRAP proteins are shown in beige . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 00610 . 7554/eLife . 24494 . 007Figure 3—figure supplement 1 . TSR domain alignment compares apicomplexan TRAP TSR domains with TSR domains from the dinoflagellates Symbiodinium minutum and S . microadriaticum , TSR 2 from human TSP1 , and ADAMTS-like TSR domains from the anemones Nematostella vectensis and Aiptasia pallida . Positioning and absence of specific cysteine residues ( colored yellow ) in TRAP and Symbiodinium TSRs will result in different patterns of disulfide bonds and three-dimensional folding . Binding sites for glycosaminoglycans ( GAGs ) and the scavenger receptors CD36/SRB1 ( annotated in red ) are somewhat conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 007 Anti-human TSR labelled two bands of 40 and 47 kDa in immunoblot analysis of homogenates from symbiotic A . pallida protein and a single band at 40 kDa in aposymbiotic A . pallida ( Figure 4A and Figure 4—figure supplement 1 ) . Immunofluorescence of S . minutum using anti-human TSR showed label on freshly isolated but not cultured cells . Dil lithophylic membrane stain labelled freshly isolated but not cultured S . minutum cells ( Figure 4—figure supplement 2 ) . Likewise , anti-TSR signal was absent from cultured S . minutum cells ( Figure 4B ) but appeared around the outside of freshly-isolated S . minutum cells ( Figure 4C ) , suggesting that it labels the host symbiosome membrane complex and/or host material associated with the freshly isolated cells . Immunofluorescent labelling of symbiotic anemone tentacle cryosections showed antibody label in host gastrodermal tissue when in close association with resident symbionts ( Figure 4D , E ) . Secondary antibody-only and IgG controls showed no labeling ( Figure 4F ) . 10 . 7554/eLife . 24494 . 008Figure 4 . Immuno-analyses using anti-thrombospondin show evidence of TSRs in symbiotic anemone host tissues . ( A ) Immunoblots of symbiotic ( SYM ) and aposymbiotic ( APO ) A . pallida label bands at 40 and 47 kDa in symbiotic anemones and a single band at 40 kDa in aposymbiotic anemones . ( B , C ) Confocal images of dinoflagellate cells taken from ( B ) culture or ( C ) freshly isolated cells taken from A . pallida homogenates . A fluorescent probe conjugated to anti-human thrombospondin does not label cells from culture ( B ) but strongly labels host cell debris and/or membranes associated with freshly isolated cells ( C ) . ( D , E ) Confocal images of cryosections from symbiotic A . pallida gastrodermal tissue stained with anti-thrombospondin at lower ( D ) and higher ( E ) magnification . Anti-thrombospondin labelling is evident in host tissues surrounding symbionts . ( F ) Confocal image of control anemone cryosections incubated with secondary antibody only . No anti-thrombospondin labeling is evident . Green = anti-thrombospondin , Red = algal autofluorescence , blue = DAPI stain of host and symbiont nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 00810 . 7554/eLife . 24494 . 009Figure 4—figure supplement 1 . A: Actin control for immunoblot blot in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 00910 . 7554/eLife . 24494 . 010Figure 4—figure supplement 2 . Lipophilic membrane staining of dinoflagellate cells using Dil . Lipophilic membrane stain Dil was absent from ( A ) cultured algae but present in ( B ) freshly isolated symbionts . This is evidence of the presence of a symbiosome membrane surrounding freshly isolated symbionts . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 010 Incubation of aposymbiotic anemones with anti-human TSR prior to and during symbiont inoculation resulted in strong and statistically significant ( mixed effects ANOVA F ( 2 , 24 ) =16 . 55 , p<0 . 0001 ) inhibition of host colonization by S . minutum ( Figure 5A ) . Levels of colonization stayed very low throughout the treatment period , rising to only 1 . 26 ± 0 . 86% . In contrast , anemones incubated in both the FSW and IgG antibody controls showed moderate rates of colonization for the first 72 hr , but a dramatic increase thereafter to 18 . 1 ± 2 . 65% and 17 . 8 ± 2 . 56% , respectively , by 120 hr post-inoculation . 10 . 7554/eLife . 24494 . 011Figure 5 . Kinetics of recolonization after antibody and peptide treatments . ( A ) Anemones pre-incubated in an anti-human thrombospondin ( green ) show decreased colonization success compared FSW-only ( light blue ) and IgG ( orange ) controls . Inset: confocal images show representative tentacle slices at 72 hr post-inoculation . ( B ) The addition of exogenous human thrombospondin-1 ( purple ) significantly increased the colonization rate during colonization , compared to control anemones in FSW ( blue ) . Inset confocal images show representative tentacle slices at 96 hr post-inoculation . ( C ) The effect of synthetic TSR peptides 1 ( blue ) and 2 ( orange ) on colonization rates compared to the control anemones in FSW . Anemones treated with both peptides 1 and 2 showed increased uptake of algae during colonization . Statistical significance of treatment effects was assessed using mixed effects models , with contrasts calculated between individual treatments and FSW at each time-point; **p<0 . 001; p<0 . 05; p<0 . 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01110 . 7554/eLife . 24494 . 012Figure 5—source data 1 . Source data used for statistical analyses described in results and depicted in Figure 5A: Long-form table with experimental results described in the results section Blocking TSR domains inhibits symbiont uptake by host anemones and shown in Figure 5A . Treatments labels are FSW: Filtered Sea Water , anti-TSR: anti-human thrombospondin antibody , Igg: IgG control . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01210 . 7554/eLife . 24494 . 013Figure 5—source data 2 . Summary statistics ( mean and s . e . ) displayed in Figure 5A . Summary statistics for results in section Blocking TSR domains inhibits symbiont uptake by host anemones as shown in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01310 . 7554/eLife . 24494 . 014Figure 5—source data 3 . Source data used for statistical analyses described in results and depicted in Figure 5B . Long-form table with experimental results described in the results section Addition of exogenous human thrombospondin-1 results in ‘super colonization’ of hosts by symbionts and shown in Figure 5B . Treatments labels are FSW: Filtered Sea Water , Hs-TSR: Homo sapiens exogenous TSR protein treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01410 . 7554/eLife . 24494 . 015Figure 5—source data 4 . Summary statistics ( mean and s . e . ) displayed in Figure 5B . Summary statistics for results in section Addition of exogenous human thrombospondin-1 results in ‘super colonization’ of hosts by symbionts as shown in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01510 . 7554/eLife . 24494 . 016Figure 5—source data 5 . Source data used for statistical analyses described in results and depicted in Figure 5C . Long-form table with experimental results described in the results section Addition of exogenous A . pallida TSR peptide fragments during inoculation increases colonization success and shown in Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01610 . 7554/eLife . 24494 . 017Figure 5—source data 6 . Summary statistics ( mean and s . e . ) displayed in Figure 5C . Summary statistics for results in section Addition of exogenous A . pallida TSR peptide fragments during inoculation increases colonization success as shown in Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 017 Addition of exogenous human thrombospondin-1 protein increased the rate of host colonization by symbionts . Anemones pre-treated with thrombospondin-1 showed markedly increased ( mixed effects ANOVA F ( 1 , 16 ) = 59 . 36 , p<0 . 0001 ) colonization success compared to FSW controls ( Figure 5B ) . Colonization success after 48 hr was 8 . 05 ± 0 . 98% in the thrombospondin-1 treatment compared to 1 . 18 ± 0 . 28% in the FSW treatment . By 96 hr post-inoculation , colonization success had risen to 25 . 1 ± 2 . 6% in the thrombosondin-1 treatment compared to just 9 . 87 ± 2 . 4% in the FSW control . By the end of the experiment , at 120 hr post-inoculation , colonization levels in control animals had almost caught up to those in treatment ones , suggesting that the stimulatory impact of thrombospondin-1 was most pronounced during the first 96 hr of symbiosis establishment . As with human thrombospondin-1 , pre-treating anemones with short synthetic A . pallida TSR peptides resulted in increased colonization success ( mixed effects ANOVA F ( 2 , 24 ) = 69 . 46 , p<0 . 0001; Figure 5C ) . At 48 hr post-inoculation , symbiont levels were higher in anemones pre-treated with either peptide ( Peptide 1: 11 . 14 ± 1 . 1%; Peptide 2: 11 . 78 ± 0 . 9% ) compared to the FSW-only controls ( 2 . 08 ± 0 . 29% ) . After 48 hr , colonization levels in the Peptide 2 treatment were consistently higher than in the Peptide 1 treatment . This difference was particularly apparent at 72 hr , where colonization levels in anemones in the Peptide 2 treatment were 5% higher than in Peptide 1 ( 20 . 2 ± 1 . 4% and 15 . 11 ± 1 . 98% , respectively ) . The peptide treatments showed the largest increase relative to the FSW control at 96 hr , with 18 . 8 ± 1 . 3% and 20 . 9 ± 1 . 68% colonization for Peptides 1 and 2 , respectively , compared to only 6 . 15 ± 0 . 75% for the FSW control . However , as in the thrombospondin-1 treatment , by the end of the experiment at 120 hr , colonization in the control animals had reached levels similar to those in the peptide-treated anemones , suggesting once again that the impact of TSR peptides was most pronounced early in the colonization process . To investigate the specific TSR proteins involved in the onset of symbiosis , gene expression of two sequences obtained from the bioinformatics searches of the A . pallida genome was measured using quantitative PCR ( qPCR ) . The first sequence , Ap_Sema5 ( AIPGENE5874 ) has a domain structure similar to the vertebrate semaphorin-5 sequence with an N-terminal Sema domain and C-terminal TSR . This sequence was selected for further investigation due to its role in tumor cell motility and invasion through modifications to the actin cytoskeleton ( Li and Lee , 2010 ) , which suggests it could play a role in cytoskeletal rearrangements during symbiont uptake . The second sequence , Ap_Trypsin-like ( similar to AIPGENE 1852 ) , represents a novel domain combination as it possesses two N-terminal ShK domains , four TSR domains , and a C-terminal trypsin domain . The peptide used in the functional experiments described above was designed specifically to this sequence , therefore making it an interesting target for further investigation . Furthermore , in the genome searches , a similar sequence was found in symbiotic species , but not the non-symbiotic Nematostella vectensis , suggesting this protein may play a role in symbiosis . Quantitative PCR results revealed similar expression trends for both Ap_Sema5 and Ap_Trypsin during the onset of symbiosis ( Figure 6 ) . Ap_Sema5 showed a significant upregulation at 12 hr post-inoculation ( estimate: −2 . 26 , 95% c . i . : [−3 . 52; −1 . 01] , p=0 . 0072 ) in the inoculated compared to aposymbiotic treatment , but by 72 hr post-inoculation it was significantly downregulated ( estimate: 1 . 98 , 95% c . i . : [0 . 73; 3 . 23] , p=0 . 015 ) . Ap_Trypsin-like displayed a downward trend in expression during the establishment of symbiosis ( ANOVA , F ( 1 , 10 ) =5 . 90 , p=0 . 036 ) , however individual pairwise comparisons were not significantly different ( see Supplementary Source code 1 for detailed outputs of individual estimates and test statistics , including all pairwise comparisons at individual time-points ) . 10 . 7554/eLife . 24494 . 018Figure 6 . Gene expression of Ap_Sema5 and Ap_Trypsin-like at the onset of symbiosis . The relative quantities from qPCR on the log2 scale are shown for animals that were inoculated with symbionts ( ‘Inoc’; solid line ) and those that remained aposymbiotic ( ‘Apo’; dashed line ) . Bars represent means ± SE ( n = 3 ) and stars represent significantly different levels of expression between the inoc and apo treatments at a particular time point ( two-way ANOVA , Tukey’s post hoc test ) . p<0 . 05 , *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 01810 . 7554/eLife . 24494 . 019Figure 6—source data 1 . Source data used for statistical analyses described in results and depicted in Figure 6 . Long-form table with experimental results described in the results section Ap_Sema-5 expression increases at early time-points during the onset of symbiosis and shown in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 019 Bioinformatic searches revealed a notable diversification of TSR-only proteins . This suggests that TSR proteins can be added to the growing list of immunity genes in cnidarians that are greatly diversified compared to their counterparts in vertebrate genomes . These include expansions of toll-like receptors in A . digitifera , ficolin-like proteins in A . pallida , NOD-like receptors in Hydra magnipapillata and A . digitifera , and scavenger receptors in a variety of cnidarians ( Lange et al . , 2011; Shinzato et al . , 2011; Hamada et al . , 2013; Baumgarten et al . , 2015; Neubauer et al . , 2016 ) . It has been hypothesized that such an expanded repertoire in basal metazoans is an alternate evolutionary strategy to vertebrate adaptive immunity that would enable complex reactions to , and management of , their microbiomes ( Hamada et al . , 2013 ) . The TSR-only repertoire expansion is of particular interest because these sequences are similar to the vertebrate complement protein properdin . This protein is known to have two interrelated functions that may be of particular relevance to the establishment of the cnidarian-dinoflagellate symbiosis . First , properdin can act as a pattern recognition receptor ( PRR ) , detecting microbe-associated molecular patterns ( MAMPs ) on invading microbes and triggering phagocytosis of microbes directly . Secondly , it can participate in the complement system alternative pathway , where it activates and stabilizes the proteolytic C3 convertase complex , which attaches to the surface of invading microbes and hence marks them for phagocytosis and/or lysis ( Hourcade , 2006; Spitzer et al . , 2007 ) . There is growing functional evidence that the complement system , which is classically thought to function in defense against pathogens , also plays a role in the onset and regulation of cnidarian-dinoflagellate symbiosis ( Kvennefors et al . , 2008 , 2010; Baumgarten et al . , 2015; Poole et al . , 2016 ) . Therefore , a testable hypothesis is that cnidarian TSR-only proteins function in a similar manner to vertebrate properdin , as either a PRR to recognize Symbiodinium or to interact with complement proteins to promote phagocytosis of symbionts . A recent transcriptomic study indicated that there was decreased expression of a transmembrane domain-containing-TSR-only protein in A . pallida larvae during the later stages of symbiosis establishment when compared to aposymbiotic larvae ( Wolfowicz et al . , 2016 ) . Therefore , it is possible that this protein may serve as a PRR , with a high level of expression in aposymbiotic larvae and during inter-partner surface recognition , but decreased expression after phagocytosis . We characterized a large repertoire of cnidarian ADAMTS metalloprotease-like proteins . The TSR domains within these cnidarian proteins are highly conserved and functional motifs are intact , including the tryptophan glycosaminoglycan-binding ( GAG ) motif ‘WXXW’ , and scavenger receptor binding motifs ‘CSVTCG’ and ‘GVITRIR’ ( Adams and Tucker , 2000; Silverstein , 2002 ) . In humans , the metalloprotease ADAMTS 13 binds to SRB1 ( Davis et al . , 2009 ) , and in C . elegans an ADAMTS protein ( AD-2 ) is responsible for initiating the TGFβ pathway , regulating body growth and maintaining cuticle formation ( Fernando et al . , 2011 ) . It is therefore conceivable that an ADAMTS-like TSR protein is involved in promoting tolerance in the cnidarian-dinoflagellate symbiosis . TSR proteins containing the trypsin domain , ShK domain and the VWA domain , were present in five of the six symbiotic cnidarians , the trypsin containing TSRs identified in A . digitifera lack the ShK domain ( Figure 2 ) . The ShK domain is found in peptides that function as potassium channel inhibitors and it has been proposed that proteins that include ShK in combination with other domains , such as trypsin , may also modulate channel activity ( Rangaraju et al . , 2010 ) . Additionally , proteins with ShK or TSR domains have previously been found in nematocysts ( Balasubramanian et al . , 2012; Rachamim et al . , 20142015 ) . Therefore , ShK plus trypsin proteins are likely toxin proteins that function in nematocysts and food acquisition . Interestingly , qPCR results indicated that Ap_Trypsin-like has a trend of decreased expression during the establishment of the symbiosis ( Figure 6 ) . Therefore , it is still unclear what role this protein plays in symbiosis . This downward trend could indicate a de-emphasis by the host on food capture , as it transitions to gaining nutritional support from its symbionts . A more detailed comparative study would need to be performed to determine whether these sequences are truly differentially distributed as a function of symbiosis . The comprehensive search for TSR-containing thrombospondin homologues found no sequences in any of the anthozoan resources examined ( Figure 2 ) . This strongly suggests that TSR-containing thrombospondins are not present in cnidarians . However , searches for TSR proteins within anthozoans revealed a rich diversity of TSR superfamily members , including some whose domain architectures bear a strong resemblance to members in other animals and others with novel domain architectures . Domain abundance and architecture show no clear pattern based on symbiotic state or anthozoan phylogeny , but instead correlate to type of resource searched: genomes provide better representation of TSR abundance than transcriptomes . It is likely that a more accurate picture of TSR protein diversity will emerge over time as more genomes become available and annotations improve . Searches of both S . minutum and S . microadriaticum genomes revealed evidence of TSR proteins , but none that had all of the hallmarks of the TRAP proteins in apicomplexans . Symbiodinium TSR sequences contain a signal peptide and multiple TSR repeats , but not the VWA or transmembrane domains found in most apicomplexan TRAPs ( Figure 3 ) . It is therefore unlikely that Symbiodinium is using TSR proteins to attach to hosts via mechanisms homologous to those used by apicomplexans . Expression profiles and localization studies of symbiont TSR proteins in culture vs . in hospite could provide insight into whether these proteins are playing a role in the symbiosis . Interestingly , the number of cysteines contained in the TSR domains differed between the two species . S . minutum TSR domains contained five cysteines , similar to apicomplexan TSRs . In contrast , S . microadriaticum TSRs contained six , similar to metazoan TSRs . We introduced dinoflagellates to aposymbiotic anemones that had been pre-treated to either block or mimic TSR proteins . Blocking TSR domain function resulted in colonization levels reduced to 1% infection and below , providing strong evidence for the involvement of TSR proteins in the establishment of the symbiosis . The anti-human TSR epitope corresponds to three TSR repeats and is therefore indiscriminate in its blocking effect of TSR proteins . Results indicate a role for host , rather than symbiont TSR proteins in symbiosis establishment , given the localization of anti-thrombospondin to host tissues , including those of aposymbiotic anemones , and not the outer surface of cultured Symbiodinium cells ( Figure 4 ) . Treatment of A . pallida with exogenous TSR domains provided further evidence for the role of host TSR proteins in the early onset of the symbiosis . Due to high levels of TSR domain conservation across taxa , synthetic peptides designed from TSR domains have been employed by a number of studies , including determining which motifs bind to CD36 ( Li et al . , 1993 ) and which Plasmodium TSR peptides bind to red blood cells ( Calderón et al . , 2008 ) . The synthetic peptide used in this study contained both the tryptophan GAG-binding motif ‘WXXW’ , and scavenger receptor binding motifs ‘CSVTCG’ and ‘GVXTRXR’ . This result suggests that one or multiples of these binding motifs are involved in successful entry to host cells by the dinoflagellates . Treatment of A . pallida with human thrombospondin and synthetic A . pallida TSR peptides resulted in ‘super colonization’ by the symbionts ( Figure 5B , C ) . These results provide evidence against the hypothesis of membrane-linked host TSRs serving as PRRs to promote inter-partner recognition . We suggest that exogenous TSRs would compete with membrane bound host TSR PRRs for Symbiodinium MAMPs , and result in decreased colonization success . Instead , our results support a hypothesis of TSRs enhancing symbiont colonization through steric interactions with a secondary molecule ( s ) , be it C3 convertase complex , SRB1 , or some other protein that promotes phagocytosis . In this case , addition of exogenous TSRs would result in binding of additional secondary proteins that would in turn promote phagocytosis and result in the ‘super colonization’ observed . This hypothesis is further supported by sequence data which indicate that the majority of cnidarian TSR proteins lack transmembrane domains ( see Supplementary file 1 ) . Our initial interest in the TSR domain was prompted by the search for a binding target for the host cell scavenger receptor SRB1 , which is upregulated in the symbiotic state of A . pallida and another sea anemone , A . elegantissima ( Rodriguez-Lanetty et al . , 2006; Lehnert et al . , 2014 ) . In other systems , SRB1-TSR interactions are implicated in promoting phagocytosis and initiating the tolerance promoting TGFβ pathway by activating latent TGFβ protein ( Khalil , 1999; Murphy-Ullrich and Poczatek , 2000; Koli et al . , 2001 ) . The addition of TSR protein may have dual functions , firstly to enhance phagocytosis of microbes and secondly to promote tolerance . Many intracellular parasites manipulate host innate immune defence mechanisms to their own advantage ( McGuinness et al . , 2003 ) . Gene expression results also provide evidence of a role for TSR proteins at the onset of symbiosis ( Figure 6 ) . Ap_Sema5 showed increased expression at early time points during onset of symbiosis , but decreased expression at later time points , indicating that it may play a role in initial recognition and uptake of symbionts , but not subsequent proliferation . Future experiments that target earlier time points during the onset of symbiosis could provide evidence to support this hypothesis . Interestingly , the decreased trend in expression at 72 hr post-inoculation is similar to the downregulation observed for several TSR protein genes and a non-TSR semaphorin ( Semaphorin-3E ) in symbiotic A . pallida larvae five to six days post-inoculation ( Wolfowicz et al . , 2016 ) . Due to the pleiotropic nature of semaphorins , further investigation of the precise role of Ap_Sema5 is needed . Intriguingly , however , vertebrate semaphorin-5a has been shown to play a role in modifications to the actin cytoskeleton , and it therefore could function in the phagocytosis of symbionts ( Li and Lee , 2010 ) . Moreover , semaphorin-5a has been shown to promote cell proliferation and to inhibit apoptosis in several cancers ( Sugimoto et al . , 2006; Pan et al . , 2010; Sadanandam et al . , 2010 ) , raising the possibility that it could promote immunotolerance of foreign Symbiodinium cells . Lastly , Ap_Sema5 could function as a PRR . In vertebrates , Semaphorin-7a has previously been shown to serve as an erythrocyte receptor for a Plasmodium TRAP protein ( Bartholdson et al . , 2012 ) , where the sema domain of semaphorin-7a interacts with a TSR domain in the TRAP protein , to promote invasion of host red blood cells by the parasite . Overall , there are a variety of roles that Ap_Sema5 may play to promote the onset of symbiosis , and future functional experiments can be used to test these . Characterization of TSR proteins in cnidarians in this study has revealed a diverse repertoire of genes whose functions remain to be fully described . Functional work provides another piece in the complex web of inter-partner signaling that supports symbiont acquisition and presents the TSR as a protein domain potentially involved in nurturing positive microbial-host interactions in the cnidarian-dinoflagellate symbiosis . Studies using antibodies , proteins , peptides and qPCR to explore TSR protein function in symbiosis suggest that one or more host-derived TSR proteins is participating in host-symbiont communication . Taken together , these studies point to these proteins , potentially working in concert with other secondary proteins , promoting phagocytosis of symbionts and enhancing colonization success . Figure 7 presents a model summarizing the evidence emerging from the immunolocalization and functional experiments . Future studies should target specific TSR homologues for further investigation using antibodies made against specific proteins and ideally using knockdown or gene-editing technologies that would empirically test the impact of these genes on host-symbiont recognition . Overall , there is mounting evidence that Symbiodinium cells can manipulate the host’s immune defenses to gain entry to , and proliferate in cnidarian cells , as occurs in parasitic infections , but how these various strands of evidence ultimately tie together is still unclear and requires further investigation . 10 . 7554/eLife . 24494 . 020Figure 7 . Model summarizing the evidence emerging from immunolocalization and functional experiments . Gastrodermal cell A depicts an aposymbiotic host cell in the process of symbiont acquisition . Results indicate that the addition of soluble TSR proteins promotes and enhances symbiont colonization . We suggest that secreted host TSR proteins may interact with MAMPs and/or secondary proteins to promote tolerance and initiate phagocytosis . Peptide experiments provide evidence against the hypothesis that membrane-linked host TSRs are serving as PRRs to promote inter-partner recognition; we hypothesize that host TSR proteins are secreted rather than membrane-anchored ( see discussion text for further explanation ) . Gastrodermal cell B depicts a symbiotic host cell . Fluorescence microscopy suggests that TSR proteins are expressed within the host-derived symbiosome membrane complex and are concentrated around the symbionts within host gastrodermal tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 020 To characterize the TSR protein repertoire in cnidarians , seven species with publically available resources were searched . These resources were selected to capture a diversity of anthozoans , with representatives from Actinaria , and the complex and robust clades of the scleractinains . Additionally , species were chosen to represent a variety of symbiotic states and symbiont transmission mechanisms . These included three anemone species: A . elegantissima ( Kitchen et al . , 2015 ) , A . pallida ( Lehnert et al . , 2012; Baumgarten et al . , 2015 ) and Nematostella vectensis ( Putnam et al . , 2007 ) , and four coral species: Acropora digitifera ( Shinzato et al . , 2011 ) , Acropora millepora ( Moya et al . , 2012 ) , Fungia scutaria ( Kitchen et al . , 2015 ) and Stylophora pistillata ( Voolstra et al . , submitted ) . These resources were derived from various developmental stages and symbiotic states ( Table 1 ) . All resources were used without manipulation , with the exception of the A . pallida transcriptome , for which raw Illumina sequence reads for accession SRR696721 were downloaded from the sequence read archive ( RRID:SCR_004891 ) entry ( http://www . ncbi . nlm . nih . gov/sra/SRX231866 ) and reassembled using Trinity ( RRID:SCR_013048 , Grabherr et al . , 2011 ) . In addition , the genomes of the symbiotic dinoflagellates Symbiodinium minutum ( ITS2 type B1 ) ( Shoguchi et al . , 2013 ) and S . microadriaticum ( Aranda et al . , 2016 ) were searched for TSR proteins , to investigate the presence of a potential TRAP-like protein . 10 . 7554/eLife . 24494 . 021Table 1 . Anthozoan and Dinoflagellate resourcesDOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 021OrganismFamilyDevelopmental stageSymbiotic stateData typePublicationNematostella vectensisEdwardsiidaeLarvaeNon-symbioticGenomePutnam et al . ( 2007 ) Anthopleura elegantissimaActiniidaeAdultAposymbioticTranscriptomeKitchen et al . , 2015Aiptasia pallidaAiptasiidaeAdultAposymbioticTranscriptomeLehnert et al . ( 2012 ) Aiptasia pallidaAiptasiidaeAdultSymbioticGenomeBaumgarten et al . ( 2015 ) Acropora digitiferaAcroporidaeSpermSymbioticGenomeShinzato et al . ( 2011 ) Acropora milleporaAcroporidaeAdult and LarvaeSymbioticTranscriptomeMoya et al . ( 2012 ) Fungia scutariaFungiidaeLarvaeAposymbioticTranscriptomeKitchen et al . , 2015Stylophora pistillataPocilloporidaeAdultSymbioticGenomeVoolstra et al . , submittedSymbiodinium minutumSymbiodiniaceaeculture ID Mf1 . 05b . 01Dinoflagellate cultureGenomeShoguchi et al . ( 2013 ) Symbiodinium microadriaticumSymbiodiniaceaestrain CCMP2467Dinoflagellate cultureGenomeAranda et al . ( 2016 ) To search for cnidarian TSR proteins , databases were queried using several search strategies to ensure that all sequences were recovered . BLASTp or tBLASTn searches with the second TSR domain from mouse and human thrombospondin-1 protein sequences , and the consensus sequence ( smart00209: TSP1 ) from the conserved domain database ( RRID:SCR_002077 , http://www . ncbi . nlm . nih . gov/cdd ) as queries were performed for each resource . Keyword searches using the terms TSP1 , thrombospondin , ADAMTS , ADAM and SEMA were also performed where genome browsers allowed keyword searches of GO , KEGG and PFAM annotations . Lastly , representative N . vectensis sequences of each protein type ( ADAMTS-like , SEMA , TRYPSIN and TSR-only ) were also used as queries for tBLASTn searches of the other six cnidarian resources . A high e-value cutoff ( 1 × 10−1 ) was used in the BLAST searches to recover divergent sequences . All BLAST searches were performed using Geneious pro version 7 . 1 . 8 ( RRID:SCR_010519 , Kearse et al . , 2012 ) with the exception of the N . vectensis , A . pallida and S . pistillata genomes , for which searches were performed through the Joint Genome Institute online portal ( RRID:SCR_002383 ) , NCBI ( RRID:SCR_004870 ) and the Reefgenomics online repository ( RRID: SCR_015009 , http://reefgenomics . org ) ( Liew et al . , 2016 ) , respectively . A list of metazoan resources searched is provided in Table 1 . Sequences identified are tabulated in Supplementary file 1 . To confirm that the sequences obtained contained TSR domains , nucleotide sequences were translated using Geneious or ExPASy translate tool ( RRID:SCR_012880 , http://web . expasy . org/translate/ ) and then annotated using the Geneious InterProScan plugin ( RRID:SCR_010519 , Kearse et al . , 2012 ) . All annotations were double checked using the online protein domain database PfamA ( RRID:SCR_004726 , http://pfam . sanger . ac . uk ) , and only sequences that showed significant PfamA matches to a TSR domain with an e-value of <1×10−4 were used . Sequences for each species were aligned and those that were identical or almost identical ( <5 aa difference in the conserved domains ) were omitted from the analysis , as they likely represented artefacts of assembly or different isoforms of the same protein . Sequences missing a start or stop codon were removed from the analysis . Diagrammatic representations of the protein domain configurations were produced using this information . Protein domain architectures were grouped together according to common domains and compared to known human TSR proteins ( Figure 2 ) . A population ( not necessarily clonal ) of Symbiodinium minutum ( clade B1 ) -containing A . pallida , originating from a local pet store , was maintained in saltwater aquaria at 26°C at a light intensity of approximately 40 µmol quanta m−2 s−1 with a 12/12 hr light/dark photoperiod , and fed twice weekly with live brine shrimp nauplii . Animals were rendered aposymbiotic by incubation for 8 hr at 4°C twice weekly for six weeks , followed by maintenance in the dark for approximately one month . Anemones were fed twice weekly with brine shrimp , and cleaned of expelled symbionts and food debris regularly . Cultured dinoflagellates - Symbiodinium minutum ( sub-clade B1; culture ID CCMP830 from Bigelow National Center for Marine Algae and Microbiota ) - were maintained in 50 ml flasks in sterile Guillard’s f/2 enriched seawater culture medium ( Sigma , St . Louis , MO , USA ) . Dinoflagellate cultures were maintained at 26°C and 70 µmol quanta m−2 s−1 with a 12/12 hr light/dark photoperiod . CCMP830 cultures were typed using Internal transcribed spacer 2 ( ITS2 ) sequencing in 2009 and 2016 to authenticate the identity of the culture . The CCMP830 cultures were not axenic and therefore Mycoplasma contamination testing was not performed . In preparation for experimental manipulation , individual anemones were placed in 24-well plates in 2 . 5 ml of 1 µm-filtered seawater ( FSW ) and acclimated for 3–4 days , with the FSW replaced daily . Well-plates containing aposymbiotic anemones were kept at 26°C in the dark , while those containing symbiotic anemones were maintained in an incubator at a light intensity of approximately 40 µmol quanta m−2 s−1 with a 12/12 hr light/dark photoperiod . Animals were not fed during the acclimation or experimental periods . Immunoblots were performed on A . pallida proteins using an anti-human thrombospondin rabbit polyclonal antibody . The thrombospondin antibody was made against an epitope corresponding to the three TSR domains of human thrombospondin proteins 1 and 2 ( Santa Cruz Biotechnology Cat# sc-14013 RRID:AB_2201952 ) . The epitope showed sequence similarity to a TSR protein identified in A . pallida ( Figure 8A ) . Groups of eight aposymbiotic or symbiotic anemones were homogenized on ice in 1 ml homogenization buffer ( 50 mM Tris–HCl , pH 7 . 4 , 300 mM NaCl , 5 mM EDTA ) with a protease inhibitor cocktail ( BD Biosciences , San Jose , CA , USA ) . Homogenates were centrifuged at 4°C for 15 min at 14 , 000 x g to pellet cell debris , supernatants were decanted and protein concentrations were determined using the Bradford assay . Protein was adjusted or diluted in RIPA buffer to a standard concentration of 50 µg total protein per well and boiled for 5 min in loading dye . Proteins were resolved on a 7% SDS–PAGE gel and then electrophoretically transferred overnight onto nitrocellulose membrane . After blocking with 5% non-fat dry milk in TBS-Tween 20 ( 0 . 1% ) for 1 hr at 37°C , membranes were incubated with anti-thrombospondin or an IgG isotype control , both at a dilution of 1:200 , for 2 hr at room temperature . The blots were washed three times in TBS-Tween 20 followed by incubation in a HRP-conjugate goat anti-rabbit IgG Alexa Fluor 546 secondary antibody ( Molecular Probes Cat# A-11030 RRID:AB_144695 ) at a 1:5000 dilution ( 0 . 2 µg ml−1; Sigma , St . Louis , MO , USA ) for 1 hr . Bands were detected by enhanced chemiluminescence ( Millipore , Temecula , CA , USA ) . Blots were stripped and re-probed with an actin loading control ( Santa Cruz Biotechnology Cat# Sc-1616 RRID:AB_630836 ) , see Figure 4—figure supplement 1 for actin control . 10 . 7554/eLife . 24494 . 022Figure 8 . Sequence information for thrombospondin antibody and TSR peptide fragments used in this study . ( A ) Alignment of the second TSR domains from human thrombospondin 1 and TSR proteins from the anemone Aiptasia pallida and the dinoflagellate Symbiodinium minutum . In red are the binding sites for glycosaminoglycans ( GAGs ) and CD36; greyscale indicates the % identity of the three sequences . Pink annotation indicates the TSR peptide sequence covering all three binding domains; inset are the synthetic peptide sequences for experimental peptides . In Peptide 2 , the cysteine residues were replaced with alanine residues , as shown in red . ( B ) A section of the antibody-binding region of the human thrombospondin 1/2 antibody ( H-300 , sc-14013 from Santa Cruz Biotechnology ) , aligned to a TSR protein fragment from Aiptasia sp . Legends for Supplementary Material . DOI: http://dx . doi . org/10 . 7554/eLife . 24494 . 022 Immunofluorescence was used to investigate the presence of TSR proteins on the surface of dinoflagellate cells . We compared anti-human TSR binding in cultured S . minutum strain CCMP830 to S . minutum cells freshly isolated from A . pallida . To obtain freshly-isolated symbiont cells with intact symbiosome membranes , anemones were homogenized in a microfuge tube with a micro-pestle and the resulting homogenate was centrifuged at a low speed ( <1000 rpm ) for 5 min to produce an algal pellet . The pellet was washed several times in FSW and re-pelleted . Algal cells were re-suspended to a final concentration of 2 . 5 × 104 cells per ml . The lipophilic membrane stain , Dil ( 1 , 1'-dioctadecyl-3 , 3 , 3' , 3'-tetramethylindocarbocyanine perchlorate , DilC18 ( 3 ) ; Molecular Probes ) , was used to test for the presence of putative symbiosome membrane surrounding freshly isolated symbiont cells and cells taken from culture . Dil was added to cells in 500 µl of FSW in a microfuge tube and gently mixed shortly before small amounts of suspended cells were placed on a well slide and imaged . Both cultured and freshly isolated S . minutum cells were incubated in the anti-human TSR conjugated to the secondary antibody Alexa Fluor 546 goat anti rabbit IgG fluorescent probe ( Molecular Probes Cat# A-11030 RRID:AB_144695 ) at a 1:1000 dilution . Anti-thrombospondin and Dil labeling in cells was imaged using a Zeiss LSM 510 Meta microscope through a Plan- APOCHROMAT 63x/1 . 4 Oil DIC objective lens . See Supplementary file 2 for a description of fluorescent dyes and the specific excitation and emission wavelengths . To localize TSR proteins in symbiotic and aposymbiotic anemone tissues , cryosections of anemone tentacles were made using methods modified from Dunn et al . ( 2007 ) . The sections were washed twice in PBS and fixed with 4% PFA for 10 min , and then washed twice in PBS . Sections were then permeabilized with 0 . 2% Triton-X-100 in PBS for 5 min and blocked in 3% BSA , 0 . 2% Triton-X-100 in PBS for 30 min , before incubation in the anti-human TSR rabbit polyclonal antibody ( described above ) at a 1:200 dilution ( in blocking buffer ) for 4 hr at 4°C . Slides were subsequently washed three times for 5 min each with 0 . 2% Triton-X-100 in PBS at rt . Alexa Fluor 546 ( Molecular Probes Cat# A-11030 RRID:AB_144695 ) secondary antibody was diluted in blocking buffer ( 1:150 dilution ) and added to the slides for 1 hr in the dark at rt . Slides were washed three times in the dark for 5 min with 0 . 2% PBS/Triton-X-100 . A drop of Vectashield DAPI hard set mounting medium was then used to stain nuclei and mount cover slips onto slides . Immunofluorescence was visualized using a Zeiss LSM 510 Meta microscope through a Plan-APOCHROMAT 63x/1 . 4 Oil DIC objective lens . The fluorescence excitation/emission was 556/573 nm for Alexa Fluor 546 and 543/600–700 nm for Symbiodinium chlorophyll autofluorescence ( see Supplementary file 2 ) . In preparation for experimental manipulation , individual anemones were placed in 24-well plates in 2 . 5 ml FSW and acclimated for 4 days , with FSW replaced daily . During this time , aposymbiotic anemones were maintained in darkness , and symbiotic anemones were maintained in an incubator at 26°C under the light regime described above . Animals were not fed during the experimental period . Aposymbiotic anemones were experimentally inoculated with S . minutum cells and colonization success was determined by quantifying the number of symbionts present in host tissues ( see below ) . Experimental treatments were initiated 2 hr prior to colonization with S . minutum . For inoculation , cultured S . minutum cells were added to each well to a final concentration of 2 × 105 cells ml−1 . After incubation with dinoflagellate cells for 4 hr , anemones were washed twice in FSW and experimental treatments were refreshed . Well-plates were then placed back into an incubator at 26°C under the light regime described above . Colonization success was assessed fluorometrically with a Zeiss LSM 510 Meta confocal microscope , following the methods detailed elsewhere ( Detournay et al . , 2012; Neubauer et al . , 2016 ) . Colonization success was expressed as the percent of pixels with an autofluorescence intensity above the background intensity . Each experimental treatment had a sample size of three anemones per treatment and time-point , with percent colonization taken as a mean of three to four tentacles per anemone . Three untreated symbiotic anemones ( three to four tentacles per anemone ) were examined to determine a baseline colonization level for symbiotic anemones . The sample size was limited by both the supply of anemones as well as the number of anemones that could be processed for confocal microscopy at each time point . The statistical significance of colonization success under the treatments described above was assessed using a mixed-effects analysis-of-variance model . As measures on multiple samples ( i . e . , tentacles ) per anemone violate independence assumptions , we treated ‘anemone’ as a random effect to account for correlation among samples within anemones . Main effects included time ( in hours ) and treatment , and their interaction was estimated to account for differences between treatments at each time point . The full model can be written as:yi , j=βXi+μj+ϵi , j Here , yi , j is the logarithm of percent colonization of tentacle i within anemone j , β is a vector of effects to be estimated , X is a design matrix encoding the treatment and time point , as well as interaction term contrasts , μj is a normally distributed random effect for anemone j , and ϵi , j are normally distributed residuals . Contrasts were specified between each treatment and controls at each time-point to assess statistical significance of treatment effects over time , using Tukey’s post-hoc test to account for multiple comparisons . The model was estimated using the NLME package ( Pinheiro et al . , 2016 ) for the statistical computing software R ( R-Core-Team , 2012 ) ( RRID:SCR_001905 , www . R-project . org ) . All datasets and code to reproduce statistical analyses and figures are given as supplementary materials ( Figure 5—source data 1–6 , and Supplementary Source code 1 ) . To investigate the specific TSR proteins that are involved in the onset of symbiosis , gene expression of two sequences obtained from the bioinformatics searches of the A . pallida genome , Ap_Sema5 and Ap_Trypsin-like was measured using quantitative PCR ( qPCR ) . First , to confirm the genome assembly , primers for each sequence were designed using Primer3plus ( RRID:SCR_003081 , http://primer3plus . com/cgi-bin/dev/primer3plus . cgi ) to amplify overlapping 700–900 bp fragments ( Supplementary file 3 ) . PCR for each primer set was performed using the Go Taq Flexi kit ( Promega , Madison , WI ) with the following protocol: 94°C for 3 min , 35 cycles of 94°C for 45 s , annealing temperature for 45 s , and 72°C for 1 min , followed by a final extension at 72°C for 10 min . PCR products were cleaned using the QiaQuick PCR purification kit ( Qiagen , Valencia CA ) and sequenced on the ABI 3730 capillary sequence machine in the Center for Genome Research and Biocomputing ( CGRB ) at Oregon State University . Sequences obtained were aligned to the original genome sequence using Geneious v 7 . 1 . 8 ( RRID:SCR_010519 , Kearse et al . , 2012 ) to verify amplification of the correct sequence and ensure that overlapping regions between fragments displayed high similarity . If a region varied greatly from the genome , the region was re-sequenced for confirmation before moving forward . Ap_Trypsin-like contained a region that was different than AIPGENE 1852 , and therefore this sequence has been submitted to GenBank ( accession # KY807678 ) . qPCR primers for products between 100–200 bp with an annealing temperature of 60°C were designed using Primer3 Plus ( Supplementary file 4 ) , and the products were amplified and sequenced as previously described to confirm the correct amplicon . The efficiency of each primer set was tested to ensure that it was at least 90% . To investigate the expression of Ap_Sema5 and Ap_Trypsin-like at the onset of symbiosis , qPCR was performed on samples from a previous experiment in which aposymbiotic specimens of A . pallida were inoculated with S . minutum strain CCMP830 ( Poole et al . , 2016 ) . The two treatment groups used in this study were aposymbiotic animals that were inoculated with symbionts ( ‘inoc’ ) and aposymbiotic animals that received no symbionts and remained aposymbiotic for the duration of the experiment ( ‘apo’ ) . The animals used in this study were sampled at 12 , 24 , and 72 hr post-inoculation ( n = 3 for each time point and treatment combination ) Symbiont quantification data indicated symbionts were taken up by 24 hr post-inoculation and levels continued to increase between 24 and 72 hr ( Poole et al . , 2016 ) . The anemones were washed at 24 hr and therefore the increase between 24 and 72 hr can be attributed to symbiont proliferation within the host . Therefore , the time points selected represent a period in which symbionts were actively engaging in recognition and phagocytosis by host cells ( 12 and 24 hr ) and as symbionts were proliferating within the host ( 72 hr ) . qPCR plates were run as previously described ( Poole et al . , 2016 ) using the ABI PRISM 7500 FAST , and resulting Ct values were exported from the machine . Triplicates were averaged and the expression of target genes was normalized to the geometric mean of the reference genes ( L10 , L12 , and PABP ) . To calculate the ΔΔ Ct , the normalized value for each sample was subtracted from the average normalized value of a reference sample , the apo at each time-point . The resulting relative quantities on the log2 scale were used for statistical analysis using R version 3 . 2 . 1 ( RRID:SCR_001905 , R Core Team , 2015 ) . Identical linear models were used to test the hypothesis of no significant difference in gene expression between ‘apo’ and ‘inoc’ anemones for both genes . The model was identical to the statistical model described above , but did not include a random effect . A two-way ANOVA was run to test for statistical significance of treatment effects , followed by Tukey’s post hoc test for pairwise comparisons . All datasets and code to reproduce statistical analyses are given as supplementary materials ( Figure 6—source data 1 and Supplementary Source code 1 ) .	Cnidarians , such as corals and sea anemones , often form a close relationship with microscopic algae that live inside their cells – a partnership , on which the entire coral reef ecosystem depends . These microalgae produce sugars and other compounds that the cnidarians need to survive , while the cnidarians protect the microalgae from the environment and provide the raw materials they need to harness energy from sunlight . However , very little is known about how the two partners are able to communicate with each other to form this close relationship , which is referred to as a symbiosis . Symbiotic relationships between a host and a microbe require a number of adaptations on both sides , and involve numerous signalling molecules . A host species is under constant pressure to develop mechanisms to recognize and tolerate the beneficial microbes without leaving itself vulnerable to attack by microbes that might cause disease . Similarly , the beneficial microbes need to be able to invade and survive inside their host . Previous research has shown that TSR proteins in hosts play a role in recognizing and controlling disease-causing microbes . Until now , however , it was unknown whether TSR proteins are involved in establishing a symbiosis between cnidarians and their algal partners . Neubauer et al . analysed six species of symbiotic cnidarians and discovered a diverse repertoire of TSR proteins . These proteins were found in the host genomes , rather than in the symbiotic algae , strongly suggesting that they originated from the host . Neubauer et al . next incubated a sea anemone species in a solution of TSR proteins and saw that it became ‘super-colonized’ with algae , meaning that over time , millions of the microalgae entered and stayed in the anemone’s tentacles . In contrast , when the TSR proteins were blocked , colonization was almost entirely stopped . This suggests that host TSR proteins play an important role for the microalgae when they colonialize corals and other cnidarians . The signals that enable microalgae to successfully colonialize cnidarians are unquestionably complex and there is still much to learn . These findings add another piece to the puzzle of how symbiotic algae bypass the cnidarian’s immune system to persist and flourish in their host . An important next step will be to test how blocking the genes that encode the TSR proteins will affect the symbiotic relationship between these species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "microbiology", "and", "infectious", "disease" ]	2017	A diverse host thrombospondin-type-1 repeat protein repertoire promotes symbiont colonization during establishment of cnidarian-dinoflagellate symbiosis
Convergent evidence suggests that the basal ganglia support reinforcement learning by adjusting action values according to reward prediction errors . However , adaptive behavior in stochastic environments requires the consideration of uncertainty to dynamically adjust the learning rate . We consider how cholinergic tonically active interneurons ( TANs ) may endow the striatum with such a mechanism in computational models spanning three Marr's levels of analysis . In the neural model , TANs modulate the excitability of spiny neurons , their population response to reinforcement , and hence the effective learning rate . Long TAN pauses facilitated robustness to spurious outcomes by increasing divergence in synaptic weights between neurons coding for alternative action values , whereas short TAN pauses facilitated stochastic behavior but increased responsiveness to change-points in outcome contingencies . A feedback control system allowed TAN pauses to be dynamically modulated by uncertainty across the spiny neuron population , allowing the system to self-tune and optimize performance across stochastic environments . When tasked with taking an action in an unknown environment , there can be considerable uncertainty about which actions will lead to the best outcomes . A principled way to resolve this uncertainty is to use previous experience to guide behavior towards actions that have led to positive outcomes in the past and away from actions that have led to negative outcomes . Convergent evidence suggests that the basal ganglia can guide behavior by incorporating positive and negative feedback in a reinforcement learning process ( O'Doherty et al . , 2003; Barnes et al . , 2005; Frank , 2005 ) . However , learning can be complicated in a changing environment , as the validity of past experiences and the relationship between actions and outcomes become uncertain as well . Mathematical models suggest that it is optimal to take uncertainty into account in learning and decision making ( Yu and Dayan , 2005; Behrens et al . , 2007; Mathys et al . , 2011 ) , but it is unclear whether the basal ganglia can directly consider uncertainty in feedback-based learning . Basal ganglia-dependent learning is often described within the normative framework of reinforcement learning ( RL ) following the observation that signaling from dopaminergic afferents matches the pattern of a reward prediction error ( RPE ) ( Montague et al . , 1996; Bayer et al . , 2007 ) . An RPE is the signed difference between the observed and expected outcomes and is often used in RL to generate point-estimates of action-values ( Sutton and Barto , 1998 ) . Phasic dopamine ( DA ) is thought to provide an RPE signal to striatal medium spiny neurons ( MSNs ) and induce learning through changes in corticostriatal plasticity ( Montague et al . , 1996; Reynolds and Wickens , 2002; Calabresi et al . , 2007 ) , with opponent learning signals in the direct and indirect pathways ( Frank , 2005; Collins and Frank , 2014 ) . Within these pathways , separate populations code for the ( positive and negative ) values of distinct action plans ( Samejima et al . , 2005 ) . Multiple lines of evidence in humans and animals support this model , including optogenetic manipulations ( Tsai et al . , 2009; Kravitz et al . , 2012 ) , synaptic plasticity studies ( Shen et al . , 2008 ) , functional imaging ( McClure et al . , 2003; O'Doherty et al . , 2003 ) , genetics and pharmacology in combination with imaging ( Pessiglione et al . , 2006; Frank et al . , 2009; Jocham et al . , 2011 ) and evidence from medication manipulations in Parkinson’s patients ( Frank et al . , 2004 ) . Despite the substantial empirical support for RPE signals conveyed by dopamine , the simple RL mechanisms used to model the basal ganglia are inflexible in the degree to which they learn in a changing environment . RL models typically adopt a fixed learning rate , such that every RPE of similar magnitude equally drives learning . However , a more adaptive strategy in a changing environment is to adjust learning rates as a function of uncertainty , so that unexpected outcomes have greater influence when one is more uncertain of which action to take ( e . g . , initially before contingencies are well known , or following a change-point ) , but less influence once the contingencies appear stable and the task is well known ( Yu and Dayan , 2005; Behrens et al . , 2007; Nassar et al . , 2010; Mathys et al . , 2011; Payzan-LeNestour et al . , 2011 ) . This Bayesian perspective presents the additional challenge for basal ganglia-dependent learning: in order to take advantage of its own uncertainty over action selection , the basal ganglia would need a mechanism to translate its uncertainty into a learning rate . Cholinergic signaling within the striatum offers a potential solution to this challenge . With few exceptions ( Tan and Bullock , 2008; Ashby and Crossley , 2011 ) , models of the basal ganglia typically do not incorporate striatal acetylcholine . Within the striatum , cholinergic interneurons are the predominant source of acetylcholine ( Woolf and Butcher , 1981 ) . These interneurons , also known as tonically active neurons ( TANs ) due to their intrinsic 2–10-Hz firing pattern , are giant , spanning large areas of the striatum with dense axonal arborization and broad synaptic input ( Goldberg and Reynolds , 2011 ) . TANs appear to be necessary to learning only when flexibility is required ( Ragozzino et al . , 2009; Bradfield et al . , 2013 ) , suggesting that they might modulate the learning rate as a function of changes in outcome statistics ( i . e . , uncertainty ) . Similar to dopaminergic neurons , TANs show sensitivity to rewarding events and develop a learned phasic response to predictive cues ( Aosaki et al . , 1994; Morris et al . , 2004 ) . This response consists of a phasic burst in TAN activity followed by a pause that lasts a few hundred milliseconds ( Aosaki et al . , 1995 ) . While the temporal pattern of the burst–pause response is temporally concomitant to the dopamine response ( Morris et al . , 2004 ) , the unidirectional TAN response is not consistent with a bivalent RPE ( Joshua et al . , 2008 ) but instead is thought to provide a permissive signal for dopaminergic plasticity ( Graybiel et al . , 1994; Morris et al . , 2004; Cragg , 2006 ) . But how would such a permissive signal be modulated by the network’s own uncertainty about which action to select ? Because TANs receive broad inhibitory synaptic input from local sources , including MSNs and GABAergic interneurons ( Bolam et al . , 1986; Chuhma et al . , 2011 ) , we hypothesized that the pause would be modulated by a global measure of uncertainty across the population of spiny neurons . Given that MSN sub-populations code for distinct action values ( Samejima et al . , 2005; Frank , 2005; Lau and Glimcher , 2008 ) , co-activation of multiple populations can signal enhanced uncertainty over action selection , which would translate into greater inhibition onto TANs . The synchrony in the TAN response suggests a global signal ( Graybiel et al . , 1994 ) , which can then be modulated by inhibitory MSN collaterals across a large range of spiny inputs . The TAN pause response is consistent with a signal of uncertainty that adjusts learning . First , it increases with the unpredictability of a stochastic outcome ( Apicella et al . , 2009 , 2011 ) . Second , pharmacological blockade or lesioning excitatory input to TANs impairs learning , specifically after a change in outcome contingencies ( Ragozzino et al . , 2002; Bradfield et al . , 2013 ) . For an optimal learner , both increases in stochasticity and changes in outcome contingencies results in an increase in uncertainty ( Yu and Dayan , 2005; Nassar et al . , 2010 ) . Here , we augment a well-established computational model of the basal ganglia ( BG ) to include a mechanism by which the effective learning rate is modulated by cholinergic signaling , and where this signaling is , in turn , modulated by uncertainty in the MSN population code via reciprocal TAN–MSN interactions . In the model , cholinergic signals dynamically modulate the efficacy of reinforcement feedback , driving changes in the number of neurons available for synaptic plasticity during reinforcement , hence the effective learning rate of the network as a whole . Thus , TANs allow the basal ganglia to tailor its learning rate to its environment , balancing the tradeoff between learning flexibility and robustness to noise by adjusting learning as a function of policy uncertainty embedded in the MSN population code . We show that this behavior is consistent with a normative account using an approximately Bayesian model and that its main functionality can be simplified in algorithmic form using a modified RL model , thus spanning Marr’s levels of implementation , algorithm , and computation . The model is consistent with several existing lines of evidence and makes falsifiable predictions . A principled way to balance this trade-off is to use the learner’s uncertainty over its action values to anneal the learning rate , similar to the gain in a Kalman filter ( Yu and Dayan , 2005 ) . For an ideal observer , the proper measure of this uncertainty depends on the parameterization of the generative processes of the task , which may require several levels of hierarchical representations ( Behrens et al . , 2007; Mathys et al . , 2011 ) that may be unavailable in a local striatal network . However , as noted above , entropy over multiple MSN populations can be used as a proxy as it can be interpreted as the uncertainty with which the network selects an action . Although we have thus far described this entropy as an observable variable to provide an explanation for how TAN pauses affect the trade-off between flexibility and stability because it can be read out from the distribution of MSN firing rates , we hypothesized that this same measure may be used to reciprocally influence TAN pauses and hence learning . In this way , the system could self-detect its own uncertainty in a way that ( unlike any given fixed pause duration ) is not dependent on the parameterization of a particular task and is implicitly available in the population response . Given their size and broad connectivity across large areas of striatum spanning many MSNs , it is plausible that TANS are sensitive to population MSN entropy or a correlate thereof by summing activity over multiple neurons coding for different actions . As noted earlier , TANs receive substantial GABAergic innervation from medium spiny neurons ( Bolam et al . , 1986; Chuhma et al . , 2011; Gonzales et al . , 2013 ) . While the initiation of a TAN pause is dependent on thalamic and dopaminergic signaling ( Aosaki et al . , 1994; Ding et al . , 2010 ) , we propose that local signaling from MSNs can modulate the response by elongating the duration of the pause ( or simply further reducing the firing rate during the pause ) through direct inhibition . If TANs are sensitive to MSN entropy and can modulate their pause durations accordingly , they may provide a mechanism to adjust to changes in uncertainty . To assess this potentially adaptive mechanism , we first conducted simulations in which the TAN pause was dynamically modulated by an analytical computation of MSN entropy ( see Materials and methods ) ; below , we will consider a more mechanistic implementation by which TANs respond to local circuits that approximate entropy . These simulations confirmed that dynamic modulations of TAN pause as a function of MSN mitigated the trade-off between flexible learning and sensitivity to noise . When trained on an 85% reward schedule , the network with a variable duration pause learned relatively quickly following reversal while still achieving a high level of asymptotic accuracy ( Figure 6 , left ) . When trained on a 40% reward schedule , performance was comparable to the best network trained with a long fixed duration pause ( Figure 6 , center ) . While in both the cases it was possible to find a particular fixed duration pause that can perform as well as a variable pause duration , networks with entropy-modulated pause durations performed better across both reward schedules than any fixed pause network [Accuracy ~ N ( μ¯ , Iσ¯ ) ; E[∥μvariable∥1 - max ∥μfixed∥1] = 5 . 76%; p ( ∥μ¯variable∥1 ≤ max ∥μ¯fixed∥1 ) = 1 . 03x10-3] . Thus , while a fixed TAN pause may perform well in any one environment , any given fixed setting is suboptimal when the environment is unknown . Varying TAN behavior with MSN entropy allows the network to learn robustly over a wider range of environment statistics , thereby allowing learning rates to be sensitive to the network’s own uncertainty . This behavioral flexibility occurs as the feedback mechanism induces longer pauses during periods of higher uncertainty . In the 85% reward environment , the feedback mechanism results in dynamic behavior: TAN pauses last greater than 250 ms both at the beginning of the task and following reversal , while pauses reach 200 ms immediately prior to reversal and at the end of the training session when less learning is required ( Figure 5 , bottom left ) . As such , the long TAN pause early in acquisition results in increased synaptic efficacy for Go units associated with the dominant response relative to the suboptimal response ( Figure 5 , bottom right ) . As the entropy and the pause duration decline , the rate of change in synaptic efficacy slows , resulting in faster reversal learning as the network can unlearn the previous reward association more quickly . In contrast , uncertainty remains elevated in the in the highly stochastic 40% reward environment and the TAN pause consequently remains elevated throughout training ( Figure 5 , bottom left ) . As such , the feedback mechanism results in changes in synaptic efficacy that are similar to changes induced by a long fixed duration TAN pause . Before further analyzing biophysical mechanisms within the model—including a mechanism for computation of entropy , the effects of M1 receptor manipulations , and additional roles of the post–pause burst on DA release—we first develop higher level ( computational and algorithmic ) models that summarize the key trade-off identified above in functional terms ( Marr and Poggio , 1976 ) . While the neural network model makes empirical predictions at the biophysical level , the core computational and algorithmic problems solved by the network can be imbedded in simpler formulations . To describe the computational and algorithmic problems solved by the network , we compared the network’s behavior with two models: ( 1 ) an approximately Bayesian model that considers the higher-level computational problem of learning under uncertainty , and ( 2 ) an algorithmic model that more closely matches learning in the basal ganglia mechanistically , modified from Collins and Frank ( 2014 ) to include the role of TANs . The core computational problem solved by the addition of TANs in the network addresses how to integrate noisy experiences in a changing environment . Behrens et al . ( 2007 ) noted that in a changing environment , a hierarchical representation of volatility can be used to adjust the learning rate in an optimal way . If rewards are distributed probabilistically with rate r and changes from trial to trial , an optimal agent can estimate the volatility of r as well as their distrust in the trial-to-trial volatility of the reward rate . It is unlikely that this hierarchical inference is implemented in the basal ganglia; thus , we consider here an approximation by which this computation need not be performed explicitly but where the striatum has access to its own uncertainty and adjust its learning rate accordingly . In the tasks simulated , rewards were of equal value and delivered stochastically . In a stationary task with binomial outcomes , the posterior distribution of the reward rate for an optimal learner is a Beta distribution ( Daw et al . , 2005 ) parameterized by the counts of rewarded and non-rewarded trials . Because reversal tasks are , by definition , non-stationary , the Beta distribution does not represent the exact posterior distribution of expected values and is slow to adjust to a reversal: it becomes too certain about reward contingencies . A Bayesian treatment allowing for the possibility that the outcome statistics can change can be approximated using a mixture distribution combining the posterior of expected values with a uniform distribution ( Nassar et al . , 2010 ) . In this approximation , the mixture component is the probability a change has occurred to some other unknown contingencies . Instead , we use a multiplicative parameter γ to decay the counts of the Beta hyper-parameters towards the prior after each trial , an approximation that has been used previously to model rodent and human behavior ( Daw et al . , 2005; Doll et al . , 2009 ) . Decaying the counts multiplicatively maintains the mode expected value for each action but increases the variance ( uncertainty ) of the distribution . This effectively reduces the model’s confidence of the expected value without changing its best estimate . Hence , a faster decay rate allows for greater effective learning rate , analogous to the mechanism in the BG model , whereby MSN weights are prevented from overlearning . We simulated the Bayesian model with both fixed values of γ as well as a γ that varied as a function of entropy in action selection . Models with fixed values of γ are analogous to neural networks with a fixed duration TAN pause , both constituting static strategies to balance asymptotic accuracy and learning speed . Varying γ as a function of action selection entropy is similar to the strategy employed by the neural network , as MSN entropy corresponds to the decision uncertainty of the network . Over both manipulations , the Bayesian model shows the same qualitative pattern of behavior as the neural network ( Figure 7 , right ) . For fixed values of γ , we found the same tradeoff between asymptotic accuracy and learning speed following reversal . Likewise , varying the decay rate with uncertainty mitigated the effects of the trade-off , facilitating faster learning following reversal than possible with a slow , fixed decay rate without the cost of asymptotic accuracy associated with high , fixed decay ( Figure 7 , right , black line ) . Given that it is an idealized statistical model and not implementational , the Bayesian learner responds with an overall higher level of accuracy than the neural network . Unlike the Bayesian learner , the neural network does not learn an estimate of each action value directly but instead learns relative preferences for actions . Nevertheless , overall , the Bayesian learner provides a normative description of the neural network as the effect of manipulating the decay parameter γ is qualitatively similar to what we see when manipulating TAN duration , suggesting that TAN duration affects the uncertainty of representations in MSNs . 10 . 7554/eLife . 12029 . 009Figure 7 . Comparison of behavior between neural network and Bayesian learner collapsed across multiple reward schedules . The performance of the Bayesian learner ( i ) is qualitatively similar to the performance of the neural network model ( left ) . A slow decay rate in the Bayesian learner ( right , blue ) has the same effect as a long TAN pause ( left , blue ) and results in higher asymptotic accuracy at a cost of slower learning following reversal . A fast decay rate ( right , red ) has the same effect as a short TAN pause ( left , red ) and results in fasters learning following reversal with lower asymptotic accuracy . Varying the decay rate and pause duration with entropy in the Bayesian learner and neural network , respectively , mitigates the trade-off . Reversal is denoted with dotted line in both panels . DOI: http://dx . doi . org/10 . 7554/eLife . 12029 . 009 The OpAL model proposed by Collins and Frank ( 2014 ) provides a more algorithmic summary of the basal ganglia network , as an expansion of an actor–critic model to include opponent ( D1/D2 ) actor values . The OpAL model provides a normative advantage over traditional RL while quantitatively capturing a variety of data across species implicating opponent processes in both learning and action selection , where dopamine manipulations affect the asymmetry with which humans and animals make decisions , in a model with few free parameters . In the OpAL model , reward prediction errors are computed within a critic that evaluates the expected value of each state , and are used to update these values and train the G and N actors , summarizing the population activity of Go and NoGo units with point values ( see Materials and methods ) . To simulate the effect of TANs in OpAL , the G and N weights were decayed multiplicatively after each trial by either a constant rate ( to simulate the effects of a fixed-duration TAN pause ) , or as a function of G and N entropy ( to simulate the effects of the proposed feedback mechanism ) . A fast rate of decay ( closer to zero ) simulated a short TAN pause whereas a slow rate of decay ( closer to one ) simulated a long TAN pause . Simulations over a range of fixed rates of decay shows a similar pattern as in the neural network and Bayesian learner , as slow rates of decay showed high asymptotic accuracy but with a decreased learning speed following reversal in a deterministic environment ( Figure 8 , left ) . Likewise , a faster rate of decay showed a similar pattern to networks with short TAN pause durations and showed increased flexibility after reversal at a cost of asymptotic accuracy . As in the neural network , the trade-off between asymptotic accuracy and speeded learning after reversal was marked by a divergence in the G weights ( Figure 8 , right ) . For models with a slow decay , the difference between G weights for the initially rewarded response and the initially sub-optimal response was much more pronounced than models with a faster decay . This divergence declined more rapidly in models with a slow decay , allowing these models to unlearn the previously rewarded response more quickly . 10 . 7554/eLife . 12029 . 010Figure 8 . Peformance of OpAL in an 85% reward environment . Left: Fast decay in synaptic weights ( red lines ) result in lower asymptotic accuracy but speeded learning following reversal while slow weight decay ( blue lines ) result in the opposite pattern . A model that varies the decay rate with policy uncertainty ( black line ) mitigates this trade-off . Right: Slow decay rates result in a large divergence between G weights for the two possible actions prior to reversal ( trial 200 ) that correlates with high asymptotic accuracy and slower learning speed following reversal , as this divergence must be unlearned . An entropy modulated decay rate shows a high initial rate of divergence sufficient to improve accuracy but slows as the model learns the task , facilitating reversal . DOI: http://dx . doi . org/10 . 7554/eLife . 12029 . 010 Allowing the decay rate to vary as a function of uncertainty also showed the same pattern of behavior found in the neural network , mitigating the trade-off and resulting in both high asymptotic accuracy as well as speeded learning following reversal ( Figure 8 , left ) . Overall , varying decay with entropy resulted in higher reward rate compared with the next best performing model with fixed decay [t ( 198 ) =3 . 7 , p<3x10–4 , Cohen’s d = 0 . 5] . Varying decay with entropy also optimized the divergence of G weights relative to fixed decay . The initial rate of divergence of the variable model is similar to that of a slow decay model , improving asymptotic accuracy . This rate of divergence declines as the model learns , preventing overlearning and facilitating faster learning following reversal . This is the same pattern observed in the divergence of Go weights in the neural model ( Figure 5 , bottom right ) , where varying pause duration with entropy optimized the divergence of weights relative to fixed duration pauses . In the neural network simulations we proposed that TANs may have access to entropy over MSNs , given that they span large regions of the striatum and receive inputs from many MSNs and GABAergic interneurons . However , the biophysical details by which spiny neuron modulate TAN activity are not fully understood . Direct transmission of entropy may not be trivial as population entropy is a nonlinear function of the units’ activity . Here we consider a more explicit mechanism by which TANs directly approximate MSN entropy through synaptic integration . First , consider the minimalist case of just two MSNs coding for alternative actions . Here , the entropy is high when both are active or inactive , but low when either of the two units alone is active . In terms of a Boolean function , the entropy of the two MSNs is the logical opposite of an 'exclusive OR' , a non-linear problem that typically requires interneurons to detect ( Rumelhart and McClelland , 1986 ) . While one potential mechanism for TANs to detect MSN entropy is by including interneurons , this non-linear detection is linearly solvable if the problem is split into two separate detections: the detection of when both neurons are active and the detection of when neither is active ( Figure 8 , top left ) . Consequently , a potential approximation for MSN entropy would be the detection of coincident activity ( or inactivity ) of pairs of MSNs . While an active readout of joint inactivity is implausible , this problem is solved by the opponent nature of MSN pathways in the BG ( Frank , 2005; Kravitz et al . , 2012 ) , where the TANs could detect either coincident activity of two D1 MSNs signaling Go activity or two D2 MSNs signaling NoGo activity . This scheme would require a synaptic organization such that two MSNs associated with different motor responses synapse close together on the dendrite of a TAN to facilitate coincidence detection ( Figure 9 , top right ) . Coincidence detection between distinct signals is thought to be an important mechanism both for cellular plasticity ( Wang et al . , 2000 ) as well as the coordination of sensory input ( Kapfer et al . , 2002 ) . If synaptic signaling from both MSNs is needed to propagate an action potential to the cell body , then the pair of synapses can be thought of as a coincidence detector . In logical terms , the synapse pairs perform Boolean 'AND' detection ( Figure 9 , top left ) . Several pair of D1 synapses located on the dendrites of a single TAN could result in the summation of these signals in the cell body of the TAN , approximating half of the entropy function signaling that there is evidence for multiple motor responses . Similarly , multiple pairs of D2 synapses contributes to the other half of the entropy function , signaling that there is evidence against multiple responses . Thus , these two measures could effectively approximate decision uncertainty in the MSN population and are detectable by a single TAN via direct synapses from MSNs . In principle , other mechanisms within the striatum could transmit the population uncertainty ( e . g . , via further interactions between MSNs and fast-spiking interneurons ) . Our exercise below thus presents but a single plausible mechanism . 10 . 7554/eLife . 12029 . 011Figure 9 . Approximations of MSN population uncertainty . ( A ) Shannon’s entropy for two MSNs can be expressed with Boolean logic . Low entropy occurs when only one MSN is active ( bottom left or top right box ) and is the exclusive-OR function . High entropy is the logical opposite . ( B ) Spatially organized synapses could allow the detection of activity in two MSNs associated with separate motor responses , indicating high entropy . Activity in pairs of D1-MSNs signal high uncertainty given evidence for multiple responses , activity in pairs of D2-MSNs signal high uncertainty given evidence against multiple responses . ( C ) Detection of 'AND' pairs in Go population approximates Shannon’s entropy across time , whereas simple summation of all Go unit activity does not . Dotted line denotes reversal at mid-point in training . ( D ) Neural network with AND detection performs well in both an 85% reward ( left ) and 40% reward ( right ) environments as compared to networks with fixed TAN behavior . ( E ) Varying pause duration with Shannon’s entropy ( green line ) or the detection of AND conjunctions ( orange line ) results in similar behavior . MSNs , medium spiny neurons; TANs , tonically active neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 12029 . 011 We simulated the network varying the TAN pause with AND detection in the Go units ( see Materials and methods ) . In a network with a fixed duration TAN pause ( 180 ms ) and trained on an 80% reward schedule , AND detection in Go units followed a similar temporal pattern to Shannon’s entropy ( Figure 9 , center left ) . Both AND detection and entropy decline as the network learns the task and rise following the reversal of the reward contingencies . Not all summary statistics of MSN activity follow this pattern: for example , the simple summation of all activity in the Go population gradually declines over training and consequently , is not suitable for an approximation of entropy ( Figure 9 , center right ) . Notably , modulating TAN pause duration with the AND detection can also mitigate the behavioral trade-off between flexibility and stability , similar to the patterns observed using entropy ( Figure 9 , bottom right ) . In the neural network model , varying the duration of the TAN pause alters the degree to which MSNs are disinhibited during reward feedback . Mechanistically , this depends solely on M2 receptors , which are responsible for inhibition of the TANs in the model , and does not involve changes in the activity of M1 receptors ( see Materials and methods ) . However , previous empirical studies have linked M1 receptors to learning after reversal . Both TAN ablations and muscarinic antagonists impair reversal learning but do not affect acquisition in deterministic tasks ( Ragozzino et al . , 2002; Witten et al . , 2010 ) , an effect that is specific to M1 receptors ( McCool et al . , 2008 ) . To investigate the effects of M1 receptors in the model , we simulated selective M1 receptor ablations ( see Materials and methods ) . In our simulations , the simulated ablation of M1 receptors only modestly decreases performance during acquisition , resulting in similar asymptotic accuracy as a control model that does not contain TANs ( Figure 3 , left ) . Following reversal , the effects are much more pronounced as simulated M1 ablation results in severely degraded accuracy ( Figure 3 , right ) . These results are qualitatively consistent with empirical findings showing impairment in reversal learning following M1 antagonists . It is noteworthy that the impairment with simulated M1 ablations only occur as a consequence of the disinhibitory effects of the TAN pause that drive MSN activity to a lower entropic state: control networks simulated without TANs do not show the same degree of impairment ( Figure 3 , right ) . Without this decrease in entropy or the increase in NoGo excitability , the network is able to learn following the reversal . However , both effects combine to allow the network to adapt more flexibly , perseverating in the correct action when it is rewarded but more sensitive to negative feedback facilitating reorienting . 10 . 7554/eLife . 12029 . 012Figure 10 . Post-pause TAN burst . An increase in phasic TAN activity following the feedback-related pause modulates asymptotic performance following reversal . Simulations shown with a fixed TAN pause of intermediate duration ( 190 ms ) in an 85% reward environment , post-pause TAN firing rates are presented in normalized units of change relative over a baseline firing rate corresponding to the tonic firing rate . TAN , tonically active neuron . DOI: http://dx . doi . org/10 . 7554/eLife . 12029 . 012 A phasic increase or rebound burst in TAN activity above tonic firing rates is commonly observed immediately following the reward-related pause ( Aosaki et al . , 2010 ) . The functional significance of this burst is an open question but one likely function of the post-pause burst is to facilitate synaptic plasticity through the release of dopamine during an important time window . Optogentic stimulation of TAN neurons leads to the release of dopamine through the activity of the nicotinic receptors on striatal dopamine terminals ( Cachope et al . , 2012; Threlfell et al . , 2012 ) . Dopamine release precipitated by post-pause TAN activity would result in the delivery of dopamine immediately following a period in which striatal spiny neurons were disinhibited . This timing has important plasticity consequences as dopamine release following glutamatergic input promotes spine enlargement ( Yagishita et al . , 2014 ) . Consequently , we hypothesize the post-pause TAN burst promotes plasticity by releasing dopamine during a sensitive time window following increased spiny neuron activity , while concurrently suppressing potentially interfering activity with muscarinic inhibition . To investigate the consequences of this hypothesis , we modeled the effects of a phasic increase in TAN activity following the feedback pause in the reversal learning task in an 85% reward environment . We found higher post-pause TAN firing rates resulted in higher asymptotic accuracy following reversal as compared to lower firing rates Figure 10 . Interestingly , this effect was selective to reversal: there was no additional effect of of post-pause TAN activity on learning speed , both prior to and following reversal , and no effects on asymptotic performance prior to reversal . Together , the effects of TAN pause and rebound burst act to enhance the BG network’s ability to reverse and stabilize newly learned contingencies . While there is some evidence that the duration of the TAN response may be an important regulator of cellular activity , the duration of the TAN pause has often not been linked to signaled probability of reward ( Morris et al . , 2004 ) . It is important to note that the uncertainty over which actions are selected is independent of signaled probability of reward . As such , we would not expect the TAN response to vary with signaled probability of reward in well-trained animals performing a task in which the reward contingencies do not change . If the animals have learned the task well , then the uncertainty about the frequency of reward in a fixed schedule will be low . Furthermore , the predictions of our model are not sensitive to the choice of TAN pause duration as a parameter of interest . In the implementation proposed , the duration of the TAN pause controls the degree to which spiny neurons are disinhibited . The algorithm we have used to simulate learning in the neural network model is sensitive to differences in excitation between stimulus presentation and feedback and these changes in disinhibition result in changes to the degree weights are updated in the neural network model . The model would make the same prediction if we varied the overall firing rate during the pause or altered inhibition directly . An additional limitation is the focus of the current work on the effects of TAN behavior on MSN excitability relative to other components of striatal cholinergic signaling . The initiation of the TAN response depends on both dopamine and thalamic signaling , ( Aosaki et al . , 1994; Ding et al . , 2010 ) both of which convey behaviorally relevant information ( Montague et al . , 1996; Matsumoto et al . , 2001 ) and which we have not considered in the current work . Rather , we have only considered the modulation of such signals by MSN activity via collaterals . TANs may play a role integrating thalamic and dopaminergic signaling , but we are unable to make predictions about changes in the TAN response motivated by cell signaling or receptor activation . The neural network model presented here was adapted from the basal ganglia model presented by Frank ( 2006 ) and implemented within the emergent neural network software ( Aisa et al . , 2008 ) . The model is available on our online repository at http://ski . clps . brown . edu/BG_Projects/Emergent_7 . 0+/ . The model uses point neurons with excitatory , inhibitory and leak conductance that contribute to an integrated membrane potential transformed into a rate-code . Learning in the model is accomplished with the Leabra algorithm ( O’Reilly and Munakata , 2000 ) and a reinforcement learning version based on dopaminergic modulation of Hebbian plasticity ( Frank , 2005 ) . The neural network model is an attractor network organized into layers of point neurons ( units ) , in which layers represent neural structures ( c . f . Frank , 2006 for a more detailed description of the model ) . The network is dynamic across time , and units within the network have stochastic behavior . The membrane potential Vm of each unit within a layer is updated at each cycle ( discrete time-point ) as a function of its net current Inet and a time constant τnet: dVmdt=τm×Inet The net current of a unit is , in turn , a stochastic function of its excitatory , inhibitory , and leak conductances ( g ) updated at each cycle as follows: Inet=ge ( t ) g¯e ( Ee−Vm ) + gitg¯iEi-Vm+ g¯iEl-Vm+ … where Vm is the membrane potential of the unit , EC is the equilibrium potential for current c , and where the subscripts e , l , and i refer to the excitatory , leak , and inhibitory currents , respectively . The total conductance of each channel gc are decomposed into the constant and time varying components g¯c and gc ( t ) . Units in the network are connected via synaptic weights , which can be excitatory or inhibitory . The total excitatory input/conductance ge ( t ) to a unit is a function of the mean of the product the firing rate of each sending unit xi and the corresponding synaptic weights wi , a time constant τg and a scaling factor κ: dgedt=τg ( k1n∑ixiwi−ge ( t ) ) Inhibitory conductance is computed similarly , but applied to synaptic inputs that come from inhibitory neurons , while leak conductance does not vary with time . Gaussian noise ( σ ) is added to the membrane potential Vm of a subset of units in the network ( units in the motor cortex and substantia nigra layers , σ = . 0015 and . 002 respectively , sampled from at each cycle of processing ) . As the conductances of each unit vary across time and between trial to trial as a function of input activity , the rate-coded activity of each unit with a layer varies with its inputs and noise . This added noise induces stochasticity in network choices ( and their latencies; Ratcliff and Frank 2012 ) for any given set of synaptic weights , encouraging exploration: noise in motor cortex induces changes in the degree to which a candidate response is active and hence subject to disinhibition by gating via striatum , and noise in the SNc facilitates within-trial variation in the balance between Go and NoGo unit activity , differentially emphasizing learned positive versus negative outcomes and their effect on gating . Together , these sources of variance yield stochastic choice and dynamics within the striatum shown in Figure 2 . The activity communicated to other units in the network , yj , is a threshold sigmoid function of the membrane potential: yj ( t ) =11+1γ[Vm ( t ) −Θ]+ where γ is a gain parameter and where [X]+ is a function that returns 0 if X≤0 and X otherwise . This function is discontinuous at Vm ( t ) =Θ , and is smoothed with a Gaussian noise kernel ( μ=0 , σ=0 . 005 ) to produce a softer threshold and represent the intrinsic processing noise in neurons: yj* ( x ) =∫−∞∞12πσe−z22σ2yj ( z−x ) dz where x is the value [Vm ( t ) −Θ]+ and yj* ( x ) is the noise-convoluted activation . The layers in the network represent neural structures within the basal ganglia and thalamus . As a first approximation , action selection in the neural network is mediated by two simulated populations of MSNs . The populations of 'Go' and 'NoGo' units , representing striatonigral ( 'D1' ) and striatopallidal ( 'D2' ) MSNs , respectively , receive excitatory input from a cortical layer corresponding to a unique stimulus . Both the Go and NoGo layer contain 18 units , half of which , through their downstream targets , are connected to one of two motor responses . Reciprocal connections with a layer of inhibitory interneurons ( simulating the GABergic fast-spiking interneurons in the striatum ) control the overall excitability of the two populations . Activity in Go units inhibits a population in the globus pallidus interal segment ( GPi ) , which results in disinhibition of the thalamus . Activity in the NoGo units inhibits a population in the globus pallidus external segment ( GPe ) , which results in disinhibition of the GPi and inhibition of the thalamus . If there is sufficient activity in the thalamus , it provides a bolus of activity to a corresponding motor cortical column , which can then inhibit its competitors via lateral inhibition , and an action will be selected . This process is stochastic at a network level and depends both on interactions between units that vary with time ( learning ) as well as noise in the network within trials . However , while this process has trial to trial variability , activity in the Go pathway will facilitate the selection of an action through its downstream effects on the thalamus while activity in the NoGo pathway will suppress response selection . Learning in the model occurs through weight updating in corticostriatal synapses without a supervised learning signal . A combination of a Hebbian learning rule and a contrastive Hebbian learning rule are used to determine the weight updates . Variants of contrastive Hebbian learning are consistent with large scale simulations of spike-time dependent plasticity ( O'Reilly et al . , 2015 ) and serve as a simplifying assumption to a more detailed mechanism of synaptic plasticity . The Hebbian component of the learning rule assumes that the coactivation of MSNs and their cortical inputs proportionally determines the synaptic weight change . The contrastive Hebbian component is determined by the difference in coactivation of pre- and postsynaptic activity across response selection ( 'minus phase' ) and outcome feedback ( 'plus phase' ) . The equation for the Hebbian weight change Δhebbwij between sending unit xi and receiving unit yj is defined: Δhebbwij=yj+ ( xi+−wij ) where yj+ refers to the activity of the receiving unit during outcome feedback ( 'plus' phase ) and xi+ is the activity of the sending unit . The contrastive Hebbian weight change is ΔCHLwij= ( yj+xi+ ) − ( yj−xi− ) Where yj− and xi− are the activity of the receiving and sending units during action selection ( 'minus' phase ) . The contrastive Hebbian term is subject to a soft-weight bound to keep between 0 and 1: ΔsbCHLwij=[ΔCHL]+ ( 1−wij ) +[ΔCHL]−wij where [X]+ is a function that returns X if X>0 and 0 otherwise and where [X]− is a function that returns X if X<0 and 0 otherwise . The Hebbian and contrastive Hebbian terms are combined additively with a normalized mixing constant κhebb ∆wij=εκhebb∆hebb+1-κhebb∆sbCHL Dopamine acts as a training signal in the model , providing a phasic increase during feedback for correct responses and a phasic pause for incorrect responses . Dopamine is simulated to act through D1 receptors in Go units , increasing excitability . In addition , D1 receptors were simulated to enhance contrast by increasing the striatal unit’s activation gain and activation threshold . This has the effect of increasing the activity of highly active Go units and decreasing the activity of weakly active units , increasing the signal-to-noise ratio . D2 receptors were simulated in NoGo units such that an increase in dopamine decreases NoGo excitability . As a result , a phasic increase in dopamine during feedback has the effect of increasing Go activity relative to NoGo activity while a phasic decrease in dopamine will have the reverse effect . Altering the activity of Go and NoGo units in response to phasic changes in dopamine during feedback alters coritcostriatal weights through the contrastive Hebbian component of the learning rule . This facilitates error driven learning without providing the network a supervised learning signal . TANs are endogenously active in the absence of synaptic activity ( Bennett and Wilson , 1999 ) and were modeled as a separate endogenously active layer in which the leak channel equilibrium potential El was typically higher than Vm . Unlike striatal Go and NoGo units , the activity of TANs were not dependent on synaptic signaling from other units in the network and TANs were simulated with little stochasticity . During stimulus presentation and action selection , TAN activity was held constant across all trials in all simulations ( Figure 2 , top left ) . Following action selection , a TAN burst–pause was simulated during outcome feedback . The burst was simulated at the onset of feedback to mirror the TAN burst by transiently increasing Vm above the equilibrium potential . The subsequent pause was generated with an accommodation current , which allows the initial burst of TAN activity to create a subsequent hyperpolarization . This simulates the after-hyperpolarization that has been found to follow a depolarization in TANs via calcium-dependent potassium current ( Wilson and Goldberg , 2006 ) . An accommodation current Ia drives the membrane potential toward a low value which is added to Inet: Ia=ga ( t ) g¯a ( Ea−Vm ) . A high accommodation current has the effect of hyperpolarizing the neuron as a function of how active it has been , simulating gated ion channels that accumulated with activity and driving the membrane potential to a low value . Consequently , the initial high firing rate at the onset of reinforcement causes the activity-dependent accommodation current to hyperpolarize the TAN units , silencing them for a length of time during feedback . The accommodation current is updated at each time step as a function of its time constant , τa: dgadt=τga ( 1−ga ( t ) ) if ba ( t ) >Θaτga ( 0−ga ( t ) ) if ba ( t ) <Θd where τga is the time constant of the channel conductance , Θa and Θd are the activation and deactivation thresholds required to invoke accommodation , respectively . The basis variable ba is a time average of the activation state: dbadt=τba ( yj−ba ( t ) ) where τba is the time basis variable time constant and yj is the unit activity . During periods of persistently high activity , the basis variable ba will increase in proportion to its time constant τba . When the basis variable exceeds the activation threshold Θa , the accommodation conductance will increase , resulting in a net decline in current and a lower TAN firing rate . This process simulates the Ca2+ dependent potassium currents in TANs: the basis variable simulates the build up of Ca2+ in the cell and the effects of the basis variable on accommodation conductance simulates the opening ( or closure ) of Ca2+ dependent channels . During the subsequent pause , dbadt is negative as activity is low , causing the accommodation current to subside as the basis variable falls below the deactivation threshold Θd , resulting in an increase in Vm at the end of feedback ( post-pause rebound , Figure 2 , bottom left ) . The behavior of the TAN pause was manipulated by varying τba . Networks with fixed TAN pauses were simulated by specifying a constant value of τba such that the duration of the TAN pause lasted a pre-specified duration . The duration of the pause is reported in milliseconds ( ms ) , which converted from cycles ( the base unit of time within the network ) at an assumed rate of 10 ms/cycle ( Ratcliff and Frank , 2012 ) . Pause durations were simulated ranging from 120 ms to 280 ms in 10 ms steps . For networks with a variable TAN pause , we simulated the impact of inhibitory collaterals from MSNs onto TANs such that the time constant τba was proportional to MSN activity ( using either entropy or a more realistic approximation thereof , see below ) . This embodies our assumption that MSNs do not directly induce the TAN pause ( which is driven by external inputs , e . g . from thalamus ) but modulate its duration via accommodation ( potentially via calcium-dependent potassium currents; Wilson and Goldberg , 2006 ) leading to an MSN activity-dependent reduction in TAN activity via direct signaling . For the initial simulations of an adaptive mechanism to control pause duration , τba was updated as a function of the Shannon’s entropy across the population of Go units on a trial to trial basis . On each time step during stimulus presentation , the rate-coded activity of each Go unit was normalized such that the normalized activity of all Go units summed to one . The normalized activity was then treated as a decision variable for the purposes of calculating entropy: the sum of activity over all units that contributed to a single response was treated as the probability of the selection of that response: pa ( t ) =∑i=1nayia ( t ) where yia ( t ) is the activity of unit i corresponding to action a at time t . This assumption conforms with other interpretations of population activity within typical network models , where a probability distribution can be created by normalizing activation levels within a finite set of units ( Hinton and Sejnowski , 1983; Rao , 2005; D'Angiulli et al . , 2013 ) . While the model presented here is more biologically complex , the same principle applies when treating the firing rates of discrete units as a probability distribution over actions . Shannon’s entropy was calculated with pa ( t ) and summing across all cycles within the action selection phase: H=−∑t∑apa ( t ) log2pa ( t ) High entropy indicates that there is more competition between actions in the Go units and low entropy indicates low competition . We leverage this same quantity for controlling TAN pauses via adaptive feedback ( as described above ) and also as a statistic to approximate the uncertainty of the network over its action selection policy , and how this evolves across learning . A second more biophysical feedback mechanism was simulated , where instead of using an analytical expression for MSN entropy , TANs directly approximate MSN population entropy through synaptic integration . In these simulations , co-activation of pairs of MSNs co-localized on TAN dendrites were assumed to drive changes in TAN pause behavior . Thus the TAN time-constant τba was adjusted as a function of supra-threshold activity in pairs of Go units , in which each member of the pair corresponded to a different action ( Figure 9 ) . The 9 Go units corresponding to each of the 2 available actions were organized into 9 pairs of units . The activity in each unit was threshold and if the activity in both units were above threshold , the pair was counted as having an 'AND conjunction' ( as the detection of supra-threshold activity is equivalent to the Boolean function AND ) : cj=1 if ( yjk > θ ) AND ( yjl > θ ) 0 otherwise where cj is the Boolean value ( 0 or 1 ) of the 'AND conjunction' for the pair of units j , yjk and yjl are the activities for the units associated with actions k and l , respectively , θ is the threshold value . The value of cj across over all the actions and across all cycles in the action selection phase of a trial: Hconj=∑t∑jcj ( t ) The time constant τba was set as a function of Hconj which accumulated during response selection and hence affected the duration of subsequent accommodation hyperpolarization during the pause induced by the initial burst in the subsequent reinforcement phase . TANs were simulated to modulate Go and NoGo activity largely through the activity of M2 and M1 muscarinic receptors . The effects of M1 and M2 receptors are relatively well understood ( Goldberg and Reynolds , 2011 ) and thus suitable for modeling . M1 and M2 receptors have opposing effects on MSN excitability: M1 activity increases dendritic excitability of indirect pathway MSNs through the post-synaptic closure of Kir2 K+ channels ( Shen et al . , 2007 ) , while M2 receptors inhibit glutamate release in presynaptic terminals of both direct and indirect pathway MSNs ( Calabresi et al . , 1998; Ding et al . , 2010 ) . Cholinergic signaling can also modulate MSN excitability through increased GABAergic inhibition ( Witten et al . , 2010 ) . Crucially , these effects act at different time scales: M1 activity is longer lasting and slower to initiate than the pre-synaptic effects of M2 receptors ( Shen et al . , 2005; Ding et al . , 2010 ) . Thus , the different temporal dynamics receptor activity may interact with the time-course of the TAN response: whereas the TAN burst activates M1 receptors and increases excitability in indirect pathway MSNs , the subsequent TAN pause reduces M2-mediated presynaptic inhibition and postsynaptic GABAergic inhibition ( Gerfen and Surmeier , 2011 ) . M2-like muscarinic receptors are located on pre- and postsynaptic glutamatergic afferents of MSNs have the effect of inhibiting corticostriatal glutamate transmission ( Goldberg and Reynolds , 2011 ) . The effects of M2-like receptors were modeled with inhibitory synaptic connections from TANs to the Go and NoGo units . Excitatory synaptic connections to inhibitory interneurons were also used to simulate the nicotinic stimulation on GABAergic interneurons ( Witten et al . , 2010 ) . These interneurons project to both Go and NoGo units and have an inhibitory effect . The effects of simulated M2 receptors and excitatory connections to the interneurons results in increased inhibition of both layers during action selection when TAN activity is constant ( Figure 2 , left ) and disinhibition during the feedback pause ( Figure 2 , center ) . M1 muscarinic receptors were simulated with a transitory increase in the leak channel Equilibrium potential El for NoGo units during feedback . This was done to simulate the effects of M1 activity on inward rectifying potassium currents ( Goldberg and Reynolds , 2011 ) . Although M1-mRNA is found in both striatopallidal and striatonigral MSNs , the effect was simulated in the NoGo Layer only as increased sensitivity to glutamatergic signaling is selective to striatopallidal MSNs ( Shen et al . , 2007 ) . In order to simulate M1 antogonists , the transitory increase in leak channel equilibrium potential in NoGo units during the simulated TAN pause was removed . No effects of M1 receptors were simulated during the action selection phase . Where noted , a rebound post–pause phasic increase in TAN activity was also modeled . A phasic increase in TAN activity is commonly reported following the feedback related TAN pause ( Aosaki et al . , 2010 ) . Mediated by nicotinic acetylcholine receptors located on dopamine terminals in the striatum , the release of acetylcholine associated with a phasic increase in TAN activity evokes dopamine release ( Cachope et al . , 2012; Threlfell et al . , 2012 ) . The release of DA during the post–pause phasic TAN burst may modulate cortico-straital plasticity in relation to activity during the feedback related pause DA release promotes spine growth specifically when stimulated after spiny neuron activity ( Yagishita et al . , 2014 ) . The effects of the post-pause TAN burst were modeled by simulating the modulation of dopamine release by nicotinic receptors through the modulation of the dopamine membrane potential during the plus phase . The observed dynamic range of effects were normed as change TAN firing rate above baseline from 0 to 1 . An approximately Bayesian reinforcement learning model was used as a computational level description of the behavior of the neural network model . For i . i . d . Bernoulli trials in a stationary task ( a reasonable approximation of the task ) , the exact posterior distribution of expected values is a beta distribution parameterized with α and β ( Daw et al . , 2005 ) Q ( s , a ) ~Beta ( α , β ) The parameters α and β can be updated online after each trial with the following rules: α ( t+1 ) =α ( t ) +r ( t ) βt+1=βt+1-rt where the reward on the current trial , r ( t ) , is either 0 or 1 . Because the task is non-stationary , a multiplicative parameter γ ranging from 0 to 1 was used to decay the parameters α and β during the update: α ( t+1 ) =γ ( α ( t ) +r ( t ) ) α ( t+1 ) =γ ( α ( t ) +1-r ( t ) ) This multiplicative decay parameter increases the flexibility of the learner and has been used previously to model human learning ( Daw et al . , 2005; Doll et al . , 2009 , 2011 ) . In simulations , γ was either held constant throughout training ( analogous to fixed pause duration ) or varied as a function of trial-to-trial changes in uncertainty with two free parameters , γ0 and γ1: logit ( γ ) =γ0+γ1⋅ΔH The free parameters γ0 and γ1 determine the baseline decay rate and sensitivity to trial-to-trial variations in uncertainty . Importantly , γ1 was negative as performance is best when the model decays more quickly ( and hence learns more from individual outcomes ) during conditions of high uncertainty/volatility , analogous to shorter TAN pauses with more MSN uncertainty . Uncertainty was defined as the Shannon’s entropy of the action selection probability , determined by integrating the expected values of both actions over the belief distributions: H=−∑i=1np ( a\|s ) log2p ( a\|s ) The difference in entropy between trials was smoothed using a delta rule algorithm with the learning rate η as a parameter . ΔH ( a ) was initialized at the 1 bit ( the maximum possible entropy ) and updated after each trial with the following rule: ΔH←ΔH+ηδ δ=∆H-Ht-Ht-1 Smoothing the difference in entropy between trials reduces the influence of trial-to-trial variance and allows for entropy for previous recent trials to influence the decay rate . Action selection was accomplished through sampling Q-values . This allows for the consideration of uncertainty during the action selection . For each trial , a Q-value from the belief distribution of each available action was randomly sampled . The action with the highest Q-value was selected for the given trial . An algorithmic model of the basal ganglia adapted from the OpAL model ( Collins and Frank , 2014 ) was used to provide a mechanistic description of the feedback-control mechanism proposed in the neural network . OpAl is an actor–critic reinforcement learning model that mimics the opponent ( D1/D2 ) actor system in the neural network . In OpAL , a single critic learns the value of a stimulus with its own learning rate ( ηc ) . A prediction error ( δ ) , the difference between the observed reward ( r ) on a trial and the learned value of the critic on trial ( Vt ) , is used to update the estimate on each trial . δ=r−Vt Vt+1=Vt+ηcδ The prediction error on each trial is also used to update to actor values , a 'Go' ( Gt ) and 'NoGo' ( Nt ) value , for the chosen action conditional on the current stimulus with two separate learning rates: Gt+1=Gt+ηGδ Nt+1=Nt+ηNδ where ηG and ηN are the learning rates for the G and N weights , respectively . These two actor values correspond to the Go and NoGo units in the neural network and model learned contributions of striatonigral and striatopallidal MSNs to action selection . Choices between actions were made using a softmax policy choice on the linear combination of actor weights: p ( a ) ∝exp{βGGa−βNNa} Here , βG and βN control the degree to which each of the weights influence action and p ( a ) is the probability action a is selected . Collins and Frank ( 2014 ) simulate a variety of documented effects with this model , including how dopamine affects the asymmetry in learning and choice incentive ( sensitivity to gains vs costs of alternative actions ) , show how its behavior converges to expected values and provide a normative interpretation . Here , we expand OpAL to include the effect of TANs on modulating learning rate , using a multiplicative decay term , γ . On each trial , following the update of the actor weights , the weights were decayed to a naïve prior with the inverse logit transform of γ as follows: Gt+1←Gt+111+e−γ+0 . 5×1−11+e−γ Nt+1←Nt+111+e−γ+0 . 5×1−11+e−γ The inverse logit transform of γ range bounds the decay rate between zero and one . Gamma was either held fixed , simulating fixed duration TAN pauses , or was allowed to vary with as a linear function of entropy in the model . Entropy in the model was the Shannon’s entropy of the policy function: H=−∑ap ( a ) log2p ( a ) Both the neural network and Bayesian models were trained on a series of two alternative forced choice tasks . In each trial , the models were presented with one of two stimuli and made one of two responses . Equal valued rewards were delivered pseudo-randomly on a fixed stochastic schedule over epochs of 20 trials . The reward schedule for the two options were not yoked , such that sampling one action did not provide full information of the other’s reward schedule . After 10 epochs of 20 trials ( or 200 trials ) , the reward contingencies for each action were exchanged and the models were trained on an additional 10 epochs . For both the neural network and the Bayesian models , each parameterization was trained on 50 instantiations . Initial weights of the neural network were randomized for each simulation . Simulations were run over eight reward schedules . The payout schedules for the best action were 85 , 80 , 75 , 70 , 65 , 60 , 55 and 40% . The payout schedules for the lowest rewarded action action were 15 , 20 , 25 , 30 , 35 , 40 , 45 and 10% . Comparisons between the neural network with and without TANs , as well as the manipulation of pause duration , were shown in networks trained on an 80/20% reward schedule .	One of the keys to successful learning is being able to adjust behavior on the basis of experience . In simple terms , it pays to repeat behaviors with positive outcomes and avoid those with negative outcomes . A group of brain regions known collectively as the basal ganglia supports this process by guiding behavior based on trial and error . However , circumstances can and do change: a behavior that routinely produces a positive outcome now will not necessarily always do so in the future . The outcome of a given behavior can also vary from time to time purely by chance: even the most appropriate action can sometimes lead to a negative outcome but should be repeated again . Exactly how the basal ganglia take into account this uncertainty over behavioral outcomes to appropriately guide learning is unclear . Franklin and Frank have now come up with a possible explanation by building on an existing computational model of the basal ganglia . Whereas the old model assumes that the rate of learning given an unexpected outcome always remains constant , in the new model learning occurs more quickly when the outcome of a behavior is uncertain . This makes intuitive sense , in that rapid learning is especially useful during the initial stages of learning a new task or following a sudden change in circumstances . The new model proposes that a group of cells called tonically active interneurons ( TANs ) , which release the chemical acetylcholine , enable the basal ganglia to take account of uncertainty . TANs are found in a basal ganglia structure called the striatum and have a characteristic firing pattern during important outcomes , consisting of a burst of activity followed by a pause lasting several hundred milliseconds . The model suggests that when the outcome of a behavior is uncertain , the length of this pause is increased . This boosts the activity of another group of neurons in the striatum , known as spiny neurons , and this in turn increases the rate of learning . Franklin and Frank found that by varying the length of the TAN pause , the basal ganglia can adjust learning rates based on the degree of uncertainty over behavioral outcomes . Comparisons show that the TAN computational model optimizes the accuracy and flexibility of learning across different environments , while also accounting for findings which show that TAN lesions induce insensitivity to changes in decision outcomes . The next step is to test some of the new predictions about uncertainty experimentally .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2015	A cholinergic feedback circuit to regulate striatal population uncertainty and optimize reinforcement learning
During inflammation and infection , hematopoietic stem and progenitor cells are stimulated to proliferate and differentiate into mature immune cells , especially of the myeloid lineage . MicroRNA-146a ( miR-146a ) is a critical negative regulator of inflammation . Deletion of miR-146a produces effects that appear as dysregulated inflammatory hematopoiesis , leading to a decline in the number and quality of hematopoietic stem cells ( HSCs ) , excessive myeloproliferation , and , ultimately , to HSC exhaustion and hematopoietic neoplasms . At the cellular level , the defects are attributable to both an intrinsic problem in the miR-146a–deficient HSCs and extrinsic effects of lymphocytes and nonhematopoietic cells . At the molecular level , this involves a molecular axis consisting of miR-146a , signaling protein TRAF6 , transcriptional factor NF-κB , and cytokine IL-6 . This study has identified miR-146a to be a critical regulator of HSC homeostasis during chronic inflammation in mice and provided a molecular connection between chronic inflammation and the development of bone marrow failure and myeloproliferative neoplasms . Hematopoietic stem cells ( HSCs ) have the ability to self-renew and replenish the entire hematopoietic repertoire during the lifetime of an organism . Balanced self-renewal vs differentiation of HSCs is intricately regulated to ensure the long-term maintenance of HSCs and the hematopoietic system ( Seita and Weissman , 2010 ) . Under stressed conditions , such as inflammation and infection , the balance is shifted in favor of hematopoietic stem and progenitor cell ( HSPC ) proliferation and differentiation to produce more mature immune cells ( King and Goodell , 2011 ) . Following the discovery that HSCs express TLRs and may sense and respond to infection and inflammatory signals directly ( Nagai et al . , 2006 ) , there has been an increasing appreciation of the role of proinflammatory cytokines and infection in the modulation of HSC activity . Numerous recent studies have shown that TLR activation or interferon stimulation leads to proliferation , skewed myeloid differentiation and impaired engraftment , and self-renewal of HSCs ( Essers et al . , 2009; Baldridge et al . , 2010; Esplin et al . , 2011 ) . Since its discovery over 25 years ago , NF-κB has been shown to be active in a wide variety of innate and adaptive immune cells as well as nonhematopoietic cells and to function as an essential player in orchestrating inflammation and immune cell functions ( Baltimore , 2011 ) . However , the function of NF-κB in HSCs remains relatively unexplored . Under stress-free conditions , NF-κB is not essential for HSC function , because mice genetically deleted for NF-κB subunits , such as Nfkb1 ( also known as p50 ) and Rel ( also known as c-Rel ) , have no apparent developmental abnormality in the hematopoietic system . Mice engrafted with p50 or c-Rel knockout HSCs or RelA knockout fetal HSCs also develop relatively normal immune system under stress-free conditions ( Gerondakis et al . , 2012 ) . However , mice with activated NF-κB signaling , as a consequence of deleting IκBα , A20 , or the inhibitory domain of p50 or p52 subunits of NF-κB , display severe inflammation , early lethality , and complex phenotypes , making studies of HSCs difficult to perform and interpret ( Lee et al . , 2000; Gerondakis et al . , 2006 ) . In recent years , microRNAs ( miRNAs ) have emerged as a class of small noncoding RNAs involved in the regulation of NF-κB ( Boldin and Baltimore , 2012 ) . Among them , miR-146a has been shown to be a particularly important negative regulator of NF-κB by targeting two upstream signal transducers , TRAF6 and IRAK1 . Mice with targeted miR-146a deletion represent one of the first genetic mouse models with NF-κB-driven chronic and low-grade inflammation that develops spontaneously with aging and can be accelerated by repeated stimulation , allowing investigation of the long-term effects of chronic inflammation and NF-κB activation on HSCs and oncogenic processes ( Boldin et al . , 2011; Zhao et al . , 2011 ) . Given this background , we have used miR-146a-deficient mice to examine the function of miR-146a and NF-κB in HSCs and progenitor cells during chronic inflammation and to directly test a long-standing hypothesis that chronic inflammation promotes excessive HSC and progenitor cell proliferation and differentiation and can lead to eventual HSC exhaustion and pathological myelopoiesis . Here , we demonstrate that this single miRNA , miR-146a , functions as a critical guardian of HSC quality and longevity during chronic inflammatory stress in mice . In the absence of miR-146a , HSC homeostasis is disrupted under physiological stresses such as aging and periodic bacterial encounters , as indicated by declines of HSC number and quality and dysregulated HSPC proliferation and differentiation . Chronically , these nominal stressors can lead to severe pathologies , such as HSC exhaustion , bone marrow failure , and myeloproliferative disease , produced by chronic NF-κB hyperactivation and IL-6 overproduction . This study speaks to a molecular pathway involving miR-146a/TRAF6/NF-κB/IL-6 that links chronic inflammatory stresses to the functional decline and depletion of HSCs and the development of myeloproliferative diseases . To examine the role of miR-146a , we first assessed the expression of miR-146a and its related family member , miR-146b , during hematopoietic differentiation . We purified by FACS various types of hematopoietic stem and progenitor cell ( HSPC ) populations from young wild-type ( WT ) mice . We found that miR-146a and miR-146b were expressed at variable levels throughout hematopoietic development . The expression of miR-146a increased by twofold as long-term HSCs ( defined as Lineage−Sca1+cKit+ CD150+CD48− ) differentiated into a mixed pool of short-term HSCs and multipotent progenitor cells ( MPPs ) ( defined by Lineage−Sca1+cKit+ , referred to as LSK cells ) . The lowest expression of miR-146a was detected in myeloid progenitor cells ( defined by Lineage−Sca1−cKit+ , referred to as L−S−K+ cells ) ( Figure 1A ) . In comparison , miR-146b expression was more uniform throughout hematopoietic development ( Figure 1A ) . This expression pattern suggests that miR-146a and miR-146b could be functional in cells as primitive as the long-term HSCs and throughout hematopoietic development . 10 . 7554/eLife . 00537 . 003Figure 1 . Accelerated HSC decline and myeloproliferation in miR-146a–deficient mice during chronic inflammation . ( A ) MiR-146a and miR-146b expression in FACS-sorted HSPC populations by Taqman RT-qPCR . Lin-BM , lineage negative bone marrow cells; L−S+K+ ( LSK ) , Lin−Sca1+cKit+; HSC , LSK CD150+CD48−; L−S−K+ , Lin−Sca1−cKit+; L−S+K− , Lin−Sca1+cKit−; miR-146a KO BM , total bone marrow cells from Mir146a−/− mice . ( B ) . Representative FACS plots of LSK cells and CD150+CD48− or EPCR+ HSCs from BM of 8-month-old wild-type ( WT ) and miR-146a KO mice . Quantification of number of BM CD45+ cells , LSK cells , LSK CD150+CD48− HSCs , LSK EPCR+ HSCs and percent of LSK CD150+CD48− HSCs of total BM from 8-month-old ( C ) and 12-month-old ( D ) WT and miR-146a KO mice by FACS . ( E ) Quantification of percent of LSK cells of total BM and percent of HSCs of LSK gate from BM of 8-month-old WT and miR-146a KO mice by FACS . ( F ) Quantification of number of LSK cells and LSK CD150+CD48− HSCs from spleen of 8-month-old WT and miR-146a KO mice by FACS . ( G ) . Total number of LSK CD150+CD48− or LSK CD150+CD48− HSCs from BM and spleen of 6-month-old WT and miR-146a KO mice . ( H ) – ( J ) 8-Week-old WT and miR-146a KO ( miR KO ) mice were subjected to repeated low-dose of intraperitoneal LPS stimulation ( 1 mg LPS/kg of body weight for 8 times ) or PBS control spread over a month . At the end the month , four groups of mice were harvested for FACS analysis . ( H ) Quantification of percent of LSK cells of total BM , CD150+CD48− HSCs of LSK gate , and number of CD11b+ myeloid cells in BM . I . Number of myeloid cells ( CD11b+ or Gr1+ ) in peripheral blood . ( J ) Quantification of total number of LSK CD150+CD48− HSCs , LSK EPCR+ HSCs and LSK cells in spleen . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 00310 . 7554/eLife . 00537 . 004Figure 1—figure supplement 1 . HSPC FACS analysis and colony-forming ability in 6-week-old and 4-month-old WT and miR-146a KO mice . ( A ) Quantification of number of white blood cells ( CD45 ) , T cells ( CD3ε ) , myeloid cells ( CD11b ) , B cells ( CD19 ) , nucleated erythrocytes ( Ter119 ) , LSK cells , LSK CD150+CD48− HSCs and LSK CD150+CD48−EPCR+ HSCs from spleen of 6-week-old WT and miR-146a KO mice by FACS . ( B ) Quantification of number of white blood cells ( CD45 ) , LSK cells , LSK CD150+CD48− HSCs , Lin-cKit+Sca1− myeloid progenitors from BM of 6-week-old WT and miR-146a KO mice by FACS . ( C ) Colony forming units ( CFU ) in vitro in methylcellulose medium per 100 , 000 total BM cells from 6-week-old WT and miR-146a KO mice . ( D ) Quantification of number or percent of white blood cells ( CD45 ) , T cells ( CD3ε ) , myeloid cells ( CD11b ) , B cells ( CD19 ) , LSK cells and LSK CD150+CD48− HSCs from BM of 4-month-old WT and miR-146a KO mice by FACS . ( E ) ( related to Figure 1H–J ) 8-Week-old WT and miR-146a KO ( miR KO ) mice were subjected to repeated low-dose of intraperitoneal LPS stimulation ( 1 mg LPS/kg of body weight for 8 times ) spread over a month . WT and miR KO mice receiving phosphate-buffered saline ( PBS ) injection were included as controls . At the end the month , four groups of mice ( WT PBS , miR KO PBS , WT LPS , and miR KO LPS ) were harvested for FACS analysis . Spleen weight , number of CD45+ , CD11b+ , and Ter119+ cells in spleen were shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 004 To characterize the physiological function of miR-146a in HSCs , we examined the consequences of miR-146a deficiency on various hematopoietic cells using mice with a targeted deletion of the Mir146a gene ( Boldin et al . , 2011; Zhao et al . , 2011 ) . We found identical numbers of phenotypically defined subsets and equal colony-forming ability in vitro in 6-week-old WT and Mir146a−/− ( miR-146a KO ) mice ( Figure 1—figure supplement 1A–C ) . These data indicate that deleting miR-146a has no detectable effect on hematopoiesis early on in life in a standard pathogen-free environment . We have previously shown that miR-146a KO mice develop spontaneous inflammation as they age ( Boldin et al . , 2011; Zhao et al . , 2011 ) . To characterize the role of miR-146a in HSCs during low level of chronic inflammation , we allowed age- and sex-matched WT and miR-146a KO mice to age over a year . By 4 months , miR-146a KO mice developed a mildly hypercellular bone marrow indicated by an increase in total bone marrow CD45+ cells , LSK cells , and long-term HSCs , with unaltered percentages of CD19+ , CD11b+ and CD3ε+ cells ( Figure 1—figure supplement 1D ) . However , the increase in bone marrow HSPCs and mature cells was not sustained . By 8 months of age , miR-146a KO mice showed a significant decrease in the number of total bone marrow cells and phenotypically defined HSPCs , including LSK cells and CD150+CD48− or EPCR+ long-term HSCs ( Figure 1B , C ) . The depletion became progressively more severe by 12 months of age when the majority of miR-146a KO mice showed only a residual number of CD45+ bone marrow cells and nearly complete exhaustion of HSCs ( Figure 1D ) . To understand the cellular process leading to a transient hypercellular marrow and eventual HSC exhaustion , we analyzed the bone marrow and spleen of aging miR-146a KO mice . Dysregulated hematopoiesis was observed in 8-month-old KO mice , as shown by an increased percent of LSK cells within the total bone marrow but a significantly decreased fraction of long-term HSCs among the LSK cells ( Figure 1E ) . By this age , miR-146a KO mice have already developed a prominent myeloproliferative phenotype ( Zhao et al . , 2011 ) . In addition , miR-146a KO mice also showed a significant increase in the number of LSK cells and long-term HSCs in their spleens , which may be due to HSPC mobilization from bone marrow to spleen and/or de novo splenic HSPC proliferation in response to bone marrow failure ( Figure 1F ) . Furthermore , when total HSCs from the marrow and spleen of the same mice were summed , a reduction in HSC number was still apparent ( Figure 1G ) . Thus , although miR-146a KO mice contain normal levels of HSPCs when they are young , as they age , they go through a hypercellular stage and then eventually start to lose bone marrow HSCs and differentiated cells , leading to HSC exhaustion and bone marrow failure . This suggests to us that a dysregulated HSC differentiation toward the myeloid lineage is taking place in the marrow , accompanied by an increased appearance of HSCs in the spleen , probably as a form of homeostatic compensation . Importantly , we found that chronic inflammatory stimulation with bacterial components in young miR-146a KO mice was sufficient to accelerate the development of the same hematopoietic defects seen during aging of these mice . We stimulated 8-week-old WT and miR-146a KO mice with LPS repeatedly for a month and observed a dysregulated HSC differentiation toward myeloid cells in miR-146a KO mice , compared to WT mice , a phenotype similar to the one that spontaneously occurs during aging of the KO mice ( Figure 1H–J and Figure 1—figure supplement 1E ) . Thus , miR-146a is needed to maintain HSC homeostasis in response to chronic inflammation . This suggests that the stress of chronic inflammation may be the physiologically relevant stimulator of HSC deficiency and myeloproliferative disease in the miR-146a KO mice . Perhaps , this becomes evident in aging miR-146a KO mice not subjected to experimental inflammatory stimulation because of a low-level continual exposure to bacterial materials even in our relatively clean conditions of animal husbandry . Whether true pathogens or commensals might be the inducing agents will require further investigation . In addition to the progressive loss of phenotypically defined HSCs in aging miR-146a KO bone marrow , we found that the functional quality of HSCs deteriorates in mice lacking miR-146a . To assess HSC function , we compared WT and miR-146a KO bone marrow HSCs in their ability to generate the entire hematopoietic repertoire competitively in vivo . Total bone marrow cells from either 6-week-old WT or miR-146a KO mice , both which were CD45 . 2+ , were transplanted along with an equal number of CD45 . 1+ WT bone marrow cells , into lethally irradiated CD45 . 1+ WT recipient mice ( Figure 2A ) . 6 months after transplant , CD45 . 2+ and CD45 . 1+ cells in both the CD45 . 2 WT/CD45 . 1 WT and CD45 . 2 KO/CD45 . 1 WT mice contributed identical proportions of cells in nearly all mature hematopoietic lineages and HSPCs ( Figure 2B ) . Furthermore , when we purified long-term HSCs ( defined by LSK CD150+CD48− ) from WT or miR-146a KO bone marrow for competitive repopulation assay , we again observed a similar contribution of WT or miR-146a KO HSCs to total white blood cells and HSPCs 6 months after transplantation ( Figure 2—figure supplement 1A ) . However , when we extended the experiment past 10 months , we began to observe a decreased contribution of miR-146a KO cells to long-term HSCs and LSK cells , but no reduction in CD45+ cells , in the bone marrow of CD45 . 2 KO/CD45 . 1 WT mice ( Figure 2C , D ) . Interestingly , while WT/WT chimera mice had similar numbers of CD45 . 2+ and CD45 . 1+ HSCs , KO/WT chimera mice showed a specific reduction in only the CD45 . 2+ KO HSCs and a slight elevation of CD45 . 1+ WT HSCs , possibly as a compensation for the loss of KO HSCs ( Figure 2E ) . These data indicate that miR-146a-deficient long-term HSCs from young mice have an intrinsic defect and are more susceptible to depletion compared to WT HSCs in the same environment; however , the intrinsic defect of miR-146a KO HSCs from young mice is rather modest and takes 10 months to become apparent in this transplant setting . 10 . 7554/eLife . 00537 . 005Figure 2 . Progressive functional decline of miR-146a-deficient HSCs . ( A ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages in pretransplanted bone marrow ( BM ) mixtures consisted of equal numbers of CD45 . 2+ WT and CD45 . 1+ WT total BM cells ( blue bar , CD45 . 2 WT:CD45 . 1 WT ) or equal numbers of CD45 . 2+ miR-146a KO and CD45 . 1+ WT total BM cells ( red bar , CD45 . 2 KO:CD45 . 1 WT ) . FACS analysis was performed on the BM mixtures before transplantation to determine the starting ratios of various lineages . ( B ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages in peripheral blood ( PB ) , spleen ( SP ) and BM 6 months after transplantation . Blue bar , CD45 . 2 WT:CD45 . 1 WT , represents mice received CD45 . 2 WT:CD45 . 1 WT BM cells; red bar , CD45 . 2 KO:CD45 . 1 WT , represents mice received CD45 . 2 KO:CD45 . 1 WT BM cells . All donor mice were 6-week-old female and recipient mice were 2-month-old CD45 . 1+ WT female . Ratio of CD45 . 2+ over CD45 . 1+ of BM LSK cells and HSCs ( C ) and total white blood ( CD45+ ) cells ( D ) 10-month after transplantation . ( E ) Number of total HSCs , CD45 . 2+ HSCs and CD45 . 1± HSCs in recipient chimera mice . WT:WT , CD45 . 2 WT:CD45 . 1 WT; KO:WT , CD45 . 2 KO:CD45 . 1 WT . n = 8 for each group . ( F ) – ( K ) A repeat of the above experiment with age-and-sex-matched 4-month-old WT and miR-146a KO female mice . ( F ) Representative FACS plots of BM mixtures before transplantation showing CD45 . 2/CD45 . 1 ratio close to 1 for both WT:WT and KO:WT BM mixtures . ( G ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages before transplantation . ( H ) Percentage of CD45 . 2+ cells of CD45+ peripheral blood nucleated cells at 1 , 2 , 3 and 6 months . ( I ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages in PB , SP , and BM 6 months after transplantation . All ratios of CD45 . 2 WT:CD45 . 1 WT are normalized to 1 . Number of total , CD45 . 2+ and CD45 . 1+ LSK cells ( J ) and HSCs ( K ) in recipient chimera mice . LSK , Lin-cKit+Sca1+; HSC , LSK CD150+CD48−; L−S−K+ , Lin−Sca1−cKit+ . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 00510 . 7554/eLife . 00537 . 006Figure 2—figure supplement 1 . Functional decline of miR-146a-deficient HSCs . ( A ) Related to Figure 2C–E . Equal numbers of BM HSCs ( LSK CD150+CD48− ) were sorted from 8-week-old CD45 . 2+ WT and CD45 . 2+ miR-146a KO mice . Fifty WT or KO HSCs were mixed with 500 , 000 CD45 . 1+ WT total BM cells and transplanted into a CD45 . 1+ WT recipient mouse . Percentage of CD45 . 2+ cells in various lineages in peripheral blood ( PB ) , spleen ( SP ) , and bone marrow ( BM ) was analyzed by FACS 8 months after transplantation . ( B ) – ( E ) Related to Figure 2F–K . Quantification of total number of cells ( including both CD45 . 2+ and CD45 . 1+ cells ) in various lineages in SP and BM of CD45 . 2 WT:CD45 . 1 WT and CD45 . 2 KO:CD45 . 1 WT recipient mice 6 months after transplantation . ( F ) – ( G ) A repeat of the experiment described in Figure 2A with sex- and age-matched 6-month-old WT and miR-146a KO female mice . ( F ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages in BM mixtures before transplantation . ( G ) Ratio of CD45 . 2+ over CD45 . 1+ cells of various lineages in PB , SP , and BM 3 months after transplantation . All ratios of CD45 . 2 WT:CD45 . 1 WT are normalized to 1 . LSK , Lin−cKit+Sca1+; HSC , LSK CD150+CD48−;L⁻S⁻K⁺ , Lin-Sca1⁻cKit⁺ . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 006 To investigate how aging affects the quality of miR-146a KO HSCs , we used 4-month-old , instead of 6-week-old , miR-146a KO and WT mice for competitive repopulation assay . At 4 months of age , miR-146a KO mice display a modestly hypercellular bone marrow and mildly increased phenotypically defined HSCs , but no disease phenotype ( Figure 1—figure supplement 1D ) . Surprisingly , miR-146a KO cells were outcompeted by their WT counterparts as early as the first month after the transplant , and there was a steady decline in their percentage over a period of 6 months ( Figure 2F–H ) . 6 months after transplant , neither group of mice showed signs of pathology , and they had identical levels of total mature hematopoietic cells and HSPCs ( Figure 2—figure supplement 1B–E ) . However , when comparing the contribution of CD45 . 1+ cells vs CD45 . 2+ cells to the total pool , we observed a significant disadvantage of miR-146a KO HSPCs and mature lineages ( Figure 2I ) . This is consistent with the preferential loss of CD45 . 2+ KO LSK cells and HSCs , but not the cotransplanted CD45 . 1+ WT cells , in KO/WT chimera mice ( Figure 2J , K ) . Similar to the progressive decline of HSC number , the functional quality of HSCs became more and more compromised as miR-146a KO mice aged . When the competitive repopulation experiment was carried out with 6-month-old WT and miR-146a KO bone marrow cells , at which time increased myeloid and LSK cells and decreased long-term HSCs were already evident ( Figure 2—figure supplement 1F ) , we observed that miR-146a KO cells were completely overwhelmed by their WT counterparts in the recipient mice , with ratios of about 1 miR-146a KO cell to 10 WT cells for nearly all mature lineages and HSPCs ( Figure 2—figure supplement 1G ) . The defective repopulating ability observed in all mature and HSPC lineages , including the long-term HSCs , in all three major hematopoietic compartments strongly suggests that the defect must originate in the most primitive HSCs . These data indicate that miR-146a-deficiency has a detrimental effect on the quality of HSCs under chronic inflammatory stress . The functional decline of HSCs in the competitive repopulation setting is evident in healthy 8-week-old miR-146a KO mice and becomes significant in 4-month-old mice in the absence of any observable pathology , indicating the physiological importance of miR-146a as a guardian of HSC quality and longevity . To directly examine both hematopoietic-intrinsic defects and extrinsic factors on hematopoiesis in the absence of miR-146a , we performed reciprocal bone marrow transplants , transferring WT bone marrow cells into miR-146a KO recipient mice ( WT to KO ) and KO bone marrow cells into WT recipient mice ( KO to WT ) . WT to WT and KO to KO transplant mice were included as controls . After 5 months , we harvested all groups for analysis . Interestingly , mice of the WT to WT and WT to KO groups had identical levels of HSPCs in their bone marrows and spleens , suggesting that the miR-146a-deficient environment is not sufficient to induce significant HSC abnormalities in WT cells during this time ( Figure 3A–G ) . In comparison , mice in the KO to WT and KO to KO groups both showed twofold to threefold reductions in bone marrow HSCs and myeloid progenitor cells , but not in LSK cells ( Figure 3A–C ) , and about a 10-fold increase in spleen HSPCs ( Figure 3D–G ) . As in the miR-146a germline KO mice , an increased percentage of LSK cells and a decreased representation of long-term HSCs within the LSK fraction was also observed in these transplantation groups , indicating a dysregulated HSC homeostasis ( Figure 3H–J ) . In addition to the HSPC abnormality , mice of the KO to WT and KO to KO groups also developed pathological features recapitulating those seen in aged miR-146a germline KO mice ( Boldin et al . , 2011; Zhao et al . , 2011 ) . By 5 months post-transplant , two out of eight mice in the KO to KO transplant group had succumbed to tumor pathology , including one case of CD4+ T-cell lymphoma in the thymus and one case of kidney tumor , with no tumors observed in any of the other groups ( Figure 3—figure supplement 1A , B ) . Necropsy also showed splenomegaly and pale bone marrows in the KO to WT and the KO to KO groups ( Figure 3—figure supplement 1C , D ) . Myeloproliferation was a prominent feature in mice of the KO to WT and the KO to KO groups , which showed an increase of spleen weight , number of white blood cells , B cells , T cells , and , most dramatically , number of CD11b+ or Gr1+ myeloid cells in their spleens , compared to the WT to WT and the WT to KO groups ( Figure 3—figure supplement 1E–L ) . Increased myeloproliferation/myelopoiesis was also observed in the bone marrow and peripheral blood ( Figure 3—figure supplement 1M , N ) . Moreover , in line with HSC exhaustion and bone marrow failure , mice of the KO to WT and KO to KO groups exhibited hypocellular bone marrow and peripheral cytopenia ( Figure 3—figure supplement 1O–U ) . 10 . 7554/eLife . 00537 . 007Figure 3 . Hematopoietic-intrinsic and hematopoietic-extrinsic contribution to hematopoietic defects . ( A ) – ( J ) Reciprocal bone marrow ( BM ) transplant performed by transferring WT donor BM cells to WT recipient mice ( WT to WT ) , WT donor BM cells to miR-146a KO recipient mice ( WT to KO ) , miR-146a KO donor BM cells to WT recipient mice ( KO to WT ) , and miR-146a KO donor BM cells to miR-146a KO recipient mice ( KO to KO ) . All donor and recipient mice were 8-week-old female mice . Mice were harvested for analysis at the end of 5 months . Quantification of total number of HSPCs in spleen and BM , including LSK CD150+CD48−EPCR+ HSCs ( A ) , LSK ( Lin−cKit+Sca1+ ) cells ( B ) , Lin−cKit+Sca1− cells ( C ) in BM , and LSK CD150+CD48− HSCs ( D ) , LSK cells ( E ) , and Lin−cKit+Sca1− cells ( F ) in spleen . Quantification of percent of HSPCs in spleen and BM , including percent LSK cells in spleen ( G ) , percent of CD150+CD48−EPCR+ HSCs in LSK gate in BM ( H ) , percent of LSK cells in total BM ( I ) , and percent of cKit+Sca1+ in Lin− gate in BM ( J ) . ( K ) – ( O ) . Serial BM transplant performed by first transplanting CD45 . 1+ WT BM cells into either CD45 . 2+ WT or miR-146a KO recipient mice for 2 months , which were then harvested and mixed with CD45 . 2+ WT BM cells for second transplantation into CD45 . 2+ WT recipient mice . Mice received CD45 . 1 WT ( WT ) :CD45 . 2 WT or CD45 . 1 WT ( KO ) :CD45 . 2 WT cells were harvested 6 months later for FACS analysis . ( K ) Schematic diagram of the experimental setup . ( L ) Ratio of CD45 . 1+ over CD45 . 2+ cells of BM HSPCs , including LSK cells , LSK CD150+CD48− HSCs and L−K+S− cells . Ratio of CD45 . 1+ over CD45 . 2+ cells of various lineages in BM ( M ) , spleen ( N ) , and PB ( O ) , including CD45+ , CD19+ , CD11b+ , CD3ε+ and Gr1+ cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 00710 . 7554/eLife . 00537 . 008Figure 3—figure supplement 1 . Hematopoietic-intrinsic and hematopoietic-extrinsic contribution to hematopoietic defects . ( A ) – ( U ) The same reciprocal BM transplant experiment as Figure 3A–J . ( A ) Kaplan–Meier survival curve of the four transplant groups ( WT to WT , WT to KO , KO to WT , and KO to KO ) . ( B ) FACS plot of a CD4+ T-cell lymphoma from thymus of a moribund KO to KO mouse . Representative photographs of spleens ( C ) and bone marrows ( D ) . ( E ) Spleen weights . Quantification of total number of CD45+ ( F ) , CD19+ ( G ) , CD3ε+ ( H ) , CD11b+ ( I ) , and Gr1+ ( J ) cells in spleen , percentage of CD11b+ ( K ) and Gr1+ ( L ) cells in spleen , percentage of CD11b+ cells in BM ( M ) , percentage of CD11b+ cells in PB , number of CD45+ ( O ) , CD11b+ ( P ) , CD19+ ( Q ) , Ter119+ ( R ) cells in BM and number of CD45+ ( S ) , CD19+ ( T ) , and CD3ε+ ( U ) cells in PB by FACS . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 008 These data indicate that it is the miR-146a deficiency in hematopoietic cells that plays the dominant role in determining the phenotype of the miR-146a knockout mouse , because transferring miR-146a-deficient bone marrow cells into WT environment , but not the reciprocal transfer , is sufficient to yield a majority of the phenotypes seen in mice with miR-146a deleted in both hematopoietic and nonhematopoietic cells . However , the miR-146a-deficient environment has a contributory role to the overall enhanced myelopoiesis because WT bone marrow cells in the KO environment gave rise to a mild but statistically significant increase in the number of CD45+ and CD11b+ marrow cells and the percentage of CD11b+ and Gr1+ cells in their spleens , compared to the WT to WT mice ( Figure 3—figure supplement 1K , L , O , P ) . More indicative is that when WT bone marrow cells that were transiently transplanted into miR-146a KO environment for 2 months were competed against WT control bone marrow cells ( Figure 3K ) , they showed a modest advantage in generating myeloid and lymphoid lineages in their spleens , peripheral blood and to a lesser extent bone marrows , whereas bone marrow HSPCs remained unaffected ( Figure 3L–O ) . Overall , we have shown that hematopoietic-intrinsic deficiency of miR-146a plays the dominant role in driving the HSC defects and pathological myeloproliferation while the miR-146a deficient environment also contributes to the overall phenotype . In the absence of the driving force from the miR-146a deficient environment , lethal tumor pathology and HSC functional decline are attenuated or delayed . We have shown that hematopoietic-intrinsic factors play the predominant role in the dysregulated hematopoiesis of miR-146a-deficient mice . But this does not tell us which cell type ( s ) in the hematopoietic compartment might influence aspects of hematopoiesis . Because miR-146a–deficient lymphocytes display a hyperactivated phenotype with dysregulated cytokine production ( Yang et al . , 2012 ) , we first determined whether dysregulation of miR-146a-deficient lymphocytes might contribute to HSC depletion in miR-146a KO mice . To this end , we crossed mice with a targeted deletion of the Rag1 gene , which is required for lymphocyte maturation , with miR-146a KO mice to generate Mir146a−/− Rag1−/− double knockout mice ( miR/Rag1 DKO ) . When WT , miR-146a KO ( miR KO ) , Rag1 KO , and miR/Rag DKO mice were allowed to age for 10 months , the miR KO mice showed significant depletion of HSPCs ( Figure 4A–D ) , a finding consistent with what we have observed previously in an independent examination of WT and miR-146a KO mice ( Figure 1 ) . Specifically , long-term HSCs and LSK cells in the miR KO marrow were reduced to only 3% and 15% of the respective WT levels . In comparison , depletion of HSPCs in miR/Rag1 DKO bone marrow was partially rescued . Long-term HSCs and LSK cells in miR/Rag1 DKO bone marrow rose to 30% and 50% of the Rag1 KO levels , respectively , and the numbers of myeloid progenitor cells and total white blood cells were normal ( Figure 4A–E ) . These data indicate that miR-146a-deficient lymphocytes contribute substantially to the overall HSPC exhaustion . In addition , miR-146a-deficient lymphocytes were shown to be a major driver of the development of bone marrow failure and myeloproliferative disease , because miR/Rag1 DKO mice showed normal bone marrow cellularity and an attenuated splenomegaly and myeloproliferative phenotype ( Figure 4E–L ) . This was also confirmed by histological analysis of femur bones of miR/Rag1 DKO mice , which showed absence of marrow fibrosis , a common and prominent feature of aged miR KO mice ( Figure 4H ) . However , the rescue was not complete because miR/Rag1 DKO mice still had mildly enlarged spleens and a twofold to threefold increase in the number of total white blood cells and myeloid cells compared to Rag1 KO mice , indicating that miR-146a deficiency in myeloid lineages has an intrinsic effect on the development of myeloproliferation ( Figure 4M ) . It is worth noting that the intrinsic effect of miR-146a deficiency in myeloid cells may either be in cis or trans: it could be the proliferating cells responding to a changed intracellular signaling or to factors secreted by myeloid cells acting in an autocrine or paracrine manner . 10 . 7554/eLife . 00537 . 009Figure 4 . MiR-146a-deficient lymphocytes contribute to the HSC defect and myeloproliferation . ( A ) – ( M ) Age-and-sex-matched WT , miR-146a KO ( miR KO ) , Rag1 KO and miR-146a/Rag1 double KO ( miR/Rag1 DKO ) mice were allowed to age to 10-month-old before harvested for analysis . ( A ) Representative FACS plots of CD150+CD48− HSCs of the LSK gate . Quantification of total number of HSCs ( B ) , LSK cells ( C ) , and Lin−cKit+Sca1− myeloid progenitor cells ( D ) in BM . Quantification of total number of CD45+ ( E ) , CD11b+ ( F ) , and Ter119+ ( G ) cells in BM . ( H ) Representative histological pictures ( H&E stain ) of femur bones . Scale bar , 40 μm . ( I ) . Representative photograph of spleens and spleen weight . Total number of CD45+ ( J ) , CD11b+ ( K ) , and Gr1+ ( L ) cells in spleen . ( M ) For comparison , various cell lineages in spleen of only Rag1 KO and miR/Rag1 DKO mice were regraphed from ( J–L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 009 To understand the molecular mechanism responsible for the stress-induced hematopoietic dysregulation , we focused on the main pathway known to be regulated by miR-146a , the NF-κB pathway . We have previously shown that aging miR-146a KO mice display hyperactivated NF-κB activity that is responsible for the development of myeloid malignancy ( Zhao et al . , 2011 ) . To study whether NF-κB may also regulate HSC homeostasis under inflammatory stress , we used a transgenic NF-κB–GFP reporter mouse to monitor NF-κB activity quantitatively and efficiently in various cell types by measuring GFP fluorescence ( Magness et al . , 2004; Lippert et al . , 2009 ) . We first tested whether NF-κB can be activated in LSK cells and HSCs in 2-month-old WT NF-κB-GFP ( WT-GFP ) reporter mouse under steady state and after LPS stimulation . We found that about 6–7% of WT LSK cells and HSCs had basally activated NF-κB activity , as measured by GFP expression . 6 hr after LPS challenge in vivo , the percentage of LSK cells and HSCs with activated NF-κB increased to more than 25% ( Figure 5—figure supplement 1A ) . Interestingly , 8-month-old WT-GFP mice also showed an increased percentage of LSK cells and HSCs with basal NF-κB activation , compared to 2-month-old mice ( Figure 5—figure supplement 1A ) . These data suggest that both LSK cells and HSCs have functional NF-κB–mediated transcription that can be augmented by LPS stimulation and aging . To analyze whether hyperactivated NF-κB activity is a feature in miR-146a KO HSPCs , we bred NF-κB-GFP reporter mice with miR-146a KO mice to generate Mir146a+/+ ( WT-GFP ) , Mir146a+/− ( miRHET-GFP ) , and Mir146a−/− ( miRKO-GFP ) NF-κB-GFP reporter mice . Unperturbed 8-week-old mice of WT-GFP , miRHET-GFP , and miRKO-GFP genotypes showed identical levels of basal NF-κB activity ( Figure 5—figure supplement 1B–D ) . After repeated LPS stimulation , miRKO-GFP mice , in comparison to WT-GFP and miRHET-GFP mice , showed increased percentages of GFP+ cells in various HSPCs ( Figure 5A–C ) and mature cells in their bone marrows , spleens and peripheral blood ( Figure 5—figure supplement 2A–C ) , demonstrating that chronic inflammatory stimulation with bacterial components leads to hyperactivated NF-κB activity in young miR-146a-deficient mice . 10 . 7554/eLife . 00537 . 010Figure 5 . NF-κB regulates HSC homeostasis during chronic inflammation . ( A ) – ( C ) 8-Week-old WT ( WT-GFP ) , Mir146a+/− ( miRHET-GFP ) , and Mir146a−/− ( miRKO-GFP ) NF-κB-GFP reporter mice were subjected to repeated intraperitoneal LPS stimulation ( 3 mg LPS/kg of body weight every other day ) for 1 week . Percent of GFP+ cells in various lineages were quantified by FACS . ( A ) Representative FACS plots of GFP+ white blood cells ( CD45+ ) in spleen ( SP ) , bone marrow ( BM ) , and peripheral blood ( PB ) . Quantification of percent GFP+ cells of HSPCs in bone marrow ( B ) and spleen ( C ) , including CD45+ , LSK cells , HSCs ( LSK CD150+CD48− ) and L−S−K+ myeloid progenitor cells . ( D ) Age- and sex-matched WT , miR-146a KO ( miR KO ) , and miR-146a/p50 double knockout ( miR/p50 DKO ) mice were allowed to age to 8–9 months before harvested for analysis . Quantification of total number of CD45+ , LSK CD150+CD48−EPCR+ HSCs in BM , percent of LSK CD150+CD48−EPCR+ HSCs of total BM , and total number of LSK CD150+CD48−EPCR+ HSCs in spleen by FACS . ( E ) – ( I ) 8-Week-old WT and miR-146a KO ( miR KO ) mice were subjected to repeated low-dose of intraperitoneal LPS stimulation ( 1 mg LPS/kg of body weight ) daily for 3 days . 1 mg of BrdU was injected intraperitoneally daily . BrdU+ and Ki-67+ HSPCs were quantified by FACS . ( E ) Representative FACS histograms of BrdU+ LSK cells and HSCs in BM and BrdU+ LSK cells in spleen . Blue: WT mice; red: miR KO mice . ( F ) Quantification of percent of BrdU+ cells in BM LSK cells , HSCs , and spleen LSK cells . ( G ) Representative FACS histograms of Ki-67+ LSK and HSCs in BM and spleen . Black: negative control; blue: WT BM; red: miR KO BM; green: WT spleen; magenta: miR KO spleen . ( H ) Quantification of Ki-67+ cells in BM LSK cells and HSCs . ( I ) Quantification of number and percent of CD11b+ myeloid cells in spleen and BM . ( J ) Representative FACS plots of BrdU+ or Ki-67+ and GFP+ cells of Lin− , L−S−K+ , LSK and HSC in BM of 8-week-old WT-GFP mice stimulated with LPS ( one dose , 1 mg/kg of body weight ) for 4 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 01010 . 7554/eLife . 00537 . 011Figure 5—figure supplement 1 . NF-κB activity in HSPCs and peripheral blood of NF-κB-GFP reporter mice . ( A ) Regular WT mice and WT NF-κB-GFP reporter mice were challenged with one dose of LPS intraperitoneally ( 2 mg LPS/kg body weight ) . GFP+ cells in various lineages were quantified by FACS 6 hr after injection . Representative FACS plots of 2-month-old regular WT mouse ( without NF-κB-GFP transgene ) with LPS stimulation , 2-month-old WT NF-κB-GFP reporter mouse at basal state , 2-month-old WT NF-κB-GFP reporter mouse with LPS stimulation , and 8-month-old WT NF-κB-GFP reporter mouse at basal state . Percent of GFP+ cells in LSK cells and HSCs were shown . Basal level of percent ( B ) , number ( C ) , and mean fluorescence intensity ( MFI ) ( D ) of GFP+ cells in various lineages in peripheral blood ( PB ) of 2-month-old WT ( WT-GFP ) , Mir146a+/− ( miRHET-GFP ) , and Mir146a−/− ( miRKO-GFP ) NF-κB-GFP reporter mice were quantified by FACS . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 01110 . 7554/eLife . 00537 . 012Figure 5—figure supplement 2 . NF-κB regulates HSPC homeostasis during chronic inflammation . ( A ) – ( C ) Related to Figure 5A–C . 8-Week-old WT-GFP , miRHET-GFP , and miRKO-GFP reporter mice were subjected to repeated intraperitoneal LPS stimulation ( 3 mg LPS/kg of body weight every other day ) for 1 week . Percent of GFP+ cells in various lineages were quantified by FACS . Quantification of GFP+ cells in various lineages , including CD45+ , CD19+ , CD11b+ , Gr1+ , and CD3ε+ , in bone marrow ( A ) , spleen ( B ) , and peripheral blood ( C ) . ( D ) – ( H ) Age- and sex-matched WT , miR KO , p50 KO , and miR/p50 DKO mice were allowed to age to up to 18 months . Mice were harvested as they became moribund or at the end of the experiment . Spleen weight ( D ) , Kaplan–Meier survival curve ( E ) , and incidence of tumors ( F ) . Representative histological images ( H&E staining ) of spleen ( G ) and femur bones ( H ) from 12-month-old female WT , miR KO , p50 KO , and miR/p50 DKO mice . Spleen from a miR-146a KO mouse with myeloid sarcoma was shown . Scale bars , 400 μm for spleens and 40 μm for bones . ( I ) Representative photographs of spleens from Rag2−/− Il2rg−/− mice transplanted with WT , miR KO , or miR/p50 DKO spleen cells . Scale bar , 1 cm . ( J ) 8-Week-old WT and miR KO mice were challenged with one dose of LPS ( 2 mg LPS/kg of body weight ) intraperitoneally for 12 hr . 1 mg of BrdU was injected intraperitoneally . BrdU+ HSPCs were quantified by FACS . Representative FACS plots of BrdU+ LSK cells and LSK CD150+CD48− HSCs in bone marrow of WT and miR KO mice . No BrdU , no BrdU injection; basal state , no LPS injection; LPS stimulation , BrdU , and LPS injection . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 012 To further investigate whether the increased NF-κB activity is responsible for driving HSC depletion in bone marrow , we deleted a main subunit of NF-κB , p50 , to determine whether reduced NF-κB activity in miR-146a/p50 double-knockout ( miR/p50 DKO ) mice might rescue the HSC exhaustion . By 8–9 months of age , miR/p50 DKO mice still exhibited levels of white blood cells and long-term HSCs comparable to that of WT mice , indicating a rescue of the HSC defects ( Figure 5D ) . We have previously shown that p50 deletion rescues the myeloproliferative disease in 6-month-old miR-146a KO mice ( Zhao et al . , 2011 ) . To determine whether myeloid cancer can also be reversed in miR/p50 DKO mice , we aged a cohort of WT , miR KO , p50 KO , and miR/p50 DKO mice to about 1½ years , by which time about 50% of miR-146a KO mice will have developed myeloid cancer . Interestingly , miR/p50 DKO mice still showed reduced spleen weight and prolonged survival ( Figure 5—figure supplement 2D , E ) . The incidence of tumors was also significantly reduced in miR/p50 DKO mice , compared to miR KO mice ( Figure 5—figure supplement 2F ) . However , there were still two cases of splenic myeloid tumors from a total of 46 miR/p50 DKO mice analyzed . Histological analysis of spleens revealed that the majority of miR/p50 DKO spleens , in contrast to the miR KO spleens with frequent myeloid sarcoma , showed preserved lymphoid follicular structures ( Figure 5—figure supplement 2G ) . Furthermore , significant marrow fibrosis seen in aged miR KO mice was also reversed in miR/p50 DKO mice ( Figure 5—figure supplement 2H ) . Transplantation of miR/p50 DKO spleen cells into Rag2−/− Il2rg−/− mice did not result in splenomegaly or myeloid pathology as transplanting miR KO splenic tumor cells did ( Figure 5—figure supplement 2I and Zhao et al . , 2011 ) . Together , these data show that p50-deficiency significantly ameliorates the myeloproliferative disease/myeloid cancer and marrow fibrosis in aging miR-146a KO mice . These findings highlight the importance of the NF-κB pathway , particularly the p50 subunit , as the primary mediator of miR-146a deficiency driven myeloid oncogenesis and marrow failure . However , the modest but significant increase in spleen weight and the occasional occurrence of myeloid tumors and marrow fibrosis in aging miR/p50 DKO mice suggest that other NF-κB subunits or other pathways regulated by miR-146a may also mediate the disease phenotype . Overall , we have shown that chronic inflammation-induced or aging-associated NF-κB activation is responsible for driving HSC exhaustion , myeloproliferative disease , and myeloid cancer . To explore whether the increased proliferation and cycling of miR-146a-deficient HSCs is an underlying cause of accelerated HSC depletion under chronic inflammation , we measured BrdU incorporation and Ki-67 expression in HSCs after LPS stimulation . Surprisingly , when examining 2-month-old WT and miR-146a KO mice , we did not observe a significant difference in the percentage of BrdU+ bone marrow LSK cells and HSCs whether the mice were unperturbed or stimulated with a single LPS injection for 12 hr ( Figure 5—figure supplement 2J ) . However , after repeated LPS stimulation for 3 days , there were significantly higher percentages of BrdU+ and Ki-67+ miR-146a KO LSK cells and HSCs , compared to the WT cells , indicating an increased proliferation ( Figure 5E–H ) . Consistent with increased HSC cycling and myeloid differentiation , miR-146a KO mice started to show a small but statistically significant increase in CD11b+ myeloid cells in their spleens and bone marrows after 3 days of LPS stimulation ( Figure 5I ) . It is also interesting to note that identically gated LSK cells and HSCs in spleen seemed to be less quiescent than their bone marrow counterparts , as indicated by a higher level of Ki-67 expression after LPS stimulation , suggesting that the bone marrow milieu may be a better environment for maintaining HSC quiescence during inflammation ( Figure 5G ) . These data show that chronic inflammation induces increased HSC proliferation and cycling in miR-146a KO mice and underscores the particular importance of miR-146a in modulating HSC activity during chronic inflammation , as opposed to in an acute setting . Because we have shown that NF-κB is activated in many other hematopoietic cells , in addition to HSCs , we next determined whether NF-κB directly regulates HSC proliferation in a cell-autonomous manner or indirectly through cytokines produced by other cells . To this end , we injected WT-GFP mice with LPS and BrdU for 4 hr and measured both NF-κB activation and proliferation simultaneously . Interestingly , we only found a small fraction of cells that were doubly positive for GFP and BrdU or Ki67 , indicating that cells with NF-κB activation and cells that were proliferating represented two largely independent populations ( Figure 5J ) . Within the HSPC subsets , myeloid progenitor cells ( L−S−K+ ) contained the largest fraction of rapidly proliferating cells , whereas HSCs and LSK cells contained the highest percentage of cells with NF-κB activation . This suggests that NF-κB does not appear to directly stimulate HSPC proliferation in a cell-autonomous manner . However , it remains to be seen whether NF-κB regulates other aspects of HSC biology , including cell death , differentiation , and trafficking . Cell intrinsic function of NF-κB in HSCs requires further study . Because of the complexity and the overwhelmingly large list of NF-κB-responsive genes , we wanted to determine whether the HSC defect and the myeloproliferative pathology involve many NF-κB–activated genes acting together or if there are key culprit genes mediating the process . Proinflammatory cytokines , such as IL-6 and TNFα , both of which are highly upregulated upon NF-κB activation , can be potent oncogenic factors , especially in epithelial cancers ( Naugler and Karin , 2008 ) . More importantly , overexpression of IL-6 in bone marrow cells in mice results in myeloproliferative or lymphoproliferative disease ( Brandt et al . , 1990; Hawley et al . , 1992 ) . Furthermore , we have shown that NF-κB appears to regulate HSC proliferation in miR-146a-deficient mice in a non-cell-autonomous manner . Therefore , we focused on the proinflammatory cytokines that were overproduced in miR-146a KO mice , believing that miR-146a-deficient mice likely suffer from a chronic inflammation-driven process , and these proinflammatory cytokines produced by other immune cells upon NF-κB activation may be the direct activators of HSCs . IL-6 and TNFα are both upregulated in aging mR-146a KO mice ( Boldin et al . , 2011; Zhao et al . , 2011 ) . However , upregulation of TNFα was only prominent in miR-146a KO spleens that have developed myeloid sarcomas , but not in ones without overt tumors , suggesting that TNFα upregulation may be a quite late event in oncogenesis . To evaluate the temporal relationship between the increase in inflammatory cytokines and the onset of HSC exhaustion and myeloproliferation , we examined younger , 6- to 10-month-old , mice . At this time , the HSC depletion and myeloproliferative phenotype start to become prominent , but no overt splenic tumors have developed . Similar to what was observed in 18-month-old mice , IL-6 expression was upregulated in both spleen and bone marrow cells of miR-146a KO mice , compared to WT mice . More importantly , miR/p50 DKO mice showed a level of IL-6 comparable to that of WT mice ( Figure 6A , B ) . However , the same trend was not observed for TNFα expression at this age ( Figure 6—figure supplement 1A ) . Furthermore , when bone marrow-derived macrophages ( BMMs ) from 8-week-old mice were stimulated with LPS in vitro , IL-6 , but not TNFα , showed consistently increased induction in miR-146a KO BMMs compared to WT BMMs . Interestingly , the exaggerated IL-6 induction in miR-146a KO BMMs was significantly more prominent with restimulation at a time when WT BMMs showed resistance to endotoxin restimulation ( Figure 6C; Nahid et al . , 2011 ) . This again suggests that miR-146a may be particularly important during chronic and repeated inflammatory challenge . In addition , induction of IL-6 , but not TNFα , in BMMs was highly dependent on p50 . In p50 KO BMMs , IL-6 induction was almost completely abolished ( Figure 6C and Figure 6—figure supplement 1B ) . These data suggest that IL-6 upregulation is an early feature in miR-146a deficiency driven HSC depletion and myeloproliferation and reduction in the IL-6 level may be an important factor underlying the reduced pathology when p50 is deleted . 10 . 7554/eLife . 00537 . 013Figure 6 . NF-κB-regulated pro-inflammatory cytokine IL-6 is an important driver of HSC depletion and myeloproliferation . Gene expression of IL-6 in bone marrow cells ( BM ) ( A ) and spleen cells ( B ) of aging WT , miR KO , and miR/p50 DKO mice measured by RT-qPCR . All mice are age- and sex-matched 6- to 10-month-old female mice . Gene expression of IL-6 ( C ) in bone marrow–derived macrophages ( BMMs ) stimulated in vitro with LPS ( 100 ng/ml ) measured by RT-qPCR . First stimulation with LPS was given at 0 hr and restimulation at 48 hr . BMMs are generated from 8-week-old WT , miR KO , and p50 KO mice . ( D ) 2-month-old WT , miR KO , and miR/IL6 DKO mice after repeated intraperitoneal injection of LPS ( 3 mg LPS/kg body weight on day 1 , 3 , 5 , and 7 ) and mice were harvested on day 8 . Spleen weight and total number of CD45+ , CD45+CD11b+ , and CD45+ Ter119+ cells in spleen were shown . ( E ) – ( G ) Age- and sex-matched WT , miR KO , IL6 KO , and miR/IL6 DKO mice were allowed to age to 6–7 months before harvested for FACS analysis . ( E ) Quantification of number of CD45+ , CD45+CD11b+ , CD45+Gr1+ , and percent of CD11b+ cells in spleen . ( F ) Quantification of number of HSPCs , including LSK cells , LSK CD150+CD48− HSCs and Lin−cKit+Sca1− myeloid progenitor cells , in spleen and percent of HSCs in spleen . ( G ) Quantification of percent of CD150+CD48−EPCR+ HSCs of LSK gate , LSK cells of total BM , and CD11b+ cells of total BM and total number of CD45+ cells in BM . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 01310 . 7554/eLife . 00537 . 014Figure 6—figure supplement 1 . NF-κB–regulated proinflammatory cytokine IL-6 is an important driver of HSC depletion and myeloproliferation . ( A ) Related to Figure 6A , B . Gene expression of TNFα in BM cells and spleen cells of aging WT , miRKO , and miR/p50 DKO mice measured by RT-qPCR . All mice are age- and sex-matched 6- to 10-month-old female mice . ( B ) Related to Figure 6C . Gene expression of TNFα in bone marrow–derived macrophages ( BMMs ) stimulated in vitro with LPS ( 100 ng/ml ) measured by RT-qPCR . First stimulation with LPS was given at 0 hr and restimulation at 48 hr . BMMs are generated from 8-week-old WT , miR KO , and p50 KO mice . ( C ) Unperturbed 2-month-old WT , miR KO , IL6 KO , and miR/IL6 DKO mice were bled for FACS analysis . Quantification of total number and percent of various cell lineages in peripheral blood under steady state . ( D ) and ( E ) Related to Figure 6D . 2-Month-old WT , miR KO , and miR/IL6 DKO mice after repeated intraperitoneal injection of LPS ( 3 mg LPS/kg body weight on day 1 , 3 , 5 , and 7 ) and mice were harvested on day 8 . ( D ) Quantification of total number of CD45+ , CD45+CD11b+ , and CD45+Gr1+ cells in peripheral blood ( PB ) . ( E ) Quantification of total number of CD45+ , CD45+CD11b+ , and CD45+Ter119+ cells in bone marrow ( BM ) . ( F ) and ( G ) ( related to Figure 6E–G . Age- and sex-matched WT , miR KO , IL6 KO , and miR/IL6 DKO mice were allowed to age to 6–7 months before harvested for FACS analysis . ( F ) Quantification of spleen weight and total number of CD45+CD19+ , CD45+CD3ε+ , and CD45+Ter119+ cells in spleen . ( G ) Representative histological pictures ( H&E stain ) of femur bones . Scale bar , 40 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 014 To determine whether IL-6 is a culprit gene downstream of NF-κB-mediating HSC depletion and myeloproliferative disease , we bred miR-146a KO mice with mice knocked out for the Il6 gene . Mir146a−/− Il6−/− ( miR/IL6 DKO ) mice were born at the expected Mendelian frequency and appeared normal . Unperturbed young WT , miR-146a KO , IL6 KO , and miR/IL6 DKO mice showed similar levels of CD11b+ and Gr1+ myeloid cells in peripheral blood , while B cells were slightly reduced in miR/IL6 DKO mice ( Figure 6—figure supplement 1C ) . When LPS was repeatedly administered in young WT , miR-146a KO , and miR/IL6 DKO mice , an accelerated myeloproliferative phenotype was observed in miR-146a KO mice that was largely absent in miR/IL6 DKO mice . Specifically , the significantly increased spleen weight , total white blood cells , myeloid cells , and erythroid precursor cells present in miR-146a KO were all reduced in miR/IL6 DKO mice to levels comparable to those of WT mice ( Figure 6D ) . The same trend was also observed in the peripheral blood ( Figure 6—figure supplement 1D ) . In the bone marrow , repeated LPS stimulation of WT and miR/IL6 DKO mice induced variable hypocellularity . However , the depletion was more severe and consistent in miR KO bone marrow ( Figure 6—figure supplement 1E ) . Furthermore , when WT , miR-146a KO , IL6 KO , and miR/IL6 DKO mice were allowed to age to 6–7 months , myeloproliferation and marrow fibrosis observed in miR-146a KO mice were also significantly reduced in miR/IL6 DKO mice ( Figure 6E and Figure 6—figure supplement 1F , G ) . More interestingly , the disrupted bone marrow HSPC homeostasis and expansion of splenic HSPCs seen in miR-146a KO mice were also partially normalized in miR/IL-6 DKO mice ( Figure 6F , G ) . These data show that in the absence of IL-6 , miR-146a KO mice display a partially reduced HSC defect and less myeloproliferative disease , indicating that upregulation of the NF-κB-responsive proinflammatory cytokine IL-6 is an important driver of the HSC defect and myeloproliferative disease in miR-146a KO mice under chronic inflammatory stress induced by aging or repeated bacterial stimulation . Because a variety of cells have the ability to produce IL-6 and IL-6 in turn has a pleiotropic effect on multiple cell types of the hematopoietic lineage , we wanted to identify the important cellular source and cellular target of IL-6 in mediating HSC defect and myeloproliferative disease in miR-146a KO mice . We have previously shown that miR-146a-deficient lymphocytes are involved in the development of the HSC defect and myeloproliferation ( Figure 4 ) , we asked whether overproduction of IL-6 by miR-146a-deficient lymphocytes represents one potential contributing mechanism . To test this , we first stimulated WT , miR-146a KO , Rag1 KO , and miR/Rag1 DKO mice with LPS and measured IL-6 production in vivo . We found increased IL-6 production in the serum of miR-146a KO mice , compared to that of WT mice . In comparison , the serum IL-6 level in miR/Rag1 DKO mice was only modestly increased in a nonstatistically significant manner , compared to that of Rag1 KO mice ( Figure 7A ) . This indicates that in the absence of lymphocytes , exaggerated IL-6 production is attenuated . However , the modest increase in IL-6 production in miR/Rag1 DKO mice suggests that cells other than lymphocytes , such as myeloid cells shown earlier ( Figure 6C ) , also exhibit enhanced IL-6 production in response to stimulation . To further study the contribution of lymphocytes , we stimulated splenocytes from WT or miR-146a KO mice in vitro with either LPS to activate both T and B cells or a combination of anti-CD3 and anti-CD28 to activate T cells specifically . In both conditions , we observed significantly increased production of IL-6 by miR-146a KO lymphocytes , compared to WT lymphocytes ( Figure 7B , C ) . Lastly , to determine whether IL-6 overproduced by miR-146a KO T cells represents an important contributor to the HSC phenotype , we transplanted total T cells from spleens of WT , miR-146a KO , or miR/IL-6 DKO mice into 10-month-old miR/Rag1 DKO mice to see whether reintroducing miR-146a-deficient T cells will promote bone marrow depletion . Interestingly , compared to miR/Rag1 DKO mice that received WT T cells , mice that received miR-146a KO T cells , but not miR/IL-6 DKO T cells , displayed bone marrow depletion of total white blood cells , myeloid progenitor cells , LSK cells , and long-term HSCs ( Figure 7D ) . In addition , mild leukopenia was also observed only in the mice that received miR-146a KO T cells ( Figure 7E ) . These data have demonstrated the particular importance of miR-146a-deficient T cells , specifically their dysregulated IL-6 production , in driving HSC and bone marrow depletion . However , because miR/Rag1 DKO mice do exhibit depleted HSCs ( Figure 4 ) , other intrinsic and extrinsic contributors of HSC defect besides T cells require further examination . In addition , it remains to be tested whether the compositions and inflammatory properties of T cells , such as percentages of regulatory T cells , Th1 , Th2 , and Th17 cells , from WT , miR-146a KO , and miR/IL-6 DKO mice are different , which may explain their different effects on promoting inflammation and bone marrow depletion . 10 . 7554/eLife . 00537 . 015Figure 7 . Analysis of cellular source and direct cellular target of IL-6 . Serum level of IL-6 measured by ELISA in 2-month-old WT , miR KO , Rag1 KO , and miR/Rag1 DKO mice stimulated with LPS ( 1 mg LPS /kg body weight ) intraperitoneally for 6 hr ( A ) IL-6 concentration measured by ELISA in the culture medium of splenocytes stimulated in vitro with LPS ( 10 μg/ml ) ( B ) or anti-CD3 ( 1 μg/ml ) /anti-CD28 ( 0 . 5 μg/ml ) antibodies ( C ) for 4 days . ( D ) and ( E ) CD3ε+ T cells were purified from spleens of 10-month-old WT , miR-146a KO ( miR KO ) , or miR/IL-6 DKO mice . 4 million T cells per mouse were transplanted into 10-month-old miR/Rag1 DKO mice intravenously . miR/Rag1 DKO mice were harvested 1 month after transplant for FACS analysis of white blood cells and HSPCs of bone marrow ( D ) and/or peripheral blood ( E ) . ( F ) cKit+ cells were purified from bone marrow of 8-week-old CD45 . 1+ WT and CD45 . 2+ WT or miR-146a KO mice . A 1:1 mixture of CD45 . 1 WT/CD45 . 2 WT or CD45 . 1 WT/CD45 . 2 KO cKit+ cells were co-cultured under IL-6 ( 50 ng/ml ) or LPS ( 100 ng/ml ) stimulation for 3 days . Percentages of CD45 . 2+ cKit+ or CD11b+ were analyzed by FACS . ( G ) – ( I ) LSK cells or long-term HSCs ( LSK CD150+CD48− ) were sorted from 8-week-old WT or miR-146a KO mice and were cultured in separate wells with IL-6 ( 50 ng/ml ) or LPS ( 1 μg/ml ) stimulation in the presence of BrdU ( 50 μM ) . After 18 hr , cells were analyzed for cell surface marker expression and BrdU incorporation by FACS . Representative FACS histograms of BrdU status of HSCs or LSK cells . Negative control represents identically gated and stained cells in the absence of BrdU pulse ( G ) . Quantification of percent BrdU+ HSCs or LSK cells under IL-6 ( H ) or LPS ( I ) stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 015 To understand whether IL-6 impacts hematopoiesis through a direct effect on HSPCs , we first stimulated a 1:1 mixture of WT or miR-146a KO CD45 . 2+ cKit+ bone marrow cells and WT CD45 . 1+ cKit+ bone marrow cells with IL-6 or LPS for 3 days . We observed increased percentages of both cKit+ and CD11b+ KO cells in the coculture , showing that miR-146a KO cKit+ cells have a proliferative/survival advantage and enhanced myeloid differentiation over WT cKit+ cells under IL-6 or LPS stimulation ( Figure 7F ) . To further determine the direct effect of IL-6 and LPS on more refined HSPC subsets , we sorted LSK cells and long-term HSCs from bone marrow of young WT and miR-146a KO mice and stimulated them with IL-6 or LPS in the presence of BrdU for 18 hr . While HSCs did not have statistically significant difference in BrdU incorporation , KO LSK cells showed higher percentage of BrdU+ cells than WT LSK cells under IL-6 stimulation ( Figure 7G , H ) . Interestingly , no proliferative differences between WT and KO LKS cells or HSCs were detected in response to a single dose of LPS stimulation at 1 μg/ml concentration ( Figure 7G , I ) . These data suggest that the hematopoietic phenotypes seen in miR-146a KO mice may be mediated by IL-6 acting directly on LSK cells , but not long-term HSCs . Furthermore , the in vivo effect of LPS on miR-146a KO mice in inducing enhanced myeloproliferation and HSPC proliferation is at least partly through stimulating the production of cytokines such as IL-6 . Two of the best-validated miR-146a targets , TRAF6 and IRAK1 , are signal transduction proteins upstream of NF-κB activation . To determine whether increased expression of TRAF6 and/or IRAK1 is responsible for the observed HSC exhaustion in the absence of miR-146a , we first measured whether their expression was derepressed in miR-146a-deficient bone marrow cells . BMMs from miR-146a KO and WT mice were stimulated with LPS for 48 hr and then restimulated with a second dose for an additional 16 hr . The transcript level of TRAF6 showed consistent derepression in miR-146a KO BMMs throughout stimulation , whereas the transcript level of IRAK1 showed perhaps an oscillating pattern but was not consistently higher than that of WT BMMs ( Figure 8A , B ) . Furthermore , when HSPCs from young WT and miR-146a KO mice were analyzed for gene expression , derepression of TRAF6 and IRAK1 were quite modest in bulk bone marrow cells but were highest in long-term HSCs , suggesting that miR-146a may play a particularly important repressive role in HSCs , acting on TRAF6 and IRAK1 ( Figure 8C ) . 10 . 7554/eLife . 00537 . 016Figure 8 . Derepression of TRAF6 , a miR-146a target , is responsible for bone marrow failure . Transcript levels of TRAF6 ( A ) and IRAK1 ( B ) in WT and miR-146a KO ( miR KO ) bone marrow–derived macrophages ( BMMs ) stimulated with LPS , which was added to the culture medium at 0 and 48 hr ( black arrow ) . ( C ) Transcript levels of TRAF6 and IRAK1 in total BM , Lin− BM , and FACS-sorted LSK cells , LSK CD150+CD48− HSCs , L−K+S− cells , and L−K−S+ cells from 8-week-old WT and miR-146a KO mice . Fold change of miR-146a KO over WT cells was graphed . ( D ) and ( E ) BM HSPCs overexpressing luciferase ( MIG-Luc ) , TRAF6 ( MIG-TRAF6 ) , or IRAK1 ( MIG-IRAK1 ) were transplanted into lethally irradiated WT recipient mice . Transduction efficiency was about 50% in all groups as measured by FACS before transplantation . ( D ) Percent of GFP+ cells in transduced HSPCs before transplantation and in peripheral blood of reconstituted mice at month 1 , 2 , 5 , 7 , and 9 were analyzed by FACS . ( E ) Representative photographs of histological analysis ( H&E stain ) of femur bones of MIG-Luc control and MIG-TRAF6 mice harvested 9-month after transplantation . Scale bar , 40 μm . ( F ) – ( I ) BM HSPCs overexpressing luciferase ( MIG-Luc ) or TRAF6 ( MIG-TRAF6 ) were transplanted into lethally irradiated WT recipient mice . Transduced HSPCs were sorted for GFP expression to ensure the transplanted HSPCs were 100% GFP+ . ( F ) Kaplan–Meier survival curve of WT recipient mice reconstituted with BM HSPCs overexpressing luciferase ( MIG-Luc ) or TRAF6 ( MIG-TRAF6 ) . Peripheral blood ( PB ) analysis of MIG-Luc and MIG-TRAF6 mice at 1 month after transplantation . ( G ) Representative photograph of 1:1000 diluted PB in phosphate-buffered saline ( PBS ) . Red blood cells in PB were counted with hemocytometer ( H ) and total number of GFP+CD45+ cells in PB ( I ) were measured by FACS . ( J ) Downregulation of miR-146a in human myelodysplastic syndromes ( MDS ) samples . Expression level of miR-146a in bone marrow samples from healthy donors ( normal ) , MDS and acute myelogenous leukemia ( AML ) patients by Taqman RT-qPCR . RNU48 was used as the normalization gene . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 01610 . 7554/eLife . 00537 . 017Figure 8—figure supplement 1 . Gene expression analysis of miR-146a targets in BM HSPCs . ( A ) Related to Figure 8D . Expression of TRAF6 and IRAK1 in BM HSPCs tranduced with pMIG-Luc , pMIG-TRAF6 , or pMIG-IRAK viruses ( ∼50% GFP+ ) . ( B ) Related to Figure 8C . Gene expression of STAT1 in various BM HSPC lineages from 2-month-old miR-146a KO and WT mice . Fold change of miR-146a KO over WT cells was graphed . DOI: http://dx . doi . org/10 . 7554/eLife . 00537 . 017 To determine the functional consequence of upregulating TRAF6 and IRAK1 , we used a GFP-expressing retroviral vector to overexpress TRAF6 ( pMIG-TRAF6 ) or IRAK1 ( pMIG-IRAK1 ) in bone marrow cells enriched for HSPCs by 5-fluorouracil ( 5-FU ) treatment . A vector expressing an irrelevant protein , luciferase ( pMIG-Luc ) , was used as the control . The vectors can consistently overexpress TRAF6 and IRAK1 by 10-fold and 100-fold , respectively , in bone marrow HSPCs ( with about 50% transduction efficiency ) ( Figure 8—figure supplement 1A ) . After transplantation , we followed the mice for 9 months . Interestingly , the percent of GFP+ cells in nucleated peripheral blood cells transduced with either pMIG-Luc or pMIG-IRAK1 remained stable or increased slightly , whereas the GFP percentage in the pMIG-TRAF6 group declined from over 40% initially to less than 10% by the end of 9 months ( Figure 8D ) . Histological analysis of harvested femur bones revealed that TRAF6-expressing mice had reduced bone marrow cellularity , compared to the control mice ( Figure 8E ) . Given the lack of apparent phenotype in IRAK1-overexpressing mice , we further pursued TRAF6 as the more relevant target of miR-146a in the context of HSC biology . When we repeated the bone marrow transplant with 100% transduced bone marrow cells by sorting the GFP+ cells , mice receiving TRAF6-overexpressing bone marrow cells rapidly succumbed to bone marrow failure ( Figure 8F ) . These mice showed severe anemia and reduced numbers of GFP+ white blood cells in their peripheral blood , compared to the pMIG-Luc control group ( Figure 8G–I ) . These data indicate that upregulation of TRAF6 in HSPCs results in bone marrow failure in mice , thus emphasizing the importance of tightly regulating TRAF6 level in HSPCs . This result suggests that derepression of TRAF6 in miR-146a-deficient HSPCs could be primarily responsible for driving HSC depletion and bone marrow failure . Myelodysplastic syndromes ( MDS ) represent a group of human hematopoietic malignancies that are thought to originate from HSCs ( Nimer , 2008a ) . HSCs in MDS patients have a defect in the ability to differentiate into mature cells , leading to peripheral cytopenia . MDS also have a predilection to progress to bone marrow failure or acute myelogenous leukemia ( AML ) . MiR-146a KO mice recapitulate several key characteristics of MDS , including the decline in function of HSCs , peripheral cytopenia , and the propensity to progress to bone marrow failure and myeloid malignancy . To assess whether miR-146a deficiency might represent a pathogenic feature of MDS , we analyzed the expression of miR-146a in bone marrow samples from healthy donors , MDS , and AML patients . The cohort of unselected MDS , but not AML , samples showed a fourfold reduction in the level of miR-146a ( Figure 8J ) , suggesting that miR-146a deficiency may be involved in MDS pathogenesis ( Starczynowski et al . , 2010; Rhyasen and Starczynowski , 2012 ) . In this study , we have shown that a single microRNA , miR-146a , is a critical regulator of HSC homeostasis following chronic exposure of mice to LPS or during the aging process . Mice lacking miR-146a develop a series of defects that can be accelerated by multiple LPS treatments . Extrapolating from this observation , we suggest that the natural exposure of miR-146a KO mice to bacteria and other infectious agents may be the cause of the ‘spontaneous’ deterioration of HSC function ( bone marrow failure ) , excessive myelopoiesis , and tumor formation that are observed in these animals . It could well be that the excess myelopoiesis seen in normal mice as they age is a consequence of the same process ( Salminen et al . , 2008; Beerman et al . , 2010 ) . One obvious test of these notions , to examine HSC function in germ-free mice , is not ideal because of the many secondary consequences of completely ablating the normal microbiota and will require more subtle approaches . The remarkable circumstance that a single microRNA plays such a crucial role in this process has allowed us to uncover it . Because of the existence of mice lacking miR-146a , we were able to use mouse genetics to elucidate the pathway by which miR-146a functions . We propose the following pathway of action of miR-146a: infections or other perturbations activate NF-κB through TRAF6 and other signal transducers; the basal and NF-κB-enhanced levels of miR-146a then limit the amount of TRAF6 , allowing the system to quickly return to basal NF-κB activity when the danger has passed; the transient NF-κB activation causes a transient increase in IL-6 and the IL-6 acts on the stem and progenitor cells to transiently promote proliferation and myeloid differentiation . Lacking miR-146a , the action of NF-κB is extended , the production of IL-6 is exaggerated , the myelopoietic stress on the HSPCs is extended , and the over time the pathological symptoms of continual myelopoiesis , bone marrow failure , and cancer emerge . This may be a pathway in humans that leads to myelodysplastic syndrome . We have demonstrated the importance of the pathway consisting of miR-146a/TRAF6/NF-κB/IL-6 in regulating HSCs during chronic inflammation at the organismal level . However , because HSCs are bathed in a complex cytokine environment and engaged in a complex interaction with other hematopoietic and nonhematopoietic cells , there are a number gaps in our knowledge that requires further study to fully disentangle which molecules are functioning cell autonomously in HSCs and which ones are working through other hematopoietic cells to indirectly affect HSCs . In addition , we have to be cautious of claims about HSC-intrinsic function due to the fact that HSCs are continuously producing more differentiated cells , which harbor the same genetic alterations and are also subject to the same external stimuli . In our effort to understand the function of each molecule in a cell-specific manner , we have undertaken various in vivo transplant experiments and in vitro cell culture studies . We have shown that miR-146a has an intrinsic function within HSCs because miR-146a-deficient HSCs have an intrinsic defect and are preferentially depleted compared to WT HSCs in the same environment . However , this intrinsic defect is modest . Other miR-146a-deficient hematopoietic cells , especially T cells , have a strong effect on HSCs . In addition , miR-146a-deficient nonhematopoietic cells also contribute to the overall HSC abnormality . TRAF6 is derepressed in miR-146a-deficient HSPCs and overexpression of TRAF6 in enriched HSPCs recapitulates the bone marrow failure phenotype . However , whether TRAF6 is functioning in HSCs or more differentiated cells require further examination . Furthermore , TRAF6 is known to regulate the MAPK and PI3K pathways in addition to NF-κB . Further work will be required to clarify their respective contributions in mediating bone marrow failure . We have also demonstrated the functional activity of NF-κB in HSCs and other hematopoietic cells . However , NF-κB does not appear to regulate HSC proliferation cell autonomously . In fact , in WT mice , NF-κB activation does not correlate with proliferative activity in any stem and progenitor cells . The functional consequence of NF-κB activation within HSPCs remains an interesting unanswered question . What is clear is that NF-κB does a significant amount of its ‘damage’ to HSCs through the proinflammatory cytokine IL-6 , produced by T cells and myeloid cells , among others . IL-6 acts directly on LSK cells to stimulate proliferation and miR-146a–deficient LSK cells are more proliferative than WT LSK cells under IL-6 stimulation . In summary , miR-146a is functional in HSCs , LSK cells , T cells , myeloid cells , as well as nonhematopoietic cells , and deficiency of miR-146a in all these cells leads to the full-blown myeloproliferative disease and HSC exhaustion during chronic inflammatory stress . Based on this study and our previous work ( Boldin et al . , 2011; Zhao et al . , 2011; Yang et al . , 2012 ) , we have shown that a circuitry involving miR-146a/TRAF6/NF-κB is operational in miR-146a-deficient T cells and myeloid cells , leading to IL-6 overproduction . IL-6 in turn directly stimulates LSK cells to proliferate and differentiate . We have also demonstrated derepression of TRAF6 and enhanced activation of NF-κB in miR-146a-deficient HSCs and LSK cells . However , it is not yet clear that the cell-autonomous function of miR-146a in HSCs is through the regulation of TRAF6 and NF-κB and the functional importance of TRAF6 and NF-κB in HSCs requires further clarification . The function of individual miRNAs in the most primitive hematopoietic stem cells is relatively unexplored , except in the case of miR-125 family , which is enriched in the long-term HSCs and can positively regulate HSC number and long-term HSC output ( Guo et al . , 2010; O’Connell et al . , 2010 ) . In this study , we added miR-146a to the list of genes that regulate HSC homeostasis during chronic inflammation and aging . Overexpression of miR-155 also causes pathological myelopoiesis similarly to that caused by deletion of miR-146a . MiR-155 appears to act by inhibiting the negative-acting phosphatase SHIP1 ( O’Connell et al . , 2009 ) . Enforced expression of TRAF6 in HSPCs results in cell death and/or engraftment failure and rapid bone marrow failure , a finding consistent with a previous study ( Starczynowski et al . , 2010 ) . The same study also shows that some key features of 5q-syndrome , a subtype of MDS , seen in TRAF6-overexpressing mice are mediated by IL-6 . It is worth noting that the level of TRAF6 expression in transduced HSPCs is about 10-fold higher than that observed in miR-146a-deficient HSCs , and thus the phenotype of enforced TRAF6 overexpression in HSPCs is more rapid and dramatic , precluding a careful analysis of HSC function and activity . It will be interesting to develop a system to modulate the level of TRAF6 overexpression in a more controlled manner for HSC analysis and to characterize the effects of downregulating TRAF6 genetically or by RNA interference in miR-146a-deficient HSCs . In addition to TRAF6 and IRAK1 , STAT1 and RelB have been identified as direct targets of miR-146a in regulatory T cells and Ly6Chi monocytes , respectively ( Lu et al . , 2010; Etzrodt et al . , 2012 ) . Recent studies have provided strong evidence for roles of both IFNα and IFNγ in regulating HSC quiescence and proliferation ( Essers et al . , 2009; Baldridge et al . , 2010 ) . As a downstream transcription factor of both types of interferons , one might speculate that derepression of STAT1 in miR-146a-deficient HSCs may have a cell-intrinsic role in promoting HSC proliferation and exhaustion . Our preliminary analysis has shown upregulation of STAT1 mRNA in miR-146a-deficient HSPCs ( Figure 8—figure supplement 1B ) . A recent study has also suggested a role of noncanonical NF-κB subunits RelB and NF-κB2 in regulating HSC self-renewal ( Zhao et al . , 2012 ) . It remains to be determined whether other putative targets of miR-146a , such as STAT1 and RelB , also contributes to the HSC defect in the context of miR-146a deficiency . In an effort to characterize the effects downstream of NF-κB activation in the absence of miR-146a , we identified IL-6 as one of the culprit NF-κB-responsive genes . Among the pleiotropic functions , IL-6 and its downstream transcription factor STAT3 are important regulators of emergency granulopoiesis ( Zhang et al . , 1998 , Zhang et al . , 2010 ) . In addition , IL-6 can also activate NF-κB in certain contexts ( Iliopoulos et al . , 2009 ) . So , IL-6 can exert its effect either directly on HSPCs and myeloid cancer cells or indirectly by contributing to the general proinflammatory and proproliferative environment through both STAT3 and NF-κB . This signaling pathway involving NF-κB/IL-6/STAT3 has been demonstrated to be important in the pathogenesis of various epithelial cancers associated with inflammation , especially gastrointestinal cancer and breast cancer ( Naugler and Karin , 2008; Grivennikov et al . , 2009; Iliopoulos et al . , 2009; Rakoff-Nahoum and Medzhitov , 2009 ) . Interestingly , a recent report shows that chronic myelogenous leukemia ( CML ) driven by the classic Bcr-Abl oncogenic fusion protein can be attenuated by genetic deletion of Il6 ( Reynaud et al . , 2011 ) . It will be interesting to identify the genes regulated by NF-κB and STATs that are directly responsible for activating the proliferation and differentiation programs within HSPCs . This study has also identified an important role of dysregulated lymphocytes in driving HSC abnormality and myeloproliferative disease in our model , suggesting that autoreactive lymphocytes can provide extrinsic stimulus to diseases like bone marrow failure and myeloproliferative neoplasms/myelodysplastic syndromes in genetically susceptible hosts . Importantly , systemic autoimmunity has long been observed in some MDS patients ( Nimer , 2008b ) . In addition , given the importance of IL-6 , a known regulator of CD4 T cell differentiation toward either Th17 or Treg cells , it will be interesting to determine whether overproduction of IL-6 contributes to abnormal hematopoiesis by altering the ratio of CD4 T cell subsets . Defining the precise role of various lymphocyte subsets , including Th1 , Th2 , Th17 , and Treg cells , using lineage-specific miR-146a conditional knockout mouse may provide further insight into how different immune cells influence physiological and pathological hematopoiesis . This study provides an insight into the function of miR-146a in regulating HSC proliferation and differentiation under the influence of physiological stressors . Given the multitude of mechanisms involved in downregulating NF-κB activity ( Liew et al . , 2005; Ruland , 2011 ) , these results highlight the critical , nonredundant role of miR-146a in the negative regulation of NF-κB activity during chronic inflammation . In this way , miR-146a acts as a guardian of the functional capabilities and longevity of murine hematopoietic stem cells . In addition , this study also has several other important implications . First , it provides direct evidence that prolonged and uncontrolled inflammation-driven hematopoiesis can ultimately exhaust the HSC pool and lead to myeloid malignancies in genetically susceptible hosts , according nicely with some recent large scale epidemiological studies showing that chronic immune stimulation from past infection or autoimmunity increases the risk of developing myeloid malignancies , including AML , MDS , and myeloproliferative neoplasms ( MPN ) ( Anderson et al . , 2009; Kristinsson et al . , 2011; Hasselbalch , 2012 ) . This study provides a possible molecular basis for these intriguing epidemiological observations and offers a unique experimental system to further explore the cellular and molecular pathways by which infection and autoimmunity can trigger myeloid malignancies; it also provides a system to test potential therapeutic interventions and chemoprevention . Second , miR-146a-deficient mice represent an excellent model to understand the pathogenesis of MDS , a hematopoietic malignancy of older adults ( median age of 70 years ) ( Sekeres et al . , 2008 ) that has consistently shown reduced expression of miR-146a . Lastly , it suggests that chronic inflammation may be a potential cause of the age-related decline in HSC function . Therapeutically , given that many of the cellular and molecular components are important in driving the overall pathology , inhibition of p50 subunit of NF-κB , hyperactivated lymphocytes , IL-6 overproduction , and TRAF6 represent multiple opportunities for therapeutic intervention to disrupt the pathogenic process leading to myeloid malignancies , and combinatorial inhibition may have even greater therapeutic impact . All mice were on a C57BL/6 genetic background and housed under specific pathogen-free condition at the California Institute of Technology . All double knockout mice were made by crossing single knockout mice . Experiments with mice were approved by the Institutional Animal Care and Use Committee of the California Institute of Technology . Mouse harvest , tumor analysis , and tumor transplant into Rag2−/− Il2rg−/− mice were performed as described ( Zhao et al . , 2011 ) . For in vivo and in vitro stimulation , Escherichia coli 055:B5 LPS ( Sigma , St . Louis , MO ) was used . Total bone marrow cells from wild-type , miR-146a−/− , or p50−/− mice were lysed with red blood cell lysis buffer ( Biolegend , San Diego , CA ) and were cultured in DMEM supplemented with 10% ( vol/vol ) Fetal Bovine Serum ( Cellgro , Manassas , VA ) , penicillin and streptomycin , and M-CSF ( 20 ng/ml ) for 6 days . On day 7 , BMMs were stimulated with E . coli 055:B5 LPS ( 100 ng/ml ) for 0 , 2 , 8 , and 24 hr . At 24 hr , BMMs were washed with phosphate-buffered saline ( PBS ) and taken off LPS stimulation for 24 hr . At 48 hr , BMMs were restimulated with 100 ng/ml LPS for additional 16 hr . Spleen , bone marrow , and peripheral blood cells were lysed with red blood cell lysis buffer . Fluorophore- or biotin-conjugated antibodies against CD45 , CD3ε , CD4 , CD8 , CD11b , Gr1 , B220 , CD19 , Ter119 , NK1 . 1 , cKit , Sca1 , CD48 , CD150 , EPCR , and Ki-67 ( Biolegend , San Diego , CA or eBioscience , San Diego , CA ) were used for staining . BrdU staining was performed with BrdU staining kit from BD Biosciences . Cells were analyzed on a MACSQuant9 or MACSQuant10 Analyzer ( Miltenyi , Auburn , CA ) for both percentage and cell number . Data analysis was performed with FloJo software ( TreeStar , Ashland , OR ) . Hematopoietic stem and progenitor cell sorting was performed by first depleting lineage+ bone marrow cells with magnetic beads ( Miltenyi ) and then stained with indicated antibodies before sorted on a FACSAria machine ( BD , Franklin Lakes , NJ ) . Analysis of BrdU incorporation and Ki-67 expression on purified LSK cells and HSCs in vitro was done with carrier cells to minimize cell loss during staining and permeabilization as previously described ( Mayle et al . , 2013 ) . Total RNA was extracted with TRIzol reagent ( Invitrogen , Carlsbad , CA ) from spleen or bone marrow cells after red blood cell lysis . cDNA was synthesized using iScript cDNA synthesis kit ( Bio-Rad ) followed by SYBR Green-based quantatitive PCR ( Quanta Biosciences , Gaithersburg , MD ) . Rpl32 was used as the normalization gene . MiRNA detection was performed with Taqman RT-qPCR probes ( Life Technologies , Carlsbad , CA ) . Sno202 was used as the normalization gene . Specific cytokines , including IL-6 and TNFα , were measured in cell culture medium or mouse serum by ELISA according to manufacturer’s protocol ( eBioscience ) . Organs were fixed in 10% neutral-buffered formalin immediately after necropsy . After fixation , organs was embedded in paraffin and processed for hematoxylin and eosin ( H&E ) staining . The histopathological analysis was performed by a board-certified hematopathologist . All figures were graphed as mean ± standard error of the mean ( SEM ) . Student t-test and Kaplan–Meier survival analysis were performed using GraphPad Prism software . In all figures , * denotes p<0 . 05 , denotes p<0 . 01 , * denotes p<0 . 001 .	Hematopoietic stem cells are cells that both renew themselves and develop into any type of blood cell , including red blood cells and the several classes of immune cells . When an injury or infection occurs , it is vital that hematopoietic stem cells replenish themselves in addition to developing into the new blood cells that are needed to help the body recover . Injury and infection also lead to the inflammatory response: tissue becomes inflamed as cytokines and other molecules are released at the site of the damage to help maintain the body’s immunity . It is thought that inflammatory molecules directly affect the rate at which stem cells become immune cells , with the protein NF-κB having an important role , but the details of this process are not fully understood . To explore the connections between hematopoietic stem cells and the inflammatory response , Zhao et al . bred mice that do not produce a type of RNA called microRNA-146a . In wild-type mice , this RNA would inhibit the production of NF-κB , so the mutant mice have abnormally high levels of NF-κB . They found that the rate at which stem cells were being converted into immune cells in the mutant mice was so high that the stores of stems cells became exhausted , which was very detrimental to the health of the mice . They also went on to identify the signaling pathways that microRNA-146a influences in order to maintain supplies of stem cells and an adequate inflammatory response in healthy mice . Zhao et al . also studied individuals with human myelodysplastic syndrome , a severe blood disorder that is associated with faulty hematopoietic stem cells , and found that these individuals produce relatively little microRNA-146a . The establishment of a link between microRNA-146a and having an adequate level of hematopoietic stem cells could have implications for human health , given the importance of these cells in both the aging process and the immune response .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "immunology", "and", "inflammation" ]	2013	MicroRNA-146a acts as a guardian of the quality and longevity of hematopoietic stem cells in mice
Multicellular development produces patterns of specialized cell types . Yet , it is often unclear how individual cells within a field of identical cells initiate the patterning process . Using live imaging , quantitative image analyses and modeling , we show that during Arabidopsis thaliana sepal development , fluctuations in the concentration of the transcription factor ATML1 pattern a field of identical epidermal cells to differentiate into giant cells interspersed between smaller cells . We find that ATML1 is expressed in all epidermal cells . However , its level fluctuates in each of these cells . If ATML1 levels surpass a threshold during the G2 phase of the cell cycle , the cell will likely enter a state of endoreduplication and become giant . Otherwise , the cell divides . Our results demonstrate a fluctuation-driven patterning mechanism for how cell fate decisions can be initiated through a random yet tightly regulated process . One of the fundamental questions in developmental biology is how patterns of specialized cell types are formed de novo from a field of identical cells . Wolpert’s French flag model proposes that a group of identical cells differentiate into different cell types based on threshold concentrations of a morphogen gradient ( Wolpert , 1996 ) . Each cell responds to the morphogen individually by expressing specific sets of downstream genes determined by the concentration sensed . This model has successfully explained the formation of various animal tissue patterns ranging from Bicoid anterior-posterior patterning in Drosophila to BMP dorsal-ventral axis patterning in Xenopus ( Eldar et al . , 2002; Houchmandzadeh et al . , 2002; Kondo and Miura , 2010; Spirov et al . , 2009; Tucker et al . , 2008 ) . In plants , traditional morphogens have yet to be observed , although it has been argued that the phytohormone auxin acts as an atypical morphogen that is actively transported to regulate plant morphogenesis ( Bhalerao and Bennett , 2003 ) . In contrast to the morphogen gradient paradigm , many patterning phenomena seem to lack specific localized signaling cues . In these cases , it is not known how identical cells become slightly different from their neighbors to initiate the patterning process . Theoretical approaches suggest a role for small differences of key transcriptional regulators , generated for example by stochastic fluctuations ( Collier et al . , 1996; Hülskamp and Schnittger , 1998; Hülskamp , 2004; Meinhardt and Gierer , 1974; Turing , 1952 ) . In these models , subtle initial differences between identical neighboring cells in activators and inhibitors are amplified and solidified through regulatory feedback loops and cell-to-cell communication to establish different cell fates ( Kondo and Miura , 2010; Meyer and Roeder , 2014 ) . For instance , in a computational model of lateral inhibition where Notch and Delta mutually inhibit one another in the same cell , small stochastic changes in Notch or Delta can flip a switch between cell identities ( Sprinzak et al . , 2010 ) . Subtle concentration changes in Notch or Delta may change a cell’s signaling ability and either push cells into a sending state ( i . e . high Delta/low Notch ) or a receiving state ( i . e . high Notch/low Delta ) . These changes subsequently are amplified through cell-to-cell Notch-Delta signaling to create ordered patterns ( Collier et al . , 1996; Formosa-Jordan and Ibañes , 2014; Sprinzak et al . , 2010 ) . While manipulating Notch-Delta levels in individual mammalian cells supports this model ( Matsuda et al . , 2015; Sprinzak et al . , 2010 ) , these dynamic fluctuations are difficult to detect during tissue patterning within a multicellular system . A similar lateral inhibition model has been proposed to explain trichome ( i . e . hair cell ) spacing in plants ( Digiuni et al . , 2008; Hülskamp and Schnittger , 1998; Hülskamp , 2004; Meinhardt and Gierer , 1974 ) . In these trichome models , initially identical cells can acquire subtle differences through brief stochastic fluctuations of transcriptional activators . These activators amplify both their own expression and the expression of faster-diffusing transcriptional repressors that move to the neighboring cell to create a non-random distribution of trichomes , following a Turing-like model ( Hülskamp , 2004; Meinhardt and Gierer , 1974; Turing , 1952 ) . Several transcriptional regulators needed for trichome patterning have been identified that support this model ( Bouyer et al . , 2008; Greese et al . , 2014; Hülskamp and Schnittger , 1998; Hülskamp , 2004; Schellmann et al . , 2002 ) . However , the stochastic fluctuations of these genes remain to be observed in vivo during trichome development . Most biological examples of stochasticity focus on how noise is buffered during development , suggesting that multiple species have evolved genetic regulatory mechanisms to offset the potentially detrimental effects of noisy gene expression ( Abley et al . , 2016; Arias and Hayward , 2006; Besnard et al . , 2014; Heisler et al . , 2005; Houchmandzadeh et al . , 2002; Howell et al . , 2012; Jönsson et al . , 2006; Meyer and Roeder , 2014; Raj et al . , 2010; Reinhardt et al . , 2003; Smith et al . , 2006 ) . However , a few studies have demonstrated the importance of stochasticity in creating the correct distribution of phenotypes within a population of cells . For instance , during Drosophila retinal development , the transcriptional regulator spineless stochastically turns on or off to generate a proportional but randomly distributed population of photoreceptor subtypes ( ~30% ultraviolet/blue sensitive and ~70% ultraviolet/green sensitive; Wernet et al . , 2006 ) . Without the stochastic dynamics of spineless expression , all cells adopt the same fate ( Wernet et al . , 2006; Johnston and Desplan , 2014 ) . Similarly , a stochastic Markov model illustrates how a tumor can maintain phenotypic equilibrium between different cancer cell subpopulations . In this model , isolated cancer subpopulations will return to their respective proportions over time through stochastic interconversions ( Gupta et al . , 2011 ) . These studies suggest that stochasticity can help different cell populations to reach or maintain the correct phenotypic equilibrium . During the development of Arabidopsis thaliana’s outmost floral organ , the sepal , equivalent epidermal cells in the primordium differentiate to produce a scattered pattern of giant cells that are interspersed between smaller cells ( Figure 1A–F; Roeder et al . , 2010 , 2012; Tauriello et al . , 2015 ) . The sepal is a useful model system because the giant cell patterning process can be live imaged from the earliest stages of initiation through giant cell differentiation . At maturity , giant cells are approximately one-fifth the length of the sepal and form when an epidermal cell undergoes multiple rounds of endoreduplication , an alternative cell cycle in which a cell replicates its DNA without undergoing mitotic division ( Figure 1C–G; Roeder et al . , 2010 ) . Mature sepals typically contain the same proportion of giant cells relative to small cells , although their spatial distribution varies from sepal to sepal and giant cells may even form adjacent to one another ( Figure 1C–F ) . The correct proportion of giant cells and small cells is needed to control the curvature of the sepal; when the proportion of giant cells is altered , sepals are unable to enclose and protect the developing floral organs ( Roeder et al . , 2010 , 2012 ) . Thus , we ask how giant cell patterning initiates and reproducibly produces the correct proportion of giant cells for proper sepal curvature ? 10 . 7554/eLife . 19131 . 003Figure 1 . The scattered pattern of giant epidermal cells . ( A ) An image of a wild-type ( WT ) Arabidopsis thaliana flower . The sepals ( s ) are the outermost leaf-like floral organs . ( B ) SEM image of developing sepals on young flower buds . The three flowers in the middle are in approximately the same orientation and stages as the live imaged sepals . Live images typically start with sepals at the youngest stage shown , exemplified by the center flower ( * ) . ( C–F ) SEM images of mature wild-type sepals . Each sepal exhibits variations in the arrangement of giant cells . Giant cells are false colored in red using Photoshop . Magnified view of E shown in F . Scale bars in B , 30 µm and in C–F , 100 µm . ( G ) A cell cycle diagram depicting the mitotic cell cycle and the endoreduplication cycle ( endocycle ) . During the mitotic cycle , a new 2C cell will enter Gap 1 ( G1 ) . In G1 , the cell will increase its size in preparation for DNA synthesis ( S ) , where it will then become 4C . After S phase , the cell will enter Gap2 ( G2 ) , where it will continue to grow in size and produce more protein in preparation for mitosis ( M ) . Completion of mitosis will result in the formation of two 2C daughter cells , which will then re-enter the mitotic cycle . Alternatively a cell may endocycle ( E ) , where a cell will go through G1 , S , G2 but bypass M to form a polyploid cell . Note that giant cells are 8C and higher polyploid epidermal cells that form through endoreduplication . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 003 We have previously shown that giant cells do not form on the sepal epidermis in plants with loss-of-function mutations in Arabidopsis thaliana MERISTEM LAYER1 ( ATML1; Roeder et al . , 2012 ) , which encodes a class IV homeodomain-leucine zipper transcription factor ( Lu et al . , 1996; Nakamura et al . , 2006; Schrick et al . , 2004 ) . Previous research has indicated that ATML1 is necessary for establishing the epidermal cell layer during early embryogenesis ( Lu et al . , 1996; Roeder et al . , 2012; Sessions et al . , 1999; Takada and Jürgens , 2007 ) . Plants doubly mutant for atml1 and its closely related paralog , protodermal factor 2 , lack an epidermal layer and are thus seedling lethal ( Abe et al . , 2003; Ogawa et al . , 2015 ) . Conversely , ectopic expression of ATML1 results in inappropriate differentiation of epidermal cell types in the inner cell layers of cotyledons ( Peterson et al . , 2013; Takada et al . , 2013 ) . This result suggests that expression of ATML1 can promote cells to adopt epidermal-specific cell identity in tissues other than the epidermis . ATML1 is required for the formation of giant cells; however , only a subset of cells expressing ATML1 become giant in the Arabidopsis sepal epidermis . This raises the question of what patterning mechanism could lead to a scattered pattern of giant cells interspersed between smaller cells . Here , we use live imaging , quantitative image analyses and computational modeling to demonstrate that fluctuations in the concentration of the transcription factor ATML1 initiate the pattern of giant and small cells in the Arabidopsis sepal . To determine how ATML1 specifies giant cells when it is expressed in every cell , we overexpressed ATML1 in the epidermis by approximately five-fold by using the PROTODERMAL FACTOR1 ( PDF1 ) promoter ( pPDF1::FLAG-ATML1; Figure 2A and G; Abe et al . , 2001 , 2003 , 2003; San-Bento et al . , 2014 ) . ATML1 overexpression lines produced sepals almost entirely covered in giant cells ( Figure 2A ) . Since giant cells endoreduplicate ( 16–32C in ploidy; Roeder et al . , 2010 ) , we tested whether ATML1 overexpression also induced endoreduplication . As expected , the proportion of highly endoreduplicated epidermal nuclei from ATML1 overexpression line sepals increased ( Figure 2I , red bars ) . These sepals contained a greater proportion of 16C and 32C giant cells than wild type , and on occasion a few cells even underwent an additional endocycle ( 64C; Figure 2I , red bars ) . In addition , we have previously demonstrated that giant and small epidermal cells can be distinguished with two molecular markers ( Roeder et al . , 2012 ) . To test whether our ATML1 overexpression line sepals confer giant cell identity , we crossed them with plants expressing the giant and small cell markers . In these crossed sepals , the giant cell marker was expressed in almost every epidermal cell and the small cell marker was expressed only in a few remaining small cells ( Figure 2J and K ) . To validate that ATML1 alone is sufficient to drive giant cell formation , we induced ATML1 expression in inflorescences using an ATML1 estradiol-inducible line . Ectopic giant cells formed on the sepal five days after being treated with 10 µM estradiol ( Figure 2—figure supplement 1 ) . Overall , these results suggest that high levels of ATML1 are sufficient to induce sepal epidermal cells to adopt giant cell identity and can force a deterministic all-giant cell pattern . 10 . 7554/eLife . 19131 . 004Figure 2 . ATML1 levels influence the quantity of giant cells that form on the sepal . ( A–F ) SEM images of sepals from an ATML1 genetic dosage series . Giant cells are false colored in red . ( A ) ATML1 overexpression line that is homozygous for the pPDF1::FLAG-ATML transgene . ( B ) ATML1 overexpression line that is hemizygous for the pPDF1::FLAG-ATML1 transgene . ( C ) ATML1 overexpression line hemizygous for the pPDF1::FLAG-ATML1 transgene crossed into a atml1–3 mutant background . ( D ) Wild type . ( E ) atml1–3/+ heterozygous mutant . ( F ) atml1–3 homozygous mutant . ( G ) qPCR on inflorescences from dosage series verifying that ATML1 mRNA levels vary between lines as expected . Fold change is calculated as the average of three biological replicates . Error bars represent the extended standard deviation . ( H ) Quantification of the average number of giant cells per sepal in ATML1 dosage series using semi-automated image processing . Giant cells are defined as cells with an area larger than 4000 µm2 . Error bars represent the standard error of mean , n = 3 sepals per genotype , with each pooled genotype having >1000 cells analyzed . ( I ) Ploidy of epidermal cells in sepals of the ATML1 dosage series determined by flow cytometry . Inset shows percentage of high ploidy nuclei . Average of 3 biological replicates with >40 , 000 nuclei analyzed per replicate; error bars represent standard error of mean . Note that epidermal cells include a large number of 2C and 4C cells on the back ( adaxial ) side of the sepal in all genotypes , which are not affected by ATML1 overexpression . ( J–K ) Confocal maximum intensity projection image of a wild-type ( J ) and ATML1 overexpression ( K ) sepal expressing the giant ( 3xvenus , nuclear localized , blue ) and small cell ( GFP , ER localized , green ) molecular markers . Cell walls are stained with propidium iodide ( PI , red ) . In the ATML1 overexpression sepal ( K ) , the giant cell marker is expressed in almost every cell and the small cell marker is extremely reduced . Note: Margin cells at the edges of the sepals are distinct cell types that are not affected by ATML1 . Scale bars in A–F , 100 µm . T-tests were performed between genetically altered dosage series and wild-type sepals . p-value ≤ 0 . 05 marked with * , p-value ≤ 0 . 01 marked with ** , and non-significant denoted by ns . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 00410 . 7554/eLife . 19131 . 005Figure 2—figure supplement 1 . ATML1 estradiol inducible transgenic plants form ectopic giant cells five days after application of 10 µM estradiol . ( A ) A confocal image of an untreated ATML1 estradiol-inducible stage 10 flower expressing an ATML1 transcriptional marker ( proATML1-nls-3XGFP ) . Note that ATML1 transcriptional reporter is only expressed in the outermost epidermal layer . The front sepal contains approximately 17 giant cells . ( B ) A confocal denoised image of a 10 µM estradiol treated ATML1 estradiol-inducible stage 10 flower expressing the ATML1 transcriptional marker . Note that now the transcriptional reporter is being expressed in multiple cell layers , suggesting that ATML1 was successfully induced . The front sepal contains approximately 30 giant cells . ( C ) Quantification of the number of giant cells for untreated ( n = 7 ) versus 10 µM estradiol treated ( n = 7 ) stage 8–10 sepals . On average , estradiol treated sepals form more giant cells than their untreated counterparts . T-tests were performed between untreated and estradiol treated sepals . p-value ≤ 0 . 05 marked with * . Inflorescences were treated with estradiol on days 1–3 and then imaged on day 5 . Associated with Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 005 Since ATML1 is expressed in every epidermal cell ( Abe et al . , 2003; Lu et al . , 1996; Roeder et al . , 2010 , 2012; Sessions et al . , 1999 ) and ATML1 overexpression leads to an ectopic all-giant cell phenotype , we wondered whether epidermal cell identity specification is sensitive to the dosage of ATML1 . We altered levels of ATML1 genetically to test whether that would change the proportion of giant cells in the sepal epidermis ( Figure 2A–F ) . First , we reduced levels of ectopic ATML1 expression by crossing our ATML1 overexpression line with wild-type plants , resulting in plants containing only one copy of the ATML1 overexpression transgene . These hemizygous plants formed ectopic giant cells , but fewer than the homozygous overexpression lines , and had more small cells ( Figure 2B ) . To reduce ectopic ATML1 levels further , we crossed ATML1 overexpression hemizygotes into an atml1–3 mutant background , removing endogenous ATML1 expression . This resulted in plants with even fewer ectopic giant cells and more small cells ( Figure 2C ) . To test dosage dependency further , we examined atml1–3 heterozygous mutant plants . These plants had fewer giant cells than WT but more than atm1–3 homozygous mutants ( Figure 2D , E and F ) . We verified through qPCR that inflorescences from each of these ATML1 dosage genotypes expressed different amounts of ATML1 as expected ( Figure 2G ) . Additionally , we used flow cytometry to quantify endoreduplication and semi-automated image processing to measure cell size ( Figure 2H and I; Cunha et al . , 2010; Roeder et al . , 2010 ) . Each dosage genotype exhibited proportional changes in ploidy and cell size . Together , these results suggest that ATML1 influences giant cell formation in a dosage-dependent manner , where the amount of ATML1 expressed will determine the proportion of giant cells that form in the sepal . The dosage dependency of ATML1 suggests that the level of ATML1 expression in each sepal is critical for establishing giant cell and small cell patterning . Furthermore , moderate overexpression of ATML1 prompts only some cells to become giant ( Figure 2A–F ) , suggesting that either cells exhibited varying responses to the same ATML1 concentration or that ATML1 concentrations varied between cells . To quantify ATML1 levels in individual cells during sepal development and distinguish between these possibilities , we created a mCitrine-ATML1 fusion protein reporter ( pATML1::mCitrine-ATML1 ) and transformed it into atml1–3 mutant plants ( Figure 3 ) . This reporter expresses mCitrine-ATML1 under the putative native ATML1 promoter and 3’ UTR . We recovered two independent transgenic lines that fully rescue the atml1–3 loss-of-giant cell mutant phenotype ( Figure 3A–D; Materials and methods ) . Both lines exhibited similar behavior , thus we focused our analysis on one of them . Overall , these results suggest that our mCitrine-ATML1 fusion protein functions similarly to endogenous ATML1 ( Figure 1C–F; Figure 3A–D ) . 10 . 7554/eLife . 19131 . 006Figure 3 . mCitrine-ATML1 expression is variable from cell to cell in the sepal but uniform in the meristem . ( A ) SEM image of a wild-type ( Col ) sepal . ( B ) SEM image of an atml1–3 mutant sepal . Note that atml1 mutants exhibit a lack-of-giant-cell phenotype . ( C–D ) SEM images showing that the pATML1::mCitrine-ATML1 transgene rescues the lack-of-giant-cell phenotype normally exhibited by the atml1–3 mutant . Additionally , both the number and spacing pattern of giant cells appear similar to wild type ( A ) . Giant cells in ( A–D ) are false colored red . ( E ) Confocal denoised images of three floral meristems expressing pATML1::mCitrine-ATML1 ( white ) . ( F ) Heat maps of mean normalized concentration levels of mCitrine-ATML1 expression in the floral meristems . ( G ) Confocal denoised images of three young sepal primordia expressing pATML1::mCitrine-ATML1 ( white ) ( right most sepal is shown later in Figure 4—figure supplement 2 as time 0 hrs of the 3rd mCitrine-ATML1 reporter sepal ) . ( H ) Heat maps of mean normalized concentration levels of mCitrine-ATML1 expression in the young sepal primordia . ( I ) Dot plot of the coefficients of variation ( CV ) of normalized fluorescent protein concentration in each sample . The CV of mCitrine-ATML1 in nuclei of young developing sepals is higher than in nuclei of floral meristems . The high CV is specific to mCitrine-ATML1 as VIP1-mCitrine ( pVIP1::VIP1-mCitrine ) , AP2-2XYpet ( pAP2::AP2-2XYpet ) and a SEC24A transcriptional reporter ( SEC24::H2B-mGFP ) have lower CVs in young sepals . n = 3 for each genotype . ( J ) Histograms of normalized mCitrine-ATML1 concentrations for sepals ( from H; red ) and meristems ( from F; blue ) . Both histograms show a unimodal distribution , however the distribution of ATML1 concentrations in single cells is broader in the sepal than in the meristem . Scale bars in A–D 100 µm; E and G , 10 µm . The number of cells analyzed for mCitrine-ATML1 meristems from left to right: n = 102 , 136 and 82 . The number of cells analyzed for each mCitrine-ATML1 sepal primodium in order from left to right: n = 91 , 48 and 142 . Denoised images and corresponding heat maps for pSEC24A::H2B-GFP , VIP1-mCitrine and AP2-2XYpet sepals are shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 00610 . 7554/eLife . 19131 . 007Figure 3—figure supplement 1 . The transcriptional reporter SEC24A:: H2B-GFP and the fusion proteins VIP1-mCitrine , and AP2-2XYpet are uniformly expressed in the developing sepal . ( A ) Confocal denoised images of three developing sepals expressing pSEC24A::H2B-GFP . ( B ) Heat maps of normalized mean concentration levels of pSEC24A::H2B-GFP expression in the developing flowers . ( C ) Confocal denoised images of three developing sepals expressing pVIP1::VIP1-mCitrine . ( D ) Heat maps of normalized mean concentration levels of pVIP1::VIP1-mCitrine expression in the developing flowers . ( E ) Confocal denoised images of three developing sepals expressing pAP2::AP2-2XYpet . ( F ) Heat maps of normalized mean concentration levels of pAP2::AP2-2XYpet expression in the developing flowers . Scalebars: A , 10 µm; C , 20 µm; E , 20 µm . pSEC24A::H2b-GFP , pVIP1::VIP1-mCitrine and pAP2::AP2-2XYpet are ubiquitously expressed in multiple cell layers . To make all three genotypes comparable to mCitrine-ATML1 flowers , only nuclei in the epidermal cell layer were used for the analysis . The number of cells analyzed for each pSEC24A::H2B-GFP sepal primordium from left to right: n = 145 , 215 and 232 . The number of cells analyzed for each VIP1-mCitrine sepal primordium from left to right: n = 73 , 80 and 180 . The number of cells analyzed for each AP2-2XYpet sepal primordium from left to right: n = 152 , 160 and 262 . To make all three genotypes comparable to mCitrine-ATML1 flowers , only nuclei in the epidermal cell layer were used for the analysis . Associated with Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 007 To quantify mCitrine-ATML1 fluorescence in each epidermal cell of early developing sepals and floral meristems , we developed and implemented an image analysis pipeline ( Box 1; Box 1—Figure 1 ) . We observed that in the developing sepal , mean normalized mCitrine-ATML1 concentrations differ between individual nuclei ( Figure 3G–J; sepals show a mean coefficient of variation ( CV ) of approximately 0 . 2 ) . Conversely in the floral meristem , which does not form giant cells , mCitrine-ATML1 concentrations are more uniform ( Figure 3E–F and I–J; meristems show a mean CV of approximately 0 . 1 ) . In particular , we can see that although unimodal , the distribution of ATML1 concentrations in individual nuclei is broader in the sepal than in the meristem , both for lower and higher values ( Figure 3J ) . This suggests that ATML1 concentration behaves differently depending on the developmental context . To see whether other genes also exhibit variable expression similarly to mCitrine-ATML1 in the developing sepal nuclei , we measured the expression of two fluorescently-tagged transcription factors , VIP1-mCitrine ( pVIP1::VIP1-mCitrine ) and AP2-2XYpet ( pAP2::AP2-2XYpet ) , and the SEC24A transcriptional reporter ( pSEC24A::H2B-GFP ) . VIP1 is a mechano-sensitive transcription factor that localizes to the nucleus upon hypo-osmotic treatment ( Tian et al . , 2004; Tsugama et al . , 2016 ) and AP2 is a master regulator of floral organ identity that is expressed in sepals ( Wollmann et al . , 2010 ) . SEC24A is a ubiquitously expressed CopII vesicle-coat protein that is involved in vesicle trafficking from the ER to the Golgi and has been previously reported to influence giant cell formation on the sepal ( Qu et al . , 2014 ) . We found that mCitrine-ATML1 concentrations in the sepal were approximately twice as variable as the other reporters ( Figure 3I , Figure 3—figure supplement 1; VIP1 sepals show a mean CV of approximately 0 . 12; AP2 sepals show a mean CV of approximately 0 . 14; SEC24A sepals show a mean CV of approximately 0 . 12 ) , suggesting that varying expression levels in sepal epidermal cells is not a common feature observed for every gene . Since ATML1 levels differ among cells and higher ATML1 levels increase the proportion of giant cells in the sepal , we hypothesized that in wild-type sepals ATML1 levels fluctuate in all epidermal cells , with only some cells passing a threshold to promote giant cell fate . According to this hypothesis , to become a giant cell , a sepal epidermal cell would need to experience a high concentration of ATML1 above a threshold . In contrast , to become a small cell , a sepal epidermal cell would experience only lower concentrations of ATML1 that fall below the threshold while fluctuating . To determine whether ATML1 fluctuates within single cells , we live imaged the mCitrine-ATML1 reporter in developing sepal primordia every 8 hr until giant cells formed and used our image analysis pipeline to track fluorescence in each nucleus over time ( Figure 4A; Figure 4—figure supplements 1A; 2A and 3A; Box 1; Box 1—Figure 1; Videos 1–4 ) . We found that during early sepal development , epidermal cells not only have varying amounts of mCitrine-ATML1 , but also that mCitrine-ATML1 levels fluctuate within individual cells over time ( Figure 4A–C; Figure 4—figure supplements 1A–C; 2A–C; and 3A–C ) . 10 . 7554/eLife . 19131 . 010Figure 4 . ATML1 fluctuates in sepal epidermal cells to initiate giant cell patterning . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing sepal . Images were taken every 8 hr for 64 hr . ( B ) Heat map showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) mCitrine-ATML1 concentrations tracked over time in cells that became giant ( red ) and cells that divided to stay small ( blue ) . ( D ) mCitrine-ATML1 peak concentration levels in each lineage preceding endoreduplication or mitotic division ( Materials and methods ) . The concentration threshold that best separates giant cells from small cells is shown as a dashed line . ( E ) Receiver operating characteristic ( ROC ) curve ( red ) for ( D ) . The ratio of correctly and incorrectly classified cells ( i . e . the true positive rate ( TPR ) and false positive rate ( FPR ) ) is calculated for a varying threshold value , providing a characteristic curve . The area under the curve ( AUC ) provides a measure of accuracy for predicting cell fate based on ATML1 concentration ( 1 being perfect and 0 . 5 no better than random classification ) . The AUC is 0 . 76 . The black dot marks the optimal concentration threshold where the difference between TPR and FPR is maximal . ( F–I ) mCitrine-ATML1 peak concentrations and ROC analysis for G1 ( 2C ) or G2 ( 4C ) phases of the cell cycle preceding endoreduplication or mitotic division . ( F ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G1 . ( G ) ROC curve for ( F ) . ( H ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G2 . ( I ) ROC curve for ( H ) . For ( G ) AUC = 0 . 52 ( not predictive ) and for ( I ) AUC = 0 . 8 ( predictive of cell fate ) . ( J–M ) Single cell lineages tracked through time ( 64 hr ) . Each denoised nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . ( J–K ) giant cell and ( L–M ) small cell lineages . ( N–Q ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineages in ( J–M ) . The ploidy at each point corresponds to the color of the dot , as above . mCitrine-ATML1 concentrations for all other cell lineages are plotted in grey for context . Note that giant cells in N and O cross the threshold while they are in G2 ( 4C ) of the cell cycle , while in Q , mCitrine-ATML1 crosses the threshold in 2C at t = 48 hr but then the cell goes onto divide . Additionally , the fate of the cell that crosses the threshold in 4C at t = 48 hr remains unknown . A total of 110 lineages were analyzed ( n = 646 cells ) . This flower is shown in Video 1 . Three similar replicate flowers are shown in the Figure 4—figure supplements 1 , 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01010 . 7554/eLife . 19131 . 011Figure 4—figure supplement 1 . Second flower that demonstrates ATML1 fluctuates in sepal epidermal cells to initiate giant cell patterning . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing sepal . Images were taken every 8 hr for 40 hr . ( B ) Heat maps showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) mCitrine-ATML1 concentrations tracked over time in cells that became giant ( red ) and cells that divided to stay small ( blue ) . ( D ) mCitrine-ATML1 peak concentration levels in each lineage preceding endoreduplication or mitotic division with a predictive concentration threshold ( dashed line ) derived from the ROC analysis . ( E ) ROC curve for ( D ) identifying a predictive threshold . AUC is 0 . 69 . The black dot marks the optimal concentration threshold where the difference between TPR and FPR is maximal . ( F–I ) A threshold in G2 stage of the cell cycle is predictive whereas a threshold in G1 is not . ( F ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G1 . ( G ) ROC curve for ( F ) . ( H ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G2 . ( I ) ROC curve for ( H ) . For ( G ) AUC = 0 . 37 ( not predictive ) and for ( I ) AUC = 0 . 8 ( predictive of cell fate ) . ( J–M ) Single cell lineages tracked through time ( 40 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . ( J–K ) giant cell and ( L–M ) small cell lineages . ( N–Q ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineages in ( J–M ) . Note that in ( P ) a 2C cell passes the giant cell threshold but then divides at t = 8 hr . Additionally in ( Q ) a 4C cell approaches the giant cell threshold but then divides at t = 32 hr . A total of 80 lineages analyzed ( n = 413 cells ) . Associated with Figure 4 and Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01110 . 7554/eLife . 19131 . 012Figure 4—figure supplement 2 . Third flower that demonstrates ATML1 fluctuates in sepal epidermal cells to initiate giant cell patterning . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing sepal . Images were taken every 8 hr for 64 hr . ( B ) Heat map showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) mCitrine-ATML1 concentrations tracked over time in cells that became giant ( red ) and cells that divided to stay small ( blue ) . ( D ) mCitrine-ATML1 peak concentration levels in each lineage preceding endoreduplication or mitotic division with a predictive concentration threshold ( dashed line ) derived from the ROC analysis . ( E ) ROC curve identifying a predictive threshold . AUC is 0 . 73 . The black dot marks the optimal concentration threshold where the difference between TPR and FPR is maximal . ( F–I ) A threshold in G2 stage of the cell cycle is predictive whereas a threshold in G1 is not . ( F ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G1 . ( G ) ROC curve for ( F ) . ( H ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G2 . ( I ) ROC curve for ( H ) . For ( G ) AUC = 0 . 43 ( not predictive ) and for ( I ) AUC = 0 . 8 ( predictive of cell fate ) . ( J–L ) Single cell lineages tracked through time ( 64 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . ( J ) Giant cell , ( K ) small cell , and ( L ) small cell and giant cell lineages . ( M–O ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineages in ( J–L ) . Note that in ( N ) a 4C cell approaches the giant cell threshold but then divides at t = 24 hr . Additionally , in ( O ) two daughter cells go on to have different cell fates . A total of 50 lineages analyzed ( n = 195 cells ) . Associated with Figure 4 and Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01210 . 7554/eLife . 19131 . 013Figure 4—figure supplement 3 . Fourth flower that demonstrates ATML1 fluctuates in sepal epidermal cells to initiate giant cell patterning . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing sepal . Images were taken every 8 hr for 64 hr . ( B ) Heat map showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) mCitrine-ATML1 concentrations tracked over time in cells that became giant ( red ) and cells that divided to stay small ( blue ) . ( D ) mCitrine-ATML1 peak concentration levels in each lineage preceding endoreduplication or mitotic division with a predictive concentration threshold ( dashed line ) derived from the ROC analysis . ( E ) ROC curve identifying a predictive threshold . AUC is 0 . 78 . The black dot marks the optimal concentration threshold where the difference between TPR and FPR is maximal . ( F–I ) A threshold in G2 stage of the cell cycle is predictive whereas a threshold in G1 is not . ( F ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G1 . ( G ) ROC curve for ( F ) . ( H ) mCitrine-ATML1 peak concentration levels and optimal concentration thresholds separating giant cells from small cells at G2 . ( I ) ROC curve for ( H ) . For ( G ) AUC = 0 . 37 ( not predictive ) and for ( I ) AUC = 0 . 84 ( predictive of cell fate ) . ( J–M ) Single cell lineages tracked through time ( 64 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . ( J–K ) giant cell and ( L–M ) small cell lineages . ( N–Q ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineages in ( J–M ) . Note that in ( Q ) two daughter cells go on to have different cell fates . A total of 80 lineages analyzed ( n = 436 cells ) . Associated with Figure 4 and Video 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01310 . 7554/eLife . 19131 . 014Figure 4—figure supplement 4 . Giant cells can be identified by their large , elongated , endoreduplicating nuclei . ( A ) Confocal image of two sepals expressing pATML1::mCitrine-ATML1 ( Green ) in the nucleus and the plasma membrane marker pML1:mCherry-RCI2A ( Red ) . Asterisks mark giant endoreduplicating cells . Note that endoreduplicated nuclei exhibit an elongated shape . ( B ) Nuclear area and cell area were quantified from ( A ) and show a linear correlation ( R2 = 0 . 87 ) . Red , blue and yellow correspond to respective ploidy classifications based on an area threshold ( Box 2 ) . Associated with Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01410 . 7554/eLife . 19131 . 015Figure 4—figure supplement 5 . Mean normalized mCitrine-ATML1 concentrations for all four pATML1::mCitrine-ATML1;atml1–3 flowers . ( A ) mCitrine-ATML1 flower number 1 ( shown in Figure 4 ) . Flower has an inferred normalized ATML1 concentration peak threshold of 1 . 21 . ( B ) mCitrine flower number 2 ( shown in Figure 4—figure supplement 1 ) . Flower has an inferred normalized ATML1 concentration peak threshold of 1 . 41 . ( C ) mCitrine-ATML1 flower number 3 ( shown in Figure 4—figure supplement 2 . Flower has an inferred normalized ATML1 concentration peak threshold of 1 . 45 . ( D ) mCitrine-ATML1 flower number 4 ( shown in Figure 4—figure supplement 3 ) . Flower has an inferred normalized ATML1 concentration peak threshold of 1 . 55 . The average normalized ATML1 concentration peak threshold for all four flowers is 1 . 4 . This threshold value was established as the common threshold . Associated with Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01510 . 7554/eLife . 19131 . 016Video 1 . A movie of a developing pATML1::mCitrine-ATML1; atm1l-3 sepal shown in Figure 4 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01610 . 7554/eLife . 19131 . 017Video 2 . A movie of a developing pATML1::mCitrine-ATML1; atm1l-3 sepal shown in Figure 4—figure supplement 1 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01710 . 7554/eLife . 19131 . 018Video 3 . A movie of a developing pATML1::mCitrine-ATML1; atm1l-3 sepal shown in Figure 4—figure supplement 2 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 01810 . 7554/eLife . 19131 . 019Video 4 . A movie of a developing pATML1::mCitrine-ATML1; atm1l-3 sepal shown in Figure 4—figure supplement 3 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 019 After specification , giant cells immediately enter endoreduplication during early sepal development and endoreduplicating nuclei can be recognized by their size and shape ( Roeder et al . , 2010 ) . We therefore classified nuclei that start to endoreduplicate and become 8C or higher as giant cell nuclei . We verified this by following giant cell differentiation throughout our live imaging series and by comparing these nuclei to nuclei of giant cells defined by cell size in sepals expressing a plasma membrane marker ( Figure 4—figure supplement 4 ) . To assess whether cells destined to be giant have fluctuations of ATML1 that reach higher peak concentrations than cells destined to be small , we tracked mCitrine-ATML1 levels in sepal primordia throughout our live imaging series ( Figure 4C; Figure 4—figure supplements 1C , 2C and 3C ) . We observed that cells that eventually become giant generally exhibit fluctuations reaching higher concentrations of mCitrine-ATML1 before endoreduplication initiates than cells that mitotically divide . However , we observed high fluctuations in some cells that divided to become small cells ( Figure 4C; Figure 4—figure supplements 1C , 2C and 3C ) . To quantitatively determine whether there was an ATML1 concentration threshold that could discriminate between cells that would become giant or cells that would remain small , we assessed how well mCitrine-ATML1 concentration peaks in each cell lineage were able to discriminate between giant cell and small cell fate . To do this , we measured the peak concentration of mCitrine-ATML1 in cells that either go on to divide ( small ) or endoreduplicate ( giant ) and performed a receiver operator characteristics ( ROC ) analysis using these two classes ( Figure 4D and E; Figure 4—figure supplements 1D and E , 2D and E and 3D and E; Chao et al . , 2015; Schröter et al . , 2015; Teles et al . , 2013 ) . In this type of analysis , the ratio of correctly and incorrectly classified cells ( i . e . the true positive rate ( TPR ) and false positive rate ( FPR ) ) is calculated for a varying threshold value , providing a characteristic curve . The area under this curve ( AUC ) provides a measure of accuracy for predicting cell fate based on ATML1 concentration peaks ( 1 being perfect and 0 . 5 no better than random classification ) . We observed an average AUC of 0 . 74 in our different datasets , highlighting the predictive power of ATML1 concentration peaks in discriminating small versus giant cell fate ( AUC = 0 . 76 , 0 . 69 , 0 . 73 , 0 . 78; Figure 4E; Figure 4—figure supplements 1E , 2E and 3E ) . Additionally , for each case we were then able to infer an optimum ATML1 concentration threshold that provides maximum separation between the cells that become giant and cells that remain small , i . e . the concentration value that maximizes the difference between TPR and FPR . We considered this threshold to be indicative of the ATML1 concentration required to trigger endoreduplication for the majority of cells in a given sepal . In summary , we show that the heterogeneity in ATML1 among cells in the sepal primordium can be explained by dynamic cell-autonomous fluctuations , where giant and small cell fate are strongly correlated with the concentration of ATML1 reached . Cells with high concentration fluctuations of ATML1 will likely endoreduplicate and become giant , whereas cells with low concentration fluctuations will likely go on to divide and remain small . Since the decision to endoreduplicate causes a cell to bypass mitosis ( Figure 1G; Inzé and De Veylder , 2006; Sugimoto-Shirasu and Roberts , 2003 ) , we wondered whether high levels of ATML1 needed to occur at a particular stage of the cell cycle to modulate cell-fate decisions . It has been previously demonstrated that in Arabidopsis there is a linear correlation between nuclear size and cell ploidy ( Jovtchev et al . , 2006 ) . Using our live imaging data , we therefore characterized cell cycle stages by ploidy at each time point , using nuclear size as a proxy , where 2C is associated with cells being in G1 and 4C is associated with cells being in G2 ( See Box 2 and Material and methods for ploidy determination ) . Next , we compared peak concentration levels of mCitrine-ATML1 in individual cell lineages during both the 2C and 4C ploidy states of the cell cycle immediately before entry into either mitosis or endoreduplication ( Figure 4F–I; Figure 4—figure supplements 1F–I , 2F–I and 3F–I ) . We found that in the preceding cell cycle , both small cells and giant cells show similar peak levels of mCitrine-ATML1 in 2C ( Figure 4F; Figure 4—figure supplements 1F , 2F and 3F ) . Our ROC analysis shows that ATML1 concentration peaks during the G1 ( 2C ) stage are not predictive of cell fate ( AUCs = 0 . 54 , 0 . 37 , 0 . 43 , 0 . 37; Figure 4G; Figure 4—figure supplements 1G , 2G and 3G ) . In contrast , most cells that experience relatively high peak concentrations of mCitrine-ATML1 while in 4C endoreduplicate and become giant cells ( Figure 4H , J–Q; Figure 4—figure supplements 1H , J–Q , 2H , J–O and 3H , J–Q ) . Our ROC analysis is consistent with this observation , showing that ATML1 concentration peaks in 4C are strongly predictive of cell fate ( AUCs = 0 . 80 , 0 . 80 , 0 . 80 , 0 . 84; Figure 4I; Figure 4—figure supplements 1I , 2I and 3I ) . Overall , these results suggest that a cell is competent to respond to high levels of ATML1 mainly during G2 to induce giant cell formation . Given that high ATML1 levels during the G2 stage of the cell cycle are associated with giant cell formation , we wondered whether all epidermal cells were expressing ATML1 above the giant cell threshold in our ATML1 overexpression sepals to produce an ectopic giant cell phenotype . To address this question , we live imaged early sepal development every 8 hr in plants that had GFP-ATML1 expressed under the PDF1 promoter , which produce the ectopic giant cell phenotype ( Figure 5A; Figure 5—figure supplements 1A and 2A; Videos 5 , 6 and 7 ) . As expected , for a promoter with an ATML1 binding site , PDF1::GFP-ATML1 levels fluctuated in individual cells ( Figure 5B and C; Figure 5—figure supplements 1B , G , 2B and G ) . 10 . 7554/eLife . 19131 . 024Figure 5 . A threshold-based mechanism is consistent with increased giant cell formation in ATML1 overexpression lines . ( A ) Raw images of pPDF1::GFP-ATML1 ( white ) from a live imaging series of a developing overexpression sepal . Images were taken every 8 hr for 48 hr . ( B ) Heat map showing corresponding GFP-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) normalized GFP-ATML1 concentrations tracked over time . Note that all cells tracked become giant . ( D ) Normalized GFP-ATML1 peak concentration levels in each lineage preceding endoreduplication for all three pPDF1::GFP-ATML1 flowers . Dashed line represents the common normalized threshold derived from pATML1::mCitrine-ATML1;atml1–3 flowers ( Figure 4—figure supplement 5 ) . Note that almost all nuclei reach high concentrations of GFP-ATML1 above the threshold before endoreduplicating . ( E–G ) Single giant cells tracked through time ( 48 hr ) . Each denoised nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . ( H–J ) Tracked normalized GFP-ATML1 concentration levels corresponding to the single cell lineages in ( E–F ) . The ploidy at each point corresponds to the color of the dot , as above . GFP-ATML1 concentrations for all other cell lineages are plotted in grey for context . Note that the giant cells cross the threshold while they are in G2 ( 4C ) of the cell cycle . A total of 23 lineages were analyzed ( n = 129 cells ) . This flower is shown in Video 5 . Two similar replicate flowers are shown in the Figure 5—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 02410 . 7554/eLife . 19131 . 025Figure 5—figure supplement 1 . Second flower demonstrating that a threshold-based mechanism is consistent with increased giant cell formation in ATML1 overexpression lines . ( A ) Raw images of pPDF1::GFP-ATML1 ( white ) from a live imaging series of a developing overexpression sepal . Images were taken every 8 hr for 56 hr . ( B ) Heat maps showing corresponding GFP-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C–D ) Single giant cells tracked through time ( 56 hr ) . Each denoised nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . Note that the giant cells cross the threshold while they are in G2 ( 4C ) of the cell cycle . Moreover , in ( D ) the cell does not pass the threshold and instead divides . In the next cell cycle one of the two daughter cells passes the threshold in 4C and starts to endoreduplicate . ( E–F ) Tracked GFP-ATML1 normalized concentration levels corresponding to the single cell lineages in ( C–D ) . The ploidy at each point corresponds to the color of the dot , as above . GFP-ATML1 concentrations in all cell lineages are plotted in grey for context . Note that the giant cells cross the threshold while they are in G2 ( 4C ) of the cell cycle . Moreover , in ( D ) the cell does not pass the threshold and instead divides . In the next cell cycle one of the two daughter cells passes the threshold in 4C and starts to endoreduplicate ( G ) Normalized GFP-ATML1 concentrations tracked over time . Note that most cells tracked are red and become giant and a few small cells area tracked in blue . Dashed line represents the common normalized threshold derived from pATML1::mCitrine-ATML1;atml1–3 flowers ( Figure 4—figure supplement 5 ) . A total of 28 lineages were tracked ( n = 198 cells ) . Associated with Figure 5 and Video 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 02510 . 7554/eLife . 19131 . 026Figure 5—figure supplement 2 . Third flower demonstrating that a threshold-based mechanism is consistent with increased giant cell formation in ATML1 overexpression lines . ( A ) Raw images of pPDF1::GFP-ATML1 ( white ) from a live imaging series of a developing overexpression sepal . Images were taken every 8 hr for 56 hr . ( B ) Heat maps showing corresponding GFP-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C–D ) Single giant cells tracked through time ( 56 hr ) . Each denoised nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C , and red = 8C and higher . Note that the giant cells cross the threshold while they are in G2 ( 4C ) of the cell cycle . ( E–F ) Tracked GFP-ATML1 normalized concentration levels corresponding to the single cell lineages in ( C–D ) . The ploidy at each point corresponds to the color of the dot , as above . GFP-ATML1 concentrations in all cell lineages are plotted in grey for context . Note that the giant cells cross the threshold while they are in G2 ( 4C ) of the cell cycle . ( G ) Normalized GFP-ATML1 concentrations tracked over time . Note that most cells tracked become giant ( red ) and a few small cells are tracked in blue . Dashed line represents the common normalized threshold derived from pATML1::mCitrine-ATML1;atml1–3 flowers ( Figure 4—figure supplement 5 ) . A total of 33 lineages were tracked ( n = 257 cells ) . Associated with Figure 5 and Video 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 02610 . 7554/eLife . 19131 . 027Video 5 . A movie of a developing pPDF1::GFP-ATML1 sepal shown in Figure 5 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 02710 . 7554/eLife . 19131 . 028Video 6 . A movie of a developing pPDF1::GFP-ATML1 sepal shown in Figure 5—figure supplement 1 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 02810 . 7554/eLife . 19131 . 029Video 7 . A movie of a developing pPDF1::GFP-ATML1 sepal shown in Figure 5—figure supplement 2 . The sepal primordium was live imaged every 8 hr until giant cells form . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 029 We next tested whether most epidermal cells surpassed the ATML1 threshold in G2 to induce endoreduplication . Since very few cells divide in our pPDF1::GFP-ATML1 sepals , we could not directly infer this threshold through ROC analysis from this data as before . Therefore , we derived a common ATML1 concentration threshold from the live imaging data of our pATML1::mCitrine-ATML1; atml1–3 flowers ( Figure 4—figure supplement 5 ) , by performing ROC analysis using mean normalized ATML1 concentrations for each flower ( see Materials and methods for details ) . Applying this threshold to the pPDF1::GFP-ATML1 data , we observed that almost all endoreduplicating cells exhibited high peak levels of GFP-ATML1 in G2 , above the common threshold ( Figure 5D–J; Figure 5—figure supplements 1C–F; and 2C–F ) . This is in contrast to wild type , where fewer cells reach the ATML1 concentration threshold ( Figure 4 ) . Combined , these data suggest that our overexpression line follows the same threshold-based cell-autonomous fluctuation patterning mechanism; the increased basal GFP-ATML1 expression from the PDF1 promoter raises ATML1 production levels such that almost all sepal epidermal cells surpass the giant cell fate-inducing threshold during G2 . We have previously published that a cyclin dependent kinase inhibitor , LOSS OF GIANT CELLS FROM ORGANS ( LGO ) , is required for giant cell formation; LGO triggers endoreduplication once giant cell fate has been established ( Roeder et al . , 2012 ) . To verify that LGO acts genetically downstream of ATML1 to establish giant cells , we crossed our ATML1 overexpression line ( pPDF1::FLAG-ATML1 ) to our lgo-2 mutant , which exhibits a loss-of-giant cell phenotype ( Figure 6C ) . Plants homozygous for both the lgo-2 mutation and the overexpression transgene do not form giant cells , demonstrating that LGO activity is required downstream of ATML1 for formation of giant cells . 10 . 7554/eLife . 19131 . 030Figure 6 . The dynamics of ATML1 fluctuations are independent of endoreduplication . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing lgo mutant sepal . Images were taken every 8 hr for 64 hr . ( B ) Heat maps showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) Genetic epistasis analysis between lgo-2 mutant and ATML1 overexpression line ( pPDF1::FLAG-ATML1 ) . Plants homozygous for both the lgo mutation and the overexpression transgene do not form giant cells , demonstrating that LGO acts genetically downstream of ATML1 to promote endoreduplication . ( D ) Quantification of the average number of giant cells in four pATML1::mCitrine-ATML1; atml1–3 sepals ( ncells = 75 , four sepals ) compared to the number of giant cells predicted to form by applying the common threshold to ATML1 concentrations observed in pATML1::mCitrine-ATML1; lgo sepals ( ncells = 59 , three sepals ) . Error bars = standard error of mean . Approximately the same number of cells would be expected to become giant cells in lgo sepals as in wild type , except that they fail to endoreduplicate . A T-test performed between the two populations yielded a non-significant ( ns ) p-value of 0 . 9 ( E ) Traces of mCitrine-ATML1 normalized concentrations of cells that do not reach the inferred threshold in G2 of the cell cycle and are predicted to remain small ( nsmall = 70 ) . ( F ) Traces of mCitrine-ATML1 normalized concentrations of cells that reach the inferred threshold during G2 of the cell cycle and are predicted to become giant ( ngiant = 25 ) . The trace ends when the cell is predicted to become giant . In ( E–F ) the dashed line represents the common normalized threshold derived from pATML1::mCitrine-ATML1;atml1–3 flowers ( Figure 4—figure supplement 5 ) . ( G–H ) Single small cell lineages tracked through time ( 64 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C . The cell marked with X is lost from our tracking . ( I–J ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineages in ( G–H ) . Cells that cross the mCitrine-ATML1 threshold fail to endoreduplicate and instead divide . A total of 149 lineages were analyzed ( n = 495 cells ) . This flower is shown in Video 8 . Two similar replicate flowers are shown in the Figure 6—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03010 . 7554/eLife . 19131 . 031Figure 6—figure supplement 1 . Second flower showing that dynamic fluctuations of ATML1 are independent of endoreduplication . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing lgo mutant sepal . Images were taken every 8 hr for 64 hr . ( B ) Heat map showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) Traces of cells that do not reach the inferred threshold ( 1 . 4 ) in G2 of the cell cycle and are predicted to remain small , nsmall = 87 . ( D ) Cell traces of cells that reach the inferred threshold ( 1 . 4 ) during G2 of the cell cycle that are predicted to become giant . ngiant = 26 . The trace ends when the cell is predicted to become giant . ( E ) Single small cell lineage tracked through time ( 64 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C . ( F ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineage in ( E ) . A total of 196 lineages were tracked ( n = 756 cells ) . Associated with Figure 6 and Video 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03110 . 7554/eLife . 19131 . 032Figure 6—figure supplement 2 . Third flower showing that dynamic fluctuations of ATML1 are independent of endoreduplication . ( A ) Raw images of pATML1::mCitrine-ATML1 ( white ) from a live imaging series of a developing lgo mutant sepal . Images were taken every 8 hr for 64 hr . Labels below the snapshots display the time after the time course was initiated , in hour units . ( B ) Heat map showing corresponding mCitrine-ATML1 concentrations ( total fluorescence divided by area ) at each time point from ( A ) . ( C ) Traces of cells that do not reach the inferred threshold ( 1 . 4 ) in G2 of the cell cycle and are predicted to remain small . nsmall=128 . ( D ) Cell traces of cells that reach the inferred threshold ( 1 . 4 ) during G2 of the cell cycle that are predicted to become giant . ngiant = 8 . The trace ends when the cell is predicted to become giant . ( E ) Single small cell lineage tracked through time ( 64 hr ) . Each nucleus image is outlined in a color associated with its ploidy: yellow = 2C , blue = 4C . The cell with the X is lost from our tracking . ( F ) Tracked mCitrine-ATML1 concentration levels corresponding to the single cell lineage in ( E ) . A total of 151 lineages were tracked ( n = 619 cells ) . Associated with Figure 6 and Video 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 032 Since LGO acts downstream of ATML1 , we hypothesized that ATML1 fluctuations should be unaltered in the lgo-2 mutant , which fail to endoreduplicate in early stage sepals . In this scenario , we would expect the same number of lgo-2 nuclei to surpass the ATML1 threshold in G2 as in wild type . Cells that pass the threshold would still divide because they are unable to endoreduplicate . To test this , we live imaged our mCitrine-ATML1 reporter in the lgo-2 mutant background ( Figure 6A; Figure 6—figure supplements 1A and 2A; Videos 8 , 9 and 10 ) . These plants still exhibited mCitrine-ATML1 fluctuations , suggesting that ATML1 fluctuates independently of LGO ( Figure 6B , E–J; Figure 6—figure supplements 1B–F; and 2B–F ) . We applied the common ATML1 concentration threshold derived from pATML1::mCitrine-ATML1; atml1–3 flowers ( see previous section; Figure 4—figure supplement 5 ) to predict the number of giant cells that would have formed exclusively based on the threshold mechanism ( ATML1 concentration peaks above threshold during G2; Materials and methods ) . We found no significant differences between the predicted number of giant cells in the lgo-2 mutant and the observed number of giant cells in wild type ( Figure 6D–F; Figure 6—figure supplements 1C–D and 2C–D ) . This suggests that a cell may still fluctuate to high levels of ATML1 in G2 but without LGO , cells cannot respond to these fluctuations to trigger endoreduplication . Since the absence of LGO does not seem to change the dynamics of ATML1 , this result further indicates that ATML1 fluctuations are independent of endoreduplication . 10 . 7554/eLife . 19131 . 033Video 8 . A movie of a developing pATML1::mCitrine-ATML1; lgo sepal shown in Figure 6 . The sepal primordium was live imaged every 8 hr throughout development . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03310 . 7554/eLife . 19131 . 034Video 9 . A movie of a developing pATML1::mCitrine-ATML1; lgo sepal shown in Figure 6—figure supplement 1 . The sepal primordium was live imaged every 8 hr throughout development . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03410 . 7554/eLife . 19131 . 035Video 10 . A movie of a developing pATML1::mCitrine-ATML1; lgo sepal shown in Figure 6—figure supplement 2 . The sepal primordium was live imaged every 8 hr throughout development . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 035 Previous studies have suggested that gene expression is inherently stochastic , where genes will experience random fluctuations in the rate in which they are transcribed and/or translated ( Elowitz et al . , 2002; Kaern et al . , 2005 ) . We therefore asked whether a simple computational model that exhibits cell-autonomous stochastic fluctuations of ATML1 is sufficient to recapitulate giant cell patterning as observed in our experimental data . In our model , we implemented a simplified regulatory network , where ATML1 stochastically fluctuates in a growing tissue ( Figure 7A and B; Materials and methods ) . In this model , we assume that in every cell there is a basal amount of ATML1 being produced as well as an amount being linearly degraded . In addition , we tested the possibility that ATML1 engages in a self-catalytic feedback loop , as ATML1 has a putative ATML1 binding site in its own promoter and ATML1 has been shown to bind this motif in vitro ( Abe et al . , 2001; Takada and Jürgens , 2007 ) . Additionally , in seedlings induction of ectopic ATML1 activity for seven days shows an increase of endogenous ATML1 expression 1 . 5 to two fold , hinting at the possibility of a feedback loop ( Takada et al . , 2013 ) . 10 . 7554/eLife . 19131 . 036Figure 7 . A plausible stochastic model for giant cell patterning . ( A ) Schematic diagram of the computational model for giant cell patterning . Top panel shows the proposed ATML1 model network in which ATML1 can prevent cell division and instead drive entry into endoreduplication and giant cell specification . Middle panel shows a cartoon of the cell cycle timer time course . When the timer exceeds a first threshold level ΘC , S , cells enter into the G2 phase and increase their ploidy to 4C . When the timer reaches a second threshold level , ΘC , D , cells divide , unless their target levels have surpassed the threshold ΘT sometime during G2 phase . Bottom panel shows a scatter plot cartoon illustrating how a ‘hard threshold’ in the target levels results in a ‘soft threshold’ in ATML1 . We refer to a hard threshold when levels right above or below the threshold will result in two different outcomes . If the target perfectly followed the dynamics of ATML1 , its upstream regulator , and obeyed a deterministic dynamics , all cells that cross the target threshold ΘT would also cross a corresponding hard ATML1 threshold . Hence , a hard threshold in the target would be effectively encoded as a hard threshold on its upstream regulator ATML1 . In contrast , in our model , the target has a finite degradation rate , and stochastic dynamics , so that it is not a perfect follower of ATML1 dynamics; thus , a hard threshold in target levels ( vertical red dashed line ) results in a soft threshold in ATML1 ( horizontal red dashed line ) . A cell close to the ATML1 soft threshold may or may not pass the target threshold and endoreduplicate to become a giant cell . Dots in the bottom panel is a cartoon of the ATML1 maxima of simulated cell lineages , with red dots indicating cells that become giant , while blue dots represent mitotically dividing small cells . ( B ) Simulation snapshots of the in silico growing sepal showing ( top ) ATML1 concentrations and ( bottom ) cell ploidies ( Video 11 ) . ( C ) Time courses of ATML1 ( left ) and its target ( right ) for a cell committing to the giant fate ( top ) and a small dividing cell ( bottom ) . Colors of the time traces represent the cell ploidy . Color code for the ploidies is the same as in panel B . Red dashed lines represent the predicted soft ATML1 threshold ΘA* , and the ΘT hard threshold imposed in the target ( Materials and methods ) . ( D ) Histogram at a final simulation time point showing ATML1 concentration levels . ( E ) Boxplot showing the percentage of cell ploidies in a simulated tissue for five simulations with different random initial ATML1 , target and timer levels . ( F–G ) ROC analysis of the ATML1 concentration maxima for the simulated lineages at ( F ) 2C and ( G ) 4C , showing that the ATML1 maximal levels at 2C is not predictive , in agreement with experimental data ( Figure 4F–I; Figure 4—figure supplements 1–3F–I ) . Parameter values are described in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03610 . 7554/eLife . 19131 . 037Figure 7—figure supplement 1 . Simulation results showing different stochastic time courses . Time courses for cells committing to the ( A and C ) giant fate and ( B and D ) small cell fate of ATML1 ( left ) , its target ( middle left for A and B; middle for C and D ) and the timer ( middle right for A and B ) . Colors of the time traces represent the cell ploidy , as in Figure 7 . Right panels in show the ATML1 time courses as in the left panel but with less time resolution , which we refer to as coarse-grained time courses . Such coarse-grained time courses are shown to emulate an experimental time courses . Dot colors in the coarse-grained time course panels represent the ploidy , following the color code in Figure 7B , while the line color refers in this case to the cell type at the end of the simulation , being orange for giant cells and blue for the small dividing cells . Horizontal red lines in the ATML1 panels represent the predicted soft threshold ΘA* and the red shaded region is a measure of its error ( Materials and methods ) . The other horizontal lines in the target and timer variables show the different thresholds set in the simulations . Coarse-grained time courses show we may lose some ATML1 peaks due to the lower time resolution , and this might slightly decrease the levels of the predicted threshold with respect to the threshold predicted from traces with higher time resolution . Cells with ATML1 levels reaching the predicted soft threshold in G2 are likely to have the corresponding downstream target levels above the target threshold , driving endoreduplication ( see A ) . However , ATML1 may even cross the predicted soft threshold , and still not drive giant cell endoreduplication , if the target does not cross its own threshold ( see C middle and right ) . On the other side , cells being close to but not reaching the ATML1 threshold might endoreduplicate , provided that the target crosses its corresponding threshold ( C left , middle ) . Parameter values are described in Table 1 . Associated with Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03710 . 7554/eLife . 19131 . 038Figure 7—figure supplement 2 . Stochastic fluctuations are essential for generating the giant cell patterning . Phase diagrams across the parameter space of basal ATML1 production rates and ATML1 auto-induction rates showing ( A and C ) the fraction of giant cells in the tissue and ( B and D ) the CVs of the ATML1 concentration in the tissue at ( A–B ) higher ( E0 = 15 ) and ( C–D ) lower ( E0 = 1500 ) noise intensities . Note that the noise intensity is inversely proportional to the characteristic cell size E0 . At higher basal ATML1 production rates , all cells cross the ATML1 soft threshold and endoreduplicate to become giant ( fraction of giant cells = 1 , colored dark red on the heat map ) . Conversely , at lower basal ATML1 production rates , all cells divide to remain small ( fraction of giant cells = 0 , colored dark blue on the heat map ) . For a certain range of basal ATML1 production rates , dynamic stochastic fluctuations create a salt-and-pepper pattern of giant cells interspersed between mitotically dividing cells ( rainbow region of the parameter space ) . At lower noise intensities , no pattern emerges in a wide visible region of the studied parameter space , and the modeled sepal has either just non-giant or giant cells . A few parameter values might still drive a salt-and-pepper pattern at low noise intensities , provided that the initial conditions are sufficiently noisy . Parameter values are described in Table 1 . Associated with Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03810 . 7554/eLife . 19131 . 039Figure 7—figure supplement 3 . Classification analysis of the simulated data shows that a weak feedback or no feedback in ATML1 reproduces the experimental observations . Analysis for ( A–D ) full and ( F–J ) coarse grained simulated time courses show we get equivalent AUC values and similar ATML1 soft thresholds . ( A and F ) AUC values of 5 simulations with different random initial conditions . ( B ) Scatter plot showing the maximal ATML1 levels and the corresponding target levels at 4C for ( red ) giant and ( blue ) small dividing cells . Dashed vertical and horizontal lines show the imposed hard threshold for the target ( ΘT ) and the predicted soft threshold for ATML1 ( ΘA* ) ( Materials and methods ) . This plot shows that a hard threshold in the target results in a soft threshold in ATML1: above the soft threshold ΘA* , we find cells having higher and lower maximal target levels than the target threshold ΘT , becoming giant cells or remaining as small cells , respectively . ( C–D , I–J ) Spread plots showing the maxima of the ATML1 at 2C and 4C and the predicted ATML1 threshold ΘA* . ( E ) Spread plot of the maximal target values at 4C and the predicted target threshold ΘT* . Notice that the predicted hard threshold for the target ΘT* accurately matches the assigned target threshold ΘT ( see panel B ) . ( G–H ) ROC curves for the coarse-grained time course ( see Figure 7F–G for the equivalent ROC curves computed with the simulated time resolution of 0 . 1 ) . ( K–N ) ROC analysis performed on simulated data with ( K–L ) higher and ( M–N ) lower time resolution in the parameter space show equivalent AUC trends . This analysis show that cells can be classified with respect to its maximal ATML1 levels at 4C but not in 2C in a wide region of the parameter space , namely , when there is no feedback , or when there is weak feedback . This is in agreement with the analysis of the experimental data . In contrast , with higher feedback strengths , maximal ATML1 levels in both 2C and 4C become predictive of giant cell fate , which does not correspond to our experimental data . The ROC analysis has been performed just in the parameter region where simulations lead to a pattern of giant and dividing cells , i . e . , where the fraction of giant cells in the tissue being between 0 and 1 . ROC analysis in higher ( K–L ) and lower ( M–N ) time resolution time courses give equivalent results . ( O–P ) ROC analysis in the parameter space for a model where the target can induce endoreduplication throughout the whole cell cycle and not just in G2 . The time resolution for the ROC analysis and threshold determination was 0 . 1 AU for A–E panels , eight for F– J and M–N panels , and 0 . 5 for K–L and O–P panels . Parameter values are described in Table 1 . Associated with Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 03910 . 7554/eLife . 19131 . 040Figure 7—figure supplement 4 . Theoretical and experimental study of the ATML1 auto-induction strength . ( A–D ) Simulation results of the model with different ATML1 auto-induction strengths show different qualitative behaviors . Simulations with different feedback strengths and different ATML1 basal production rates that lead to similar percentages of giant cells in the tissue ( 7 to 9% ) are shown . Histograms of ATML1 concentrations ( left ) , ATML1 time traces for cells becoming giant cells ( middle left ) , ROC analysis ( middle right ) and percentages of cells with the different ploidies . Boxplots are obtained from five simulations with different initial conditions ( middle right and right panels ) . The stronger is the feedback , the higher is the CV for the ATML1 concentrations . No feedback or a weak feedback give rise to a unimodal distribution of ATML1 concentrations ( A–C ) , while a stronger feedback can give rise to a bimodal distribution ( D ) . For no feedback ( A ) or weaker feedback ( B ) , small and fast fluctuations drive the singling out of cells for endoreduplication . Stronger feedback strengths make larger fluctuations appear , whose time-scales are larger ( C–D ) . This makes the ATML1 peak levels at 2C equivalent to 4C levels . In these cases , the AUC values are also high at 2C , which does not correspond to the experimental data . Similar fractions of giant cells are produced by all four induction strengths ( A–D ) . Color codes of the time traces are as in Figure 7—figure supplement 1 . The red lines and shaded bands in the time traces represent the predicted ATML1 soft threshold ΘA* and its error , respectively ( Materials and methods ) . Feedback auto-induction strengths and basal production rates for the different panels are ( A ) VA = 0 , PA = 1 . 41 , ( B ) VA = 1 . 25 , PA = 1 . 14 , ( C ) VA = 1 . 75 , PA = 1 . 01 and ( D ) VA = 2 . 5 , PA = 0 . 88 , respectively . Left and right panels in ( B ) are also shown in Figure 7 , and the AUCs panel in ( B ) is shown in Figure 7—figure supplement 3 . Other parameter values are described in Table 1 . ( E–F ) QPCR results testing feedback strength of ATML1 induction on endogenous ATML1 48 hr after application of 10 µM , 1 µM or 0 . 1 µM estradiol ( inducing agent ) compared to mock treated inflorescences . Endogenous ATML1 transcript levels increase approximately 1 . 5-fold within 48 hr after ATML1 is induced with 10 µM estradiol as compared to mock-treated plants . To put this fold change in context , we examined the 10 µM estradiol induction of other genes downstream of ATML1 including CER5 ( 2 . 3-fold ) , FDH ( 1 . 6-fold ) , PDF2 ( 1 . 5-fold ) and PDF1 ( 1 . 2-fold ) . Note that these downstream genes are induced with very similar fold changes as the endogenous ATML1 . CER5 , FDH , and PDF1 do not encode transcription factors and therefore cannot act in a feedback loop . This suggests that the feedback of ATML1 on itself is not activating ATML1 further than other targets at the 48 hr time point . The transgene was induced about 700-fold by 10 µM estradiol ( F ) . Induction with 0 . 1 µM or 1 µM estradiol produced intermediate levels of induction and activation of downstream genes , also consistent with a weak positive feedback loop . Wilcoxon 1-tailed tests were performed between the corresponding mock treated and estradiol treated plants . p-value ≤ 0 . 05 marked with * . Associated with Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 040 ATML1 is a transcription factor that regulates the expression of downstream genes . Therefore , to induce endoreduplication , ATML1 likely directly or indirectly regulates the expression of a downstream cell cycle regulator ( e . g . cyclin/CDK/cyclin-dependent kinase inhibitor ) . We therefore assigned ATML1 to activate a downstream target that inhibits cell division and promotes entry into endoreduplication . Only if the downstream target passes its own specific threshold in G2 , does it successfully drive a cell to endoreduplicate to form a giant cell ( Figure 7A , C; Figure 7—figure supplement 1 ) . Hence , we expect a few cells to divide even if their ATML1 concentrations go above the threshold because the target’s threshold is not reached . This is consistent with our live imaging data , where in some cases mCitrine-ATML1 concentrations exceed the giant cell threshold in 4C but the cells go on to divide ( Figure 4—figure supplement 2N ) . Furthermore , we expect that a few giant cells will form when ATML1 approaches but does not exceed the threshold because the target stochastically passes its own threshold ( Figure 7—figure supplement 1C ) . These circumstances create what we term a soft ATML1 threshold ( Figure 7A ) . In the model , different ploidy and cell division checkpoints were determined using a linearly increasing timer variable , which represents the cell cycle . The timer resets at every cell division checkpoint with a small amount of noise ( Figure 7A; Material and methods; Figure 7—figure supplement 1A , B; Video 11 ) . 10 . 7554/eLife . 19131 . 041Video 11 . Simulation results showing ATML1 , target , timer levels and cell ploidies throughout time in a growing tissue . Cells that cannot divide , increase their ploidy , becoming giant cells . The time resolution of the displayed movie ( 0 . 5 ) is lower than the actual simulation time step ( 0 . 1 ) , so fluctuations in ATML1 and in the target may be missed . Color scales in the ATML1 and target variables have been truncated for the sake of better visualizing the fluctuations . Parameter values are described in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 041 The model qualitatively reproduced our experimental data and led to a scattered pattern of giant cells in a growing tissue ( Figure 7B , Video 11 ) . Specifically , dynamic fluctuations in ATML1 and in the target during G2 enable a subset of cells from the developing tissue to become giant cells ( Figure 7C–E; Figure 7—figure supplement 1 ) . We found parameter values that produced wild-type-like sepals , in which the distributions of ATML1 levels and the number of giant cells were similar to those observed experimentally ( Figures 3I–J and 7D; Materials and methods ) . Furthermore , lowering the intensity of the stochastic fluctuations in the model prevented it from matching the experimental data ( Figure 7—figure supplement 2 ) . To test whether our model could recapitulate G2-mediated giant cell fate specification , we performed a ROC analysis on the simulated time traces , mimicking the analysis performed on the experimental data ( Figure 7F–G and Figure 7—figure supplement 3 ) . Consistent with our experimental observations , we found lower AUC values in 2C stages than in 4C . This supports our hypothesis that ATML1 levels during the G2 phase of the cell cycle are important for giant cell fate commitment ( Figure 7F–G and Figure 7—figure supplement 3A–E ) . To further study whether our model could recapitulate our experimental data , in which some fluctuations might be missed due to the 8 hr interval live imaging , we tested whether our AUC analysis would still give similar results when studying the simulated time traces with lower time resolution . We therefore subsampled our simulated data to generate coarse time series , with 80 times lower time resolution than the simulated time step , and we still detected the same trends ( Figure 7—figure supplement 3F–J ) . As previously mentioned , ATML1 might act in a positive feedback loop . We therefore explored different feedback strengths in the parameter space to determine the robustness of our model . We modeled the different feedback strengths by varying the ratio between ATML1 dependent and basal production rates , whilst keeping the number of predicted giant cells close to experimental values ( Materials and methods ) . With no feedback or low feedback strengths , we could qualitatively match the experimental ROC analysis ( Figure 7—figure supplements 3K–N; and 4A–B ) . In contrast , we were unable to match our experimental data with high feedback strengths because AUC values were predictive of giant cell identity in both 2C and 4C ( Figure 7—figure supplements 3K–N; and 4C–D ) . Higher feedback strengths lead to bistability in the system , inducing large and slow fluctuations between high and low levels ( Figure 7—figure supplement 4C–D ) . To test the type of feedback of ATML1 on itself , we examined the effects of induction of ATML1 on the transcription of the endogenous ATML1 gene in inflorescences using qPCR ( Peterson et al . , 2013; Takada et al . , 2013 ) . We found that ATML1 induction with 10 µM estradiol lead to total ATML1 levels 7 . 1 times higher than the mock treated samples , and increased endogenous ATML1 expression 1 . 5-fold within 48 hr ( Figure 7—figure supplement 4E–F ) . This level of induction was similar to that observed in other downstream genes , suggesting that the feedback of ATML1 on itself is not activating ATML1 further than other targets at the 48 hr time point ( Figure 7—figure supplement 4E–F ) . The results are also consistent with a previous study carried out in seedlings after 7 days , where endogenous ATML1 levels increased to 1 . 7-fold after induction ( Takada et al . , 2013 ) . To further test the properties of the feedback , we also induced with 0 . 1 µM or 1 µM estradiol and achieved intermediate levels of induction and activation of downstream genes . In our strong feedback simulations , the parameters chosen are on , or close to , the bistability region in the system , leading to a long-tailed or bimodal distribution of ATML1 expression ( Figure 7—figure supplement 4C–D ) , which we do not observe experimentally ( Figure 3J ) . Our experimentally observed gradual increase in induced ATML1 with increasing levels of estradiol further supports the case for weak feedback in the system , as endogenous ATML1 levels are not sensitive to small increases in exogenous ATML1 . In the strong feedback case , sensitivity to ATML1 induction increases as the system is bistable and easily reaches the high value state . Thus , our results are consistent with weak feedback in the system . In order to confirm that endoreduplication can occur only if the target reaches a threshold in G2 , we simulated a simpler model where cells could commit to endoreduplication if the target reaches its threshold at any point throughout the cell cycle . In contrast to our experimental data , these simulations led to ATML1 exhibiting high AUC values in both 2C ( G1 ) and 4C ( G2 ) ( Figure 7—figure supplement 3O–P ) . These results reaffirm our hypothesis that a cell’s ability to respond to the target must be restricted to G2 in order for ATML1 to be predictive only in the G2 phase of the cell cycle . We then asked whether our model could qualitatively reproduce the ATML1 dosage phenotypes we had observed with our genetic dosage series . We found that changing the basal ATML1 production rate was sufficient to gradually increase the total amount of the ATML1 in the modeled tissue , and accordingly , the fraction of giant cells in the sepal ( Figure 8 ) . These results , together with our dosage analysis , show that there is a positive relationship between graded ATML1 levels and the fraction of giant cells produced in the tissue ( Figures 2 and 8 ) . 10 . 7554/eLife . 19131 . 042Figure 8 . The model recapitulates ATML1 dosage dependency . ( A ) Snapshots showing the resulting patterns of giant cells ( 8C , 16C , 32C and 64C cells ) and small cells ( 2C and 4C cells ) at the final time point of the simulations when the basal ATML1 production rate is modified . Values chosen for the ATML1 basal production rate from the parameter exploration shown in panels B-G are , from left to right: PA = 1 . 58 , PA = 1 . 25 , PA = 1 . 17 , PA = 1 . 14 , PA = 1 . 01 and PA = 0 . 99 . ( B–G ) Simulation results for different basal ATML1 production rates for ( B–D ) a model with a weak auto-induction ATML1 feedback loop ( VA = 1 . 25 ) and for ( E–G ) a model with no feedback ( VA = 0 ) . ( B and E ) Total amount of ATML1 in the tissue . The total ATML1 amount is the sum of the area of each cell multiplied by the ATML1 concentration in that cell . The feedback drives a sharper increase of ATML1 amount for a certain range of basal ATML1 production rates . ( C and F ) Fraction of giant cells ( 8C , 16C , 32C and 64C cells ) in the tissue with respect to the total amount of ATML1 . The gradual increase of the fraction of cells with respect to the total ATML1 amount in the tissue is qualitatively consistent with the different phenotypes shown in Figure 2 . The model with feedback has a slightly more gradual increase in fraction of giant cells with respect to the total amount of ATML1 . ( D and G ) CVs of the ATML1 concentrations in the tissue . In the cases of having a weak feedback or not having a feedback , there is a plateau of CV values for intermediate ATML1 total amounts in the tissue . Stronger feedback levels will lead to non-monotonic CVs with respect to the total amount of ATML1 ( see Figure 7—figure supplement 2B ) . Other parameter values are described in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 042 Hence , our model shows that fast and relatively small stochastic fluctuations of ATML1 are sufficient to pattern giant and small cells in the sepal . ATML1 activates a downstream target , which if activated in G2 , will induce endoreduplication . The dynamics of the ATML1-target network creates a soft ATML1 threshold during G2 . Here , we have identified a cell-autonomous fluctuation patterning mechanism for specifying cell fate in a multicellular system ( Figure 9 ) . During Arabidopsis sepal development , the pattern of giant cells and small cells in the epidermis is initiated through fluctuations of the transcription factor ATML1 . Using live-imaging , quantitative image analyses and mathematical modeling , we have revealed that cells in which ATML1 levels surpass a soft threshold during the G2 phase of the cell cycle have a high probability of establishing giant cell identity and entering endoreduplication . A sepal epidermal cell is only competent to respond to ATML1 fluctuations during a window of time defined by G2 stage of the cell cycle . 10 . 7554/eLife . 19131 . 043Figure 9 . Fluctuations of ATML1 around a soft threshold pattern giant cells and small cells in the sepal . ATML1 fluctuates in every young sepal epidermal cell . However , cells only respond to high levels of ATML1 during G2 phase of the cell cycle . ( A ) Schematic showing that in G1 , cells are impervious to high concentrations of ATML1 . In G2 , cells can respond to ATML1 to become a giant cell if levels surpass a soft threshold . If a cell does not receive a high enough level then the cell will divide . ( B ) Schematic demonstrating a cell progression from 2C ( G1 phase of the cell cycle ) to 4C ( G2 stage of the cell cycle ) . The cell will then either become an 8C cell , if it receives a high level of ATML1 , or to divide to make two 2C cells if ATML1 levels are low . In the G2 phase , our inferred mCitrine-ATLML1 threshold level is about 80% accurate in predicting giant cells versus small identity correctly . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 04310 . 7554/eLife . 19131 . 044Figure 9—figure supplement 1 . ACR4 and DEK1 act in the giant cell patterning pathway . ( A–B ) SEM images of a sepal overexpressing ( OX ) ATML1 under the PDF1 promoter ( pPDF1::FLAG-ATML1 ) . ( C–D ) SEM images of a wild-type sepal . ( E–F ) SEM images of sepal homozygous for both ATML1 OX transgene and acr4 mutation . ( G–H ) SEM images of acr4–2 mutant sepal . Note that the number of giant cells is severely reduced . ( I–J ) SEM images of a sepal homozygous for both ATML1 OX transgene and dek1–4 mutation . Sepal contains no giant cells . ( K–L ) SEM image of dek1–4 mutant sepal . Note that no giant cells form . All giant cells are false colored red . Scalebars in A–L , 100 µm . Associated with Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 044 Strikingly , our fluctuation-patterning model resembles Wolpert’s French flag model in that each individual cell makes an autonomous fate decision based on the concentration of a key developmental regulator . Our model however deviates from the French flag model because it utilizes internal fluctuations instead of a diffusible morphogen to generate concentration differences . Concentration threshold-based patterning mechanisms have been traditionally viewed as being non-robust because they are sensitive to small perturbations in concentrations . Often additional mechanisms are needed to achieve robustness ( Eldar et al . , 2002 , 2003; Kondo and Miura , 2010 ) . This sensitivity to small changes in concentration is consistent with our results in the sepal , where giant cell formation is highly responsive to changes in the basal production of ATML1 . Interestingly however , in wild-type plants , the number of giant cells varies only slightly from sepal to sepal , falling within a small range ( 10-30 ) . This indicates that these fluctuations together with a threshold must be tuned to ensure that the correct proportion of giant cells form on the sepal . Our data suggests that the cell cycle acts as a stabilizing factor to restrict giant cell fate decisions similarly to secondary mechanisms used in other biological systems . A few recent studies have similarly demonstrated that the cell cycle provides a window of opportunity for making cell fate decisions . However , these studies suggest that G1 is the critical phase for specification . During G1 , there is a growth factor-dependent restriction point , where a cell determines whether to enter quiescence ( G0 ) or progress through the cell cycle . Cyclin/Cyclin Dependent Kinase ( CDK ) activity is normally reduced during the restriction point , providing a window for cells to receive extracellular signals necessary for cell fate decisions ( Blagosklonny and Pardee , 2002; Blagosklonny et al . , 2002 ) . This has been nicely demonstrated in human embryonic stem cells , where a stem cell’s ability to differentiate into an endodermal cell is dependent upon receiving TGF-β-Smad2/3 signals during this restriction point in early G1 , when CyclinD levels are low ( Pauklin and Vallier , 2013 ) . In addition to transient Cyclin/CDK expression , some studies have found that cells extend their G1 phase immediately before differentiation . This may allow cell fate inducing factors to reach sufficient levels to induce differentiation ( Calegari et al . , 2005; Collart et al . , 2013 ) . How G1 lengthening occurs is still under debate . However , one recent study showed that increasing a cell’s nuclear to cytoplasmic ratio dilutes the concentration of DNA replication factors which results in a prolonged G1 phase ( Collart et al . , 2013 ) . Additionally , Singh et al . showed that chromatin changes associated with the M-G1 transition cause transcriptional leakiness of many prodifferentiation genes , which prime cells to respond to cellular differentiation signals ( Singh et al . , 2013 ) . We have found that a sepal epidermal cell’s window to differentiate occurs not in G1 but in G2 , suggesting a different manner of regulation than in G1-gated determination . For instance , cell fate decisions governed by the G1 phase of the cell cycle must often receive an extracellular signal to activate prodifferentiation genes instead of going into G0 quiescence . In contrast , our model suggests ATML1 fluctuations could be sufficient to pattern the sepal without a need for an extracellular signal . Alternatively , ATML1 could be priming the cell to receive a signal during the G2 phase of the cell cycle . We have previously reported that ACR4 ( a transmembrane receptor kinase; Gifford et al . , 2003 , 2005; Watanabe et al . , 2004; Roeder et al . , 2012 ) and DEK1 ( a transmembrane calpain protease; Liang et al . , 2013; Lid et al . , 2002 , 2005; Roeder et al . , 2012 ) act in the giant cell formation pathway , suggesting that intercellular signaling may assist in promoting giant cell fate decisions . An epistasis analysis between ACR4 , DEK1 and ATML1 reveals that during giant cell formation , ACR4 acts upstream of ATML1 but that DEK1 acts downstream ( Figure 9—figure supplement 1 ) . These results are in opposition to what has been previously published about these genes during embryogenesis , where DEK1 acts upstream of ATML1 and ACR4 acts downstream ( Abe et al . , 2003; Gifford et al . , 2003; Johnson et al . , 2008; San-Bento et al . , 2014; Takada et al . , 2013; Tanaka et al . , 2002 ) . One possibility for these results is that ACR4 and DEK1 may act together with ATML1 in a feedback loop ( Galletti and Ingram , 2015 ) . As previously discussed ( see Introduction ) , computational models propose that in tissues where no localized signals are present , stochastic fluctuations of transcriptional regulators create subtle differences between identical cells which initiate feedback loops including intercellular signaling to create the pattern ( Meyer and Roeder , 2014 ) . While our current model suggests that giant cell fate can be predicted through cell autonomous mechanisms , it will be interesting to see if ACR4 and DEK1 act to help establish or maintain giant cell fate or to propagate giant cell patterning in the developing sepal . To facilitate the entry into endoreduplication , ATML1 may need to activate a downstream target that only functions during G2 phase of the cell cycle . One possible ATML1 target is the Siamese-related CDK inhibitor LGO . LGO acts genetically downstream of ATML1 in the giant cell pathway to promote endoreduplication once giant cell identity is acquired ( Figure 6C; Roeder et al . , 2012 ) . It is not yet exactly understood how CDK inhibitors like LGO function in promoting endoreduplication because some evidence suggests that they interact with cyclin-CDK complexes during both G1-S and G2-M transitions , while other studies suggest specificity for G2-M ( Boudolf et al . , 2009; Churchman et al . , 2006; Kumar et al . , 2015; Van Leene et al . , 2010 ) . It is hypothesized that SIAMESE and LGO control the entry into endoreduplication by inhibiting G2-M transitions ( Kalve et al . , 2014; Roodbarkelari et al . , 2010; Van Leene et al . , 2010 ) . It will be interesting to test whether the G2 responsiveness of ATML1 arises due to direct or indirect regulation of LGO . There are a few examples that support the idea that G2 can be important for post-mitotic cell differentiation . For instance in Drosophila , changes in protein levels of the homeobox transcription factor Pax6 during the G2-M transition will cause neurogenic progenitor cells to specify into different types of post-mitotic neurons ( Hsieh and Yang , 2009 ) . Although Pax6 behaves similarly to ATML1 through controlling cell fate in a dosage dependent manner , Pax6 expression remains relatively constant in neurogenic progenitor cells until the G2/M phase . This indicates that Pax6 does not undergo random fluctuations like ATML1 , but is likely regulated by an upstream factor . Other examples of G2 mediated cell fate decisions include the development of secondary vulval precursor cells , where precursor cells require high levels of LIN-12 mediated signaling during G2 to commit to secondary cell fates ( Ambros , 1999 ) , and Drosophila mechanosensory precursor cells , where cells enter a temporary quiescence in G2 to provide a small window for proneural determinant gene products to accumulate ( Nègre et al . , 2003 ) . Although both systems use G2 as a window to initiate cell fate decisions , neither has been reported to experience fluctuations similar to ATML1 . Our theoretical model has shown that dynamic stochastic fluctuations in protein expression levels can provide a mechanism for singling out cells in the developing sepal to adopt the giant cell fate . It would be interesting to examine whether other sources of noise can shape such fluctuations and contribute to the process of giant cell fate commitment . In our giant cell patterning model , a hard threshold in the downstream target produces a soft threshold in the upstream regulator ( i . e . ATML1 ) . A soft but still reliable threshold can emerge when a target follows the dynamics of its upstream regulator . Indeed , our experimental data shows that the ATML1 threshold is soft , but robust across different plants . We have described a cell-autonomous fluctuation-driven patterning mechanism , where fluctuations of the transcription factor ATML1 must reach a concentration threshold during the G2 stage of the cell cycle to regulate cell fate decisions . This overall demonstrates that stochastic processes can be important for creating spatial patterns necessary for reproducible tissue development . Columbia ( Col ) plants were used as the wild-type accession for all genotypes except pSEC24A::H2B-GFP which was in Landsberg erecta ( Qu et al . , 2014 ) . atml1–3 ( SALK_033408 ) ; exhibits a lack of giant cell phenotype . The atml1–3 mutation is a dosage dependent mutation that contains a T-DNA insertion in the homeodomain . The atml1–3 mutation can be PCR genotyped by amplifying with oAR272 ( CAGGCAGAAGAAAATCGAGAT ) , oAR273 ( GAAACCAGTGTGGCTATTGTT ) and LBb1 ( GCGTGGACCGCTTGCTGCAACT ) . lgo-2 ( SALK_039905 ) ; exhibits a lack of giant cell phenotype . The lgo-2 mutation is a recessive mutation , containing a T-DNA insertion . The lgo-2 mutation can be PCR genotyped by amplifying with oAR284 ( CTTCCCTCTCACTTCTCCAA ) , oAR285 ( CCGAACACCAACAGATAATT ) , and JMLB2 ( TTGGGTGATGGTTCACGTAGTGGG ) ( Roeder et al . , 2010 ) . dek1–4 plants do not form giant cells . The dek1–4 mutation can be PCR genotyped by amplifying with oAR448 ( TGTTGGTGGAACAGACTATGTGAATTCA ) and oAR449 ( TGAAGACTGAAAGGACAAAAGGTGC ) with a 60°C annealing temperature followed by a 4 hr product digest using BsaAI . acr4–24 plants have a severe reduction in the number of giant cells that form . The acr4–24 mutation can be PCR genotyped by amplifying with oAR302 ( ATAGAAGTCCCTGTGAGAACTGCG ) and oAR303 ( TATGATCATAGTGCGGTCTGTTGG ) with a 60°C annealing temperature followed by a 4 hr product digest using HhaI . pAP2::AP2-2XYpet plants were provided by Jeff Long ( Wollmann et al . , 2010 ) . pVIP1::VIP-mCitrine plants were provided by the ABRC ( CS36991 ) ( Tian et al . , 2004 ) . ATML1 estradiol inducible lines were provided by Shinobu Takada ( proRPS5A-ATML1/pER8 and proATML1-nls-3xGFP ) and Keiko Torii ( pKMP151 line #134 ) ( Peterson et al . , 2013; Takada et al . , 2013 ) . All plants used for this analysis were grown in Percival growth chambers with 24 hr light conditions at 22°C to minimize any diurnal effect on plants . ATML1 , AT4G21750; giant cell enhancer trap marker , YJ158; small cell enhancer trap marker , CS70134; LGO , AT3G10525; atml1–3 , CS68906 , SALK_033408; lgo-2 , CS69160 , SALK_039905; pPDF1::GFP-ATML1 , GIL91–4; pPDF1::FLAG-ATML1 , GIL90–5; SEC24 , AT3G07100; pVIP1::VIP-mCitrine , CS36991; CER5 , AT1G51500; FDH , AT2G26250; PDF2 , AT4G04890; PDF1 , AT2G42840 . To create genetically altered lines of ATML1 for our dosage series , we first crossed PDF1::FLAG-ATML1 plants , which exhibit an all ectopic giant cell phenotype to Columbia plants , resulting in F1 plants that were hemizygous for the PDF1::FLAG-ATML1 transgene ( PDF1::FLAG-ATML1/+ ) . To lower amounts of ectopic ATML1 even further , PDF1::FLAG-ATML1/+ plants were crossed into the atml1–3 mutant background . Using genetic segregation and PCR genotyping , plants containing the PDF1::FLAG-ATML1 transgene in an atml1–3 mutant background were recovered and analyzed ( PDF1::FLAG-ATML1/+; atm1–3/atml1–3 ) . Next , to look at the effects of atml1–3 heterozygotes , the Columbia plants were crossed with atml1–3 mutants . The resulting F1 plants were analyzed . To assess whether ectopic sepal giant cells from PDF1::FLAG-ATML1 plants confer giant cell identity , PDF1::FLAG-ATML1 plants were crossed with plants expressing the giant and small cell marker ( PAR111 and CS70134; Roeder et al . , 2012 ) . Plants homozygous for all three transgenes were analyzed . To look at the effects of ATML1 in flowers that lack giant cells , pPDF1::FLAG-ATML1 plants were crossed with giant cell patterning mutants lgo-2 , acr4–24 , and dek1–4 . Genotyping PCR was used to identify plants homozygous for pPDF1::FLAG-ATML1 and either lgo-2 , acr4–24 , or dek1–4 . To see how pATML1::mCitrine-ATML1 behaved in lgo-2 mutants , pATML1::mCitrine-ATML1 plants were crossed into lgo-2 mutants . Genetic segregation analysis and confocal microscopy was used to find pATML1::mCitrine-ATML1; lgo-2 plants . Scanning electron microscopy was performed as previously described ( Roeder et al . , 2010 ) . Briefly , Stage 14 flowers were fixed in an FAA solution ( 50% ethanol , 5% acetic acid , and 3 . 7% formaldehyde ) for 4 hr and dehydrated using an ethanol series . Flowers were critical point dried and sepals were dissected . Sepals then were sputter-coated with platinum palladium and imaged using a LEICA 440 scanning electron microscope . Analysis of the giant and small cell enhancer fluorescent reporters was performed as previously described in ( Roeder et al . , 2012 ) . Stage 12 medial abaxial sepals were stained with Propidium Iodide ( PI ) and imaged with a Zeiss 710 laser scanning confocal microscope . The small cell marker was excited with a 488 nm laser and emission was collected with a 493–516 filter whereas the giant cell enhancer was excited with a 514 nm laser and emission was collected with a 519–565 filter . PI emission was collected with a 599–651 filter . Images were taken with a 10x objective . pSEC24A::H2B-GFP was imaged using a Zeiss 710 laser scanning confocal microscope . The GFP marker was excited with a 488 nm laser and collected with a 493–548 filter . Nuclear fluorescence was then calculated using our quantification pipeline . pVIP1::VIP1-mCitrine was imaged using a Zeiss 710 laser scanning confocal microscope . The mCitrine marker was excited with a 514 nm laser and collected with a 519–564 filter . VIP1 is a bZIP transcripton factor that is cytoplasmically localized under stable conditions but will become nuclear localized upon hypoosmotic treatment ( Tsugama et al . , 2016 ) . To nuclear localize VIP1 , VIP1-mCitrine inflorescences were submerged in a hypoosmotic solution ( H2O and 0 . 001% triton-X ) for approximately 10 min prior to confocal imaging . Nuclear fluorescence was then calculated using our quantification pipeline . pAP2::AP2-2XYpet was imaged using a Zeiss 710 laser scanning confocal microscope . The 2XYpet marker was excited with a 514 nm laser and collected with a 519–564 filter . Nuclear fluorescence was then calculated using our quantification pipeline . Live imaging of each fluorescent reporter line in developing sepals was performed as previously described ( Roeder et al . 2010 ) , except for the experimental setup . Transgenic plants including pHM44 pATML1::mCitrine-ATML1 ( ex . 514 nm at 2% , em . 519–564 nm ) , lgo;pHM44 pATML1::mCitrine-ATML1 ( ex . 514 nm at 2–2 . 2% , em . 519–564 nm ) or GIL91–4 pPDF1::GFP-ATML1 ( ex . 488 nm at 1–1 . 5% , em . 493–598 ) were imaged either every 8 hr or every hour using a Zeiss 710 laser scanning confocal microscope with a 20x water-immersion objective ( numerical aperture = 1 . 0 ) . Before imaging , plant inflorescences were dissected down to early stage flowers and meristems and then taped onto slides . Dissected inflorescences were then stained with PI and mounted with a cover slip and imaged . Inflorescences were unmounted , dried , and plants were placed upright in the growth chamber for 8 hr before remounting and imaging . The resulting images were 3D cropped with ImageJ ( Schindelin et al . , 2012; Schneider et al . , 2012 ) to remove neighboring flowers . mCitrine-ATML1 fluorescence was quantified in each nucleus throughout the live imaging series with our pipeline ( see below ) . Flow cytometry was conducted as previously done in ( Roeder et al . , 2010 ) using an Accuri C6 flow cytometer . 50–100 stage 12 sepals were dissected from transgenic plants containing epidermal GFP-tagged nuclei ( pAR180 pML1::H2B-mGFP ) . Nuclei were stained with PI and gated as described previously ( Roeder et al . , 2010 ) to isolate epidermal nuclei ( GFP positive ) from internal tissue nuclei ( GFP negative ) . PI fluorescence histograms showed the relative DNA content of each population analyzed . Ploidy and nuclear area were quantified from DAPI stained sepals as previously described ( Roeder et al . , 2010 ) and imaged with a Zeiss 700 . DAPI was excited with a 405 nm laser and emission collected with a 410–584 nm filter . Images were cropped in ImageJ and quantified using ImageJ or our quantification pipeline . Cell size analysis was performed by imaging pAR169 ( pML1::mCitrine-RCI2A ) sepals with a Zeiss 710 confocal microscope . mCitrine was excited with a 514 nm laser and emission was collected with a 519–621 filter . Imaged sepals were semi-automated image processed using a MATLAB module , which has been previously published ( Cunha et al . , 2010; Roeder et al . , 2010 ) to determine cell area . To create pHM44 ( pATML1::mCitrine-ATML1 ) , a 6160 bp fragment upstream of the ATML1 protein coding region was PCR amplified using oHM23 ( ACCGACAATGTATGAATGTACTCT ) and oHM24 ( cggtaccggcgcgccGATGATGATGGATGCCTATCAATTT ) and cloned into a pGEM-T Easy vector to create pHM20 . Additionally , a 992 bp region downstream of the ATML1 protein coding region was PCR amplified using oHM25 ( cggtaccTCGATGTTTTCGGGTAAGCTTTTT ) and oHM26 ( TTTGATGACTTGGTCTCCATAATTTC ) and cloned into pGEM-T easy to create pHM21 . pHM21 was cut with SacII and KpnI and cloned into pHM20 to make pHM22 . A gateway cassette from pXQ ( AscI-GW-KpnI in pGEM-T easy ) was cut with AscI and KpnI and cloned into pHM22 to make pHM23 ( Qu et al . , 2014 ) . Then , pHM23 was cut with NotI and cloned into the pART27 binary vector to make pHM43 ( pATML1::GW:ATML1 3’UTR ) . Next , mCitrine was PCR amplified using oHM42 ( CACCAAAATGGTGAGCAAGGGCGAGGAGCTG ) and oHM39 ( atACTAGTGGCCGCTGCCGCAGCGGCAGCCGCAGCTGCTCCGGACTTGTAC ) and cloned into pENTR/D-TOPO vector to make pHM30 . ATML1 was PCR amplified using oHM40 ( tcggcgcgccCACCCTTTTAGGCTCCGTCGCAGGCCAGAGCGGCT ) and oHM41 ( ccactagtATGTATCATCCAAACATGTTCGAATCTCATC ) and cloned into pGEM-T easy to make pHM28 . ATML1 was cut using SpeI and AscI and cloned into pHM28 to make pHM25 ( pENTR mCitrine-ATML1 ) . LR reaction between pHM25 and pHM43 to make pHM44 ( pATML1::mCitrine-ATML1 ) . The atml1–3 rescue line was generated by transforming the pHM44 pATML1::mCitrine-ATML1 transgene into atml1–3 mutants using Agrobacterium-mediated floral dipping methods ( Clough and Bent , 1998 ) . We recovered lines with varying numbers of giant cells , presumably due to varying levels of transgene expression . From the lines recovered , two produced the wild-type number of giant cells , rescuing the mutant phenotype . Both of these lines showed differing levels of ATML1 among cells . Therefore , we characterized one of them . To test whether ATML1 acts in a feedback loop , inflorescences of ATML1-estradiol inducible plants ( proRPS5A-ATML1/pER8 and proATML1-nls-3xGFP line #7 provided by Takada et al . , 2013 ) were cultured in apex culturing media ( 1/2x MS , 1% sucrose , 0 . 5 g/L MES , pH 5 . 7 , 0 . 8% agar; Hamant et al . , 2013 ) containing either 0 . 1 µM , 1 µM , or 10 µM estradiol or a mock solution ( ethanol equivalent to the solvent of estradiol ) . Tissue was then collected 48 hr later and prepared for qPCR . Three or five biological replicates were analyzed for each treatment ( estradiol and mock ) . To test whether inducing ATML1 could increase the number of giant cells that form on the sepal , we dipped inflorescences expressing proRPS5A-ATML1/pER8 and proATML1-nls-3xGFP provided by Shinobu Takada ( Takada et al . , 2013 ) in 10 µM estradadiol ( with 0 . 01% silwet ) for three consecutive days and then examined seven sepals ( stage 8–10 ) five days later and compared them to untreated sepals at equivalent developmental stages . To perform qPCR , 3–4 inflorescences were collected per sample and total RNA was extracted using RNeasy Plant Mini Kit ( Qiagen , Venlo , Netherlands ) . Next , 1 microgram of total RNA was DNAse treated with amplification grade DNAse I ( Invitrogen , Carlsbad , USA ) and reverse transcribed using Superscript II reverse transcriptase ( Invitrogen ) with oligo dT primers . Real-time PCR was performed using 480 SYBR Green I Master ( Roche , Indianapolis , IN ) on a Roche LightCycler 480 system . At least three biological replicates were analyzed per genotype and ROC1 ( AT4G38740 ) was used as a reference gene to normalize gene expression . Furthermore , three technical replicates were used to ensure the validity of each biological replicate . qPCR primers: In order to accurately quantify mCitrine-ATML1 levels at the single cell level and track individual cells during sepal growth , we developed an integrated image analysis pipeline incorporating modules from different available sources . Raw fluorescence intensity images were denoised using the PureDenoise ImageJ plugin ( Blu and Luisier , 2007; Luisier et al . , 2009 , 2010 ) , optimized for the mixed Poisson-Gaussian noise that typically affects fluorescence microscopy images ( parameters: frames = 4; cycle spins = 3 ) . Denoised images were imported into MorphoGraphX ( Barbier de Reuille et al . , 2015 ) in order to produce binary masks for individual sepal nuclei while simultaneously removing non–relevant meristematic and border cell nuclei ( parameters: brighten/darken: 1–4; Gaussian Blur: 0 . 3–1; Binarize: 5000–8000 ) . Since different genotypes show different proportions of giant and small cells , and segmentation parameters are globally applied to the whole tissue , slight adjustments were made for each genotype in order to fit the binary masks as well as possible to all nuclei across all genotypes . For each individual time course , parameter values were kept constant for all time points . Binary mask images were used as input for the final nuclear segmentation , performed with the Costanza ( COnfocal STack ANalyZer Application ImageJ plugin ( http://www . plant-image-analysis . org/software/costanza ) . Costanza performs segmentation following the steepest descent algorithm , providing high-resolution three-dimensional segmentation of each individual sepal nucleus . Denoised images were processed using the FeatureJ ImageJ plugin ( http://imagej . net/FeatureJ ) for edge detection by applying the Canny method ( parameters: gradient-magnitude image smoothing scale = 0 . 25 ) . To track the cell nuclei between two successive nuclei segmentations , Nt and Nt+Δt ( where Δt corresponds to the time interval between two consecutive acquisitions ) , the block matching framework ( Michelin et al . , 2016 , Commowick et al . , 2008; Ourselin et al . , 2000 ) was used to non-linearly register the corresponding denoised images , It and It+Δt , ( floating and reference images respectively ) . The registered floating image and the reference image were merged with different colors into a double channel image in ImageJ ( Box 1 , 3D projection of the merged image , red: reference image , green: registered floating image ) . This allowed a visual inspection of registration quality . The non-linear transformation computed by block matching , TIt ← It+Δt , was then applied to Nt ( i . e . Nt ○ ( TIt ← It+Δt ) . Using ALT ( Fernandez et al . , 2010 ) we computed optimal cell-cell pairing between Ntand Nt ○nTIt ← It+Δt . Given the spatial complexity of the tissue and the large time interval between consecutive images ( Δt=8 hr ) , registration was not always successful for all nuclei . Incorrectly tracked nuclei were manually corrected using the MorphoGraphX parent labels tools , making use of the ALT-generated optimal pairing tables , describing the mother/daughter relations between time points . A set of MATLAB ( The MathWorks , Inc . , Natick , Massachusetts , United States ) functions and scripts was developed to quantify signal intensity , as well as size and shape properties of individual nuclei from sets of confocal microscopy images processed as described above ( Source code 1 ) . We did not use a secondary nuclear marker to detect nuclear size because mCitrine-ATML1 levels are low and may experience bleed through from a nuclear marker in a different channel . Additionally , given imaging artifacts observed when using three-dimensional images , which include extension of nuclei in the Z-axis , we chose to perform quantification in two-dimensional images , in order to maximize result accuracy . Two-dimensional nuclei were obtained by scanning , for each nucleus , through each individual Z slice of the Costanza-segmented images and selecting the slice with the largest area , where segmentation is most accurate . For all 2D nuclei , shape parameters such as eccentricity were quantified using the regionprops function in the MATLAB Image Processing Toolbox , which was also used to quantify areas . For each 2D nucleus , absolute fluorescence intensity was quantified by summing intensity of all pixels in the respective region of the raw intensity image . Both absolute intensity and area were corrected for possible magnification changes during the time course by taking into account pixel sizes , and concentrations were calculated based on the corrected absolute intensity and area values . ATML1 concentration , area and eccentricity plots for all cells in the time course were generated with custom functions that make use of the corrected parental correspondence information . From the complete set of tracked lineages , we selected for lineages that exhibited high quality segmentation and tracking data that allowed us to follow a given cell either until the last point of the time course , or a until fate became apparent ( division or endoreduplication; see Supplemental file 1 for examples ) . Calculations were performed using the perfcurve function of the Statistics and Machine Learning Toolbox in MATLAB ( Source code 1 ) . Classes were defined based on their final identity ( small or giant ) and cell cycle stage ( 2C for G1 or 4C for G2 ) . After ploidy was assigned , we identified peaks in mCitrine-ATML1 concentration at G1 and G2 stages of the cell cycle for ROC analysis . Individual cell lineages were included in the analysis only if a cell passed through both the G1 and G2 stages of the cell cycle before entry into either mitosis or endoreduplication or was first detected in G2 and remained in G2 for more than two consecutive time points before entry into mitosis or endoreduplication . For these lineages we used the highest concentration level of mCitrine-ATML1 during both the G1 and G2 stages of the cell cycle before entry into either mitosis ( small cell ) or endoreduplication ( giant cell ) . For each sepal , the ATML1 concentration value that maximized the difference between true positive rate ( TPR ) and false positive rate ( FPR ) when classifying small versus giant cells , was taken as the threshold ATML1 concentration required for triggering giant fate decision in individual cells . For sepals where such a threshold could not be inferred , whether due to the absence of sufficient numbers of dividing ( pPDF1::GFP-ATML1 flowers ) or endoreduplicating cells ( lgo-2 mutant flowers ) , a common threshold was inferred from the wild-type pATML1::mCitrine-ATML1 flowers ( Figure 4—figure supplement 5 ) . For each of the four analyzed sepals , ATML1 concentrations were normalized by dividing by the mean concentration for all nuclei , over the entire time course . This mean normalization had the objective of taking into account systematic differences between time courses due to experimental variation . After normalization , ATML1 concentration peak selection , ROC analysis and concentration threshold inference were performed as described above . The final common ATML1 concentration threshold was defined as the mean of the four individual thresholds . In the lgo-2 mutant sepals , giant cell fate prediction was performed by comparing the normalized ATML1 concentrations for each lineage ( calculated using the mean concentration for all nuclei , over the entire time course ) with the previously inferred common threshold . A lineage was considered to be a giant cell lineage if the ATML1 concentration of a given nucleus surpassed the common threshold concentration during 4C at any point of the time course . If this event never occurred , the lineage was considered to correspond to a small cell lineage . We implemented a stochastic computational model for ATML1 mediated giant cell fate decisions in a 2D idealized growing tissue . The model has a core simplified ATML1 regulatory network that can prevent cell division , driving cell endoreduplication . We modeled ATML1 cell concentration dynamics as a basally produced protein that self-activates and is linearly degraded ( Frigola et al . , 2012; Weber and Buceta , 2013 ) . ATML1 expression activates a downstream target , which can prevent cell division when expression passes a threshold . The deterministic expression for the dynamics of ATML1 and its downstream target concentrations in cell i , whose variables are [ATML1]i and [Target]i respectively , reads ( 1 ) d[ATML1]idt=PA+VA[ATML1]inAKAnA+[ATML1]inA−GA[ATML1]i ( 2 ) d[Target]idt=VT[ATML1]inTKTnT+[ATML1]inT−GT[Target]i , where PA is a basal ATML1 production rate , VX is the maximal ATML1-dependent production rate for the X ( either ATML1 or Target concentration ) variable , KX is the ATML1 concentration at which the ATML1-dependent production rate has its half-maximal value , nX is the Hill coefficient and GX is the linear degradation rate for the X variable . For simplicity , we will refer to VA as the ATML1 auto-induction rate , so no feedback is considered when VA = 0 . A cell is defined by a set of vertices in 2D and we set the tissue to grow exponentially and anisotropically by moving vertices outwards from the center of mass of the tissue . Hence , all cells grow anisotropically , and they divide according to a timer variable present in each cell . We implemented dilution of ATML1 and its target variables due to growth . During sepal development , nuclear and cell area of epidermal cells are correlated ( Figure 4—figure supplement 4 ) . We used cell area growth to implement the dilution effect into the ATML1 and target variables . The timer linearly increases with time and is reset when it reaches a specific threshold ( Figure 7A ) . Hence , its equation reads ( 3 ) dTimeridt=PC , where Timeri is a variable in cell i , and PC is the basal timer production rate . The timer resetting was performed at each time step according to the following equation: ( 4 ) Timeri ( t ) →{Uiif Timeri ( t ) ≥ΘC , DTimeri ( t ) otherwise , where Ui is a uniform randomly distributed number in the interval [0 , 0 . 5 ) and ΘC , D is a cell division threshold for the timer . Cell ploidy was modeled as a discrete variable dependent on the timer and cell division , which also depends on the ATML1 network . Specifically , cell ploidy increases from 2C to 4C when the timer reaches a threshold ΘC , S , which represents S phase , and decreases again to 2C if the cell divides . Cell division occurs at the 4C stage , when the timer reaches a second threshold ΘC , D , unless cells have reached [Target] levels higher than a specific threshold ΘT during the 4C stage . In that case , endoreduplication occurs , and cells reset their timer when they reach the ΘC , D threshold , but keeping its ploidy to 4C . We imposed that 4C cells having endoreduplicated once cannot undergo cell division anymore . As a consequence , these cells will increase their ploidy every time they pass the timer threshold ΘC , S representing entry into S phase . Our experimental data shows that the nuclear area scales with the DNA content and ploidy in the cell ( Box 2; Box 2—Figure 1; Box 2—Videos 1–2 ) . Previous data in tomato has shown that expression levels positively correlate with cell ploidy ( Bourdon et al . , 2012 ) , so one could also assume there is a linear correlation between ploidy and expression levels . Because of these two assumptions , the production rates of the ATML1 and Target concentration variables become independent of the cell ploidy . For the sake of simplicity , production rates remain constant throughout cell cycles . Dynamic stochasticity was introduced in the ATML1 , Target and Timer variables by extending its deterministic dynamics to its Langevin form ( Adalsteinsson et al . , 2004; Gillespie , 2000 ) . In particular , for every ATML1 , Target and Timer variable X in cell i , the resulting stochastic equations would read ( 5 ) dXidt=FXi+−FXi−+FXi++FXi−2εi ( t ) ηXi ( t ) , where Fxi+ and Fxi– are positive functions that represent the birth and death processes for the species X in cell i . Hence , we take into account stochasticity coming from production and degradation of the modeled species . εi ( t ) is a normalized cell area; we assume εi ( t ) =E0Ei ( t ) , where E0 is an effective cell area used to normalize noise , and Ei ( t ) is the area of cell i in arbitrary units . ηXi is a random Gaussian variable with zero mean that fulfills ⟨ηXi ( t ) ηX’j ( t’ ) ⟩=δ ( t-t’ ) δXXδij , where i and j are cell indices , X and X’ the modeled variables , δXX and δij are Kronecker deltas and δ ( t-t’ ) is the Dirac delta . Note that , as the standard chemical Langevin equation ( Gillespie , 2000 ) , Equation 5 recovers the deterministic limit when the cell sizes go to infinity . Due to the presence of stochasticity and the fact of having a target that is able to follow the dynamics of its upstream regulator , the threshold on the target ΘT results in a soft threshold on the ATML1 variable ( see Figure 7 ) . A soft threshold means that there is a range of ATML1 values in which a cell being in 4C will be likely to prevent mitosis , and therefore , become giant . The higher the ATML1 value the cell has in this range , the more likely will for a cell to become giant . Integration of the resulting Langevin equations with the Îto interpretation was performed by using a variation of the Heun algorithm ( Carrillo et al . , 2003 ) with an absorptive barrier at 0 to prevent negative values of the modeled variables . Growth and its dilution-derived effects were considered deterministic , and were integrated with an Euler algorithm . The integration time step was set to dt = 0 . 1 . Note that stochasticity was also introduced in the initial conditions of the modeled variables and when resetting the timer variable after cell division ( Equation 4 ) . Cells divide according to a shortest path rule in which the new wall pass through the center of mass of the dividing cell ( Sahlin and Jönsson , 2010 ) . Daughter cells have the same initial ATML1 and Target concentrations at birth , but can have different sizes . After dividing , these cells will acquire different initial timer values due to the noise term in Equation 4 . For the sake of simplicity , no mechanical interactions were implemented to the simulated tissue . Unless otherwise stated , simulation parameters were set as described in Table 1 . We set uniformly distributed random initial conditions for ATML1 and Target variables within the interval [0 , 1 ) and [0 , 0 . 1 ) , respectively . Timer initial conditions were set in correlation to the cell size of the initial template , following the expression . ( 6 ) Timeri ( t=0 ) =0 . 8 θC , DEMax−EMin ( Ei ( t=0 ) −EMin ) +0 . 1 θC , D ( 1− Ui′ ) \ , being U’i an uniformly distributed random number defined in the interval [0 , 0 . 1 ) , and EMin and EMax the minimal and maximal areas of the cells at the start of the simulation . This made larger cells being initiated at more advanced stages of the cell cycle , and hence , being more likely to divide . Ploidies were initially set to either 2C or 4C , depending on whether the initial timer values set by Equation 6 were lower or higher than the S-phase timer threshold ΘC , S . 10 . 7554/eLife . 19131 . 045Table 1 . Main parameter values used for simulations in Figures 7 and 8 and Figure 7—supplements 1–4 . We omit time and concentration units , since all are considered arbitrary . DOI: http://dx . doi . org/10 . 7554/eLife . 19131 . 045ParameterDescriptionValuesPA ATML1 basal production rate 1 . 14VA ATML1 auto-induction rate1 . 25KA ATML1 concentration for half ATML1 auto-induction maximal rate1 . 9nA Hill coefficient for ATML1 auto-induction5GA ATML1 degradation rate1VT Target maximal production rate 10KT ATML1 concentration for half ATML1-mediated target maximal production rate 2nT Hill coefficient for ATML1-mediated target induction1GT Target degradation rate10ΘT Target threshold for inhibiting mitosis0 . 6ΘC , S Timer threshold for synthesis2ΘC , D Timer threshold for timer resetting3PC Timer basal production rate 0 . 1E0Characteristic effective volume15Exponential radial growth rate0 . 007Exponential added growth rate to the vertical direction0 . 012 We assigned different parameter values based on experimental evidence when available . Threshold values of the timer for the synthesis phase and division checkpoint ( ΘC , S and ΘC , D respectively ) were assigned so that we could recover 2C and 4C percentages of cells in atml1–3 mutants ( Figure 2G ) in regions of the parameter space in which no giant cells were formed . Given the chosen timer threshold values and an arbitrary basal timer production rate , simulations were integrated throughout 105 arbitrary time units , so that cells could undergo around three cell cycles ( Roeder et al . , 2010 ) . Simulations scanning the parameter space were performed by using logarithmic spaced values of the ATML1 basal production rate ( PA ) and linearly spaced values of the ATML1 auto-induction production rate ( VA ) . Specifically , we performed simulations on 121 logarithmically spaced PA values between 0 and 2 . 2 , and 11 linearly spaced VA values between 0 and 2 . 5 . From these parameter scans , PA and VA parameters were chosen for the simulations shown in Figure 7 and Figure 7—figure supplement 4 . PA and VA parameter values for representing the wild-type sepal in Figure 7 were chosen so that there was a unimodal distribution of ATML1 concentration with similar CVs to the experimental CVs , giving rise to the same number of giant cells found in developing sepals . In particular , we aimed to have sepals that developed a total of 30 giant cells with 8C and higher ploidy , with approximately 17 of those cells being 16C and higher ploidies ( see Figure 2H ) . To ensure that the target approximately followed and mimicked the dynamics of ATML1 , we simulated a target with a higher degradation rate than ATML1 itself . To grow the sepal in a realistic manner , we provided a certain degree of anisotropy on the tissue growth parameters , as previously reported experimentally ( Hervieux et al . , 2016 ) . The computational implementation of the model was performed through the open source C++ Organism package , ( http://dev . thep . lu . se/organism/; Bozorg et al . , 2014; Jönsson et al . , 2006 ) . Data analysis and plots from simulation output were performed with Python 2 . 7 , the Matplotlib package ( Hunter , 2007 ) and MATLAB . See Source code 2 for further details on the implementation of the model and the analysis of the simulated data . The visualization of the simulated growing sepals was performed with Paraview software ( http://www . paraview . org ) . ROC analysis was also applied to the ATML1 concentration maxima across the different simulated lineages , by following a similar procedure as for the experimental data ( see Receiver operator characteristics ( ROC ) analysis section and Source code 2 for details ) . Classes were also defined based on their final identity; lineages having 2C ploidy at the end of the simulation were considered small cells , while lineages having 8C ploidy or higher were considered giant cells . Lineages remaining in 4C ploidy at the end of the simulated time course were excluded of the analysis , given their unknown final fate . The soft ATML1 threshold ΘA* was determined by finding the threshold assigned to the optimal ( maximized difference between TPR and FPR ) operating point of the ROC curve . Specifically , we used 30 different random subsamples of the small cell population with as much cells as the pool of giant cells , so that the total cost of misclassification of positive and negative cases for the threshold determination would remain equivalent and similar to the experimental analysis . As a result , the computed soft threshold ΘA* was defined as the mean of the 30 different optimal thresholds found using random subsamples . This subsampling method , when applied to the target maxima throughout 4C time courses , could accurately predict the hard threshold of the target variable imposed in the simulations ΘT , which we denote by ΘT* ( Figure 7—figure supplement 3B , E ) . We represented the predicted thresholds as a dashed red line within a red shaded red region . This red region shows the standard deviation of the 30 optimal thresholds computed in the subsampling method . Note that sometimes the shaded red region is too small to be seen ( e . g . see ΘA* in Figure 7C ) .	Plant and animal organs contain several types of cells that perform different roles . As a plant or animal develops , these different cell types can form intricate patterns . To start the pattern , a few cells within a group of identical cells must somehow become different from their neighbors . Some patterns of cells are very well organized and easily reproduced . However , sometimes cells spontaneously become different from their neighbors , producing a less ordered pattern . In a plant called Arabidopsis ( commonly known as Thale cress ) , a scattered pattern of giant cells and small cells spontaneously forms within a part of the developing flower called the sepal . A protein called ATML1 is a key regulator in the formation of giant cells , but because it is found in both giant cells and small cells , it is not clear how this regulation works . Mathematical models of this process suggest that identical cells could initially acquire subtle differences , potentially from random fluctuations in the activity of key regulatory molecules , to start the patterning process . Meyer , Teles , Formosa-Jordan et al . used a combination of microscopy , image analysis and mathematical modeling to investigate how the level of ATML1 fluctuates in cells to give rise to the pattern within the sepal . The experiments show that early in the development of the sepal , the levels of ATML1 fluctuate up and down in every sepal cell . If ATML1 reaches a high level specifically when a cell is preparing to divide , that cell will decide to become a giant cell , whereas if the level of ATML1 is low at this point , then the cell will divide and remain small . Overall , the findings of Meyer , Teles , Formosa-Jordan et al . demonstrate that fluctuations of key regulators while cells are preparing to divide are important for creating patterns during development . A future challenge is to examine whether other tissues in plants , or tissues in other organisms , use a similar mechanism to generate patterns of cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2017	Fluctuations of the transcription factor ATML1 generate the pattern of giant cells in the Arabidopsis sepal
Exceptionally high rates of tooth fracture in large Pleistocene carnivorans imply intensified interspecific competition , given that tooth fracture rises with increased bone consumption , a behavior that likely occurs when prey are difficult to acquire . To assess the link between prey availability and dental attrition , we documented dental fracture rates over decades among three well-studied populations of extant gray wolves that differed in prey:predator ratio and levels of carcass utilization . When prey:predator ratios declined , kills were more fully consumed , and rates of tooth fracture more than doubled . This supports tooth fracture frequency as a relative measure of the difficulty of acquiring prey , and reveals a rapid response to diminished food levels in large carnivores despite risks of infection and reduced fitness due to dental injuries . More broadly , large carnivore tooth fracture frequency likely reflects energetic stress , an aspect of predator success that is challenging to quantify in wild populations . Competition for food among large carnivores is expected to increase when dietary overlap is high and the availability of preferred prey declines . Under such conditions , large carnivores are likely to spend more time feeding on their kills and consume them more completely , as well as scavenge more often . Greater carcass utilization often involves the consumption of less nutritious parts of a kill including bone , a behavior that increases tooth wear and the probability of tooth fracture , and might reflect higher levels of energetic stress . The association between tooth wear and bone consumption suggests that prey availability , carcass utilization , and tooth fracture are linked . When prey is difficult to kill , fewer prey are taken , and therefore carcasses might be more fully utilized with more bone consumed , and consequent increases in tooth fracture . If so , then rates of tooth fracture in large carnivores could be used as an index of prey availability in modern , historic , and ancient ecosystems . Studies of tooth fracture frequencies in large Pleistocene carnivorans have revealed substantially higher rates of tooth breakage and wear in multiple species in both the Old and New Worlds . For example , the mean rate of tooth fracture on a per tooth basis is 2 . 3 ± 1 . 3% among 13 extant large ( >21 kg ) felids , canids , and hyaenids , whereas the same figure is 8 . 1 + 3 . 5% for a sample of five large North American late Pleistocene felids and canids that represent distinct geographic locations ( Alaska , California , Mexico , Peru ) ( Van Valkenburgh , 2009 ) . Similarly , a study of temporal variation in dental wear and breakage in Pleistocene gray wolves of Great Britain found that per tooth fracture rates ranged from 2 . 5% to 8% , and noted that the higher rate was associated with lower prey diversity and the presence of a likely competitor , the brown bear ( Ursus arctos ) ( Flower and Schreve , 2014 ) . In both studies , the authors suggested that elevated tooth fracture frequencies in large Pleistocene carnivorans reflect intensified competition for kills and consequent increases in heavy carcass utilization and scavenging . However , the inference of an association between prey availability , competition , increased bone consumption and higher tooth fracture rates is complicated by several factors . First , because the probability of having at least one broken tooth increases with age ( Van Valkenburgh , 2009 ) , populations dominated by older individuals are likely to have higher rates of tooth fracture and heavier tooth wear than those dominated by younger individuals . Most museum collections of skulls do not have associated age data so it is difficult to control for this potential bias . Second , museum collections almost never have data on levels of prey availability or carcass consumption behavior for the sampled predators . Third , the most commonly broken teeth are canines , and these teeth have multiple functions as weapons in combat and predation as well as in feeding . Consequently , increases in canine tooth breakage are more problematic to interpret than increases in premolar or molar breakage . To better understand the causes of variation in tooth fracture frequency among fossil and living large carnivorans , we collected dental wear and fracture data from gray wolves ( Canis lupus ) representing three well-studied populations that had associated data on prey availability , and in some cases the degree of carcass consumption and age of death for each wolf . These include samples of wolves from Isle Royale National Park ( USA ) , Yellowstone National Park ( USA ) and Scandinavia . In both Isle Royale and Scandinavia , greater than 90% of the prey of wolves is moose ( Alces alces ) , but the mean ratio of moose to wolves is 499:1 in Scandinavia and only 55:1 in Isle Royale ( Sand et al . , 2012 ) . Kill rates ( #kills/wolf/day ) are three times greater in Scandinavia than Isle Royale ( Sand et al . , 2012 ) , and in association with this , Scandinavian wolves tend to consume less of their kills ( about 70% ) than Isle Royale wolves ( about 90% ) ( Vucetich et al . , 2012 ) . Consequently , wolf tooth wear and fracture rates are expected to be greater in Isle Royale than Scandinavia . In Yellowstone ( YNP ) , over 90% of the prey are elk ( Cervus canadensis ) ( Metz et al . , 2012 ) , and the ratio of elk to wolves has declined sharply from over 600:1 to around 100:1 since the initial reintroduction of wolves in 1995 ( Figure 1 ) . As the availability of elk declined in Yellowstone over the last 20 years , it is possible that carcass consumption levels increased , as well as rates of tooth wear and fracture . Because age at death was recorded for most of the preserved Yellowstone wolves , we can compare tooth wear and fracture frequency between wolves of similar age during times of prey abundance and times of relative prey scarcity . The three wolf populations , Isle Royale , Scandinavia , and Yellowstone provide windows into three different but overlapping scenarios of prey availability with consequent implications for dental attrition , one with abundant prey ( Scandinavia ) , a second with more limited prey ( Isle Royale ) , and the special case of Yellowstone in which there is the potential to track predator tooth damage alongside a decline in prey numbers . The tooth wear and fracture data from these three populations were also compared with data collected previously from historic populations of 223 North American gray wolf skulls of known provenance but without associated data on prey availability or feeding behavior . Previous work on tooth wear and fracture in mammals has focused on the relationship between diet and dental attrition . Early experimental studies on laboratory populations of rodents ( Carlsson et al . , 1966 ) , primates ( Teaford and Oyen , 1989 ) , and carnivorans ( Berkovitz and Poole , 1977 ) established the fact that abrasive diets that include grit or other tough materials increase rates of tooth wear relative to diets composed of relatively soft foods . Studies of wild populations are rare , but a recent comparison of tooth wear and fracture in wild coyotes relative to a matched age sample of captive-reared coyotes fed a relatively soft diet lacking any bones found more rapid rates of wear and higher tooth fracture in the wild sample ( Curtis et al . , 2018 ) . The authors suggested that this likely reflected a greater inclusion of bone in the wild coyotes’ diets , but behavioral data to confirm this were not available . In all these studies , the primary question has been the association between dental attrition and diet , with little or no consideration of how tooth wear might provide a window into levels of food stress or competition . Our comparison of dental attrition and carcass utilization in three wild populations of gray wolves expands the application of tooth fracture analysis to larger ecological questions , such as variation in the intensity of intra- and interspecific competition within carnivore guilds . The percentage of individuals with at least one tooth broken in life averaged 51% across all five samples , but ranged from less than 38% in the Scandinavian and early Yellowstone ( 1995–2006 ) populations to 72% in Isle Royale wolves ( Table 1 ) . The Isle Royale and later Yellowstone samples did not differ significantly from one another in the fraction of individuals with at least one broken tooth , 72% and 64% , respectively ( p=0 . 3 ) . Both of these samples exhibited significantly more individuals with broken teeth than either of the samples with known high prey:predator ratios , Scandinavia and early Yellowstone ( p<0 . 001 ) . Notably , the early and later Yellowstone samples differed significantly in the percentage of individuals with at least one broken tooth with and without including canine teeth ( p=0 . 001 ) ; the fraction was 38% with canines ( 35% without ) between 1996 and 2006 , whereas in the following decade it rose to 64% with canines ( 63% without ) . It should be noted that the number of broken teeth per individual in the later Yellowstone sample is significantly greater than in the earlier sample but still usually represents less than ten percent of the entire tooth row ( e . g . less than five of the 42 teeth per individual ) . Results were similar for fracture frequency on a per tooth basis . Isle Royale wolves displayed the highest fracture frequency ( 8 . 6% ) and Scandinavian and early Yellowstone wolves the lowest , 1 . 7% and 1 . 8% , respectively ( Table 1 ) . Although the fracture rates per individual were similar between Isle Royale and later Yellowstone wolves , Isle Royale wolves exhibit significantly more tooth fracture on a per tooth basis , 8 . 6% as opposed to 4 . 6% in later Yellowstone wolves ( p<0 . 001 ) . As was the case for breakage per individual , later Yellowstone wolves significantly exceeded early Yellowstone wolves in fracture frequency ( 4 . 6% as opposed to 1 . 8%; p<0 . 001 ) when canines were included and also when only incisors and cheek teeth were considered ( 3 . 9% vs . 1 . 3%; p<0 . 001 ) , thus removing any bias due to a prevalence of canine tooth fracture . Within each of the five samples , tooth fracture frequencies calculated with and without canine teeth were very similar and did not differ significantly ( p=0 . 1–0 . 5 , Table 1 ) . The differences in fracture frequency among the five samples were not due to higher fracture frequency at any single tooth position such as the canines . Instead , elevated fracture frequencies usually occurred across the entire tooth row . For example , the two samples with the highest fracture frequencies , Isle Royale and later Yellowstone , broke all their teeth , incisors , canines , premolars , carnassials and post-carnassial molars , more often than was observed in the remaining three samples ( Figure 2 , top; Supporting Information Supplementary file 1 Table S2 ) . Differences between these two samples and the remaining three were significant ( p<0 . 05 ) at all tooth positions except the canines in the case of the two Yellowstone samples . Notably , the two Yellowstone samples differed most in the fracture frequencies of all teeth except the canines . Whereas canine tooth fracture rate increased approximately 50% between the first and second decade , fracture rates for incisors , premolars and carnassials more than tripled ( Figure 2 , top; Supporting Information Supplementary file 1 Table S2 ) . Relative to both Scandinavian and other North American wolves , early Yellowstone wolves were unusual in exhibiting relatively high canine tooth fracture incidence along with low carnassial and post-carnassial molar fracture incidence . The high fracture frequencies for Isle Royale and later Yellowstone wolves are associated with a greater percentage of individuals with moderate and heavy tooth wear ( Figure 2 , middle; Supporting Information Supplementary file 1 Table S2 ) . Over half ( 56%–64% ) of the Scandinavian , early Yellowstone and other North American wolf samples , respectively , were classified as slight wear . By contrast , only 2% ( 1/64 ) of Isle Royale individuals fell into the slight category , and just 29% ( 24/83 ) of the later Yellowstone fell into the slight category ( Supporting Information Supplementary file 1 Table S2 ) . Because wear stage is positively correlated with an individual’s age , the preponderance of moderate and heavily worn individuals and associated higher tooth fracture rates within Isle Royale and the later Yellowstone samples might reflect a bias toward older individuals in these two samples rather than differences in diet or feeding behavior . Fortunately , it is possible to control for this potential bias because both the Yellowstone and Isle Royale samples have associated age data . Analysis of the age distributions for the two Yellowstone and Isle Royale samples does reveal significant differences in the predicted direction . Both the Isle Royale and later Yellowstone sample have a greater proportion of older ( >4 years ) individuals than the early Yellowstone sample ( p<0 . 001; Figure 2 , bottom ) . Nevertheless , when rates of tooth wear are compared within similar age classes ( 1–3 years , 4–6 years , and 7+ years ) , it is apparent that both the Isle Royale and later Yellowstone samples wore their teeth more rapidly as they aged ( Figure 3 ) . For example , the early Yellowstone sample had individuals with slight wear in all three age groups , whereas the later Yellowstone and Isle Royale groups had almost no slightly worn individuals among 4–6 year-old wolves , and none at all after age seven . Instead , the proportion of individuals assigned to the heavy wear class was greater in both of these samples in the two later age classes ( 4–6 years , 7+ years ) . Tooth fracture rates also differed by age at death among the three samples with or without canine teeth included ( Figure 4 ) . The Isle Royale wolves have significantly more fractured teeth at any age relative to both groups of Yellowstone wolves . Between the two Yellowstone samples , the 2007–2016 sample consistently exhibited higher fracture frequencies . The difference approaches significance in the 4–6 year-old age class ( p=0 . 08 ) and is highly significant ( p<0 . 01 ) for the 7+ year-old age class . Whereas in the later Yellowstone sample , 10% ( 66/671 ) of the teeth in individuals older than 6 years were broken , the same figure was only 3% ( 16/562 ) for the earlier Yellowstone wolves . Between 1996 and 2016 , necropsies were performed on 2372 adult elk largely from Yellowstone’s northern range representing a total of 99 , 624 possible skeletal elements ( excluding the pelvis and cranium ) as categorized by the observers ( Supplementary file 1 , Tables S3 , S4 ) . Of these elements , approximately 17% were missing and presumed removed by wolves . Forelimb elements ( scapula , humerus , radius + ulna , metacarpus ) appear often to have been removed together and were more likely to be taken than hindlimb elements ( Supplementary file 1 , Table S3 ) . This is not surprising given that the hindlimb has a bony articulation with the inominate ( pelvis ) bone , whereas the forelimb has a muscular attachment to the ribcage that is more easily severed . Like the hindlimb , vertebrae and dentary bones are more difficult to remove given their ligamentous and bony articulations and they were more likely to remain at kill sites on average than forelimb elements . The proportion of the skeleton that was removed varied over time , ranging from a low of 7% in 1997 to a high of 35% in 2016 ( Figure 5 , Supplementary file 1 , Table S4 ) . Changes over time in the overall proportion of the skeleton removed were mirrored by those of the humerus alone , highlighting the apparent preference for the forelimb mentioned above . There is considerable scatter in the plot of skeletal utilization over time ( Figure 5 ) with more variance in the first decade than the second ( i . e . , heteroscedasticity ) . Consequently , the data are not appropriate for linear regression and a three-year simple moving average was estimated . There is the suggestion of an upward trend in the last decade , indicating that Yellowstone wolves processed carcasses more fully in the second decade after elk numbers declined . When the necropsy data are split into two chronological samples , 1997–2006 and 2007–2016 , the proportion of skeletal elements removed is 15% for the early sample and significantly less than that for the later sample , 20% ( p=0 . 001 ) . In the sampled wolves , a rise in tooth fracture frequency is associated with a decline in the availability of their primary prey , as estimated by the ratio of prey to predator abundance . The two samples from areas with high prey to predator ratios , Scandinavia and early Yellowstone , are similar in exhibiting low tooth fracture frequencies , both on a per tooth and per skull basis . In contrast , the two samples from areas with much lower prey to predator ratios , Isle Royale and later Yellowstone , have greater rates of tooth wear and fracture , even if canine teeth are excluded and differences in age distribution are considered . Although the per tooth fracture frequency in later Yellowstone ( 4 . 6% ) is not as high as that of Isle Royale ( 8 . 6% ) , it is significantly larger than the fracture frequency for early Yellowstone wolves ( 1 . 8% ) , suggesting a change in feeding behavior . The change in feeding behavior that is most likely to have caused the rise in tooth fracture is increased bone consumption . This is supported by previous interspecific comparisons of carnivoran tooth fracture and diet ( Van Valkenburgh , 1988; Van Valkenburgh , 2009; Mann et al . , 2017 ) , as well as by the distribution of fracture across the tooth row in the wolves sampled for this paper . As has been shown in previous studies of tooth fracture , the canine teeth are the most likely to be broken , probably due to their elongate shape and use in combat and killing prey . Nevertheless , teeth involved in gnawing ( incisors ) and food processing ( premolars and molars ) are broken more often in Isle Royale and later Yellowstone samples than in the early Yellowstone , Scandinavian , or other North American samples . Notably , the two Yellowstone samples differ much less in canine tooth fracture frequency than carnassial tooth fracture frequency . Whereas canine tooth breakage rose by 50% between the early and later cohorts , carnassial tooth breakage went up 500% . Additionally , Cubaynes et al . ( 2014 ) found density-dependent rates of aggression in Yellowstone’s northern range wolves , with significantly lower rates of aggression corresponding with later Yellowstone samples showing greater canine tooth breakage . Furthermore , the proportion of winter biomass acquired by later Yellowstone wolves from bison ( Bison bison ) increased significantly ( Metz et al . , 2016 ) , largely from scavenging winter-killed carcasses . Consequently , utilization of bison carcasses with larger , thicker bones may explain some of the increased frequency in tooth fracture . These patterns reinforce the idea that it is a shift in feeding behavior rather than one in levels of aggression or a change in predatory methods that is responsible for the bump up in tooth breakage . The distribution of individuals by tooth wear stage also differed between the high and low prey to predator samples in ways that are consistent with more bone consumption in the latter . There were greater proportions of individuals with slight wear in the Scandinavian , early Yellowstone , and other North American samples relative to both Isle Royale and later Yellowstone , in which individuals with moderate to heavy wear predominated . These differences might reflect differences among the samples in age distribution rather than feeding behavior , given that older individuals have more heavily worn teeth , but this was not the case . Age data were available for both the Yellowstone and Isle Royale samples , and when individuals of similar age are compared , it is clear that later Yellowstone and Isle Royale wolves wore their teeth more rapidly as they aged and were more likely to fracture their teeth at any age than early Yellowstone wolves ( Figures 3 and 4 ) . Even though both Isle Royale and later Yellowstone wolves exhibit elevated rates of tooth wear and fracture relative to the other wolf samples , the Isle Royale wolves have much higher rates of fracture on a per tooth basis than the later Yellowstone wolves . On a per tooth basis , the frequency of tooth fracture for Isle Royale wolves is similar to that of sampled late Pleistocene gray wolves and dire wolves . Isle Royale wolf per-tooth fracture rates are nearly double that observed for later Yellowstone wolves and about three to four times the rates observed in all the other extant samples . This may reflect more extreme food limitation in Isle Royale , but there is another factor that might be relevant . Isle Royale wolves are the most inbred of the three focal samples , the other two being Yellowstone and Scandinavia . Previous work on congenital malformities in both the highly inbred Isle Royale wolves and moderately inbred Scandinavian population found a significantly greater incidence of skeletal malformations in the former ( Räikkönen et al . , 2009; Räikkönen et al . , 2013 ) . Perhaps inbreeding depression has affected tooth strength in Isle Royale wolves , although there are no other studies of inbred mammals that document impacts on tooth strength or breakage . Alternatively , perhaps the higher fracture frequencies in Isle Royale are due to a greater size difference between predator and prey . Both Isle Royale and Scandinavian wolves kill moose , but the Scandinavian wolves are significantly larger ( Sand et al . , 2012 ) . Given their smaller size , Isle Royale wolves may have more difficulty killing adult moose and therefore need to consume kills more completely . Moreover , although wolves prefer to kill calves , the number of available calves each year is much less in Isle Royale than Scandinavia and consequently Isle Royale wolves kill a greater proportion of adults ( Sand et al . , 2012 ) . Finally , it is possible that the probability of breaking a tooth increases with the number of broken teeth . The presence of broken teeth might lead to malocclusion or other abnormalities that then increase the probability of heavier wear and/or tooth breakage , producing more rapid increases in tooth fracture rates and wear as individuals age . Isle Royale wolves may be further along this trajectory than the later Yellowstone wolves . The data presented here support the idea that rates of cheek tooth fracture in large carnivores can be used as an index of prey availability in modern and ancient ecosystems . Consequently , the elevated rates of tooth fracture in large Pleistocene carnivores might be interpreted as evidence of relatively high predator to prey ratios and greater top-down forcing on large herbivore populations than are typically observed today ( Ripple and Van Valkenburgh , 2010; Van Valkenburgh et al . , 2016 ) . However , ‘prey availability’ , defined as the relative ease of acquiring and consuming a kill , can be more complicated than the metric used here , the numbers of prey relative to predators . For example , in the case of wolves , prey are easier to catch in severe winters with deep snow than milder winters ( Mech et al . , 2001 ) , as well as during late winter versus early winter when prey are in better nutritional condition ( Metz et al . , 2012 ) . In addition , prey availability can be affected by interference as well as exploitative competition . Intraguild competition is relatively intense within guilds of large mammalian carnivores , and manifests itself in multiple ways , including carcass theft ( kleptoparasitism ) and intraguild predation , both of which tend to increase when food is limited ( Palomares and Caro , 1999; Donadio and Buskirk , 2006 ) . Studies of extant carnivores have revealed the significant impact that kleptoparasitism can have on subordinate species that regularly lose their kills to dominant apex predators . As a consequence of calories lost due to kleptoparasitism , European lynx , cheetahs , pumas and wild dogs all have been documented to kill more frequently ( Creel and Creel , 1996; Carbone et al . , 1999; Hayward et al . , 2006; Krofel et al . , 2012; Broekhuis et al . , 2013 ) . Moreover , in the case of wild dogs and pumas , kleptoparasitism was associated with consuming carcasses more completely despite the presence of abundant prey ( Carbone et al . , 1999; Elbroch et al . , 2015 ) . Carbone et al . ( 1999 ) observed that wild dogs spent more time feeding on carcasses and ‘consuming the poorest sections’ when spotted hyena ( Crocuta crocuta ) numbers were higher . It is difficult to estimate the added energetic costs incurred by both losing a kill and having to hunt more frequently , but our gray wolf data suggest that the associated need for , or at least advantage of , consuming carcasses more fully results in more rapid wear and risk of tooth fracture , the latter of which can result in debilitating or fatal infections . Thus , rates of tooth wear and fracture in large carnivores can be used as indicators of food limitation and energetic stress load in extant and extinct populations . For example , Pleistocene guilds of large carnivores were much more species-rich than their present-day equivalents , and often included at least three very large , in some cases social , carnivorous species , such as sabertooth and non-sabertooth cats , large canids , hyaenids , and ursids . It seems likely that kleptoparasitism was a relatively common occurrence , and this would have contributed to the need to utilize carcasses more fully and perhaps scavenge the kills of others , including the consumption of less nutritious and potentially damaging portions , such as bones . Consistent with this , the very high tooth fracture frequencies observed in a number of large Pleistocene carnivores in the Old and New Worlds ( Van Valkenburgh and Hertel , 1993; Van Valkenburgh , 2009; Flower and Schreve , 2014 ) suggest that they experienced ‘tough times’ or food limitation at more frequent intervals than historic or current populations of similar or the same species . Elevated levels of food stress among apex predators could have had ecosystem wide impacts . Intense food competition and frequent kleptoparasitism would have favored higher kill rates among subordinate species , and probably an overall greater supply of large carcasses to ecosystems . In addition , it may have been associated with intensified negative effects on herbivore population growth ( top-down forcing ) as well , but this is difficult to determine without data on fossil herbivore population dynamics . Finally , our study demonstrates the great value of preserving skeletons , or at least skulls of well-studied mammal populations whenever possible . In our case , dental wear data allowed us to gain insights into the relative difficulty of killing and consuming prey , a critical aspect of large carnivore success that is challenging to quantify in wild populations . Initial studies of tooth fracture in carnivores focused largely on biomechanical questions relevant to tooth strength and consequently , the discovery that tooth fracture can provide insights into levels of food limitation and associated energetic stress was unexpected . There are certain to be more such discoveries based on natural history collections , especially when they are associated with extensive metadata such as those used here ( Schmitt et al . , 2018 ) . Sadly , natural history collections worldwide are under threat due to insufficient funding and a lack of appreciation of their value . Studies such as ours might help reverse this disturbing trend . Isle Royale National Park ( IRNP ) is an approximately 544 km2 island in Lake Superior within the boundaries of the state of Michigan , USA . The island is inhabited by wolves and moose , both of which have been under intensive study since 1959 . There are no other large carnivorans or ungulates on the island and it is essentially isolated from emigration or immigration . Other carnivorans on the island include red fox , pine marten , weasel , mink and river otter , none of which are likely competitors , although foxes are known to scavenge wolf kills . The numbers of wolves and moose are estimated annually from fixed wing aircraft or via ground surveys ( Peterson et al . , 2014 ) . The wolf population has fluctuated , from a maximum of 50 in 1981 to its current minimum of two individuals at the time of this writing ( Peterson et al . , 2018 ) . Over the last 50 years , the ratio of moose to wolf has varied over 8-fold , ranging from 20:1 to 160:1 with an average of 55:1 ( Peterson et al . , 2018 ) . The 64 adult wolf skulls sampled for this paper span 1963–2009 and are housed at Michigan Technological University , Houghton , MI ( Table 1 ) . Whenever possible , an estimated age at death was recorded for each wolf . Between 1995 and 2008 , necropsies were performed on 239 moose killed by wolves , and an estimate of carcass utilization was made that focused largely on the degree of skeletal disarticulation and proportion of bone consumed ( Vucetich et al . , 2012 ) . Over the course of the study , increases in per capita kill rate ( #kills/wolf/unit time ) were significantly associated with decreases in the proportion of carcass utilized . The Scandinavian wolf sample consists of 94 skulls collected between 1998 and 2010 and housed in the Swedish Royal Museum of Natural History , Stockholm . Gray wolves were regarded as functionally extinct in Scandinavia in 1966 , but began to recolonize south-central Scandinavia in the 1980’s , and by 2010 , numbered between 250–290 individuals in 52 packs ( Wabakken et al . , 2001 ) . As in IRNP , moose are the primary prey of wolves , but many more calves are taken in southern Scandinavia ( Norway , Sweden ) than IRNP because they are more abundant due to selective human hunting for bulls and forest management ( Sand et al . , 2012 ) . Sand et al . ( 2005 ) estimated the proportion of edible biomass ( not including bones , rumen , hide and guts ) consumed during winter for moose kills made within two wolf territories in southern Scandinavia between 2000–2002 . The Yellowstone National Park sample consists of 160 skulls of adult wolves that are housed in the Yellowstone Heritage and Research Center near Gardiner , Montana . The wolves died between 1996 and 2016 , and almost all were from packs inhabiting the northern range of the park . Whenever possible , the age in years and months at death was recorded . Since their initial reintroduction , wolves in this section of the park have been monitored intensively through the application of VHF telemetry and GPS radio-collars , year-round behavioral observations , and genetic analyses as part of the Yellowstone Wolf Project ( https://www . nps . gov/yell/learn/nature/wolfreports . htm ) . Skulls from individuals that were members of the original 31 Canadian founders were excluded from our study as they may have experienced tooth damage while in fenced enclosures that were essential at the start of the project . At the time of wolf reintroduction in 1995 , elk numbers in the northern range exceeded 18 , 000 and wolves had been absent from the park for over 60 years ( Peterson et al . , 2014 ) . Within two decades after the return of wolves , the number of elk declined to about 5000 . Over the same interval , wolf numbers on the northern range rose to exceed 100 but have since declined to between 40 and 50 . As a result , the elk:wolf ratio dropped from greater than 600:1 in 1996 to around 100:1 by 2004 and has remained near the lower level for the last 14 years ( Figure 1 ) . To better understand patterns of wolf predation , Yellowstone Wolf Project personnel completed field necropsies on prey killed by wolves whenever possible . Data collected include the number of skeletal elements present as well as additional data on characteristics of the prey ( e . g . species , sex , estimated age , nutritional condition ) . Data from field necropsies of over 2300 adult elk killed by wolves between 1997 and 2016 were analyzed for this paper to examine whether carcass utilization has changed as elk numbers declined . To explore whether tooth wear and fracture rate shifted over this same interval in Yellowstone wolves , the data were divided into two groups , wolves that died between 1996 and 2006 ( n = 74 ) , and those that died between 2007 and 2016 ( n = 82 ) . As can be seen from Figure 1 , the elk:wolf ratio declined rapidly in the first 10 years and then fluctuated around 100:1 for the second decade . The historic sample of 223 wolf skulls without associated prey data includes individuals collected between 1874 and 1952 from three regions; 1 ) Alaska , 2 ) Texas and New Mexico , and 3 ) Idaho and adjacent Canada ( Table 1 ) . All these skulls are housed in the National Museum of Natural History , Washington , DC ( see Supplementary file 1 , Table S1 ) . As in previous studies ( Mann et al . , 2017; Van Valkenburgh , 1988; Van Valkenburgh , 2009 ) , individual skulls and associated mandibles were examined for dental wear and fracture . To avoid counting teeth that were broken post-mortem or just prior to death due to trap or other damage , teeth were recorded as broken only if there was clear evidence of a fracture ( e . g . , partially or fully broken cusp ) and a blunted surface due to subsequent wear . Missing teeth were not counted as broken , even when alveolar resorption suggested tooth loss due to injury . Consequently , the number of broken teeth are likely underestimates . In addition to recording tooth condition , a qualitative estimate of overall wear stage for the individual was made as follows: 1 ) ‘slight’ , little or no wear on shear facets or blunting of cusps; 2 ) ‘moderate’ , shear facets apparent on carnassial teeth and cusps blunted on most teeth; or 3 ) ‘heavy’ , carnassial teeth with strong shear facets and/or blunted cusps , premolars and molars with well-rounded cusps ( Figure 2—figure supplement 1 ) . Tooth fracture incidence was assessed on both a per-individual ( percentage of individuals with one or more broken teeth ) and a per-tooth ( percentage of all teeth that are broken ) basis . The distribution of tooth fracture across the tooth row was quantified by tooth type ( incisors , canines , premolars , carnassials , post-carnassial molars ) . Because we were especially interested in detecting the impacts of increased bone consumption , tooth fracture rates were quantified and compared with canine teeth excluded as well as included . Although canine teeth may fracture when feeding on bones , they also may break during intra- and inter-specific combat , as well as when killing prey . Incisors and cheek teeth are used for gnawing and cracking bones , respectively ( Van Valkenburgh , 1996 ) , and thus are expected to wear more quickly when bone consumption increases . Tooth fracture rates were compared among the study samples using chi-square statistics and the software package SPSS version 25 . For the analysis of the effects of age on tooth fracture and wear stage , linear regressions of the number of teeth broken against age for all individuals were performed to test for differences among populations , but results did not achieve significance due to a lack of normality and substantial scatter in the data ( Figure 4—figure supplement 1 ) . Consequently , comparisons were made among individuals placed into one of three age groups , 1–3 years , 4–6 years , and seven or more years . In the case of the Scandinavian and Isle Royale samples , carcass utilization data were taken from the literature . In Scandinavia , Sand et al . ( 2005 ) observed that approximately 70% of the edible biomass of each adult moose kill was consumed by the Scandinavian wolves in their sample , whereas Vucetich et al . ( 2012 ) estimated that on average , Isle Royale wolves consumed 90% of each moose kill . This difference is greater than it appears given that the Scandinavian estimate of the proportion of edible biomass consumed did not include bones as potential edible biomass . Had this been done , the proportion consumed would be less than 70% . The relatively low utilization rate in Scandinavia was ascribed to three factors , moose density was high , the moose were naïve to wolf predation , and human disturbance ( Sand et al . , 2005 ) . For Yellowstone , a skeletal utilization index for over 2300 adult elk kills was constructed based on the percentage of the skeleton that remained after wolves abandoned a kill . The data are derived from the field necropsies mentioned above . At each carcass , observers recorded the number of mandibles , scapulae , vertebrae , humeri , radii plus ulni , femora , tibiae , and presence or absence of metacarpi , metatarsi , skull , and pelvis . The skeletal utilization index used here was calculated as one minus the ratio of the number of mandibles and limb bones found at kill sites to the total number expected if skeletons were complete . Thus if 80% of the skeleton remained , the utilization index was 20% . Data on the status of the skull and pelvis were excluded for this analysis as they were almost always present . For the purposes of this study , it is assumed that missing skeletal elements were most likely removed by wolves , but it is possible that other scavengers , such as bears and coyotes , also may have taken bones . Consequently , the data should be viewed with this in mind . If the skeletal utilization data indicate an increase in bone removal that is not associated with increased tooth wear among wolves , then it might be the case that other carnivores were responsible for the missing bones . However , three-quarters ( 76% ) of the elk necropsies were done during winter when bears are hibernating so their role in carcass utilization is not a factor for the bulk of the sample .	Gray wolves roam many European and American landscapes , where they prey on large animals such as elk and moose . A healthy dentition is essential for these predators to kill , eat and defend themselves . As a result , they tend to avoid biting down on tough body parts , such as bones , so that their teeth do not break . If food becomes scarce however , the wolves may resort to consuming these hard elements , eating more of the carcasses and leading to more damaged teeth . It could therefore be possible to assess the food levels available to existing ( or even extinct ) wolf populations based on how many broken teeth the animals have . However , older individuals are also more likely to have more damaged teeth , so age would need to be taken into consideration . Van Valkenburgh et al . decided to evaluate whether it was indeed possible to deduce how much food was available to groups of wolves based on teeth damage . Tooth wear and fracture were quantified in three current populations of gray wolves whose skulls had been collected and preserved in natural history collections . For each group , there were data available about the variations of number of moose per wolf over time , and how much of the carcasses the wolves were consuming . The analyses showed that indeed , when prey became less abundant , the wolves ate more of the remains – including the bones – and therefore broke more teeth . These conclusions can be applied to other large predators and even to extinct species such as dire wolves or sabertooth cats . Tapping into the potential of museum specimens could help to retrace environmental conditions and the history of animals now long gone .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2019	Tooth fracture frequency in gray wolves reflects prey availability
The CRISPR-Cas9 bacterial surveillance system has become a versatile tool for genome editing and gene regulation in eukaryotic cells , yet how CRISPR-Cas9 contends with the barriers presented by eukaryotic chromatin is poorly understood . Here we investigate how the smallest unit of chromatin , a nucleosome , constrains the activity of the CRISPR-Cas9 system . We find that nucleosomes assembled on native DNA sequences are permissive to Cas9 action . However , the accessibility of nucleosomal DNA to Cas9 is variable over several orders of magnitude depending on dynamic properties of the DNA sequence and the distance of the PAM site from the nucleosome dyad . We further find that chromatin remodeling enzymes stimulate Cas9 activity on nucleosomal templates . Our findings imply that the spontaneous breathing of nucleosomal DNA together with the action of chromatin remodelers allow Cas9 to effectively act on chromatin in vivo . The recent development of CRISPR ( clustered regularly interspaced short palindromic repeats ) systems , particularly the type II CRISPR-Cas9 mechanism from Streptomyces pyogenes , as an artificial tool for genome engineering , gene regulation , and live imaging is a remarkable achievement with profound impact in a wide variety of research fields and applications ( Makarova et al . , 2015; Doudna and Charpentier , 2014; Cong et al . , 2013; Jinek et al . , 2012; 2013; Mali et al . , 2013 ) . Despite its successful adoption across numerous eukaryotic organisms , relatively few details are known of the mechanism by which bacterial CRISPR-Cas9 systems operate in eukaryotic cells ( Doudna and Charpentier , 2014; Ghorbal et al . , 2014; Vyas et al . , 2015 ) . CRISPR-Cas9 originated in bacteria , where genomic DNA generally consists of supercoiled circular molecules associated with nucleoid-associated proteins ( Travers and Muskhelishvili , 2005 ) . In contrast , eukaryotic chromosomes are linear , packaged with histone octamers into nucleosomes , and further organized into higher-order structures ( Luger et al . , 1997; Olins and Olins , 1974; Woodcock et al . , 1976; Dixon et al . , 2012 ) . The packaging of DNA into nucleosomes generally inhibits the binding of sequence specific DNA binding factors . In the simplest model , nucleosomes would analogously inhibit Cas9 action . Further , in eukaryotes ATP-dependent chromatin remodelers reposition , remove , or restructure nucleosomes to regulate the access of DNA binding factors ( Clapier and Cairns , 2009; Narlikar et al . , 2013 ) . It can therefore be imagined that the action of remodelers also regulates the action of Cas9 on nucleosomes . To quantitatively test the above models we performed biochemical studies to measure Cas9 activity on nucleosomes assembled with native and artificial nucleosome positioning sequences . We find that the combination of nucleosome breathing , by which DNA transiently disengages from the histone octamer , and the action of chromatin remodeling enzymes allow Cas9 to act on nucleosomal DNA with rates comparable to naked DNA . The results provide a biochemical explanation for the efficacy of Cas9 in eukaryotic cells . To determine if a nucleosome inhibits the ability of Cas9 to scan , recognize , and cleave sgRNA-directed DNA targets , we performed in vitro Cas9 cleavage assays using mononucleosomes ( single nucleosomes on short dsDNA molecules ) reconstituted using the Widom 601 positioning sequence with 80 base pairs of flanking DNA on both sides ( referred to as 601 80/80 particles , Figure 1A ) ( Lowary and Widom , 1998 ) . The 601 sequence is an artificially derived sequence with high affinity for the histone octamer and has proved a valuable tool for assembling well positioning nucleosomes for biochemical studies . Using sgRNAs targeting the nucleosomal dyad , entry/exit sites , and flanking DNA , we measured the rates of Cas9 cleavage with naked 601 DNA and the 601 80/80 particles . Targeting the DNA flanking the nucleosome showed cleavage rates comparable to those of naked DNA . Cleavage rates at entry/exit sites of the nucleosome were much lower compared to naked DNA ( ~23–28x decrease cleavage rate vs . DNA alone ) ( Figure 1B , C ) . Targeting near the nucleosomal dyad resulted in further inhibition of cutting by Cas9 ( ~1000x decrease vs . DNA alone ) ( Figure 1C , D ) . Previous work has shown that nucleosomal DNA transiently disengages from the histone octamer , a process termed as nucleosomal DNA unpeeling or breathing . The equilibrium for DNA unpeeling gets progressively more unfavorable the closer the DNA site gets to the dyad ( Polach and Widom , 1995; Li and Widom , 2004; Luger et al . , 2012 ) . The nucleosome-mediated inhibition of Cas9 activity is more pronounced near the dyad suggesting that Cas9 cleavage occurs on DNA that is transiently disengaged from the histone octamer . 10 . 7554/eLife . 13450 . 003Figure 1 . Cas9 DNA nuclease activity is hindered by nucleosomes . ( A ) Schematic of sgRNAs designed against the assembled 601 80/80 nucleosome substrates targeting the flanking regions , entry/exit sites , and near the nucleosomal dyad . ( B ) Cleavage assay comparing Cas9 cleavage on 80/80 DNA and 80/80 nucleosomes when loaded with sgRNA #3 . ( C ) Kinetics of cleavage with sgRNA #3 . ( D ) Comparison of the relative rates of cleavage on nucleosomes to DNA at various positions along the 80/80 nucleosome construct . The position reported is the site of cleavage by Cas9 . Represented values are mean ± SEM from three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00310 . 7554/eLife . 13450 . 004Figure 1—source data 1 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNAs #2 and #6 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00410 . 7554/eLife . 13450 . 005Figure 1—source data 2 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNAs #2 and #6 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00510 . 7554/eLife . 13450 . 006Figure 1—source data 3 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNAs #2 and #6 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00610 . 7554/eLife . 13450 . 007Figure 1—source data 4 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #5 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00710 . 7554/eLife . 13450 . 008Figure 1—source data 5 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #5 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00810 . 7554/eLife . 13450 . 009Figure 1—source data 6 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 00910 . 7554/eLife . 13450 . 010Figure 1—source data 7 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01010 . 7554/eLife . 13450 . 011Figure 1—source data 8 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01110 . 7554/eLife . 13450 . 012Figure 1—source data 9 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01210 . 7554/eLife . 13450 . 013Figure 1—source data 10 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #4 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01310 . 7554/eLife . 13450 . 014Figure 1—source data 11 . Replicate gels of Cas9 cleavage of 80/80 601 DNA and nucleosomes with sgRNA #4 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01410 . 7554/eLife . 13450 . 015Figure 1—source data 12 . Quantification of Figure 1 Cas9 cleavage gels . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01510 . 7554/eLife . 13450 . 016Figure 1—figure supplement 1 . Nucleosome positioning blocks Cas9 from binding PAM sites on DNA . ( A ) Schematic illustrating the stepwise mechanism of Cas9 binding to DNA targets and subsequent nucleolytic cleavage . ( B ) Gel shift assay comparing dCas9 binding to 0/0 DNA and nucleosomes while loaded with sgRNA #3 targeting the nucleosome dyad . Band shift pattern appears as discrete lower band ( dCas9 bound to sgRNA-specified target , arrowhead ) and higher , super shift bands ( additional dCas9 PAM-only binding events ) . ( C ) Quantification of ( B ) . Represented values are mean ± SEM from three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01610 . 7554/eLife . 13450 . 017Figure 1—figure supplement 1—source data 1 . -3Replicate gels of dCas9 binding to 0/0 601 DNA and nucleosomes with sgRNA #3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01710 . 7554/eLife . 13450 . 018‌‌Figure 1—figure supplement 1—source data 2 . -3Replicate gels of dCas9 binding to 0/0 601 DNA and nucleosomes with sgRNA #3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01810 . 7554/eLife . 13450 . 019Figure 1—figure supplement 1—source data 3 . -3Replicate gels of dCas9 binding to 0/0 601 DNA and nucleosomes with sgRNA #3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 01910 . 7554/eLife . 13450 . 020Figure 1—figure supplement 1—source data 4 . Quantification of Figure 1—figure supplement 1 gel shifts . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 020 Nucleosomes block the ability of Cas9 to cleave DNA , but it is unclear at which step of Cas9 activity this occurs . Cas9 recognizes DNA target sites in a two-step process that begins with binding to the DNA protospacer adjacent motif ( PAM , in this case 'NGG' ) through its C-terminal PAM-interacting region , followed by sequential melting of the DNA double strand and annealing of the sgRNA guide segment to the unwound target DNA strand ( Figure 1—figure supplement 1A ) ( Sternberg et al . , 2014; Jiang et al . , 2015 ) . Complete annealing of the 20-nt guide RNA to the target DNA is required to drive a progressive conformational transformation that authorizes Cas9 to simultaneously cleave both DNA strands ( Sternberg et al . , 2015; Josephs et al . , 2016 ) . Given this order of events , it is conceivable that nucleosomes can interfere with any of the steps preceding and including DNA cleavage . To identify the point at which nucleosomes disrupt Cas9 function , we assessed binding of nuclease-dead Cas9 ( dCas9 ) to mononucleosomal particles by an electrophoretic mobility shift assay . We performed dCas9 binding assays using 601 0/0 nucleosomal particles which are devoid of naked DNA overhangs . Binding of dCas9 pre-loaded with core targeting sgRNA with 601 0/0 nucleosomes is undetectable whereas binding to naked DNA control is still robust ( Figure 1—figure supplement 1B ) . The presence of super shifts in the gel migration pattern suggests that multiple dCas9 molcules are engaging the same DNA substrate molecule . We investigated this further and determined that , in our binding assay , the highly transient dCas9 binding to PAMs within target DNA is observable as super shifts , likely due to a combination of a high number of PAMs on the target DNA ( 23 NGG PAMs present in 601 0/0 sequence ) and common caging effects of gel binding assays . The absence of a super shift binding pattern with 0/0 nucleosomes ( Figure 1—figure supplement 1B , right ) suggests that dCas9 cannot stably interact with PAMs located on nucleosomes , in a manner consistent with a recently published study ( Hinz et al . , 2015 ) . The artificial Widom 601 is an atypically strong nucleosome positioning sequence that shows ~100-fold less breathing dynamics compared to physiological nucleosome positioning sequences , such as the 5S rRNA gene ( Anderson et al . , 2002; Partensky and Narlikar , 2009 ) . To determine how Cas9 contends with nucleosomes assembled on this physiological positioning sequence , we performed cleavage experiments with nucleosomes assembled from 5S rRNA gene sequences from Xenopus borealis ( Figure 2A ) . Cas9-mediated cleavage at sites near the entry/exit of the nucleosome is substantially enhanced ( 700–fold ) with 5S nucleosomes compared to 601 particles ( Figure 2B–D ) . In the context of 601 , cutting at this site is 1000-fold slower than in naked DNA . In contrast , with 5S nucleosomes , cutting at the comparable site is only 1 . 5-fold slower than in naked DNA . However , Cas9 cleavage near the dyad is inhibited to a similar extent on both 5S and 601 nucleosomes , showing that the 5S-specific enhancement of Cas9 activity does not extend all the way to the nucleosomal dyad . These results support our interpretation that nucleosomal DNA breathing substantially enhances Cas9 binding to nucleosomes and demonstrate that nucleosomal DNA sequence , through its influence on nucleosome stability , can regulate Cas9 activity over a large dynamic range . 10 . 7554/eLife . 13450 . 021Figure 2 . Higher nucleosomal breathing dynamics enhance Cas9 cleavage . ( A ) Schematic illustrating nucleosome breathing and how it can enable Cas9 binding to a target in the nucleosome . ( B ) Cleavage assay comparing Cas9 cleavage of 601 and 5S 0/0 nucleosomes when loaded with sgRNAs targeting comparable positions at core and entry sites . ( C ) Quantification of ( B ) . ( D ) Cas9 cleavage rates on 601 and 5S nucleosomes when targeted to core and entry sites . Values were normalized against naked DNA control rates . Represented values are mean ± SEM from three replicates . Additional gel panels shown in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02110 . 7554/eLife . 13450 . 022Figure 2—source data 1 . Replicate gels of cleavage of 0/0 5S DNA and nucleosomes with sgRNA core . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02210 . 7554/eLife . 13450 . 023Figure 2—source data 2 . Replicate gels of cleavage of 0/0 5S DNA and nucleosomes with sgRNA core . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02310 . 7554/eLife . 13450 . 024Figure 2—source data 3 . Replicate gels of cleavage of 0/0 5S DNA and nucleosomes with sgRNA entry . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02410 . 7554/eLife . 13450 . 025Figure 2—source data 4 . Replicate gels of cleavage of 0/0 5S DNA and nucleosomes with sgRNA entry . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02510 . 7554/eLife . 13450 . 026Figure 2—source data 5 . Replicate gels of cleavage of 0/0 601 DNA and nucleosomes with sgRNA entry . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02610 . 7554/eLife . 13450 . 027Figure 2—source data 6 . Replicate gels of cleavage of 0/0 601 DNA and nucleosomes with sgRNA entry . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02710 . 7554/eLife . 13450 . 028Figure 2—source data 7 . Quantification of Figure 2 Cas9 cleavage gels . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02810 . 7554/eLife . 13450 . 029Figure 2—source data 8 . Quantification of Figure 2 Cas9 cleavage gels . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 02910 . 7554/eLife . 13450 . 030Figure 2—figure supplement 1 . Cas9 cleavage assay with 601 and 5S 0/0 nucleosomes . Representative gel images of Cas9 cleavage experiments with 601 ( left ) and 5S ( right ) 0/0 particles using sgRNAs targeting entry ( top ) or core ( bottom ) sites , including DNA control experiments . Samples were resolved on 12% ( entry ) or 8% ( core ) polyacrylamide gels . Cleavage with sgRNAs targeting the entry site on both 5S and 601 substrates generates a short cleavage product ( 13 bp ) which partially denatures and runs as two bands ( single and double stranded ) on 12% polyacrylamide gels . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 030 We next investigated whether chromatin remodeling could enhance Cas9 activity towards chromatin substrates . Nucleosome positioning in vivo is strongly dependent on ATP-dependent chromatin remodelers , which are capable of loading , repositioning , and removing nucleosomes from the chromatin fiber . To measure how chromatin remodelers can influence Cas9 activity , we performed experiments where we pre-incubated 601 nucleosomes with remodeler enzymes prior to Cas9-mediated cleavage . For our experiments with the human ISWI-family remodeler SNF2h , we used asymmetric nucleosomes that possess flanking DNA only on the entry side ( 601 80/0 particles ) . When incubated with 601 80/0 particles , SNF2h promotes sliding of the nucleosome towards the center of the DNA molecule ( Figure 3A–B , Figure 3—figure supplement 1 ) ( Längst et al . , 1999; He et al . , 2006; Yang et al . , 2006 ) . We then performed in vitro cleavage experiments where 80/0 particles , pre-remodeled with SNF2h , were incubated with Cas9:sgRNA complex with its target site located within the nucleosome exit region . Remodeling 80/0 nucleosomes by SNF2h resulted in a strong enhancement of Cas9 cleavage to ~34% , showing that SNF2h slides the nucleosome enough to improve Cas9’s accessibility to the target site and that Cas9 is now able to bind and cleave at a higher rate ( Figure 3A–D ) . 10 . 7554/eLife . 13450 . 031Figure 3 . Chromatin remodeling improves Cas9 cleavage of nucleosomal substrates . ( A ) Schematic of Cas9 cleavage assay with remodeling . Cas9 is presented with 601 nucleosomes either untreated or previously remodeled with SNF2h or RSC remodelers . ( B ) Assay comparing cleavage on untreated and remodeled 80/0 nucleosomes when Cas9 is targeted to exit site ( depicted in green ) . These asymmetric nucleosomes are recentered by SNF2h , exposing the exit target site to Cas9 ( C ) Quantification of ( B ) . ( D ) Cleavage rates of 80/0 nucleosomes by Cas9 relative to naked DNA , in the presence or absence of SNF2h . SNF2h improves Cas9 cleavage to ~35% of the naked DNA cleavage rate . ( E ) Assay comparing Cas9-mediated cleavage at entry site of 80/80 symmetric 601 nucleosomes , either untreated or previously treated with RSC remodeler . RSC can destabilize nucleosome structure and reposition nucleosomes towards the DNA ends . ( F ) Quantification of ( E ) ( G ) Comparison of the rates of cleavage of nucleosomes normalized to DNA control with and without the action of RSC chromatin remodeler . Mean enhancement rates of Cas9 activity by chromatin remodeling are shown . ( H ) Cleavage rates of 80/80 nucleosomes by Cas9 relative to naked DNA , in the presence or absence of RSC . Cas9 cleavage is substantially enhanced by RSC , attaining ~63% of the naked DNA cleavage rate . Represented values are mean ± SEM from three replicates . Additional gel panels shown in Figure 3—figure supplement 1 . ( I ) Model of Cas9 activity in vivo in eukaryotes . Left , stable and strongly positioned nucleosomes impede Cas9 activity ( downward arrows ) . However , nucleosomes in vivo are generally more dynamic ( breathing ) , allowing Cas9 opportunities to target underlying DNA ( center ) . Cas9 accessibility to nucleosomal DNA can be further enhanced by the activity of chromatin remodelers that destabilize and/or reposition nucleosomes ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03110 . 7554/eLife . 13450 . 032Figure 3—source data 1 . Replicate gels of cleavage of 80/0 DNA and nucleosomes using sgRNA #4 with or without prior remodeling by Snf2h . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03210 . 7554/eLife . 13450 . 033Figure 3—source Data 2 . Replicate gels of cleavage of 80/0 DNA and nucleosomes using sgRNA #4 with or without prior remodeling by Snf2h . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03310 . 7554/eLife . 13450 . 034Figure 3—source data 3 . Replicate gels of cleavage of 80/0 DNA and nucleosomes using sgRNA #4 with or without prior remodeling by Snf2h . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03410 . 7554/eLife . 13450 . 035Figure 3—source data 4 . Quantification of Cas9 cleavage gels from Figure 3—source data 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03510 . 7554/eLife . 13450 . 036Figure 3—source data 5 . Replicate gels of cleavage of 80/80 DNA and nucleosomes using sgRNA 601_2 with or without prior remodeling by RSC . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03610 . 7554/eLife . 13450 . 037Figure 3—source data 6 . Replicate gels of cleavage of 80/80 DNA and nucleosomes using sgRNA 601_2 with or without prior remodeling by RSC . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03710 . 7554/eLife . 13450 . 038Figure 3—source data 7 . Replicate gels of cleavage of 80/80 DNA and nucleosomes using sgRNA 601_2 with or without prior remodeling by RSC . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03810 . 7554/eLife . 13450 . 039Figure 3—source data 8 . Quantification of Cas9 cleavage gels from Figure 3—source data 5–7 . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 03910 . 7554/eLife . 13450 . 040Figure 3—figure supplement 1 . Cas9 cleavage assays with SNF2h and RSC chromatin remodelers . ( A ) Representative gel images of Cas9 cleavage experiments with 601 80/0 asymmetric particles and SNF2h chromatin remodeler , including DNA control experiments . Cas9 was loaded with sgRNA targeting the exit site of the nucleosome . SNF2h re-centers asymmetric nucleosomes such as 80/0 ( Figure 3—figure supplement 3 ) , exposing the Cas9 target site . ( B ) Quantification of ( A ) . ( C ) Representative gel images of Cas9 cleavage experiments with 601 80/80 symmetric nucleosomes and RSC chromatin remodeler , including DNA control experiments . Cas9 was targeted to the entry site of the nucleosome . RSC destabilizes nucleosomal structure and repositions nucleosomes to the ends of the DNA molecule . ( D ) Quantification of ( C ) . Represented values are mean ± SEM from three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 04010 . 7554/eLife . 13450 . 041Figure 3—figure supplement 2 . Simultaneous chromatin remodeling and Cas9 cleavage of nucleosomal substrates . ( A ) Assay comparing Cas9 cleavage of 601 80/0 nucleosomes simultaneously with chromatin remodeling byr SNF2h . The 601 80/0 asymmetric nucleosomes are recentered by SNF2h , exposing the exit target site to Cas9 ( B ) Quantification of ( A ) . ( C ) Cleavage rates of 80/0 nucleosomes by Cas9 relative to naked DNA , in the presence or absence of SNF2h . SNF2h improves Cas9 cleavage to ~40% of the naked DNA cleavage rate , similarly to sequential remodeling and cleavage assays shown in Figure 3A–D . Results shown for single experiment performed . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 04110 . 7554/eLife . 13450 . 042Figure 3—figure supplement 2—source data 1 . Gel of cleavage of 80/0 DNA and nucleosomes using sgRNA #4 with or without simultaneous remodeling by Snf2h . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 04210 . 7554/eLife . 13450 . 043Figure 3—figure supplement 3 . SNF2h and RSC remodel nucleosomes prior to Cas9 cleavage . Gel‐shift nucleosome remodeling assay comparing positioned and SNF2h‐remodeled 80/0 nucleosomes ( left ) or RSC-remodeled 80/80 nucleosomes ( right ) . Migration pattern for all three forms ( centered , end-positioned nucleosomes and free DNA ) is illustrated ( center ) . Remodeling reactions were carried out 1 hr before being quenched and run on a 5% Acrylamide 0 . 5x TBE gel . SNF2h remodels 80/0 end‐positioned nucleosomes in a range of nucleosome positions biased towards the center of the DNA molecule , whereas RSC has the opposite bias of repositioned centered nucleosomes towards the ends of the DNA molecule . Images shown for single experiments performed , as previously described ( Yang et al . , 2006; Rowe and Narlikar , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 04310 . 7554/eLife . 13450 . 044Figure 3—figure supplement 3—source data 1 . Test remodeling gel of 80/0 nucleosomes with Snf2h . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 04410 . 7554/eLife . 13450 . 045Figure 3—figure supplement 3—source data 2 . Test remodeling gel of 80/80 nucleosomes with RSC . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 045 We also performed this experiment by simultaneously adding SNF2h and Cas9 and found a similar rate enhancement ( Figure 3—figure supplement 2 ) . While the ISWI remodeler SNF2h predominantly slides nucleosomes , remodelers from the SWI/SNF class have multiple outcomes , which include generation of DNA loops and eviction of the histone octamer in addition to nucleosome sliding ( Rowe and Narlikar , 2010; Narlikar et al . , 2001; Lorch et al . , 1998; Schnitzler et al . , 1998; Clapier and Cairns , 2009 ) . To determine if the types of remodeled products generated influence Cas9 activity , we performed similar experiments using 601 80/80 particles and the yeast chromatin remodeler RSC . We find that RSC activity also dramatically enhances cleavage on 601 80/80 nucleosomes when Cas9 is targeted to the entry site , negating most of the inhibitory influence of the nucleosome on Cas9 ( Figure 3E–F ) . These results demonstrate that two different classes of chromatin remodeling enzymes can significantly enhance Cas9 access to DNA targets normally obscured by nucleosomes . Here we demonstrate , using detailed biochemical studies with a variety of nucleosomal templates , that ( i ) the intrinsic stability of the histone-DNA interactions , ( ii ) the location of the target site within the nucleosomes ( nucleosome positioning ) , and ( iii ) the action of chromatin remodeling enzymes play critical roles in regulating the activity of S . pyogenes Cas9 . Below we discuss the implications of our results . Nucleosomes have been shown to inhibit the action of DNA binding factors . Recent work using nucleosomes assembled on the 601 sequence has led to the qualitatively similar conclusion that nucleosomes are refractory for Cas9 action ( Hinz et al . , 2015; Horlbeck et al . , 2016 ) . The comparison here between Cas9 action on 601 nucleosomes vs . nucleosomes assembled on the native 5S sequence suggests a more refined model for how nucleosomes regulate Cas9 action . We find that Cas9 sites near the entry/exit sites of 5S nucleosomes are cleaved ~700-fold better than the corresponding sites within 601 nucleosomes . Given that DNA breathing occurs at least 100-fold more in 5S nucleosomes than 601 nucleosomes we propose that Cas9 gains access to nucleosomal DNA when the DNA is transiently unpeeled from the histone octamer . This model also explains why sites closer to the entry/exit sites are cut more readily by Cas9 than sites near the dyad . This is because DNA unpeeling up to the dyad is substantially less favored ( 100-fold ) for both the 601 and 5S nucleosomes than DNA unpeeling near their respective entry/exit sites ( Anderson and Widom , 2000 ) . In vivo , as in vitro , the precise position of nucleosomes can greatly affect DNA factor binding . Chromatin remodeling enzymes can move nucleosomes away or towards the factor binding sites to respectively enhance or inhibit factor binding . We find that Cas9 activity can also benefit from chromatin remodeling to access nucleosomal DNA , as evidenced by the strong enhancements of Cas9 cleavage resulting from the action of the chromatin remodelers SNF2h and RSC . These two remodelers produce distinct nucleosomal arrangements yet still substantially alleviate nucleosome-mediated occlusion of Cas9 activity . In combination , our data lead to a comprehensive model that reconciles both biochemical evidence and in vivo observations to explain how Cas9 is able to access nucleosomal DNA in live cells ( Figure 3I ) . In vivo , the majority of nucleosomes are not located on strong positioning sequences , and therefore may be permissive to Cas9 binding , especially at target sites that are readily accessible by DNA unpeeling . Chromatin remodeling activities can further provide diverse mechanisms to potentiate Cas9 activity at sites located close to the nucleosomal dyad or within more strongly positioned nucleosomes , which would otherwise be refractory to Cas9 action . We hypothesize that the combination of spontaneous DNA unpeeling and remodeling contributes to the widespread success of CRISPR-Cas9 in eukaryotic cells . Interestingly , most applications of CRISPR-Cas9 in vivo have focused on genome engineering of protein-coding genes and other functional genomic elements associated with gene expression , which are typically associated with high rates of nucleosome remodeling ( Clapier and Cairns , 2009 ) . It is also conceivable that Cas9 can temporarily gain access to less accessible regions of the genome during specific points of cell cycle ( e . g . DNA replication ) , leading to sufficient DNA cleavage events to promote NHEJ-mediated mutagenesis or HDR-mediated DNA integration at appreciable rates . Recent studies on Cas9’s behavior by single molecule imaging have also demonstrated that Cas9 favors more accessible euchromatin regions but is not completely excluded from transcriptionally silent heterochromatin ( Knight et al . , 2015 ) . For other CRISPR applications that require stable binding of nuclease-deficient dCas9 to DNA , such as transcriptional regulation and live-cell imaging with fluorescent dCas9 , even modest nucleosome phasing could have a dramatic impact ( Gilbert et al . , 2013; Mali et al . , 2013; Chen et al . , 2013; Ma et al . , 2015 ) . For example , the +1 nucleosome in RNA pol II-transcribed genes is strongly positioned with phasing that dissipates gradually with each following nucleosome . Several high resolution studies conducted in parallel to our work have established that the +1 nucleosome and resulting nucleosome phasing can exert a strong influence on dCas9’s DNA-binding ability for transcriptional regulation , but the effect is less striking on genome editing with Cas9 ( Horlbeck et al . , 2016; Smith et al . , 2016 ) . Our observations suggest that sgRNA design strategies that avoid targeting near the dyad of strongly phased nucleosomes are likely to be more successful than current methods . Large scale nucleosome positioning or DNA accessibility maps are now readily available and can inform CRISPR sgRNA design in order to avoid targeting regions of low accessibility ( Jiang and Pugh , 2009; Thurman et al . , 2012; Wu et al . , 2014; Hsieh et al . , 2015 ) . Alternatively , whole cell chromatin de-condensation or de-repression using chromatin factor drugs such as HDAC or DNA methyltransferase inhibitors may be an alternative and attractive strategy for improving CRISPR-Cas9 activity towards densely compact regions of chromatin ( Haaf , 1995; Tóth et al . , 2004 ) . Wild-type Streptococcus pyogenes Cas9 and catalytically-inactive Cas9 ( dCas9 ) containing D10A and H840A mutations were individually cloned into a custom pET-based expression vector encoding an N-terminal 6xHis-tag followed by a small ubiquitin-related modifier ( SUMO ) fusion tag and a Ulp1 protease cleavage site . Recombinant Cas9 variants were then expressed in Escherichia coli strain BL21 ( DE3 ) ( Novagen ) and further purified to homogeneity as previously described ( Jiang et al . , 2015 ) . Single guide RNAs ( sgRNAs ) were prepared by in vitro run-off transcription using recombinant His-tagged T7 RNA polymerase and PCR product templates . Briefly , the DNA templates containing a T7 promoter , a 20-nt target DNA sequence ( listed in Table 1 ) and an optimal 78-nt sgRNA scaffold were PCR amplified using Phusion Polymerase ( NEB ) according to manufacturer’s protocol . The following PCR products were used directly as DNA templates for in vitro RNA synthesis in 1x transcription buffer ( 30 mM Tris-HCl pH 8 . 1 , 20 mM MgCl2 , 2 mM spermidine , 10 mM DTT , 0 . 1% Triton X-100 , 5 mM each NTP , and 100 μg mL-1 T7 RNA polymerase ) . After incubation at 37°C for 4–8 hr , the reactions were further treated with RNase-free DNase I ( Promega ) at 37°C for 30 min to remove the DNA templates . The synthesized sgRNAs were purified by Ambion MEGAclear kit and eluted into DEPC-treated H2O for downstream experiments . 10 . 7554/eLife . 13450 . 046Table 1 . Spacer sequences for sgRNAs used in biochemistry experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 046sgRNA #Guide sequencePAMTarget strandFigures where used601_1CGAGTTCATCCCTTATGTGATGGAntisenseFigure 1D601_2 ( entry ) AATTGAGCGGCCTCGGCACCGGGSenseFigure 1D , Figure 2B–D , Figure 2—figure supplement 1 , Figure 3E–H , Figure 3—figure supplement 1D–E601_3 ( core ) CCCCCGCGTTTTAACCGCCAAGGAntisenseFigure 1B–D , Figure 1—figure supplement 1B–C , Figure 2B–D , Figure 2—figure supplement 1601_4GTATATATCTGACACGTGCCTGGSenseFigure 1D601_5TCGCTGTTCAATACATGCACAGGSenseFigure 1D601_6GCGACCTTGCCGGTGCCAGTCGGAntisenseFigure 1D5S_1 ( entry ) TCTGATCTCTGCAGCCAAGCAGGSenseFigure 2B–E , Figure 2—figure supplement 15S_2 ( core ) TATGGCCGTAGGCGAGCACAAGGAntisenseFigure 2B–E , Figure 2—figure supplement 1 Gradient salt dialysis was used to assemble mono-nucleosomes on DNA templates containing the 147 bp long 601 or the 5S positioning sequence from X . borealis ( listed in Table 2 ) , and labeled with fluorescein on the 5’ upstream end . Histones and histone octamers were prepared as previously described ( Luger et al . , 1999 ) . 10 . 7554/eLife . 13450 . 047Table 2 . Sequences for DNA molecules used for biochemical assays ( Positioning sequence highlighted in grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13450 . 047NameSequence601 80/80CGGGATCCTAATGACCAAGGAAAGCATGATTCTTCACACCGAGTTCATCCCTTATGTGATGGACCCTATACGCGGCCGCCCTGGAGAATCCCGGTGCCGagGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTGCATGTATTGAACAGCGACCTTGCCGGTGCCAGTCGGATAGTGTTCCGAGCTCCCACTCTAGAGGATCCCCGGGTACCGA601 0/0CTGGAGAATCCCGGTGCCGagGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGT601 80/0CGGGATCCTAATGACCAAGGAAAGCATGATTCTTCACACCGAGTTCATCCCTTATGTGATGGACCCTATACGCGGCCGCCCTGGAGAATCCCGGTGCCGagGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGT5S 0/0GGCCCGACCCTGCTTGGCTGCAGAGATCAGACGATATCGGGCACTTTCAGGGTGGTATGGCCGTAGGCGAGCACAAGGCTGACTTTTCCTCCCCTTGTGCTGCCTTCTGGGGGGGGCCCAGCCGGATCCCCGGGCGAGCTCGAATT Cleavage assays were conducted as previously described with the following modifications ( Anders and Jinek , 2014 ) . Cas9:sgRNA complexes were reconstituted by incubating Cas9 and sgRNA for 10 min at 37°C . Reactions contained 5 nM fluorescein labeled DNA or nucleosomes and 100 nM Cas9:sgRNA . In combined cleavage and remodeling experiments , 25 nM SNF2h or 3 nM RSC was first incubated with 5 nM naked DNA or nucleosomes for 45 min at 37°C ( Narlikar et al . , 2001 ) . Cleavage assays were carried out in reaction buffer ( 20 mM Tris-HCl pH 7 . 5 , 70 mM KCl , 5 mM MgCl2 , 5% Glycerol , and 1 mM DTT ) at 25°C . For SNF2h and RSC remodeling experiments , 0 . 2 mM ATP was also added . For RSC remodeling experiments , 1 mM MgCl2 was used . Time points were quenched using stop buffer ( 20 mM Tris-HCl pH 7 . 5 , 70 mM EDTA , 2% SDS , 20% glycerol , and 0 . 2 mg/mL xylene cyanol and bromophenol blue ) . Proteins were digested with 3 mg/mL of Proteinase K and incubated at 50°C for 20 min . Samples were resolved on 1x TBE , 10% Polyacrylamide gels for 4 hr at 140 V before visualizing using a Typhoon scanner ( GE Healthcare ) and quantifying with Image J ( Schneider et al . , 2012 ) . For band quantification , background intensity was first subtracted after averaging the intensity of three areas . For cleavage gels , fraction uncleaved was determined by measuring the intensity of the uncleaved band compared to the total intensity for the lane . Similarly , fraction unbound was determined by measuring the intensity of the unbound band compared to the total intensity for the lane . All experiments were performed in triplicate . Experiment variability is presented as the standard error of the mean , calculated by the standard deviation divided by the square root of N . Propagation of error for Rates of Cleavage on Nucleosomes to Rates of Cleavage on DNA was calculated as follows:Error=kNucleosomekDNA ( SEMNucleosomeskNucleosomes ) 2+ ( SEMDNAkDNA ) 2 Data were fit on Graphpad Prism using a standard one phase decay model:Y= ( Y0−Plateau ) e−kt+Plateau where Y is the fraction of uncleaved DNA , Y0 is the value of Y at time = 0 , k is the observed rate constant ( min-1 ) and t is time ( min ) . dCas9 and a 2x molar ratio of sgRNA were incubated for 10 min at 37°C . Various concentrations of dCas9:sgRNA complex were incubated with 20 nM naked DNA or nucleosomes in binding buffer ( 20 mM Tris-HCl pH 7 . 5 , 100 mM KCl , 5 mM MgCl2 , 5% Glycerol , 1 mM DTT , and 0 . 02% NP-40 ) . Samples were incubated at room temperature for 1 hr before being run on native 0 . 5X TBE 6% polyacrylamide gels , visualized on a Typhoon scanner , and quantified using ImageJ . Fraction unbound was measured as the intensity of all unbound species divided by the total intensity . Fraction unbound was then converted to fraction bound:FractionBound=1−FractionUnbound , and binding curves were fit with:FractionBound=[Cas9:sgRNA]n ( [Cas9:sgRNA]n+K1/2n )	CRISPR is a method of editing the genetic material inside living cells and has enabled dramatic advances in a broad variety of research fields in recent years . The method relies on a bacterial enzyme called Cas9 that can be programmed , via short guide molecules made from RNA , to target specific sites in the cell’s DNA . Once bound to its target , the Cas9 enzyme cuts the DNA molecule; this often leads to changes in the DNA sequence . In nature , bacteria use the CRISPR-Cas9 system to defend themselves against viruses . However , this system also works in other cell types and can be reprogrammed to target almost any site in the DNA . To date , the CRISPR-Cas9 system has been used in fungi , worms , flies , plants , mammals and other eukaryotes . Yet , unlike in bacteria , much of the DNA in eukaryotes is wrapped around proteins called histones to form units referred to as nucleosomes . This means eukaryotic DNA is often tightly packaged , which makes it less accessible to other proteins . Nevertheless , eukaryotic DNA will spontaneously detach and reattach to the histones – a phenomenon that is commonly known as DNA “breathing” . Also , protein machines known as chromatin remodelers can move , assemble and take apart the nucleosomes in eukaryotic cells . However , because there is much still to learn about how CRISPR-Cas9 works in eukaryotic cells , it is not clear how nucleosomes affect this system’s activity . Isaac et al . have now used a simplified biochemical system to test how nucleosomes and chromatin remodelers affect CRISP-Cas9 activity . The system comprised purified Cas9 enzymes , short guide RNA molecules and nucleosomes . The experiments revealed that the Cas9 enzyme was able to cut DNA on nucleosomes when the DNA sequence allowed more spontaneous breathing or when chromatin remodelers were present to destabilize or move the nucleosome out of the way . These results suggest that by taking the placement of the nucleosomes into account , researchers can better predict how effective the CRISPR-Cas9 system will be at targeting a specific DNA sequence in a eukaryotic cell . The findings also suggest ways to make genome editing with CRISPR-Cas9 even more efficient .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "short", "report", "biochemistry", "and", "chemical", "biology" ]	2016	Nucleosome breathing and remodeling constrain CRISPR-Cas9 function
The multi-domain splicing factor RBM5 regulates the balance between antagonistic isoforms of the apoptosis-control genes FAS/CD95 , Caspase-2 and AID . An OCRE ( OCtamer REpeat of aromatic residues ) domain found in RBM5 is important for alternative splicing regulation and mediates interactions with components of the U4/U6 . U5 tri-snRNP . We show that the RBM5 OCRE domain adopts a unique β–sheet fold . NMR and biochemical experiments demonstrate that the OCRE domain directly binds to the proline-rich C-terminal tail of the essential snRNP core proteins SmN/B/B’ . The NMR structure of an OCRE-SmN peptide complex reveals a specific recognition of poly-proline helical motifs in SmN/B/B’ . Mutation of conserved aromatic residues impairs binding to the Sm proteins in vitro and compromises RBM5-mediated alternative splicing regulation of FAS/CD95 . Thus , RBM5 OCRE represents a poly-proline recognition domain that mediates critical interactions with the C-terminal tail of the spliceosomal SmN/B/B’ proteins in FAS/CD95 alternative splicing regulation . An essential step during the regulation of eukaryotic gene expression is the removal of non-coding intron sequences from pre-mRNA transcripts through the process of pre-mRNA splicing . The catalytic steps of pre-mRNA splicing are carried out by the spliceosome , a large and dynamic assembly of five small nuclear ribonucleoprotein ( snRNP ) complexes and more than 150 additional splicing factor proteins ( Wahl et al . , 2009 ) . Many splicing factors are involved in early steps of the assembly of the spliceosome through the recognition of short regulatory RNA motifs and/or through protein-protein interactions . Alternative splicing is the mechanism by which particular intronic or exonic regions are included or excluded to produce diverse mRNAs from the same gene ( Blencowe , 2006 ) . It is thought that more than 90% of human multi-exon genes undergo alternative splicing ( Pan et al . , 2008; Wang et al . , 2008 ) . The genomic diversity of eukaryotic gene expression is thus greatly expanded by alternative splicing of mRNA transcripts . Often , the protein products of alternative splicing have antagonistic roles in cellular functions and are implicated in human diseases ( Cooper et al . , 2009 ) . Notably , mutations in splicing factors that modulate alternative splicing decisions have been implicated in cancer ( David and Manley , 2010; Bonnal et al . , 2012 ) . A biologically important example of alternative splicing is found in the FAS gene ( also known as CD95 or APO-1 ) . The FAS gene encodes a transmembrane signaling protein that stimulates a pro-apoptotic signaling cascade upon binding of the FAS ligand at the cell surface ( Krammer , 2000 ) . Alternatively spliced FAS transcripts that exclude exon 6 encode a soluble Fas isoform that lacks the transmembrane domain . This soluble isoform can be secreted outside of the cell where it sequesters the FAS ligand and inhibits downstream activation of apoptosis ( Cheng et al . , 1994; Cascino et al . , 1995 ) . Thus , regulation of the alternative splicing of FAS can either stimulate or inhibit cell survival . The pro-apoptotic Fas protein plays an important role during T-lymphocyte maturation ( Liu et al . , 1995; Papoff et al . , 1996; Van Parijs et al . , 1998; Roesler et al . , 2005 ) and additional evidence implicates this isoform in the proliferation of cancer cells ( Chen et al . , 2010 ) . A number of splicing factors have been shown to modulate FAS alternative splicing , including RBM5 . The multi-domain RNA-binding protein 5 ( RBM5 ) regulates FAS splicing by promoting skipping of exon 6 ( Bonnal et al . , 2008 ) . RBM5 is a 92 kDa , multi-domain protein with an arginine-serine ( RS ) -rich region , two RNA Recognition Motifs ( RRM1 and RRM2 ) , two Zinc-Finger domains ( ZF1 and ZF2 ) , a C-terminal OCtamer REpeat ( OCRE ) domain ( Callebaut and Mornon , 2005 ) in addition to a glycine patch , and KEKE ( lysine/glutamate ) repeats ( Figure 1 ) . It belongs to the family of RNA Binding Motif ( RBM ) proteins , including RBM6 and RBM10 , which share a similar domain organization with RBM5 and have 30% and 50% amino acid identity , respectively ( Sutherland et al . , 2005 ) . The RBM5 ( also known as H37 and LUCA-15 ) and RBM6 genes are located in a chromosomal region 3p21 . 3 , which is frequently deleted in heavy smokers and lung cancers ( Oh et al . , 2002; Zabarovsky et al . , 2002 ) . RBM5 is known to regulate the alternative splicing of apoptosis–related genes , such as FAS and Caspase-2 ( Bonnal et al . , 2008; Fushimi et al . , 2008 ) . It has also been reported to suppress metastasis by modulating the expression of Rac1 , β-catenin , collagen and laminin ( Oh et al . , 2010 ) . However , RBM5 has also been found to be up-regulated in some aggressive forms of breast and ovarian cancers ( Oh et al . , 1999; Rintala-Maki et al . , 2007 ) . The possible dual role of RBM5 in cancer progression may be linked to its various activities as splicing regulator , including differential recognition of its pre-mRNA targets . In the regulation of FAS alternative splicing , RBM5 inhibits mature spliceosome formation by blocking association of the U4/U6 . U5 tri-snRNP complex ( in complex B ) after the splice sites flanking exon 6 have been recognized by U1 and U2 snRNPs ( Bonnal et al . , 2008 ) . RBM5 therefore appears to influence the pairing between the splice sites after complex A formation and thus promote tri-snRNP assembly at the distal splice sites , leading to exon 6 skipping ( Bonnal et al . , 2008 ) . Moreover , it was shown that the OCRE domain of RBM5 interacts with components of the tri-snRNP and is essential for the function of RBM5 in regulating FAS splicing ( Bonnal et al . , 2008 ) . OCRE domains , which are present in different proteins in various organisms ( Xiao et al . , 2013 ) , were identified by bioinformatic sequence analysis as a tandem of five imperfect octameric repeat sequences with triplets of aromatic residues ( Callebaut and Mornon , 2005 ) , expected to predominantly adopt β-strand secondary conformation . Due to the presence of OCRE domain in RNA binding proteins such as RBM5 , they were presumed to facilitate RNA binding ( Callebaut and Mornon , 2005 ) . However , at least in the case of RBM5 , the OCRE domain appears to play a role in mediating protein-protein interactions ( Bonnal et al . , 2008 ) . In order to provide molecular insights into the function of the OCRE domain in splicing regulation , we have combined structural and biochemical experiments with mutational analyses in vitro and in vivo . We show that the RBM5 OCRE domain directly binds to the spliceosomal SmN/B/B’ proteins . Sm proteins are core components of most spliceosomal snRNPs and form a heptameric ring composed of SmD1 , SmD2 , SmF , SmE , SmG , SmD3 and SmN/B/B’ , which jointly recognize the uridine-rich Sm site RNA motif in U1 , U2 , U4 and U5 snRNAs ( Kambach et al . , 1999; Will and Lührmann , 2001; Pomeranz Krummel et al . , 2009; Weber et al . , 2010; Leung et al . , 2011; Matera and Wang , 2014; Kondo et al . , 2015 ) . SmB and SmB' are encoded by two transcript variants from the SNRPB gene , while a different gene , highly expressed in brain and heart , encodes the homologous protein SmN ( Schmauss et al . , 1992 ) . SmN/B/B’ , SmD1 , and SmD3 have C-terminal extensions that include symmetrically dimethylated RG repeats ( Brahms et al . , 2001; Tripsianes et al . , 2011 ) . The SmN/B/B’ C-terminal tails contain additional proline-rich sequences , where these regions in SmN , SmB and SmB’ are 93% homologous ( van Dam et al . , 1989 ) . Here , we show that the RBM5 OCRE domain binds to the C-terminal proline-rich motifs present in SmB and SmN . The structure of the RBM5 OCRE domain adopts a unique β-sheet fold that recognizes the proline-rich C-terminal tails of the SmN/B/B' proteins through aromatic-CH interactions . We demonstrate that disruption of these interactions by mutations in the OCRE domain or in the proline-rich motifs of its ligands decreases the affinity between SmN/B/B’ and RBM5 in vitro and affects alternative splicing regulation of the FAS gene . Our results demonstrate that OCRE is a novel protein-protein interaction domain that mediates interactions with the core spliceosome in alternative splicing regulation . To gain insights into the molecular functions of the RBM5 OCRE domain , we determined the three-dimensional structure for the human RBM5 OCRE domain using solution NMR techniques . Previous reports ( Callebaut and Mornon , 2005 ) and a multiple sequence alignment of the related splicing factors RBM5 , RBM6 , and RBM10 indicated that the OCRE domain spans amino acid residues 451–511 of RBM5 ( Figure 1A , B ) . NMR chemical shift and 15N relaxation analyses of a construct with a C-terminal extension ( 451–532 , including the KEKE region ) did not reveal additional structural elements , suggesting that residues 451–511 indeed define the OCRE fold ( Figure 1—figure supplement 1A , B ) . An analysis of 13C secondary chemical shifts shows that the secondary structure of the RBM5 OCRE domain comprises six β-strands between residues 460 and 505 ( Figure 1B; Figure 1—figure supplement 1C ) . Notably , residues 452–459 preceding β1 , and the loops connecting the β-strands are well-structured and not flexible ( Figure 1—figure supplement 1B , C ) . The solution structure of the RBM5 OCRE domain is shown in Figures 1C , D and Figure 1—figure supplement 1D , E; structural statistics are provided in Table 1 . Residues 462–465 ( β1 ) , 470–473 ( β2 ) , 478–481 ( β3 ) , 486–489 ( β4 ) , 494–498 ( β5 ) , and 504–506 ( β6 ) make up six consecutive anti–parallel β-strands . An N-terminal extension ( residues 452–459 ) packs against one side of the β−sheet and is connected to the β1 strand by a 310 helix ( residues 459–461 ) ( Figures 1C , D ) . The six β–strands form a twisted β–sheet where aromatic side chains are exposed on opposite surfaces of the β–sheet ( Figure 1C , D ) . On one side , the aromatic side chains of Tyr464 ( in strand β1 ) , Tyr 471 ( β2 ) , Tyr480 ( β3 ) , Tyr487 ( β4 ) form extended aromatic side chain interactions on the surface . Tyr487 forms an additional hydrophobic cluster with Tyr462 and Pro457 from the N-terminal extension . Trp498 and Tyr505 from the C-terminal extension , and the N-terminal Tyr454 are also involved in this cluster ( Figure 1C ) . This forms a compact fold that is further stabilized by a salt bridge involving Lys453 and Glu503 and thus brings the N- and C-terminal regions of the OCRE domain in close spatial proximity . Due to the interactions with residues from the N-terminal extension this aromatic surface of the OCRE domain is shielded from the solvent . In contrast , on the other side of the β-sheet , the interactions of the aromatic side chains of Tyr470 ( β2 ) , Tyr479 ( β3 ) , Tyr486 , Tyr488 ( β4 ) , and Tyr495 , Tyr497 ( β5 ) form an aromatic surface with the tyrosine hydroxyl groups exposed to the solvent ( Figure 1D ) . The N-terminal extension has an extended conformation consistent with secondary 13C chemical shifts ( Figure 1—figure supplement 1C ) . Although the topology of the six antiparallel β-strands is rather simple , the twisted β-sheet of the OCRE domain appears unique . Structural similarity searches with Dali ( Holm et al . , 2008 ) and SSM ( Krissinel and Henrick , 2004 ) did not reveal any significant structural homologs ( Z-scores <2 ) , indicating that the OCRE domain represents a unique fold . While the β-sheet fold is quite simple , the specific arrangement of the N-terminal extension represent the unique features of the OCRE fold . The electrostatic surface potential of the OCRE domain is predominantly negatively charged with some hydrophobic patches ( Figure 1—figure supplement 1E ) , consistent with a potential role in protein–protein interactions and less likely for nucleic acid binding . To get initial insights into functionally relevant residues and subdomains , we focused on surface-exposed amino acids conserved between OCRE domains from different RBM proteins . These include Tyr495 , Tyr497 on one surface of the fold and Tyr454 , Glu501 and Asp458 on the opposite side . To assess the functional relevance of these residues , we generated triple and double point mutations ( Y495/Y497/E501 ( YYE ) >AAK ) and ( Y454/458 ( YD ) >AR ) , respectively . The mutants were expressed in HeLa cells , co-transfected with a FAS alternative splicing reporter ( Forch et al . , 2000 ) and the pattern of Fas exon 6 inclusion/skipping analyzed by RT-PCR . While the YD mutant had similar activity as the WT protein in promoting Fas exon 6 skipping , the YYE mutation failed to induce significant levels of FAS exon 6 skipping , similar to the effect of deleting the entire OCRE domain ( Figure 1E ) . To further investigate the importance of the residues in the YYE cluster , single point mutants were generated ( Y495A , Y497A , E501D/K ) and tested by co-transfection . Replacement of the aromatic tyrosines that are exposed on one surface of the β-sheet to alanine impaired the function of the protein in FAS alternative splicing regulation , arguing that the tyrosine residues in the surface of the β-sheet are important for the function of the OCRE domain in splicing regulation ( Figure 1E ) . It was previously shown that the RBM5 OCRE domain mediates interactions with protein components of the U5 snRNP complex , but the direct binding partners were not determined ( Bonnal et al . , 2008 ) . SmN , a core snRNP protein highly homologous to SmB/B' , was among proteins , identified by mass-spectrometry , pulled down with GST-OCRE RBM5 from HeLa cell extracts ( Bonnal et al . , 2008 ) . We also found RBM5 as an interacting partner in a two-hybrid screen using the C-terminal tail of human SmB as bait ( unpublished results ) . SmN/B/B' are components of the Sm core , a heptamer of Sm proteins that associates with all spliceosomal snRNPs except U6 . While SmB and SmB’ are ubiquitous , SmN is primarily found in neuronal and cardiac cells but has also been found to be expressed in HeLa cells ( Sharpe et al . , 1990 ) . The three SmN/B/B’ isoforms have a similar domain architecture , comprising a small globular N-terminal Sm domain ( Kambach et al . , 1999 ) and a C-terminal region that is expected to be intrinsically disordered . The C-terminal region starts with an arginine-glycine ( RG ) -rich region at ~residue 90 that harbors symmetrically dimethylated arginine residues , and additionally comprises a 60–70 residue proline-rich region of unknown function beyond residue 167 ( Figure 2A ) . As the globular N-terminal Sm domain in SmN/B/B’ proteins participates in the formation of the heptameric Sm core in the U snRNP complexes ( Kambach et al . , 1999; Pomeranz Krummel et al . , 2009 ) it seems unlikely that it could interact with the RBM5 OCRE domain . We thus considered the possibility that , in particular , the C-terminal domain of SmN/B/B’ could directly bind to the RBM5 OCRE domain . As an initial test , we verified that recombinant purified GST-RBM5 , or a derivative containing the C-terminal ( OCRE-containing ) domain , pulled down in vitro translated SmB , while a GST-fusion of the N-terminal domain of RBM5 did not ( Figure 2B ) . Next we carried out pull-down experiments using recombinant GST-tagged RBM5 OCRE domains ( wild type and mutants where surface-exposed tyrosine residues are replaced by alanine ) and assessed the ability to pull down purified recombinant N-terminal T7 epitope-tagged-SmN protein produced in mammalian cells ( Figure 2C; Figure 2—figure supplement 1A ) . These experiments demonstrate that recombinant GST-tagged OCRE domain binds SmN in vitro and that the mutations Y479A , Y488A , or Y495A diminish SmN binding ( Figure 2C ) , consistent with the decrease in activity of Y495A in splicing regulation ( Figure 1E ) . The mutations Y486A , Y497A and E501K do not significantly impair the interaction ( Figure 2C ) . Interestingly , mutation of Y495 to tryptophan or phenylalanine did not compromise binding , arguing that an aromatic residue at this position is required for the interaction ( Figure 2—figure supplement 1 ) . To further characterize these interactions by an independent method , we performed yeast-two-hybrid assays and tested whether the C-terminal tail of SmB binds to different variants of RBM5 . These experiments demonstrate a robust interaction between RBM5 and the SmB tail , where the OCRE domain is found to be necessary and sufficient for the interaction ( Figure 2D ) . Moreover , mutations of surface-exposed tyrosines to alanine reduce the interaction , while mutation to another aromatic amino acid ( phenylalanine ) maintains the interaction ( Figure 2E ) as was observed for the full-length SmN protein ( Figure 2C; Figure 2—figure supplement 1 ) . We next wished to identify the region in the Sm tails that mediates the OCRE interaction by using NMR and isothermal titration calorimetry ( ITC ) experiments . We studied binding of different constructs of the C-terminal tail of SmN , which harbors a region comprising Arg-Gly motifs ( residues 95–134 , not shown ) and a proline rich region ( residues 167–240 , up to the C-terminal end , Figure 2A ) ( Weber et al . , 2010 ) . The interaction of the RBM5 OCRE domain was assessed by monitoring NMR 1H , 15N chemical shifts of 15N-labeled OCRE domain during a titration with different SmN-derived peptides ( Figure 2—figure supplement 2 ) . Titrations with two different SmN fragments ( residues 97–196 and 167–196 ) that both comprise parts of the proline-rich region gave rise to significant and comparable chemical shift perturbations ( CSPs ) ( Figure 2—figure supplement 2A , B ) , suggesting that it is the proline-rich region , which is common to both peptides , that mediates the interaction . A peptide comprising the complete proline-rich region ( residues 167–240 ) shows comparable chemical shift changes . SmN ( 167–196 ) and ( 167–240 ) , both of which comprise about one half and the complete proline-rich region , bind to RBM5 OCRE with dissociation constants of KD = 195 μM and 41 μM , respectively , as determined by ITC ( Figure 2—figure supplement 2F; Table 2 ) . SmB and SmN share a poly-proline-rich region in their C-terminal tails , with 80% sequence identity and 98% sequence similarity ( Figure 2A ) . Upon titration of the SmB C-terminal region ( residues 167–231 ) to 15N-labeled OCRE , significant NMR CSPs are observed ( Figure 2—figure supplement 2E ) . Moreover , the observed CSPs are very similar to those seen with the SmN titration and affect the same residues , indicating that SmB and SmN bind to the same site in the RBM5 OCRE domain ( Figure 2—figure supplement 2G ) . The binding affinity of the SmB and SmN peptides to OCRE , determined by isothermal titration calorimetry ( ITC ) gave similar KD values of 21 μM and 41 μM , respectively ( Table 2 ) . Considering the high sequence similarity between SmB , SmB’ and SmN , it is expected that OCRE will interact with proline-rich sequences present in SmB and SmB’ as well . It is interesting to note that the NMR CSPs observed for the OCRE domain when titrated with a peptide comprising just one proline-rich motif ( PRM , residues 219–229 ) ( Figure 2F ) or the larger fragments of SmN are very similar ( Figure 2—figure supplement 2A–D ) . This indicates that all SmN fragments harboring PRMs utilize the same binding site on the OCRE domain . The strongest CSPs are observed for the amides of Y470 , Y488 , and Y495 , which map to the cluster of exposed aromatic tyrosines on β2 , β4 , and β5 of the OCRE domain , respectively ( Figure 2G ) . However , the short peptide containing four consecutive proline residues induces comparable CSPs only at much higher peptide:OCRE ratio compared to the longer Sm tail peptides , thus indicating higher binding affinity for the longer Sm tail constructs ( Figure 2—figure supplement 2 ) . Nearly identical spectral changes are induced by the peptide , suggesting that all PRMs contribute to the overall affinity for the OCRE domain . Thus , the interaction with multiple motifs is enhanced by avidity , e . g . , interaction of multiple PRMs within the Sm tails provides increased local concentration of PRM ligand motifs and thereby contributes to the significantly higher affinity of Sm-tails compared to a single PRM . In order to assess the sequence requirements of PRMs for binding to the OCRE domain , we compared the amino acid sequences of the SmN/B tails ( Figure 3A ) . This analysis reveals the presence of multiple ( five to six ) PRMs harboring three or four consecutive prolines in SmB or SmN , which are flanked by conserved arginine residues within ±3 residues on either side of the poly-proline motif ( Weber et al . , 2010 ) . The role of these conserved sequence features was probed by compared binding affinities of wild type and variant Sm tails to the RBM5 OCRE domain using ITC experiments ( Figure 3A ) . Reducing the number or PRMs significantly decreases the binding affinity , consistent with avidity effects and contribution to the overall affinity by the presence of multiple PRMs . Replacing four prolines by four alanines has little effect . This observation is consistent with the fact that consecutive stretches of alanines adopt a poly-proline type II ( PPII ) helical conformation , as confirmed by CD spectra ( Figure 3—figure supplement 1 ) . As the SmN/B/B’ PRMs adopt a PPII helical conformation when bound to the OCRE domain ( see below ) , the replacement by four alanines can thus be tolerated to some extent . Notably , charge reversal of two arginines flanking the poly-proline motif on either side significantly decreases the binding affinity . These data suggest that a PPII helical conformation flanked by positively charged arginine residues is specifically recognized by the OCRE domain . The contribution of individual residues within the PRM motif was determined by comparing relative binding affinities of wild type and mutant 11-mer peptides that exhibit sequence features observed in the Sm tails ( Figure 3B ) . As the binding affinities of these peptides are beyond the detection limit of ITC , we resorted to a semi-quantitative NMR CSP-based approach . We devised a normalized CSP score for a set of seven amide signals surrounding the binding pocket showing significant CSPs in the OCRE domain to obtain a proxy for the relative binding affinities ( see Materials and methods for details ) ( Figure 3B—figure supplement 2 ) . As the comparison between different peptides is based on the CSP score derived from the same set of residues , it indirectly reflects their relative binding affinities . Several themes emerge from this comparison suggesting that the OCRE domain has a preferred binding motif , but can also accommodate some sequence variants . First , wild type peptides with three prolines induce smaller CSPs than those with four proline motifs and binding saturation is achieved only at higher peptide concentration , indicating a lower affinity . PRMs harboring only three consecutive proline residues show 2–3-fold reduced affinity relative to four proline motifs . Second , an APAP motif , which disrupts PPII conformations , has a strongly reduced CSP score , suggesting that the structure and composition of the PPII are important for interaction . Third , both arginines flanking the PRM motif are important for the interaction , as a double charge reversal ( R→E ) strongly reduces the binding affinity . Neutralizing the charge with R→A mutations had a smaller effect , presumably because no charge clashes are introduced with the negatively charged surface of the OCRE domain ( Figure 1—figure supplement 1E ) as is the case for R→E mutations . Taken together , the analysis shown in Figure 3 indicates that a PPII conformation mediated by four consecutive proline residues with flanking positively charged residues is optimal for OCRE binding . Notably , the arginine residue preceding the PRM appears more important for the interaction than the one located C-terminal to the PRM . To determine molecular details of the recognition of SmN/B/B’ by the RBM5 OCRE domain , we determined its structure in complex with a proline-rich peptide derived from the C-terminal region of SmN ( GMRPPPPGIRG ) corresponding to SmN residues 219–229 ( Figure 4; Figure 4—figure supplement 1A ) . This motif is also found within residues 219–229 in SmB with only one amino acid difference ( GMRPPPPGMRG ) . The structure of the complex is defined by numerous NOEs , and is supported by 109 distance restraints derived from intermolecular NOEs ( Figure 4—figure supplement 1B , Table 2 ) . The structure shows that the central proline stretch ( SmN Pro222-Pro225 ) of the bound peptide adopts a PPII helix ( Figure 4 ) . The PPII conformation is also indicated by the 13C chemical shifts of the peptide . Considering the large excess of peptide , the chemical shifts largely reflect the unbound state . This indicates that the PPII conformation is already preformed in the absence of the OCRE domain , consistent with our CD data ( Figure 3—figure supplement 1 ) . The central part of the proline-rich peptide interacts with conserved tyrosine residues in the OCRE domain in strands β3 ( Tyr470 , Tyr479 ) , β4 ( Tyr486 , Tyr488 ) and β5 ( Tyr495 ) ( Figure 4A–C ) . On one side , Tyr488 stacks with SmN Pro222 and on the other with Tyr479 , which itself contacts SmN Pro224 by T-stacking . The Tyr486 side chain stacks with SmN Pro225 and interacts with Tyr495 , which in turn contacts Pro223 in the SmN peptide . A unique binding orientation of the peptide is defined by a combination of the stacking interactions with the PPII conformation of the peptide , and interactions with the arginine residues flanking the proline-rich stretch on either side , as well as SmN Ile227 . The side chain of SmN Arg221 , preceding the poly-proline stretch is poised to interact with the hydroxyl groups of OCRE Tyr472 , Tyr479 and Ser490 in loop β1-β2 ( Figure 4D ) . The side chain of SmN Arg228 can form potential hydrogen bonds with the side chains of Asp481 and Ser484 in the OCRE domain ( Figure 4E ) . Both residues are conserved in loop β3-β4 in the OCRE domains of RBM5 and RBM10 ( Figure 1B ) . In addition , the hydrophobic side chain of SmN Ile227 , located C-terminal of the proline-rich motif packs against SmN Pro223 , consistent with numerous intermolecular NOEs between the two side chains , and thus shields the hydrophobic PPII helix from solvent exposure ( Figure 4C ) . To enable these contacts involving the Ile227 side chain , the presence of a preceding Gly226 , enables a specific backbone conformation by allowing unusual backbone torsion angles . Considering the specific spacing of the arginine residues to the four-proline-stretch in the SmN ligand , these interactions define a unique , unambiguous orientation of the peptide along the OCRE β-sheet surface . The structure of the OCRE domain does not undergo significant conformational changes upon binding to the SmN peptide , with a backbone coordinate r . m . s . d . of 1 . 7Å between the free and SmN-bound OCRE domain . However , the tyrosine side chains involved in the SmN interaction slightly rearrange to optimize contacts with the proline-rich SmN ligand . To summarize , the RBM5 OCRE domain uses an array of tyrosine side chains that are exposed from one side of its β-sheet to recognize a proline-rich motif in a PPII conformation . Additional specific interactions are mediated by two positively charged side chains flanking the N- and C-terminal sides of the poly-proline helix in the bound SmN peptide , which in part involve the side chain hydroxyl groups of the exposed tyrosines , thus consistent with the strong conservation of tyrosines in the OCRE domain . These structural insights reveal that the OCRE domain represents a novel proline-rich motif binding domain . To evaluate the importance of contacts observed in the OCRE-SmN structure we carried out a mutational analysis of the OCRE domain . We compared the binding of wild-type and mutant OCRE domains to SmN/B/B’ in vitro using GST pull-downs , ITC and NMR titrations , and analyzed the functional activity of wild-type and OCRE domain mutants in alternative splicing regulation . We focused on the effects of mutation of tyrosine residues located within the proline-rich motif binding pocket , including Tyr470 , Tyr479 , Tyr486 , Tyr488 and Tyr495 and control mutations of residues that are not involved in the SmN interactions , i . e . Tyr497 , Glu501 and Tyr454/Asp458 . To analyze whether the effects of the OCRE mutations could induce ( partial ) disruption of the OCRE structure , recombinant OCRE proteins were expressed and purified and structural integrity analyzed by NMR ( Figure 4—figure supplement 2 ) . The NMR spectra of the Y495F , Y495W and E501K OCRE domains suggest that these proteins are globular folded . Small changes in position and intensities of the NMR signals for the Y495F/W mutations likely reflect that aromatic side chains have significant contributions to the chemical shifts of surrounding residues , but are still consistent with the integrity of the overall fold . NMR spectra of the Y495A , Y495T , Y497A and the double mutant Y454A/D458K suggest partial unfolding of strand β5 and the N-terminal extension , respectively , even though the overall fold appears to remain intact . Larger changes observed for the Y479A , Y486A or Y488A OCRE domains suggest more significant destabilization of the fold ( Figure 4—figure supplement 2 ) . GST-pulldown experiments ( Figure 2—figure supplement 1 ) performed with Tyr495 mutations in the RBM5 OCRE domain confirm the importance of the aromatic side chain of Tyr495 for Sm binding . Next , we compared the binding affinities of wild type OCRE domains with Y495A , Y495F , Y488A and of the double mutant Y454A/D458K using ITC ( Table 2 ) . Consistent with the structural analysis and the observed effects in splicing regulation , the Y495A and Y488A mutants show strongly reduced binding to the SmN-derived proline-rich ligands , KD = 172 and 220 μM , respectively , compared to the wild type OCRE domain ( KD = 41 μM ) . The Y495F mutation does not strongly affect the SmN interaction KD = 87 μM . The double mutant Y454A/D458K shows a slightly reduced affinity to SmN KD = 106 μM , which , however , might reflect partial destabilization of the mutant OCRE fold ( Figure 4—figure supplement 2 ) . Yeast two-hybrid assays confirmed the importance of an aromatic residue at this position also for the interaction of the OCRE domain with SmB , as the Y495A mutation reduced the interaction between RBM5 OCRE and SmB while the Y495F retained binding competence ( Figure 2E ) . To probe the functional effects of the OCRE domain mutations , we tested the splicing activity of full-length RBM5 harboring mutations in the ligand binding region of the OCRE domain ( Figure 5; Figure 5—figure supplements 1 and 2 ) . As shown in Figure 1E , substitutions of conserved tyrosines that are involved in SmN interactions ( Tyr495 and Tyr497 ) impair the activity of RBM5 in Fas alternative splicing regulation ( Figures 5A , C; Figure 5—figure supplement 2 ) . Further mutations of aromatic residues , including tyrosines 470 , 479 , 486 and 488 by alanine compromise the Fas exon 6 skipping activity of RBM5 to different extents , consistent with the reduced binding of these mutants to SmN-derived proline-rich ligands ( see above ) and with the location of these residues in the binding pocket for the SmN poly-proline motif ( Figures 4 , 5A–C ) . Substitution of the important Tyr495 residue by threonine or glutamate also compromised activity , while replacement by other aromatic residues ( phenylalanine or tryptophan ) retained full activity in splicing assays ( Figure 5A–C; Figure 5—figure supplement 2 ) , consistent with sustained SmN interaction ( Figure 2; Figure 2—figure supplement 1 ) . This argues that , in agreement with the structural analysis , the aromatic nature of the tyrosine is a key feature of the splicing regulatory properties of the OCRE domain . In contrast , mutations of residues , Tyr454 , Asp458 , Glu501 ( Figure 1E ) and Ser468 and Asn483 ( Figure 5—figure supplement 1 ) , which are remote from the SmN binding site , had only moderate effects on the activity of RBM5 on FAS splicing . All the mutant proteins accumulated to levels similar as the wild type ( Figure 5B ) . These data further argue that specific recognition of the proline-rich SmN peptide by a cluster of aromatic residues in the RBM5 OCRE domain is required for the function of RBM5 as a splicing regulator . We next wished to probe the contributions of different regions in the SmN/B/B’ proteins for FAS alternative splicing . Consistent with previous results ( Saltzman et al . , 2011 ) , we observed that the knock down of the SmN/B/B’ proteins by siRNA in HeLa cells led to an increase in the level of FAS exon 6 skipping in endogenous transcripts or in transcripts derived from a Fas reporter ( Figure 5—figure supplement 3A , C ) . These effects were attenuated by strengthening FAS exon 6-associated 5’ splice site ( Figure 5—figure supplement 3C ) , as has been previously reported ( Saltzman et al . , 2011 ) . Co-expression of SmN with a FAS alternative splicing reporter bearing a mutation ( Fas U-20C ) that increases FAS exon 6 skipping ( Izquierdo et al . , 2005 ) , led to increased levels of exon inclusion , while neither expression of the amino terminal part nor of the C-terminal part of the protein did ( Figure 5—figure supplement 3B ) . While expression of SmN variants harboring mutations in the proline-rich stretches ( either to glycine or to APAP motif ) does not significantly compromise exon inclusion , mutation of the arginines flanking stretches of four prolines does ( Figure 5—figure supplement 3D ) . However , the results of SmN/B/B’ overexpression experiments are highly variable and difficult to interpret , perhaps because of additional complex effects of Sm protein overexpression on snRNP biogenesis/activity . Taken together , the structural and functional analyses show that three tyrosine residues in the RBM5 OCRE domain ( Tyr479 , Tyr488 and Tyr495 ) play crucial roles in the recognition of proline-rich regions present in the conserved C-terminal tails of SmN/B/B’ and reveal a tight correlation between this binding and the activity of RBM5 as a regulator of FAS alternative splicing . We have previously shown that the alternative splicing activity RBM5 depends on the C-terminal region of RBM5 and that recombinant RBM5 inhibits the transition from the pre-spliceosomal complex A to spliceosomal complex B ( Bonnal et al . , 2008 ) . To assess which region of the RBM5 protein is required , we performed in vitro spliceosome assembly assays with AdML pre-mRNA , using the N- and C-terminal halves of RBM5 as well as with a C-terminal version that lacks the OCRE domain or a fragment that includes only the OCRE and KEKE domains ( Figure 6—figure supplement 1 ) . These and additional transient transfection experiments ( Figure 6—figure supplement 2 ) confirm that the inhibition of complex B formation depends on the C-terminal region of RBM5 and reveal that the OCRE domain is necessary but not sufficient for this activity . This is consistent with a key role for the interaction of OCRE with the PRM in the spliceosomal tails in the tri-snRNP , but argues that additional interactions involving the C-terminal region of RBM5 are important , including for example the previously reported interaction with U2AF65 ( Bonnal et al . , 2008 ) . The Octamer Repeat ( OCRE ) domain of RBM5 adopts a unique three-dimensional fold with a large number of conserved tyrosine residues . Although the primary sequence signature could suggest a linear sequence of octamer repeats , our structural analysis clearly reveals that the OCRE domain adopts a small globular fold . We show that the twisted β-sheet of the OCRE domain exposes conserved tyrosine residues that provide a platform for the recognition of a proline rich motif ( PRM ) . This interaction is based upon a network of tyrosine residues as well as anchoring residues located in the loops of the β-strands . The SmN PRM is flanked by a positively charged arginine residue preceding the four consecutive prolines , followed by a hydrophobic residue and another arginine . Taking into account the sequence conservation , binding studies ( Figure 3 ) , and the details of molecular recognition of the SmN peptide ( Figure 4 ) , we propose that the OCRE domain recognizes a RPPP ( P ) GϕR consensus motif . A key feature of the motif is a central PPII helix , which is recognized by a network of parallel and aromatic T-stacking of Tyr470 , Tyr479 , Tyr482 , Tyr488 and Tyr495 with the proline side chains in the PPII helix . These tyrosines are exposed at one side of the OCRE β-sheet and collectively embed the PPII helix . Specific contacts that define the orientation of the bound peptide are mediated by two flanking arginine residues , with a stronger contribution of the N-terminal arginine ( SmN Arg221 , Figure 3B ) . The glycine residue may enable an unusual backbone conformation such that hydrophobic residues ( ϕ , Ile/Met/Val flanking different PRMs in the SmN/B/B’ C-terminal tails , Figure 3A ) can shield the central hydrophobic PPII motif from solvent exposure . Notably , unbound PRMs harboring 3–4 consecutive prolines already adopt ( at least partially ) a preformed PPII conformation and thus reduce the entropy loss associated with binding to the OCRE domain . The somewhat reduced affinity for peptides with only three prolines may reflect a smaller propensity of forming PPII conformation in the free ligands and the lack of a preceding arginine , which is consistently present in all four proline PRMs . The tertiary fold and the mode of PRM recognition by the OCRE domain is distinct from other PRM binding domains , such as GYF , SH3 , and WW domains ( Figure 7A ) . Although – as a common feature – aromatic residues are used to interact with proline residues and to recognize the PPII helix ( Ball et al . , 2005 ) , different secondary structure elements are found in GYF ( α-helical ) , SH3 ( α/β fold ) , WW ( β-sheet ) and OCRE domains ( β-sheet ) . Interestingly , a somewhat related PRM motif in the CD2 protein ( PPPPGHR ) was reported to interact with the GYF domain of the CD2BP2 protein ( Freund et al . , 2002 ) . The CD2BP2 GYF domain and the FBP21 WW domain have been previously implicated in binding to the PRM sequences of SmB , suggesting that interactions between PRMs and these factors may be critical for the function of these proteins in spliceosome regulation ( Bedford et al . , 1998; Klippel et al . , 2011 ) . An interesting aspect of the OCRE/Sm interaction is that the presence of multiple PRM motifs greatly enhances the overall affinity ranging from KD ≈ mM for a single PRM peptide to KD = 20–40 µM for complete Sm tails , and thus comparable to other PRM binding domains . The relatively weak affinity of a single PRM found in the Sm tails argues that avidity effects due to the presence of multiple motifs play an important role for high affinity binding by increasing the local concentration of PRM motifs available for OCRE binding . We note that avidity effects provided by multiple binding sites have also been observed for other PRM binding domains ( Varani et al . , 2000; Klippel et al . , 2011 ) , and that this feature is reminiscent of the recognition of dimethyl-arginine residues in the Sm tails by Tudor domains ( Tripsianes et al . , 2011 ) . On the other hand , the weak binding affinity of a single motif indicates that the interaction between RBM5 and SmN/B/B’ may have been selected to be transient , as expected for a regulatory interaction that needs to promote interactions with the spliceosome but also needs to be disrupted at later steps of the splicing reaction . Previously , we have shown that the splicing regulation by RBM5 acts at the stage of spliceosomal B complex formation ( Bonnal et al . , 2008 ) . The role of RBM5 and its OCRE domain in this context could be to attract the U4/U6 . U5 tri-snRNP to the splice sites . The fact that the SmN/B/B’ tails are intrinsically disordered and extend very far away from the seven-membered Sm core ring in U snRNPs ( Figure 7B , C ) , is consistent with such an activity . In fact , the recent EM structure of the tri-snRNP ( Nguyen et al . , 2015 ) revealed that the Sm cores of U4 and U5 , as well as the LSm ring of U6 are located at the outside of this large RNP ( Figure 7C ) . Thus , the C-terminal tails of SmN/B/B’ of the U4 and U5 components are clearly accessible and could scan for possible binding partners of their proline-rich motifs , such as the RBM5 OCRE domain . This could help to localize the tri-snRNP to the splice sites that are regulated by RBM5 . SmB was described to participate in alternative splicing of its own pre-mRNA and many additional genes , including the FAS gene ( Saltzman et al . , 2011 ) . Recently , mutations in SmB were linked to cerebro-costo-mandibular syndrome ( Bacrot et al . , 2015 ) . The multi-domain protein RBM5 promotes FAS exon 6 skipping and it has been proposed that for this activity the protein modulates splice site pairing after the competing 5’ and 3’ splice sites have been recognized by U1 and U2 snRNPs , respectively ( Bonnal et al . , 2008 ) . This model is based upon the observation that RBM5 inhibits the transition from A ( U2 snRNP binding ) to B ( U4/U6 . U5 tri-snRNP binding ) complex formation in the introns flanking exon 6 . It is conceivable that interactions mediated by the C-terminal tails of SmN/B/B’ , which are present in U1 , U2 and the tri-snRNP , facilitate the transition from A to B complex and that binding of the OCRE domain of RBM5 prevents these interactions and therefore the progression of spliceosome assembly . Direct or indirect association of RBM5 with particular regions of the pre-mRNA – likely involving other regions of the protein including RRM or zinc finger domains – may prevent A to B transition of pre-spliceosomes assembled on the flanking introns of exon 6 , but fail to prevent tri-snRNP assembly on the distal splice sites , thus facilitating exon skipping . Alternatively , or in addition , RBM5 interaction with the C-terminal tails of SmN/B/B’ may lead to a general decrease in snRNP function , possibly more acute for U2 snRNP . As decreased activity of U2 snRNP – e . g . by individual knock down of its protein components – is known to result in increased FAS exon skipping ( Papasaikas et al . , 2015 ) , RBM5-mediated reduction in U2 function could potentially explain at least part of the effects of the protein on FAS splicing regulation . Taken together , our study provides a molecular understanding of how the OCRE domain of RBM5 interacts with proline-rich sequences of the SmN/B/B' tail and thus identifies a key interaction essential for regulation of alternative splicing . It is interesting to note that , based on primary sequence alignments , the OCRE domains of RBM5 and RBM10 are expected to adopt highly similar structures and possibly functions . In fact , the recently reported structure of the RBM10 OCRE domain is highly similar to RBM5 OCRE ( backbone coordinate r . m . s . d . 1 . 1Å ) ( Martin et al . , 2016 ) . Interestingly , mutation of a conserved tyrosine residue ( corresponding to Tyr470 in RBM5 , where it contributes to the recognition of the PRMs by the OCRE domain ) is frequently found associated with lung carcinoma ( Imielinski et al . , 2012 ) . It is thus tempting to speculate that splicing defects that are linked to impaired interactions with PRM in SmN/B/B’ proteins contribute to the pathogenesis . RBM5 ( Swiss Prot P42756 ) OCRE domain ( 451–511 ) and all the mutants were subcloned into a pETM11 vector ( with a N-terminal His6-tag ) from the corresponding GST-RBM5 full-length DNA . The several SmN and SmB constructs were cloned into a pETM30 vector , containing a N-terminal His6-tag followed by glutathione S-transferase protein . In both vectors , a TEV site is present before the corresponding protein . All the proteins were produced by overexpression in E . coli BL21 cells at 20°C for 16 hr after induction with 0 . 5 mM of IPTG when cells were at approximately O . D . 0 . 7 in media supplemented with 30 μg/ml kanamycin . For unlabeled proteins , bacteria were grown in Luria broth . For isotope-labeled proteins , bacteria were grown in M9 minimal media supplemented with 13C-glucose and/or 15NH4Cl . Cell lysates were suspended in buffer containing 20 mM Tris pH 8 , 300 mM NaCl , 5% glycerol , 10 mM imidazole , and 2 mM of 2-mercaptoethanol and purified with Ni-NTA Superflow beads ( Qiagen , Hilden , Germany ) using standard conditions . After overnight cleavage of the fusion protein with tobacco etch virus protease , proteins were purified in a gel filtration column 26/60 sephadex II ( GE Healthcare , München , Germany ) and buffer was exchanged to NMR buffer; 20 mM sodium phosphate pH 6 . 5 , 50 mM NaCl . For measurements in D2O , the protein was lyophilized and dissolved in D2O . The OCRE domain used for NMR studies thus comprises residues 451–511 preceded by a GAM tripeptide that results from the TEV cleavage . The SmN 219–229 peptide ( Peptide Specialty Laboratory , Heidelberg , Germany ) was dialyzed against water , lyophilized and then dissolved in NMR buffer . GST-tagged RBM5 OCRE domain proteins were expressed in BL21 cells . Bacteria were transformed and grown in one liter LB to an absorbance at 600 nm of 0 . 6 before the induction of the expression with 1 mM IPTG for 3 hr at 37°C . The pellet was resuspended in cold buffer X ( 20 mM Tris pH 7 . 5; 1 M NaCl; 0 . 2 mM EDTA; 1 mM DDT and protease inhibitors cocktail ( Roche Diagnostics , reference 11697498001 , Mannheim , Germany ) and sonicated . The supernatant was collected after centrifugation for 20 min at 10 , 000 rpm at 4°C and incubated with Glutathione Sepharose 4B beads for 15 min on a rotating wheel at 4°C . The beads were washed with 50 ml buffer X 3 times 10 min and eluted on column with 50 mM glutathione , 100 mM Hepes pH 8 , 0; 1 mM DTT . The selected fractions were dialyzed against buffer D ( 20 mM Hepes pH 8 . 0; 20% glycerol; 0 . 2 mM EDTA; 0 . 1 M KCl; 1 mM DTT and 0 . 01% NP40 ) , frozen in liquid nitrogen and stored at −80°C . T7-tagged SmN protein was expressed and purified from HEK 293T cells ( Cazalla et al . , 2005 ) . The N-terminal ( aa 1–318 ) and C-terminal ( 320–815 ) regions of RBM5 were produced by PCR amplification using specific primers and cloned into the pET15b vector ( Novagen ) generating plasmids N-term and C-term , respectively . The plasmid C-term ΔOCRE was obtained by removing the aa 452 to 511 of plasmid C-term using site-directed mutagenesis . Recombinant His6-tagged proteins were expressed in E . coli ( BL21-CodonPlus ) , solubilized under denaturing conditions ( 6M Guanidine hydrochloride ) ( Wingfield et al . , 2001 ) and purified by Ni-NTA affinity chromatography . After purification , the proteins were dialysed overnight against buffer D using Slide-A-Lyser devices ( Pierce ) . Spectra were recorded at 298 K in DRX500 , 600 and 900 spectrometers with cryogenic triple resonance probes using 13C , 15N labeled OCRE sample ( 1 mM ) for the apo structure , and a 13C , 15N labeled OCRE and unlabeled peptide ( 2 mM and 14 mM , respectively ) for the complex . Data were processed with NMRPipe ( Delaglio et al . , 1995 ) and analyzed using NMRView ( Johnson and Blevins , 1994 ) . For backbone resonance assignments , standard experiments were recorded , including 3D HNCA , HNCACB , HN ( CO ) CACB experiments . Side chain resonances assignments were made with 3D HCCH-TOCSY and 3D 15N-HSQC TOCSY experiments . Distance restraints were derived from 15N- and 13C-resolved three-dimensional , 1H homonuclear two-dimensional ( mixing time 120 ms ) . In addition , for the complex structure , assignment and distance restraints information for the peptide were obtained by recording a 15N , 13C filtered TOCSY with the mixing times of 15 , 30 , 60 , 90 ms and NOESY experiments with NOE mixing times of 120 ms . H/D exchange experiments were recorded after lyophilization . 15N relaxation experiments ( T1 , T2 and {1H}–15N heteronuclear NOEs ) were measured for both the apo and bound form of OCRE . Chemical shift mapping on OCRE was done by monitoring the 2D 1H , 15N HSQC , 2D 15N-labeled OCRE ( 0 . 2 mM ) with an excess of unlabeled SmN constructs until no further changes in chemical shifts were observed in the 2D 1H , 15N HSQC spectra . Combined CSPs were calculated as Δδ ( ppm ) = ( ( 10*ΔδHN ) 2 + ( ΔδN ) 2 ) 1/2 . A semi-quantitative approach was taken to assess the contribution of the amino acid sequence of PRM motifs in Sm tails to the binding interaction . Due to the relatively weak interaction , NMR titrations were used to compare relative affinities between wild type peptides and peptides with specific mutations . The SmN derived wild type and mutant peptides were titrated into OCRE at a 1:10 ratio of OCRE:peptide . CSPs from 7 OCRE residues ( Y470 , Y471 , Y479 , D481 , N483 , S490 , Y495 ) were added for each of the peptide titration and normalized with that of the wild type peptide ( GMRPPPPGIRG ) . The score is used to compare the relative affinities of the various peptides with the wild type ( Figure 3B ) . An overlay of 1H , 15N-HSQC spectra of OCRE bound to wild type ( GMRPPPPGIRG ) or variant peptides -GMAPPPPGIRG , GMEPPPPGIEG illustrates the different extent of CSPs observed for seven residues in the OCRE domain that are most strongly affected by the binding ( Figure 3—figure supplement 2 ) . These residues were selected to define the CSP score . For the apo structures , automatic NOE assignments and structure calculations were initially performed by CYANA3 ( Güntert , 2009 ) . Subsequently , NOEs were manually checked and applied as distance restraints together with dihedral angle restraints in a simulated annealing protocol using ARIA ( Linge et al . , 2001 ) and CNS ( Brunger et al . , 1998 ) . Dihedral restraints were derived from TALOS+ ( Shen et al . , 2009 ) , hydrogen bond distance restraints were applied based on secondary structure identified by TALOS+ , and added during structure calculations . For the complex structure , manual assigned NOEs were applied as distance restraints together with dihedral angle and measure hydrogen bond restraints in a simulated annealing protocol using ARIA ( Linge et al . , 2001 ) and CNS ( Brunger et al . , 1998 ) . Water refinement was performed on the final ensembles of NMR structures ( Linge et al . , 2003 ) . The structural quality of the 10 lowest energy structures out of 100 calculated structures was evaluated using ProcheckNMR ( Laskowski et al . , 1996 ) , and the iCING ( Doreleijers et al . , 2012 ) and PSVS ( Bhattacharya et al . , 2007 ) servers . Ramachandran statistics for the free RBM5 OCRE domain and the complex structure with SmN ( 219–229 ) are 91 . 5%/8 . 5%/0%/0% and 88 . 0%/9 . 8%/2 . 0%/0 . 2% in the most favored/additionally/generously/disallowed regions , respectively . Ribbon representations and the electrostatic surface potential were prepared with PYMOL ( DeLano Scientific , San Carlos , CA , USA ) . ITC experiments were performed using an ITC200 instrument ( MicroCal , Wolverton Mill , UK ) at 24°C . SmN constructs ( 168–196 , 168–240 ) and an SmN peptide ( 219–229 ) at 1 mM , 1 mM and 5 mM , respectively , were titrated into OCRE ( 100 μM , 100 μM and 1 mM , respectively ) . For the mutation analysis , purified OCRE mutants ( 2 mM ) were titrated to wild type SmN ( 29 . 5 μM ) or SmB ( 40 mM ) . All purified proteins were in the same buffer , 20 mM sodium phosphate pH 6 . 5 , 50 mM NaCl . The lyophilized peptide was dialyzed against water , lyophilized again , and then dissolved in the same buffer as the protein . The heat of dilution was measured by titrating SmN or OCRE mutants into buffer . The titration protocol consisted of one initial injection of 0 . 4 μL followed by 38 injections of 1 μl of the ligand into the protein sample with intervals of 120 s , allowing the titration peak to reach the baseline . Data were calculated using the program Origin v7 . 0 ( MicroCal ) and duplicates were measured for all the experiments . All CD spectra were recorded on a JASCO-J715 spectropolarimeter and analyzed with Spectramanager version 1 . 53 . 00 ( Jasco Corp . ) . The temperature was regulated using a Peltier type control system ( PTC-348WI ) . The spectra were recorded at 5°C in 20 mM sodium phosphate , 100 mM NaCl , pH 6 . 5 buffer from 190–260 nm wavelength with a 1 . 0 nm bandwidth , 0 . 5 nm pitch at a scan speed of 50 nm/min , in cuvettes with 0 . 1 cm path length . All spectra are presented as an average of 20 scans , obtained after buffer subtraction and plotted as mean residue ellipticity ( deg cm2 dmol−1 ) vs wavelength ( nm ) . All peptides were measured at 0 . 3 mM concentration in a buffer containing 20 mM sodium phosphate , 100 mM NaCl , pH 6 . 5 at 5°C . GST RBM5 OCRE and T7 SmN or 35S-labeled SmB proteins were incubated for one hour at 4°C degrees on a rotating wheel in 1 ml PBS supplemented with 0 . 1% Triton X100 . 45 µl of packed and equilibrated GSH beads ( Glutathione Sepharose 4B , GE Healthcare , reference 17-0756-05 ) were added and the samples were incubated for one hour more as before . The beads were then washed four times with 1 ml PBS-0 . 1% Triton X100 and the proteins were directly eluted in SDS loading dye at 95°C for 5 min under shaking , loaded on SDS gels , separated by electrophoresis . Proteins were revealed by autoradiography or analyzed by western blot using the following antibodies: anti-T7 ( T7-Tag Antibody HRP conjugate , Novagen , reference 69048 ) and anti-GST ( GST ( B14 ) -HRP mouse monoclonal , Santa Cruz Biotechnology , reference sc-138 HRP ) . Protocol of co-transfection , RT-PCR and western blot analysis were carried out as previously described ( Bonnal et al . , 2008 ) . siRNA against Sm proteins were carried out as described ( Saltzman et al . , 2011 ) . Cy5-CTP labeled AdML RNA bearing exon 1 – intron 1 – exon 2 was in vitro transcribed using T7 Megascript kit ( Ambion ) . The spliceosome assembly reaction was performed as described previously ( Mackereth et al . , 2011 ) with 10 ng/ul fluorescently labeled RNA and the indicated recombinant protein . After electrophoresis , the gel was analyzed directly with a PhosphorImager Typhoon . Quantification was carried out using Image Quant . The C-terminal domain of human SmB ( aminoacids 84–231; Accession n°: NM_003091 ) was PCR-amplified using oligonucleotides carrying EcoRI ( Forward ) and SalI ( Reverse ) restriction sites and the pGFP-huSmB plasmid as template ( Mouaikel et al . , 2003 ) . After agarose gel electrophoresis , the DNA fragment was purified using the GeneClean procedure and transferred into EcoR1-Sal1 cut pGBT9 vector carrying the DNA binding domain of Gal4 ( Fields and Song , 1989 ) . The various pACT2-RBM5 plasmids containing the RBM5 full length and mutated coding sequences in frame with the Gal4AD ( activation domain ) were constructed using the Gateway system and a pACT2-based vector according to the manufacturer's instructions ( Invitrogen ) . The plasmids used for amplification have been described previously ( Bonnal et al . , 2008 ) . All pDonor constructs were sequenced prior to proceed to LR recombination . The sequences of cloning junctions and coding sequences of all plasmids were verified to ensure the absence of any unwanted mutations . Yeast strains were grown using standard procedures and media . For 2YH assays , the pGBT-CterSmB plasmid and the appropriate pACT2 plasmids were transformed into the CG1945 strain ( Fromont-Racine et al . , 1997 ) . Transformants were selected on double selectable media ( -Leu -Trp ) and further grown in minimal -Trp -Leu liquid medium . Growth of yeast was measured by spotting serial dilutions of liquid cultures on -Leu -Trp -His plates which enables selection of interacting partners . The atomic coordinates for the NMR ensembles of the RBM5 OCRE domain and the complex with the SmN ( 219–229 ) are deposited in the Protein Data Bank under accession numbers 5MFY and 5MF9 , respectively . The chemical shift assignments have been deposited in the Biological Magnetic Resonance Data Bank under accession numbers 34068 and 34067 .	The information required to produce proteins is encoded within genes . In the first step of creating a protein , its gene is “transcribed” to form a pre-messenger RNA molecule ( called pre-mRNA for short ) . Both the gene and the pre-mRNA contain regions called exons that code for protein , and regions called introns that do not . The pre-mRNA therefore undergoes a process called splicing to remove the introns and join the exons together into a final mRNA molecule that is “translated” to make the protein . Many pre-mRNAs can be spliced in several different ways to include different combinations of exons in the final mRNA molecule . This process of “alternative splicing” allows different versions of a protein to be produced from the same gene . Changes that alter the pattern of alternative splicing in a cell affect various cellular and developmental processes and have been linked to diseases such as cancer . The pre-mRNA transcribed from a gene called FAS can be alternatively spliced so that it either does or does not contain an exon that enables the protein to embed itself in the cell membrane . The protein produced from mRNA that includes this exon generates a cell response that leads to cell death . By contrast , protein produced from mRNA that lacks this exon is released from cells and promotes their survival . A splicing factor called RBM5 promotes the removal of this exon from FAS pre-mRNA . RBM5 binds to some of the proteins that make up the molecular machine that splices pre-mRNA molecules . Mourão , Bonnal , Soni , Warner et al . have now used a technique called nuclear magnetic resonance spectroscopy to solve the three-dimensional structure formed when RBM5 binds to one of these proteins , called SmN . Further experiments introduced specific mutations to the proteins to investigate their effects in human cells . This revealed that mutations that impaired the association between RBM5 and SmN compromised the activity of RBM5 to regulate the alternative splicing of FAS pre-mRNA molecules . Future research could examine how RBM5 associates with pre-mRNAs and other components of the splicing machinery , and investigate whether proteins that are closely related to RBM5 act in similar ways .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods", "Accession", "codes" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Structural basis for the recognition of spliceosomal SmN/B/B’ proteins by the RBM5 OCRE domain in splicing regulation
Following learning , increased coupling between spindle oscillations in the medial prefrontal cortex ( mPFC ) and ripple oscillations in the hippocampus is thought to underlie memory consolidation . However , whether learning-induced increases in ripple-spindle coupling are necessary for successful memory consolidation has not been tested directly . In order to decouple ripple-spindle oscillations , here we chemogenetically inhibited parvalbumin-positive ( PV+ ) interneurons , since their activity is important for regulating the timing of spiking activity during oscillations . We found that contextual fear conditioning increased ripple-spindle coupling in mice . However , inhibition of PV+ cells in either CA1 or mPFC eliminated this learning-induced increase in ripple-spindle coupling without affecting ripple or spindle incidence . Consistent with the hypothesized importance of ripple-spindle coupling in memory consolidation , post-training inhibition of PV+ cells disrupted contextual fear memory consolidation . These results indicate that successful memory consolidation requires coherent hippocampal-neocortical communication mediated by PV+ cells . Rhythmic oscillations that occur during sleep and periods of quiet wakefulness are thought to be important for memory consolidation ( Diekelmann and Born , 2010 ) . Specifically , during periods of rest , hippocampal sharp-wave ripples , a form of high frequency network oscillation ( 100–250 Hz ) , are observed in temporal proximity to prefrontal cortical oscillations called spindles ( 12–15 Hz ) ( Siapas and Wilson , 1998 ) . This temporal correlation , referred to as ripple-spindle coupling , is thought to support communication between the hippocampus and prefrontal cortex required for memory consolidation ( Buzsáki , 1989 , 1996; Clemens et al . , 2011; Dudai et al . , 2015; Frankland and Bontempi , 2005; Girardeau and Zugaro , 2011; Igarashi , 2015; Peyrache et al . , 2009; Schwindel and McNaughton , 2011; Siapas and Wilson , 1998; Sirota et al . , 2003; Staresina et al . , 2015; Wierzynski et al . , 2009; Wilson and McNaughton , 1994 ) . Consistent with this hypothesis , cortical electrical stimulation both enhances ripple-spindle coupling and improves performance on an object-location task ( Maingret et al . , 2016 ) . However , whether increased ripple-spindle coupling following learning is necessary for memory consolidation is unknown . Furthermore , the specific cell types that underlie this phenomenon have not yet been identified . In the hippocampus , parvalbumin-positive ( PV+ ) interneurons play a key role in regulating temporal correlations in activity . More specifically , in the CA1 region of the hippocampus , PV+ cells are not required for the generation of ripple oscillations , but appear to be important for the timing of ripples and the synchronization of spiking during ripples . PV+ cells exhibit phase-locked firing with ripples ( Klausberger et al . , 2003 ) , and optogenetic inhibition of CA1 PV+ cells disrupts this phase-locking ( Gan et al . , 2017 ) and the coherence of spiking during ripples in CA1 ( Stark et al . , 2014 ) , without impacting the probability of ripple occurrence ( Gan et al . , 2017 ) . Less is known about the role of PV+ cells in regulating temporal correlations during oscillations in the mPFC . But , as with ripples in CA1 , PV+ cell activity is phase-locked to spindles in the mPFC ( Averkin et al . , 2016; Hartwich et al . , 2009; Peyrache et al . , 2011 ) , suggesting a similar role of PV+ cells in promoting coherent cortical population activity . The promotion of temporal coherence by PV+ cells during ripples and spindles matches previous findings showing that PV+ basket cells can act as a ‘clocking mechanism’ in circuits to ensure specific cell populations fire at appropriate times ( Freund and Katona , 2007 ) . Given the importance of spike-synchrony for communication between circuits ( Wang et al . , 2010 ) , such mechanisms may be critical for inter-regional communication events such as increased ripple-spindle coupling following learning . This raises the possibility that increased ripple-spindle coupling depends on the activity of PV+ cells . If so , then inhibition of PV+ cell activity in either CA1 or mPFC should perturb inter-regional communication by altering ripple and spindle coherence . To test the hypotheses that ( 1 ) PV+ cells mediate increases in ripple-spindle coupling following learning , and ( 2 ) that this increase in coupling is necessary for memory consolidation , we trained mice using contextual fear conditioning . This form of learning engages plastic processes in the hippocampus , including CA1 ( Johansen et al . , 2011; Maren et al . , 2013 ) , and the mPFC , including the anterior cingulate cortex ( ACC ) ( Vetere et al . , 2011; Zhao et al . , 2005 ) . We used PV+ cell-specific Cre driver mice to express chemogenetic constructs allowing us to selectively inhibit PV+ cells in the ACC or CA1 following training . To investigate the role of PV+ cells in promoting increased ripple-spindle coupling , we performed in vivo electrophysiological recordings in mice post-training . As expected , we observed an increase in the probability of ripple-spindle coupling following contextual fear conditioning . Notably , post-training inhibition of PV+ cell activity in the ACC or CA1 did not alter ripple or spindle incidence , but eliminated the learning-induced increase in ripple-spindle coupling . Consistent with this finding , inhibition of PV+ cell activity in either ACC or CA1 also impaired contextual fear memory consolidation . These data indicate that PV+ cells play an important role in enhancing hippocampal-neocortical dialogue following learning , and that this communication is important for memory consolidation . To target PV+ interneurons in the ACC or CA1 , we micro-infused an adeno-associated virus ( AAV ) that expresses the inhibitory Designer Receptor Exclusively Activated by Designer Drugs ( DREADD ) hM4Di with a fluorescent reporter ( mCherry ) in a Cre-recombinase-dependent manner ( AAV-DIO-hM4Di-mCherry ) in mice expressing Cre-recombinase only in PV+ cells ( PV-Cre mice ) ( Armbruster et al . , 2007; Hippenmeyer et al . , 2005; Sohal et al . , 2009 ) . Four weeks following surgery , numerous mCherry+/PV+ interneurons were observed in the ACC or CA1 , respectively ( Figure 1a; Figure 1—figure supplement 1a; Figure 1—figure supplement 2 ) . Over 85% of endogenous PV+ cells were mCherry+ , reflecting efficient infection rates ( Figure 1b , n = 10 ) . Moreover , >93% of mCherry+ cells expressed PV , indicating that infection was limited to the target cell type ( Figure 1c , n = 10 ) ( Sohal et al . , 2009 ) . DREADDs are activated by the synthetic ligand , clozapine-N-oxide ( CNO ) . To verify that CNO-induced activation of hM4Di suppresses PV+ interneuron activity , we used whole-cell patch clamp to record from ACC slices from PV-Cre mice infected with the DREADD viral vector , AAV-DIO-hM4Di-mCherry . To further control for any off-target effects of CNO , or any effects caused by the metabolic conversion of CNO to clozapine ( Gomez et al . , 2017 ) , we also performed the same experiments using the control vector , AAV-DIO-mCherry ( Figure 1d; hM4Di-mCherry+ n = 12 , hM4Di-mCherry-n=10 , mCherry+ n = 13 , mixed-model permutation test , 1000 permutations , [hM4Di-mCherry+ versus hM4Di-mCherry- versus mCherry+]: p=0 . 001 ) . mCherry+ cells from both hM4Di- and control vector-infused mice exhibited much higher spiking rates than mCherry− cells across all current levels tested prior to CNO application , verifying that infection was limited to fast-spiking PV+ interneurons ( Klausberger et al . , 2003 ) . CNO induced hyperpolarization of hM4Di-infected PV+ cells , as bath application of CNO decreased firing rates of hM4Di-mCherry+ , but not mCherry− , or mCherry+ cells in mice micro-infused with the control vector ( Figure 1e; mixed-model permutation test , 1000 permutations , [hM4Di-mCherry+ versus hM4Di-mCherry- versus mCherry+] x [pre-CNO versus post-CNO]: p=0 . 001; individual cell firing rates pre- and post-CNO are shown in Figure 1—figure supplement 3 ) . Furthermore , CNO decreased the input resistance of hM4Di-mCherry+ cells only ( Figure 1f; −80 pA current injection , two-way ANOVA , [hM4Di-mCherry+ versus hM4Di-mCherry- versus mCherry+] x [pre-CNO versus post-CNO]: F32 , 1 = 13 . 14 , p=6 . 8×10−5 , post hoc paired t-test with Bonferroni correction hM4Di-mCherry+ [pre-CNO versus post-CNO] , t11 = 4 . 9 , p=0 . 001 , hM4Di-mCherry- [pre-CNO versus post-CNO] , t9 = −2 . 3 , p=0 . 12 , mCherry+ [pre-CNO versus post-CNO] , t12 = 0 . 67 , p=1 . 0 ) , consistent with the interpretation that activation of hM4Di opens inwardly-rectifying K+ channels . There were no changes in the excitability of mCherry- cells following bath application of CNO . This is likely because pyramidal cells in ex vivo slices do not receive inhibitory input from PV+ cells at baseline , and therefore inhibiting PV+ cells with bath application of CNO has no further effect on pyramidal cell excitability . These experiments also demonstrate that the effect of our manipulation ( i . e . , CNO-mediated inhibition ) is specific for hM4Di+ cells . Ripple-spindle coupling was previously found to increase following training in an odor-reward task ( Mölle et al . , 2009 ) . Here , we tested whether coupling is similarly increased following training in an aversively-motivated task , contextual fear conditioning ( Kim and Fanselow , 1992 ) . We micro-infused the AAV-DIO-hM4Di-mCherry vector in either the ACC or CA1 of PV-Cre mice , and recorded local field potentials ( LFPs ) in both regions to simultaneously detect spindles and ripples ( Figure 1—figure supplement 1b ) . Mice were trained in contextual fear conditioning and immediately following training administered either CNO or vehicle . ACC and CA1 activity was recorded both pre-training ( one day before training ) and post-training ( Figure 2a ) . Because ripple-spindle coupling is observed most commonly during sleep , we measured ripples ( 100–250 Hz ) and spindles ( 12–15 Hz ) during non-REM ( NREM ) periods in the pre- and post-training recording sessions using previously established criteria ( Boyce et al . , 2016; Klausberger et al . , 2003; Phillips et al . , 2012 ) ( Figure 2b ) . Inhibiting PV+ cells in either the ACC or CA1 with CNO did not alter the incidence of ripples ( Figure 2c; Virus-ACC: n = 8 per group; two-way repeated measures ANOVA pre-training versus post-training x Vehicle ( Veh ) versus CNO; pre-training versus post-training F1 , 14 = 1 . 77 , p=0 . 20; Veh versus CNO F1 , 14 = 0 . 0007 , p=0 . 98; interaction F1 , 14 = 2 . 91 , p=0 . 11; Virus-CA1: n = 8 per group; pre-training versus post-training F1 , 14 = 1 . 317 , p=0 . 27; Veh versus CNO F1 , 14 = 3 . 63 , p=0 . 077; interaction F1 , 14 = 0 . 10 , p=0 . 76 ) , consistent with previous reports using genetic manipulation of PV+ cells ( Gan et al . , 2017; Rácz et al . , 2009 ) . This finding contrasts with a previous study in which inhibiting CA3 PV+ cells disrupted ripple generation ( Schlingloff et al . , 2014 ) , and suggests that PV+ cells may play region-specific roles in modulating ripple oscillations . CNO-mediated inhibition of PV+ cells in either the ACC or CA1 did not alter the incidence of spindles ( Figure 2d; Virus-ACC: n = 8 per group; pre-training versus post-training F1 , 14 = 1 . 48 , p=0 . 24; Veh versus CNO F1 , 14 = 2 . 25 , p=0 . 16; interaction F1 , 14 = 3 . 54 , p=0 . 081; Virus-CA1: n = 8 per group; pre-training versus post-training F1 , 14 = 0 . 039 , p=0 . 85; Veh versus CNO F1 , 14 = 0 . 002 , p=0 . 96; interaction F1 , 14 = 2 . 74 , p=0 . 12 ) . Furthermore , CNO did not affect ripple or spindle amplitude ( Figure 2—figure supplement 1a–b ) , induce seizure-like activity ( i . e . , high frequency oscillations ) ( Figure 2—figure supplement 1c–d ) , nor alter sleep architecture ( total NREM , NREM epoch duration ) ( Figure 2—figure supplement 1e–f ) . Having established that CNO-induced inhibition of PV+ cells does not alter ripple or spindle incidence , we next asked whether inhibition of PV+ cells affects the co-incidence of these two oscillations . We computed the cross-correlation between ripple and spindle amplitudes and observed a conditioning-dependent increase in ripple-spindle coupling in vehicle-treated mice . CNO-induced inhibition of PV+ cells post-training eliminated the conditioning-dependent increase in coupling ( Figure 3; Figure 3b: ACC: top; n = 8 per group; pre-training versus post-training F1 , 14 = 2 . 88 , p=0 . 11; Veh versus CNO F1 , 14 = 0 . 15 , p=0 . 70; interaction F1 , 14 = 6 . 68 , p=0 . 022; post hoc Bonferroni’s test , Veh pre-training versus Veh post-training p=0 . 018 , CNO pre-training versus CNO post-training p>0 . 999; CA1: bottom; n = 8 per group; pre-training versus post-training F1 , 14 = 0 . 46 , p=0 . 51; Veh versus CNO F1 , 14 = 0 . 09 , p=0 . 77; interaction F1 , 14 = 8 . 42 , p=0 . 012; post hoc Bonferroni’s test , Veh pre-training versus Veh post-training p=0 . 048 , CNO pre-training versus CNO post-training p=0 . 28; Figure 3c: ACC: top; n = 8 per group; Welch’s t-test t9 . 24 = 2 . 46 , p=0 . 035; Veh versus one one-sample t-test t7 = 2 . 59 , p=0 . 036; CNO versus one one-sample t-test t7 = 0 . 17 , p=0 . 87; CA1: bottom; Pre-training-normalized peak correlation coefficients , n = 8 per group; Mann-Whitney p=0 . 015; Veh versus one one-sample Wilcoxon signed rank test , p=0 . 008; CNO versus one one-sample Wilcoxon signed rank test , p=0 . 31 ) . An identical pattern was observed using other measures of coupling ( cross-correlation of ripple and spindle events [Figure 2—figure supplement 1g–h] and ripple-spindle joint occurrence rate [Figure 2—figure supplement 1i] ) . The peak levels of ripple-spindle coupling , during both Pre- and Post-training , were significantly higher than chance in all ACC- and CA1-infused mice ( an example is shown in Figure 3—figure supplement 1a ) . This suggests that the baseline coupling still likely reflected a significant , continuous communication between ACC and CA1 , but this level was dynamically modulated by fear learning . Importantly , CNO treatment had no effect on this conditioning-dependent increase in ripple-spindle coupling in mice micro-infused with the control vector ( AAV-DIO-mCherry ) into the ACC , indicating that the combination of hM4Di and CNO administration was necessary for the observed effects in vivo ( Figure 3—figure supplement 1b ) . Our findings that post-conditioning inhibition of PV+ cells in either the ACC or CA1 eliminated ripple-spindle coupling indicates that intact PV+ cell activity in both regions is necessary for coordinating the enhanced hippocampal-neocortical communication following learning . We additionally examined the relationship between ripples and ACC delta oscillations since ripples are also coupled to delta oscillations ( Sirota et al . , 2003 ) , and enhancement of cortical delta oscillations is associated with improved memory ( Marshall et al . , 2006 ) . Similar to the effects of inhibiting PV+ cells on disrupting ripple-spindle coupling , we observed that the post-conditioning increase in coupling between ripple and ACC delta oscillations was eliminated by inhibition of PV+ cells in either the ACC or CA1 ( Figure 3—figure supplement 1c–d ) . Importantly , inhibiting PV+ cells did not affect the time lag between baseline ripple and spindle , or between ripple and delta , peak correlation ( Figure 3—figure supplement 1e–f ) . Thus , inhibition of PV+ cells prevents learning-induced increases in the probability of coupling of hippocampal-neocortical oscillations , but not the baseline interactions . If increased ripple-spindle coupling is essential for memory consolidation ( Igarashi , 2015 ) , then post-training inhibition of PV+ interneurons should impair memory consolidation . We first assessed whether PV+ interneurons were activated following learning . Analysis of the activity-regulated gene , Fos , shows that following fear conditioning , PV+ cell activity was elevated in both CA1 and ACC ( compared to untrained control mice ) , indicating that this population of cells is strongly activated by learning ( Figure 4—figure supplement 1a ) . These results are consistent with previous studies showing strong activation of inhibitory interneurons following learning ( Pi et al . , 2013; Sparta et al . , 2014 ) , and , specifically , PV+ cells following fear conditioning ( Donato et al . , 2013; Restivo et al . , 2015; Ruediger et al . , 2011 ) . To directly assess whether intact PV+ cell activity in the CA1 or ACC is required for memory consolidation , we trained mice in contextual fear conditioning and then administered CNO or vehicle for 4 weeks . Mice were then tested drug-free . Inhibition of PV+ cells in the ACC impaired consolidation of contextual fear memory , with CNO-treated mice freezing less compared to vehicle-treated controls . Similarly , chronic , post-training suppression of PV+ cells in CA1 impaired consolidation of contextual fear memory ( Figure 4a; ACC: Veh n = 6 , CNO n = 8 , Mann-Whitney test p=0 . 028; CA1: Veh n = 7 , CNO n = 9 , t-test t14 = 3 . 42 , p=0 . 004 ) . Inhibiting PV+ interneurons in either region immediately prior to testing did not affect freezing during test ( Figure 4b; ACC: Veh n = 9 , CNO n = 8 , t-test t15 = 0 . 44 , p=0 . 66; CA1: Veh n = 6 , CNO n = 5 , t-test t9 = 0 . 28 , p=0 . 78 ) , indicating that PV+ cell activity is not necessary for memory retrieval . Using ex vivo patch-clamp experiments , we verified that chronic ( month-long ) CNO treatment inhibited hM4Di-infected neurons without altering baseline neuronal excitability ( Figure 4c–e; Figure 4d: mCherry+ Veh n=14 , CNO n = 20 , mCherry- Veh n = 14 , CNO n = 15 , mixed-model permutation test , 1000 permutations , CNO versus Veh: p=0 . 77; Figure 4e: mCherry+ Veh n=14 , CNO n = 20 , mCherry- Veh n = 14 , CNO n = 15 , voltage clamp , mixed-model permutation test , 1000 permutations , CNO versus Veh: p=0 . 88 ) . Furthermore , analysis of the activity-regulated gene , Fos , confirmed that CNO water treatment reduced retrieval-induced activation of hM4Di-infected neurons in both CA1 and ACC ( Figure 4f–h , Figure 4—figure supplement 1c; Figure 4g: Veh n = 4 , CNO n = 5 , t-test t7 = 1 . 37 , p=0 . 21; Figure 4h: Veh n = 4 , CNO n = 5 , t-test t7 = 2 . 54 , p=0 . 039 ) . The ACC also modulates pain affect ( Bliss et al . , 2016 ) . Therefore , it is possible that our PV manipulations in the ACC impact pain processing post-learning , rather than disrupting memory consolidation . To address this potential confound , we trained mice in a cued fear conditioning paradigm in which a tone was paired with a shock . This form of fear learning does not depend on either the CA1 or ACC ( Fanselow , 2010; Rajasethupathy et al . , 2015 ) . In contrast to the effects observed in contextual fear conditioning , chronic CNO-induced suppression of ACC PV+ cell activity did not affect consolidation of tone fear conditioning ( Figure 4—figure supplement 2d ) , suggesting that post-shock pain processing was not altered . Moreover , similar chronic CNO-induced suppression of ACC PV+ cell activity did not alter general exploratory or anxiety-related behaviours ( Figure 4—figure supplement 2a–b ) . In these experiments , the activity of PV+ cells was chemogenetically suppressed for one month following training . However , in recording experiments , we detected increases in ripple-spindle coupling immediately following contextual fear conditioning , and not 7 or 14 days later ( Figure 3—figure supplement 2 ) . This suggests that increased ripple-spindle coupling may transiently contribute to memory consolidation , and , furthermore , that shorter periods of PV suppression might be sufficient to impair consolidation . To test this idea , mice were fear conditioned and tested 28 days later , as above . However , CNO was administered either during the first or last post-training week to temporally restrict inhibition of PV+ interneurons ( Figure 5a–b; Figure 5a: ACC: Veh n = 7 , CNO n = 6 , Welch’s t-test t7 . 48 = 2 . 51 , p=0 . 038; CA1: Veh n = 9 , CNO n = 9 , t-test t16 = 2 . 87 , p=0 . 011; Figure 5b: ACC: Veh n = 7 , CNO n = 7 , Mann-Whitney test p=0 . 90; CA1: Veh n = 8 , CNO n = 9 , t-test t15 = 0 . 62 , p=0 . 55 ) . CNO-induced suppression of PV+ cell activity in the ACC in the first , but not last , post-training week impaired consolidation of contextual fear memory . Similarly , post-training suppression of PV+ interneuron activity in CA1 during the first , but not last , post-training week impaired consolidation of contextual fear memory . Suppression of PV+ interneuron activity in either the ACC or CA1 produced a similar pattern of results using a weaker conditioning protocol ( Figure 5—figure supplement 1 ) . More importantly , we observed the same pattern of behavioral results in mice that underwent in vivo recording ( Figure 5c; ACC: Veh n = 8 , CNO n = 8 , Mann-Whitney test p=0 . 05; CA1: Veh n = 8 , CNO n = 8 , t-test t14 = 2 . 64 , p=0 . 020 ) . Furthermore , analysis of the activity-regulated gene , Fos , confirmed that activation of hM4Di-infected neurons was reduced by week-long CNO treatment in both CA1 and ACC ( Figure 4—figure supplement 1b ) . The absence of effects on retrieval ( Figure 4b ) , as well as at time points remote to training ( Figure 5b ) , suggests that PV+ interneuron suppression in the ACC or CA1 does not simply interfere with the ability of mice to freeze . Indeed , chronic pre-training suppression of PV+ interneurons does not alter subsequent learning or retrieval ( Figure 4—figure supplement 2c ) . Together , these results indicate that the increase in ripple-spindle coupling within a relatively narrow time window following training is required for successful memory consolidation . To further narrow down the window in which PV+ cell activity in ACC and CA1 contributes to memory consolidation , we conducted an additional set of experiments . In these experiments , mice were fear conditioned and tested 1 day later . Immediately following training , mice received a single injection of CNO or Veh ( Figure 6a ) . Inhibition of PV+ cells in CA1 impaired consolidation of contextual fear memory ( Veh n = 7 , CNO n = 10 , t-test t15 = 2 . 75 , p=0 . 015 ) , consistent with a recent report ( Ognjanovski et al . , 2017 ) . Similarly , inhibition of PV+ cells in ACC impaired consolidation of contextual fear memory ( Veh n = 12 , CNO n = 16 , t-test t26 = 3 . 10 , p=0 . 0046 ) . In contrast , inhibiting PV+ interneurons in either region immediately prior to testing did not affect freezing during test ( Figure 6b; ACC: Veh n = 7 , CNO n = 12 , t-test t17 = 0 . 71 , p=0 . 48; CA1: Veh n = 6 , CNO n = 6 , t-test t10 = 0 . 74 , p=0 . 94 ) , indicating that PV+ cell activity is not necessary for memory retrieval 24 hr following training . Ripple-spindle coupling has been proposed to facilitate memory consolidation , and is increased following odor-reward learning ( Mölle et al . , 2009 ) . Furthermore , promoting ripple-spindle coupling enhances consolidation of an object-location memory ( Maingret et al . , 2016 ) . However , previous studies did not directly test whether this form of hippocampal-neocortical communication is necessary for successful memory consolidation , nor identify the cellular bases for mediating learning-dependent changes in ripple-spindle coupling . Here we found that contextual fear learning increased ripple-spindle coupling , and , furthermore , that chemogenetic inhibition of PV+ cells in the ACC or CA1 both eliminated this learning-induced increase in ripple-spindle coupling and impaired memory consolidation . Both mono- and multi-synaptic pathways between ACC and CA1 can support bidirectional communication between these two regions via ripple-spindle coupling . We observed an average lag between ripple and spindle peak amplitude of ~70 ms , consistent with ranges previously reported ( 40–244 ms; e . g . , [Peyrache et al . , 2009; Phillips et al . , 2012; Siapas and Wilson , 1998; Wang and Ikemoto , 2016; Wierzynski et al . , 2009] ) . This suggests that these two events are more likely coordinated via multiple synapses . Although the exact mechanism is unclear , there are several possibilities for bidirectional modulations . For example , ACC can modulate dorsal CA1 activity via thalamic regions , including nucleus reuniens ( e . g . , [Varela et al . , 2014; Xu and Südhof , 2013] ) . Interestingly , mPFC neurons that project to the nucleus reuniens preferentially synapse onto hippocampus-projecting reuniens cells ( Vertes et al . , 2007 ) . In addition , a subset of neurons in the nucleus reuniens project to inhibitory interneurons in CA1 ( Dolleman-Van der Weel and Witter , 2000 ) . Furthermore , a group of nucleus reuniens cells also has collaterals in both CA1 and mPFC , potentially coordinating activities between the two regions ( Varela et al . , 2014 ) . CA1 can , in turn , modulate ACC via subiculum ( Varela et al . , 2014 ) , ventral hippocampus , retrosplenial cortex ( e . g . , [Cenquizca and Swanson , 2007] ) , infralimbic cortex [Swanson , 1981] , and/or prelimbic cortex [Thierry et al . , 2000] ) . PV+ cells likely coordinate ripple-spindle coupling by facilitating synchronized spiking during ripples and spindles . In CA1 and mPFC , PV+ cell activity is phase-locked to ripples ( Klausberger et al . , 2003 ) and spindles ( Averkin et al . , 2016; Hartwich et al . , 2009; Peyrache et al . , 2011 ) , respectively . In CA1 , inhibition of PV+ cells disrupts phase-locked firing of PV+ cells to ripples , and ripple coherence ( Gan et al . , 2017; Stark et al . , 2014 ) . This is consistent with the proposed role of PV+ cells acting as a ‘clocking mechanism’ in circuits , ensuring that specific cell populations fire at appropriate times ( Freund and Katona , 2007 ) . Inhibition of PV+ cells in the ACC or CA1 did not affect baseline probability of ripple-spindle coupling , but prevented learning-induced increases in ripple-spindle coupling . In the absence of learning , PV+ cells show moderate levels of activation . However , following learning we observed strong activation of PV+ cells in both regions , as well as a corresponding increase in the probability of ripple-spindle coupling . Importantly , CNO-mediated inhibition did not eliminate PV+ cell activity , but reduced it to pre-learning or home cage levels ( as shown in our ex vivo and in vivo experiments ) . Therefore , we would expect that chemogenetic inhibition of PV+ cells following learning should not eliminate ripple-spindle coupling altogether , but instead , reduce it to the levels that occur in the absence of training , which is what we observed . Consistent with this idea , fear conditioning increases hippocampal network stability ( Donato et al . , 2013 ) , and chemogenetic inhibition of PV+ cells in CA1 blocks this learning-induced increase ( Ognjanovski et al . , 2017 ) . Notably , when PV+ activity levels are driven below baseline levels via other techniques , there is an associated reduction in the probability of ripple-spindle coupling , even in the absence of learning ( Phillips et al . , 2012 ) . This suggests that the overall levels of PV+ cell activity regulate the probability of ripple-spindle coupling . Accordingly , strong activation of PV+ cells during learning ( Donato et al . , 2013; Restivo et al . , 2015; Ruediger et al . , 2011 ) may increase coherence both within and across brain regions . Synchronous activity , such as ripple-spindle coupling , is particularly effective at driving inter-regional communication and plasticity required for consolidation ( Fell and Axmacher , 2011; Igarashi , 2015; Wang et al . , 2010 ) . Therefore , inhibition of PV+ cell activity in either the CA1 or the mPFC likely prevented this learning-induced increase in coupling , by perturbing intra-regional synchrony of action potentials during ripples and spindles , and consequently , the coordination of inter-regional communication . In contrast , inhibition of PV+ cells in either ACC or CA1 immediately prior to testing did not affect recall ( at 1 or 28 days post-training ) . Since overall activity in ACC and CA1 are known to be important for retrieval of contextual fear memories , these observations suggest that the activity of non-PV+ cells was not affected by our PV manipulations . Consistent with this , the c-Fos levels in mCherry- cells in these regions following CNO treatment were not altered . Ripples are associated with simultaneous memory trace reactivation in the hippocampus and neocortex ( Peyrache et al . , 2011; Peyrache et al . , 2009; Schwindel and McNaughton , 2011 ) . Therefore , impaired ripple coherence following CA1 inhibition of PV+ cells ( Stark et al . , 2014 ) likely reduced coordinated hippocampal output to the neocortex , and consequently decreased the probability of simultaneous memory trace reactivation in the neocortex . In the mPFC , memory trace reactivation is often followed by occurrence of spindles , and increased activation of local PV+ cells ( Peyrache et al . , 2011 ) . This is thought to favor the consolidation of recently modified synapses during memory reactivation , while suppressing interfering inputs to the neocortex . Since ACC inhibition of PV+ cells was sufficient to disrupt ripple-spindle coupling ( without changing the overall incidence of spindles or ripples ) , this suggests that our manipulation interfered with the timely occurrence of spindles following ripples/memory reactivation . Therefore , inhibition of ACC PV+ cells likely prevented the strengthening of synapses in the neocortex that is necessary for memory consolidation . Our findings provide support for the idea that PV+ cells are necessary for learning-associated increases in ripple-spindle coupling probability , and consequently , successful memory consolidation . Ripple-spindle coupling is also increased following odor-reward learning ( Mölle et al . , 2009 ) , and therefore it seems plausible that the role of PV+ interneurons is similar during consolidation of appetitively-motivated ( as well as aversively-motivated ) tasks . There are , however , alternative possibilities for why our PV manipulation resulted in consolidation deficit . For example , it is possible that the effects of inhibition of PV+ cells outside of the sleep period ( i . e . , the ripple-spindle coupling window ) could contribute to the consolidation deficits that we observed . Moreover , inhibition of PV+ cells may have increased lateral disinhibition and disrupted local circuit activity , in addition to disrupting global communication ( i . e . , ripple-spindle coupling ) . While we cannot definitively exclude this possibility , three pieces of evidence suggest that the observed consolidation deficits are mediated primarily by disruption of global communication . First , we found that inhibition of PV+ cells in either ACC or CA1 immediately following training impaired memory tested 24 hr later . Activity in CA1 , but not ACC , is critical for expression of contextual fear memory at this time point ( Frankland and Bontempi , 2005 ) . Therefore , if our manipulation of PV+ cells activity only affected local activity , we would not predict the memory deficits following inhibition of ACC PV+ cells . Second , inhibition of PV+ cells had no effect on retrieval of contextual fear memories , tested either 24 hr or 28 days post-training , suggesting again that the overall local activity is relatively undisturbed . This reinforces the idea that our PV manipulation is distinct from other manipulations that more profoundly impact pyramidal cell activity in these regions . Third , consistent with this , we did not observe increased activation in mCherry- cells in targeted regions following inhibition of PV+ interneurons . Therefore , the more plausible explanation is that the observed deficits are caused by disrupted global synchrony ( i . e . , ripple-spindle coupling ) . We used a chemogenetic approach to manipulate PV+ cell activity in ACC and CA1 . One advantage of this approach is that chemogenetic-induced inhibition does not completely eliminate the activity of infected cells ( e . g . , compared to some forms of optogenetic silencing ) , and therefore is less likely to produce large-scale changes in overall circuit activity . Consistent with this , we did not observe a detectable increase in activation of mCherry- cells in either in vivo or ex vivo experiments . This may also explain why our PV manipulation did not produce broad changes in local field potential at theta ( Amilhon et al . , 2015 ) or gamma ( Sohal et al . , 2009 ) frequencies , as previously observed using optogenetic silencing of PV+ cells . The absence of changes in the activity of non-infected neurons may also be related to the fact that PV+ cells represent only a subpopulation of GABAergic interneurons in both ACC and CA1 ( Bezaire and Soltesz , 2013; Rudy et al . , 2011; Tremblay et al . , 2016 ) , and therefore it is plausible that non-infected cells in the circuit can still maintain homeostasis of spiking activity when the activity of PV+ cells is suppressed . Moreover , reducing PV-mediated inhibition could lead to disinhibition of other inhibitory cell types ( e . g . , [Lovett-Barron et al . , 2012] ) , thereby producing little overall change in excitation or inhibition . In conclusion , here we showed that contextual fear learning increased the probability of ripple-spindle coupling . Inhibition of PV+ cells in either ACC or CA1 eliminated this learning-induced enhancement and impaired fear memory consolidation . These data indicate that temporally correlated activities across brain regions are necessary for contextual fear memory consolidation , and our study provides evidence for an integral role for PV+ cells in this process . All procedures were approved by the Canadian Council for Animal Care ( CCAC ) and the Animal Care Committees at the Hospital for Sick Children and the University of Toronto . Experiments were conducted on 8–12 week old male and female PV-Cre knock-in transgenic mice where Cre-recombinase was targeted to the Pvalb locus , without disrupting endogenous PV expression ( RRID:IMSR_JAX:017320 ) . The PV-Cre mice were originally generated by Silvia Arber ( Hippenmeyer et al . , 2005 ) , and obtained from Jackson Lab . The mice were bred as homozygotes , weaned at 21 days , and group housed with 2–5 mice per cage in a temperature-controlled room with 12 hr light/dark cycle ( light on during the day ) . All experiments were performed between 8 am and 12 pm . Mice were given ad libitum access to food and water . Mice were randomly assigned to experimental groups . The experimenter was aware of the experimental group assignment , as the same experimenter conducted the training and testing of all mice , but was blinded during behavioral assessment and cell counting experiments . Mice were excluded from analysis based on post-experimental histology: only mice with robust expression of the viral vector ( hM4Di-mCherry ) specifically in the targeted region were included . The spread of virus was estimated to be the following: CA1: AP −1 . 2 ~ −2 . 4 mm , ML ±0 . 2 ~ 3 mm , DV −1 . 5 ~ −2 mm; ACC: AP 1 . 2 ~ −0 . 2 mm; ML ±0 . 1 ~ 0 . 8 mm , DV −0 . 7 ~ −2 mm ( Figure 1—figure supplement 2 ) . For the in vivo electrophysiology experiments , only mice with correct electrode placements in both the ACC and CA1 , as well as robust viral vector expression in the targeted region were included . Specifically , only mice where we could reliably detect sharp-wave ripples during the Pre-training recording sessions were included , to ensure that the electrodes were in CA1 cell layer . In rare cases where electrodes deteriorated prior to the completion of all experiments , and hence resulting in high noise background and no viable signals , subsequent recordings were not included in the analysis ( Figure 3—figure supplement 1g . ACC-Veh , 2 mice ) . AAV8-hSyn-DIO-hM4Di-mCherry and AAV8-hSyn-DIO-mCherry viruses were obtained from UNC Vector Core ( Chapel Hill , NC ) . In the DREADD receptor virus , AAV8-hSyn-DIO-hM4Di-mCherry , the double-floxed inverted open reading frame of hM4Di fused to mCherry can be expressed from the human synapsin ( hSyn ) promoter after Cre-mediated recombination . Similarly , in the control viral vector , AAV8-hSyn-DIO-mCherry , the double-floxed inverted open reading frame of the mCherry fluorescence tag can be expressed from the hSyn promoter after Cre-mediated recombination . Four weeks prior to behaviour or electrophysiology experiments , PV-Cre mice were micro-infused bilaterally with one of these viral vectors ( 1 . 5 µl per side , 0 . 1 µl/min ) in the ACC ( +0 . 8 mm AP , ±0 . 3 mm ML , −1 . 7 mm DV , from bregma according to Paxinos and Franklin [2001] ) or CA1 ( −1 . 9 mm AP , ±1 . 3 mm ML , - 1 . 5 mm DV ) . Similar to the previously described protocol ( Richards et al . , 2014 ) , mice were pretreated with atropine sulphate ( 0 . 1 mg/kg , intraperitoneal ) , then anesthetized with chloral hydrate ( 400 mg/kg , intraperitoneal ) . Mice were then placed on a stereotaxic frame , and holes were drilled in the skull at the targeted coordinates . Viral vector was micro-infused at 0 . 1 μl/min via glass pipettes connected to a Hamilton microsyringe with polyethylene tubing . After micro-infusion , the glass pipette was left in the brain for another 5 min to allow sufficient time for the virus to diffuse . We have found that this infusion procedure produces high infection in the targeted region , without significant spread outside the region of interest ( Rashid et al . , 2016; Richards et al . , 2014 ) . Mice were then treated with analgesic ( ketoprofen , 5 mg/kg , subcutaneous ) and 1 ml of 0 . 9% saline ( subcutaneous ) . Clozapine-N-oxide ( CNO , kindly provided by Dr . Bryan Roth , University of North Carolina ) was dissolved in dimethyl sulfoxide ( DMSO ) to produce a 10 mg/ml CNO stock solution . For i . p . injections , CNO stock solution was mixed with 0 . 9% saline , and injected at a dose of 5 mg/kg . The Vehicle ( Veh ) control group received equivalent amount of DMSO solution dissolved in 0 . 9% saline . For administration of CNO in the drinking water , preliminary experiments were first carried out to determine the amount of water a mouse consumes per day ( approximately 3–5 ml of water/day ) . Based on the number of mice per cage , the amount of water required for 7 days was calculated for each cage , and 5 mg/kg of CNO/mouse/day was added to the water . We added sucrose ( 1% ) to the drinking water to encourage CNO consumption . The control group received vehicle in 1% sucrose . For experiments that required more than 7 days of CNO/vehicle water , the water was changed every 7 days . Immunofluorescence staining was conducted as previously described ( Restivo et al . , 2015 ) . Specifically , at the end of behaviour experiments , mice were transcardially perfused with 1x PBS followed by 10% paraformaldehyde . For the c-Fos experiment ( Figure 4f–g , Figure 4—figure supplement 1 ) , mice were perfused 90 min after behaviour test or training . Brains were fixed overnight at 4∘C , and transferred to 30% sucrose solution for 48 hr . Brains were sectioned coronally using a cryostat ( Leica CM1850 ) , and 50 µm sections were obtained for the entire medial prefrontal cortex or hippocampus , for ACC- or CA1-infused animals , respectively . For PV and c-Fos immunostaining , free-floating sections were blocked with PBS containing 2 . 5% bovine serum albumin and 0 . 3% Triton-X for 30 min . Afterwards , sections were incubated in PBS containing mouse monoclonal anti-PV primary antibody ( 1:1000 dilution; Sigma-Aldrich Cat# P3088 RRID:AB_477329 ) and rabbit polyclonal anti-c-Fos primary antibody ( 1:1000 dilution; Santa Cruz Biotechnology Cat# sc-52 RRID:AB_2106783 ) for 48 hr at 4°C . Sections were washed with PBS ( 3 times ) , then incubated with PBS containing goat anti-mouse ALEXA Fluor 488 ( for PV , 1:500 dilution; Thermo Fisher Scientific Cat# A-11001 RRID:AB_2534069 ) and goat anti-rabbit ALEXA Fluor 633 ( for c-Fos , 1:500 dilution , Thermo Fisher , USA Scientific Cat# A-21070 RRID:AB_2535731 ) secondary antibody for 2 hr at room temperature . Sections were washed with PBS , mounted on gel-coated slides , and coverslipped with Vectashield fluorescent mounting medium ( Vector Laboratories ) . Images were obtained using a confocal laser scanning microscope ( LSM 710; Zeiss ) with a 20X objective . For cell counting experiments ( Figures 1 and 4 and Figure 4—figure supplement 1 ) , every second section in either ACC or CA1 was assessed for mCherry+ , PV+ and c-Fos+ cells . Approximately 4–6 sections/mouse were counted and averaged , with 3–6 mice/group . Transduction specificity ( total numbers of PV+ cells total numbers of mCherry+ cells x 100 ) , and efficiency ( total numbers of mCherry+ cells/total numbers of PV+ cells x 100 ) were calculated . To evaluate the effectiveness of CNO in vivo , c-Fos co-localization in mCherry+ cells ( total numbers of c-Fos+ and mCherry+ co-localized cells/total numbers of mCherry+ cells x 100 ) was calculated . To assess the activity in mCherry- cells , c-Fos+ cells that are not co-localized with mCherry+ cells in the region was also counted , and normalized to the area in the same section ( total numbers of c-Fos+ and mCherry- cells/10 , 000 μm2 ) . To evaluate the activity of PV+ cells during learning , c-Fos co-localization in PV+ cells in each region ( total numbers of c-Fos+ and PV+ co-localized cells/total numbers of PV+ cells x 100 ) was calculated . PV-Cre mice were micro-infused with the DREADD receptor virus ( AAV-DIO-hM4Di-mCherry ) or the control vector ( AAV-DIO-mCherry ) in the ACC ( as above ) . Mice were separated into two groups: ( 1 ) acute tests , to assess the excitability of ACC neurons upon direct application of CNO ( Figure 1 ) , or ( 2 ) chronic tests , to assess whether lasting changes arise in the excitability of neurons after 28 days of continuous CNO delivered in drinking water ( Figure 4c–e ) . For the acute group , 4 weeks following viral micro-infusion mice were anesthetized with 1 . 25% tribromoethanol ( Avertin ) and underwent cardiac perfusion with 10 mL of a chilled cutting solution ( containing , in mM: 60 sucrose , 83 NaCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 . 5 KCL , 0 . 5 CaCl2 , 6 MgCl2 , 20 D-glucose , 3 Na-pyruvate , 1 ascorbic acid ) , injected at a rate of approximately 2 mL/min . After perfusion , the brain was quickly removed and cut coronally ( 350 μm thickness ) with a vibratome ( Leica , VT1200S ) in chilled cutting solution in order to obtain live , healthy slices containing the ACC . Slices were transferred to a recovery chamber comprising of a 50:50 mix of warm ( 34°C ) cutting solution and aCSF ( containing , in mM: 125 NaCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 . 5 KCl , 1 . 3 CaCl2 , 1MgCl2 , 20 D-glucose , 3 Na-pyruvate , 1 ascorbic acid ) . Following 40–60 min of incubation , slices were transferred into a different incubation chamber with room temperature aCSF . Within the recording chamber , aCSF was heated to 32°C using an in-line heater ( Warner Instruments , SF-28 ) . Whole-cell current clamp recordings were made using glass pipettes filled with internal solution ( comprising , in m ) : 126 K D-Gluconate , 5 KCl , 10 HEPES , 4 MgATP , 0 . 3 NaGTP , 10 Na-phosphocreatine ) . Glass capillary pipettes were pulled with a flaming brown pipette puller ( Sutter , P-97 ) to tip resistances between 3–8 MΩ . We determined the effects of acute CNO application by patching individual mCherry+ or mCherry− cells and injecting square 500 ms current pulses into the cell ( in 40 pA steps , ranging from −80 pA to 400 pA ) , both before and after CNO application ( washing aCSF containing 10 μM CNO onto the slice for 10 min ) . We calculated the difference in firing rate ( using the positive current injections ) and input resistance ( using the negative current injections ) pre- and post-CNO application . For the chronic group , 4 weeks following viral micro-infusion , mice were given either CNO or vehicle in their drinking water for 28 days . On the 29th day , mice received clean drinking water for 24 hr , to flush out the CNO in their system and allow testing in drug-free conditions . Extraction and incubation procedures followed those above . In addition to the current clamp recordings , voltage clamp recordings were obtained by clamping the voltage for 500 ms in 20 mV steps from −90 mV to +30 mV . To estimate the strength of the active , non-inactivating K+ currents ( which may have been altered by chronic CNO exposure ) we measured the steady state current in the final 400 ms of the voltage step . Four weeks after micro-infusion of hM4Di-mCherry or mCherry virus in the ACC or CA1 in PV-Cre mice , custom-made local field potential ( LFP ) electrodes were implanted in the ACC ( +0 . 8 mm AP , ±0 . 3 mm ML , −1 . 8 mm DV ) and CA1 ( −1 . 9 mm AP , ±1 . 3 mm ML , 1 −1 . 7 mm DV ) . Similar to described above , mice were first anesthetized with 2% isoflurane and placed on a stereotaxic frame . Holes were drilled in the skull at the targeted coordinates , and virus was delivered as described above . Four weeks following viral vector micro-infusion , mice were implanted with LFP electrodes . Mice were anesthetized with 2% isoflurane and mounted onto a stereotaxic frame . Miniature stainless steel screw was placed in the cerebellum for ground , and a stripped stainless steel wire was inserted into the neck muscle for recording electromyogram ( EMG ) activity . Holes were drilled at the targeted coordinates , and custom made Teflon-coated stainless steel LFP electrodes ( A-M Systems , Carlsborg , WA ) bundled in 23–25G stainless steel cannulas were slowly lowered to the ACC ( bipolar electrode with 0 . 3 mm distance between electrodes ) and CA1 ( tripolar electrode with 0 . 3 mm distance between electrodes ) , at the rate of 0 . 1 mm/s . LFP signals are referenced locally within the ACC or CA1 . All wires were soldered to gold pins and inserted into to a plastic cap ( PlasticsOne ) . The electrodes and cap were secured on the skull using dental cement . Mice were given ketoprofen ( 5 mg/kg , subcutaneous ) and 1 ml 0 . 9% saline ( subcutaneous ) for 2 days following surgery . Mice were single-housed following surgery , to prevent potential fighting that could damage the cap . Three days after surgery , mice were habituated to the recording chamber for two days ( 2 hr/day ) . The sound-attenuated chamber was dimly lit , and contained a tall Plexiglass cylinder , inside which mice were placed and allowed to sleep for the duration of the recording . All recording session were carried out during ZT 2–6 , and LFP activities were recorded using the RZ-5 recording system ( Tucker-Davis Technologies ) . Signal was amplified 1000 times , filtered between 1 and 400 Hz , and digitized at 2 kHz . On the second day of habituation , baseline ( pre-training; Figure 2a ) LFP activity was obtained . On the following day , mice were fear conditioned , similar to as described above . Immediately afterward , mice were given CNO ( 5 mg/kg ) or vehicle i . p . , and within 5–10 min , placed into the recording chamber to record the post-conditioning LFP activity ( post-training , Figure 2a and 2 hr ) . We chose this specific delay ( 5–10 min ) , because data from many other groups show that neural activity in chemogenetic-infected cells is altered within 10–60 min following CNO injection ( e . g . , ( Alexander et al . , 2009 ) [Figure 5c]; ( Ryan et al . , 2015 ) [Figure S12] ) . For PV+ cells specifically , a previous study used an identical chemogenetic-based approach to inhibit PV+ cells ( AAV-DIO-hM4Di in PV-Cre mice , same dose of CNO ) ( Kuhlman et al . , 2013 ) . They measured calcium transients following CNO injection , and observed a decrease in PV+ cell activity , beginning 30–60 min following CNO injection . The delay we chose therefore allows us to capture the earliest onset of CNO-mediated effects on LFP activity . Following the post-training recording session , mice were returned to the home cage , and given CNO or vehicle in drinking water for the next 7 days . The first consolidation recording session took place 7 days after fear conditioning ( Con . 1 , Figure 3—figure supplement 1g–h and 2 hr ) . All mice were then placed on clean drinking water for another 7 days , and at the end , the second consolidation recording session took place ( Con . 2 , Figure 3—figure supplements 1g–h and 2 hr ) . Mice were then placed back into the fear training context for 4 min without shock , to examine their fear memory ( Figure 5c ) . At the end of the experiments , mice were anesthetized and electrolytic lesions ( 20 µA for 30 s for each electrode tip ) were performed to verify the locations of electrodes . Mice were then transcardially perfused , and brains were sectioned and imaged to verify the spread of virus , similar to as described above . In addition , cresyl violet staining was performed on every other section in the ACC and dorsal CA1 , to verify electrode locations ( Figure 1—figure supplement 1b ) . All analyses were performed offline using MATLAB ( The MathWorks ) and previously established methods as detailed below . No statistical tests were used to pre-determine sample size , but the sample sizes used are similar to those generally used within the field . Data were tested for normality and variance . If data from neither group were significantly non-normal and if variances are not significantly unequal , data were analyzed using parametric two-way repeated measures ANOVA , or two-sample Student’s unpaired t-test . For comparisons between two groups , if the groups had significantly different variances ( with ɑ = 0 . 05 ) , Welch’s t-test was used . For comparisons to a hypothetical mean of 1 , one-sample t-test was used . Where appropriate , ANOVA was followed by post hoc pairwise comparisons with Bonferroni correction . If data were significantly non-normal ( with ɑ = 0 . 05 ) or variances were significantly unequal , mixed-model permutation test , Kruskal-Wallis test or Mann-Whitney test ( between-group comparisons ) , and Wilcoxon signed-rank test or Friedman test ( within-group comparisons ) were used accordingly . All tests were two-sided . Statistical analyses were performed using R and Graphpad Prism V6 .	Sleep contributes to the strengthening of memories . During non-dreaming sleep , neurons in regions of the brain that form and store memories – such as the hippocampus and prefrontal cortex – fire in rhythmic waves . The neurons in the hippocampus tend to fire during a wave that repeats up to 250 times per second , called sharp-wave ripples . Meanwhile , in the prefrontal cortex , the neurons tend to fire during a lower frequency wave that repeats 12 to 15 times per second , called spindles . During sleep and quiet wakefulness , hippocampal ripples often synchronize with prefrontal spindles; that is , both waves tend to occur at approximately the same time . Many neuroscientists think this allows the brain regions to better communicate with one another , which in turn should help the brain to strengthen memories . Consistent with this possibility , rodents that learn a new task show more synchrony between ripples and spindles afterwards . But no one had actually tested whether this increase in ripple-spindle synchrony does strengthen the rodent’s memory of the task . It was also unclear how the brain achieves such an increase . Xia et al . suspected that this process involved a group of inhibitory brain cells called parvalbumin-positive interneurons . These cells act like timekeepers , and help to synchronize the firing of groups of neurons . Xia et al . now show that training mice to associate an environment with a mild electric shock made it more likely that the animals would show ripple-spindle synchrony . Yet , inhibiting the activity of parvalbumin-positive interneurons in either the hippocampus or prefrontal cortex blocked this effect . It also prevented sleep from strengthening the animals’ memory of the link between the environment and the shock . Patients with Alzheimer’s disease have fewer parvalbumin-positive interneurons . By showing that these neurons help strengthen new memories , these findings may explain why losing them can impair memory . Restoring or replacing interneuron activity could be a promising therapeutic avenue to explore .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Parvalbumin-positive interneurons mediate neocortical-hippocampal interactions that are necessary for memory consolidation
Organizers play important roles during the embryonic development of many animals . The most famous example is the Spemann organizer that sets up embryonic axes in amphibian embryos . In spiders , a group of BMP secreting mesenchymal cells ( the cumulus ) functions as an organizer of the dorsoventral axis . Similar to experiments performed with the Spemann organizer , transplantation of the cumulus is able to induce a secondary axis in spiders . Despite the importance of this structure , it is unknown which factors are needed to activate cumulus specific gene expression . To address this question , we performed a transcriptomic analysis of early embryonic development in the spider Parasteatoda tepidariorum . Through this work , we found that the transcription factor Pt-Ets4 is needed for cumulus integrity , dorsoventral patterning and for the activation of Pt-hunchback and Pt-twist expression . Furthermore , ectopic expression of Pt-Ets4 is sufficient to induce cell delamination and migration by inducing a mesoderm-like cell fate . The self-regulatory capacities of vertebrate embryos were most famously demonstrated by Spemann and Mangold . They found that by grafting the dorsal-lip of an amphibian embryo ( now known as the Spemann Organizer ) to the ventral side of the host gastrula embryo it was possible to induce a secondary body axis ( Spemann and Mangold , 2001; De Robertis , 2009; Anderson and Stern , 2016 ) . Intriguingly , spider embryos also have high self-regulatory capacities , even to the extent that twinning can occur spontaneously ( Napiórkowska et al . , 2016; Oda and Akiyama-Oda , 2008 ) . During spider embryogenesis a group of migratory cells ( the cumulus ) is needed to break the radial symmetry of the early embryo and to induce the dorsoventral body axis ( Oda and Akiyama-Oda , 2008; Akiyama-Oda and Oda , 2003 , 2006; McGregor et al . , 2008; Hilbrant et al . , 2012; Mittmann and Wolff , 2012; Schwager et al . , 2015 ) . Similar to the vertebrate experiments , Holm showed that transplanting cumulus material was able to induce a secondary axis in spider embryos ( Holm , 1952 ) . Modern work has shown that the cumulus signals via BMP signaling ( again , similar to vertebrates ) . The mesenchymal cumulus cells are the source of the BMP receptor ligand Decapentaplegic ( Akiyama-Oda and Oda , 2006 ) . Interfering with the BMP signaling pathway by gene knockdown results in the loss of dorsal tissue identity , which in turn leads to completely radially-symmetric and ventralized embryos ( Akiyama-Oda and Oda , 2006 ) . The cumulus forms in the center of the so-called germ-disc ( the embryonic pole of the embryo ) and migrates underneath the ectoderm towards the rim of the disc . Arrival of the cumulus at the rim induces the opening of the germ-disc ( Oda and Akiyama-Oda , 2008; Akiyama-Oda and Oda , 2003 , 2006; McGregor et al . , 2008; Hilbrant et al . , 2012; Mittmann and Wolff , 2012; Schwager et al . , 2015 ) . Cumulus migration is dependent on the Hh-signaling pathway ( Akiyama-Oda and Oda , 2010 ) and it was shown that the knockdown of components of this signaling pathway results in cumulus migration defects and in the ectopic opening of the germ-disc ( Akiyama-Oda and Oda , 2010 ) . How the cumulus is specified and forms is still under debate . During the formation of the germ-disc a small cluster of cells ingress and form an indentation where the future center of the fully formed germ-disc will be located . This cluster of cells appears as a visible spot and is called the primary thickening ( Akiyama-Oda and Oda , 2003; Hilbrant et al . , 2012 ) . However , it is not clear whether all or only a subset of the cells of the primary thickening give rise to the cumulus , or if cumulus cells arise from subsequent cell invagination at the site of the primary thickening ( Oda and Akiyama-Oda , 2008; Akiyama-Oda and Oda , 2003 ) . Cell tracing ( Holm , 1952; Edgar et al . , 2015 ) , as well as the expression of the endodermal marker forkhead ( Oda et al . , 2007 ) within the primary thickening/cumulus cells led to the suggestion that the primary thickening/cumulus cells are central endodermal cells ( Hilbrant et al . , 2012; Oda et al . , 2007 ) . However , these studies could not completely rule out that the labeled cumulus cells develop into cells of the visceral mesoderm ( Edgar et al . , 2015 ) . During the last 15 years , research focused on candidate genes known to be involved in development in Drosophila melanogaster has revealed several aspects of how spider embryos pattern their main body axis . However , there are many open questions regarding the early regulation of cumulus specific gene expression , cumulus establishment and maintenance . To overcome the limitations of the candidate gene approach , we have carried out transcriptome sequencing of carefully staged embryos to find new genes involved in cumulus and axial patterning in the spider Parasteatoda tepidariorum . From this work , we have identified the transcription factor Pt-Ets4 as a new gene expressed during early development and have found it to be expressed exclusively within the central primary thickening and the cells of the migrating cumulus . Our combined genetic and cellular analyses show that Pt-Ets4 is needed for the integrity of the cumulus . We found that the knockdown of this gene leads to embryos that show axis patterning defects reminiscent of BMP knockdown phenotypes , suggesting that an intact cumulus is needed to induce the formation of the bilaterally symmetric spider embryo . Importantly , Pt-Ets4 is necessary and sufficient for driving the early expression of twist ( a gene involved in gastrulation and mesoderm formation in Drosophila ) and hunchback , and the ectopic expression of Pt-Ets4 is sufficient to induce cell delamination . Prior to germ-disc condensation , Pt-Ets4 is expressed within the cluster of cells that will form the center of the future germ-disc at early stage 3 ( Figure 1A ) . Expression persists throughout germ-disc formation and , at stage 4 , Pt-Ets4 is strongly and exclusively expressed within the central cluster of cells ( the so-called primary thickening , Figure 1B ) that has delaminated during germ-disc formation ( Figure 1C ) . During stage 5 , the cumulus starts to migrate from the center of the germ-disc to its periphery ( Oda and Akiyama-Oda , 2008; Akiyama-Oda and Oda , 2003 , 2006; McGregor et al . , 2008; Hilbrant et al . , 2012; Mittmann and Wolff , 2012; Schwager et al . , 2015 ) . At this stage Pt-Ets4 is strongly and exclusively expressed in the migrating cumulus cells ( Figure 1D and D’ ) . We were not able to detect Pt-Ets4 transcripts in ovaries and early stage 1 embryos via RNA in situ hybridization ( Figure 1—figure supplement 3A and B ) . In addition , our sequencing data shows that Pt-Ets4 is only expressed at a very low level during early stage 1 , but is mildly up-regulated at late stage 2 and strongly up-regulated at early stage 3 ( Figure 1—figure supplement 1 ) . From this we conclude that Pt-Ets4 transcripts are not maternally provided . Time-lapse imaging and cross-sectioning revealed that the knockdown of Pt-Ets4 neither affected formation of the germ-disc nor of the primary thickening/cumulus ( Video 1B; middle column in Figure 2 , Figure 3A and B ) . However , during stage 5 , cumulus integrity was affected in Pt-Ets4 knockdown embryos ( Video 1B , middle column in Figure 2 ) . While in control embryos the cumulus migrated towards the rim ( Video 1A , left column in Figure 2 ) , the cumulus of Pt-Ets4 RNAi embryos remained at the center of the germ-disc until early stage 5 and disappeared soon after gastrulation was initiated at the center and at the rim of the disc ( Video 1B ( 15h onwards ) , middle column in Figure 2 ) . Analysis of mid-stage 5 Pt-Ets4 RNAi embryos for the expression of the cumulus marker Pt-fascin ( Akiyama-Oda and Oda , 2010 ) revealed that although the cells of the cumulus were still in the center of the germ-disc , they appeared to be more loosely organized ( Figure 3C and D ) . 10 . 7554/eLife . 27590 . 007Figure 2 . Pt-Ets4 is required for cumulus integrity . Stills from the embryos shown in Video 1 . The cumulus ( asterisk ) migrates in the control , disappears in the Pt-Ets4 RNAi and stays in the center of the germ-disc in Pt-ptc RNAi embryo . Ectopic , central opening ( induction of the dorsal field ) of the germ-disc is depicted via the dotted line ( Pt-ptc RNAi st . 6 ) . Posterior ( P ) and anterior ( A ) tube formation in Pt-Ets4 and Pt-ptc knockdown embryos is also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 00710 . 7554/eLife . 27590 . 008Figure 2—figure supplement 1 . Knockdown efficiency after RNAi with Pt-Ets4 . In comparison to the control embryo ( A–A’’ ) the expression of Pt-Ets4 is strongly silenced after Pt-Ets4 RNAi ( B–B’’ ) . Nascent Pt-Ets4 transcripts are still detectable in the nuclei of the primary thickening . The boxed region in A and B is magnified in A’’ and B’’ . ( C ) Schematic representation of the AUGUSTUS prediction for Pt-Ets4 including the location of the untranslated regions ( 5’ and 3’ UTR ) , the coding sequence ( CDS ) and the primers ( T7-Pt-CDS-Ets4-Fw ( Fw1 ) , T7-Pt-CDS-Ets4-Rev ( Rev1 ) , T7-Pt-3’Ets4-Fw ( Fw2 ) , T7-Pt-3’Ets4-Rev ( Rev2 ) , see Material and methods ) that were used to generate two non-overlapping fragments of Pt-Ets4 that were used in the RNAi experiments ( see Material and methods ) . ( D–F ) Statistical analysis of the knockdown efficiency after pRNAi with the CDS fragment and the 3’UTR fragment of Pt-Ets4 . As many embryos were able to completely recover from the Pt-Ets4 knockdown at later stages of development ( see embryo shown in G ) the embryos were analyzed for their phenotypes at embryonic stages 6 and 7 . ( G ) Stills from a time-lapse imaging experiment showing a strongly affected Pt-Ets4 RNAi embryo at developmental stages 6 and 7 . The tube-like structure elongates from the posterior , folds back onto the yolk and re-establishes a DV axis during stages 8 and 9 . The embryo has fully recovered from the early DV phenotype at stage 9 of development . This demonstrates the regulatory capacities of spider embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 00810 . 7554/eLife . 27590 . 009Figure 3 . Cumulus integrity and signaling is affected in Pt-Ets4 knockdown embryos . ( A and B ) Cross-section through the central cumulus of ubiquitously stained ( via Pt-arm RNA in situ hybridization ) control ( A ) and Pt-Ets4 RNAi ( B ) embryos . ( C and D ) control and Pt-Ets4 RNAi embryos stained for the cumulus marker Pt-fascin . Cells of the cumulus are dispersing in the Pt-Ets4 knockdown embryo ( D ) . ( E and F ) Single color double stain of anterior Pt-otd expression ( anterior ring ) and nuclear localized pMAD in the cells overlaying the cumulus . pMAD signal is absent in Pt-Ets4 RNAi embryos ( F ) . In situ hybridization for the ventral fate marker Pt-sog ( G , G’ , J and J’ ) or the segmental marker Pt-en ( H , I , K and L ) in control ( G–I ) and Pt-Ets4 knockdown embryos ( J–L ) . The same embryos in fluorescence vs . bright field channel are shown in G and G’ as well as in J and J’ . Nuclear stain ( DAPI ) /bright field overlay is shown in A-F , H , I , K and L . Flat mounted embryos in C-F . Lateral-ventral view ( G , G’ , J–K ) , lateral view ( H ) , ventral view ( I ) . Abbreviations: ch: cheliceral segment; L1-L4: walking leg bearing segments 1–4; O1: opisthosomal segment 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 00910 . 7554/eLife . 27590 . 010Figure 3—figure supplement 1 . BMP pathway activity in wt embryos . ( A–D ) The BMP signaling pathway ( visualized via pMad antibody staining; false color images ) is active from early stage 5 onwards . However , the germ-disc does not open up before late stage 5 , after the cumulus has reached the rim of the disc . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 01010 . 7554/eLife . 27590 . 011Video 1 . Knockdown of Pt-Ets4 and Pt-ptc . Live imaging of a control ( A ) , a Pt-Ets4 RNAi ( B ) and a Pt-ptc RNAi ( C ) embryo under transmitted light conditions . The video starts at stage 3 and ends at stage 9 of embryonic development . Cumulus migration and normal germ-band formation is visible in the control embryo ( A ) . The cells of the cumulus disperse in the Pt-Ets4 knockdown embryo ( B ) . The ventralized Pt-Ets4 RNAi embryo stays radially symmetric and posterior tube formation is initiated ( 30 hr onwards ) . The cumulus of the Pt-ptc RNAi embryo does not migrate ( C ) . The germ-disc opens up at the central position and the radially symmetric embryo overgrows the yolk and anterior tube formation is initiated ( 48 hr onwards ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 011 In Pt-Ets4 knockdown embryos the radial symmetry of the germ-disc was not broken ( as shown by the formation of a tube-like germ band in Video 1B ( 30h onwards ) , and the middle column in Figure 2 ) , presumably due to the loss of the cumulus . To investigate this phenotype in more detail , we knocked down another gene that also results in defects in radial symmetry breaking , Pt-ptc . In Pt-ptc knockdown embryos , cumulus migration is lost but the cumulus itself is otherwise unaffected ( Akiyama-Oda and Oda , 2010 ) . After pRNAi with Pt-ptc , the cumulus stays in the center of the germ-disc ( Video 1C; right column in Figure 2 ) and BMP signaling is ectopically activated ( as shown via antibody staining against the phosphorylated form of mothers against dpp ( pMAD ) ( Akiyama-Oda and Oda , 2010 ) ) . As a result , the germ-disc ectopically opened at the center and the dorsal field was induced at the posterior pole of the Pt-ptc RNAi embryos ( right column in Figure 2 ) . This ectopic induction of the dorsal field in the center of the germ-disc was never observed in the Pt-Ets4 RNAi embryos . To test if the disappearing cumulus was at least partially able to activate BMP signaling in the germ-disc of Pt-Ets4 RNAi embryos , we performed a pMAD antibody staining in both control and Pt-Ets4 RNAi embryos . In control embryos the cumulus reached the periphery of the germ-disc at late stage 5 and a strong pMAD staining was visible in the overlying ectodermal cells ( Figure 3E , Figure 3—figure supplement 1 ) . At this stage , the anterior marker Pt-orthodenticle ( Pt-otd ) was expressed in a ring , which had a width of 3–5 cells ( Akiyama-Oda and Oda , 2003; Pechmann et al . , 2009; Akiyama-Oda and Oda , 2016 ) ( Figure 3E ) . In Pt-Ets4 knockdown embryos Pt-otd expression was unaffected but nuclear pMAD was not detectable ( Figure 3F ) . This lack of BMP signaling explains why Pt-Ets4 RNAi embryos do not induce the dorsal field and stay radially symmetric . Indeed , the knockdown embryos were completely ventralized; the expression of the ventral marker Pt-short-gastrulation ( Pt-sog ) was uniform around the embryonic circumference and the segmental marker Pt-engrailed ( Pt-en ) was expressed in symmetric rings demonstrating the radial symmetry of the embryo ( Figure 3G–L ) . During later development the embryonic tissue either grew completely over the yolk ( Figure 3J–L ) , or a tube like structure elongated at the posterior of the embryo ( Video 1B , middle column Figure 2 ) . This was in contrast to Pt-ptc RNAi embryos , where the germ-disc opened up centrally and a tube like structure formed at the anterior of the embryo ( right column Figure 2; Video 1C ) . The Pt-Ets4 knockdown phenotype is rather similar to that of Pt-dpp , although cumulus migration/integrity is not affected in Pt-dpp RNAi embryos ( Akiyama-Oda and Oda , 2006 ) . This observation , plus the early and strong expression of Pt-Ets4 and the fact that BMP signaling disappeared upon Pt-Ets4 knockdown , led us to hypothesize that Pt-Ets4 functions as an activator of Pt-dpp transcription within the cells of the cumulus . However , there was no obvious difference in the expression of Pt-dpp in stage 4 control and Pt-Ets4 RNAi embryos ( Figure 4A and E ) . Vice versa , Pt-dpp appears not to be regulating the expression of Pt-Ets4 ( Figure 4—figure supplement 1A ) . In addition , Hh signaling seems not to be involved in the regulation of Pt-Ets4 expression and Pt-Ets4 appears not to regulate the expression of Pt-ptc within the primary thickening ( Figure 4—figure supplement 1B and C ) . 10 . 7554/eLife . 27590 . 012Figure 4 . Analysis of ‘cumulus marker’ genes . Flat mount preparation of in situ stained stage 4 embryos . While Pt-dpp and Pt-fkh expression is unaffected ( compare A to E and B to F ) , Pt-fascin expression is slightly ( compare C to G ) and Pt-hb expression is strongly down regulated in Pt-Ets4 RNAi embryos ( compare D to H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 01210 . 7554/eLife . 27590 . 013Figure 4—figure supplement 1 . Regulation of Pt-Ets4 and Pt-ptc . ( A ) Pt-dpp does not regulate the expression of Pt-Ets4 . ( B ) Single color double staining detecting cumulus-specific Pt-Ets4 and anterior Pt-otd transcripts in a Pt-ptc RNAi embryo . Strong Pt-Ets4 expression ( asterisk ) is detectable in a Pt-ptc RNAi embryo . As already described ( Akiyama-Oda and Oda , 2010 ) , Pt-otd is ectopically activated in the center of the germ-disc ( arrowheads ) . ( C ) Pt-Ets4 does not regulate the expression of Pt-ptc . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 013 In order to determine what other genes Pt-Ets4 may regulate , we studied the expression of genes normally expressed in the primary thickening in Pt-Ets4 RNAi embryos . We found that Pt-forkhead ( Pt-fkh ) expression was unaffected ( Figure 4B and F ) , while the cumulus marker Pt-fascin was slightly down regulated , and Pt-hunchback ( Pt-hb ) was strongly down regulated in Pt-Ets4 knockdown embryos ( Figure 4C , D , G and H ) . As Pt-Ets4 is strongly expressed in the cumulus and is required for cumulus integrity , we wondered how ectopic expression of Pt-Ets4 would affect cell behavior . To generate small cell clones ectopically expressing Pt-Ets4 within the germ-disc , we micro-injected late stage 1 embryos ( via single cell/blastomere injections [Hilbrant et al . , 2012; Kanayama et al . , 2010] ) with in vitro synthesized capped mRNA coding for an EGFP-Pt-Ets4 fusion protein ( see Figure 5—figure supplement 1 ) . Our first observation was that the EGFP-Pt-Ets4 fusion protein localizes to the nuclei of the injected cells , suggesting that the nuclear-localization-signal of Pt-Ets4 is functioning normally ( Figure 5A ) . EGFP-Pt-Ets4 marked cell clones resembled wild-type until stage 4 . As soon as a dense germ-disc formed , however , the cell clones expressing EGFP-Pt-Ets4 seemed to move beneath the germ-disc and the EGFP signal became occluded by the opaque cells of the germ-disc ( Figure 5B ) . We have never seen such behavior following injection of EGFP-NLS constructs alone ( Figure 5F–F’’’ ) , indicating that the change in cell behavior is due to the ectopic expression of Pt-Ets4 . 10 . 7554/eLife . 27590 . 014Figure 5 . Ectopic expression of Pt-Ets4 causes the delamination and migration of cells . ( A and A’ ) A stage 4 embryo in which a cell clone has been marked via the ectopic expression of an EGFP-Pt-Ets4 fusion construct . The fusion protein localizes to the nuclei . ( B and B’ ) The cell clone has delaminated four hours later . As the overlaying epithelium is highly light-scattering , the nuclear EGFP signal is no longer visible . The inset shows the magnification of the boxed region in A and B . ( C and D ) Stills from Video 2A and D ( magnifications of the lynGFP positive regions ) . Cells expressing lynGFP alone ( C–C’’ ) stay at the surface epithelium of the germ-disc . Cells expressing lynGFP/EGFP-Pt-Ets4 ( D–D’’ ) constrict and delaminate . ( E ) EGFP antibody staining ( maximum intensity projection is shown in E and the orthogonal view is shown in E’ ) of a fixed embryo ectopically expressing EGFP-Pt-Ets4 . ( F–F’’’ ) Stills from Video 3A ( control ) . A cell clone marked via the ectopic expression of nuclear localized EGFP ( EGFP-NLS ) marks the ectoderm during germ-band formation . ( G–G’’’ ) Stills from Video 3B ( ectopic expression of Pt-Ets4 ) . A cell clone ectopically expressing EGFP-Pt-Ets4 in combination with EGFP-NLS delaminates after germ-disc formation ( G’ ) . The cells of this cell clone start to disperse during later stages of development ( G’’ and G’’’ ) . Scale bar is 100 µm in E . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 01410 . 7554/eLife . 27590 . 015Figure 5—figure supplement 1 . Constructs . Schematic representation of the constructs that were used for the production of capped mRNA . The EGFP-Pt-Ets4 fusion construct was synthesized and cloned into the pUC57 vector by Eurofins Genomics . This construct was modified ( see Material and methods ) to generate the nuclear localized form of EGFP ( via insertion of the SV40-NLS sequence ) . Location of used restriction sites ( used for modification and linearization ) is indicated above the EGFP-Pt-Ets4 fusion construct . The PolyA tail consists of 25 residues . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 015 To further visualize the process of cell clone delamination , we marked cell membranes by co-injecting capped mRNA coding for EGFP-Pt-Ets4 together with capped mRNA coding for lynGFP ( Köster and Fraser , 2001 ) . In control embryos ( ectopic expression of lynGFP alone ) , lynGFP strongly marked the cell outlines and cell clones stayed at the surface epithelium of the germ-disc ( Video 2A , Figure 5C–C’’ ) . Regardless of the position and the shape of the cell clone , cells expressing lynGFP/EGFP-Pt-Ets4 constricted and delaminated shortly after the formation of the germ-disc ( Video 2B–D , Figure 5D–D’’ ) . The detection of EGFP-Pt-Ets4 cell clones in fixed embryos using an antibody against EGFP confirmed that the labeled cells were below the epithelium of the germ-disc ( Figure 5E and E’ ) . 10 . 7554/eLife . 27590 . 016Video 2 . Ectopic expression of Pt-Ets4 causes the delamination of cells . Ectopic expression of lynGFP ( A ) and lynGFP in combination with EGFP-Ets4 ( B-D ) . The lynGFP positive cell clone ( A ) stays in the ectodermal cell layer of the germ-disc . Regardless of the shape of the cell clones , Pt-Ets4 positive cells apically constrict and delaminate ( B-D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 016 The nuclear signal of the EGFP-Pt-Ets4 fusion construct ( and the membrane signal of the lynGFP ) is hardly visible after the delamination process . Therefore , we co-injected capped mRNA for EGFP-Pt-Ets4 with capped mRNA for nuclear localized EGFP ( EGFP-NLS , see Figure 5—figure supplement 1 ) , a construct that we have found to produce a very bright and persistent fluorescent signal . This experiment resulted in a strong nuclear localized EGFP signal within the marked cell clone , which we used to perform time-lapse imaging . While in the control embryos ( injected with EGFP-NLS alone ) the marked cell clones persisted at the surface of the germ-disc ( Video 3A , Figure 5F and F’ ) , the cell clone expressing both EGFP-NLS and EGFP-Pt-Ets4 delaminated shortly after the formation of the germ-disc ( Video 3B , Figure 5G and G’ ) . As previously shown ( Kanayama et al . , 2010 , Kanayama et al . , 2011 ) , germ-disc cells continue to divide and undergo convergent extension during the formation of the germ-band , which causes cell clones to become thin and elongated as seen in our control EGFP-NLS clones ( Video 3A , Figure 5F’’ and F’’’ ) . In contrast to this , cell clones expressing both EGFP-NLS and EGFP-Pt-Ets4 stopped dividing as soon they delaminated . In addition , when the cumulus started to migrate , the delaminated cells of the EGFP-NLS/EGFP-Pt-Ets4 marked cell clone lost contact with each other and spread out underneath the germ-disc/germ-band epithelium ( Video 3B , Figure 5G’’ and G’’’ ) . This observation was consistent and reproducible in multiple analyzed embryos ( Video 4 ) . 10 . 7554/eLife . 27590 . 017Video 3 . Ectopic expression of Pt-Ets4 causes the delamination and migration of cells . Ectopic expression of EGFP-NLS ( A ) and EGFP-NLS in combination with EGFP-Ets4 ( B ) . The cell clone positive for EGFP-NLS ( A ) stays in the ectodermal cell layer of the germ-disc . Cells further divide and form a long and thin cell clone ( via convergent extension [Kanayama et al . , 2011] ) at stage 8 of embryonic development . The Pt-Ets4 positive cell clone delaminates at stage 4 ( B ) . Cells stop dividing and start to disperse as soon as the cells of the cumulus start to migrate ( st . 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 01710 . 7554/eLife . 27590 . 018Video 4 . Ectopic expression of EGFP-NLS and EGFP-Pt-Ets4 in multiple embryos . Ectopic expression of EGFP-NLS ( A-D ) and EGFP-NLS in combination with EGFP-Ets4 ( E-H ) . EGFP-NLS ( A-D ) and ectopic clones expressing Pt-Ets4 ( E-H ) have equal positions within the germ-disc . While all of the control cell clones ( A-D ) form long and thin stretched cell clones at stage 8 of embryonic development , the Pt-Ets4 positive cells clones ( E-H ) delaminate and the cells disperse from stage 5 onwards . Only the EGFP channel is shown for all embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 018 These results demonstrate that Ets4 expression is sufficient to induce cell delamination and migration in embryos of P . tepidariorum . As already mentioned , Pt-Ets4 seems to have no influence on the early expression of Pt-dpp itself ( Figure 4A and E ) . Furthermore , we were not able to detect Pt-dpp transcripts in cells ectopically expressing Pt-Ets4 ( Figure 6A and A’ ) . In contrast to Pt-dpp , we did find that Pt-Ets4 is regulating the expression of Pt-hb . While Pt-hb expression was nearly absent in Pt-Ets4 RNAi embryos ( Figure 4H ) , Pt-hb transcripts were present in the ectopically Pt-Ets4 positive cell clone ( Figure 6B and B’ ) . 10 . 7554/eLife . 27590 . 019Figure 6 . Pt-Ets4 regulates Pt-hb and Pt-twi expression within the primary thickening . Live stage 4 embryos in which a cell clone is marked via the ectopic-expression of EGFP-Ets4 are depicted in A-C . The same embryos have been fixed and analyzed for their expression of Pt-dpp ( A’ ) , Pt-hb ( B’ ) and Pt-twi ( C’ ) , respectively . Expression of Pt-twi in a control ( D ) and a Pt-Ets4 knockdown embryo ( E ) . A’ , B’ , C’ , D and E are false-color overlays of in situ hybridization images . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 01910 . 7554/eLife . 27590 . 020Figure 6—figure supplement 1 . Controls for the ectopic expression of EGFP-Ets4 . ( A–C ) Live stage 4 control embryos in which a cell clone is marked via the ectopic expression of EGFP-Ets4 ( A ) or EGFP-NLS ( B and C ) . The same embryos have been fixed and analyzed for their expression of Pt-Ets4 ( A’ ) , Pt-twi ( B’ ) and Pt-hb ( C’ ) , respectively ( false-color overlays of in situ hybridization images ) . Injected EGFP-Ets4 mRNA is detectable within the marked cell clone ( A’ ) . The cell clones ectopically expressing EGFP-NLS are negative for Pt-twi and Pt-hb transcripts ( B’ and C’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 02010 . 7554/eLife . 27590 . 021Figure 6—figure supplement 2 . Expression of Pt-twi in wt and Pt-Ets4 RNAi embryos . Newly discovered expression of Pt-twi in the central cell cluster of the forming germ-disc ( arrow in A ) and the primary thickening ( pt in B ) is depicted in ( A–C ) . The already published expression of Pt-twist ( Yamazaki et al . , 2005 ) is depicted in ( D–G ) . The expression of Pt-twi in stage 7 Pt-Ets4 RNAi embryos is not affected ( H and I ) . Anterior is to the left ( where possible ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27590 . 021 The behavior of the Pt-Ets4 positive cells is reminiscent of migrating gastrulating cells that invade the germ-disc from the center and from the rim of the germ-disc ( Mittmann and Wolff , 2012; Kanayama et al . , 2011; Yamazaki et al . , 2005 ) ( see Video 1A ) . As Pt-Ets4 activates the expression of Pt-hb ( a gene that is known to be expressed in mesodermal cells in diverse animals [Schwager et al . , 2009; Franke and Mayer , 2015; Kerner et al . , 2006] ) we wondered if Pt-Ets4 misexpression is inducing a mesoderm-like cell fate . For this reason , we tested whether the ectopic expression of Pt-Ets4 is also inducing the expression of the key mesodermal marker Pt-twist ( Pt-twi ) . Indeed , Pt-twi was detectable within the cell clone that ectopically expressed Pt-Ets4 ( Figure 6C and C’ , compare to controls in Figure 6—figure supplement 1 ) . Interestingly , in the stage 4 embryos ectopically expressing Pt-Ets4 , we could not only detect Pt-twi transcripts within the ectopic Pt-Ets4 expressing cell clone , but also within the central primary thickening ( Figure 6C’ ) . This comes as a surprise , as it was reported that Pt-twi expression is not initiated before the end of stage 5 ( Yamazaki et al . , 2005 ) . For this reason , we reanalyzed the full expression series of Pt-twi in wild-type embryos and we were able to confirm that Pt-twi is expressed in the developing primary thickening of stage 3 and 4 embryos ( Figure 6—figure supplement 2A–C ) . Finally , we confirmed the regulation of Pt-twi via Pt-Ets4 by analyzing the expression of Pt-twi in Pt-Ets4 knockdown embryos . Pt-twi transcripts were no longer detectable in the primary thickening of stage 4 Pt-Ets4 RNAi embryos ( Figure 6D and E ) . However , late segmental mesoderm specification was unaffected , as the expression of Pt-twi was unchanged in stage 7 Pt-Ets4 knockdown embryos ( Figure 6—figure supplement 2H–I’ ) . This confirms that the formation and migration of the cumulus and later gastrulation events from central and peripheral parts of the germ-disc are two independent processes ( Hilbrant et al . , 2012; Oda et al . , 2007 ) . Overall , we suggest that the activation of Pt-twi and Pt-hb by Pt-Ets4 in cell clones initiated a mesoderm-like cell fate that led to the migratory behavior of the ectopic Pt-Ets4 positive cells . The cumulus is a fascinating example of a migrating and signaling organizer . Thus far , cumulus-related dorsoventral patterning defects have been observed in spider embryos that either completely lack BMP signaling or are deficient for cumulus migration ( Akiyama-Oda and Oda , 2006 , Akiyama-Oda and Oda , 2010 ) . Here , we show that the knockdown of the transcription factor Pt-Ets4 generates a novel dorsoventral phenotype that is dependent on cumulus integrity . Our results show that formation of the bilaterally symmetric spider embryo is a precisely timed process that relies on the presence of an intact , migrating and signaling cumulus . Parasteatoda tepidariorum adults and embryos were obtained from our laboratory culture at the University of Cologne . Spiders were kept in plastic vials at room temperature and fed with Drosophila melanogaster and crickets ( Acheta domesticus and Gryllus bimaculatus ) . Embryos were staged according to ( Mittmann and Wolff , 2012 ) . PCR amplification and cloning of Pt-Ets4 was performed using standard techniques . Pt-twist ( AB167807 . 1 ) , Pt-hunchback ( FM956092 . 1 ) , Pt-engrailed ( AB125741 . 1 ) , Pt-fork-head ( AB096073 . 1 ) , Pt-armadillo ( AB120624 . 1 ) , Pt-orthodenticle ( AB096074 . 1 ) , Pt-short-gastrulation ( AB236147 . 1 ) , Pt-decapentaplegic ( AB096072 . 1 ) and Pt-fascin ( AB433905 . 1 ) have been isolated previously . The transcriptomes of the embryonic stages 1–3 ( early stage 1 , late stage 2 and early stage 3; see Figure 1—figure supplement 1 ) were sequenced ( HiSeq2000 ) at the Cologne Center for Genomics . The total RNA of three cocoons per stage was pooled and sequenced in each case . The sequence reads ( deposited to the Sequence Read Archive ( http://www . ncbi . nlm . nih . gov/sra/ ( RRID:SCR_004891 ) , BioProject ID: PRJNA383558 ) were mapped to the AUGUSTUS gene predictions ( RRID:SCR_008417 ) ( https://i5k . nal . usda . gov/Parasteatoda_tepidariorum , Schwager et al . , 2017 ) using Bowtie 2 ( Langmead and Salzberg , 2012 ) . New candidates were picked according to their expression profile . These new candidates ( including Pt-Ets4; AUGUSTUS prediction: aug3 . g4238 ) were up-regulated in a similar manner as Pt-decapentaplegic , Pt-hedgehog and Pt-patched ( genes that show a defect in dorsoventral patterning or cumulus migration upon knockdown ( Akiyama-Oda and Oda , 2006 , Akiyama-Oda and Oda , 2010 ) ( see Figure 1—figure supplement 1 ) . A 1094 bp fragment of Pt-Ets4 was amplified using the primer Pt-g4238-Fw ( 5’-GTA CAC AGC ACC TTC TAT TAT GG-3’ ) and Pt-g4238-Rev ( 5’-CCT TCT TGT AAT ATT GGC GA-3’ ) in an initial PCR . For the production of dsRNA a T7 promoter sequence was added to the 5’ and 3’ end of the sequence by performing a nested PCR with the primer T7-Pt-g4238-Fw ( 5’-GTA ATA CGA CTC ACT ATA GGG CCA CAA AAG ATG GCC-3’ ) and T7-Pt-g4238-Rev ( 5’-GTA ATA CGA CTC ACT ATA GGG GAA CGG CTG AGT TTG-3’ ) . This nested PCR yielded a 1046 bp fragment that was used for the initial knockdown of Pt-Ets4 . Double stranded RNA ( dsRNA ) was produced using the MEGAscript T7 Kit ( ThermoFisher SCIENTIFIC ) . Within one week , adult females of Parasteatoda tepidariorum were injected three to four times with 2 µl dsRNA solution ( 2–3 µg/µl ) . Water injections served as a control . The knockdown of Pt-Ets4 was performed several times ( nexperiments > 5; ninjected females >24 ) and always resulted in the same phenotype . For the statistical analysis ( Figure 2—figure supplement 1D–F ) two non-overlapping fragments , targeting the CDS and the 3’UTR of Pt-Ets4 , were used ( see Figure 2—figure supplement 1C ) . The coding sequence of Pt-Ets4 was amplified using the primer Pt-CDS-Ets4-Fw ( 5’-GTA GTC TTG AAC TTC AGT TAT CAA AG-3’ ) and Pt-CDS-Ets4-Rev ( 5’-GGT TTA CTT CAA GAA CTG GAC-3’ ) and was cloned into the pCR4 vector ( ThermoFisher SCIENTIFIC ) . The 3’UTR of Pt-Ets4 was amplified using the primer Pt-3’Ets4-Fw ( 5’-CAC TAT GGT TTC AAA CAT CGA TTG-3’ ) and Pt-3’Ets4-Rev ( 5’-GTC ATA TCC CCT CTA TAG CTA AC-3’ ) and was cloned into the pCRII-Blunt vector ( ThermoFisher SCIENTIFIC ) . For the production of dsRNA the T7 promoter sequence was added to both ends of the CDS and the 3’UTR fragment by using the primer T7-Pt-CDS-Ets4-Fw ( 5’-GTA ATA CGA CTC ACT ATA GGG GTA GTC TTG AAC TTC AGT TAT C-3’ ) and T7-Pt-CDS-Ets4-Rev ( 5’-GTA ATA CGA CTC ACT ATA GGG GTC TGA AGT AAT CTT CTG ATA G-3’ ) and T7-Pt-3’Ets4-Fw ( 5’-GTA ATA CGA CTC ACT ATA GGG CAC TAT GGT TTC AAA CAT CG-3’ ) and T7-Pt-3’Ets4-Rev ( 5’-GTA ATA CGA CTC ACT ATA GGG CCT AAA ACA CAG TTT TAG GAG-3’ ) , respectively . We observed a similar knockdown efficiency for both fragments with the highest penetrance in the third and fourth cocoons ( Figure 2—figure supplement 1E and F ) . As many embryos were able to recover from the knockdown of Pt-Ets4 ( one analyzed cocoon had a recovery rate of 71%; n = 66 ) during later stages of development ( >stage 8 , see Figure 2—figure supplement 1G ) , the number of affected embryos was analyzed during the embryonic stages 6 and 7 . A gene fragment of Pt-decapentaplegic was amplified using the primers Pt-dpp-Fw ( 5’-GTG ATC ATA ACA GGT TCC TGA CC-3’ ) and Pt-dpp-Rev ( 5’-GAC AAA GAA TCT TAA CGG CAA CC-3’ ) . The resulting 1147 bp Pt-dpp fragment was cloned into pCRII-Blunt vector . dsRNA template was generated by using T7 and T7Sp6 primer . We injected three adult females of P . tepidariorum with Pt-dpp dsRNA and the knockdown resulted in the same phenotype as published ( Akiyama-Oda and Oda , 2006 ) . We used the Pt-dpp pRNAi embryos of a fourth cocoon ( the development of 73 embryos of this cocoon were monitored under oil; one embryo died and 72 embryos showed a strong BMP signaling defect phenotype ( Akiyama-Oda and Oda , 2006 ) during stages 6–8 ) to perform the Pt-Ets4 in situ staining shown in Figure 4—figure supplement 1A . Two gene fragments of Pt-patched were amplified from a plasmid ( containing a 2 kb fragment of Pt-ptc ) using the primer T7-Pt-ptc-Fw1 ( 5’-GTA ATA CGA CTC ACT ATA GGG GGG TAG AAG ACG GCG G-3’ ) and T7-Pt-ptc-Rev1 ( 5’-GTA ATA CGA CTC ACT ATA GGG GAG ACT CTT TAG CTA TAA TCT C-3’ ) and T7-Pt-ptc-Fw2 ( 5’-GTA ATA CGA CTC ACT ATA GGG GAG ATT ATA GCT AAA GAG TCT C-3’ ) and T7-Pt-ptc-Rev2 ( 5’-GTA ATA CGA CTC ACT ATA GGG GAT TTG TTT GTC GAC CAC C-3’ ) . dsRNA of both Pt-ptc fragments were combined and injected into three adult P . tepidariorum females . The knockdown resulted in the same phenotype as published ( Akiyama-Oda and Oda , 2010 ) ( see right column in Figure 2 , Video 1 ) . Amino acid sequences were obtained from FlyBase ( RRID:SCR_006549 ) ( dos Santos et al . , 2015 ) , WormBase ( RRID:SCR_003098 ) ( WormBase release Version: WS257 ) , or translated from the P . tepidariorum AUGUSTUS predictions online ( https://i5k . nal . usda . gov/Parasteatoda_tepidariorum ) . Amino acid sequences were aligned using MUSCLE ( RRID:SCR_011812 ) ( Edgar , 2004 ) , alignments were trimmed using TrimAl with the GappyOut setting ( Capella-Gutiérrez et al . , 2009 ) , and maximum likelihood based phylogenies were constructed using PhyML at ‘phylogeny . fr’ ( Dereeper et al . , 2008 ) . Full amino acid sequences were used for all genes except for Pt-aug3 . g5814 . t1 , which is missing the N-terminus but still contains the ETS domain ( as predicted online [de Castro et al . , 2006] ) . Final phylogenies were generated with the WAG substitution model and 1000 bootstrap replicates ( Whelan and Goldman , 2001 ) . Phylogenetic analysis was also performed using the ETS domains alone , and while tree topology changed in some ways , the Ets4 genes from D . melanogaster and C . elegans still branched together with strong support , and the gene we have named Pt-Ets4 was the only P . tepidariorum gene branching together with this clade . Experiments have been performed by injecting capped mRNA into late stage 1 embryos of P . tepidariorum . Embryonic microinjections were performed as described previously ( Pechmann , 2016 ) . For the production of capped mRNA , the mMASSAGE mMACHINE Kit ( T7 or Sp6 , ThermoFisher SCIENTIFIC ) was used . Capped mRNA was injected at a concentration of 2–3 µg/µl . For the ectopic expression of Pt-Ets4 an EGFP-Pt-Ets4-PolyA fusion construct was synthesized at Eurofins Genomics ( see Figure 5—figure supplement 1; see Supplementary file 1 for full sequence ) . For the production of capped mRNA , the construct contained a T7 and a Sp6 promoter at its 5’ end and could be linearized via NotI , PstI or EcoRI restriction enzyme digest . In addition , the coding sequence of EGFP-Pt-Ets4 was flanked by the 5’ and the 3’ UTR of the Xenopus beta-globin gene ( also used in Tribolium [Benton et al . , 2013] ) . For the ectopic expression of NLS-EGFP , the Pt-Ets4 sequence of the EGFP-Pt-Ets4-PolyA construct was removed ( via BglII , SalI double digest ) and replaced by the sequence MAKIPPKKKRKVED ( contains the SV40 T antigen nuclear localization signal [Kanayama et al . , 2010] ) . For this , the primer BglII-NLS-SalI-Fw ( 5’-TTT AGATCT ATG GCT AAA ATT CCT CCC AAA AAG AAA CGT AAA GTT GAA GAT TAA GTCGAC TTT-3’ ) and BglII-NLS-SalI-Rev ( 5’-AAA GTCGAC TTA ATC TTC AAC TTT ACG TTT CTT TTT GGG AGG AAT TTT AGC CAT AGATCT AAA-3’ ) ( coding for the NLS sequence ) were annealed to each other , digested with BglII and SalI and inserted to the already cut vector . This resulted in an EGFP-NLS-PolyA construct ( see Figure 5—figure supplement 1 ) . The function of Pt-Ets4 was analyzed either by injecting capped mRNA of the Pt-Ets4-EGFP fusion construct alone or by injecting capped mRNA of Pt-Ets4 ( the EGFP was removed from the EGFP-Pt-Ets4-PolyA construct via an XhoI digest ) together with capped mRNA of EGFP-NLS . Ectopic expression of Pt-Ets4 co-injected with EGFP-NLS resulted in the same phenotype as shown for the EGFP-Pt-Ets4 fusion construct . To obtain a stronger signal during live imaging , capped mRNA of EGFP-Pt-Ets4 was co-injected with EGFP-NLS . To mark the membranes of the embryonic cells and to visualize the delamination process of the Pt-Ets4 positive cell clones , capped mRNA coding for lynGFP ( Köster and Fraser , 2001 ) was co-injected with capped mRNA coding for EGFP-Pt-Ets4 . Next to the ectopic expression of lynGFP we tried to ectopically express GAP43YFP ( another marker that was shown to localize to the cell membranes of Tribolium embryos [Benton et al . , 2013 , Benton et al . , 2016] ) . However , we observed a much stronger signal for lynGFP . Embryos used for RNA in situ hybridization were fixed in a two phase fixative containing 1 . 5 ml of PBS , 1 . 5 ml 10% formaldehyde and 3 ml heptane . Prior to fixation embryos were dechorionated for 3–5 min in a 2 . 8% hypochlorite solution ( DanKlorix ) . The embryos were washed with H2O several times and were transferred to the fixative , subsequently . Embryos were fixed for over night at room temperature and 50–100 rpm . After the fixation embryos were gradually transferred ( 30% , 50% , 80% ) to 100% methanol . In situ hybridization was performed as previously described ( Prpic et al . , 2008 ) with minor modifications ( proteinase K treatment was not carried out ) . Fluorescent FastRed staining was performed as described in ( Benton et al . , 2016 ) . To analyze BMP pathway activity in control and Pt-Ets4 knockdown embryos a pMAD antibody staining was performed in embryos that were already stained for Pt-otd ( via in situ hybridization , n > 10 for control and Pt-Ets4 RNAi embryos ) . Embryos were fixed as described above . In situ stained embryos were washed in PBST ( 3 × 15 min ) and blocked in PBST containing 0 , 1% BSA and 5% goat serum ( 1 hr at RT ) . Subsequently , the embryos were transferred to a fresh solution of PBST containing 0 , 1% BSA and 5% goat serum . The Phospho-Smad1/5 ( Ser463/465 ) ( 41D10 ) Rabbit mAb ( Cell Signaling Technology , Inc . ( RRID:AB_491015 ) ) was added to this solution ( antibody concentration: 1:1000; 4°C o . n . ) . On the next day the embryos were washed in PBST ( 3 × 15 min ) and were blocked again in PBST containing 0 , 1% BSA and 5% goat serum ( 1 hr at RT ) . The secondary antibody ( Anti-Rabbit IgG , couplet to alkaline phosphatase ( AP ) , produced in goat; A3687 SIGMA ( RRID:AB_258103 ) ) was added to the blocking solution at a 1:1000 concentration . After incubating the secondary antibody for 2–3 hr ( RT ) excessive antibody was removed by washing the embryos several times in PBST ( 6 × 15 min; final washing step at 4°C o . n . ) . Finally , a regular NBT/BCIP staining was carried out ( see whole mount in situ hybridization ) . For the control and the Pt-Ets4 RNAi embryos , the staining reaction was stopped at identical time points We used an anti green fluorescent protein mouse IgG antibody ( A11120; ThermoFischer SCIENTIFIC ( RRID:AB_221568 ) ; final concentration 1:1000 ) as primary and an Alexa Fluor 488 goat anti mouse IgG ( A11001; ThermoFischer SCIENTIFIC ( RRID:AB_2534069 ) ; final concentration 1:400 ) as secondary antibody . To stain all of the embryonic cells , in situ hybridization with the ubiquitously expressed gene Pt-arm was carried out in early stage 5 embryos of control and Pt-Ets4 RNAi embryos . Embryos were then stained with Sytox Green ( 1:5000 in PBST , ThermoFischer SCIENTIFIC ) . Embryos were then gradually ( 50% , 70% , 90% ) transferred to 100% EtOH . After a washing step in 1:1 EtOH/acetone , the embryos were transferred to 100% acetone . Single embryos were transferred to microtome embedding molds in a 1:1 durcupan/acetone solution . Acetone was removed by incubating the embryos at room temperature ( o . n . ) . The embedding molds were filled with fresh durcupan ( Fluka ) . Polymerisation of the durcupan was carried out at 65°C ( 16–20 hr ) . Cross sectioning ( 8 µm ) was performed on a LEICA RM 2255 microtome ( n = 3 for control and Pt-Ets4 RNAi embryos ) . RNA from stage 1–3 embryos was extracted and sequenced as described in the ‘Identification of Pt-Ets4’ section above . These sequences were made available for us to download from the Cologne Centre for Genomics server , and FastQC ( RRID:SCR_014583 ) ( Andrews , 2010 ) was used for initial assessment of read quality . This was excellent ( lower quartile Phred quality above 30 until the last base in the read , no residual adapter sequence noted ) and as such no trimming was performed . Comparative expression analysis was performed by mapping reads to Parasteatoda_tepidariorum AUGUSTUS gene predictions ( https://i5k . nal . usda . gov/Parasteatoda_tepidariorum ) using RSEM 1 . 2 . 28 ( Li and Dewey , 2011 ) and Bowtie 1 . 0 . 0 ( RRID:SCR_005476 ) ( Langmead , 2010 ) as packaged in the Trinity 2 . 2 . 0 module ( RRID:SCR_013048 ) ( -est_method RSEM—aln_method bowtie [Grabherr et al . , 2011] ) . Cross sample normalization was performed using Trimmed Mean of M-values , and edgeR ( RRID:SCR_012802 ) ( Robinson et al . , 2010 ) was run to determine differential expression with a dispersion ratio fixed at 0 . 1 . Those differentially expressed genes with a p-value cut off for FDR of 0 . 001 and min abs ( log2 ( a/b ) ) change of 2 were then chosen for annotation and further examination , with target gene results provided in Figure 1—figure supplement 1 . Pictures were taken using an Axio Zoom . V16 that was equipped with an AxioCam 506 color camera . Confocal imaging was performed on a LSM 700 ( Zeiss ) . Live imaging was carried out on the Axio Zoom . V16 , a Zeiss AxioImager . Z2 ( equipped with an AxioCam MRm camera and a movable stage ) and on a Leica CLSM SP8 ( Imaging facility Biocenter Cologne ) . Projections of image stacks were carried out using Helicon Focus ( HeliconSoft ( RRID:SCR_014462 ) ) or Fiji ( Schindelin et al . , 2012 ) ( RRID:SCR_002285 ) . All movies have been recorded at room temperature and images have been adjusted for brightness and contrast using Adobe Photoshop CS5 ( RRID:SCR_014199 ) . For false-color overlays of in situ hybridization images a bright field image of the NBT/BCIP staining was inverted . This inverted picture was pasted into the red channel of the nuclear stain image . The input levels ( Adobe Photoshop CS5; Levels function ) of the red channel were adjusted in a way that only the signal of the NBT/BCIP staining remained visible .	At the earliest stages of animal development , embryos consisting of only a handful of cells must figure out where each part of their body will come from . The first step in this process is to determine what will be their head versus their tail , and what will be their front versus their back . Many animals use specialized groups of cells , called “organizers” , to make this decision . This occurs in backboned animals – including humans – and also in distantly related animals such as spiders . In spiders , the developing embryo must form an organizer called the “cumulus” or the spiderling will not develop correctly . In order to form and maintain the cumulus , various genes must be turned on in a carefully controlled order in exactly the right cells . Pechmann et al . have now discovered the role of a previously unknown gene ( called Pt-Ets4 ) that marks the spot where the cumulus forms . This gene is required for cumulus maintenance and it also helps to activate a number of other cumulus-specific genes . When this gene is disrupted , the spider embryo does not properly differentiate its front from its back . The findings presented by Pechmann et al . add to a growing foundation of studies aiming to understand how genes ‘talk’ to one another and organize embryos as they develop . In years to come , the unraveling of these gene pathways , where genes sequentially turn other genes on and off , will allow us to more fully understand how a single cell can grow into a complete adult animal .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	A novel role for Ets4 in axis specification and cell migration in the spider Parasteatoda tepidariorum
Cardiac neural crest cells contribute to important portions of the cardiovascular system including the aorticopulmonary septum and cardiac ganglion . Using replication incompetent avian retroviruses for precise high-resolution lineage analysis , we uncover a previously undescribed neural crest contribution to cardiomyocytes of the ventricles in Gallus gallus , supported by Wnt1-Cre lineage analysis in Mus musculus . To test the intriguing possibility that neural crest cells contribute to heart repair , we examined Danio rerio adult heart regeneration in the neural crest transgenic line , Tg ( −4 . 9sox10:eGFP ) . Whereas the adult heart has few sox10+ cells in the apex , sox10 and other neural crest regulatory network genes are upregulated in the regenerating myocardium after resection . The results suggest that neural crest cells contribute to many cardiovascular structures including cardiomyocytes across vertebrates and to the regenerating heart of teleost fish . Thus , understanding molecular mechanisms that control the normal development of the neural crest into cardiomyocytes and reactivation of the neural crest program upon regeneration may open potential therapeutic approaches to repair heart damage in amniotes . The neural crest is an important stem cell population characterized by its multipotency , migratory behavior , and broad ability to differentiate into derivatives as diverse as elements of the cardiovascular system , craniofacial skeleton , and peripheral nervous system . However , not all neural crest cells are alike , with distinct populations existing along the body axis . One of the most unique neural crest populations is the ‘cardiac neural crest’ that contributes to the outflow septum and smooth muscle of the outflow tract of the heart . Ablation studies in chick embryos show that removal of the cardiac crest results in a broad range of defects , including persistent truncus arteriosus , abnormal myocardium function , and misalignment of the arch arteries ( Kirby et al . , 1983; Waldo et al . , 1999; Bockman et al . , 1987 ) . These defects are highly reminiscent of some of the most common human congenital heart defects . Importantly , other neural crest populations cannot rescue the effects of cardiac neural crest ablation even when grafted in its place , exemplifying the uniqueness of this population ( Kirby , 1989 ) . Classically , quail-chick transplantation experiments have been used to uncover contributions of the cardiac neural crest to the heart , with some more recent attempts using antibody staining of migratory neural crest cells or LacZ retroviral lineage analysis as well as transgenic lines such as Wnt1-Cre driven β-galactosidase in mammals ( Kirby et al . , 1983; Kuratani and Kirby , 1991; Boot et al . , 2003; Jiang et al . , 2000 ) . The results suggest that the cardiac neural crest contributes to smooth muscle cells lining the great arteries , outflow tract septum and valves , mesenchyme that remodels pharyngeal arch arteries , and parasympathetic innervation of the heart , such as the cardiac ganglion . However , inconsistencies remain between different lineage approaches , most of which suffer from high background and low cellular resolution . To reconcile these differences , here , we use a multi-organismal approach to examine the lineage contributions of cardiac neural crest to the heart . Using a novel retroviral labeling approach in chick and confirmed by Wnt1-Cre reporter lines in mouse , we reveal a previously undetected contribution of the amniote cardiac neural crest to the trabecular myocardium of the ventricles , a derivative previously thought to be confined to non-amniotic vertebrates ( Sato and Yost , 2003; Li et al . , 2003; Cavanaugh et al . , 2015 ) . The homologous cardiac neural crest contribution to cardiomyocytes across diverse species raised the intriguing possibility that these cells may contribute to cardiac repair . As the adult zebrafish heart exhibits extensive regenerative capacity , we turned to this model to test whether the neural crest may contribute to heart regeneration ( Poss et al . , 2002 ) . Intriguingly , we show that resected adult zebrafish hearts reactivate many genes of a neural crest gene regulatory program during the regeneration process . Taken together , these results demonstrate an evolutionarily conserved contribution of neural crest cells to cardiomyocytes across vertebrates and a previously unappreciated role during heart regeneration . To specifically label cardiac neural crest cells prior to their emigration from the neural tube and identify novel progeny of chick cardiac crest , we use a replication-incompetent avian retrovirus ( RIA ) that indelibly and precisely marks neural crest progenitors for long term lineage analysis at single cell resolution and without the need for tissue grafting . To this end , the post-otic neural tube of the hindbrain adjacent to somites 1–3 was injected at Hamburger and Hamilton ( HH ) stage 9–10 with high-titer ( 1 × 107 ifu/mL ) RIA ( Figure 1A ) , which drives expression of nuclear localized H2B-YFP under control of a constitutive RSV promoter ( Li et al . , 2017; Tang et al . , 2019; Fields-Berry et al . , 1992; Chen et al . , 1999; Hamburger and Hamilton , 1951 ) . At this stage in the development , premigratory cardiac neural crest cells are positioned within the dorsal neural tube and about to emigrate . Accordingly , this labeling approach solely marks hindbrain neural tube cells including premigratory cardiac neural crest cells that subsequently delaminate from the dorsal neural tube during a two-hour time window when the virus remains active . Virally infected embryos were then allowed to develop for 1–9 days post injection , cryo-sectioned , and analyzed using confocal microscopy . One day after injection , whole mount imaging revealed RIA-labeled cells migrating in a stream along pharyngeal arch 3 ( Figure 1B , B’ ) , that subsequently accumulated in pharyngeal arches 3 , 4 and 6 two days after infection ( Figure 1C ) . Next , we confirmed that all labeled cells in the periphery co-localized with the migratory neural crest marker , HNK-1 , demonstrating that the neural crest is the only population labeled with H2B-YFP outside the neural tube , thus verifying specificity of infection ( Figure 1D , Figure 1—figure supplement 1A ) . With time , labeled cardiac crest cells were observed in numerous and diverse derivatives , populating the cardiovascular system in a proximal to distal progression ( Figure 1E–I , Supplementary file 1a ) . Consistent with quail-chick chimera , we observed RIA-labeled cells adjacent to and within the walls of pharyngeal arch arteries , in the aorticopulmonary septum , outflow tract , and cardiac cushion . Moreover , we definitively observed YFP-labeled cells in the superior interventricular septum , a site for which the neural crest contribution has been controversial , although ventricular septal defects are common after cardiac neural crest ablation ( Kirby et al . , 1985 ) . The cells of the outflow tract septum and pharyngeal arch arteries differentiated into smooth muscle actin ( SMA ) positive cells on embryonic day ( E ) 5 ( Figure 2A , B ) . Importantly , by E3 and onward , virally labeled neural crest cells were observed in the myocardium of both the outflow tract and the ventricles , where they expressed the myocardial markers , Troponin T and Myosin Heavy Chain ( Figure 1H , Figure 1—figure supplement 1B , outflow tract; Figure 2C , D , Figure 2—figure supplement 1A , B , ventricles ) . These neural crest-derived cardiomyocytes were not actively undergoing cell division or programmed cell death ( Figure 2E , F ) , consistent with the stable presence of cells observed over time ( Figure 1I , Supplementary file 1a , 1b ) . Supplementary file 1a and 1b present quantification of contributions of virally labeled cells in the chick ventricular myocardium . While previous lineage tracing experiments in zebrafish showed that a stream of neural crest cells integrate into the myocardium of the primitive heart tube to give rise to cardiomyocytes , our results present the first evidence of a homologous neural crest contribution to cardiomyocytes in chick embryos ( Sato and Yost , 2003; Li et al . , 2003; Cavanaugh et al . , 2015 ) . To test whether the contribution of cardiac neural crest cells to the myocardium was conserved in mammals , we examined Wnt1-Cre;ZsGreenfl/fl transgenic mice in which neural crest cells were labeled with cytoplasmic GFP ( Chai et al . , 2000 ) . Embryos were fixed at E15 . 5 ( similar to E7 in chick ) . Analogous to the results in chick embryos , we observed a large number of ZsGreen-positive myocardial cells in the outflow tract and ventricles , as confirmed by Troponin T expression ( Figure 3A–C ) . To avoid ectopic expression that has been associated with the Wnt1-Cre;ZsGreenfl/fl transgenic line due to endogenous Wnt1 activation caused by in-frame ATG located upstream of Wnt1 start codon , we tested an improved Wnt1 line ( Wnt1-Cre2+; R26mTmG mouse line ) without ectopic activation of canonical Wnt/β-catenin pathway ( Lewis et al . , 2013 ) . The results were similar to those observed with the Wnt1-Cre;ZsGreenfl/fl transgenic mice ( Figure 3D , E ) . As in the chick embryos , murine neural crest derived cells were present in the outflow tract , interventricular septum , and myocardium of both ventricles . The numbers of neural crest-derived cells appear to decrease with distance along the proximal-to-distal axis ( Figure 3—figure supplement 1A ) , such that no neural crest-derived cardiomyocytes were observed in the apex of the heart ( Figure 3—figure supplement 1D , E ) . As in the chick , the numbers of Wnt1+ cells remain stable with time , and the cells do not appear to undergo active cell division or apoptosis ( Figure 3—figure supplement 1B , B’ , C , C’ ) . This contribution persists postnatally , as Wnt1+ cells are present at postnatal day 2 ( Figure 3—figure supplement 1F–H ) . These results are consistent with previous studies using less specific P0-cre lines and demonstrate that comparable cardiac crest contributions occur in birds and mammals ( Tomita et al . , 2005; Tamura et al . , 2011 ) . Quantification of numbers of neural crest lineage labeled cells in the trabeculated myocardium of mice reveals that they represent approximately 17% of the population in the proximal half of the ventricle ( Supplementary file 1a ) . The lineage contributions of neural crest-derived cells in chick and mouse are remarkably similar to those previously shown in zebrafish ( Sato and Yost , 2003; Li et al . , 2003; Cavanaugh et al . , 2015 ) . In all three species , neural crest cells contribute to cardiomyocytes of the trabecular myocardium . This homologous lineage contribution in both amniotes and anamniotes raised the intriguing possibility that neural crest cells may represent a cell population that could contribute to heart repair in adults . In adult birds and mammals , cardiac injury leads to scarring with little regeneration , whereas heart regeneration is common in amphibians and fish ( González-Rosa et al . , 2017 ) . For example , adult zebrafish have the capacity to regenerate their hearts after removal of up to 20% of the ventricle . This has been shown to occur by dedifferentiation and proliferation of pre-existing cardiomyocytes ( Poss et al . , 2002; Jopling et al . , 2010 ) . Given that cardiac neural crest cells give rise to a portion of zebrafish cardiomyocytes during development ( Sato and Yost , 2003; Li et al . , 2003; Cavanaugh et al . , 2015 ) similar to those we report here in chick and mouse , we next asked whether the progeny of these cells might have the ability to contribute to heart regeneration in adult zebrafish . To address this possibility , we first turned to a transgenic line expressing GFP under the control of a sox10 promoter , Tg ( −4 . 9sox10:eGFP ) , that labels all embryonic migratory neural crest lineages to address whether neural crest-derived cardiomyocytes reactivated their developmental program upon injury ( Carney et al . , 2006 ) . While sox10 is expressed in migrating zebrafish cardiac neural crest cells , it is down-regulated in the embryo shortly after these cells reach the heart ( Cavanaugh et al . , 2015 ) . We confirmed this in adult hearts , finding that very few cells within the apex of the adult myocardium of control adult fish expressed sox10 one month post-sham injury , in which the body cavity was opened but no resection was made ( Figure 4A , Supplementary file 1c , n = 3 ) . However , after surgical removal of ~20% of the ventricular apex , cells in the heart reactivated the sox10 promotor sequence and began to re-express GFP in cardiomyocytes of the trabeculated myocardium near the injured site by 7 days post resection ( dpa ) ( Figure 4A; n = 6 ) . GFP expression was not limited to the regenerating tissue but was also observed in the uninjured part of the ventricle . By 21dpa , the hearts had undergone vast regeneration and morphologically were nearly indistinguishable from controls ( Figure 4A; n = 6 ) . Interestingly , consistent with our prediction , the regenerating apex was comprised of more sox10+ positive cells ( Figure 4A , B , Supplementary file 1c ) , suggesting that these cells had proliferated and redeployed a neural crest gene regulatory program during the heart regeneration process . To test if sox10 and other bona fide neural crest markers such as tfap2a , were upregulated endogenously , we performed in situ hybridization on paraffin sections of regenerating and uninjured ventricles . The results reveal upregulation of expression of sox10 and tfap2a transcripts after injury , whereas they were mostly absent from uninjured ventricles ( Figure 4B , C ) . Furthermore , we observed co-localization of sox10 transcripts with a Tg ( sox10:GAL4-UAS-Cre;ubi-Switch ) , which permanently labels all sox10-derived lineages with mCherry ( Figure 4—figure supplement 1 , n = 2 ) . The Tg ( sox10:GAL4-UAS-Cre;ubi-Switch ) is a double transgenic line for the sox10:GAL4-UAS-Cre transgene and the ubi:Switch reporter in which the sox10 promoter drives expression of Cre recombinase . Upon activation of sox10 expression in neural crest cells , eGFP is excised and so cells of the sox10 lineage are permanently labeled with mCherry ( Cavanaugh et al . , 2015 ) . All cells expressing sox10+ transcripts also had mCherry , though not all mCherry positive cells were sox10+ at the 7 day time point ( Figure 4—figure supplement 1 , insets 1 and 2 ) . Our results are consistent with recent findings from Abdul-Wajid and colleagues , who observed that ablation of the embryonic neural crest yields few or no sox10+ cells in the adult heart and results in severe heart defects ( Abdul-Wajid et al . , 2018 ) . This suggests there are no subsequent post-embryonic neural crest additions to the heart and that the population we observe re-expressing neural crest genes are embryonic-derived neural crest progeny . These results raise the intriguing possibility that the neural crest developmental gene regulatory network was being redeployed in neural crest-derived cells of the heart during regeneration . To test this , we performed transcriptional profiling of sox10:mRFP+ cells in the regenerating zebrafish hearts at 21dpa . To this end , we dissected and dissociated injured ventricles ( n = 12 per replicate ) into single cell suspensions and performed FAC-sorting of sox10:mRFP+ cells ( Figure 4—figure supplement 2A ) . The results were compared with mRFP negative cardiac cells from the same injured , isolated ventricles . This led to the identification of 1093 genes that are significantly enriched ( p-adj <0 . 05 ) in regenerating sox10+ cells compared to sox10- cells of the same injured ventricles ( Figure 4D , Figure 4—figure supplement 2 ) . We then compared the differentially expressed genes of isolated 21dpa sox10+ cells to: 1 ) our recently published chick developmental cardiac neural crest gene regulatory program , 2 ) known zebrafish neural crest genes , and 3 ) core neural crest gene regulatory network genes expressed at all axial levels ( Tani-Matsuhana et al . , 2018; Martik and Bronner , 2017; Lukoseviciute et al . , 2018 ) . The results revealed upregulation of many genes of the embryonic neural crest gene regulatory network at the time of regeneration ( Figure 4D and E ) . Interestingly , numerous genes known to be responsible for cardiomyocyte proliferation also are expressed in sox10+ cells upon heart injury ( Figure 4—figure supplement 2E ) ( González-Rosa et al . , 2017 ) . The co-expression of these genes as well as an upregulation of a cell proliferation gene signature suggests a role for sox10-derived cells in cardiomyocyte proliferation during regeneration ( Figure 4—figure supplement 2C ) . Furthermore , these results suggest that the population of proliferating cardiomyocytes in the regenerating heart is heterogeneous and comprised of both neural crest- and mesoderm-derived cardiomyocytes ( González-Rosa et al . , 2017; Sánchez-Iranzo et al . , 2018; Schindler et al . , 2014; Kikuchi et al . , 2010 ) . While much attention has been paid to the molecular signals that promote myocardial dedifferentiation and proliferation during regeneration , far less is known about the cell lineages that contribute to the regeneration process . Based on our observation on the lineage relationship between cardiac neural crest cells and cardiomyocytes during development , we propose that neural crest-derived cells ( progenitors and/or pre-existing cardiomyocytes ) may represent a key population that proliferates and differentiates into new cardiomyocytes after injury . Our cell lineage labeling results provide direct evidence for a neural crest contribution to the undamaged myocardium of the amniote heart . Furthermore , consistent with previous lineage tracing experiments in zebrafish ( Cavanaugh et al . , 2015 ) , where a proportion of cardiac crest derived-cells were located in the trabeculated myocardium in adult fish , we show that after injury , there is activation of numerous neural crest gene regulatory transcription factors and other neural crest genes during regeneration ( Figure 4 ) . While the underlying gene regulatory network of neural crest cells is responsible for formation of cardiomyocytes during normal development , we speculate that it also does so in a similar manner upon injury by redeploying sox10 and other neural crest gene regulatory network genes . The finding that sox10-derived cells are primarily in the proximal trabecular myocardium of the zebrafish heart suggests that these cells must be migrating into to the wound site after injury . Of course , we cannot rule out the possibility that the cells that reactivate sox10 and the neural crest program may come from another adult lineage . But in the adult , their molecular signature strongly correlates with that of embryonic neural crest cells ( Figure 4D and E ) . Whereas our data clearly show that the sox10+ cells contribute to cardiomyocytes ( Figure 4A ) , whether they also might contribute to other lineages ( e . g . hematopoietic cells ) within the regenerated tissue remains to be explored . Why was the contribution of neural crest cells to cardiomyocytes in amniotes previously missed ? Interspecific quail-chick chimera are generated via transplantation of donor tissue into the host , which requires time to heal ( Kirby et al . , 1983 ) . If the neural crest cells that migrate to the ventricles are the earliest migrating cells , this population may have been delayed after grafting due to wound healing and hence unable to migrate as far . Alternatively , the labeled cells may have been missed since it can be challenging to identify a small population of dispersed quail cells amongst many more numerous chick cells . Furthermore , cell behavior might be altered when transplanted quail cells are introduced into a chick environment . Our retroviral lineage labeling circumvents these issues by indelibly labeling an endogenous neural crest population without the need for grafting . Moreover , the labeled cells are easily detectable due to their fluorescent readout . For lineage labeling in mice , there were hints in the literature regarding a possible neural crest contribution to cardiomyocytes . However , the experiments were either indirect or used lineage tracing techniques that were not specific to the neural crest . For example , Tomita et al . showed that cells isolated from ‘cardiospheres’ can behave like neural crest cells when injected into chick embryos ( Tomita et al . , 2005 ) . In addition , lineage analysis in mouse using a P0-cre line revealed EGFP-positive cells in the myocardium that gather at the ischemic border upon injury ( Tomita et al . , 2005 ) . However , P0 is not a neural crest specific marker , making these results inconclusive at the time . In contrast , Wnt1 is the ‘gold standard’ for neural crest labeling and the improved Wnt1 line ( Wnt1-Cre2+; R26mTmG ) corrects possible ectopic expression problematic in the original Wnt1-Cre;ZsGreen line ( Jiang et al . , 2000; Chai et al . , 2000; Lewis et al . , 2013 ) . In chick and mouse , neural crest-derived cells comprise a significant portion ( ~17% ) of the trabeculated myocardium in the proximal part of both ventricles . Interestingly , this percentage is similar to what has been reported in zebrafish ( Cavanaugh et al . , 2015; Abdul-Wajid et al . , 2018 ) . In amniotes , we find that the density of the cells decreases along the proximal-distal axis and appears to be stable through time ( Figure 1I , Supplementary file 1a , b ) . The presence of neural crest-derived cardiomyocytes across vertebrates and the redeployment of a sox10+ cell population in zebrafish heart regeneration suggest that the neural crest-derived myocardium might also play a role in heart regeneration in neonatal mice , which requires further testing . In summary , the present results show , for the first time , the common ability of cardiac neural crest cells across diverse vertebrates to contribute to heart muscle . Moreover , these cells appear to be critical for cardiac regeneration in zebrafish . If the results extrapolate to other species , the mechanisms that control the normal development of the neural crest into cardiomyocytes may be harnessed to stimulate these cells to proliferate and regenerate new cardiomyocytes , thus offering potential therapeutic approaches to repair heart damage in mammals including humans . Using a standard transfection protocol , chick DF1 cells ( ATCC , Manassas , VA; #CRL-12203 , Lot number 62712171 , RRID:CVCL_0570 , Certificate of Analysis with negative mycoplasma testing at the ATCC website ) were transfected with RIA-H2B-YFP plasmid ( RRID:Addgene_96893 ) and ENV-A plasmid in 15 cm culture dishes . Cell culture medium was collected 24 hr post-transfection , and twice per day for four days , then centrifuged at 26 , 000 rpm for 1 . 5 hr . The supernatant was dried with aspiration , and the pellet was dissolved in 20–30 μl of DMEM to a final titer of 1 × 107 ifu/mL . Viral aliquots were stored in −80°C until the time of injection . Viral stock was diluted 1:2 with Ringer’s solution ( 0 . 9% NaCl , 0 . 042%KCl , 0 . 016%CaCl2 • 2H2O wt/vol , pH7 . 0 ) to generate the working solution , which was mixed with 0 . 3 μl of 2% food dye ( Spectral Colors , Food Blue 002 , C . A . S# 3844-45-9 ) as indicator . The lumen of the neural tube adjacent to the middle of the otic vesicle to the level of somite three was injected with 0 . 2 μl of working in HH8-10 chicken embryos . Embryos were sealed with surgical tap and incubated at 37°C for 1–9 days , harvested at HH14 ( n = 5 ) , HH18 ( n = 5 ) , HH21 ( n = 4 ) , HH25 ( n = 4 ) , HH28 ( n = 12 ) , HH32 ( n = 4 ) and E10 ( n = 4 ) . At the time of harvesting , chick embryos were dissected , fixed in 4%PFA in PBS for 30 mins at 4°C , then embedded in gelatin and sectioned ( Microm HM550 cryostat ) . The Wnt1-Cre; ZsGreenfl/fl mice described in Chai et al . ( 2000 ) ( gift from Drs . Xia Han and Yang Chai at University of Southern California , Center for Craniofacial Molecular Biology ) were harvested and fixed at E15 . 5 ( n = 8 ) and P2 ( n = 2 ) . The hearts were dissected , fixed in 4%PFA in PBS for 30mins at 4°C . E15 . 5 Wnt1-Cre2+; R26mTmG mice ( Lewis et al . , 2013 ) ( 129S4-Tg ( Wnt1-cre ) 1Sor/J , gift from Dr . Jeffrey Bush at University of California , San Francisco , n = 3 ) were fixed with 4% PFA overnight before dissection . The hearts were embedded in gelatin , and sectioned . To quantify RIA-labeled cells in chick embryos , three consecutive sections of the same axial level were imaged per embryo . The number of YFP-positive cells was averaged to account for variability due to sampling . n = 4–6 embryos were analyzed at each stage as biological replicates . The results are presented as presence or absence of virally labeled cardiac neural crest derivatives at different anatomical locations in Figure 1I and as numerical values in Supplementary file 1a , 1b . To quantify Wnt1-Zsgreen+ cells in E15 . 5 mouse heart , three consecutive sections of the same axial level were imaged per embryo ( n = 4 ) . Automated particle analysis was conducted with FIJI program to estimate the total number of Zsgreen+ cells in the image . For the percentage of neural crest-derived cells in the ventricle , the same procedure was performed with the DAPI channel which represents total cell population . % Zsgreen/DAPI was calculated , and averaged to the result presented in the text of Supplementary file 1a . Same analysis was conducted to estimate the number of sox10:eGFP+ cells in 7dpa ( n = 3 ) , 21dpa ( n = 3 ) and sham operated ( n = 3 ) hearts in an area of 2 × 105 μm2 at the apex . One section per heart at the middle of the apex was quantified and presented in Supplementary file 1c . Adult zebrafish heart resection was conducted with the Tg ( −4 . 9sox10:eGFP ) or Tg ( sox10:mRFP ) line , according to published protocols ( Poss et al . , 2002 ) . Resected and sham operated fish hearts ( n = 24 ) were collected at 7 days post injury ( dpi ) ( n = 18 ) , and 21 dpi ( n = 53 ) at which time the fish were euthanized and the hearts were removed for further analysis . The hearts were fixed in 4%PFA in PBS for overnight at 4°C prior to processing for staining . Adult zebrafish were maintained in the Beckman Institute Zebrafish Facility at Caltech , and all animal and embryo work were completed in compliance with California Institute of Technology Institutional Animal Care and Use Committee ( IACUC ) protocol 1764 . After cryosectioning , slides were incubated in 1xPBS at 42°C to remove gelatin . 0 . 3% vol/vol Triton-X100in 1xPBS was used to permeabilize the tissue . Sections were incubated with primary antibody underneath a parafilm layer at 4°C overnight ( primary antibody dilutions: 1:10 Troponin T CT3 , DSHB ( RRID:AB_528495 ) ; 1:10 Myosin Heavy Chain ALD58 , DSHB ( RRID:AB_528361 ) ; 1:10 Myosin Heavy Chain F59 , DSHB ( RRID:AB_528373 ) ; 1:500 Mouse anti-smooth muscle actin , Sigma-Cat# A5228-200uG; 1:500 Mouse anti phospho-histone H3 , Abcam-ab14955; 1:500 rabbit anti caspase-3 , R and D Systems # AF835; 1:500 goat anti GFP , Abcam Cat#ab6673 , all in blocking reagent 1xPBS with: 5% vol/vol normal donkey serum , 0 . 3% vol/vol Triton-X100 ) . Subsequently , sections were washed for 3 times with 1xPBS , incubated with secondary antibody for 40 mins at room temperature and counterstained with DAPI . Secondary antibodies include: Goat anti-mouse IgG2a Alexa-568 ( RRID:AB_2535773 ) , Goat anti-mouse IgG1 Alexa-568 ( RRID:AB_2535766 ) , Goat anti-rabbit IgG Alexa-568 ( RRID:AB_2534121 ) , Donkey anti-goat IgG Alexa-488 ( RRID:AB_2534102 ) ; 1:1000 , Molecular Probes . Zeiss AxioImager . M2 with Apotome . 2 and Zeiss LSM 800 confocal microscope were utilized for imaging . Images were cropped , rotated , and intensity was linearly adjusted for visualization . After fixation , hearts were embedded in paraffin and sections were prepared at 10 µm thickness on a Zeiss microtome . After paraffin removal with histosol , sections were washed and then hybridized with 1 ng/µl anti-sense digoxygenin-labeled probes overnight at 70°C in a humidifying chamber . After hybridization , sections were washed with 50% formamide/50% 1X SSCT buffer followed by washes with MABT and a blocking step in 1% Roche blocking reagent . Sections were then incubated overnight at room temperature with a 1:2000 dilution of anti-DIG-Alkaline Phosphatase antibody ( Roche ) . After several washes with MABT , chromogenic color was developed using NBT/BCIP precipitation ( Roche ) . For each replicate ( n = 2 ) , regenerating ventricles ( n = 12 ) were isolated at 21 days post injury and dissociated into a single cell suspension using a pestle-A tissue homogenizer followed by incubation in Accumax ( Innovative Cell Technologies , Inc ) at 30°C . sox10-mRFP-positive and sox10-mRFP-negative cells were collected by FAC-sorting on a BD Biosciences FACSAriaFusion Cell Sorter . cDNA from mRFP-positive and negative cells was prepared using SMART-seq Ultra Low Input RNA Kit V4 ( Takara ) according to the manufacturer’s protocol . Sequencing libraries were built according to Illumina Standard Protocols and sequenced using an Illumina HiSeq2500 sequencer at the Millard and Muriel Jacobs Genetics and Genomics Laboratory ( California Institute of Technology , Pasadena , CA ) . 50 million , 50 bp , single-ended reads from two biological replicates were mapped to the zebrafish genome ( GRCz10 ) using Bowtie2 ( Langmead and Salzberg , 2012 ) . Transcript counts were calculated using featureCounts ( Subread ) and differential gene expression analysis was performed using DESeq2 ( Liao et al . , 2014; Love et al . , 2014 ) . Protein classification analysis was performed using PANTHER ( Mi et al . , 2019 ) . Heatmaps of normalized counts were generated using Heatmap2 . Databases have been deposited to NCBI ( BioProject # PRJNA526570 ) .	Before birth , unspecialized stem cells go through a process called differentiation to form the many types of cells found in the adult . Neural crest cells are a group of these stem cells found in all animals with backbones ( i . e . vertebrates ) including humans . These cells migrate extensively during development to form many different parts of the body . Due to their contributions to diverse organs and tissues , neural crest cells are very important for healthy development . The heart ventricle is one of the tissues to which neural crest cells contribute during embryonic development in fish and amphibians . However , it was unclear whether this is also the case for birds or mammals or whether neural crest cells have any roles in the regeneration of the adult heart after injury in fish and amphibians . To address these questions , Tang , Martik et al . used cell biology techniques to track neural crest cells in living animals . The experiments revealed that neural crest cells contribute to heart tissue in developing birds and mammals and help repair the heart in adult zebrafish . Further results showed that the contribution of neural crest cells to the heart is controlled by the same genes during both the growth of the embryonic heart and the repair of the adult heart . These results provide new insights into the repair and healing of damaged heart muscle in fish . They also show that similar processes could exist in mammals , including humans , suggesting that activating neural crest cells in the heart could treat damage caused by heart attacks and related conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "short", "report" ]	2019	Cardiac neural crest contributes to cardiomyocytes in amniotes and heart regeneration in zebrafish
The Drosophila midgut is maintained throughout its length by superficially similar , multipotent intestinal stem cells that generate new enterocytes and enteroendocrine cells in response to tissue requirements . We found that the midgut shows striking regional differentiation along its anterior-posterior axis . At least ten distinct subregions differ in cell morphology , physiology and the expression of hundreds of genes with likely tissue functions . Stem cells also vary regionally in behavior and gene expression , suggesting that they contribute to midgut sub-specialization . Clonal analyses showed that stem cells generate progeny located outside their own subregion at only one of six borders tested , suggesting that midgut subregions resemble cellular compartments involved in tissue development . Tumors generated by disrupting Notch signaling arose preferentially in three subregions and tumor cells also appeared to respect regional borders . Thus , apparently similar intestinal stem cells differ regionally in cell production , gene expression and in the ability to spawn tumors . Stem cells allow a tissue to maintain homeostasis by regularly replacing old and damaged cells ( reviewed in Losick et al . , 2011 ) . For example , intestinal stem cells ( ISCs ) of the adult Drosophila midgut ( Ohlstein and Spradling , 2006; Micchelli and Perrimon , 2006 ) generate the two major differentiated cell types of the gut , enterocytes and enteroendocrine cells , throughout life . ISCs with high levels of the Delta ( Dl ) ligand specify daughters to be enterocytes via unidirectional Notch signaling , while ISCs with low Delta levels specify enteroendocrine cell daughters ( Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2006 , 2007 ) . Although , the exact nature of their niche remains unclear , ISCs are highly responsive to signals from neighboring cells . The animal’s nutritional state , as sensed in part by insulin signaling , influences ISC division and whether divisions are symmetric or asymmetric ( Choi et al . , 2011; O’Brien et al . , 2011 ) . Dietary toxins , pathogenic bacteria , or other stressors stimulate enterocytes to release Egfr and JAK/STAT ligands that trigger nearby stem cells to multiply and produce tissue hyperplasia ( reviewed in Jiang and Edgar , 2011 ) . ISCs that follow a generally similar program of enterocyte and enteroendocrine cell production based on Notch signaling were detected along the entire length of the midgut by clonal analysis and marker gene expression ( Ohlstein and Spradling , 2006; Micchelli and Perrimon , 2006; Ohlstein and Spradling , 2007 ) . Whether these stem cells are precisely equivalent , however , has not been critically tested since most studies have focused on the posterior midgut , where ISC divisions are frequent . Some differences are plausible , since histological studies have long distinguished the middle midgut from the anterior and posterior segments . The middle midgut has an acidic pH , and is itself tripartite in organization . The anterior portion contains specialized enterocytes known as interstitial cells ( Filshie et al . , 1971 ) , interspersed with morphologically distinctive ‘copper cells’ , which selectively fluoresce following exposure to copper ions ( Poulson and Bowen , 1952; McNulty et al . , 2001 ) . There follows a zone of ‘large flat cells’ ( Poulson and Waterhouse , 1960 ) and an ‘iron region’ enriched in the iron-storage protein Ferritin ( Poulson and Waterhouse , 1960; Poulson and Bowen , 1952; Mehta et al . , 2009 ) . Copper cells depend uniquely on the homeotic gene labial during embryonic differentiation ( Panganiban et al . , 1990; Hoppler and Bienz , 1994; Dubreuil et al . , 2001 ) and function in acid production using vacuolar H+ ATPase pump proteins localized on their apical membranes ( Dubreuil , 2004; Shanbhag and Tripathi , 2009 ) . A recent study of ISCs in the copper region ( Strand and Micchelli , 2011 ) found that they are capable , like posterior ISCs , of replenishing all the major cell types , including copper , interstitial and enteroendocrine cells . However , copper region ISCs were reported to differ from posterior ISCs in lacking the Notch ligand Delta , and in being normally quiescent in the absence of stress ( Strand and Micchelli , 2011 ) . Thus , the regulation of ISCs differs in the copper region compared to other studied regions of the midgut . The possible existence of regional variation is further suggested by the restricted spatial localization of some digestive enzymes in midguts from a variety of insects ( reviewed by Terra and Ferreira , 1994 ) and from Drosophila larvae . Some enzymes , such as the lipase Magro ( Sieber and Thummel , 2012 ) , may be trafficked into the midgut from the proventriculus via the peritrophic matrix ( King , 1988 ) . Others such as , α-amylase , which is expressed primarily in the anterior and posterior midgut regions ( Thompson et al . , 1992 ) probably indicate true regional differences in enterocyte expression . Some of the strongest evidence for further regionalization comes from studies showing that unique neuropeptides are secreted by enteroendocrine cells located in specific gut regions ( Ohlstein and Spradling , 2006; Veenstra et al . , 2008 ) . These spatial differences in gene expression might be induced downstream of the ISC by region-specific signals , or they might reflect intrinsic differences in regional stem cell programming . Here we document extensive regionalization along the length of the Drosophila midgut , at the level of morphology , cell behavior and gene expression . Each subregion displays a sharp boundary with its neighbors , suggesting that it carries out distinctive functions . ISCs likely contribute to these differences , since stem cells from most tested regions did not produce adjacent region cells even when located at the border . Regional stem cell differences likely influence tumorigenesis , since midgut tumors caused by attenuating Notch signaling arose at very different rates in the different subregions . Thus , tissue stem cells may comprise a wider variety of types , each with a more limited therapeutic scope , than previously appreciated . The Drosophila intestine varies significantly in cellular content and activity based on age , sex , mating status , and nutritional and environmental conditions ( Ohlstein and Spradling , 2006; O’Brien et al . , 2011; reviewed in Jiang and Edgar , 2011 ) . We used stringent animal husbandry strategies to minimize such variation . Only , fertilized adult females 4–14 days of age were employed to avoid the final steps of gut maturation that take place in young adults ( Takashima et al . , 2013a ) , as well as age-induced decline ( O’Brian et al . , 2011 ) . Flies were kept at a controlled density in fresh vials , at 25°C and provided with a uniform level of nutrition before and during the study period . Under these conditions , the cellular structure of the midgut was stable and reproducible as assessed by cell counts along its length ( Figure 1 ) . Our rationale was to understand a ‘steady state’ gut before analyzing the more complex situations where the gut is changing its structure ( O’Brien et al . , 2011 ) . 10 . 7554/eLife . 00886 . 003Figure 1 . The Drosophila midgut is comprised of multiple subregions . ( A ) Drawings are shown to scale of representative enterocytes from each of 10 consecutive regions located along the anterior-posterior ( a/p ) axis of the midgut , as determined from electron microscopy , including parietal cells ( orange ) and interstitial cells ( sky blue ) . Cells enriched in glycogen ( navy blue ) , lipid ( purple or teal ) , or iron ( green ) are shown . ( B ) A midgut from a Fer-GFP protein trap stained anti-GFP ( red ) to highlight the Fe region , and with Nile red ( green ) to highlight lipid-rich regions A2 , P1 , and P3 . Inset: lipid staining in the indicated zones is shown at higher magnification . ( C ) Midgut drawing showing the average number of enterocyte cell diameters along the a/p axis within the 10 subregions ( ± SEM ) . ( D ) Gene expression borders in enterocytes ( ECs , purple ) or enteroendocrine cells ( ees , blue ) at the indicated subregion junctions are totaled for 49 Janelia lines with patterned midgut expression . ( E ) Gene expression borders in midgut–associated muscle relative to subregion junctions are totaled for the eight Janelia lines with patterned circular muscle expression . ( F and G ) Midguts from a R45D10-GAL4; UAS-GFP female ( F ) or R50A12-GAL4; UAS-GFP female ( G ) stained with anti-Cut ( red polyploid Cu cells ) , anti-Prospero ( red diploid cells throughout ) , and anti-GFP antibody shows expression restricted to A1 or to Cu , P2 , P3 and P4 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 00310 . 7554/eLife . 00886 . 004Figure 1—figure supplement 1 . Electron microscopic analysis of midgut enterocytes reveals ten subregions along the anterior-posterior axis ( see diagram above ) . ( A–J ) Representative electron micrographs of enterocytes within the indicated regions are shown . Scale bars are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 00410 . 7554/eLife . 00886 . 005Figure 1—figure supplement 2 . Gene expression patterns of Janelia and other useful Gal4 lines . For each GAL4 line shown on the left , colored bars indicate its pattern and cell type specificity of expression within the ten midgut subregions ( columns ) . Cell type is indicated by color: enterocytes ( EC:purple ) , enteroendocrine cells ( ee:blue ) , intestinal stem cells ( ISC:green ) , enteroblasts ( EB:orange ) , enteric neurons ( N:burgundy ) , circular visceral muscle ( M:grey ) . A continuous line indicates strong expression ( over 50% of cells ) , whereas a dashed line indicates weaker expression ( less than 50% of cells ) . Purple wavy lines indicate parietal cell expression . Not all patterns corresponded to regional boundaries , as indicated by lines that cross their borders . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 005 Under these conditions , we looked for differences in enterocyte morphology along the a/p axis of the Drosophila midgut using light and electron microscopy ( EM ) . EM analyses of longitudinally sectioned midguts revealed many more reproducible regional differences in enterocyte structure than those described previously in the middle midgut , including differences in cell type , cell size , membrane invaginations , microvillar length , and the presence of ferritin particles , glycogen or lipid droplets ( Figure 1—figure supplement 1 ) . Differences were summarized in scale drawings that define 10 distinct zones along the a/p axis ( Figure 1A ) . Furthermore we used nutrient-specific and antibody stains , as well as protein trap lines to validate many of the subregions . Ferritin-GFP selectively stained the Fe region , while Nile red , which stains lipid droplets , marked the A2 , P1 and P3 zones ( Figure 1B ) , the same regions showing enterocytes with high lipid content in EM images . The number of enterocyte cell diameters along the a/p axis within each section was recorded for multiple guts and these proved to be highly reproducible ( Figure 1C ) . We searched for gene expression differences that would mirror these variations in enterocyte structure . Since GAL4 driver lines would be particularly useful for subsequent functional tests , we screened the midgut expression patterns of 931 Gal4 lines from the Janelia Farm collection ( Jenett et al . , 2012 ) and eight additional GAL4 strains . The GFP expression patterns driven by the most useful lines in 5–10 day old adult female midguts ( Figure 1—figure supplement 2 ) are summarized in Figure 1D–E . One example ( Figure 1F ) shows expression only in A1 , while another labels the Cu region , as well as in P2-4 ( Figure 1G ) . By counting the number of enterocytes along the a/p axis , bands of GFP expression were approximately mapped to subregions ( Figure 1F–G , dashed lines ) in each particular cell type ( Figure 1—figure supplement 2 ) . The correspondence of gene expression boundaries in endodermal cells with the morphological junctions of the 10 subregions was striking ( Figure 1D ) . In contrast , gene expression boundaries in gut muscle were mostly offset from these boundaries ( Figure 1E ) . We next investigated whether physiologically meaningful differences in endogenous gene expression occur in the different regions of the midgut . For this purpose , we used appropriate GAL4 lines to drive UAS-GFP fluorescence , and manually isolated specific gut subregions under a fluorescence microscope . Dissected regions were rapidly isolated in small quantities , and total cellular RNA was extracted using a protocol that prevented RNA degradation . RNA samples from 30 separate preparations derived from 10 different single or grouped regions ( Figure 2A ) were analyzed by RNAseq using paired-end reads of 100 × 100 nucleotides and a depth of 30–110 million reads ( Supplementary file 1A ) . 10 . 7554/eLife . 00886 . 006Figure 2 . RNAseq analysis of midgut subregions . ( A ) Schematic of the 10 samples collected . Bars represent isolated portions of gut tissue , either containing single regions ( A1 , Cu , Fe , P1 ) or pooled regions ( A1-3 , Cu-LFC-Fe , P1-4 , A2-3 , LFC-Fe , P2-4 ) . ( B ) Expression profile of select hormones showing regional specificity . ( C ) Expression profile of PGRP isoforms showing regional specificity . ( D ) Regionalized gene expression validates RNAseq method and rules out cross contamination . For each gene ( key at right ) , its mean expression level ( fpkm ) by RNAseq in the three replicate isolated regions ( x axis ) is plotted . CG10725 ( peritrophic membrane ) , Gal ( beta-Galactosidase ) , lab ( homeobox transcription factor ) , CG15570 ( unknown ) , ZIP1 x 0 . 5 ( zinc-iron transporter ) , CG3106 x 0 . 1 ( acyl transferase ) , Ast ( Allatostatin ) . ( E ) Regionalized expression of caudal ( cad ) and two PGRP genes implicated in immune interactions with the midgut microbiome . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 006 The sequence data confirmed that specific regions had been isolated free of significant cross-contamination and revealed a much greater level of regionalized gene expression than previously supposed . For example , the expression of the neuropeptides Npf and Ast-A is patterned within the midgut ( Veenstra et al . , 2008 ) and we observed expression of the corresponding mRNAs within the middle and posterior regions , as predicted , along with patterned expression of many other previously unlocalized neuropeptide mRNAs ( Figure 2B , D ) . The labial ( lab ) gene encodes a homeobox transcription factor that is only expressed in the copper cell region , where it is required for copper cell differentiation ( Panganiban et al . , 1990; Hoppler and Bienz , 1994; Dubreuil et al . , 2001; Strand and Micchelli , 2011 ) . The RNAseq data showed that labial reads were detected only within the middle midgut in experiments that divided the midgut into anterior ( A1 , A2 and A3 ) , middle ( Cu , LFC and Fe ) or posterior ( P1 , P2 , P3 and P4 ) sections . Revealingly , lab fpkm was 40–100 times higher in isolated Cu region RNA than in any other adult midgut region ( Figure 2D ) . All seven of the midgut segments we were able to isolate showed highly specific gene expression . CG10725 was enriched in A1 ( Figure 2D ) , along with many other genes involved in chitin metabolism , where it likely encodes a component of the peritrophic membrane . Although their levels of spatial specificity were not previously known , Beta-galactosidase ( Gal ) mRNA was selectively found in the A2-A3 sample , the ZIP1 iron transporter mRNA in the Fe region , and the CG3106 acyl transferase mRNA in P1 . This high specificity proves that cross-contamination was minimal and that gene expression boundaries correspond to the regional boundaries used in tissue isolation , while the correspondence of the RNAseq data to previous studies of gene expression validates the reliability of this large new body of information . Many genes were expressed in previously unappreciated regional patterns , suggesting they are relevant to diverse aspects of midgut function . For example , PGRPs are membrane proteins that mediate interactions between the midgut and the microbiome that are essential for normal digestion ( Cash et al . , 2006; Ryu et al . , 2008 ) . PGRP-SC1b and PGRP-SC2 expression was found to be highly enriched in the LFC and Fe regions , respectively , along with their regulator , caudal ( Figure 2E ) . The localized expression of these and other PGRP family members ( Figure 2C ) suggest that the microbiome interacts with enterocytes in a highly organized way and that some bacterial species may be localized to specific regions of the gut lumen by the selective expression of host factors . Regionally expressed genes were surprisingly common and specific . Each isolated subregion expressed 50–150 genes 10 times higher than in any other selected subregion of the midgut ( Supplementary file 1B ) . The anterior , middle and posterior zones each expressed 259–474 genes at least five times more highly than in the other two zones , and differentially expressed mRNAs represented some of the most highly expressed genes in each region ( Supplementary file 1B ) . Many differentially expressed genes were annotated as enzymes likely to be relevant to digestion or metabolism , or their expression had previously been mapped specifically to larval or adult midgut by the Fly Atlas microarray project ( Chintapalli et al . , 2007 ) . Interestingly , many regionally expressed midgut genes reside in genomic clusters like highly expressed genes in many other differentiated cells ( Spellman and Rubin , 2002 ) . These included trypsins , lysozymes , serine proteases , alpha-glucosidases , lipases and Jonah genes ( Akam and Carlson , 1985 ) . We carried out gene ontology analyses of the differentially expressed genes and abundant genes from each region and identified candidate biochemical pathways to which they likely contribute using the Database for Annotation , Visualization , and Integrated Discovery ( DAVID ) ( Huang et al . , 2008a , 2008b ) . Figure 3 shows GO terms enriched in the anterior , middle or posterior midgut , as well as in each of the individual regions that were isolated . These findings suggest several general insights into midgut function . Food entering the anterior midgut from the proventriculus begins to break down due to the action of enzymes expressed in the anterior midgut . In addition to digestive enzymes , anterior enterocytes may secrete other substances such as lipids to aid in nutrient solubilization . Antibacterial activity is also important in this region . In the three middle midgut regions food is further digested releasing sugars , amino acids , fatty acids , etc aided by the low pH . Additionally , metal ions that must be reduced before they can be absorbed , and other micronutrients are likely taken up for storage or metabolism . Finally , upon entry into the posterior midgut , nutrients begin to be massively absorbed by enterocytes and transported to other regions of the body for storage , modification or utilization . The lipid-rich zone in the P1 region probably represents such nutrient uptake . The large flux of ready-to-use nutrients in the posterior region may explain why their stem cells divide more rapidly than other regions ( Figure 4S; Ohlstein and Spradling , 2006; Micchelli and Perrimon , 2006 ) . Both the middle and posterior regions likely maintain large populations of commensal microorganisms to aid in carrying out their functions . 10 . 7554/eLife . 00886 . 007Figure 3 . Regional gene expression . Pathways and gene ontology analysis using DAVID ( see ‘Materials and methods’ ) enriched in the anterior ( A1–3 pooled sample ) , middle ( Cu-LFC-Fe pooled sample ) , and posterior ( P1–4 pooled sample ) midgut are listed in the upper panel . Below , a schematic diagram illustrates sequential strategy of digestion suggested by RNAseq analysis . Orange arrows from orange parietal cells in copper region indicate acidification process . Bottom panel indicates pathways and gene ontology analysis using DAVID enriched in subregions . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 00710 . 7554/eLife . 00886 . 008Figure 4 . Stem cells also differ between subregions . ( A–C ) Electron micrographs of an intestinal stem cell from regions: A3—representing typical stem cell morphology— ( A ) , A1 showing its flatter shape ( B ) , P1 showing the presence of lipid droplets . ( D ) Light micrograph of a P1 ISC stained with anti-Delta ( green ) , Nile red ( red ) and DAPI ( blue ) . ( E ) Expression boundaries in two GAL4 lines with ISCs expression . ( F–O ) Light micrograph of ISCs ( green cytoplasm ) , ees ( green nuclear ) , and enterocytes ( polyploid DAPI nuclei ) displaying regional cell ratios . ( F ) A1 , ( G ) A2 , ( H ) A3 , ( I ) Cu , ( J ) LFC , ( K ) Fe , ( L ) P1 , ( M ) P2 , ( N ) P3 , ( O ) P4 , ( P–Q ) Light micrograph of midgut regions stained for Delta ( green ) and for Notch reception activity as assayed by GbeSu ( H ) -LacZ ( red ) . White arrow: Dl+ ISC . Red arrow: GbeSu ( H ) -LacZ+ EB . White asterisk: GbeSu ( H ) -LacZ+ EC . ( P ) A2 region showing recently decided call pairs with Delta staining ( green ) confined to ISCs ( white arrows ) and Notch reporter activity ( red ) in EBs . ( Q ) P1 region stained as in ( P ) , showing that Notch activation ( red ) and Delta expression ( green ) persists downstream from the EB into young ECs . ( R ) Ratio of enteroendocrine cells to ISCs ( blue , top ) and ratio of enterocytes to ISCs ( purple , bottom ) in different midgut regions . ( S ) ISC divisions per day in different midgut regions as determined from clonal analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 008 We analyzed the ISCs within all 10 subregions to look for morphological and gene expression differences that might contribute to regional specialization . ISCs can be recognized in electron micrographs due to their basal location , extensive basement membrane contact , and triangular shape ( Figure 4A ) . In A1 , where enterocytes are more squamous and shorter than in other regions , the stem cells were flatter than the upward-directed triangular shape seen in all other regions ( Figure 4B ) . Additionally , in the most lipid-rich region , P1 , ISCs often contained many lipid droplets and reported lipid synthesis using the SREBP-GAL4 reporter line ( Matthews et al . , 2009 ) ( Figure 4C ) . We verified that these cells were ISCs by staining for Delta and with Nile red ( Figure 4D ) . Lipid droplets were also observed in A2 ISCs but in much fewer numbers ( not shown ) . Two lines uncovered in the screen , R42G03 ( Pdp1 ) and R44E05 ( Stat92E ) , were differentially expressed in ISCs ( Figure 4E; Figure 1—figure supplement 2 ) , and the boundaries of expression coincided with regional boundaries . Thus , stem cells can be regionally distinctive , but it remained unclear if these differences were a cause or effect of other regional features . Studies of stem cell behavior throughout the midgut uncovered further regional differences . The frequency of ISCs was investigated by staining for Delta ( Figure 4F–O ) . Using improved staining conditions ( ‘Materials and methods’ ) , we detected Dl+ cells with the characteristics of ISCs in all 10 regions ( Figure 4F–O ) , including the Cu region ( Figure 4I ) , where Delta expression in ISCs was previously reported to be absent ( Strand and Micchelli , 2011 ) . However , ISCs divided at different rates in different regions based on clonal marking ( Figure 4S ) . While it has been reported that heat shock can alter ISC division rates measured by lineage labeling ( Strand and Micchelli , 2011 ) , we minimized these effects by using a single brief heat shock and a marking system that lacks GAL4 . Moreover , all subregions should have been affected equally , and the data should be accurate as relative measurements . The most rapidly dividing ISCs are found in P1-P3 , which divide about once per day , while ISC division is slightly less frequent in the anterior midgut , and much slower in the middle region . However , in contrast to a previous report ( Strand and Micchelli , 2011 ) , Cu region ISCs were not entirely quiescent but divided regularly every 4–5 days . ISCs in different regions maintain differing numbers of enteroendocrine cells and enterocytes ( Figure 4R ) . Interestingly , in P1 ( and to a lesser extent in P3 and A2 ) , the enteroblast and even the youngest EC continued to stain for Dl ( Figure 4L′ ) . Although ISCs generated the same cells as elsewhere in the midgut , the persistent Delta in enteroblasts in the P1 region likely leads to persistent Notch activation during early enterocyte differentiation; the Notch activity reporter GbeSu ( H ) -lacZ was expressed in young ECs in P1 and P3 , but not in other regions ( Figure 4Q vs 4P ) . If ISCs are intrinsically different in different subregions , then their progeny might only be able to differentiate into cells from that subregion . In contrast , if regional differences are induced by signals received from the gut lumen , muscle , enteric nerves or other sources , then ISCs near a border should on occasion generate daughter cells from both regions . We noticed that clones in the midgut sometimes run for a considerable distance perpendicular to the a/p axis ( Figure 5A ) , but rarely if ever parallel to the axis . To investigate quantitatively whether there are restrictions on cell movement and/or differentiation along the a/p axis , we marked stem cells at a relatively low frequency to avoid generating two adjacent clones derived from different regions . Progeny of individual clones were then analyzed with respect to particular regional boundaries defined using appropriate GAL4 lines or other region-specific markers . Among more than 3000 sparse clones generated throughout the midgut , we identified rare ‘boundary clones’ that contained a single stem cell , at least four total cells and that contacted a marked boundary from one side or the other . Each boundary clone was carefully analyzed by confocal microscopy so that the location of its stem cell ( marked by Delta ) , the number , cell type and location of the downstream cells ( marked by lacZ and DAPI ) and their precise relationship to the boundary in question ( marked by GFP ) could be determined ( Figure 5—figure supplement 1 ) . We then recorded clone size , the number of cells that contacted the boundary , and the number of cells , if any , that crossed the boundary and successfully differentiated into cells of the adjacent region as defined by regional marker expression ( Figure 5B , C ) . It was difficult to obtain large numbers of valid boundary clones due to the low rate of clone induction , regional differences in stem cell activity , and the difficulty of analyzing boundary clones accurately using different regional markers . However , we characterized 49 unambiguous boundary clones contacting six different regional boundaries ( Table 1 ) . 10 . 7554/eLife . 00886 . 009Figure 5 . Stem cells are frequently compartmentalized . ( A ) Two clones from a 25dphs gut showing elongation perpendicular to the a/p axis . ( B–C ) Expected stem cell clonal ( red circles ) distribution near a regional boundary ( green line ) in the absence ( B ) or presence ( C ) of stem cell regional autonomy . ( B ) Regionally multipotent stem cells are predicted to produce progeny on both sides of the boundary . ( C ) Regionally autonomous stem cells generate clones that do not cross the boundary . The diagrams show how boundary clones were scored: boundary cells ( purple outline ) crossed cells ( orange outline ) , enterocyte ( large red circle ) , stem cell ( small red circle ) , enteroendocrine cell ( open red circle ) . ( D–K ) Fluorescence micrographs showing ISC clones ( red ) , region specific EC expression ( green ) , ISCs ( white cytoplasmic ) , ee’s ( white nuclear ) , and DAPI ( blue ) . Green lines: regional boundary determined by region-specific GFP expression ( green ) . White dashed line: outline of clone . White arrow: Isc . ( D and E ) Green: 45D10-Gal4 , UAS-GFP labels A1 . ( D ) Clone originating and remaining in A1 . ( E ) Clone originating and remaining in A2 . ( F and G ) Green: 46B08-Gal4 , UAS-GFP labels Fe . ( F ) Clone originating in the Fe region that crossed the LFC-Fe boundary . ( G ) Clone originating in the LFC region that crossed the LFC-Fe boundary . ( H and I ) Green: Ferritin-GFP labels Fe region . ( H ) Clone originating in and remaining in Fe . ( I ) Clone originating and remaining in P1 . ( J ) Green: 46B08-Gal4 , UAS-GFP labels P1 . Clone originating and remaining in P1 . ( K ) Green: 50A12-Gal4 , UAS-GFP labels P2 . Clone originating and remaining in P2 . ( D′–K′ ) Schematic diagrams of clones in respect to regional boundaries extrapolated from ( D–K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 00910 . 7554/eLife . 00886 . 010Figure 5—figure supplement 1 . Analysis of boundary clones at single cell resolution . The composite image , and the separated channels ( GFP: green , LacZ: red , DAPI; blue , Prospero: white nuclear , Delta: white cytoplasmic ) are shown for the fluorescence micrographs of Figure 5E–J . The white arrow indicates the ISC , white the green line indicates the region boundary , as determined from GFP staining . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 01010 . 7554/eLife . 00886 . 011Table 1 . Clonal analysis of stem cell autonomyDOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 011Region 1 ( ISC ) Region 2ClonesBoundary clonesClones crossedTotal cellsBoundary cellsCrossed cellsp value1NoteA1A2522021100p<0 . 05A2A12504034120p<0 . 01A3Cu863029140p<0 . 01CuA300000n . a . CuLFC94301670p<0 . 05LFCCu2410830n . s . LFCFe6777523723FeLFC235108633625FeP11724026120p<0 . 01P1Fe833302490p<0 . 01P1P25607062280p<0 . 01P2P19525045250p<0 . 012 dying The A1/A2 boundary was typical of most that were studied . Boundary clones that resulted from stem cell labeling within A1 ( Figure 5D , D′ ) or A2 ( Figure 5E , E′ ) contacted the border zone defined by R45D10::UAS-GFP expression , but never included cells from the opposite side . Either the clones did not cross the boundary , or they crossed but were unable to differentiate properly and rapidly died . Since the boundaries are not smooth or precisely perpendicular to the a/p axis at the cellular level , it is also possible that cells from an A1 stem cell that pushed into the anterior edge of the A2 zone simply caused a bulge in the boundary at that location ( see Figure 5D , D′ ) . Over time such cell movement might cause the boundary between the two regions to fluctuate slightly in shape as cells die and are replaced by cells from one side or the other . Similarly , clones failed to differentiate heterologously across four other studied boundaries ( Table 1 ) , including Fe/P1 ( Figure 5H , I ) and P1/P2 ( Figure 5J , K ) . In P1/P2 , we verified clonal behavior using more than one boundary marker ( Figure 5J , K ) . Clones readily crossed one boundary , between LFC and Fe , and differentiated or transformed into cells characteristic of the opposite side . We induced ISC clones and identified 17 boundary clones touching the LFC/Fe border , using Fe region-specific expression of Ferritin-GFP to mark the boundary . The behavior of clones at this boundary was very different than at A1/A2 . 15 of 17 clones crossed the boundary , and did so readily from either direction ( Figure 5F , G ) . This was not an artifact of Ferritin-GFP expression because crossing was also observed using R46B08 to mark the LFC/Fe boundary , whereas clones failed to cross from Fe into P1 using Ferritin-GFP . Thus , despite the regional differences in enterocyte morphology and gene expression described earlier , the stem cells within LFC and Fe can each generate at least some cells characteristic of the other region . We concluded that our method efficiently identified clones that produced cells from two different regions across a boundary when such behavior occurred . Since the average size and border contact of boundary clones was similar in all the regions ( Table 1 ) , the behavior of clones in the other regions could be compared quantitatively to clones at the LFC/Fe boundary to determine if their failure to differentiate across a boundary was statistically significant . For example , 88% of boundary clones on the LFC/Fe boundary crossed and generated cells on the other side ( N = 17 ) . In comparison , 0% of boundary clones crossed the A1/A2 border ( N = 6 ) , 0% of boundary clones crossed the A3/Cu border ( N = 3 ) , 0% of boundary clones crossed the Cu/LFC border ( N = 4 ) , 0% of boundary clones crossed the Fe/P1 border ( N = 7 ) , and 0% of boundary clones crossed the P1/P2 border ( N = 12 ) . These differences are all statistically significant , since the probability that even three boundary clones would fail to cross a junction equivalent to the LFC/Fe boundary is only ( 1–0 . 88 ) 3 = 0 . 0017 . To further test the significance of the results , we reasoned that there should be a correspondence between the number of boundary cells and the number of boundary-crossing cells , on average , in the absence of a constraint . In boundary clones at the LFC/Fe border , there were 0 . 63 crossing cells for every boundary cell ( N = 115 ) . In contrast , there were 0 crossing cells for 22 boundary cells at the A1/A2 border , 0 crossing cells for 14 boundary cells on the A3/Cu border , 0 crossing cells for 10 boundary cells on the Cu/LFC border , 0 crossing cells for 21 boundary cells on the Fe/P1 border , and 0 crossing cells for 53 boundary cells on the P1/P2 border . The absence of crossing cells differs significantly from expectation based on LFC/Fe ( e . g . , at the Cu/LFC border with ten boundary cells , p=0 . 014; χ2 = 6 . 15 ) . However , the limited number of clones and boundary cells that could be obtained at the Cu/LFC and A3/Cu borders does limit our ability to detect cell crossing is inhibited in both directions . At the Cu/LFC border , Cu stem cell progeny cannot cross , but it is not possible to conclude that LFC progeny lack this capacity . The restriction at the A3/Cu border also is known to apply only from the A3 side . A summary of the compartmentalization of stem cell clones is shown in Figure 6A . 10 . 7554/eLife . 00886 . 012Figure 6 . Midgut subregions differ in the production and movement of ISC/enteroendocrine tumor cells . ( A ) Summary of barriers to cross-regional stem cell differentiation . The results of testing whether stem cell clones are able to span six of the midgut regional boundaries are indicated graphically . Blue arrows indicate that stem cell progeny could cross the indicated boundary in the direction shown , and differentiate into regionally appropriate cells . Red inhibitor symbols indicate that clonal crossing and differentiation from the indicated directions were not observed . Maroon shading shows the regions of the midgut in which enteroendocrine tumors arise preferentially when Notch signaling is inhibited . ( B ) Low magnification view of the entire midgut from an animal expressing UAS-NRNAi driven by esg-GAL4 showing accumulation of tumor cells in regions A2 , P1 and P3 . ( Green: UAS-GFP expression , Red cytoplasmic: Delta expression in ISCs , Red nuclear: Prospero expression in ee’s , Blue: DAPI in all nuclei , White lines: regional boundaries ) . ( C ) Control expression in the Fe and P1 regions in the absence of UAS-NRNAi . ( D and E ) Higher magnification of a midgut as in ( B ) ee and ISC-like cells are evident in P1 but not in Fe; ( F–H ) tumor cells induced by esg-GAL4 driven expression of Notch[DN] appear to respect regional borders between P1/P2 ( F ) , P2/P3 ( G ) , and P3/P4 ( H ) . Markers are as in ( B–E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00886 . 012 Disrupting Notch signaling in ISCs leads to enteroendocrine tumor formation . We investigated whether these tumors arise with the same frequency throughout the midgut by knocking down Notch signaling in ISCs and EBs using the esg-GAL4 driver , or by feeding flies the Notch inhibitor DAPT . Several days after reducing Notch activity , clusters of diploid tumor cells began to appear in the tissue . Surprisingly , tumorous cells arose and proliferated much more extensively in some midgut subregions than in others ( Figure 6B ) . The P1 region consistently had the largest tumors and earliest onset , although tumors also arose rapidly in P3 and the posterior part of A2 . In contrast , we never observed tumors within A1 , even after long periods of exposure to these agents . As the tumor cells expanded they appeared to respect the cellular boundaries with adjacent midgut regions . For example , the anterior border of P1 tumor cells corresponded closely with the Fe/P1 boundary based on counts of cell diameters from the Cu region ( Figure 6D–E ) . The posterior boundary of P1 tumor expansion mapped near or at the P1/P2 border ( Figure 6F ) , while the anterior and posterior boundaries of P3 tumors also corresponded with their normal junctions with P2 ( Figure 6G ) and P3/P4 ( Figure 6H ) . The same regional specificity was observed when Notch signaling was inhibited using N[RNAi] , treatment with the Notch inhibitor DAPT , or by expressing a Notch-dominative negative allele NDN . These results indicate that regional epigenetic differences have a large effect on the development of midgut enteroendocrine tumors and that regional boundaries restrict tumor cell movement into adjacent tissue compartments . Our experiments significantly expand previous knowledge of regional variation within the Drosophila midgut . At the levels of cell morphology , cell behavior and gene expression , the midgut is much more highly organized than a uniform cellular tube containing an acidic middle region or ‘stomach’ . Regionalization likely supports the complex metabolic tasks carried out by the midgut . Ingested food goes through multiple intermediate stages during digestion and these steps may be most efficient if carried out in a controlled sequence ( Figure 3 ) . Some of these steps are associated with an array of bacterial species that constitute the normal intestinal microbiome ( Ryu et al . , 2008; Shin et al . , 2011 ) . Our work will allow the function of genes , pathways , cells and regions within the midgut to be tested in digestion , tissue maintenance , microbiome function and immunity . After this work was submitted for publication , Buchon et al . ( 2013 ) also described midgut regionalization; they classified five major midgut divisions ( R1–R5 ) , 13 total subregions ( e . g . , R1a and R1b ) , and 2 microregions ( BR2–3 , BR3–4 ) . Converting our counts of cell number within the 10 regions ( Figure 1C ) to the fractional length coordinates used by Buchon et al . ( 2013 ) suggests a close correspondence between the studies: A1 = R1a + R1b; A2 = R2a + R2b; A3 = R2c; Cu = R3a + R3b; LFC = R3c; Fe = BR3–R4 , P1 = R4a + R4b; P2 = R4c; P3 = R5a; P4 = R5b . In the case of single regions in our system that Buchon et al . split in two , we had also usually noted differences in gene expression; for example , between the anterior and posterior Cu region ( Figure 1D , Figure 1—figure supplement 2; Strand and Micchelli , 2011; Buchon et al . , 2013 ) or the subregions of A2 and P1 that differ in lipid accumulation . We emphasized enterocyte morphology and qualitative gene expression differences in defining subregions . In contrast to our study , Buchon et al . ( 2013 ) view constrictions to be of primary importance and define two constrictions as micro-regions in their own right . We found that enterocyte subregions abut directly , and favor the view that constrictions are mesodermal features , not microregions . Our results argue strongly that the striking regionalization of structure and gene expression within the midgut is maintained at least in part by regional differences between their resident stem cells . In the midgut subregions surrounding five different boundaries , we did not detect a single stem cell that produced differentiated cells on the opposite side of the boundary , that is from a region different from the one in which it resided . All the ‘non-crossing’ clones contacted the regional boundary , and in 78% the founder stem cell was located on or within one cell of the boundary , such that progeny cells could have reached the adjacent region prior to differentiation . In contrast , boundary clones with the same general properties almost always crossed the LFC/Fe boundary , showing that ‘non-crossing’ behavior would not occur by chance . Consistent with the existence of epigenetic differences in stem cells that limit trans-regional differentiation , clones frequently pushed into adjacent regions at the boundary , but retained their autonomous identity based on marker expression . Mechanical and/or adhesive forces may also contribute to maintaining some regional boundaries . Indeed , the tendency of tumor cells to respect regional boundaries suggests that cell–cell interactions at boundaries are likely to be important as well as stem cell programming . The behavior of ISC clones at regional boundaries is reminiscent of the behavior of clones in developing imaginal discs at ‘compartment’ boundaries ( reviewed in Dahmann et al . , 2011 ) . In the developing Drosophila embryo and imaginal discs , the engrailed gene and hedgehog signaling play important roles in defining posterior compartments ( Dahmann and Basler , 2000 ) . We did not detect expression of engrailed or the closely related gene invected in any midgut region , and expression of hedgehog pathway components was similar on both sides of non-crossing boundaries such as Fe/P1 . Dorsal ventral compartments in the developing wing are mediated by apterous and by Notch signaling ( Milan and Cohen , 2003 ) . Apterous was expressed only at very low levels throughout all regions and Serrate was found at significant levels only in posterior regions 2–4 where the gene is dispensable for cell differentiation ( Ohlstein and Spradling , 2007 ) . In the developing vertebrate brain , Hox genes are important in specifying developmental compartments ( Kiecker and Lumsden , 2005 ) . However , Hox genes are only expressed at very low levels in endodermal cells during embryogenesis ( Hoppler and Beinz , 1994 ) and these genes were very weakly expressed in our RNAseq studies , perhaps due to expression in non-endodermal cells within these samples . Consequently , the genetic basis for adult midgut compartmentalization probably differs from previously studied examples of tissue regionalization . The homeotic transcription factor labial ( lab ) is an outstanding candidate for a regional regulatory factor . In the embryonic and larval gut , lab is required for Cu cell specification , differentiation and maintenance ( Hoppler and Beinz , 1994; Dubreuil et al . , 2001 ) . The gene is expressed in copper cells , but not elsewhere in the larval midgut , and we observed similar specificity of lab expression in the adult . When lab is mis-expressed during embryonic development in other midgut regions , the copper region can expand ( Hoppler and Bienz , 1994 ) . Endodermal cell identity along the a/p axis may be determined by signals from adjacent mesoderm during embryogenesis ( Bilder and Scott , 1998 ) , and then fixed by the induction of secondary factors such as lab . Gene expression within the midgut muscles might play a similar role in the adult midgut , however , expression boundaries of muscle genes were frequently offset with respect to endodermal regions ( Figure 1E ) . Whether this bears any relationship to the documented offset in homeotic gene expression between the ectoderm and visceral endoderm in embryonic development ( Lawrence and Morata , 1994 ) remains unclear . A key question is whether individual or combinations of differentiation regulators analogous to lab specify other midgut subregions in which the ISCs fail to generate cells across regional boundaries . The RNAseq data should provide a valuable resource in identifying such factors . For example , one potential candidate , the homeotic gene defective proventriculus ( dve ) , functions in copper cells ( Nakagawa et al . , 2011 ) and its expression was observed to fall sixfold between Fe and P1 . One remaining question is whether a pre-existing pattern of larval midgut subdivision plays any role in the origin of adult midgut organization . The larval gut has a middle acidic region containing copper cells and an iron region like the adult tissue ( Poulson and Bowen , 1952 ) , and EM studies show additional morphological differences ( Shanbhag and Tripathi , 2009 ) . However , it is not known whether regions analogous to other eight midgut domains described here exist in larvae . The larval midgut contains nests of diploid intestinal precursors that proliferate following pupariation to build the adult gut and establish its ISCs ( Jiang and Edgar , 2009; Mathur et al . , 2010 ) . Larval midgut domains might serve as a template for adult regionalization if gut precursor cells within each region already differ autonomously and do not mix during pupal development . However , cells do cross boundaries between the hindgut and midgut during pupal gut development ( Takashima et al . , 2013b ) . Identical regionalization within larval and adult guts might be disadvantageous to species with very different larval and adult diets , hence many adult midgut regions are likely to be established de novo or to be re-specified during pupal development . Many mammalian tissues such as skin , muscle , lung , liver , and intestine contain thousands of spatially dispersed stem cells , like the Drosophila midgut . Our studies raise the question of whether these tissues exhibit finer grained regional patterns of gene expression than has been previously recognized , patterns that might be supported by small autonomous differences in their stem cells . Currently , the strongest indication for such regionalization comes from studies of the intestine . Lineage labeling shows that similar stem cells expressing Lgr5 exist along the mammalian gut ( Barker et al . , 2007 ) despite the fact that enterocytes , enteroendocrine cells and bacterial symbionts differ regionally ( Rindi et al . , 2004; Bradley et al . , 2011 ) . For example , iron absorption in mammals takes place primarily in the duodenum ( Fuqua et al . , 2012 ) , a specialized subregion of the small intestine located just downstream from the acidic stomach . This is similar to the position of the midgut iron region just downstream from the acid-producing parietal cells of the Cu and LFC regions . The antibacterial lectin RegIIIγ , which like Drosophila PGRPs recognizes bacterial peptidoglycans , is expressed most prominently in the distal region of the small intestine ( Cash et al . , 2006 ) . The existence of tissue and stem cell regionalization in other mammalian tissues deserves further detailed investigation . The human large intestine is much more prone to cancer than the small intestine . Our studies suggest that regional differences in the properties of apparently similar stem cells and tissue cells contribute to such differences . The midgut zones most favorable for the expansion of Notch-deficient cells showed pre-existing differences in Notch signaling within the early enterocyte lineage . Delta expression did not decrease shortly after ISC division , as in other regions , and Notch signaling persisted throughout enterocyte development ( Figure 4Q ) . Curiously , same tumor-prone regions with persistent Notch signaling also were enriched in lipid droplets ( Figure 1 ) . At present it is not clear how the altered signaling , regional metabolic activity and tumor susceptibility are related . Additionally , regional differences in the microbiome , as suggested by our observation of domain-specific expression of PGRP proteins , may also influence the occurrence of cancer . Gastric bacteria such as Helicobacter pylori contribute to stomach cancer ( Uemura et al . , 2001 ) , while colonic Bacteroides fragilis likely promote gut cell DNA damage and colon cancer ( Wu et al . , 2009 ) . Regional tissue differences likely also affect rates of tumor progression and metastasis . These observations emphasize the importance of understanding tissues region by region . In sum , the Drosophila midgut provides an outstanding tissue in which to explore and understand the significance of intrinsic stem cell differences . We identified GAL4 drivers that allow gene expression to be manipulated in all intestinal cell types , including cells such as circular muscle and enteric neurons that are thought to contribute to niche function . Will altering the expression of genes that normally differ between regions cause ISCs to generate cells with heterotypic characteristics ? Such studies might eventually make it possible to stimulate medically useful responses from the endogenous stem cells that remain within a diseased tissue . Strains of Drosophila melanogaster were obtained from Bloomington Stock Center ( http://flystocks . bio . indiana . edu/ ) unless otherwise indicated . Genetic symbols are described in Flybase ( McQuilton et al . , 2011 ) . Flies for all experiments were maintained either at a constant 25°C or between 23–25°C with 15–25 flies per vial and tossed onto new food every 2–3 days to ensure that midguts behaved in a reproducible manner that was minimally influenced by external variables . Janelia lines ( Jenett et al . , 2012 ) were obtained from JFRA; further information about each line is available online ( www . janelia . org/gal4-gen1 ) . Sources of other strains were: esg-GAL4;tub-GAL80ts , UAS-GFP ( B Edgar ) ; UAS-NotchRNAi ( S Bray ) ; UAS-Notch Dominant Negative ( UAS-NDN ) ( I Rebay ) , and SREBP-reporter line ( R Rawson; Matthews et al . , 2009 ) . We dissected the gut tissue in Grace’s buffer and fixed on a nutator shaker for 1 hr at room temp ( or 16–20 hr at 4°C for Delta staining ) in 4% paraformaldehyde ( Electron Microscopy Sciences , Hatfield , PA; 32% paraformaldehyde , EM grade , cat:15 , 714 ) and 2% antibody wash ( 0 . 3%TritonX and 0 . 3% of 30%BSA in 1XPBS ) in Grace’s buffer . After fixation , guts were washed in antibody wash with continuous shaking three times for at least 20 min followed by primary and secondary antibody application for at least 12 hr each at 4°C with at least 12 hr of washing with antibody wash in between . After the secondary antibody incubation , samples were washed in antibody wash three times for at least 20 min . Finally , nuclei were stained with DAPI ( 1:5000 of Sigma , St . Louis , MO; cat:D 9542 ) diluted in 1XPBS for 30 min with shaking . Primary antibodies: mouse-α-Prospero ( 1:20 ) , mouse-α-Delta ( 1:100 ) , and mouse-α-Cut ( 1:20 ) from Developmental Studies Hybridoma Bank ( http://dshb . biology . uiowa . edu/ ) , chick-α-βGal from Abcam ( Cambridge , MA; ab9361 ) ; and rabbit-α-GFP ( 1:1000 ) from Invitrogen ( cat:A11122 ) . Secondary antibodies from Invitrogen ( Burlingame , CA ) all used at 1:500: goat-α-mouse AlexaFluor488 ( A11001 ) , goat-α-mouse AlexaFluor568 ( A11004 ) , goat-α-mouse AlexaFluor633 ( A21052 ) , goat-α-rabbit AlexaFluor488 ( A11034 ) , goat-α-chick AlexaFluor568 ( A11041 ) . Nile red ( 72 , 485; Sigma ) and Prussian blue ( 03 , 899; Sigma ) were applied at the secondary antibody wash step , and utilized at 1:5000 . Preparations were mounted in Vectashield ( Vector Labs Inc . cat:H-1000 ) , and examined with a LeicaSP5 confocal microscope . 931 lines from the Janelia GAL4 enhancer trap collection were crossed to the UAS-RFP , UAS-FLP , Ubip63–FRT-stop-FRT-nEGFP/CyO ( Evans et al . , 2009 ) and guts were dissected from at least three progeny females aged 3–7 days from each strain . Guts were fixed , stained to reveal GFP , Prospero and Delta , and examined in a confocal microscope . We found 63 lines with expression in the adult proventriculus , midgut epithelium , midgut muscle , and/or enteric nervous system . The expression of these lines were analyzed further by crossing to UAS-GFP . S65T ( Bloomington 1522 ) and mated female flies were reared at 25°C with 15–25 flies per vial . Flies were transferred onto new food every 2–3 days and fed with dry yeast 2 days prior to dissection ( between 6–10 days post eclosion ) . At least 10 guts were analyzed from each line to document expression . Figure 1—figure supplement 2 presents the spatial pattern and cell types expressing GFP in those lines with midgut expression . Eight strains with region-specific GFP expression , R50A12-GAL4 , R45D10-GAL4 , R46B08-GAL4 , R42C06-GAL4 , R43D03-GAL4 , R47G08-GAL4 , R40B12-GAL4 and Ferritin-GFP ( Morin et al . , 2001 ) , were crossed to UAS-GFP . S65T ( Bloomington 1522 ) and progeny flies were reared at 25°C for 6–10 days . Mated females fed on dry yeast for 2 days prior to collection were dissected in Grace’s buffer and appropriate midgut regions were isolated under a fluorescent dissecting microscope . At least five samples of each region were fixed for 1 hr at 4°C in 1% gluteraldehyde , 1% OsO4 , 0 . 1 M cacodylate buffer , 2 mM Ca ( pH 7 . 5 ) . Following washing in cacodylate buffer ( 3X × 5 min ) tissue was embedded in agarose at 55°C , rinsed for 5 min in 0 . 05 M maleate ( pH 6 . 5 ) , and stained for 1 . 5 hr in 0 . 5% uranyl acetate , 0 . 05 M maleate pH 6 . 5 . After rinsing in H2O , samples were dehydrated in an ethanol series ( 35% twice for 5 min , 50% for 10 min , 75% for 10 min , 95% for 10 min and 100% three times for 10 min ) , incubated in propylene oxide ( 2X × 10 min ) and then in 1:1 propylene oxide:resin ( Epon 812:Quetol 651 [2:1] ) , 1% silicone 200 , 2% BDMA for 1 hr . After three changes of resin ( 1 hr each ) , resin was allowed to polymerize overnight at 50°C and then at 70°C overnight . Images were captured with a Phillips Tecnai 12 microscope and recorded with a GATAN multiscan CCD camera using Digital Micrograph software . We analyzed at least three longitudinal sections from each of the 10 proposed regions . The strains Hs-FLP;X15-29 and w;X15-33 that in combination allow lineage marking ( see Ohlstein and Spradling , 2006 ) were crossed to UAS-GFP and a regional-GAL4 line , respectively , and their F1 progeny were intercrossed to generate the starting strains used . For regional proliferation rate studies , mated females 4 days post eclosion with the following genotype: Hs-FLP , UAS-GFP;X15-33/X15-29; regional-GAL4 , UAS-GFP/+ or Hs-FLP;X15-33/X15-29; Fer-GFP/+ , were heat shocked for 30 min at 37°C and dissected at multiple times thereafter . In each case , midguts were stained to reveal lacZ ( clonal marker ) , Dl ( to identify ISCs ) and Prospero ( to identify enteroendocrine cells ) , and the size and cellular content of ISC clones was analyzed . The ability of stem cell progeny to differentiate into cells from an adjacent subregion ( ‘boundary crossing’ ) was analyzed using the same flies and experimental design , except that flies were heat shocked for 10 min at 37°C ( to ensure the clonal induction rate was low enough such that even large clones rarely touched nearby clones ) and dissected at multiple time points up to 25 days after heat shock . Regional boundaries were marked with region-specific protein trap lines , GAL4-driver lines , or antibody stains . The A1/A2 boundary was indicated by R45D10 ( A1 expression ) , A3/Cu by anti-Cut antibody or R50A12 ( Cu parietal cell expression ) , Cu/LFC by anti-Cut antibody or R50A12 ( Cu parietal cell expression ) , LFC/Fe by Ferritin-GFP or R46B08 ( Fe expression ) , P1/P2 by R46B08 ( P1 expression ) or R50A12 ( P2 expression ) , and P2/P3 by R50A12 ( P2 expression ) . Clones containing a single ISC , at least four total cells , and contained at least one cell that contacted the boundary under study were termed ‘boundary clones’ and were analyzed further to determine the number and location of all cells relative to the boundary . A ‘boundary cell’ is any cell in the clone that neighbors midgut cell in a different subregion . A ‘crossed cell’ is any cell in the clone that is located past the boundary and that displays the marker expression of a midgut subregion different from that of its founder stem cell . A crossed clone is a clone that contains at least one crossed cell . Several classes of potentially ambiguous clones were not included in our tabulation to ensure accuracy . Clones lacking a single , clearly stained stem cell were excluded , as were clones that could not be accurately scored because they wrapped around the periphery of the squashed tissue . Additionally , clones touching regions where boundary marker expression could not be clearly scored were excluded . For each boundary clone , the total size , number of boundary-touching cells , number of crossed cells , if any , and the location of the ISC with respect to the border , was recorded . To perturb the Notch pathway we crossed UAS-N[RNAi} , and UAS-NDN to yw;esg-Gal4/CyO;tub-Gal80ts , UAS-GFP/TM3 at 18°C to restrict transgene expression during larval and pupal stages . UAS-N[RNAi];esg-GAL4/+;tubulin-GAL80ts , UAS-GFP/+ , yw;esg-GAL4/UAS-Nact;tubulin-GAL80ts , UAS-GFP/+ , and yw;esg-GAL4/UAS-NDN;tub-GAL80ts , UAS-GFP/+ flies were moved to 29°C ( permissive temperature for GAL4-mediated knockdown ) upon eclosion , and dissected at multiple time points ( no later than 16 days ) . Boundary limits were determined by counting cell diameters from the cut+ copper cells , and by gut morphology . We identified GAL4 lines with regional expression , then dissected and sectioned gut regions based on the presence or absence of GFP under a fluorescent dissecting microscope . Flies were maintained at 22–25°C for 6–8 days post eclosion , and given dry yeast for the 2 days prior to dissection . Gut regions were isolated , no more than five at a time , and moved to iced Tripure reagent ( Roche/Boehringer Mannheim cat: 11667157001 ) to avoid RNA degradation ( which occurred unless guts were processed within 30 min of isolation ( or 10–15 min in the case of the copper region samples ) . After 25 or 50 gut regions were collected in 200 μl Tripure , they were homogenized . 600 μl of fresh Tripure was added , mixed , and allowed to stand at room temperature for 5–10 min . After adding 180 μl chloroform , samples were vortexed 2 × 45 s , and incubated at room temperature for 10 min . Samples were then centrifuged for 15 min at 12 , 000 rpm at 4°C , and the aqueous layer was moved to a fresh tube . RNA was precipitated by adding 400 μl of isopropanol , vortexing for 15 s , and incubating at room temperature for 15 min , and centrifuging at 12 , 000 rpm for 15 min at 4°C . After washing the pellet with 75% EtOH , the RNA was dried in air dry for 5 min , re-suspended in 50 μl nuclease free H2O , and stored at −80°C . cDNA libraries were constructed from poly ( A ) -selected RNA using Illumina TruSeq RNA Library Prep Kit v2 , and sequenced using a HiSeq2000 . Fpkm values for genes in all samples were calculated using Bowtie2 v 2 . 0 . 6 , TopHat v 2 . 0 . 7 , and Cufflinks version 2 . 02 , with Refseq annotation file dm3 . Triplicate samples independently prepared and analyzed from each region gave fpkm values between 50 , 000 and 0 . 1 for about 10 , 000 of 15 , 600 annotated features . Expression values were highly reproducible between replicates ( R2 > 0 . 95 ) , except for a small subclass of outlier genes , including many RNA genes . Genes in which the fpkm standard deviation ( SD ) /mean >1 within the replicates were excluded , since analyses of reads suggested that the divergence in such cases was usually due to inconsistent read alignment in the presence of nearby or overlapping genes . A small subset of genes showed possible physiological variation between replicates . Gene ontology analysis based on average fpkm was carried out for differentially expressed and for highly expressed gene sets within each subregion ( Supplementary file 1B ) using DAVID software ( Huang et al . , 2008a , 2008b ) . These data are available at the NIH Geo Website under accession GSE47780 .	Many cells in the body accumulate wear and tear over time , and a fraction of them are always nearing the end of their lives . However , in some tissues there are stem cells that can divide into daughter cells which then differentiate and replace the damaged cells . Unlike embryonic stem cells , these ‘adult tissue stem cells’ normally differentiate into only a few related cell types , but their ability to produce replacement cells keeps the tissue functioning normally . Here , Marianes and Spradling have investigated a type of adult stem cell , known as intestinal stem cells , that resides in the midgut of fruit flies . The midgut is the major site of digestion in fruit flies , and functions much like the small intestine in mammals . This tissue is a long tube that is lined with two types of cells: digestive cells and hormone-producing cells . These cell types are maintained by thousands of apparently similar intestinal stem cells , and it has long been thought that the stem cells give rise to cells throughout the midgut by responding to the same set of signals . However , certain digestive processes—such as the breakdown or uptake of particular nutrients—are known to occur only in a specific portion of the intestine . For example , in fruit flies , a region in the middle of the intestine is acidified , and may act like an extra stomach . And in both fruit flies and mammals , iron is taken up mostly in the area of the gut just after the stomach . These regional differences in function have led to uncertainty over how midgut cells both arise and are replaced . Marianes and Spradling now show , based on a detailed study of tissue cells and stem cells , that the midgut contains at least ten subregions that occur in a specific order . The cells in these subregions have distinct features , including shape , size and contents ( e . g . , stores of carbohydrates or nutrients ) . Each subregion appears to perform specific functions during digestion , and the cells in these subregions also transcribe genes that reflect their roles in breaking down or storing various nutrients . Interestingly , the stem cells in most subregions are distinct , and do not differentiate into the cells from adjacent subregions . The subregions also differ in their incidence of cancer: when a particular signal was inhibited in stem cells in all ten subregions , aggressive tumors formed in only three subregions and the tumor cells did not cross into neighboring subregions . These observations may inform future studies of the mammalian small intestine and improve our understanding of its susceptibility to cancer and other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2013	Physiological and stem cell compartmentalization within the Drosophila midgut
Passage through mitosis is driven by precisely-timed changes in transcriptional regulation and protein degradation . However , the importance of translational regulation during mitosis remains poorly understood . Here , using ribosome profiling , we find both a global translational repression and identified ∼200 mRNAs that undergo specific translational regulation at mitotic entry . In contrast , few changes in mRNA abundance are observed , indicating that regulation of translation is the primary mechanism of modulating protein expression during mitosis . Interestingly , 91% of the mRNAs that undergo gene-specific regulation in mitosis are translationally repressed , rather than activated . One of the most pronounced translationally-repressed genes is Emi1 , an inhibitor of the anaphase promoting complex ( APC ) which is degraded during mitosis . We show that full APC activation requires translational repression of Emi1 in addition to its degradation . These results identify gene-specific translational repression as a means of controlling the mitotic proteome , which may complement post-translational mechanisms for inactivating protein function . Genome-wide microarray , RNA sequencing ( RNA-seq ) , and protein-based mass spectrometry studies have revealed changes in the abundance of hundreds of proteins during the cell cycle ( Cho et al . , 2001; Whitfield et al . , 2002; Aviner et al . , 2013; Grant et al . , 2013; Lane et al . , 2013; Stumpf et al . , 2013; Ly et al . , 2014 ) , many of which are specifically expressed in G2 phase and mitosis ( G2/M ) . Transcriptional regulation plays an important role in this temporal expression pattern , as many genes show cell cycle-stage specific expression of their mRNA level ( Cho et al . , 2001; Whitfield et al . , 2002 ) . Regulated protein degradation also plays a key role in sculpting the proteome during the cell cycle , particularly at the end of mitosis when a large set of proteins is ubiquitinated by the E3 ubiquitin ligase , the Anaphase Promoting Complex ( APC ) , and then degraded by the proteasome ( Peters , 2006 ) . In addition to regulation of transcription and protein degradation , changes in translation efficiency ( TE ) also can modify protein abundance during the cell cycle . Translational regulation plays a particularly important role during the specialized cell division cycles of meiosis and early embryonic development , since transcription is largely silent at this stage ( Mendez and Richter , 2001; Groisman et al . , 2002; Tadros and Lipshitz , 2009; Weill et al . , 2012 ) . For example , initial work on meiosis in vertebrate oocytes revealed that lengthening of the poly ( A ) tails of dormant mRNAs through polyadenylation results in translational activation , which plays a critical role in meiotic progression ( Weill et al . , 2012; Subtelny et al . , 2014 ) . While somatic cells appear to use transcriptional regulation as a major mechanism for modulating protein expression during the cell cycle , translational regulation has been described in somatic cells as well ( Sivan and Elroy-Stein , 2008; Novoa et al . , 2010; Kronja and Orr-Weaver , 2011; Aviner et al . , 2013; Stumpf et al . , 2013 ) . As in meiosis , changes in the poly ( A ) tail length can occur during G2 phase of the mitotic cell cycle ( Novoa et al . , 2010 ) , but these effects appear to be less important for translational regulation during the somatic cell cycle than during early embryonic development ( Subtelny et al . , 2014 ) . Perhaps the most striking example of translation regulation in the somatic cell cycle occurs during mitosis , when translation is thought to be globally repressed ( Fan and Penman , 1970 ) . This global translational repression appears to be sequence-independent , and likely affects most mRNAs ( Fan and Penman , 1970 ) . While such global translational repression was observed many decades ago , its functional significance remains unknown . Several studies suggested that this global translational repression may facilitate the selective synthesis of a small number of proteins that can escape global inhibition through a non-canonical , mRNA cap-independent translation initiation mechanism dependent on internal ribosome entry sites ( IRESes ) ( Cornelis et al . , 2000; Pyronnet et al . , 2000; Qin and Sarnow , 2004; Schepens et al . , 2007; Wilker et al . , 2007; Marash et al . , 2008; Ramirez-Valle et al . , 2010 ) . However , recent work challenged the view that translation in mitosis is mediated to a significant extent by IRESes , and instead found that canonical , cap-dependent translation dominates in mitosis ( Shuda et al . , 2015 ) . Therefore , it is unclear whether IRES-dependent translation represents a general mechanism of translational regulation during mitosis , and whether such IRES-dependent translational activation represents , a minor , or the dominant mechanism of gene-specific translational regulation during mitosis . The recent development of ribosomal profiling , a method that uses deep sequencing of ribosome-protected mRNA fragments to quantify ribosome occupancy on individual mRNAs , allows the TE of single mRNA species to be examined at a system-wide level ( Ingolia et al . , 2009 , 2011 ) . A recent study applied this technology to investigate the mechanism of cell cycle-dependent translational control and found hundreds of genes that show changes in TE during the cell cycle ( Stumpf et al . , 2013 ) . Many mRNAs showed altered translation rates in mitosis when compared to either G1 or S phase cells . However , this study did not include analysis of G2 phase cells , which is required to determine whether observed changes are mitosis-specific , as G2 phase cells are very similar to mitotic cells , but very different from G1- or S-phase cells , at least with respect to mRNA levels ( Cho et al . , 2001; Whitfield et al . , 2002 ) . Thus , it is imperative to compare mitotic cells with both G2 cells ( pre-mitotic entry ) and G1 cells ( post-mitotic exit ) in order to identify the translational changes that occur specifically during mitosis . It is also critical to obtain a minimally perturbed population of mitotic cells . Previous studies of mitotic translational regulation relied on synchronizing cells in mitosis with microtubule targeting drugs ( Fan and Penman , 1970; Pyronnet et al . , 2001; Stumpf et al . , 2013 ) , but recent work has shown that translation rates are dramatically affected by treatment with such drugs ( Sivan et al . , 2011; Coldwell et al . , 2013 ) , thereby complicating the annotation of mitosis-specific translation effects . Here , using metabolic labeling , combined with ribosome profiling and a tight cell cycle synchronization protocol that does not require microtubule drug treatment , we have identified two distinct translational programs that occur during mitosis . First , using metabolic labeling of non-transformed RPE-1 cells , we find a modest ( ∼35% ) global translational repression of the bulk of mRNAs during mitosis , which is consistent with findings in other studies ( Fan and Penman , 1970; Bonneau and Sonenberg , 1987; Pyronnet et al . , 2001 ) . In addition to this modest global repression , using ribosomal profiling , we identify a subset ∼200 of mRNAs that show much larger ( >threefold ) , gene-specific changes in their TE during mitosis . The large majority of the latter group of mRNAs are translationally repressed at mitotic entry and then translationally re-activated at mitotic exit , highlighting the precise temporal specificity of this gene-specific translational regulation . Thus , translational repression , rather than IRES-dependent activation of mRNA translation , is the dominant method of gene-specific translational regulation in mitosis , although minor effects of IRES-dependent translation cannot be ruled out . Follow-up studies on one of these translationally repressed genes , Early mitotic inhibitor 1 ( Emi1 ) , a potent inhibitor of the APC , reveals that translational repression in mitosis is important to prevent new Emi1 protein synthesis at a time when the existing Emi1 protein pool is degraded . These results lead to a model in which the combined activities of protein degradation and translational repression ensures the complete removal of Emi1 protein , which enhances APC activation and the degradation of APC substrates at the end of mitosis . These results provide the first genome-wide view of the translational changes that occur specifically in mitosis and reveal that translational regulation can enhance the efficiency of post-translational protein inhibition , which may represent a more general function for translational repression . To study mRNA translation at a genome-wide level during mitosis , we used ribosome profiling , a recently developed technique that enables precise measurements of translation of each mRNA in the cell ( Ingolia et al . , 2009 , 2011 ) . In ribosome profiling , the small fragments of mRNA ( ∼30 nt ) that are associated with a ribosome ( called the ribosome footprints [FPs] ) are isolated and quantitatively analyzed by deep sequencing . This sequence information allows the calculation of the average number of ribosomes per mRNA , which reports on the TE of each mRNA . In parallel , we analyzed total mRNA content by RNA-seq . As many cell cycle-regulated pathways are deregulated in cancer , we used a non-transformed human epithelial cell line , RPE-1 , for these studies , as these cells can be precisely synchronized in the cell cycle ( see below ) . We synchronized RPE-1 cells in late G2 ( G2 ) , mitosis ( M ) or early G1 ( G1 ) with a specific small molecule CDK1 inhibitor , RO-3306 , using a previously established synchronization protocol ( Vassilev et al . , 2006 ) ( Figure 1A ) . In this protocol , cells are first arrested in G2 using the CDK1 inhibitor . CDK1 is largely inactive during G2 and only becomes activated at the end of G2/prophase ( Jackman et al . , 2003; Gavet and Pines , 2010 ) . Thus , the inhibitor prevents the progression out of G2 , but the low CDK1 state in the presence of the CDK1 inhibitor reflects the normal G2 phase in unsynchronized cells ( Jackman et al . , 2003; Gavet and Pines , 2010 ) . To release cells from G2 , the CDK1 inhibitor is washed out and the cells progress into M and then G1 , and pure ( 95% ) populations of M and G1 cells can be obtained based upon the timing of release from G2 arrest ( Figure 1B , C ) . Thus , this cell synchronization protocol is minimally perturbing and avoids the use of drugs that arrest cells in mitosis by targeting microtubules , which are known to affect translation ( Sivan et al . , 2011; Coldwell et al . , 2013 ) . 10 . 7554/eLife . 07957 . 003Figure 1 . Cell synchronization and analysis of translation efficiency during the cell cycle . ( A ) Schematic overview of RPE-1 cell synchronization protocol . RO-3306 ( 6 μM ) was used as the CDK1 inhibitor . ( B , C ) G2 , M and G1 samples were prepared as outlined in ( A ) . ( B ) FACS analysis ( Hoechst staining of DNA ) reveals that the samples are effectively synchronized in the respective cell cycle phase . ( C ) The number of mitotic cells was scored by microscopy based on chromosome condensation ( DNA stained with DAPI ) . Graph is average of 3 independent experiments with ∼50 cells scored per experiment . Error bars represent standard error of the mean ( SEM ) . ( D ) RPE-1 cells were synchronized as described in ( A ) . Before harvesting , cells were incubated with S35-methionine for 10 min to radioactively label newly synthesized proteins . The left panel shows the autoradiograph of newly synthesized proteins . The middle panel shows total protein content of the cells stained by Coomassie . Right panel shows quantification of autoradiographs of 3 independent experiments , normalized to total protein . Mean and standard deviation ( SD ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 003 Using S35-methionine labeling , we found a global decrease in mRNA translation during mitosis when compared to G2 and G1 cells respectively ( Figure 1D ) . This finding confirms the global reduction in translation during mitosis reported by other studies ( Fan and Penman , 1970; Bonneau and Sonenberg , 1987; Pyronnet et al . , 2001 ) ; however , the magnitude of the effect seen here is smaller ( ∼35% reduction in our study vs 70% found previously [Fan and Penman , 1970] ) . This difference might be due to the use of microtubule inhibitors in cells arrested in mitosis in previous studies , which can cause global translational repression ( Sivan et al . , 2011; Coldwell et al . , 2013 ) . Next , we analyzed genome-wide mRNA levels in G1 , G2 and M phase cells by deep sequencing ( Supplementary file 2 ) . Comparing mRNA abundance in G2 vs M cells , we detected only 25 genes that were down-regulated in M and not a single up-regulated gene ( >threefold , see ‘Materials and methods’ section ) ( Figure 2A , red bars , left graph ) . In contrast , when G2 cells were compared to G1 cells , 220 genes were down-regulated and 82 genes up-regulated in G1 ( Figure 2A , red bars , middle graph ) . Similarly , when M cells were compared to G1 cells , many mRNAs were changed ( 138 up and 156 down in M ) ( Figure 2A , red bars , right graph ) . The mRNAs that changed between G1 and G2 were mostly the same set that also changed between G1 and M ( Figure 2—figure supplement 1A ) ; 96% of mRNAs that are up-regulated in G1 vs G2 are also increased at least twofold in G1 vs M . Similarly , 80% of mRNAs that are down-regulated in G1 vs G2 , are also decreased at least twofold in G1 vs M . Together , these results indicate that the mRNA content of G2 and M phase cells is very similar , but distinct from G1 phase cells . 10 . 7554/eLife . 07957 . 004Figure 2 . Transcriptional and translational regulation affect distinct cell cycle transitions . ( A ) For each gene the ratio of mRNA levels ( red bars ) and translation efficiency ( TE ) ( blue bars ) was determined for G2 vs M ( left ) or G2 vs G1 ( middle ) and M vs G1 ( right ) . The number of genes that showed changes in mRNA levels ( >threefold difference , red bars ) or changes in TE ( or >threefold difference in TE combined with >twofold difference in the ribosome footprint value , blue bars ) was plotted in a pair of histograms . ( B ) mRNA levels are plotted for all genes that are translationally regulated in M vs G2 . The left graph shows the mRNA levels in M compared to G2 , the right graph shows mRNA levels of M compared to G1 . Note that mRNA levels of translationally regulated mRNAs are similar in G2 , M and G1 . ( C ) 19 well characterized cell cycle proteins that show strong cell cycle-dependent regulation of mRNA levels were manually selected . Fold difference between G2 and G1 in mRNA levels ( red bars ) and TE ( blue bars ) is shown . While there is a large change in mRNA levels , the TE is similar in G2 and G1 for the majority of these mRNAs . ( D ) The subset of genes that was translationally repressed in M compared to G2 ( 182 genes ) was selected and the fold difference in TE for M vs G2 was plotted against the fold difference in TE for M vs G1 . Results show that genes which are translationally repressed in M compared to G2 , or repressed to similar levels in M when compared to G1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 00410 . 7554/eLife . 07957 . 005Figure 2—figure supplement 1 . Transcriptional and translational regulation affect distinct cell cycle transitions . ( A ) The median mRNA RPKM value was determined for the set of genes for which the mRNA was higher ( >threefold ) in G2 compared to G1 ( left ) , or higher in G1 compared to G2 ( right ) . Graphs show that mitotic mRNA levels are more similar to G2 than to G1 . ( B ) The subset of genes that was translationally repressed in M compared to either G2 ( 182 genes , blue bars ) or compared to G1 ( 86 genes , red bars ) was selected . For both these gene sets , the median TE value was determined for G2 , M and G1 . Results show that genes which are translationally repressed upon mitotic entry , are re-activated upon mitotic exit . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 00510 . 7554/eLife . 07957 . 006Figure 2—figure supplement 2 . Translational regulation affects many cell cycle-dependent processes . ( A , B , C , E , F ) mRNA levels ( red bars ) and TE ( blue bars ) were compared between M and G2 for multiple genes in the same biological pathway . ( D ) Since most histone mRNAs do not contain poly ( A ) tails and are thus not enriched in the total mRNA sample , the footprint RPKM values were compared between M and G2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 00610 . 7554/eLife . 07957 . 007Figure 2—figure supplement 3 . Excessive PP1γ and PP2aβ activity perturbs chromosome segregation . ( A–D ) RPE-1 cells expressing either H2B-GFP alone or together with mCherry and PP1γ ( A , C ) or mCherry and PP2aβ ( B , D ) were analyzed by time-lapse microscopy . Simultaneous expression of mCherry and untagged PP1γ or PP2aβ was accomplished by inserting a P2A ribosome skipping sequence in between mCherry and the phosphatase sequence . ( A ) Stills from representative videos of control cell ( upper panel ) or PP1γ overexpressing cell ( lower panel ) . Time is shown in min . ( B ) Stills from representative video of control cell ( upper panel ) or PP2aβ expressing cells ( lower two panels ) . The top PP2aβ expressing cell shows a prometaphase arrest with misaligned chromosomes , while the bottom cell shows a cell with a prometaphase delay and subsequent cytokinesis without chromosome segregation , known as a ‘cut’ phenotype . Time is shown in min . ( C ) The fraction of cells in which one or more chromosomes mis-segregated was determined for control cells and cells expressing PP1γ . ( D ) shows the average time from NEB to anaphase for control cells and cells expressing PP2aβ . All graphs are the mean and SD of 3 independent experiments with 20–40 cells analyzed per experiment . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 007 Next , we subjected ribosome FPs to deep sequencing to examine whether individual mRNAs are differentially translationally regulated at different stages of the cell cycle ( Supplementary file 3 ) . We refer to this regulation as gene-specific translational to distinguish it from the global translational repression described above . The number of ribosome FPs ( which reports on the amount of total translation ) was determined for each mRNA and was divided by the total mRNA abundance to obtain the TE . The vast majority of gene-specific changes in TE were observed when M phase transcripts were compared with either G2 or G1; 199 and 92 genes were translationally regulated between M and either G2 or G1 , respectively . In contrast , only 13 genes showed changes in translation between G2 and G1 ( Figure 2A , blue bars; transcripts with >threefold difference in TE , and >twofold difference in ribosome footprint ( FP ) density were scored as translationally controlled , see ‘Materials and methods’ for more details ) . Thus , in contrast to mRNA abundance , which is similar in G2 and M , but distinct in G1 , TE is similar in G2 and G1 , but very different in M . When we analyzed mRNA abundance of the 199 genes that showed gene-specific regulation in M , we found that their mRNA levels were largely constant throughout the cell cycle ( Figure 2B ) . Similarly , the TE of genes known to be transcriptionally regulated was largely constant ( Figure 2C ) . These results indicate that gene-specific translational regulation affects a different set of genes than transcriptional regulation . The vast majority of the 199 mRNAs that show gene-specific translational regulation in M compared to G2 were repressed rather than activated; comparing M to G2 , 182 were translationally downregulated in M and only 17 were upregulated ( Figure 2A , blue bars , middle graph; Figure 2—figure supplement 1B ) . Similarly , of the 92 mRNAs that translationally regulated between M and G1 , 86 were repressed in M , and only 6 were activated ( Figure 2A , blue bars , right graph; Figure 2—figure supplement 1B ) . To test whether the same set of mRNAs that was translationally repressed at mitotic entry were de-repressed at mitotic exit , we compared the overlap in mRNAs repressed in M vs G2 and M vs G1 . The genes that were translationally repressed in M vs G2 were mostly also repressed in M vs G1; of the 182 genes that were repressed in M compared to G1 , 87% were repressed >twofold in M compared to G1 . Furthermore , there is a good correlation in the fold change in TE between G2 vs M and G1 vs M for individual mRNAs ( Figure 2D ) . In summary , when cells progress from G2 to M , gene-specific translational regulation is dominated by repression , and the genes that are translationally repressed as cells enter mitosis are mostly re-activated upon mitotic exit . It is important to note that fold change values noted above are relative to the average mRNA of the biological sample ( as ribosome profiling only reports on relative changes ) . Thus , specific mRNAs that are translationally repressed threefold relative to other mRNAs in mitosis , are repressed ∼fourfold relative to the same gene in G2 phase ( given the global ∼35% translational repression that acts on all mRNAs during mitosis ) . Similarly , the small number of mRNAs that are translationally activated by threefold in mitosis , are only expressed ∼twofold higher than in G2 phase . Thus , we conclude that the vast majority of mRNAs that undergo gene-specific regulation are translationally repressed in mitosis . Next , we examined whether there were particularly types of genes that were predominantly regulated by translational vs transcriptional control , so we performed gene ontology enrichment analysis using the functional annotation tool DAVID ( Huang da et al . , 2009 ) . Many genes that exhibited variations in mRNA levels during the cell cycle are involved in cell division ( p-values < 10−9 , see ‘Materials and methods’ ) . In contrast , the translationally regulated genes were functionally very different from the transcriptionally regulated genes and included many signaling molecules , transcription factors , and transmembrane proteins ( significantly enriched with p-values < 10−8 ) ( see Figure 2—figure supplement 2 ) . Manual curation of the mRNAs that showed gene-specific translational repression during mitosis revealed regulation of several components of the same pathway . For example , multiple components of the PI3 kinase pathway were translationally repressed during mitosis ( Figure 2—figure supplement 2A ) . We also found mitosis-specific translational downregulation of the mitotic phosphatases PP1γ and PP2aβ ( Figure 2—figure supplement 2B ) . During mitosis , these phosphatases are strongly inhibited through phosphorylation and binding to inhibitory proteins ( Wurzenberger and Gerlich , 2011 ) , suggesting that mitotic translational repression may represent an additional back-up mechanism to inhibit protein function ( see Discussion ) . Consistent with this notion , we found that overexpression of either PP1γ or PP2aβ phosphatase strongly disrupted normal cell division ( Figure 2—figure supplement 3A–C ) . We also found a strong translational repression of two key regulators of centriole duplication , Plk4 and CP110 ( Figure 2—figure supplement 2C ) ( Chen et al . , 2002; Habedanck et al . , 2005 ) . Finally , the majority of histones showed a strong reduction in protein synthesis in M compared with G2 . Newly synthesized histones may not incorporate readily into highly condensed mitotic chromosomes , which perhaps could explain why their translation is reduced during mitosis , although we cannot completely rule out a small contamination of S-phase cells in the G2 sample which might give rise to an apparent high level of translation of histone mRNAs in G2 . We also found a few mRNAs that were translationally increased in mitosis as compared to both G1 and G2 , although most were below our threshold of threefold change , indicating the changes were subtle . Included in this list are genes involved in cytoskeleton function and DNA replication initiation ( Figure 2—figure supplement 2E , F ) , the latter of which may reflect the ability of cells to license DNA replication at the end of mitosis ( Clijsters et al . , 2013 ) . Taken together , these results show that transcriptional and gene-specific translational control dominate at different stages of the cell cycle ( transcription at the G1-to-G2 transition and translational regulation dominating at the more rapid G2-to-M transition ) and regulate a distinct set of genes . To validate the results obtained in our ribosome profiling experiments in living cells , we developed a fluorescence-based reporter to analyze translation of individual transcripts in living cells . In brief , the reporter consists of a GFP fused to an inducible degron , which is continuously degraded until a small molecule stabilizer is added ( Iwamoto et al . , 2010 ) ( Figure 3A ) . A mCherry protein expressed independently from the same transcript was used for normalization ( Figure 3—figure supplement 1A , B ) . Our assay is similar to another recently developed method for measuring translation rates in cells ( Han et al . , 2014 ) , although our system included an mCherry protein that was expressed from the same mRNA using a P2A site . Upon addition of the stabilizer drug , time-lapse microscopy revealed an increase in the GFP/mCherry ratio over time , which reflects the translation rate of the GFP ( Figure 3B , C , see ‘Materials and methods’ ) . 10 . 7554/eLife . 07957 . 008Figure 3 . Analysis of translation efficiency in living cells using a fluorescence-based translation reporter . ( A ) Schematic representation of the live-cell translation reporter . An inducible degron ( DHFR-Y100I ) fused to sfGFP and an NLS is separated from an NLS-mCherry protein by a P2A ribosome skipping sequence , which allows these two proteins to be synthesized as separate proteins from a single transcript . Upon addition of the small molecule stabilizer trimethoprim ( TMP ) , newly synthesized DHFR-sfGFP-NLS is stabilized and GFP fluorescence increases over time due to new GFP protein synthesis . Thus , GFP fluorescence increase reports on translation efficiency . The mCherry signal is used to normalize for the plasmid copy number per cell . ( B ) RPE-1 cells stably expressing the reporter were treated with 50 μM TMP and followed by time-lapse microscopy . Scale bar , 20 μm . Time is indicated in min . ( C ) Quantification of GFP/Cherry ratio ( mean and standard deviation , n = 8 cells ) with or without cycloheximide treatment . ( D ) 5′ and 3′ UTRs from indicated genes were inserted in the reporter . Cells expressing the different reporters were blocked in mitosis with taxol , treated with TMP and imaged for 4 hr . To determine the translation rate , the GFP/mCherry ratio was calculated at the start and end of each video for both interphase and mitotic cells . The ratio of translation rates in mitosis and interphase for each reporter is shown . For Emi1 , mitotic cells were compared with G2 phase cells only , as translation was also reduced in G1 ( unpublished observation ) . Results are mean and SEM of 3 independent experiments with 10–20 cells analyzed per condition per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 00810 . 7554/eLife . 07957 . 009Figure 3—figure supplement 1 . Analysis of mRNA sequences that confer translational regulation . ( A ) RPE-1 cells expressing the fluorescent reporter with control UTRs were treated with TMP for 3 hr and changes in GFP and mCherry signals were followed by time-lapse microscopy . The total increase in GFP fluorescence was measured per cell and after background subtraction was plotted in the graph ( blue diamonds , each diamond represents 1 cell ) . mCherry signal was measured at the start of the video and the absolute increase in GFP fluorescence was divided by the initial mCherry signal to normalize for reporter copy number per cell ( red squares ) . Note that the spread after normalization is smaller . ( B ) Western blot of RPE1 cells expressing the translational reporter and treated with TMP for 24 hr probed for GFP . Note that two bands are visible; the bottom bands represents the protein product after successful ribosome skipping at the P2A site , while the weak upper band is a fusion protein that is generated after failure of ribosome skipping at the P2A site . ( C , D ) Visual representation of RNA-seq data of the 5′ and 3′ UTRs of indicated genes . Individual sequencing reads are mapped onto the UTR sequences , in which the 5′ nucleotide of each sequencing read is shown . RNA-seq data are shown for G2 samples ( C ) or for both both G2 and M cells ( D ) . ( E ) Untreated RPE-1 cells were fixed and Emi1 mRNA was detected by single molecule FISH . The number of mRNAs was counted in either prophase or metaphase ( as determined by DNA condensation and chromosome alignment ) . ( F ) Either control UTRs or indicated Emi1 UTRs were inserted in the fluorescence translation reporter ( note: 5′ UTR numbering refers to numbered isoforms shown in [D] ) . Cells expressing the different reporters were left unsynchronized or blocked in mitosis with taxol , treated with TMP and imaged for 8 hr . Cells were defined as G2 cells if they entered mitosis within 5 hr . To determine the translation rate , the GFP/mCherry ratio was calculated at the start and end of a 3 hr interval . The ratio of translation rates in M and G2 for each reporter is shown . Results are mean and SD of 3 independent experiments with 10 cells analyzed per condition per experiment . Note that complete translational inhibition is expected to result in a slightly negative value for the normalized translation rate , because GFP levels will actually be lower at the end of the 3 hr measurement interval if no new GFP is synthesized , due to the decay of existing GFP protein . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 009 To test for translational regulation , the 5′ and 3′UTRs of the control reporter were replaced by the 5′ and 3′ UTRs of genes that were identified by ribosomal profiling as translationally repressed in mitosis; UTRs of two genes that were not translationally regulated were tested as controls . For these experiments , we used the UTRs of mRNA isoforms that closely matched the isoforms expressed in RPE1 cells , as determined by our RNA-seq data ( Figure 3—figure supplement 1C , D ) While the UTRs of control transcripts did not lead to translational regulated of the GFP reporter , time-lapse imaging revealed that the transcripts with UTRs from translationally repressed mRNAs showed decreased synthesis in mitosis compared with interphase ( Figure 3D ) . These results provide strong , independent validation of the ribosome profiling dataset and show that regulatory elements present in 5′ and 3′ UTRs confer translational repression in mitosis . To study the function of translational repression in mitosis , we focused on Emi1 , a highly repressed gene identified in our ribosome profiling dataset ( Supplementary file 3 ) and translational reporter assay . ( Figure 3D ) . First we wished to confirm that regulation of Emi1 expression is due solely to translational regulation , not to changes in Emi1 mRNA expression . To establish that mRNA levels of Emi1 remain unchanged as cells progress into mitosis , we performed single molecule FISH and found that Emi1 mRNA was present at similar levels in G2 and M ( Figure 3—figure supplement 1E ) . However , since these FISH probes detect all known Emi1 isoforms , we also wished to determine whether isoform specific regulation may occur . First , we confirmed that the same mRNA isoforms were expressed in G2 and M ( Figure 3—figure supplement 1D ) . Furthermore , we mapped the regulatory element in the Emi1 UTRs , and found that the translational repression was conferred by the 3′UTR of Emi1 alone , which is identical in all Emi1 mRNA isoforms ( Figure 3—figure supplement 1D , F ) . Consistent with this , all 5′UTR isoforms in combination with the Emi1 3′UTR were regulated in a similar manner ( Figure 3—figure supplement 1D , F ) . Together , these results confirm the conclusion that the inhibition of Emi1 protein synthesis in M is due to translational repression , and show that the 3′UTR of Emi1 is sufficient for this translational regulation . Emi1 is a potent inhibitor of the APC bound to its activator Cdh1 ( APC/Cdh1 ) . In G2 phase , Emi1 inhibits APC/Cdh1 , and this inhibition is required to allow accumulation of cyclins in G2 and for mitotic entry ( Reimann et al . , 2001; Hsu et al . , 2002; Di Fiore and Pines , 2007 ) . However , in early mitosis , Emi1 protein is inactivated both through protein degradation and CDK1-dependent phosphorylation ( Reimann et al . , 2001; Guardavaccaro et al . , 2003; Margottin-Goguet et al . , 2003; Moshe et al . , 2011 ) , which is likely required for APC/Cdh1 activation and APC substrate degradation in the subsequent telophase/G1 . Given these post-translational mechanisms for inhibiting Emi1 , it is unclear whether translational repression of Emi1 also serves a role in suppressing its activity during mitosis . To test the importance of translational inhibition of Emi1 in mitosis , we expressed mCherry-Emi1 ( a functional fusion protein [Di Fiore and Pines , 2007] ) from a plasmid lacking native UTRs and thus lacking translational regulation , and examined APC/Cdh1 activation as cells progressed through mitosis into the next G1 phase . First , we confirmed that mCherry-Emi1 was degraded in late G2/early mitosis ( Figure 4—figure supplement 1A ) , as found previously ( Guardavaccaro et al . , 2003; Margottin-Goguet et al . , 2003 ) . As a readout of APC/Cdh1 activity , we expressed fluorescently-tagged Aurora A , Plk1 or CDC20 , all of which are APC/Cdh1 substrates . ( Peters , 2006 ) . These fluorescence reporter substrates were all rapidly degraded in anaphase/telophase in control cells as expected ( Figure 4A–C ) . When Emi1 lacking its normal 5′ and 3″ translation regulatory elements was expressed , cells progressed through mitosis normally and the bulk of Emi1 protein was degraded; however , APC/Cdh1 activation in telophase was partially inhibited , as indicated by the decreased degradation of APC/Cdh1 substrates ( Figure 4A–C and Video 1 ) . Importantly , exogenous Emi1 was expressed at similar levels as the endogenous gene in these experiments , as determined by transcript counting by single molecule FISH ( Figure 4—figure supplement 1B ) . These results show that Emi1 can inhibit APC/Cdh1 activation at the end of mitosis and suggest that Emi1 protein degradation is insufficient to completely inhibit Emi1 activity in the presence of continued Emi1 synthesis . 10 . 7554/eLife . 07957 . 010Figure 4 . Translational inhibition of Emi1 during mitosis promotes APC/Cdh1 activation in telophase . ( A–C ) RPE-1 cells expressing Aurora A-GFP ( A ) , mCherry-Plk1 ( B ) or mCherry-CDC20 ( C ) alone or combined with fluorescently tagged Emi1 were analyzed by time-lapse microscopy and the protein degradation rates were assayed through quantification of fluorescence intensities over time as cells progressed through mitosis . ( D–F ) RPE-1 cells stably expressing Aurora-GFP and , where indicated , mCherry-Emi1 with indicated UTRs , were analyzed be time-lapse microscopy . Representative images ( D ) and quantification of Aurora A-GFP levels ( E ) and mCherry-Emi1 levels ( F ) are shown . Asterisks and dotted boxes mark dividing cells . Degradation of Aurora A normally occurs between anaphase onset and telophase . For quantification only cells were included that had very low mCherry-Emi1 fluorescence to ensure low expression level of exogenous Emi1 ( see also Figure4—figure supplement 1C , D ) . Scale bar , 10 μm . Mean and standard error of 3 independent experiments , with ∼10 cells analyzed per experimental condition per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 01010 . 7554/eLife . 07957 . 011Figure 4—figure supplement 1 . Translational repression of Emi1 by its UTRs facilitates APC activation . ( A ) RPE-1 cells stably expressing mCherry-Emi1 were followed by time-lapse microscopy . Representative images are shown of a cell expressing mCherry-Emi1 undergoing cell division when Emi1 protein is degraded . ( B ) Either control RPE-1 cells or RPE-1 cells stably expression mCherry-Emi1 were fixed and mRNA was visualized by single molecule FISH . The number of endogenous Emi1 mRNA molecules or exogenous mCherry-Emi1 mRNA molecules was quantified in G2 phase of the cell cycle . ( C ) RPE-1 cells stably expressing Aurora A-GFP together with mCherry-Emi1 with either control UTRs or Emi1 UTRs were followed by time-lapse microscopy . Fluorescence intensities of Aurora A-GFP in prophase and telophase/early G1 were measured , and the fraction of GFP fluorescence remaining in telophase/early G1 compared to prophase was plotted on the y-axis . The fluorescence intensity of mCherry-Emi1 was also measured in prophase ( before Emi1 degradation ) and was plotted on the x-axis . Each dot represents a single cell . Graph shows that at similar mCherry-Emi1 protein expression Aurora A is degraded more efficiently when Emi1 is under control of its own UTRs . ( D ) Untreated RPE-1 cells were fixed and Emi1 mRNA was visualized by single molecule FISH . Microtubules and DNA were stained with an anti-α-tubulin antibody and DAPI , respectively . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 01110 . 7554/eLife . 07957 . 012Video 1 . Video of two RPE-1 cells going through cell division . The first cell to divide expresses both Aurora A-GFP and mCherry-Emi1 , while the second cell only expresses Aurora A-GFP Note that upon division , Aurora A-GFP degradation is perturbed in the cell expressing mCherry-Emi1 . Time interval between images is 20 min . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 012 To test whether translational repression of Emi1 would allow APC activation under these conditions , we replaced the control UTRs of mCherry-Emi1 with Emi1's native UTRs and examined APC activation at mitotic exit . Strikingly , translational repression of Emi1 in mitosis allowed a substantially higher degree of APC/Cdh1 activation at comparable Emi1 expression levels ( Figure 4D , E and Figure 4—figure supplement 1C ) . To confirm that the UTRs of Emi1 were enhancing APC/Cdh1 activation solely through mitosis-specific translational repression , rather than through an alternative mechanism , for example by stimulating specific mRNA localization , we sought to confer mitosis-specific translational repression on Emi1 mRNA independently of the Emi1 UTRs . To this end , we replaced the Emi1 UTRs with the UTRs of an unrelated mRNA , ARHGAP5 , which have a completely different sequence , but confer translational repression in mitosis to a similar extent as Emi1 UTRs ( Figure 3D ) . Indeed , mitosis-specific translational repression of mCherry-Emi1 by ARHGAP5 UTRs allowed APC activation at the end of mitosis to a similar extent as Emi1 UTRs ( Figure 4D , E ) . Furthermore , examination of Emi1 mRNA localization during mitosis did not reveal localization to a specific site in the cell ( Figure 4—figure supplement 1D ) , arguing against a role for the UTRs of Emi1 in mRNA localization , but rather indicating that their primary function is to repress translation during mitosis . To understand how translational repression of Emi1 contributes to APC activation , we examined Emi1 protein levels during mitosis in the presence and absence of translation repression . Interestingly , Emi1 was more completely depleted when protein degradation was combined with translational repression ( Figure 4F ) . These results demonstrate that inactivation of existing Emi1 protein , through degradation and potentially inhibitory phosphorylation ( Moshe et al . , 2011 ) , is insufficient to completely inhibit Emi1 activity , but when post-translational inactivation is combined with translational repression , Emi1 is more thoroughly inhibited , which allows robust APC activation . Previous work had found a global translational repression during mitosis ( Fan and Penman , 1970 ) . Here we confirm a modest ( 35% ) and general repression of translation in mitosis , but in addition , our genome-wide analyses revealed much larger ( >300% ) effects on TE of several hundred specific mRNAs , the large majority of which involve translational repression rather than activation . Thus in summary , the mitotic translational program is dominated by a modest global translational repression , combined with potent repression of a subset of mRNAs , which acts on top of the global repression . Our follow-up studies on Emi1 suggests that its translational repression acts as a mechanism to enhance post-translational protein inactivation , which could represent a more general function of translational repression in mitosis and in other processes . The evidence for mitosis-specific translational repression of specific mRNAs is supported by two separate methodologies . The first evidence comes from ribosomal profiling which measures ribosome occupancy of native transcripts , but does not provide a direct measure of translation itself . The second method involves the use of an introduced ( thus non-native ) reporter mRNA that provides a real-time , live-cell readout of protein production from translation . In the test cases where both methods were applied , we find good agreement in the results , providing strong support for widespread gene-specific translational repression in mitosis . In this work , we avoided microtubule depolymerizing drugs for cell cycle synchronization , as employed by many other studies of translation in mitosis , since microtubule depolymerization has been recently shown to have a profound effect on mRNA translation ( Coldwell et al . , 2013 ) . However , one possible concern is whether the CDK1 inhibitor RO-3306 used here to arrest cells in G2 might influence translation as well and thus affect the conclusions of this study . CDK1 , for example , is known to phosphorylate a number of translation factors ( Heesom et al . , 2001; Dobrikov et al . , 2014; Shuda et al . , 2015 ) and this could affect TE . However , we feel that the major conclusions regarding selective translational repression in mitosis are unlikely to be due to the use of this CDK1 inhibitor for cell synchronization for three reasons . First , the genome-wide comparison of TE of G1 vs M cells ( neither of which have any drug present ) yields very similar results to the comparison of G2 ( RO-3306 present ) vs M cells . Second , we confirmed translational regulation of a selected subset of mRNAs using a single cell fluorescence-based assay that does not involve the use of the CDK1 inhibitor . Third , the CDK1 inhibitor is only present in our G2 sample , a time at which CDK1 is mostly inactive in unperturbed cells ( Jackman et al . , 2003; Gavet and Pines , 2010 ) . Consistent with this , the translation inhibitor 4E-BP1 , which is phosphorylated by CDK1 in mitosis , is not detectably phosphorylated in G2 ( Heesom et al . , 2001 ) . Thus , our synchronized G2 sample closely mimics the G2 phase of unsynchronized cells with respect to CDK1 activity . Together , these results strongly support the conclusion that the observed translational effects reported in this study are due to distinct cell cycle states and not an effect of drug treatment . Interestingly , the sets of genes that show altered TE and mRNA abundance ( likely due , at least in part , to transcriptional regulation ) are largely non-overlapping ( Figure 2B , C ) . Furthermore , the cell cycle timing of transcriptional and translational regulation also differ; the majority of gene-specific translational changes occur at entry and exit from mitosis ( G2-to-M and M-to-G1 transitions ) , while transcriptional changes predominate between G1 and G2 . We also show that the vast majority of the ∼200 translationally-regulated mRNAs are repressed ( several fold below the modest global down-regulation of translation in mitosis ) rather than activated . Thus , we conclude that the dominant function of gene-specific translational regulation during mitosis is to potently inhibit synthesis of a relatively small subset of the proteome . Our work complements another cell cycle ribosomal profiling study by Stumpf et al . ( Stumpf et al . , 2013 ) , which focused on translational changes associated with S phase , and together these studies provide a complete overview of translational regulation during the cell cycle . The distinct use of transcriptional and translational regulatory mechanisms is consistent with the timing of the cell cycle transitions for which they are used; regulating transcription is a relatively slow mechanism for altering protein synthesis rates , because of the time required for transcription , mRNA processing/nuclear export and mRNA turnover . This limits the usefulness of transcriptional regulation to longer time transitions in the cell cycle , specifically from G1 to S or G2 . On the other hand , translational control alters protein synthesis rates almost instantaneously , which makes it well suited to the short time scale of mitosis , which is generally less than 1 hr in most somatic cells with doubling times of 1–2 days . Interestingly , a recent study of mouse dendritic cell activation , which happens at a time-scale of many hours , similar to the G1-S transition , found that the vast majority of changes in protein synthesis were due to altered mRNA abundance , rather than due to changes in translation rates ( Jovanovic et al . , 2015 ) . A second unique feature of translational control is its rapid reversibility , enabling protein synthesis to restart quickly when cells exit from mitosis and enter G1 . Thus , translational control may be employed when acute changes in protein synthesis are needed ( e . g . , entry into and exit from mitosis ) , while transcriptional control is used to affect slower changes in gene expression . A surprising finding of this study is that very few mRNAs showed substantial ( >threefold ) , gene-specific translational activation during mitosis . In previous reported cases of translational regulation of specific mRNAs during mitosis , the findings were predominantly of translational activation through an IRES-dependent mechanism ( Cornelis et al . , 2000; Pyronnet et al . , 2000; Qin and Sarnow , 2004; Wilker et al . , 2007; Marash et al . , 2008; Ramirez-Valle et al . , 2010 ) . While we also found a small number of mRNAs that were translated more efficiently during mitosis ( 16 and 6 mRNAs were translationally upregulated >threefold in mitosis compared to G2 and G1 , respectively ) , the vast majority of regulated mRNAs were translationally repressed ( 187 and 97 mRNAs were downregulated in mitosis compared to G2 and G1 , respectively ) . One possible explanation for this discrepancy is that the IRES-mediated upregulation of translation in mitosis may be relatively weak compared to the cut-off ( >threefold ) that we have used in this study to identify changes in gene-specific TE . A previous study of IRES-mediated translational upregulation found that only one of three proteins examined was upregulated by > threefold ( Qin and Sarnow , 2004 ) . Furthermore , a recent study found that translation in mitosis is dominantly cap-dependent ( Shuda et al . , 2015 ) , suggesting that IRES-mediated translation may only make a minor contribution to the translational landscape in mitosis . The results from our work and Shuda et al . , while not ruling out IRES-dependent translational activation of some genes in mitosis , indicates that the dominant function of gene-specific translational regulation is to shut down rather than activate synthesis of proteins during mitosis . Our results , along with others , also reveal a number of interesting differences when comparing mitosis in somatic cells to studies of translational regulation during meiosis and early embryonic development ( Mendez and Richter , 2001; Groisman et al . , 2002 ) . Oocytes stockpile maternal mRNA , as new transcription does not occur until the mid-blastula stage of development . Due to the absence of transcriptional control , translational regulation is the main mechanism by which protein synthesis rates are tuned . In these systems , translational regulation occurs largely by modulating poly ( A ) tail length ( Weill et al . , 2012; Subtelny et al . , 2014 ) . While changes in poly ( A ) tail length also occur on a subset of mRNAs during the mitotic cell cycle ( Novoa et al . , 2010 ) , regulation of poly ( A ) tail length does not appear to be a general mechanism for controlling TE in somatic cells ( Subtelny et al . , 2014 ) . Thus , translational control in meiosis is most likely mechanistically distinct from the gene-specific translational regulation during mitosis described here . Consistent with this , we find that the large majority of regulated mRNAs are translationally repressed during mitosis , while in meiosis regulatory mechanisms mainly function to specifically activate a subset of mRNAs ( Mendez and Richter , 2001 ) . Furthermore , the sets of genes that are regulated during meiosis and mitosis are largely non-overlapping . Interestingly , Emi1 is an exception to this rule , as it is translationally regulated during both meiosis ( Belloc and Mendez , 2008 ) and mitosis ( this study ) . However , in meiosis it is translationally activated through control of its poly ( A ) tail length by the RNA binding protein CPEB ( Belloc and Mendez , 2008 ) . This does not appear to be the case during mitosis , as mutation of all CPEB binding sites in the 3′UTR of Emi1 does not prevent translational repression in mitosis ( unpublished observation ) . Therefore , we conclude that the gene-specific translational control program in mitosis identified here is distinct from the meiotic translation program . What might be the function of translational repression during mitosis ? The average protein half-life is many hours ( Schwanhausser et al . , 2011 ) , while mitosis takes less than an hour , so inhibition of protein synthesis for this short period of time is not expected to have a major impact on overall protein levels . Our analysis of Emi1 offers a clue to the role of gene-specific translational repression during mitosis . During mitosis pre-existing Emi1 protein is inactivated through phosphorylation and ubiquitin-mediated degradation ( Guardavaccaro et al . , 2003; Margottin-Goguet et al . , 2003; Moshe et al . , 2011 ) . However , we find that in the presence of continued protein synthesis , degradation cannot remove all of the Emi1 , resulting in a small amount of residual Emi1 protein levels at the end of mitosis , which can interfere with full APC activation . In contrast , when new Emi1 protein synthesis is inhibited through translational repression , Emi1 is eliminated more completely ( Figure 5 ) . Transcriptional inhibition at the G2/M transition would not afford a similar effect , since protein synthesis rates would not decline substantially within the required time-scale due to relatively slow mRNA turnover . 10 . 7554/eLife . 07957 . 013Figure 5 . Model of Emi1 regulation in mitosis . Model of Emi1 regulation in G2 and M . In G2 , new Emi1 protein is continuously synthesized and existing Emi1 protein is stable allowing robust buildup of Emi1 protein levels and inhibition of APC/Cdh1 ( left ) . In mitosis , Emi1 synthesis is repressed and pre-existing Emi1 protein is inactivated through protein degradation and CDK1-dependent phosphorylation , resulting in full inhibition of Emi1 activity , allowing APC/Cdh1 activation in telophase ( middle ) ( Note- additional Emi1-independent mechanisms ( not depicted ) keep APC/Cdh1 inactive in ( pro ) metaphase ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07957 . 013 Our model for the inhibition of Emi1 during mitosis involves the synergistic effects of protein degradation along with the inhibition of protein synthesis via translational repression . However , of the proteins that are translationally repressed at the G2-to-M transition , very few are known to be degraded during mitosis , leaving open the question of why they are translationally repressed . One possibility is that these proteins are inactivated during mitosis through other post-translational mechanisms , such as phosphorylation . As newly synthesized proteins are unphosphorylated and thus active , inhibition of translation will limit the formation of such new , active protein and thus enhance post-translational protein inactivation . In summary , our work provides a genome-wide view of translationally controlled mRNAs in mitosis and provides a new hypothesis for the role of translational repression during mitosis; as a mechanism to augment post-translational protein inactivation . For ribosome profiling , cells were treated with cycloheximide before lysis for 2 min as previously described , which does not substantially alter overall mRNA read density ( Ingolia et al . , 2011 ) . Cells were lysed in lysis buffer ( 20 mM Tris pH 7 . 5 , 150 mM KCl , 5 mM MgCl2 , 1 mM dithiothreitol , 8% glycerol ) supplemented with 0 . 5% Triton X-100 , 30 U/ml Turbo DNase ( Ambion , Life Technologies , CA , United States ) and 100 µg/ml cycloheximide ( Sigma Aldrich , MO , United States ) ; ribosome-protected fragments were then isolated and sequenced as previously described ( Ingolia et al . , 2011 ) . Total RNA was isolated from cells using Trizol Reagent ( Ambion ) . Polyadenylated RNA was purified from total RNA using magnetic oligo ( dT ) beads . The resulting mRNA was modestly fragmented by partial hydrolysis in bicarbonate buffer so that the average size of fragments was ∼80 bp . The fragmented mRNA was separated by denaturing PAGE and fragments 50–80 nt were selected . The sequencing libraries were prepared and sequenced as previously described ( Ingolia et al . , 2011 ) . Prior to alignment , linker and poly ( A ) sequences were removed from the 3′ ends of reads . Bowtie v0 . 12 . 7 ( Langmead et al . , 2009 ) ( allowing up to 2 mismatches ) was used to perform the alignments . First , reads that aligned to human rRNA sequences were discarded . All remaining reads were aligned to the human ( hg19 ) genome . Finally , still-unaligned reads were aligned to known canonical mRNA . Reads with unique alignments were used to compute the total number of reads obtained for each transcript . FP alignments were assigned to specific P site nucleotides by using the position and total length of each alignment , calibrated from FPs at the beginning and the end of CDSes . FP and mRNA densities were calculated in units of reads per kilobase per million ( RPKM ) in order to normalize for gene length and total number of reads per sequencing run . The density of ribosome FPs is used as a measure of the rate of translation and TE is defined as the ratio of FP density/total mRNA . For each sample , two biological replicates were generated and the sum of the number of reads for each gene over the two replicates was calculated . We found that genes that had more than 200 reads total showed very strong reproducibility between replicates ( Supplementary file 1A ) . Therefore , we excluded from the analysis genes that had less than 200 reads . In addition , genes for which the RPKM value of the two biological replicates showed a difference of more than threefold , were excluded from further analysis . For genes that passed both filters , the RPKM values of the two biological replicates were averaged . R2 values and standard deviations for the ratio of the replicates are listed in Supplementary file 1 . For TE analysis , only genes for which mRNA samples had at least 200 reads were included . Unless stated otherwise , we used a cutoff of threefold difference between samples . For differences in TE , we used a >threefold difference in TE combined with a >twofold difference in FP RPKM . We reasoned that an increase in TE is only biologically meaningful if it is accompanied by an increase in FP RPKM , as this indicates that the total amount of protein synthesis is increased on the specific mRNA . A >threefold difference cutoff was >sixfold the standard deviation between replicates of all samples and was therefore highly significant . Functional classification of genes was done using the online webserver DAVID ( Huang da et al . , 2009 ) . The data described in this study have been deposited in NCBI's Gene Expression Omnibus ( GEO ) and are accessible through GEO Series accession number GSE67902 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE67902 ) . RPE-1 cells were grown in DMEM:F-12 medium supplemented with 10% FCS and antibiotics . For all stable cell lines , lentiviruses were made in 293 cells using the pHR vector and separate packaging vectors . For transduction , RPE-1 cells were incubated with virus for 24 hr . RO-3306 ( Axon medchem , The Netherlands ) was dissolved in DMSO and used at 6 μM for cell synchronization . CHX ( Sigma ) was dissolved in ethanol and used at 100 μg/ml . TMP ( Sigma ) was dissolved in DMSO and was used at 50 μM . All live cell imaging experiments were performed at 37°C on a Nikon TI widefield microscope using epifluorescence illumination , a 40× 0 . 95 NA air objective and a Hamamatsu sCMOS camera . Cells were grown and imaged in 96-well glass bottom plates and 1 hr before imaging normal growth medium was replaced with DMEM:F-12 with HEPES and without phenol red , supplemented with 10% FCS and antibiotics . Microscopes were controlled by Micro-Manager software ( Edelstein et al . , 2010 , 2014 ) and image analysis and quantification was performed using Micro-Manager and ImageJ . For image quantification , images were first corrected for unevenness in illumination using control images of a homogenously fluorescent slide . Images were then corrected for bleaching using the ImageJ plugin bleach correction ( by J . Rietdorf ) . Fluorescent intensities were then measured in ImageJ and corrected for background . For single molecule FISH , Stellaris FISH probe sets for Emi1 were used ( Biosearch technologies ) , consisting of 48 fluorescently labeled probes . Fixation and hybridization were performed as recommended by manufacturer . RPE-1 cells were synchronized by treatment with a CDK1 inhibitor , RO-3306 , which blocks cells in G2 ( see Figure 1A ) . Cells were either harvested at this point to obtain a G2 sample or the CDK1 inhibitor was removed to allow cells to enter mitosis . Upon removal of the inhibitor , ∼50% of cells progressed into mitosis in a highly synchronous manner . At 45 min after RO release , mitotic cells were isolated using mechanical shake-off and either harvested to generate the mitotic sample or re-plated to allow mitotic exit and then harvested as a G1 sample 3 hr after re-plating . Global translation rates were measured using a 10 min incubation with S35-methionine . Cells were then washed and lysed and total protein was analyzed on a denaturing gel . Total radioactive S35 incorporation was quantified by measuring the intensity of the entire lane on the gel . We generated an inducible , fast-maturing green fluorescent protein ( sfGFP ) by fusing the inducible degron DHFR ( Iwamoto et al . , 2010 ) to sfGFP . An NLS was also added , concentrating the protein in the nucleus , which simplifies analysis and increases the signal-to-noise ratio . The DHFR-sfGFP-NLS protein was continuously degraded in the absence of the small molecule stabilizer , but rapidly accumulated upon stabilization by trimethoprim ( TMP ) . Indeed , addition of TMP resulted in a rapid increase in GFP fluorescence , which was due to synthesis of new protein . Thus , the increase in GFP fluorescence is a good readout for the rate of translation of the reporter . However , different cells within a population usually contain different copy numbers of the reporter , and therefore showed different rates of GFP fluorescence increase . We therefore inserted a short viral P2A sequence after the GFP-NLS , which allows expression of a second independent protein from the same transcript with very high efficiency ( >90% , [Kim et al . , 2011] ) , due to the inability of the ribosome to form a peptide bond at the end of the P2A sequence . Downstream of the P2A sequence , we inserted a NLS-mCherry protein , which is insensitive to the degron , as the NLS-mCherry protein is physically separated from the DHFR-GFP-NLS protein . Levels of mCherry can thus be used to normalize for the total amount of reporter per cell ( Figure 3—figure supplement 1 ) . Aurora A , Emi1 , PP1γ and PP2aβ were PCR amplified from a cDNA library generated from RPE-1 cells . Aurora A was cloned upstream of GFP into the pHR lentiviral expression vector using MluI-NotI restriction sites ( hereafter called pHR vector ) with a truncated SV40 promoter to reduce expression levels . Emi1 was inserted downstream of mCherry in the pHR vector using BamHI-NotI . For the pHR-mCherry-P2A-PP1γ , mCherry was first PCR amplified and a P2A sequence was inserted into the 3′ primer with an extra RsrII restriction site . mCherry-P2A was then inserted in pHR using BstX1-NotI , after which PP1γ was inserted downstream of the P2A sequence with RsrII-NotI . For pHR-mCherry-P2A-strepII-PP2aβ , mCherry was first PCR amplified with P2A-StrepII-tag sequence , as well as an extra 3′SpeI site in the 3′ primer . mCherry-P2A-strepII was then inserted into pHR , after which PP2aβ was inserted in this vector using SpeI-Not1 . The fluorescence-based translation reporter was cloned using fusion PCR of three parts: 1 . DHFR ( Y100I ) , 2 . sfGFP-NLS-P2A 3 . NLS-mCherry . The product was cloned into the pHR vector using BstXI-NotI . All 5′ UTRs were inserted using BstXI-BsiWI and 3′UTRs were inserted using RsrII-NotI . Sequences of the entire reporter and all 5′ and 3′ UTRs used in this study can be found in the supplemental methods section . Primers to amplify the UTRs used in this study were based on the RNA-seq data to represent the most common UTR splice variant in RPE-1 cells .	The human body contains billions of cells , most of which formed via a process called mitosis in which a single cell divides to produce two new daughter cells . Actively dividing cells pass through a series of events ( or phases ) that are collectively known as the cell cycle . These phases allow the cell to grow in size , copy its genetic material , and then make preparations for cell division before taking the final decision to divide . Many proteins are involved in regulating the cell cycle and each protein has a particular role in specific phases . The levels of these proteins in cells may change during the cycle , which is often crucial to allow the cell to progress to the next phase . For example , cells need a group of proteins called the anaphase-promoting complex ( or APC for short ) to destroy other specific proteins at the end of mitosis . Another way in which the amount of protein in a cell can be adjusted is by controlling how much new protein is made during a process known as translation . During this process , a molecule called a messenger RNA ( mRNA ) —which contains information copied from a particular gene—is used as a template to assemble a new protein . However , it is not clear whether regulation of translation is involved in control of the cell division . Tanenbaum et al . now address this question using a technique called ribosome profiling to measure the translation of individual mRNA molecules . The experiments analysed the changes in protein production before , during and after mitosis . The overall level of translation of all the mRNAs was about 35% lower during mitosis . However , some mRNAs in particular experienced a very large reduction in the level of translation ( between three- and ten-fold less than the levels before mitosis ) . One example of an mRNA whose translation is turned off in mitosis is the mRNA that makes a protein called Emi1 . It is known from previous work that Emi1 inhibits the activity of the APC . Therefore , Emi1 needs to be inactivated in mitosis so that the APC can become active and promote progression to the next phase of the cell cycle . It was previously shown that Emi1 is destroyed during mitosis to allow the APC to operate . Tanenbaum et al . found that translation of the Emi1 mRNA must also be suppressed during mitosis in order to keep Emi1 protein levels very low and allow the APC to become fully active . These findings uncover a new role for the control of protein production in regulating the cell cycle . The next challenge will be to find out whether suppression of translation is also used in other biological systems where proteins need to be rapidly inactivated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2015	Regulation of mRNA translation during mitosis
Regulation of cell wall assembly is essential for bacterial survival and contributes to pathogenesis and antibiotic tolerance in Mycobacterium tuberculosis ( Mtb ) . However , little is known about how the cell wall is regulated in stress . We found that CwlM , a protein homologous to peptidoglycan amidases , coordinates peptidoglycan synthesis with nutrient availability . Surprisingly , CwlM is sequestered from peptidoglycan ( PG ) by localization in the cytoplasm , and its enzymatic function is not essential . Rather , CwlM is phosphorylated and associates with MurA , the first enzyme in PG precursor synthesis . Phosphorylated CwlM activates MurA ~30 fold . CwlM is dephosphorylated in starvation , resulting in lower MurA activity , decreased cell wall metabolism , and increased tolerance to multiple antibiotics . A phylogenetic analysis of cwlM implies that localization in the cytoplasm drove the evolution of this factor . We describe a system that controls cell wall metabolism in response to starvation , and show that this regulation contributes to antibiotic tolerance . Mycobacterium tuberculosis ( Mtb ) , a bacterium that causes 1 . 5 million deaths a year ( Sharma et al . , 2006; World Health Organization , 2014 ) , differs from many bacterial pathogens in its infection strategy , which depends on an unusual cell envelope ( Cambier et al . , 2014 ) . The importance of this structure is evident from the fact that at least a quarter of the genes in the Mtb genome are involved in its construction and regulation ( Doerks et al . , 2012 ) . The mycobacterial cell wall consists of peptidoglycan ( PG ) covalently bound to a layer of arabinogalactan sugars , which is in turn covalently bound to a layer of mycolic acid lipids that make up an outer membrane ( Alderwick et al . , 2015; Cambier et al . , 2014 ) . While the chemistry of this cell envelope is increasingly well defined , its regulation is poorly understood . Regulated cell wall changes are essential for Mtb to cause disease ( Doerks et al . , 2012; Sharma et al . , 2006 ) and are thought to contribute to phenotypic antibiotic tolerance . TB therapy requires 6–12 months of drug treatment . This is likely due , in part , to drug tolerance like that observed in stressed , non-growing Mtb cultures ( Wallis et al . , 1999 ) . Because stressed Mtb also exhibits cell wall changes such as thickening , differential staining and chemical alterations ( Bhamidi et al . , 2012; Cunningham and Spreadbury , 1998; Seiler et al . , 2003 ) , it seems probable that the cell wall changes contribute to antibiotic tolerance . Importantly , starved , antibiotic tolerant Mtb cells have been shown to be less permeable to antibiotics ( Sarathy et al . , 2013 ) . However , the regulatory mechanisms that induce cell wall changes in response to stress and contribute to impermeability and tolerance have not been described . The peptidoglycan ( PG ) layer provides protection and shape-defining structure to cells of almost all bacterial species . This layer is constructed by transpeptidases and glycosyltransferases , which attach new precursors to the existing cell wall; and several types of catabolic PG hydrolases , which break bonds in the existing PG . PG enzymes must be tightly regulated to promote cell growth and septation without compromising the wall integrity , and to restructure the cell wall to withstand stresses ( Kieser and Rubin , 2014 ) . Accordingly , a large number of regulators in the cytoplasm , inner membrane and periplasm coordinate and control the activities of the PG enzymes , either directly or indirectly ( Kieser and Rubin , 2014; Typas et al . , 2011 ) . In addition to the dedicated regulators , many PG synthases and hydrolases work together in complexes and regulate each others’ enzymatic activity through protein-protein interactions ( Banzhaf et al . , 2012; Hett et al . , 2010; Smith and Foster , 1995 ) . Notably , some PG hydrolases have lost their enzymatic activity and function only as regulators: EnvC in E . coli is missing catalytic residues from its active site but contributes to cell septation by activating the PG amidases AmiA and AmiB ( Uehara et al . , 2010; Yang et al . , 2011 ) . In this work we study the predicted PG hydrolase , CwlM , and find that its essential function is regulatory rather than enzymatic . We find that CwlM is located in the cytoplasm , is phosphorylated by the essential Serine Threonine Protein Kinase ( STPK ) PknB , and functions to stimulate the catalytic activity of MurA , the first enzyme in the PG precursor synthesis pathway . In nutrient-replete conditions CwlM is phosphorylated . Using in vitro biochemistry we show that phosphorylated CwlM ( CwlM~P ) increases the rate of MurA catalysis by ~30 fold . In starvation , CwlM is dephosphorylated and in this state does not activate MurA , which has very low activity alone ( Xu et al . , 2014 ) . Importantly , we find that over-activation of MurA in the transition to starvation causes increased sensitivity to multiple classes of antibiotics . Finally , a phylogenetic analysis implies that CwlM protein evolution was driven by localization to the cytoplasm . To confirm the essentiality of cwlM in Msmeg we constructed a strain ( See Supplementary file 1 for full descriptions of all strains ) , Ptet::cwlM , in which the only copy of cwlM is under control of an anhydrotetracyline ( Atc ) -inducible promoter . Depletion of CwlM by removing Atc results in cell death ( Figure 1B ) . Microscopy of CwlM-depleted cells shows that they are short , implying a defect in elongation . To assess polar elongation ( Aldridge et al . , 2012; Thanky et al . , 2007 ) , we stained cells with an amine reactive dye ( ARD ) ( Aldridge et al . , 2012 ) and cultured them to allow new , unstained polar cell wall to form before imaging ( Figure 1C ) . We found that CwlM-depleted cells fail to elongate normally ( Figure 1D , E , F ) . 10 . 7554/eLife . 14590 . 003Figure 1 . CwlM’s essential function promotes polar growth , but not through enzymatic activity . ( A ) Alignment of the active-site proximal regions of the enzymatic domain of PG amidases . The Zn2+-coordinating residues are boxed: the conserved aspartate/ glutamates in green , and the degenerate histidine/ arginines in red . ( B ) Colony forming units ( CFU ) of the Ptet::cwlM ( CB82 ) strain grown with ( +CwlM ) or without ( −CwlM ) Atc inducer from Time 0 . Data from all experiments throughout is from three biological replicates and error bars are the standard deviation , unless otherwise noted . ( C ) Cartoon of the amine reactive dye ( ARD ) staining method showing polar growth after staining , with a diagram of how cell length and polar growth were measured for Figures 1EF and 2EF . ( D ) Overlaid micrographs of the Ptet::cwlM strain grown with ( +CwlM ) or without ( −CwlM ) Atc for 9 hr , stained with ARD , grown 3 hr and imaged with a 488/530 filter ( green ) and in phase ( red and black ) . Representative cells from several images used in E and F were pasted together , the scale is conserved between images . ( E ) The length of ~230 cells from each condition in ( D ) . Two biological replicates were used for all microscopy experiments , unless noted . ( F ) The length of the longer unstained pole of the cells in ( E ) , line indicates the median in ( E ) and ( F ) . **** p value <0 . 0001 . ( G ) Growth curves showing the OD660 of WT ( L5::cwlM , CB236 ) and amidase-ablated ( L5::cwlM D209A E331A , CB239 ) strains growing in 7H9 , and a bar graph showing the calculated doubling times of these strains ( inset ) . n . s . – not significant . The p value was 0 . 51 by the student’s t-testDOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 00310 . 7554/eLife . 14590 . 004Figure 1—figure supplement 1 . Overexpression of CwlM . Colony forming units of Msmeg strains overexpressing wild-type ( CB193 ) , the amidase ablated allele of cwlM ( CB194 ) , and the amidase domain alone ( CwlM ∆PG binding domains , CB202 ) , and an α-His western blot showing that these proteins are expressed . Proteins were expressed from Atc-inducible promoters on episomal vectors . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 004 To determine if CwlM requires amidase activity for its essential function , we mutated the active site aspartate and glutamate that coordinate the catalytic Zn2+ ( Prigozhin et al . , 2013 ) and have been shown to be required for catalysis in related proteins ( Prigozhin et al . , 2013; Shida , 2001 ) . We performed allele swapping at the L5 phage integrase site ( Pashley and Parish , 2003 ) to replace the wild-type ( WT ) cwlM with the mutant allele cwlM E209A D331A . cwlM E209A D331A has zero out of four conserved Zn2+-coordinating residues , but is able to complement the WT allele in a growth curve assay ( Figure 1G ) . Thus , CwlM is essential for cell survival and elongation in Msmeg , but its presumed amidase active site does not appear to be essential . CwlM may therefore have another function . A proteomic screen in Mtb found that CwlMTB is phosphorylated at T43 and T382 ( Prisic et al . , 2010 ) . To determine whether the Msmeg CwlM is also phosphorylated , we constructed a strain with an epitope-tagged allele ( cwlM::FLAG ) , immunoprecipitated the protein from Msmeg lysate and separated it with 2D gel electrophoresis . We found several CwlM species with different isoelectric points , a hallmark of phosphorylation ( Figure 2A ) . Mass spectrometry of the gel spots identified phosphorylation at T35 and T374 ( equivalent to T43 and T382 in the Mtb protein ) , as well as T376 and T378 , and acetylation at K362 and K369 . Most of the modifications are found at the C-terminal tail of the protein , which is highly conserved across actinomycetes ( Figure 2B ) , despite being outside of a conserved domain . To assess whether phosphorylation is important for function , we replaced WT cwlM at the L5 site with mutants in which phosphorylated threonine residues were replaced by alanines . Most mutants grew normally but the L5::cwlM T374A strain grew slowly ( Figure 2C; Figure 2—figure supplement 1A ) , despite containing normal amounts of CwlM protein ( Figure 2D ) . The L5::cwlM T374A cells were short ( Figure 2E ) and defective for polar elongation ( Figure 2F ) . Because the phenotypes of the L5::cwlM T374A strain are similar to the CwlM depletion ( Figure 1 ) , it seems likely that absence of phosphorylation on T374 inhibits the essential function of CwlM , though the T374A mutation could interrupt CwlM function in another way . 10 . 7554/eLife . 14590 . 005Figure 2 . CwlM is phosphorylated and cytoplasmic . ( A ) 2-D gel of CwlM-FLAG from Msmeg ( CB100 ) with the isoelectric point ( pI ) indicated on the bottom . ( B ) Weblogo diagrams of the C-terminus of CwlM homologues . Acetylation ( Ac ) and phosphorylation ( PO4 ) sites are indicated . The colors of the three phosphates match the colors of the corresponding phospho-ablative strains in C , E and F . ( C ) Doubling times of L5::cwlM-FLAG WT ( CB300 ) and phospho-ablative mutants ( T374A = CB300; T376A = CB345; T378A = CB348 ) . All multiple comparisons throughout were performed by One-way ANOVA with a Dunnett multiplicity correction . The adjusted p values for each mutant compared to WT are: T374A =< 0 . 0001; T376A = 0 . 88; T378A = 0 . 15 . ( D ) α-strep western blot of L5::cwlM-strep WT ( CB663 ) and phospho-ablative mutants ( CB666 , 669 , 672 ) . A background band on the same blot probed with α-FLAG is used as a control . Westerns of the L5::cwlM-FLAG strains could not be used because of high background . ( E ) Cell length of 200–300 cells from L5::cwlM-FLAG WT and phospho-ablative mutants . The adjusted p values for each mutant compared to WT are: T374A =< 0 . 0001; T376A = 0 . 99; T378A = 0 . 01 . ( F ) Length of the longer unstained pole of cells in ( E ) . Cells were stained with ARD and grown for three hours before imaging . The adjusted p values for each mutant compared to WT are: T374A =< 0 . 0001; T376A = 0 . 59; T378A = 0 . 36 . ( G ) Substituted cysteine accessibility . α-strep western of cells with strep-tagged Pgm ( CB886 , MSMEG_4579 , cytoplasmic control ) , lprG ( CB706 , Rv1411 , periplasmic control ) and CwlM1cys ( CB457 ) that were cysteine-blocked with MTSEA ( membrane permeable ) or MTSET ( membrane impermeable ) then alkylated at unblocked cysteines . Control samples: ( − ) = not alkylated or blocked , ( + ) = alkylated but not blocked . % alkylation ( %alk . ) = high band/ total for each sample . The periplasmic localization score = ( %alk . MTSEA/% alk . MTSET ) and is indicated below the gel for each protein ± the 95% confidence interval from two biological replicates . Scores < 0 . 5 are cytoplasmic , near 1 are periplasmic . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 00510 . 7554/eLife . 14590 . 006Figure 2—figure supplement 1 . Growth rates of strains with different cwlM alleles . ( A ) Doubling times of ∆cwlM Msmeg strains with different cwlM-FLAG alleles at the L5 phage integrase site . WT = CB300; T35A = CB265; T374A = CB319; T376A = CB345; T378A = CB348; K362A = CB351; K369A = CB354 . p values for each mutant against the WT were calculated with the students t test: all p values were insignificant except T374A > 0 . 0001 . ( B ) Doubling times of wild-type M . smegmatis mc2155 ( CB1 ) , mc2155 ∆cwlM L5::cwlM-FLAG ( CB300 ) and mc2155 ∆cwlM L5::cwlM ( CB236 ) . p values: CB1 vs . CB300 > 0 . 0001; CB300 vs . CB236 = 0 . 8 . ( C ) Doubling times of mc2155 ∆cwlM L5::cwlM-strep ( CB663 ) and mc2155 ∆cwlM L5::cwlM K362A K369A T376A T378A-strep ( CB675 ) , p value = 0 . 033 . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 006 Peptidoglycan hydrolases and their direct regulators are found in the periplasm; however , CwlM lacks classical sec or tat secretion signal sequences ( Bagos et al . , 2010; Petersen et al . , 2011 ) . Moreover , phosphorylation of periplasmic proteins has not been described . To determine the compartmental location of CwlM , we used the Substituted Cysteine Accessibility Method ( SCAM ) ( Karlin and Akabas , 1998 ) . We measured the accessibility of a cysteine on CwlM to agents with different membrane permeability and found that CwlM is only accessible to agents that can reach the cytoplasm ( Figure 2G ) . Thus , CwlM is a cytoplasmic protein whose essential function is likely activated by phosphorylation . Mtb has 11 STPKs , which have been shown to phosphorylate many proteins involved in stress and growth ( Molle and Kremer , 2010 ) . To identify the cognate kinase of CwlMTB , we performed phospho-transfer profiling ( Baer et al . , 2014 ) using proteins from the pathogen Mtb , which have high sequence conservation with the Msmeg homologues . We expressed and purified his-CwlMTB and his-MBP ( Maltose Binding Protein ) fusions to the Mtb kinase domains ( KD ) of PknA , B , D , E , F , H , J , K and L , and measured phosphorylation of his-CwlMTB using Western blotting with an α-phosphothreonine ( α-thr~P ) antibody . We found that his-MBP-PknBTB ( KD ) phosphorylates his-CwlMTB more rapidly than the other kinases ( Figure 3A; Figure 3—figure supplement 1A , B ) . We used mass spectrometry to show that in vitro , his-MBP-PknBTB phosphorylates his-CwlMTB at threonines 382 , 384 and 386 ( equivalent to T374 , 376 and 378 in M . smegmatis , Figure 3—figure supplement 1C ) . Thus , the phosphorylation sites in the C-terminal tail of CwlM are conserved in Mtb and Msmeg , which supports the idea that the strong sequence conservation of the CwlM C-terminal tail across actinomyecetes ( Figure 2B ) is functionally relevant . 10 . 7554/eLife . 14590 . 007Figure 3 . CwlM is phosphorylated by PknB . ( A ) α-thr~p western blots of in vitro kinase reactions with the kinase domains of 9 STPKs from Mtb fused to MBP and his-CwlM from Mtb . Reactions were stopped at 2 min . Longer time points are in Figure 3-figure supplement 1B . ( B ) α-thr~p and α-FLAG western blot of lysates from a strain expressing CwlM-FLAG in a PknBTB overexpression ( CB418 ) background . PknBTB was uninduced or induced for 30 min with 100 ng/ml Atc . Ratio = signal of ( α-thr~p / α-FLAG ) PknB OE / ( α-thr~p / α-FLAG ) uninduced . Experiments in ( A ) and ( B ) were both performed twice , representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 00710 . 7554/eLife . 14590 . 008Figure 3—figure supplement 1 . Phosphorylation of CwlM . ( A ) α-phospho-threonine western blot of kinase reactions of the kinase domains ( KD ) of STPKs from Mtb fused to his-MBP incubated overnight at room temperature with his-GarATB , a non-specific substrate . ( B ) α-phospho-threonine western blot of kinase reactions of his-CwlMTB incubated with the kinase domains of STPKs from Mtb fused to his-MBP . ( C ) Alignment of CwlM from Mtb and Msmeg . Yellow boxes indicate the position of the PG binding 1 domains , green boxes indicate the position of the PG amidase domain . Residues highlighted in red were found to be phosphorylated in CwlM-FLAG pulled down from Msmeg cell lysates . Residues in orange were acetylated in these same lysates . Residues highlighted in blue were found to be phosphorylated on his-CwlMTB expressed and purified from E . coli and incubated with his-MBP-PknB ( KD ) TB . All the modifications were identified by mass spectrometry . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 008 To test whether PknB affects CwlM phosphorylation in vivo we immunoprecipitated CwlM-FLAG from an Msmeg strain before and after induction of PknBTB overexpression , and determined the degree of phosphorylation by comparing reactivity with α-thr~P and α-FLAG antibodies ( Figure 3B ) . We found that PknBTB overexpression increases CwlM phosphorylation . Thus , PknB is likely the primary kinase of CwlM . This is consistent with the known essential role of PknB in phosphorylating proteins involved in cell growth and division ( Fernandez et al . , 2006; Kang , 2005 ) . PknE may also contribute to CwlM phosphorylation . Phosphorylated CwlM appears to be required for normal cellular growth . To identify the molecular role of CwlM , we screened for spontaneous suppressor mutants that grow rapidly despite carrying only the growth-restrictive , phospho-ablative cwlM T374A allele . We passaged this strain until we isolated rapidly growing mutants , which were mapped using whole genome sequencing . Three of five suppressor strains contained an identical S368P mutation in the gene murA . MurA is a cytoplasmic enzyme that catalyzes the first dedicated step in PG precursor synthesis: transfer of an enolpyruvyl group from phosphoenol pyruvate ( PEP ) onto UDP-N-acetylglucosamine ( UDP-GlcNAc ) ( Marquardt et al . , 1992 ) . The S368 residue is near the active site , according to an alignment with a structure of MurA from L . monocytogenes ( Halavaty et al . , 2011 ) ( Figure 4A ) . 10 . 7554/eLife . 14590 . 009Figure 4 . CwlM binds to and regulates MurA . ( A ) Crystal structure of MurA from L . monocytogenes ( Halavaty et al . , 2011 ) . Orange: alanine that corresponds to S368 in Msmeg and Mtb MurA; gray: catalytic cysteine ( aspartate in Msmeg and Mtb . ) . ( B ) Fitness of cwlM alleles in murA WT ( CB737 ) and S368P ( CB762 ) backgrounds , assessed by the percentage of colonies in which the WT cwlM allele was replaced by the indicated allele ( WT = pCB277; T374A = pCB255; ∆CT = pCB557; ∆cwlM = pCB558 ) . Four biological replicates of each transformation were performed , and 96–198 colonies were counted for each replicate . The Sidak correction for multiple comparisons was used to calculate p values using one-way ANOVA . The adjusted p values among the murA WT strains were calculated compared to cwlM WT: T374A =< 0 . 0001; ∆CT =< 0 . 0001; ∆cwlM =< 0 . 0001 . The adjusted p values among the murA S368P strains were calculated compared to cwlM WT: T374A = 0 . 99; ∆CT = 0 . 006; ∆cwlM =< 0 . 0001 . ( C ) α-FLAG and α-strep western blots of immunoprecipitates from the indicated strain ( top three rows , from top: CB300 , CB737 , CB779 ) with the indicated antibody-conjugated beads ( bottom two rows ) . The pulldown was performed twice , a representative image is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 00910 . 7554/eLife . 14590 . 010Figure 4—figure supplement 1 . Growth rates of strains with different cwlM and murA regulatory alleles . Doubling times of strains . cwlM WT murA WT = CB779; cwlM WT murA S368P = CB782; cwlM T374A murA WT = CB319; cwlM T374A murA S368P = CB785 . In the experiment described in Figure 4 , we were not able to attain any cwlM T374A murA WT transformants , so the strain with this genotype is not isogenic with the other strains presented here . We were only able to attain colonies with the cwlM T374A murA WT genotype when the L5 allele swap was conducted in such a way that the correct swap was genetically forced . CB319 and the other strains presented in Figure 2 were made by using a parental strain ( CB233 ) in which the only copy of cwlM was under the control of an Atc-inducible promoter on an L5 vector that contained the tetR gene: this strain is Atc-dependent , because TetR will repress transcription of cwlM unless Atc is added . The alternate alleles of cwlM were under the control of Atc-inducible promoter on L5-integrating vectors that do not contain the tetR gene; they were transformed into CB233 and plated on plates that do not contain Atc . If the original L5 vector remains at the L5 site in addition to the new vector with the alternate allele , the TetR expressed from the original vector will repress transcription of both alleles of cwlM and cause cell death: thus , the loss of the original vector is forced , and we were able to recover colonies even with very slow growth rates . In the transformations described in Figure 4 , resulting in strains CB779 , 782 and 785 , there was no tetR on either of the L5 vectors; thus , the swap was not forced , and at least half of the colonies from any transformation had both the original L5 vector with WT cwlM and the newly-transformed vector . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 010 To confirm that the murA S368P allele suppresses the growth defect of the cwlM T374A allele in a different genetic background , we performed L5 allele swapping of cwlM in strains encoding either WT murA or murA S368P . In these strains a WT allele of cwlM is encoded on an L5-integrated construct marked with the nourseothricin resistance gene . We transformed these cells with kanamycin-marked L5 vectors with alternate alleles of cwlM . Appropriate recombination can be scored by gain of the kanamycin marker and loss of the nourseothricin marker . We found that the cwlM T374A allele and a deletion of the C-terminal tail are poorly tolerated in a WT murA background , but can be recovered at near WT levels in a murA S368P background . The entire cwlM gene cannot be deleted in either background ( Figure 4B ) . The strain carrying murA S368P alone grows at the same rate as WT , the strain with both mutant alleles has a slight growth defect , but grows much more rapidly than the strain with cwlM T374A and WT murA ( Figure 4—figure supplement 1 ) . These results confirm that cwlM and murA interact genetically . To test whether the proteins physically interact , we immunoprecipitated proteins from a strain containing cwlM-FLAG and murA-strep with both α-FLAG and α-strep beads . We found that CwlM-FLAG could be precipitated with α-strep beads , but only in the presence of MurA-strep , and vice versa ( Figure 4C ) . Thus , there is a genetic link between cwlM and murA , and the two proteins interact . We hypothesized that CwlM~P activates MurA and that MurA is less active when CwlM is less phosphorylated , resulting in the inhibition of cell growth seen in the cwlM T374A mutant . This hypothesis is based on the observation that the phospho-ablated cwlM T374A is suppressed by murA S368P ( Figure 4B ) , which is likely to be a gain-of-function mutation because murA is an essential gene ( Griffin et al . , 2011 ) and the mutation does not affect growth rate ( Figure 4—figure supplement 1 ) . To test this model we expressed and purified his-MurATB and his-CwlMTB from E . coli and performed kinetic assays of his-MurATB activity by measuring the accumulation of the MurA product , enolpyruvyl-UDP-N-acetylglucosamine ( EP-UDP-GlcNAc ) , by HPLC ( Figure 5A ) . We compared the activity of his-MurATB alone and in the presence of his-CwlMTB and his-CwlMTB~P phosphorylated on T382 , 384 and 386 by his-MBP-PknBTB ( Figure 3—figure supplement 1B ) . We found that the reaction rate of his-MurATB is 20–40 times faster in the presence of equimolar his-CwlMTB~P ( Figure 5 ) than it is alone or with unphosphorylated his-CwlMTB . We found that nearly all his-CwlMTB incubated with his-MBP-PknBTB is phosphorylated on at least one site ( Figure 5—figure supplement 1F ) ; however , because there are three phosphorylation sites , it is likely a heterogeneous mixture . his-CwlMTB with different degrees of phosphorylation may stimulate MurA activity differently , thus the rates in Figure 5D represent a population average of the rates from his-CwlMTB in different states . 10 . 7554/eLife . 14590 . 011Figure 5 . CwlM~P activates MurA in vitro . ( A ) An HPLC trace from the MurATB kinetic assays , with substrate UDP-GlcNAc and product EP-UDP-GlcNAc indicated . Assays were done on his-MurATB alone , his-MurATB with equimolar his-CwlMTB~P in an active kinase reaction ( +ATP ) with his-MBP-PknBTB ( KD ) ( a his-Maltose binding protein fusion to the kinase domain ( KD ) of PknB from Mtb ) , or his-MurATB with his-CwlMTB in an inactive kinase reaction ( -ATP ) with his-MBP-PknBTB ( KD ) . ( B ) Rate of MurA activity vs . the concentration of phosphoenol pyruvate ( PEP ) . Lines are the fit to the Michaelis Menten model . ( C ) Rate of MurA activity vs . the concentration of UDP-GlcNac for each MurA reaction condition . MurA and MurA+CwlM did not follow Michaelis Menten kinetics , ( lines only connect the dots ) , for MurA+CwlM~P the line is the Michaelis Menten fit . The assays shown in ( B ) and ( C ) were done twice with two separate preps of his-MurATB and his-CwlMTB . Data from one replicate is shown here , full data in Figure 5 - figure supplement 1A–D . ( D ) Michaelis Menten kinetic constants from experiments in ( B ) and ( C ) . Constants are the average ± the 95% confidence interval from the two replicates . ( E ) Amount of EP-UDP-GlcNAc produced after 30 min at 37°C in reactions with 2 mM PEP and 15 mM UDP-GlcNAc and his-MurATB at 20 µg/ml; his-CwlMTB at 20 . 2 µg/ml; his-MBP-PknBTB ( KD ) at 2 µg/ml and ATP at 1 mM where indicated . his-CwlMTB was incubated with his-MBP-PknBTB and ATP for 1 hr at RT before initiating the MurA reaction . T382A indicates that the singly-phosphoablative mutant protein his-CwlMTB T382A was used . 3T->A indicates that the triply-phosphoablative mutant protein his-CwlMTB T382A T384A T386A was used . Experiment was performed 2 times . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 01110 . 7554/eLife . 14590 . 012Figure 5—figure supplement 1 . MurA enzyme kinetics . Rates of MurA activity vs . [PEP] concentration with ( A ) the first preparation of his-MurATB and his-CwlMTB . ( B ) the second prep . Rates of MurA activity vs . [UDP-GlcNAc] concentration with ( C ) the first preparation of his-MurATB and his-CwlMTB and ( D ) the second prep . Data in ABC and D were fit to the Michaelis Menten formula where possible , the lines indicate this fit . In C and D , the data for MurA alone and MurA +CwlM could not be fit to the Michaelis Menten formula . These reactions apparently operate under atypical kinetic parameters . ( E ) Concentration of EP-UDP-GlcNAc formed in a 30 min incubation at 37°C of 2 mM PEP , 15 mM UDP-GlcNAc , 20 µg/ml his-MurATB and his-CwlMTB~P in an active kinase reaction at a range of CwlM:MurA molar ratios . ( F ) Two-dimensional gel of purified His-CwlMTB after 1 hr incubation with his-MBP-PknBTB in the absence ( top ) or presence ( bottom ) of ATP . The shift of the entire CwlM spot to a higher pI indicates that ~100% of the protein is phosphorylated . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 012 his-CwlMTB with phosphoablative mutations does not appreciably stimulate his-MurATB activity after incubation with his-MBP-PknBTB and ATP ( Figure 5E ) . his-CwlMTB alone cannot synthesize EP-UDP-GlcNAc; MurA is not thought to be phosphorylated ( Prisic et al . , 2010 ) and his-MBP-PknBTB cannot stimulate MurA activity in the absence of his-CwlMTB ( Figure 5E ) . We measured the stoichiometry of the CwlM:MurA reaction and found that his-MurATB is maximally activated in a reaction with a 2-fold molar excess of fully phosphorylated his-CwlMTB~P ( Figure 5—figure supplement 1E , F ) . Thus , CwlM~P activates MurA to initiate PG synthesis . These data with Mtb proteins confirm our model based on the genetic data in Msmeg , and show that CwlMTB~P is a direct activator of MurATB . MurA from other species have been studied in vitro and shown to have much more rapid kinetics and lower Km values than isolated MurATB ( Figure 5B , C; Figure 5—figure supplement 1; Supplementary file 1D [Xu et al . , 2014] ) . Why would an essential housekeeping enzyme evolve to be almost inactive by default and require activation by another factor ? We hypothesized that this MurA regulatory pathway could exist to restrict PG synthesis in stresses such as nutrient starvation , because cell length decreases in starvation ( Figure 6—figure supplement 1A ) and when CwlM is deactivated ( Figures 1D , E , 2E ) . To test this , we immunoprecipitated CwlM-FLAG from Msmeg cultures in log phase and starvation and measured the degree of phosphorylation . We found that CwlM phosphorylation is quickly reduced upon starvation in PBS ( Figure 6A ) . 10 . 7554/eLife . 14590 . 013Figure 6 . MurA is downregulated in starvation , contributing to drug tolerance . ( A ) α-thr~P western blot showing the level of CwlM phosphorylation in the cwlM::FLAG strain ( CB100 ) in log . phase and upon starvation in PBS . Ratio = signal of ( α-thr~p / α-FLAG ) starvation / ( α-thr~p / α-FLAG ) log phase ( B ) Amount of EP-UDP-GlcNAc produced after 30 min in a reaction with 20 µg/ml of his-MurAMSWT or his-MurAMSS368P , 2 mM PEP , 15 mM UDP-GlcNAc and either no or equimolar his-CwlMTB or his-CwlMTB~P . ( C ) Fluorescent intensity of WT ( CB779 ) and murA S368P ( CB782 ) cells pulse stained with TADA in log . phase HdB culture ( t = 0 ) and at time points after initiation of starvation in PBS+tween 80 , as measured by flow cytometry . The median fluorescence for 10K+ cells is shown . On the right are representative images of cells from the t = 0 ( log . phase ) and t = 24 ( starved ) time points . Phase and red fluorescent images of each are shown . The images were taken with identical exposure settings and the brightness and contrast was adjusted to identical levels . Pictures of several cells from images processed identically were pasted together . The scale bar applies to all the images . ( D ) Colony forming units per ml of strains with the murA WT ( CB779 ) and S368P ( CB782 ) alleles after transfer to PBS+tween80 starvation media and treatment with 100 µg/ml isoniazid , 20 µg/ml meropenem or 50 ug/ml rifampin . The p value for the indicated time point in each experiment was calculated by the student’s t test . Isoniazid = 0 . 0087; meropenem = 0 . 0082; rifampin = 0 . 0067 . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 01310 . 7554/eLife . 14590 . 014Figure 6—figure supplement 1 . Phenotypes of murA S368P . ( A ) Phase micrographs and quantification of Msmeg mc2155 ( CB1 ) cell lengths in log . phase 7H9 and after 5 hr in HdB-C carbon starvation media . 500 cells from single cultures were measured for each condition . p<0 . 0001 by the student’s t test . ( B ) . α-strep and α–Hsp65 western blots of A280-normalized lysates from murA WT ( CB779 ) and S368P ( CB782 ) cultures starved for four hours in Hdb-C . The murA allele in both these strains has a C-terminal strep tag . ( C ) Controls for flow cytometry . Cells were grown in HdB as in the t = 0 time point of Figure 6C , and either stained and then fixed as in Figure 6C , or fixed first and then stained , or fixed but not stained , and then subjected to flow cytometry . The median fluorescence for 10K+ cells is shown . ( D ) Fluorescence intensity of TADA stained murA WT and murA S368P cell fractions . Cells were extracted first with water and then SDS . The remaining cell wall pellets were photographed ( inset ) , and all fractions measured by fluorimetry . Two replicates of each culture were used . ( E ) Colony forming units ( CFU ) of murA WT and S368P strains after being grown to log phase in 7H9 , then pelleted and resuspended in PBS+tween80 . ( F ) CFU of murA WT and S368P strains grown to log phase in 7H9 then treated with either 50 µg/ml isoniazid , 20 µg/ml rifampin or 5 µg/ml meropenem . ( G ) Colony forming units of murA WT and S368P strains after being grown to log phase in 7H9 , then pelleted and resuspended in in PBS+tween80 . At the 18 hr time point either 100 µg/ml isoniazid , 40 µg/ml rifampin or 20 µg/ml of meropenem was added . Error bars represent the standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 014 These results , together with the biochemical data and the observation that MurA protein levels do not change substantially in starvation ( Figure 6—figure supplement 1B ) , suggest that phosphorylation of CwlM converts MurA to an 'on' state and that during starvation , decreased CwlM phosphorylation blocks further PG synthesis - because MurA alone or in the presence of unphosphorylated CwlM has very low activity ( Figure 5 ) . To assess the importance of this post-translational regulation of MurA on mycobacterial physiology , we further studied the murA S368P mutant . The genetic data predict that murA S368P is active without CwlM phosphorylation . To test this , we expressed and purified his-MurAMS WT and S368P and measured their activity alone and with his-CwlMTB and his-CwlMTB~P . We found that his-MurAMSS368P is more active than the WT protein in all conditions by about 2-fold ( Figure 6B ) . This suggests that the mutant enzyme should produce more PG precursors than the WT strain , even in starvation . We assessed PG activity using a fluorescently labeled D-alanine analogue TAMRA-D-alanine ( TADA ) , which is thought to be incorporated primarily into metabolically active PG ( Kuru et al . , 2012 ) . We observed increased fluorescence in the strain carrying murA S368P as compared to the WT allele during the first 12 hr of PBS starvation , but , by 24 hr , the staining was equivalent between the strains ( Figure 6C ) . Controls show that TADA staining is specific to the cell wall ( Figure 6—figure supplement 1C , D ) . We interpret this to mean that post-translational regulation of MurA is important to control PG metabolism in logarithmic phase growth and during the transition to starvation , but that after prolonged starvation other regulatory mechanisms predominate to control PG metabolism . The abrupt cessation of growth ( Figure 6—figure supplement 1E ) and decreased CwlM phosphorylation ( Figure 6A ) imply that incorporation of new PG decreases upon starvation in the WT strain; however , TADA staining remains constant ( Figure 6C ) . This suggests that TADA staining in these conditions may be due more to exchange of D-alanine in the periplasm by Ldts ( Cava et al . , 2011 ) than to incorporation of new , labeled PG transported from the cytoplasm . Regardless , the over-activation of MurA clearly alters PG metabolism during the transition to starvation . Decreased synthesis of PG precursors and cell wall metabolism is an important adaptation to stress: subverting this regulation should be disadvantageous . Many models suggest that decreased metabolism contributes to Mtb’s antibiotic tolerance during treatment . This is based on the observation that many antibiotics are less effective during starvation and stationary phase in Mtb ( Betts et al . , 2002; Herbert et al . , 1996; Wallis et al . , 1999; Xie et al . , 2005 ) and other bacteria , likely because the activity of drug targets is reduced , or because permeability is decreased ( Kester and Fortune , 2014; Sarathy et al . , 2013 ) . We hypothesized that continued synthesis of PG during the transition to starvation could reduce antibiotic tolerance . To test this , we subjected cells to PBS starvation and measured antibiotic sensitivity in strains with murA WT and S368P alleles . We found a 3–100 fold increase in killing of the strain carrying the murA S368P allele when the strains were treated with isoniazid , meropenem or rifampin at the initiation of starvation ( Figure 6D ) . The mutant also had a significant antibiotic tolerance defect in nutrient rich media , but not when the antibiotics were added late in starvation ( Figure 6—figure supplement 1F , G ) , after the period in which post-translational regulation of MurA affects PG metabolism ( Figure 6C ) . We conclude that the PknB-CwlM-MurA signaling cascade functions to quickly turn off PG synthesis during nutrient restriction ( Figure 7 ) , and that this regulation contributes to antibiotic tolerance during growth and under changing conditions . 10 . 7554/eLife . 14590 . 015Figure 7 . Phylogenetics and model for the PknB-CwlM-MurA regulatory pathway . ( A ) Domain structure of CwlM from mycobacteria . ( B ) Phylogenetic analysis of cwlM in the actinobacteria . Tree is based on a 16S rRNA alignment for each species . Species indicated in gray do not contain cwlM homologues , and are representatives of larger clades . All homologues lack a predicted secretion signal . The conservation of an intact or degenerate amidase active site and the presence of a PTG or TGT phosphorylation motif are indicated by color . ( C ) Model for the regulation of MurA . In nutrient rich conditions PknB phosphorylates CwlM , which can then promote the activity of MurA , resulting in increased flux through the PG biosynthetic pathway ( indicated by the series of arrows ) and promoting cell growth . In starvation , PknB is poorly expressed , CwlM is not phosphorylated , and MurA is therefore not activated , resulting in reduced production of PG precursors , which helps to halt cell growth . DOI: http://dx . doi . org/10 . 7554/eLife . 14590 . 015 CwlM’s function could not have been predicted from available knowledge about the function of its conserved domains . Based on domain structure , it seems likely that an ancestral cwlM homologue was a periplasmic enzyme involved in PG hydrolysis . How could such a protein evolve into a cytoplasmic , phosphorylation-responsive regulator of PG biosynthesis ? To address this , we searched for cwlM homologues that retained both PG binding domains and the PG amidase domain in the same arrangement as is seen in the mycobacterial proteins ( Figure 7A ) , and found that almost all such homologues were found in members of the phylum Actinobacteria . We chose representatives from each family and searched for the presence of a secretion signal , the conservation of the four Zn2+-coordinating residues , and the presence of a PTG or TGT phosphorylation motif in the C-terminal tail . We constructed a phylogenetic tree based on 16S rRNA sequence of the species that have cwlM homologues and some representative species from families that do not , and mapped onto it the conservation of the three cwlM features ( Figure 7B ) . We found that all actinobacterial cwlM homologues lack secretion signals . Potential phosphorylation motifs are conserved in most clades , irrespective of whether or not they conserve the canonical Zn2+-coordinating residues . These data imply that both CwlM cytoplasmic localization and , possibly , regulation by phosphorylation were present in an ancestor of the Actinobacteria . CwlM is essential in M . smegmatis , but its amidase activity does not appear to be essential ( Figure 1 ) . However , a strain with a phospho-ablative mutation , cwlM T374A , exhibits defects in elongation ( Figure 2 ) and reduced viability in a manner reminiscent of CwlM depletion . The cwlM T374A mutant strain has such low viability that it could not be constructed except in a genetic background that ‘forced’ the L5 allele swap by repressing transcription of cwlM in transformants that maintained a wild type copy of the cwlM allele ( See Figure 2 and 4B ; Figure 4—figure supplement 1;Supplementary file 1 ) . We show that the murA S368P mutation restores normal growth of the cwlM T374A mutant ( Figure 4B; Figure 4—figure supplement 1 ) which implies that phosphorylation of CwlM activates MurA . The short cells seen in the CwlM depletion and the cwlM T374A mutant look similar to Msmeg cells with a temperature sensitive MurA mutation , which are also short and have compromised cell walls ( Xu et al . , 2014 ) . The loss of viability in the cwlM mutants and temperature sensitive murA strain ( Xu et al . , 2014 ) could be due to uncoordinated synthesis of the various layers of the cell wall: peptidoglycan biosynthesis is downregulated but other cell wall biosynthetic and metabolic processes may remain active , leading to a cell wall with poor structural integrity . This differs from wild-type regulation during starvation , when PG regulation is likely accompanied by regulation of other cell wall layers . In fact , PknB , the kinase likely responsible for physiologic CwlM phosphorylation , has been implicated in the regulation of many cell wall processes ( Khan et al . , 2010; Molle et al . , 2006; Parikh et al . , 2009; Vilchèze et al . , 2014 ) , and probably helps to coordinate regulation of the various cell wall layers . The enzymatic activity of other PknB substrates changes less upon phosphorylation ( Khan et al . , 2010; Molle et al . , 2006; Parikh et al . , 2009 ) than the ~30-fold activation of MurA by CwlM~P . It is possible that PknB may regulate other factors like it does MurA – indirectly through intermediary substrates . Both PknB and PknE can phosphorylate CwlM in vitro . Because PknB acts more rapidly ( Figure 3A ) it is likely to be the primary kinase of CwlM . In nutrient replete conditions PknB apparently activates MurA by phosphorylating CwlM . How is the phosphorylation cascade regulated in starvation ? PknB can be transcriptionally ( Kang , 2005 ) and post-translationally regulated ( Baer et al . , 2014 ) . However , the rapid decrease in CwlM phosphorylation ( Figure 6B ) implies that there is a CwlM phosphatase . CwlM is a cytoplasmic PG hydrolase homologue . While most PG hydrolases are in the periplasm , there are cytoplasmic PG hydrolases that are involved in enzymatic processing of recycled anhydro-muropeptides ( Jacobs et al . , 1994; 1995; Park and Uehara , 2008 ) . The level of cytoplasmic anhydro-muropeptides can be a signal reporting on environmental conditions ( Jacobs et al . , 1994; 1995; Park and Uehara , 2008;Núñez et al . , 2000 ) . If CwlM is involved in processing or detection of anhydro-muropeptides , it could integrate information about their levels into its regulation of PG precursor synthesis . Because MurA is the first committed step in PG biosynthesis , its regulation is an efficient point at which to regulate the pathway . This is certainly true in other species . MurA from E . coli is subject to feedback regulation ( Mizyed et al . , 2005 ) , and many Gram positive species have two MurA homologues which are differentially regulated ( Blake et al . , 2009; Kock et al . , 2003 ) . Mycobacteria have only one MurA homologue , but the regulation by CwlM allows it to quickly and precisely adjust its enzymatic activity for different conditions without being proteolyzed and requiring re-synthesis ( Figure 6—figure supplement 1B ) . Our data imply that the post-translational regulation of MurA is important for cell wall regulation at early time points during stress ( Figure 6C ) . It is likely that transcriptional ( Dahl et al . , 2003 ) and proteolytic ( Festa et al . , 2010 ) systems downregulate cell wall enzymes after prolonged starvation or stress . The Kcat values we measured for MurA alone by varying PEP concentration are comparable to the values measured previously ( Xu et al . , 2014 ) , and the Kcat of MurA with CwlM~P was ~30 times higher . We and others have measured high Km values for UDP-GlcNAc ( Figure 5D; Supplementary file 1D; Figure 5—figure supplement 1 and [Xu et al . , 2014] ) , which surely exceed the concentrations in the cell; thus , there are probably other factors that regulate MurA and either reduce the Km or increase the local concentration of UDP-GlcNAc . One of the major limitations in treating tuberculosis is the extended course of therapy required in part because of phenotypic drug tolerance , which is thought to be a result of decreased metabolic activity of drug targets , and decreased permeability . Subverting signals that downregulate metabolism in these bacteria could re-sensitize them to antibiotics . We find that this is true during the transition to starvation in Msmeg grown in culture . A strain carrying the murA S368P allele is unable to properly downregulate PG metabolism , and is more sensitive to several antibiotics early in starvation ( Figure 6D ) . Because this increased killing was seen in the presence of isoniazid and meropenem , which target the cell wall , as well as rifampin , which does not , we think that the increased cell wall metabolism results in higher permeability , and that the increased killing is due to greater drug uptake ( Sarathy et al . , 2013 ) . Our data imply that MurA inhibitors ( De Smet et al . , 1999 ) might contribute to death of actively growing mycobacteria early in treatment , but that they could also contribute to drug tolerance of non-growing bacteria , which are likely to dominate later in treatment . While our experiments were conducted in vitro with Mtb proteins , the conservation of these proteins among mycobacteria suggest that the PknB-CwlM-MurA regulatory pathway is conserved , although it is likely that different stresses activate MurA regulation in Mtb compared to Msmeg . Treatments that interfere with the regulation of the mycobacterial cell wall during infection could shorten treatment times and improve patient outcomes . Biological complexity is predicated on the diversity of protein function . How do novel protein functions evolve ? One hypothesis is that compartmental re-targeting of duplicated genes is critical to the evolution of complexity in eukaryotic cells ( Bright et al . , 2010; Gabaldón and Pittis , 2015 ) . It is less clear that this same mechanism would drive the evolution of complexity in bacteria . Here , though , we find a bacterial example of evolutionary repurposing through new compartmentalization . Because CwlM consists of three functional domains , all of which are predicted to function in the periplasm , we assume that an ancestral homologue was periplasmic . However , all of the available cwlM homologues lack a predicted secretion signal . These homologues are found throughout the Actinobacteria , implying that a cwlM ancestral homologue was re-targeted to the cytoplasm before the actinobacterial phylum split off from its relatives . The conservation of the putative phosphorylation motif implies that cwlM homologues in most Actinobacteria have a role in signal transduction . Because periplasmic proteins are not known to be phosphorylated , this signaling role likely evolved after CwlM was re-targeted to the cytoplasm . CwlM provides cells with an additional layer of regulation of PG synthesis . MurA from Gram negative ( Dai et al . , 2002; Krekel et al . , 2000 ) and positive species ( Blake et al . , 2009; Du et al . , 2000 ) are regulated transcriptionally and by proteolysis ( Blake et al . , 2009; Kock et al . , 2003 ) . In mycobacteria , enzymes in PG biosynthesis are regulated transcriptionally ( Dahl et al . , 2003 ) and proteolytically ( Festa et al . , 2010 ) , in addition to the regulation of MurA through phosphorylation of CwlM described here . Thus , compartmental re-targeting of the ancestral CwlM protein has allowed for the evolution of a complex regulatory system by which mycobacteria , and possibly related Actinobacteria , control synthesis of PG precursors in response to environmental conditions . Mycobacterium smegmatis mc2155 was grown in 7H9 salts ( Becton-Dickinson , Franklin Lakes , NJ ) supplemented with: 5 g/L albumin , 2 g/L dextrose , 0 . 85 g/L NaCl , 0 . 003 g/L catalase , 0 . 2% glycerol and 0 . 05% Tween80 , or plated on LB agar . Hartmans-de Bont ( HdB ) media was made as described ( Hartmans and De Bont , 1992 ) with 0 . 05% tween80 , HdB-C was made without glycerol . E . coli DH5α was used for cloning and E . coli BL21 codon plus or ArcticExpress DE3 RP were used for protein expression . Antibiotic concentrations for M . smegmatis were: 25 µg/ml kanamycin , 50 µg/ml hygromycin , 20 µg/ml zeocin and 20 µg/ml nourseothricin . Antibiotic concentrations of E . coli were: 50 µg/ml kanamycin , 100 µg/ml hygromycin , 25 µg/ml zeocin , 40 µg/ml nourseothricin , 20 µg/ml chloramphenicol and 140 µg/ml ampicillin . All strains were grown at 37°C . Knockouts of cwlM and murA were made by first complementing the genes at the phage integrase sites L5 ( Lewis and Hatfull , 2003 ) and Tweety ( Pham et al . , 2007 ) , and then using recombineering to delete the endogenous copy . ~500 base pair regions upstream and downstream of the gene were amplified by PCR and PCR stitched to either side of a zeocin resistance cassette . The assembled PCR fragment was transformed into Msmeg strains expressing the RecET proteins . The resulting colonies were screened for deletion of the gene ( van Kessel and Hatfull , 2008 ) . To make CB737 , the zeocin resistance cassette in the deletions was removed by the Cre recombinase between each subsequent deletion . After complementing murA at the Tweety site and deleting the endogenous murA , we found that there was still another copy of murA in its native genomic context . This led to our discovery of an IS1096-mediated ( Wang et al . , 2008 ) genomic duplication which comprises MSMEG_4928–4944 ( murA is MSMEG_4932 ) and is present in most of our Msmeg strains . We deleted the second copy of murA by removing the zeocin cassette in the first deletion via Cre recombinase , and using the same knockout construct to delete the second copy . The cwlM::FLAG ( CB100 ) strain was also made by recombineering . ~500 base pair regions upstream and downstream of the stop codon of cwlM were amplified and PCR stitched to a zeocin cassette , which was transformed as described above . Different alleles of cwlM were attained by swapping the nourseothricin resistance-marked vector with WT cwlM for a kanamycin resistance-marked mutant allele , as described ( Pashley and Parish , 2003 ) . We also exchanged the murA allele at the Tweety integrase site in the same way as is done at the L5 site . Full strain details in Supplementary file 1A–C . For polar growth quantification , cells were stained with AlexaFluor488 carboxylic acid succinimidyl ester as described ( Aldridge et al . , 2012 ) , resuspended in 7H9 media , shook at 37°C for 3 hr , immobilized on agarose pads and imaged in phase and GFP channels with a Nikon TE-200E microscope with a 60X objective and Orca-II CCD camera ( Hamamatsu , Japan ) . The Ptet::cwlM strain was grown uninduced for 9 hr before staining . Images were analyzed in ImageJ ( NIH ) . The length of the longer unstained pole and the total cell length was measured manually for each cell . For the TADA experiment , cells were grown in HdB . 1 µl of 10 mM TADA ( synthesized according to [Kuru et al . , 2014] ) was added to 1 mL of culture and incubated for 5 min before washing with PBS+tween80 and fixing for 10 min with 1% paraformaldehyde . The rest of the culture was pelleted , resuspended and cultured in PBS+tween80 , and aliquots were stained the same way at 1 , 4 , 8 , 12 and 24 hr after resuspension . Controls were performed on cells growing in HdB: cells were either fixed first and then stained , or fixed and not stained at all . The samples were filtered through a 10 µm filter and analyzed by flow cytometry ( MACSQuant VYB Excitation: 561 nm; Emission filter: 615/20 ) . The settings for FSC , SSC , and cell density were adjusted to sample single cells . Three biological replicates were used for each strain and time point , and the median fluorescence for 50 , 000 cells from each replicate was calculated and averaged . 75 mL replicate cultures of murA WT and murA S368P growing in HdB were stained for 15 min at 37°C with 10 µM TADA . Cells were pelleted and washed twice with PBS+tween80 , resuspended in water and boiled for 1 hr , and spun at 100 K rpm for 20 min . The supernatants were collected , and the pellets resuspended in PBS+2 . 5% SDS , boiled for 1 hr , nutated O/N at 37°C , and spun at 21 K rpm for 15 min . The supernatants were collected and the pellets were washed twice with PBS . The fluorescence of each sample was measured with a Spectra284 plate reader ( Excitation: 541 , Emission: 568 ) . The water and SDS extracted supernatants were measured undiluted . The cell wall pellet suspension was diluted 120 fold , which was required for measurements to be within the linear range . The fluorescence values of the cell wall fractions measured by the plate reader were therefore multiplied by 120 . N-terminally his-MBP-tagged kinase domains of the nine canonical serine-threonine protein kinases from Mtb were expressed and purified as described ( Kieser et al . , 2015 ) . His-CwlM was expressed with 1 mM IPTG at 14° for 40 hr in ArcticExpress DE3 RP ( Agilent , Lexington , MA ) . Cells were resuspended and French-pressed in Ni Wash Buffer ( 50 mM NaHPO4 pH 8 . 0 , 300 mM NaCl , 20 mM imidazole ) , and supernatants were poured over TALON affinity resin ( Clontech , Mountain View , CA ) . Bound proteins were washed and eluted with Ni Wash Buffer + 200 mM imidazole . Soluble proteins were separated from aggregates on a Superdex S200 gel filtration column ( GE Healthcare , Westborough , MA ) in 20 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM DTT . Soluble proteins were concentrated and stored in 50 mM NaPO4 pH 7 . 5 , 150 mM NaCl , 20% glycerol , 2 mM DTT , 1 mM PMSF . Kinase reactions , α-phospho-threonine western blots and mass spectrometry were performed as described ( Kieser et al . , 2015 ) . Growth curves of M . smegmatis strains were done in triplicate , the OD600 of each culture was measured every 30 min in a Bioscreen growth curve machine ( Growth Curves USA , Piscataway , NJ ) . To determine the conserved features of the C-terminus of CwlM , the entire CwlM protein sequence was BLASTed against the genus Mycobacteria , and the C-terminal tail from the 34 most similar mycobacterial , and 98 most similar Actinomycetes CwlM proteins was analyzed using Weblogo ( Crooks et al . , 2004 ) . For 2D gels and phosphothreonine quantification of CwlM , CwlM-FLAG was immunoprecipitated by bead beating cells in 50 mM Tris pH7 . 5 , 300 mM NaCl with Protease Inhibitor Cocktail ( Roche , Switzerland ) . α-FLAG M2 Magnetic Beads ( Sigma Aldrich , Natick , MA ) were added to supernatants and washed with 50 mM NaHPO4 , 300 mM NaCl . CwlM was eluted from the beads with 0 . 5 mg/ml FLAG peptide in TBS . Samples were separated in 2D using the ReadyPrep 2D Starter Kit ( BioRad , Hercules , CA ) , according to the manufacturer’s protocols . SDS-PAGE was done with 4–12% NuPAGE Bis Tris precast gels ( Life Technologies , Beverley , MA ) . Mouse α-FLAG ( Sigma Aldrich ) was used at 1:10 , 000 in TTBS , Rabbit α-strep ( Genscript , China ) and Rabbit α-phosphothreonine ( Cell Signaling Technology , Danvers , MA ) were used at 1:1000 in TTBS + 0 . 5% BSA . Cultures of the L5::cwlM1cys-strep , L5::lprG-strepC and L5::pgm-strepC strains were grown to late log phase , washed twice and resuspended in PBS + 0 . 25% tween80 , split and treated with 0 or 0 . 6 mg/ml of MTSEA and MTSET ( Biotium , Hayward , CA ) for 20 min at RT . Blocking was quenched with cysteine at 50 mM; cells washed in PBS + 0 . 05% tween80 , and resuspended in PEGylation buffer ( 600 mM Tris pH 7 . 4 , 10 M Urea , 2% SDS , 1 mM EDTA ) , heated at 85° for 20 min and spun . 5 . 6 mg of MalPEG5K ( Sigma Aldrich ) was added to all the supernatants except ( - ) controls and incubated at 37° for 2 hr . Proteins were TCA precipitated and visualized by α-strep western blot . To confirm the suppression of the cwlM T374A allele by the murA S368P allele , the L5::cwlM ( nuoR ) tw::murA ( CB737 ) and L5::cwlM ( nuoR ) tw::murA S368P ( CB762 ) strains were transformed with pCB255 , pCB277 , pCB557 and pCB558 and plated on kanamycin . For each of 3–4 transformations , kanamycin and nourseothricin resistance was assessed for 96-192 colonies . Cultures of CB779 , CB737 and CB300 were grown to mid-log phase , washed , resuspended in PBS + 0 . 25% paraformaldehyde , incubated at 37° for 2 hr , quenched with 40 mM glycine , washed and lysed by French press . α-strep was conjugated to Protein G dynabeads ( Novex , Cambridge , MA ) which were added to the CB300 and the CB779 lysate . α-FLAG beads were added to the rest of the CB779 lysate and the CB737 lysate . Pull-downs were done according to the manufacturers protocols , and proteins visualized by western blot . his-MurA was purified like his-CwlM . Reactions with 20 µg/ml of his-MurA , equimolar his-CwlM in the kinase reaction , and varying concentrations of substrate were run as in ( Brown et al . , 1994 ) . MurA kinetic assays were done with MurATB alone in MurA reaction buffer ( 50 mM Tris pH 8 . 0 , 2 mM Kcl , 2 mM DTT ) or with equimolar CwlMTB in a kinase reaction with His-MBP-PknBKD . The kinase reactions contained: 202 µg/ml CwlMTB , 50 µg/ml His-MBP-PknBKD , 2 mM ATP , 2 mM MnCl2 and were brought up to 100 µl with Buffer C ( 20 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM DTT ) . Inactive kinase reactions had extra buffer instead of ATP . Kinase reactions were run for 1 hr at room temperature before being added to the MurA reactions . MurA reactions contained: 20 µg/ml MurA and 20 . 2 µg/ml CwlM in the kinase reaction ( 1/10 of MurA reaction was CwlM kinase reaction , or buffer for MurA alone kinetics ) . In MurA reactions with varying PEP , UDP-GlcNac was at 10 mM and PEP concentrations were: 25 , 100 , 250 , 500 and 1000 µM; these reactions were started with the addition of UDP-GlcNAc . In MurA reaction with varying UDP-GlcNac , PEP was at 2 mM and UDP GlcNac concentrations were: 0 . 5 , 1 , 3 , 5 , 8 , 10 and 15 mM; these reactions were started with the addition of PEP . 2 , 5 and 8 min time points were taken for the MurA + CwlM~P reaction; 5 , 15 and 25 min time points were taken for the MurA and MurA +CwlM reactions . At the indicated time point , a 25 µl aliquot of the reaction was removed and quenched in an equal volume of 400 mM KOH . The quenched reactions were spun in Microcon 10 K devices ( EMD Millipore , Billerica , MA ) and chilled until being injected into the Agilent HPLC . HPLC separation was performed on a MonoQ 5/50 GL anion exchange column ( GE Healthcare ) with the following protocol: 0 . 6 ml/min; 2 min of 20 mM tetraethylammonium bicarbonate pH 8 . 0 , an 18 min gradient from 20–500 mM tetraethylammonium bicarbonate pH 8 . 0 , 5 min of 500 mM tetraethylammonium bicarbonate pH 8 . 0 , 5 min of 20 mM tetraethylammonium bicarbonate pH 8 . 0 . The area under the A254 curve peak corresponding to EP-UDP-GlcNAc was integrated using Agilent ChemStation software . Because EP-UDP-GlcNac is not commercially available , we were not able to perform a standard curve of product concentrations to determine the relationship between peak area and product concentration . Instead , we measured the peak area of a range of UDP-GlcNac concentrations , and found that a peak area of 11 . 73 in the A254 curve was consistently equivalent to 1 uM of UDP-GlcNac . Because UDP-GlcNac and EP-UDP-GlcNac differ only in one enol-pyruvyl group , we reasoned that the relationship between A254 peak area and concentration should be comparable between these two molecules . We therefore divided the values for peak area for each EP-UDP-GlcNac peak by 11 . 73 to calculate the approximate µM concentration of EP-UDP-GlcNac . These values were plotted vs . time for each substrate concentration and the rate of product produced vs . time was calculated based on the linear portions of the curves . The resulting rates were plotted against substrate concentration in the curves shown in Figure 5 and Figure 5—figure supplement 1 . The peak that corresponds to EP-UDP-GlcNAc was identified because it appeared only when MurA and UDP-GlcNAc were incubated with PEP . The data were fitted to the Michaelis Menten formula using GraphPad Prism . Biological replicates of CB779 ( murA ) and CB782 ( murA S368P ) were grown in 7H9 , transferred to PBS+tween80 , diluted to OD = 0 . 05 , treated with 100 µg/ml isoniazid , 50 µg/ml rifampin or 20 µg/ml meropenem , and rolled at 37° during the CFU measurement period . For the unstarved kill curves ( Figure 6—figure supplement 1C ) , log . phase cultures in 7H9 were diluted to OD = 0 . 05 in 7H9 and treated with 50 µg/ml isoniazid , 20 µg/ml rifampin or 5 µg/ml meropenem . For Figure 6—figure supplement 1D , the log . phase 7H9 cultures were transferred to PBS+tween80 , diluted to OD = 0 . 05 and rolled at 37° for 18 hr before 100 µg/ml isoniazid , 50 µg/ml rifampin or 20 µg/ml meropenem was added . A CDART search was used to identify cwlM homologues with the same domain structure . All but 3 of these were in the Actinobacteria , the remaining 3 are in the Firmicutes . The ~700 CDART hits were separated by family , and a phylogenetic tree was made of each family . Between 3 and 16 homologues , representing the breadth of the phylogenetic distribution , were chosen from each family , leaving 97 homologues . The secretion signal identification programs Phobius ( Kall et al . , 2007 ) and PRED-TAT ( Bagos et al . , 2010 ) were used to assess the N-terminus of each homologue for the presence of a Sec or Tat secretion signal . The COBALT tool on NCBI was used to align the amidase domain of each homologue to active amidase domains of the same family; the conservation of the four Zn+2-coordinating residues was assessed for each . Each homologue was then searched for the presence of either PTG or TGT motifs in the C-terminal region after the amidase domain . 16S rRNA sequences were collected from the species with the 97 CwlM homologues , and from representative species from across the clade of Actinobacteria ( Lang et al . , 2013 ) that do not have CwlM homologues . The 16S sequences were aligned with Clustal Omega and a distance-matrix tree was constructed and visualized in ITOL ( Letunic and Bork , 2011 ) .	Bacterial cells are surrounded by a protective cell wall . Some bacteria , including the species Mycobacterium tuberculosis that causes tuberculosis , have the ability to change the properties of their cell wall when exposed to stressful conditions . These changes may also help the bacteria to resist the antibiotics used to treat the infections they cause . However , little is known about how changes to the cell wall are regulated . By carrying out genetic experiments in a non-infectious relative of M . tuberculosis and by performing biochemical assays with M . tuberculosis proteins , Boutte et al . have now investigated the role of a bacterial protein called CwIM . This protein was predicted to be an enzyme that cuts peptidoglycan , a network of sugars and short proteins that forms part of the bacterial cell wall . Such an enzyme would allow the peptidoglycan to expand and remodel . However , Boutte et al . found that CwlM does not act as an enzyme . Instead , it regulates an enzyme called MurA , which is present inside bacteria and helps to make the peptidoglycan network . Thus , CwIM actually helps to determine how much peptidoglycan a cell produces . When activated , the CwlM protein binds to MurA and stimulates it to start producing peptidoglycans . However , Boutte et al . observed that CwlM is active only when cells have plenty of nutrients . When nutrients are scarce , CwlM deactivates and reduces the activity of the MurA enzyme . This quickly shuts down the production of peptidoglycan and makes starved cells tolerant to antibiotics . Notably , increasing peptidoglycan production in starved cells , by enhancing the activity of the MurA enzyme , makes the cells more vulnerable to several antibiotics . Future work could now investigate the conditions under which CwlM is activated and deactivated in M . tuberculosis during an infection . It also remains to be seen whether other enzymes are regulated in a similar way to MurA .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2016	A cytoplasmic peptidoglycan amidase homologue controls mycobacterial cell wall synthesis
Notch is a critical regulator of T cell differentiation and is activated through proteolytic cleavage in response to ligand engagement . Using murine myelin-reactive CD4 T cells , we demonstrate that proximal T cell signaling modulates Notch activation by a spatiotemporally constrained mechanism . The protein kinase PKCθ is a critical mediator of signaling by the T cell antigen receptor and the principal costimulatory receptor CD28 . PKCθ selectively inactivates the negative regulator of F-actin generation , Coronin 1A , at the center of the T cell interface with the antigen presenting cell ( APC ) . This allows for effective generation of the large actin-based lamellum required for recruitment of the Notch-processing membrane metalloproteinase ADAM10 . Such enhancement of Notch activation is critical for efficient T cell proliferation and Th17 differentiation . We reveal a novel mechanism that , through modulation of the cytoskeleton , controls Notch activation at the T cell:APC interface thereby linking T cell receptor and Notch signaling pathways . T cell activation is mediated by antigen recognition through the T cell receptor ( TCR ) . To allow physiological adaptation , the TCR signal is modulated by co-regulatory signals . Here , we address co-stimulation through Notch . Notch family proteins are large , heterodimeric transmembrane receptors . Upon Notch ligation by one of a family of Notch ligands ( Osborne and Minter , 2007 ) , a plasma membrane-tethered matrix metalloproteinase , ADAM10 or ADAM17 , removes the Notch extracellular domain . Subsequently , the plasma membrane-embedded γ-secretase complex liberates the Notch intracellular domain ( NICD ) . The NICD constitutively translocates to the nucleus where it displaces transcriptional repressors and recruits enhancers to genomic loci characterized by binding of the transcription factor RPBJκ ( Borggrefe and Oswald , 2009 ) . An essential role for Notch1 in T cell thymic development is well established ( Robey et al . , 1996; Washburn et al . , 1997 ) ; a Notch1 deficient hematopoietic compartment yields no T cells ( Radtke et al . , 1999 ) . Notch1 signaling also plays a pivotal role in mature T cells . Notch1 is activated following TCR stimulation ( Amsen et al . , 2004; Guy et al . , 2013; Ong et al . , 2008 ) and is required for effector cell development ( Amsen et al . , 2004; Keerthivasan et al . , 2011 ) . The degree of Notch1 activation is directly proportional to the strength of the TCR signal ( Guy et al . , 2013 ) . Antigen-induced Notch1 activation in T cells may be ligand independent ( Adler et al . , 2003 ) ( Ayaz and Osborne , 2014; Palaga et al . , 2003 ) . However , the cellular mechanism coupling proximal T cell signaling to Notch activation is unresolved . Here we reveal a spatially constrained mechanism of Notch1 activation . We demonstrate that PKCθ , a serine/threonine kinase integrating TCR and CD28 signals ( Altman and Kong , 2016 ) , enhances T cell actin dynamics through localization and phosphorylation of the negative actin regulator Coronin1A ( Coro 1A ) and thus mediates actin-based recruitment of ADAM10 to the T cell:APC interface for efficient Notch activation . To study the role of TCR/CD28-proximal signaling in Notch1 activation , we bred PKCθ-deficient Tg4 mice ( Tg4KO ) . PKCθ integrates TCR and CD28 signals . PKCθ-deletion renders peripheral T cells hyporesponsive but allows normal thymic selection ( Sun et al . , 2000 ) . Tg4 CD4+ T cells ( Liu et al . , 1995 ) recognize the acetylated N-terminal peptide of myelin basic protein Ac1-9[4K] and its high affinity MHC-binding analogue Ac1-9[4Y] . To determine whether Notch activation could play a role in mature T cells that is comparable to that in thymocytes , where Notch drives critical developmental decisions ( Radtke et al . , 2013 ) , we assayed NICD expression in Tg4 thymocytes , naïve and primed T cells in response to anti-CD3 and anti-CD28 . NICD expression and changes thereof upon cellular activation were similar ( Figure 1—figure supplement 1A and B ) . However , Notch activation was impaired in mature T cells from Tg4KO mice . Tg4KO mice showed reduced Notch1 expression sixteen hours after in vivo T cell activation by injecting mice with 80 µg [4Y] peptide s . c . ( Figure 1A , B ) even though Tg4KO mice were grossly normal with a reduced number and proportion of CD4+ splenocytes but unaffected Tg4 TCR expression ( Figure 2A; Figure 2—figure supplement 1 ) . Reduced Notch expression was confirmed by Western blot analysis 60 min after s . c . administration of [4Y] peptide ( Figure 1C ) and through analysis of Notch1-dependent hes1 expression ( Figure 1D ) . Corroborating these data in non-TCR transgenic T cells , Notch1 cleavage in naïve CD4+ T cells ( CD4+ , CD44− , CD25− ) from PKCθ-deficient B10 . PL mice was diminished following overnight activation with anti-CD3 and anti-CD28 ( Figure 1E ) . Such Notch activation was CD28-dependent ( Figure 1F ) , consistent with an important role of PKCθ downstream of CD28 ( Huang et al . , 2002; Kong et al . , 2011; Yokosuka et al . , 2008 ) . As biochemical signaling activity in T cell activation peaks within the first few minutes , we verified that PKCθ-dependent Notch activation can also occur at this time scale . Increased nuclear Notch enrichment could be detected 5 to 20 min after in vitro T cell activation or after injecting mice with 80 µg [4Y] peptide s . c . ( Figure 1G–I ) . This effect was corroborated using a second TCR transgenic system , 5C . C7 ( Singleton et al . , 2009 ) ( Figure 1J ) . 10 . 7554/eLife . 20003 . 003Figure 1 . PKCθ enhances antigen-induced Notch activation . ( A ) Tg4WT and Tg4KO mice were injected subcutaneously with 80 µg of MBPAc1-9[4Y] peptide or PBS . After 18 hr splenocytes were immunostained to assess intracellular Notch1 expression and analyzed by flow cytometry . Gated on live , CD4+ cells . ( B ) The expression ( geometric mean fluorescence intensity , gMFI ) of intracellular Notch1 in CD4+ T cells from spleens of Tg4WT and Tg4KO mice treated as in A is shown as the mean ± SEM . p=0 . 002 , ns p=0 . 06 ( ANOVA ) . One experiment of 2 , n = 3 mice per condition . ( C ) Tg4WT and Tg4KO mice were injected subcutaneously with 80 µg of MBPAc1-9[4Y] peptide or PBS . CD4+ T cells were isolated from the spleen after 60 min by MACS and protein extracts analyzed by Western blotting with anti-NICD , anti c-myc and GAPDH . One representative Western blot of three . ( D ) Naïve Tg4WT and Tg4KO CD4+ T cells were isolated from splenocytes and stimulated with plate-bound anti-CD3 and anti-CD28 for 30 min . Expression of Hes1 was determined by RT-PCR . One representative experiment of four . ( E ) Naïve CD4+ T cells were isolated from B10 . PL PKCθ WT or KO splenocytes by magnetic selection and stimulated for 18 hr with plate-bound anti-CD3 and anti-CD28 as indicated . An equal amount of protein extract from each sample was analyzed for expression of the NICD and GAPDH by Western blotting . ( F ) Naïve CD4+ T cells isolated from Tg4WT and Tg4KO mice were stimulated for 18 hr with a titration of plate-bound anti-CD3 ±2 µg/ml anti-CD28 , as indicated . Expression of NICD and GAPDH was assessed by Western blotting . One representative Western blot of two . ( G , H ) Tg4WT and Tg4KO T cell blasts were incubated for 15 min with [4Y]-loaded PL8 cells before fixation and immunostaining against the IC domain of Notch1 . The cells were counterstained with phalloidin and DAPI before imaging by confocal microscopy . The proportion of NICD staining in the nucleus ( defined by DAPI staining ) and the cytoplasm ( defined by phalloidin staining ) was measured and the ratio of nuclear:cytoplasmic NICD calculated . p=0 . 0014 , p=0 . 02 , ns p=0 . 2 ( ANOVA ) . 32–58 cells analyzed per condition , combined data from two independent experiments . ( I ) Tg4WT mice were injected subcutaneously with 80 µg of MBPAc1-9[4Y] peptide or PBS . CD4+ T cells were isolated from the spleen after 5 or 20 min by MACS , fixed and immunostained against the IC domain of Notch1 . The ratio of nuclear:cytoplasmic NICD is given . T cell treatment with 2 mM EDTA serves as a positive control of Notch activation . The difference between the 0 min time point and the EDTA control is significant with p=0 . 02 ( ANOVA ) . One representative experiment of 4 . ( J ) 5C . C7 mice were injected subcutaneously with 80 µg of MCC ( 89–103 ) peptide or PBS . CD4+ T cells were isolated from the spleen after 5 or 20 min by MACS , fixed and immunostained against the IC domain of Notch1 . The ratio of nuclear:cytoplasmic NICD is given . Differences between the 0 versus 5 and 20 min time points are significant with p=0 . 005/0 . 002 , respectively ( ANOVA ) . One representative experiment of 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00310 . 7554/eLife . 20003 . 004Figure 1—figure supplement 1 . NICD expression is comparable across Tg4 thymocytes , naïve and primed T cells and substantially enhanced upon retroviral expression . A is a representative NICD Western blot of protein extracts from Tg4WT T cells stimulated or not with plate-bound anti-CD3 and anti-CD28 for 1 hr . ( B ) Data from four experiments are presented . ( C ) Tg4WT T cells were retrovirally transduced to express the NICD in parallel with GFP as a sorting marker and GFP positive and negative T cells were analyzed . The retrovirally expressed NICD is slightly smaller than the NICD generated by γ-secretase processing , consistent with the upper band in the NICDrv lanes . The molecular identity and functional capacity of the lower band is uncertain . Both bands are included in the quantification in D . ( D ) Given is the quantification of three independent experiments , as in C , representing the mean expression of the NICD relative to that in non-activated primed Tg4WT T cells ± SEM . Differences between NICD expression in Tg4WT and Tg4KO T cells were not significant ( 19 ± 2 . 5 fold versus 18 . 5 ± 1 fold ) and data were therefore pooled . indicates p<0 . 001 as determined by Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00410 . 7554/eLife . 20003 . 005Figure 2 . Constitutively active Notch rescues defective proliferation , Th17 polarization and IL-17A secretion in PKCθ deficient T cells . ( A ) Splenocytes from Tg4WT and Tg4KO mice were stained for the indicated molecules and absolute cell numbers or percentages are given as indicated . Each data point represents one mouse . Data are combined from mice assayed across three experiments . As previously reported ( Gupta et al . , 2008; Sun et al . , 2000 ) in non-TCR transgenic mouse strains , PKCθ deficiency in Tg4 mice resulted in a reduced number and proportion of CD4+ T cells . p values by Student’s unpaired two-tailed t-test . ( B ) The proliferation of naïve Tg4WT and Tg4KO CD4+ T cells stimulated with irradiated B10 . PL splenocytes and a titration of MBPAc1-9[4K] is given . n = 8 Tg4WT , n = 15 Tg4KO mice , each assayed in triplicate for each peptide concentration . Shown is the mean ± SEM . p<0 . 0001 by Student’s unpaired two-tailed t-test . ( C ) Naïve Tg4WT and Tg4KO mice were injected subcutaneously with 80 µg of MBPAc1-9[4Y] peptide or PBS . After 18 hr splenocytes were immunostained to assess CD69 expression and analyzed by flow cytometry . Gated on live , CD4+ cells . The mean expression ( geometric mean fluorescence intensity , gMFI ) ± SEM of CD69 is given . ns p=0 . 08 . One experiment of 2 , n = 3 mice per condition . ( D ) Tg4WT or Tg4KO mice were injected subcutaneously with 80 µg [4Y] peptide . After 60 min , CD4+ splenocytes were isolated by MACS , RNA was isolated and the expression of c-myc and IL-2 determined by RT-PCR . n = 3 mice per condition , shown is mean ± SEM . ns = 0 . 68 by unpaired Student’s t-test . ( E–G ) Splenocytes from Tg4WT or Tg4KO mice were stimulated with 10 µg/ml [4K] peptide , IL-12 and IL-2 for 7–9 days before restimulation with PMA and ionomycin in the presence of monensin . The proportion of IFNγ and IL-2-producing CD4+ T cells was determined by intracellular cytokine staining . Shown are representative FACS plots , gated on live , CD4+ cells and the combined data from all replicates shown as mean ± SEM n = 7–11 independent biological replicates ns = 0 . 71 ( F ) and 0 . 73 ( G ) by unpaired Student’s t-test . ( I–J ) Splenocytes from Tg4WT or Tg4KO mice were stimulated with 10 µg/ml [4K] peptide , IL-6 , IL-1β , IL-23 , anti-IFNγ and anti-IL-4 for 7–9 days before restimulation with PMA and ionomycin in the presence of monensin . The proportion of IFNγ and IL-17A-producing CD4+ T cells was determined by intracellular cytokine staining . Shown are representative FACS plots , gated on live , CD4+ cells and the combined data from all replicates shown as mean ± SEM . n = 8 independent biological replicates , p<0 . 0001 ( I ) p=0 . 03 ( J ) by Student’s t-test . ( K ) Splenocytes from Tg4WT and Tg4KO mice were stimulated with [4K] peptide and IL-2 before transduction with a retrovirus encoding NICD and GFP or GFP alone . After 72 hr the incorporation of 3H thymidine was measured . n = 3 replicate transductions per condition , mean values ± SEM . * = 0 . 0005 , ns = 0 . 31 . One representative experiment of four . ( L , M ) Splenocytes from Tg4WT and Tg4KO mice were stimulated with [4K] peptide in the presence of IL-6 , IL-1β , TGFβ and IL-23 before transduction with a retrovirus encoding NICD and GFP or GFP alone . After 96 hr of further culture with IL-6 , IL-1β , TGFβ and IL-23 the cells were restimulated with PMA and ionomycin in the presence of monensin before intracellular staining for the expression of IL-17A . ( L ) shows representative FACS data . The mean ± SEM of three replicates from one experiment of four is shown in M . ns = 0 . 06 by Student’s t-test . ( N ) The mean concentration of IL-17A was measured in supernatants from triplicate cultures of Tg4WT and Tg4KO cells T cells transduced with NICD or GFP alone under Th17-polarising conditions . p values by Student’s t-test . One representative experiment of three . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00510 . 7554/eLife . 20003 . 006Figure 2—figure supplement 1 . Tg4KO mice display largely unperturbed immune cell distributions . Splenocytes from Tg4WT and Tg4KO mice were stained for the indicated molecules and absolute cell numbers or percentages are given as indicated . Each data point represents one mouse . Data are combined from mice assayed across three experiments . As previously reported ( Gupta et al . , 2008; Sun et al . , 2000 ) in non-TCR transgenic mouse strains , PKCθ deficiency in Tg4 mice resulted in no defect in the proportion of peripheral CD8+ T cells or B cells but did result in a reduced number of FoxP3+ Treg . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 006 To determine functional outcomes of diminished Notch processing in PKCθ-deficient T cells , we analyzed T cell proliferation and differentiation . In accordance with published data ( Keerthivasan et al . , 2011; Marsland et al . , 2004; Tan et al . , 2006 ) , T cell proliferation , CD69 and c-Myc upregulation were defective in Tg4KO T cells whereas IL-2 was unaffected ( Figure 2B–D ) . Tg4KO T cells showed no defect in Th1 cell differentiation ( Figure 2E–G ) . Conversely , the proportion of IL-17A+ T cells was significantly reduced under conditions favoring Th17 development ( Figure 2H–J ) . Overexpression of NICD at 18 . 5 ± 1 . 5 fold the level in non-activated primed Tg4WT cells ( Figure 1—figure supplement 1C and D ) so as to constitutively activate Notch in Tg4KO cells led to restoration of T cell proliferation ( Figure 2K ) , Th17 cell differentiation and IL-17A secretion ( Figure 2L–N ) . Notch1 thus enhances T cell proliferation and differentiation downstream of PKCθ . We investigated ADAM10 as a key signaling molecule potentially employed by PKCθ for Notch1 activation . We visualized ADAM10 recruitment to the interface between the T cell and the APC by virally transducing Tg4WT and Tg4KO CD4+ T cells with ADAM10-GFP . ADAM10-GFP+ T cells were imaged as they interacted with H-2u PL8 lymphoma APC presenting the Ac1-9[4Y] antigen . Tg4KO T cells , whether transduced to express ADAM10-GFP or other sensors , formed tight cell couples upon APC contact albeit with a slightly reduced frequency compared to Tg4WT T cells ( 30 ± 5% versus 50 ± 4% , p=0 . 02 ) with comparable gross T cell morphology , as characterized in the next paragraph . Such effective cell coupling allowed an analysis of the interface recruitment of GFP-tagged signaling intermediates and the spatiotemporal patterns thereof . ADAM10-GFP was recruited rapidly and transiently to the interface of Tg4WT T cells and APC ( Figure 3A , B; Figure 3—figure supplement 1A , Video 1 ) consistent with previous work in AND T cells ( Guy et al . , 2013 ) . In contrast , ADAM10-GFP was not enriched at the interface of Tg4KO T cells ( Figure 3A , B; Figure 3—figure supplement 1A , Video 2 ) . In Tg4WT cells , ADAM10 was enriched in the interface lamellum , an actin-based signaling structure ( Roybal et al . , 2015b ) . Impairment of lamellum formation with 40 nM Jasplakinolide ( Figure 3—figure supplement 1B ) ( Roybal et al . , 2015a ) prevented ADAM10 interface recruitment ( Figure 3A , B; Figure 3—figure supplement 1A ) and Notch cleavage following stimulation with anti-CD3/28 ( Figure 3C ) . The defect in lamellal ADAM10 recruitment upon PKCθ-deficiency was selective since the lamellal accumulation of Themis , a protein with substantially more prominent and persistent lamellal accumulation than ADAM10 , was only moderately impaired ( Figure 3—figure supplement 1C–E ) . Together , these data suggest that actin-dependent ADAM10 recruitment to the T cell:APC interface at the early peak of T cell signaling activity mediates efficient Notch1 activation downstream of PKCθ . In AND T cells strong stimuli cause concerted accumulation of the TCR , Vav and Notch at the T cell/APC interface as related to efficient Notch activation ( Guy et al . , 2013 ) . Actin dynamics may thus mediate coordinated interface accumulation of both ADAM10 and Notch . It needs to be determined how PKCθ and Vav-dependent actin dynamics are related . 10 . 7554/eLife . 20003 . 007Figure 3 . PKCθ mediates transient actin-dependent recruitment of ADAM10 to the T cell lamellum . ( A ) Tg4WT CD4+ T cells , treated with 40 nM Jasplakinolide ( bottom ) or not ( top ) , and Tg4KO ( middle ) CD4+ T cells expressing ADAM10-GFP were activated with PL8 cells presenting the Ac1-9[4Y] peptide . Given are representative images showing pseudocolored ( purple to red ) maximum projections of the ADAM10-GFP fluorescence and a reference DIC bright field image at times relative to the formation of a tight couple between T cell and APC . The entire image sequences are given in Video 1 ( Tg4WT ) and 2 ( Tg4KO ) . ( B ) The graph shows the percentage of T cells with lamellal accumulation of ADAM10-GFP at the time relative to couple formation ± SEM . Differences in lamellal accumulation between Tg4WT and Tg4KO and Jasplakinolide-treated Tg4WT T cells at time points 0:00 and 0:20 were each significant with p≤0 . 005 ( Tg4KO versus Tg4WT 0:00 p=0 . 001 , 0:20 p=0 . 005; Tg4WT +Jasp versus Tg4WT 0:00 p=0 . 004 , 0:20 p=0 . 001 , by proportions z-test ) . 18–28 cell couples were analyzed per condition ( 57 total ) . Full pattern analysis is given in Figure 3—figure supplement 1A . ( C ) CD4+ blasts from Tg4WT and Tg4KO mice ( four days after stimulation ) were restimulated for 18 hr with anti-CD3 and anti-CD28 ±40 nM Jasplakinolide or left unstimulated as indicated . NICD and GAPDH expression in protein extracts was measured by Western blotting . One representative experiment of three . ( D ) Tg4WT and Tg4KO CD4+ T cells expressing F-tractin-GFP were activated with PL8 cells presenting the Ac1-9[4Y] peptide . Representative images are given as in A . The entire image sequences are given in Video 3 ( Tg4WT ) and 4 ( Tg4KO ) . ( E ) The percentage of cell couples with predominantly peripheral F-tractin-GFP accumulation is given as in B . The difference in peripheral accumulation between Tg4WT and Tg4KO T cells at joint time points 0:00 and 0:20 was significant ( p=0 . 01 by proportions z-test , 31 , 47 cell couples were analyzed per condition ) . Full pattern analysis is given in Figure 3—figure supplement 2A . ( F ) An example image of a T cell exhibiting the ‘bottleneck’ phenotype , defined as having a diameter minimum between the interface and the widest part of the cell body , is given as a grey scale F-tractin-GFP maximum projection together with a matching DIC bright field image . Measurement positions to determine the interface width ( yellow ) relative to the cell body ( red ) or the presence of a necking phenotype ( blue ) are shown . ( G ) The relative interface diameter was determined by relating the interface diameter to the widest part of the cell body and is given relative to the time of tight cell coupling . Shown is the mean ratio ± SEM . ns p=0 . 33 ( 0 s ) and 0 . 149 ( 20 s ) by unpaired , two-tailed Student’s t-test . 49 ( WT ) 35 ( KO ) cell couples were analyzed per condition . ( H ) The percentage of T cells displaying a bottleneck phenotype in at least one timepoint during the first 60 s after coupling is given . **p<0 . 001 by proportions z-test . 35 cell couples were analyzed per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00710 . 7554/eLife . 20003 . 008Figure 3—figure supplement 1 . PKCθ enables transient recruitment of ADAM10 to the T cell lamellum . ( A ) For the ADAM10-GFP experiments displayed in Figure 3A , B the proportion of T cells with accumulation in one of the six interface patterns at the time relative to couple formation is given ± SEM . ( B ) Given are fractions of T cells forming a tight cell couple upon contact with an activating APC in the presence of Jasplakinolide at the indicated concentrations for Tg4WT T cells activated with PL8 APCs presenting the Ac1-9[4Y] peptide and 5C . C7 T cells activated with CH27 APCs presenting the MCC peptide ( 5C . C7 data are taken from [Roybal et al . , 2015a] ) . Differences in cell coupling between Tg4WT and 5C . C7 T cells at each Jasplakinolide concentration were not significant . The cell coupling data establish that Tg4 and 5C . C7 T cells respond comparably to increasing concentrations of Jasplakinolide . The extensive characterization of the effect of 40 nM Jasplakinolide on T cell actin dynamics in ( Roybal et al . , 2015a ) thus applies to the Tg4WT T cells . ( C ) Tg4WT and Tg4KO CD4+ T cells expressing Themis-GFP were activated with PL8 cells presenting the Ac1-9[4Y] peptide . Representative images are given as in Figure 3A . ( D ) The percentage of cell couples with predominantly lamellal Themis-GFP accumulation is given as in Figure 3B . Differences in lamellal accumulation between Tg4WT and Tg4KO T cells are not significant ( p>0 . 05 ) . 43 , 24 cell couples were analyzed per condition . Full pattern analysis is given in D . ( E ) For the experiments displayed in B , C the proportions of T cells with accumulation in one of the six interface patterns at the time relative to couple formation are given ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00810 . 7554/eLife . 20003 . 009Figure 3—figure supplement 2 . PKCθ enhances interface actin dynamics . ( A ) For the F-tractin-GFP experiments displayed in Figure 3D , E the proportions of T cells with accumulation in one of the six interface patterns at the time relative to couple formation are given ± SEM . ( B ) Representative electron micrographs of Tg4WT and Tg4KO CD4+ T cell:PL8 APC interactions are given . The left image is representative of a large lamellum , the right one of a short and substantially ‘necked’ one . The black lines indicate the distance between the interface and the nucleus as displayed in C . ( C ) Quantification of electron micrographs is given . On the left , the interface diameters are shown at an early ( less than 2 min after cell coupling ) and late ( 3–5 min ) time point as a box and whisker plot . Across both time points interface diameters are significantly smaller ( p=0 . 05 ) in Tg4KO CD4+ T cells . 19 to 62 cell couples were analyzed per time point and condition ( total 155 ) . On the right , for the same cell couples the distances between the nucleus and the interface are given as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 00910 . 7554/eLife . 20003 . 010Video 1 . ADAM10-GFP accumulates rapidly and transiently at the interface between Tg4WT CD4+ T cells and PL8 APCs . A representative interaction of a Tg4WT CD4+ T cell expressing ADAM10-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown . Top: DIC images . Bottom: Top-down maximum projections of 3D fluorescence data are shown in a rainbow-like , false-color intensity scale ( increasing from blue to red ) . 20 s intervals in video acquisition are played back as two frames per second . Tight cell coupling occurs in frame 3 ( 1 s indicated video time ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 01010 . 7554/eLife . 20003 . 011Video 2 . ADAM10-GFP does not accumulate at the interface between Tg4KO CD4+ T cells and PL8 APCs . A representative interaction of a Tg4KO CD4+ T cell expressing ADAM10-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown as in Video 1 . Tight cell coupling occurs in frame 5 ( 2 s indicated video time ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 011 Previous work has linked PKCθ to actin regulation ( Sasahara et al . , 2002; Sims et al . , 2007; Villalba et al . , 2002 ) . By visualizing actin dynamics in Tg4WT and Tg4KO T cells with F-tractin-GFP ( Johnson and Schell , 2009 ) ( Videos 3 and 4 ) , multiple elements of the actin-dependent establishment of a tight T cell:APC interface were modestly impaired in cells lacking PKCθ . The spreading of F-actin to the periphery of the interface was delayed in Tg4KO T cells ( Figure 3D , E; Figure 3—figure supplement 2A ) . The interface diameter of Tg4KO T cells was significantly ( p<0 . 05 ) reduced across multiple time points ( Figure 3F , G ) , as confirmed by electron microscopy ( Figure 3—figure supplement 2B ) . The lamellum connecting the T cell body to the interface was smaller in Tg4KO T cells as it showed a significantly ( p<0 . 001 ) larger constriction or ‘neck’ ( Figure 3F , H ) . Long lamella ( >2 . 5 µm ) did not occur ( Figure 3—figure supplement 2C ) . Tg4KO T cells thus displayed modest defects across multiple elements of actin-driven cell spreading consistent with the slightly reduced efficiency of tight cell coupling . 10 . 7554/eLife . 20003 . 012Video 3 . F-tractin-GFP accumulates rapidly at the interface between Tg4WT CD4+ T cells and PL8 APCs . A representative interaction of a Tg4WT CD4+ T cell expressing F-tractin-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is show as in Video 1 . Tight cell coupling occurs in frame 6 ( 4 s indicated video time ) . Immediate spreading of the majority of F-actin to the edge of the interface is visible . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 01210 . 7554/eLife . 20003 . 013Video 4 . F-tractin-GFP accumulates rapidly at the interface between Tg4KO CD4+ T cells and PL8 APCs . A representative interaction of a Tg4KO CD4+ T cell expressing F-tractin-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown as in Video 1 . Tight cell coupling occurs in frame 4 ( 2 s indicated video time ) . Delayed spreading of the majority of F-actin to the edge of the interface is visible . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 013 We sought to identify actin regulators mediating the modest actin modulation by PKCθ . Coronin1A inhibits the Arp2/3 complex ( Humphries et al . , 2002; Oku et al . , 2005 ) and regulates clearance of actin from the NK immune synapse ( Mace and Orange , 2014 ) . Furthermore , Coronin is an established interactor with and substrate of PKC ( Cai et al . , 2005; Siegmund et al . , 2015 ) . Coronin1A was highly enriched at the interface of Tg4WT and Tg4KO T cells ( Figure 4A , B; Figure 4—figure supplement 1A; Videos 5 and 6 ) . Similar to actin , Coronin1A spreading to the interface periphery was delayed in Tg4KO cells , leaving substantially more Coronin 1A in the lamellum . Given that Coronin 1A is a negative regulator of actin dynamics its enhanced enrichment in the lamellum is consistent with the concurrent , localized impairments in actin , T cell morphology and ADAM10 recruitment . As a specificity control , the spatiotemporal distribution of the dominant F-actin severing protein Cofilin ( Roybal et al . , 2016; Singleton et al . , 2011 ) was unaffected by PKCθ deficiency ( Figure 4—figure supplement 1A ) . 10 . 7554/eLife . 20003 . 014Figure 4 . PKCθ phosphorylates and localizes Coronin1A . ( A ) Tg4WT and Tg4KO T cell expressing coronin1A-GFP were activated with PL8 cells presenting the Ac1-9[4Y] peptide . Representative images are given as in Figure 3A . The entire image sequences are given in Video 5 ( Tg4WT ) and 6 ( Tg4KO ) . ( B ) The percentage of cell couples with predominantly lamellal ( top ) and peripheral ( bottom ) Coronin1A-GFP accumulation is given as in Figure 3B . The differences in lamellal accumulation between Tg4WT and Tg4KO T cells at time points −40 to 80 were each significant with p≤0 . 05 by proportions z-test . 50 , 34 cell couples were analyzed per condition . Full pattern analysis is given in Figure 4—figure supplement 1A . ( C ) Shown is a representative Phos-tag western blot of protein extracts from Tg4WT or Tg4KO T cells stimulated with PMA and/or Calyculin A ( Caly ) for 5 min as probed with anti-coronin1A . ( D ) Given is the quantification of four independent experiment as in C as the mean ratio of the top ( phospho ) and lower ( non-phospho ) Coronin 1A bands ± SEM . indicates p<0 . 05 Tg4WT versus Tg4KO T cells by two-way ANOVA with Sidak’s correction for multiple comparisons . ( E ) A graphical summary of the proposed mechanism of the enhancement of Notch activation by PKCθ is given . The top and bottom rows illustrate Tg4WT or Tg4KO T cells , respectively . Each individual panel shows the interface part of the T cell that contacts the APC ( not shown on top ) . Separate panels are drawn from left to right for PKCθ ( as also included in the other panels ) , Coronin1A , F-actin and ADAM10 . Colors denote preferential accumulation patterns , central ( red ) , lamellal ( green ) and peripheral ( blue ) . Shade of color denotes the extent of accumulation . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 01410 . 7554/eLife . 20003 . 015Figure 4—figure supplement 1 . Interface recruitment of signaling intermediates peaks within the first three minutes in Tg4KO CD4+ T cells . ( A ) In the top row , for the Coronin1A-GFP experiments displayed in Figure 4A , B the proportions of T cells with accumulation in one of the six interface patterns at the time relative to couple formation are given with SEM . In the bottom row Tg4WT and Tg4KO T cell expressing Cofilin-GFP were activated with PL8 cells presenting the Ac1-9[4Y] peptide and pattern distributions are given . 45 , 51 cell couples were analyzed per condition . ( B ) Tg4WT T cell expressing the indicated sensors was activated with PL8 cells presenting the Ac1-9[4Y] peptide and pattern distributions are given . 31 ( TCRζ-GFP ) , 48 ( Lck-GFP ) , 49 ( LAT-GFP ) , 79 ( PKCθ-GFP , a representative image sequence is given as Video 7 ) , 53 ( PLCδ PH-GFP for PIP2 ) , and 61 ( Vav1-GFP ) cell couples were analyzed ( 321 total ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 01510 . 7554/eLife . 20003 . 016Video 5 . Coronin1A-GFP accumulates rapidly at the interface between Tg4WT CD4+ T cells and PL8 APCs . A representative interaction of a Tg4WT CD4+ T cell expressing Coronin1A-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown as in Video 1 . Tight cell coupling occurs in frame 4 ( 2 s indicated video time ) . Immediate spreading of the majority of Coronin1A-GFP to the edge of the interface is visible . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 01610 . 7554/eLife . 20003 . 017Video 6 . Coronin1A-GFP accumulates rapidly at the interface between Tg4KO CD4+ T cells and PL8 APCs . A representative interaction of a Tg4KO CD4+ T cell expressing Coronin1A-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown as in Video 1 . Tight cell coupling occurs in frame 3 ( 1 s indicated video time ) . Transient lamellal accumulation of the majority of Coronin1A-GFP is visible . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 017 Next , we investigated phosphorylation of Coronin 1A by PKCθ in Tg4 T cells . Coronin activity is negatively regulated by serine/threonine phosphorylation , which can be induced by phorbol ester treatment ( Cai et al . , 2005; Oku et al . , 2008 , 2012 ) . To allow detection of changes in Coronin1A phosphorylation , we prevented Coronin1A dephosphorylation by treating cells with the phosphatase inhibitor Calyculin A ( Oku et al . , 2008 , 2012 ) . Treating Tg4WT T cells with PMA and Calyculin A resulted in a shift in the ratio of phosphorylated to non-phosphorylated Coronin1A from 0 . 3 ± 0 . 1 to 2 . 8 ± 1 . 1 fold , indicative of efficient Coronin1A phosphorylation ( Figure 4C , D ) . This shift was significantly ( p<0 . 05 ) smaller in Tg4KO cells ( 0 . 25 ± 0 . 05 to 0 . 8 ± 0 . 25 fold ) ( Figure 4C , D ) demonstrating that PKCθ is required for efficient PMA-induced phosphorylation of Coronin1A . Together , our data suggest ( Figure 4E ) that in Tg4WT T cells PKCθ ( Figure 4—figure supplement 1B; Video 7 ) inactivates Coronin 1A selectively in the region of most intense stimulating signaling , i . e . the center of the T cell:APC interface with effects extending across the entire lamellum but not reaching the peripheral actin ring . Thus PKCθ locally inhibits Coronin 1A-mediated attenuation of actin dynamics , promoting the formation of a strong actin-based lamellum . With regard to Notch1 activation this allows for the efficient actin-driven recruitment of ADAM10 to the T cell:APC interface . This mechanism of enhanced Notch processing peaks within the first few minutes of T cell activation . Such early signaling emphasis is consistently observed in Tg4WT cells ( Figure 4—figure supplement 1B ) and other TCR transgenic systems , where it extends to the nuclear localization of other transcription factors , NFAT and NFκB ( Roybal et al . , 2015b; Singleton et al . , 2009 ) . We have thus identified a spatially restricted , actin-dependent mechanism of Notch activation downstream of PKCθ ( Figure 4E ) . Future work will determine how the enhancement of Notch activation by PKCθ is integrated with PKCθ-dependent NFκB activation ( Gruber et al . , 2009; Sun et al . , 2000 ) in the regulation of T cell differentiation . 10 . 7554/eLife . 20003 . 018Video 7 . PKCθ-GFP accumulates at the center of the interface between Tg4WT CD4+ T cells and PL8 APCs . A representative interaction of a Tg4WT CD4+ T cell expressing PKCθ-GFP with a PL8 APC presenting the Ac1-9[4Y] peptide is shown as in Video 1 . Tight cell coupling occurs in frame 4 ( 2 s indicated video time ) . Central accumulation of the majority of PKCθ-GFP is visible . DOI: http://dx . doi . org/10 . 7554/eLife . 20003 . 018 A key feature of our mechanism of PKCθ function is that it connects signaling at the time scale of minutes to outcomes in cellular differentiation over days . While causally connecting such divergent time scales is a great challenge , there is precedent . 15 min of contact between a primed T cell and a professional APC is sufficient to trigger T cell proliferation 24 hr later ( Iezzi et al . , 1998 ) . Similarly , 1 hr of ZAP-70 activity can trigger substantial negative selection ( Au-Yeung et al . , 2014 ) . Differential signaling kinetics may also regulate Treg induction ( Miskov-Zivanov et al . , 2013 ) . On an even shorter time scale 5 min of TGFβ incubation saturates Smad2 phosphorylation at 1 hr ( Vizán et al . , 2013 ) . In T cell activation , it has been argued that a time delay in the onset of activating versus inhibitory signaling from 2 to more than 5 min , respectively , may play an important role in the induction of anergy in response to high doses of antigen ( Wolchinsky et al . , 2014 ) . In B cells a single pulse of BCR engagement can trigger the nuclear accumulation of NFκB for 6 hr ( Damdinsuren et al . , 2010 ) . While mechanisms linking rapid proximal signaling to later cell function largely remain to be determined , the well-supported existence of such causal links is consistent with our model of PKCθ-dependent Notch activation . All mice were maintained under SPF conditions with ad libitum access to water and standard chow at the University of Bristol . All animal experiments were carried out under the UK Home Office Project Licence number 30/2705 held by David Wraith and the study was approved by the University of Bristol ethical review committee . B10 . PL , 5C . C7 ( Seder et al . , 1992 ) and Tg4 ( Liu et al . , 1995 ) mice were bred in-house at the University of Bristol . PKCθ-deficient Tg4 mice were generated by cross-breeding Tg4 mice with C57BL/6 prkcq−/−mice ( a gift of A . Poole , University of Bristol , originally generated by D . Littman ( Sun et al . , 2000 ) for >8 generations . The genetic status of each animal was assessed by PCR as previously described ( Sun et al . , 2000 ) . B10 . PL PKCθ KO mice were obtained by breeding Tg4KO mice with B10 . PL mice . The PL8 H-2u expressing , antigen-presenting B cell lymphoma was prepared in our laboratory ( Wraith et al . , 1992 ) . The H-2k expressing CH27 B cell lymphoma was prepared as described previously ( Haughton et al . , 1986 ) and obtained from Mark Davis , Stanford University . Both cell lines proved mycoplasma free by PCR and were validated by staining for MHC class II expression and assessing their ability to present antigen to relevant T cell lines . Lymphoid tissue was dissociated using standard protocols and red blood cells removed using Red Cell Lysis buffer ( Sigma ) . Unless otherwise stated , cells were cultured in complete RPMI 1640 ( Lonza; supplemented with 25 mM HEPES , 50 U/ml Pen/Strep , 2 mM L-Glutamine and 50 µM 2-mercaptoethanol ) with 5–10% FCS ( BioSera , Hyclone ) . PL8 and CH27 cells were maintained in complete RPMI with 10% FCS . Th17 cells were generated and maintained in IMDM ( Lonza; supplemented with 50 U/ml Pen/Strep , 2 mM L-Glutamine and 50 µM 2-mercaptoethanol ) containing 10% FCS . For isolation of double negative thymocytes , thymi were gently disaggregated on ice in 5% FCS/PBS . Cells were stained at 4°C with CD4-FITC and CD8a-APC antibodies . Propidium iodide was added immediately prior to flow cytometric sorting of viable CD4-CD8a- thymocytes using a BD Influx cell sorter . For in vitro stimulation experiments , naïve CD4+ T cells were isolated from spleen and axillary , brachial and inguinal lymph nodes using either EasySep Mouse Naïve CD4+ T cell isolation kit ( Stem Cell Technologies ) or MagniSort Mouse Naïve CD4+ T cell enrichment kit ( eBioscience ) according to the manufacturers’ instructions . For ex-vivo analysis of activated T cells , CD4+ cells were enriched with Mouse CD4+ T cell enrichment kit II ( Miltenyi Biotech ) according to the manufacturer’s instructions . Naïve or pre-activated CD4+ T cell cells were stimulated with plate-bound anti-CD3 ( 2C11 , eBioscience or BioExcel , 1 µg/ml or as indicated ) and anti-CD28 ( 37 . 51 , eBioscience or BioExcel , 2 µg/ml ) . Alternatively , cells were stimulated with PL8 cells and MBPAc1-9 [4K] or [4Y] peptide ( GL Biochem ) or CH27 cells and MCC 88–103 peptide at the concentration indicated . Where indicated , cells were incubated with PMA ( Sigma , 20 ng/ml ) , Calyculin A ( Sigma , 100 nM ) or Jasplakinolide ( Tocris , 40 nM ) . Suspensions of splenocytes from Tg4WT and Tg4KO mice were stimulated with 10 µg/ml MBP Ac1-9 [4K] peptide . For TH1 generation , cells were cultured in complete RPMI containing 10 ng/ml IL-12 ( Peprotech ) and 20 U/ml rhIL-2 ( R and D systems ) . For TH17 cells , culture was performed in complete IMDM containing 25 ng/ml IL-6 , 10 ng/ml IL-1β , 2 ng/ml TGFβ ( all Peprotech ) , 10 ng/ml IL-23 ( eBioscience ) , 50 µg/ml anti-IFNγ ( XMG1; BioExcel ) and 10 µg/ml anti-IL-4 ( 11B11; BioExcel ) . CD4+ T cells were stimulated as indicated and washed with ice-cold PBS before protein was extracted in RIPA buffer supplemented with protease and phosphatase inhibitor cocktails ( all from Pierce ) ( 1−2 × 107 cells/ml ) . Lysates were centrifuged for 10 min at 17 , 000xg and the soluble fraction denatured in Laemmli buffer before resolution by SDS-PAGE on 4–12% gels ( NuPAGE ) , transfer to PVDF membrane and immunodetection using standard ECL protocols . Where indicated , samples were resolved on 12 . 5% gels supplemented with 50 µM PhosTag reagent ( Wako ) and 100 µM ZnCl2 . For PhosTag experiments , cells were washed with HBSS instead of PBS . The following antibodies were used Notch1; ( D1E11 , Cell Signaling ) , GAPDH ( D16H11 , Cell Signaling ) , c-myc ( E910 , Santa Cruz Biotechnology ) , Coronin1A ( H300 , Santa Cruz Biotechnology ) , anti-rabbit and anti-mouse HRP conjugates ( Sigma ) . Non-viable cells were excluded from all analyses using Live/Dead eF780 dye ( 1:1000 , eBioscience ) . Surface staining was performed in PBS containing 0 . 5% FCS and 2 mM EDTA . Intracellular cytokine staining ( ICCS ) was performed on cells stimulated with PMA ( 10 ng/ml ) and ionomycin ( 1 µg/ml ) in the presence of GolgiStop ( BD Bioscience , 1:1000 ) for 4 hr . Cells were surface stained before fixation and permeabilization using eBioscience reagents . Staining for intracellular Notch1 and FoxP3 was performed after fixation and permeabilization using FoxP3 staining kit reagents ( eBioscience ) . The following antibodies , all purchased from eBioscience and/or Biolegend , were used; CD4 Alexa700 ( GK1 . 5 , 1:100 ) , CD69 FITC ( H1 . 2F3 , 1:100 ) , Notch1-PE ( mN1A , 1:100 , Biolegend ) , CD8a APC ( 53–6 . 7 , 1:200 ) , CD19 ( 1D3 , 1:200 ) , B220 FITC ( RA3-6B2 , 1:100 ) , Vb8 . 1/2 FITC ( KJ16-133 , 1:100 ) , FoxP3 PE ( FJK-16S , 1:100 , eBioscience ) , CD25 PE-Cy7 ( PC61 . 5 , 1:300 ) , IFNγ PE-Cy7 ( XMG1 , 1:400 ) , IL-2 eF450 ( JES6-5H4 , 1:100 ) and IL-17A PE or PE-Cy7 ( 17B7 , 1:2–400 ) . Soluble IL-17A was detected in culture supernatant by ELISA using Ready-Set-Go ELISA kit ( eBioscience ) . The cDNA encoding the IC domain of murine Notch1 was obtained from Addgene ( plasmid number 20183 ) . The IC domain was amplified by PCR with the primers ACCGCGGTGGCGGCCATGCAGCATGGCCAGCTCT and CGGGCTAGAGCGGCCTTATTTAAATGCCTCTGGAATGT and cloned into the Not1 site of pGC-IRES-GFP ( Costa et al . , 2000 ) using In-Fusion HD reagents ( Clontech ) . cDNA encoding murine Adam10 , obtained from Sinobiological ( Genbank number NM_007399 . 3 ) , was amplified with primers ACCGCGGTGGAGGCCAAGATGGTGTTGCCGACAGT and GGCGACCGGTGGATCTCCACCGCGTCGCATGTGTCCCATT and cloned into BamH1 and Not1 sites of pGC-GFP using In-Fusion reagents such that the C-terminus of ADAM10 was fused to GFP . The pGC-IRES-GFP vector was used as a negative control . The constructs used to express other sensors including Coronin1A-GFP , Cofilin-GFP , Themis-GFP , F-Tractin-GFP and PKCθ–GFP have been previously described ( Table 1 in [Roybal et al . , 2015b] and [Roybal et al . , 2016] ) . Retrovirus was generated by transfecting Phoenix-E cells using calcium phosphate precipitation . For imaging experiments , Tg4 T cells were infected by centrifugation with viral supernatant 24 hr after stimulation with 10 µg . ml−1 [4K] and 20 U . ml−1 rhIL-2 . Immediately following transduction culture media was replaced with complete RPMI supplemented with 20–40 U/ml rhIL-2 . For TH17 Notch rescue experiments , Tg4WT and Tg4KO splenocytes were cultured for 24 hr under TH17-polarising conditions ( as described above ) before transduction . Immediately following transduction , the culture medium was replaced with complete IMDM containing 20 ng/ml IL-23 and 50 µg/ml anti-IFNγ . T cells were analyzed by imaging or flow cytometry 4–5 days after transduction . 72 hr following transduction , 2 . 5 µCi 3H thymidine/ml ( Perkin Elmer ) was added to culture wells . After 16 hr incubation , incorporation of 3H thymidine was measured by scintillation counting . CD4+ T cells were stimulated and isolated as indicated in figure legends and RNA was extracted using either RNeasy Mini kit ( Qiagen ) or TRI Reagent ( Sigma Aldrich ) . cDNA was generated using Superscript III polymerase ( Life Technologies ) and real time PCR performed using a SYBR green PCR Mastermix ( Life Technologies ) . Primers; il2 sense; AGCAGCTGTTGATGGACCTA , il2 antisense; CGCAGAGGTCCAAGTTCAT , cmyc sense; TTGAAGGCTGGATTTCCTTTGGGC , cmyc antisense; TCGTCGCAGATGAAATAGGGCTGT , Hes1 sense; AAAGATAGCTCCCGGCATTC , Hes1 antisense; TGCTTCACAGTCATTTCCAGA , β2M sense; GCTATCCAGAAAACCCCTCAA , β2M antisense; CGGGTGGAACTGTGTTACGT . Data were analysed using the 2-△△CT method , normalized to β2microglobulin . Live cell imaging was performed as described in detail before ( Singleton et al . , 2009 ) . Using FACS , GFP+ transductants were sorted to a five-fold range of expression around 2 µM , the lowest concentration visible by microscopy and often within the range of endogenous protein amounts ( Roybal et al . , 2016 ) . PL8 cells were pre-loaded with 10 µg/ml [4Y] for >4 hr and combined with pre-sorted GFP+ Tg4 T cells in a glass-bottom plate on the stage of a spinning disk microscope system ( UltraVIEW 6FE system , Perkin Elmer; DMI6000 microscope , Leica; CSU22 spinning disk , Yokogawa ) . GFP data were collected as 21 z-sections at 1 µm intervals every 20 s . All imaging was performed at 37⁰C in PBS containing 10% FCS , 1 mM CaCl2 and 0 . 5 mM MgCl2 . Images were exported in TIFF format and analyzed with the Metamorph software ( Molecular Devices ) . Cell couples were identified using the differential interference contrast ( DIC ) bright field images . The subcellular localization of GFP-tagged protein sensors at each time point was classified into one of six previously defined stereotypical patterns ( Singleton et al . , 2009 ) that reflect cell biological structures driving signaling organization ( Roybal et al . , 2013 ) . Briefly , interface enrichment of fluorescent proteins at less than 35% of the cellular background was classified as no accumulation . For enrichment above 35% the six , mutually exclusive interface patterns were: accumulation in a large protein complex at the center of the T cell:APC interface ( central ) , accumulation in a large T cell invagination ( invagination ) , accumulation that covered the cell cortex across central and peripheral regions ( diffuse ) , accumulation in a broad actin-based interface lamellum ( lamellum ) , accumulation at the periphery of the interface ( peripheral ) or in smaller membrane protrusions ( asymmetric ) . Pre-activated Tg4WT and Tg4KO CD4+ T cells ( 4 days after activation ) were combined with PL8 APC pre-incubated with 10 µM MBPAc1-9[4Y] for 15 min before fixation with 4% PFA . Alternatively , Tg4 or 5C . C7 T cells were activated in vivo by s . c . injection with 80 µg MBPAc1-9[4Y] or MCC ( 88-103 ) respectively before cell isolation and fixation . Following permeabilization with 0 . 05% Triton X-100 cells were immunolabelled with anti-Notch1 IC domain ( D1E11 , Cell Signaling ) with an anti-rabbit Alexa488-conjugated secondary antibody ( Life Technologies ) and counterstained with DAPI and Phalloidin Alexa647 ( Life Technologies ) . Alternatively , cell couples were stained with anti-Notch Alexa647 ( Abcam , ab194122 ) and anti-CD4 FITC . Images were acquired on a Leica SP5 confocal microscope and image analysis was performed in Metamorph and Volocity ( Perkin Elmer ) . Electron microscopy experiments were executed as described in detail in Roybal et al . ( 2015b ) . Briefly , Tg4WT or Tg4KO CD4+ T cells and peptide-loaded PL8s were centrifuged together for 30 s at 350 g to synchronize cell coupling , the cell pellet was immediately resuspended to minimize unspecific cell coupling and cellular deformation and the cell suspension was further incubated at 37 degree C . After 2 and 5 min for early and late time points , respectively , the cell suspension was high pressure frozen and freeze substituted to Epon . Ultrathin sections were analyzed in an FEI Tecnai12 BioTwin equipped with a bottom-mount 4*4K EAGLE CCD camera . T cell:APC couples were identified in electron micrographs through their wide cellular interface . As described in detail in Roybal et al . ( 2015b ) , the time point assignment of cell couples was filtered with morphological criteria post acquisition using the presence of a uropod and T cell elongation . No statistical methods were used to predetermine the sample size . The significance of pairwise comparisons was measured by Student’s t-test . Where multiple comparisons were made , significance was determined by ANOVA with Tukey correction . The statistical significance in differences in percentage occurrence was calculated with a proportions z-test .	The body’s immune system recognizes and responds to foreign agents such as bacteria and viruses . Immune cells known as T cells recognize foreign substances through a protein on their surface called the T cell receptor . Specifically , the T cell receptor binds to fragments of foreign proteins displayed on the surface of other cells , which sets in motion a chain of events that leads to the T cell becoming activated . An activated T cell divides to form new cells that develop into “effector” T cells , which can mount an effective immune response . The T cell engages with the cell displaying the foreign proteins via an interface referred to as the immunological synapse . This zone of contact brings together the signaling machinery of the T cell . Like many other cells , T cells contain an internal skeleton-like structure made up of actin filaments . These filaments are crucial for the formation of the immunological synapse , in part because they help to transport the T cell receptor and other signaling proteins to the immunological synapse . Recent research suggests that a signaling protein called Notch plays an important role in instructing activated T cells to develop into effector cells . Notch is found on the surface of many cells , including T cells , and it becomes activated when it is cut by a specific enzyme . However , it was not entirely clear how T cell signaling drives the activation of the Notch protein . Britton et al . have now investigated the mechanism that leads to Notch activation in T cells from mice . The results show that a protein found inside the T cell , called PKCθ , is a major contributor to Notch activation when T cells become activated . So how does the PKCθ protein control the activation of Notch ? Britton et al . observed that PKCθ inactivates a protein that normally inhibits actin filaments from forming , and does so specifically at the center of the immunological synapse . This inhibition promotes the generation of a large actin-rich structure known as the lamellal actin network . This structure is required to recruit the Notch-cutting enzyme to the immunological synapse . Further analysis revealed that Notch gets cut and activated during the first few minutes of T cell activation leading to cell division and the development of effector T cells . Following on from this work , the next challenge will be to explore if altering signaling from the T cell receptor – for example , using drugs or small molecules – can modify the activation of Notch . If so , it will be important to explore if the chemicals could potentially be used to treat diseases that develop when T cells go awry , such as rheumatoid arthritis , psoriasis and Crohn’s disease .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology", "immunology", "and", "inflammation" ]	2017	PKCθ links proximal T cell and Notch signaling through localized regulation of the actin cytoskeleton
Store-operated calcium entry ( SOCE ) by calcium release activated calcium ( CRAC ) channels constitutes a primary route of calcium entry in most cells . Orai1 forms the pore subunit of CRAC channels and Stim1 is the endoplasmic reticulum ( ER ) resident Ca2+ sensor . Upon store-depletion , Stim1 translocates to domains of ER adjacent to the plasma membrane where it interacts with and clusters Orai1 hexamers to form the CRAC channel complex . Molecular steps enabling activation of SOCE via CRAC channel clusters remain incompletely defined . Here we identify an essential role of α-SNAP in mediating functional coupling of Stim1 and Orai1 molecules to activate SOCE . This role for α-SNAP is direct and independent of its known activity in NSF dependent SNARE complex disassembly . Importantly , Stim1-Orai1 clustering still occurs in the absence of α-SNAP but its inability to support SOCE reveals that a previously unsuspected molecular re-arrangement within CRAC channel clusters is necessary for SOCE . Store-operated calcium entry or SOCE is activated in response to the depletion of endoplasmic reticulum ( ER ) calcium stores , and constitutes a primary mechanism of calcium influx in excitable and non-excitable cells ( Parekh and Putney , 2005 ) . Store-operated calcium release activated calcium ( CRAC ) channels have been extensively studied for their role in the activation of nuclear factor of activated T cells ( NFAT ) regulated gene expression in lymphocytes and other cells ( Lewis , 2001; Crabtree and Olson , 2002; Feske et al . , 2005; Parekh and Putney , 2005; Vig and Kinet , 2009 ) . Previously , genome-wide RNAi screens in Drosophila cell lines led to the identification of ER-resident STIMs as the store sensors and plasma membrane ( PM ) resident Orai/CRACMs as the pore forming subunit of CRAC channels ( Liou et al . , 2005; Zhang et al . , 2005 , 2006; Feske et al . , 2006; Peinelt et al . , 2006; Prakriya et al . , 2006; Yeromin et al . , 2006; Vig et al . , 2006a , 2006b ) . Stim1 , and its homolog Stim2 , contain a calcium-sensing EF-hand like domain facing the ER lumen . Upon store-depletion , Stim1 oligomerizes and concentrates in regions of ER directly adjacent to the PM , frequently referred to as junctional ER ( Liou et al . , 2005; Zhang et al . , 2005; Stathopulos et al . , 2006; Wu et al . , 2006; Liou et al . , 2007; Luik et al . , 2008; Stathopulos et al . , 2008 ) . While some factors involved in store-dependent and store-independent Stim1 clustering have been described ( Honnappa et al . , 2009; Smyth et al . , 2009; Srikanth et al . , 2010 , 2012; Walsh et al . , 2010b; Krapivinsky et al . , 2011; Lewis , 2011; Singaravelu et al . , 2011 ) , little is known about the final steps in CRAC channel activation . For instance , Orai1 multimers are thought to diffuse freely until trapped by Stim1 clusters in the ER–PM junctions , with active CRAC channels consisting of Orai1 hexamers ( Penna et al . , 2008; Park et al . , 2009; Madl et al . , 2010; Walsh et al . , 2010a; Hou et al . , 2012 ) , however , it remains unexplored whether mere trapping of Orai1 by Stim1 in ER–PM junctions is sufficient or additional molecular steps enable optimal activation of SOCE . We hypothesized that additional cytosolic proteins might be necessary to impart functionality to Stim1–Orai1 clusters that constitute the CRAC channel complex and adopted a candidate-based approach based on our earlier genome-wide RNAi screen ( Vig et al . , 2006b ) . Here we identify a direct and unexpected role for α-SNAP in the activation of SOCE . α-SNAP is a cytosolic protein with tetra-tricopeptide repeat ( TPR ) -like helical domains that bridges N-ethylmaleimide sensitive fusion protein ( NSF ) to soluble NSF attachment protein receptor ( SNARE ) complexes to promote their disassembly and SNARE recycling ( Clary et al . , 1990; Chang et al . , 2012 ) . We find that α-SNAP directly and independently binds Stim1 and Orai1 to regulate the function and molecular composition of Stim1–Orai1 clusters that form the CRAC channel complex . This role for α-SNAP in SOCE is independent of its conventional role in facilitating NSF mediated SNARE complex disassembly . Thus , we have identified a novel role for α-SNAP and our data define a new α-SNAP dependent step in the physiological activation of SOCE via CRAC channels . To test our hypothesis that additional proteins are needed to facilitate SOCE under physiological conditions , we used ∼200 to 500 base pair long double stranded RNA ( dsRNA ) to deplete the expression of six genes that we previously identified as candidates in a genome-wide RNAi screen ( Vig et al . , 2006b ) in Drosophila Kc cells . We found that knockdown of soluble NSF attachment protein ( SNAP ) strongly reduces SOCE by day 3 ( Figure 1A and Supplementary file 1 ) . Given the almost complete inhibition of SOCE in SNAP deficient Drosophila cells , we hypothesized that SNAP might engage in novel , SOCE specific protein–protein interactions that may or may not involve NSF and SNAREs . α- and β-SNAP are the two mammalian proteins most closely related to Drosophila SNAP ( Figure 1—figure supplement 1 ) . α-SNAP is ubiquitously expressed while β-SNAP expression is largely restricted to the brain . We depleted α-SNAP in HEK-293 ( Figure 1B ) , Jurkat T cells ( Figure 1C ) and U2OS cells ( Figure 1—figure supplement 2 ) using lentivirus-based RNAi constructs ( Supplementary file 1 ) and found SOCE to be strongly inhibited by day three compared to cells treated with control RNAi . By day five , depletion of α-SNAP caused cell rounding in adherent cell lines; we therefore restricted our analysis to adherent cells at early time-points . We confirmed the efficiency of α-SNAP knockdown for each experiment on western blots of whole cell lysates ( WCLs ) ( Figure 1D ) and by immunostaining cells with anti-α-SNAP antibody ( Figure 1—figure supplement 3 ) . Importantly , when we reconstituted α-SNAP deficient HEK 293 cells with an RNAi resistant version of α-SNAP we found that SOCE was largely restored ( Figure 1E ) . Defective SOCE in α-SNAP deficient HEK 293 cells could also be restored by over-expressing β-SNAP ( Figure 1—figure supplement 4 ) . γ-SNAP is a third SNAP protein widely expressed in mammalian tissues but is less similar to Drosophila SNAP and α-SNAP than β-SNAP ( Figure 1—figure supplement 1 ) . γ-SNAP depletion failed to inhibit SOCE in HEK 293 cells ( Figure 1—figure supplement 5 ) and γ-SNAP over-expression did not compensate for α-SNAP depletion ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 00802 . 003Figure 1 . α-SNAP depletion inhibits SOCE and NFAT activation . ( A ) Average Fura-2 ratios of Drosophila Kc cells treated with dsRNA targeting SNAP ( red ) or Rho-1 ( black ) for 3 days , and stimulated with 1 μM TG to measure SOCE using flexstation . ( n > 3 ) . ( B and C ) Average Fura-2 ratios of α-SNAP ( red ) or scramble ( scr ) ( black ) RNAi treated HEK 293 cells ( B ) or Jurkat T cells ( C ) stimulated with 1 μM TG to measure SOCE using flexstation . ( n > 20 ) ( D ) A representative Western Blot for α-SNAP . WCLs of α-SNAP and scr RNAi treated HEK 293 cells were subjected to western blot analysis for each experiment using α-SNAP monoclonal antibody followed by anti-mouse secondary antibody . ( E ) Reconstitution of α-SNAP deficient cells with α-SNAP . Average Fura-2 ratios of RNAi treated HEK 293 cells stimulated with 1 μM TG to measure SOCE using flexstation , analyzed 3–4 days post RNAi transduction and 24 hr post-transfection with α-SNAP . ( Black ) cells transduced with scr RNAi and transfected with empty vector; ( Red ) cells transduced with α-SNAP RNAi and transfected with empty vector; ( Blue ) cells transduced with scr RNAi , transfected with α-SNAP; and ( Purple ) cells transduced with α-SNAP RNAi , transfected with α-SNAP . ( n = 3 ) ( F ) Western blot for nuclear NFATc1 in RNAi treated Jurkat T cells . Nuclear extracts were prepared from RNAi treated Jurkat T cells , unstimulated , or stimulated with 1 μM TG and 10 ng/ml PMA and subjected to western blot using anti-NFATc1 antibody . c-Myc was used as a loading control . ( n = 3 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 00310 . 7554/eLife . 00802 . 004Figure 1—figure supplement 1 . Alignment of Drosophila and human SNAP amino acid sequences . Purple ( basic ) , Blue ( acidic ) , Red ( hydrophobic ) , Green ( polar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 00410 . 7554/eLife . 00802 . 005Figure 1—figure supplement 2 . α-SNAP depletion using two different RNAi sequences in U2OS cells . Average single cell Fura-2 ratios of U2OS cells treated with two different α-SNAP RNAi or scr RNAi and stimulated with 1 μM TG ( n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 00510 . 7554/eLife . 00802 . 006Figure 1—figure supplement 3 . Immuno-staining for α-SNAP in α-SNAP depleted and control cells . Epifluorescence images of α-SNAP and scr RNAi treated HEK 293 cells stained using α-SNAP monoclonal antibody followed by anti-mouse IgG secondary antibody . Scale bar 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 00610 . 7554/eLife . 00802 . 007Figure 1—figure supplement 4 . Reconstitution of α-SNAP deficient cells with β- or γ-SNAP . Average Fura-2 ratios of RNAi treated HEK 293 cells stimulated with TG to measure SOCE ( using flexstation ) , analyzed 3–4 days post RNAi transduction and 24 hr post-transfection with β- or γ-SNAP . ( Black ) cells transduced with scr RNAi and transfected with empty vector; ( Red ) cells transduced with α-SNAP RNAi and , transfected with empty vector; ( Blue ) scr RNAi , transfected with β-SNAP ( Left Panel ) or γ-SNAP ( Right Panel ) ; and ( Purple ) α-SNAP RNAi , transfected with β-SNAP ( Left Panel ) or γ-SNAP ( Right Panel ) . ( n = 2 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 00710 . 7554/eLife . 00802 . 008Figure 1—figure supplement 5 . RNAi mediated depletion of γ-SNAP and measurement of SOCE . ( Left Panel ) Average Fura-2 ratios of HEK 293 cells transduced with scr RNAi or five different RNAi targeting γ-SNAP , and stimulated with TG to measure SOCE ( using flexstation ) . ( Right Panel ) Semi-quantitative PCR on total RNA to assess the level of γ-SNAP mRNA depletion compared to GAPDH in RNAi treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 008 SOCE is essential for the activation of nuclear factor of activated T cells ( NFAT ) , which in turn regulates gene expression in lymphocytes and most other cells ( Lewis , 2001; Crabtree and Olson , 2002 ) . Western blot of nuclear extracts of α-SNAP deficient Jurkat T cells showed a strong reduction in nuclear translocation of endogenous NFATc1 in response to the store-depleting agent thapsigargin ( TG ) in combination with phorbol 12-myristate 13-acetate ( PMA ) ( Figure 1F ) . These data demonstrate that α-SNAP is a novel regulator of SOCE and downstream signaling required for NFAT activation . To understand how α-SNAP is involved in SOCE , we first asked whether α-SNAP binds Stim1 and/or Orai1 . Co-immunoprecipitations ( co-IPs ) from lysates of store-depleted HEK 293 cells co-expressing either Stim1-Myc and YFP-α-SNAP ( Figure 2A ) or Flag-Orai1 and YFP-α-SNAP ( Figure 2B ) showed that α-SNAP co-precipitates with both Stim1 and Orai1 and vice versa . To determine if the interaction is direct , we incubated 100 nM purified α-SNAP with Orai1 and Stim1 immunoprecipitates ( IPs ) in vitro . α-SNAP bound to both Stim1 and Orai1 but not to an unrelated calcium channel , TRPC6 ( Figure 2C ) . 10 . 7554/eLife . 00802 . 009Figure 2 . α-SNAP directly binds Stim1 and Orai1 and co-localizes with CRAC channel clusters . ( A and B ) Co-immunoprecipitation of α-SNAP with Stim1 and Orai1 . WCLs of store-depleted HEK 293 cells expressing Stim1-Myc and YFP-α-SNAP ( A ) or Flag-Orai1 and YFP-α-SNAP ( B ) were subjected to immunoprecipitation using anti-Myc , anti-Flag , or anti-GFP antibodies and Western Blot as indicated ( n = 10 ) . ( C ) In vitro binding of recombinant α-SNAP to full length Stim1 and Orai1 . Flag-Orai1 , Flag-TRPC6 , and Stim1-Myc immunoprecipitates were incubated with purified α-SNAP protein ( 100 nM ) . Post-incubation , beads were washed , boiled , and subjected to western blot analysis using anti-α-SNAP antibody . ( n > 10 ) . ( D , E and F ) Representative confocal images of resting HEK 293 cells co-expressing YFP-α-SNAP and CFP-Stim1 ( D ) or store-depleted cells showing significant co-localization ( E ) . Store-depleted HEK 293 cells co-expressing YFP-α-SNAP and Orai1-CFP ( F ) . Scale bar 10 μm . ( n = 3 ) ( G ) TIRF images of live , store-depleted HEK 293 cells , co-expressing CFP-Stim1 , Orai1-CFP , and YFP-α-SNAP . Scale bar 10 μm . ( n = 3 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 009 We next sought to determine whether α-SNAP , a predominantly cytosolic protein , co-localizes with ER localized Stim1 and/or PM localized Orai1 under resting or store-depleted conditions . Some α-SNAP co-localized with Stim1 in almost all of the CFP-Stim1 expressing cells imaged even under resting conditions ( Figure 2D ) . Upon store-depletion , we observed distinct co-localization of α-SNAP with Stim1 clusters in nearly 30–40% of cells ( Figure 2E ) . Although there was no obvious co-localization between α-SNAP and Orai1-CFP expressed alone under resting ( not shown ) or store-depleted conditions ( Figure 2F ) , α-SNAP co-localized at a higher frequency with clusters containing Stim1 as well as Orai1 in Stim1 , Orai1 co-expressing stable cells , with nearly 75% cells showing distinct co-localization ( Figure 2G ) . Taken together , these data show that a part of the total cellular pool of α-SNAP constitutively binds Stim1 . Upon store-depletion , Stim1 likely brings α-SNAP to CRAC channel clusters where α-SNAP engages with Orai1 . α-SNAP is well known for its ability to promote SNARE complex disassembly and SNARE protein recycling by bridging NSF to SNARE complexes ( Clary et al . , 1990; Hanson et al . , 1995; Jahn et al . , 1995; Xu et al . , 1999; Marz et al . , 2003; Barszczewski et al . , 2008; Wickner and Schekman , 2008; Winter et al . , 2009; Chang et al . , 2012 ) . Therefore , α-SNAP depletion could affect Orai1 trafficking to the PM or Stim1 localization in the ER , thereby contributing to defective SOCE . Epifluorescence imaging of Orai1-YFP and YFP-Stim1 expressing α-SNAP depleted cells showed normal localization of both Orai1 and Stim1 ( Figure 3A and 3B ) . To specifically quantify Orai1 protein levels at the PM , we tagged the extracellular loop of Orai1 with a α-bungarotoxin ( BTX ) binding site ( BBS ) ( Sekine-Aizawa and Huganir , 2004 ) ( Figure 3—figure supplement 1 ) . Stable expression of Orai1-BBS-YFP followed by exogenous application of labeled BTX provided an estimate of cell surface Orai1 levels . α-SNAP depleted , resting , or store depleted cells did not show any significant difference in the levels of cell surface Orai1 when compared to controls ( Figure 3—figure supplement 2 ) . Consistently , immunostaining of α-SNAP depleted cells with an ER marker , calreticulin ( Figure 3—figure supplement 3 ) , and a Golgi-marker , giantin ( Figure 3—figure supplement 3 ) showed no apparent changes in the ER-Golgi morphology suggesting that at the time points used in our studies , α-SNAP depletion had not perturbed overall organelle structure . 10 . 7554/eLife . 00802 . 010Figure 3 . Regulation of SOCE by α-SNAP is NSF independent . ( A and B ) Resting localization of Orai1 and Stim1 in control and α-SNAP depleted cells . Epifluorescence images of α-SNAP and scr RNAi treated , resting HEK 293 cells stably expressing Orai1-YFP ( A ) or YFP-Stim1 ( B ) . Scale bar 10 μm . ( n > 5 ) ( C ) Quantification of intracellular transferrin-alexa 555 fluorescence . U2OS T-REx cells stably transfected with inducible WT-NSF ( black ) or E329Q-NSF ( mut ) ( red ) were induced with doxycycline for 8 hr . Post induction , cells were incubated with transferrin-alexa 555 , washed , and fixed . Alexa 555 fluorescence was quantified in cells using ImageJ . ( n = 2 with ∼80–100 cells scored per group ) *p<0 . 001 ( D ) SOCE in live and adherent U2OS cells expressing WT-NSF ( black ) vs E329Q-NSF ( mut ) ( red ) . Fura-2 ratios averaged from cells induced with doxycycline for 16–20 hr and stimulated with TG . ( n = 3 ) ( E ) Western blot on WCLs of WT- or E329Q-NSF expressing adherent U2OS cells analyzed in panel D . ( F ) Localization of WT-NSF with respect to Stim1 clusters in store-depleted cells . TIRF images of Myc-tagged WT-NSF expressing U2OS T-REx cells transiently transfected with YFP-Stim1 and stimulated with TG prior to imaging . Scale bar 10 μm . ( G ) Quantification of intracellular transferrin-alexa 555 fluorescence in RNAi treated HEK 293 cells . Cells transduced with scr RNAi and reconstituted with YFP alone ( EV ) ( black bars ) or transduced with α-SNAP RNAi and reconstituted either with YFP-alone ( EV ) ( red ) , with YFP-tagged WT α-SNAP ( blue ) or with YFP-tagged α-SNAP-L294A mutant ( gray ) . Post RNAi transduction and reconstitution , cells were incubated with transferrin-alexa 555 , washed , and fixed . Alexa 555 fluorescence was quantified in YFP positive cells using ImageJ . ( n = 2 , with each experiment scoring ∼100 cells ) p<0 . 001 ( H ) Average single cell Fura-2 ratios of scr RNAi treated ( black ) or α-SNAP RNAi treated HEK 293 cells reconstituted with empty vector ( red ) or α-SNAP-L294A mutant ( gray ) showing SOCE in response to stimulation with TG . Cells were analyzed 3 days post RNAi transduction and 16 hr post-reconstitution . ( n = 2 ) ( I and J ) Co-immunoprecipitation of Stim1 ( I ) and Orai1 ( J ) with WT α-SNAP or α-SNAP-L294A . WCLs of store-depleted HEK 293 cells co-expressing Stim1-Myc with YFP-α-SNAP or Stim1-Myc with YFP-α-SNAP- L294A ( I ) and Flag-Orai1 with YFP-α-SNAP or Flag-Orai1 with YFP-α-SNAP-L294A ( J ) were subjected to immunoprecipitation using anti-GFP antibodies and Western blot as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01010 . 7554/eLife . 00802 . 011Figure 3—figure supplement 1 . Strategy for quantifying Orai1 levels in the plasma membrane . Schematic showing the site of insertion of α-bungarotoxin ( BTX ) binding site ( BBS ) in the second extracellular loop of Orai1 and the C-terminal YFP tag . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01110 . 7554/eLife . 00802 . 012Figure 3—figure supplement 2 . Quantification of cell surface Orai1 in α-SNAP depleted and control cells . U2OS cells stably expressing Orai1-BBS-YFP were transduced with scr ( black ) or α-SNAP RNAi ( red ) , stimulated with 0 ( hashed line ) , or 1 μm TG ( solid line ) , and incubated with α-bungarotoxin alexa 647 ( BTX ) . Shown here is BTX binding on YFP-positive cells measured using FACS caliber . BTX binding to Orai1-YFP expressing cells was used as a control ( gray line ) ( n > 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01210 . 7554/eLife . 00802 . 013Figure 3—figure supplement 3 . Morphology of ER and Golgi in control and α-SNAP depleted cells . Epifluorescence images of α-SNAP or scr RNAi treated HEK 293 cells , stained with anti-calreticulin ( Left Panel ) , or anti-giantin ( Right Panel ) primary antibodies followed by respective secondary antibodies . Scale bar 10 μm . ( n = 3 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 013 Since the only well defined function of α-SNAP , thus far , is to bridge NSF to SNARE complexes ( Clary et al . , 1990; Chang et al . , 2012 ) , we next asked whether NSF is involved in SOCE . We used inducible expression of a dominant negative NSF mutant , E329Q-NSF , which lacks the ATPase activity required for dissociating the SNARE complex ( Dalal et al . , 2004 ) and blocks transferrin flux through the endocytic pathway by 8 hr ( Figure 3C ) . Surprisingly , SOCE was unaffected in E329Q-NSF expressing live and adherent cells even 20 hr after inducing the expression of the mutant ( Figure 3D and 3E ) and unlike α-SNAP , WT-NSF failed to co-localize with Stim1 clusters in ER–PM junctions ( Figure 3F ) . Most importantly , reconstituting α-SNAP deficient cells with a previously reported mutant of α-SNAP , L294A ( Barnard et al . , 1997 ) ( Figure 4A ) , that is unable to activate NSF and promote SNARE complex disassembly , failed to reconstitute transferrin flux through the endocytic pathway ( Figure 3G ) but fully restored SOCE ( Figure 3H ) . Accordingly , like WT-α-SNAP , the L294A mutant co-immunoprecipitates Stim1 ( Figure 3I ) as well as Orai1 ( Figure 3J ) . This clearly separates the function of α-SNAP in SOCE from its function in membrane trafficking demonstrating that α-SNAP regulates SOCE by a novel mechanism that is independent of its role in bridging NSF to SNARE complexes and more sensitive to its depletion . 10 . 7554/eLife . 00802 . 014Figure 4 . α-SNAP requires its hydrophobic loop for regulating SOCE . ( A ) Schematic showing domains of interest in human α-SNAP protein sequence and strategy for the generation of α-SNAP mutants . ( B ) Representative confocal images of store-depleted CFP-Stim1 expressing HEK 293 cells; transiently transfected with YFP-α-SNAP-NT ( 1–160 aa ) or YFP-α-SNAP-CT ( 161–295 aa ) Scale bar 10 μm ( n = 3 ) . ( C ) Average single cell Fura-2 ratios of scr ( black ) or α-SNAP RNAi treated HEK 293 cells , reconstituted either with empty vector ( red ) or with α-SNAP-NT ( gray ) , showing SOCE in response to TG . Cells were analyzed three days post RNAi transduction and 16 hr post-reconstitution ( n = 2 ) . ( D ) Quantification of intracellular transferrin-alexa 555 fluorescence . HEK 293 cells were transduced with scr RNAi and reconstituted with YFP alone ( black bars ) or with α-SNAP RNAi and either reconstituted with YFP-alone ( red ) or with YFP-tagged α-SNAP-LOOP mutant ( gray ) . Post RNAi transduction and reconstitution cells were incubated with transferrin-alexa 555 , washed , and fixed . Intracellular alexa 555 fluorescence was quantified in YFP positive cells using ImageJ . ( n = 2 , with each experiment scoring ∼100 cells ) p<0 . 01 , p<0 . 001 ( E ) Average single cell Fura-2 ratios of scr ( black ) or α-SNAP RNAi treated HEK 293 cells reconstituted either with empty vector ( red ) or with α-SNAP LOOP mutant ( gray ) , showing SOCE in response to TG . Cells were analyzed three days post RNAi transduction and 16 hr post-reconstitution . ( n = 2 ) ( F and G ) Co-immunoprecipitation of Stim1 ( F ) and Orai1 ( G ) with WT-α-SNAP or α-SNAP-LOOP mutant . WCLs of store-depleted HEK 293 cells expressing Stim1-Myc and YFP-α-SNAP or YFP-α-SNAP-LOOP mutant ( F ) and HEK 293 cells expressing Flag-Orai1 and α-SNAP or α-SNAP-LOOP mutant ( G ) were subjected to immunoprecipitation using anti-GFP or anti-α-SNAP antibody and western blot as shown . ( n = 2 ) ( H ) Representative confocal images of store-depleted CFP-Stim1 expressing HEK 293 cells; transiently transfected with YFP-α-SNAP-LOOP mutant . Scale bar 10 μm . ( n = 3 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 014 To explore the mechanism of regulation of SOCE by α-SNAP , we next set out to identify the domains involved in binding to the CRAC channel cluster and regulating SOCE by generating deletion and point mutations in α-SNAP as shown in Figure 4A . The N-terminal half of α-SNAP ( 1–160 aa ) or α-SNAP-NT is predicted to contain a hydrophobic loop along with three putative TPR domains and the C-terminal half ( 161–295 aa ) or α-SNAP-CT contains the fourth putative TPR domain ( Rice and Brunger , 1999 ) . Previous domain analysis of α-SNAP in the context of SNARE complex disassembly has shown that α-SNAP-CT but not α-SNAP-NT is necessary and sufficient for binding and activating the NSF ATPase activity ( Barnard et al . , 1997; Rodriguez et al . , 2011 ) . In contrast , upon store-depletion , α-SNAP-NT but not α-SNAP-CT co-localized with Stim1 clusters in CFP-Stim1 expressing cells ( Figure 4B ) . Accordingly , α-SNAP-NT was able to partially restore SOCE in α-SNAP depleted cells ( Figure 4C ) suggesting that a specific domain within the N-terminal half ( 1–160 aa ) of α-SNAP is crucial for binding to Stim1 and recruitment to the CRAC channel cluster . To further define the involved region within the N-terminal fragment , we introduced four point mutations ( F27S , F28S , L31S , F32S ) inside a putative hydrophobic loop within the N-terminal half of α-SNAP ( Figure 4A ) . Remarkably , reconstitution of α-SNAP deficient cells with this α-SNAP hydrophobic loop mutant ( α-SNAP-LOOP ) was able to restore transferrin uptake and recycling ( Figure 4D ) but failed to rescue SOCE ( Figure 4E ) . Accordingly , α-SNAP-LOOP showed significantly reduced binding to Stim1 ( Figure 4F ) and Orai1 ( Figure 4G ) and failed to co-localize with Stim1 clusters in store-depleted HEK 293 cells stably expressing CFP-Stim1 ( Figure 4H ) . Taken together these data demonstrate that the domains of α-SNAP required for SOCE are different from those essential for NSF activation and SNARE complex disassembly , conclusively demonstrating a distinct role for α-SNAP in SOCE . We next sought to identify the domains of Stim1 and Orai1 that α-SNAP binds within the CRAC channel cluster . Given the predominant cytosolic localization of α-SNAP , we expressed four different fragments of Stim1 cytosolic tail tagged with GST ( Lewis , 2011; Soboloff et al . , 2012 ) ; the full length cytosolic tail ( CT ) , the inhibitory coiled coil domain CC1 , the CRAC activation domain ( known as CAD or SOAR ) , and the CAD plus CC1 domain in Escherichia coli and performed an in vitro binding assay with purified α-SNAP . α-SNAP showed strong binding to the CAD domain of Stim1 and faint binding to the CAD plus CC1 domain ( Figure 5A ) . The CAD domain is a ∼100 amino acid long , multifunctional cytosolic domain of Stim1 , which when expressed with Orai1 , engages with specific regions within the C- as well as the N-terminal cytosolic tails of Orai1 to activate SOCE ( Muik et al . , 2009; Park et al . , 2009; Yuan et al . , 2009 ) . To identify the domains of Orai1 that bind α-SNAP , we tagged the cytosolic N- and C-terminal tails of Orai1 with GST , expressed them in E . coli and found that purified α-SNAP predominantly binds to the C-terminal tail of Orai1 ( Orai1-CT ) although faint binding to the N-terminal tail of Orai1 ( Orai1-NT ) was also detected ( Figure 5B ) . 10 . 7554/eLife . 00802 . 015Figure 5 . α-SNAP directly binds the CRAC activation domain of Stim1 and the C-terminal tail of Orai1 . ( A ) In vitro binding of α-SNAP to cytosolic domains of Stim1 . ( Top panel ) GST-tagged Stim1 fragments , expressed in E . Coli and immobilized on resin , were incubated with purified α-SNAP protein ( 10 nM ) . Post-incubation , beads were washed , boiled , and subjected to western blot analysis using anti-α-SNAP antibody . ( Bottom panel ) Ponceau S staining showing the input of GST-tagged fragments and their expected sizes . ( n = 3 ) ( B ) In vitro binding of α-SNAP to cytosolic tails of Orai1 . ( Top panel ) GST-tagged Orai1 cytosolic tails , expressed in E . Coli and immobilized on resin , were incubated with purified 10 nM α-SNAP protein as described above and subjected to western blot analysis using anti-α-SNAP antibody . ( Bottom panel ) Ponceau S staining showing GST-tagged fragment input and their expected sizes . ( n = 3 ) ( C and E ) SOCE in RNAi treated HEK 293 cells stably expressing Orai1 and STIM1 ( C ) or Orai1 and CAD ( E ) . Average , single cell Fura-2 ratios of cells transduced with α-SNAP ( red ) or scr ( black ) RNAi and stimulated with TG . ( n = 3 ) ( D and F ) Measurement of the rate of increase in intracellular Ca2+ . The rate of store-operated Ca2+ influx in ( C ) and ( E ) was estimated by measuring the maximal rate of initial rise in Fura-2 ratios after replenishing Ca2+ in the extracellular buffer . ( p value<0 . 001 ) ( G ) Basal intracellular Ca2+ concentration in unstimulated HEK 293 cells ( Ctrl ) or RNAi treated HEK 293 cells stably expressing Orai1 and CAD . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 015 Given that α-SNAP directly binds the two critical domains involved in the activation of SOCE via the CRAC channel complex , α-SNAP could facilitate a functional coupling between the ER localized Stim1 and Orai1 in the PM . To test this hypothesis and directly place α-SNAP in the molecular sequence of SOCE , we first asked whether the final molecular step of CAD domain mediated activation of SOCE via Orai1 is intact in α-SNAP depleted cells . To address this , we generated an inducible stable HEK 293 cell line , co-expressing Orai1 along with either full length Stim1 or just the soluble CAD domain of Stim1 . As expected , α-SNAP depletion caused significant inhibition of SOCE in cells co-expressing full length Stim1 and Orai1 ( Figure 5C and 5D ) . However , SOCE in cells co-expressing the cytosolic CAD fragment of Stim1 along with Orai1 , although overall smaller , was resistant to α-SNAP depletion ( Figure 5E and 5F ) . Similarly , basal calcium levels in cells co-expressing the cytosolic CAD fragment of Stim1 along with Orai1 , although higher compared to WT HEK 293 cells , were unaffected by α-SNAP depletion ( Figure 5G ) . Importantly , the lack of effect of α-SNAP depletion on CAD mediated constitutive and SOCE conclusively rules out any possible unknown effects on membrane potential or membrane trafficking of other components contributing to the activation of SOCE . Taken together , these data demonstrate that α-SNAP directly binds specific domains within Stim1 and Orai1 to enable a molecular step that precedes CAD mediated activation of SOCE in cells expressing full length Stim1 . Because α-SNAP binds the CAD domain of Stim1 , we wondered whether α-SNAP contributes to the formation of store-depletion induced Stim1 clusters in the junctional ER . To observe these plasma membrane proximal regions of ER in live cells and to determine whether α-SNAP depletion affects Stim1 clustering in the junctional ER , we used Total Internal Reflection Fluorescence ( TIRF ) Microscopy . While the ability of Stim1 to oligomerize and translocate to the junctional ER upon store-depletion appeared normal in α-SNAP deficient cells , the size of Stim1-Orai1 clusters appeared larger ( Figure 6A ) . We adopted a localized thresholding method to detect Stim1-Orai1 cluster boundaries ( Figure 6—figure supplement 1 ) , used it to quantify the size and intensity of individual Stim1–Orai1 clusters , and found that α-SNAP depletion increased the overall size of CRAC channel clusters ( Figure 6B ) . Because Orai1 is thought to be largely dependent on Stim1 for its clustering in ER–PM junctions ( Liou et al . , 2005; Zhang et al . , 2005; Stathopulos et al . , 2006; Wu et al . , 2006; Liou et al . , 2007; Varnai et al . , 2007; Luik et al . , 2008; Stathopulos et al . , 2008 ) , we hypothesized that this increase in cluster size could result from an increase in the density of Stim1 in the junctional ER or could reflect a specific defect in the ability of Stim1 to efficiently co-cluster Orai1 . To distinguish between these possibilities , we first quantified the average intensity of CFP-Stim1 in individual Stim1–Orai1 clusters . We did not find an appreciable difference in the average intensity of CFP-Stim1 estimated from Stim1–Orai1 clusters in control or α-SNAP depleted , HEK 293 cells stably expressing CFP-Stim1 and Orai1-YFP ( Figure 6—figure supplement 2 ) . These data suggest that α-SNAP depletion does not affect the density of Stim1 in junctional clusters . Surprisingly , we found a significant increase in the average Orai1-YFP intensity in α-SNAP depleted clusters ( Figure 6—figure supplement 2 ) , resulting in a significant decrease in the ratio of CFP ( Stim1 ) to YFP ( Orai1 ) across all CRAC channel clusters ( Figure 6C ) as well as in the majority of individual clusters ( Figure 6D ) . These changes and their inhibitory effect on SOCE ( Figure 6—figure supplement 3 ) suggest that α-SNAP regulates an active molecular rearrangement within CRAC channel clusters that is necessary for obtaining optimal Stim1/Orai1 ratios required for the physiological activation of SOCE through CRAC channels . Indeed , two independent recent studies that experimentally manipulated Stim1:Orai1 ratios in junctional clusters showed that the amplitude of calcium selective CRAC currents correlates well with the ratio of Stim1 to Orai1 or CAD to Orai1 within CRAC channel complex ( Mcnally et al . , 2012; Hoover and Lewis , 2011 ) . 10 . 7554/eLife . 00802 . 016Figure 6 . α-SNAP regulates the molecular composition of CRAC channel clusters in ER–PM junctions . ( A ) Stim1 translocation and Orai1 clustering in control and α-SNAP deficient cells . TIRF images of resting and store-depleted , control , and α-SNAP deficient HEK 293 cells stably expressing CFP-Stim1 and Orai1-YFP . Scale bar 10 μm . ( n > 5 ) ( B ) Quantification of Stim1-Orai1 cluster size in control and α-SNAP deficient cells . TIRF images were acquired as in ( A ) and boundaries for individual Stim1-Orai1 clusters were detected ( as shown in Figure 6—figure supplement 1 ) , pooled from nine cells and compared between control ( black ) and α-SNAP depleted cells ( red ) . Histograms show the size comparison of ∼1000–1600 clusters . ( C and D ) Quantification of the relative intensities of CFP-Stim1 vs Orai1-YFP in CRAC channel clusters of RNAi treated cells . Stim1-Orai1 cluster boundaries were detected as described in ( B ) and average CFP vs YFP intensity of each cluster was calculated , pooled from nine cells and compared between control ( black ) and α-SNAP depleted cells ( red ) . ( C ) Average CFP/YFP ratio of pooled clusters from RNAi treated cells and ( D ) CFP/YFP ratio of individual clusters from Figure 6C . ( E and F ) Independent quantification of relative intensities of CFP-Stim1 and Orai1-YFP in CRAC channel clusters over-expressing α-SNAP or empty vector . TIRF images were acquired from Orai1-Stim1 co-expressing HEK 293 cells transiently transfected with α-SNAP ( red ) or empty vector ( black ) . Stim1-Orai1 cluster boundaries were detected and average intensity of CFP-Stim1 vs Orai1-YFP per cluster was calculated as described in Figure 6B–D above , pooled from six cells per group and compared . ( E ) Average CFP/YFP ratio of pooled clusters from α-SNAP ( red ) and empty vector ( black ) over-expressing cells . ( F ) CFP/YFP ratio of individual clusters from Figure 6E . ( **p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01610 . 7554/eLife . 00802 . 017Figure 6—figure supplement 1 . A representative Orai1-YFP image and its corresponding mask . TIRF image was acquired as described in Figure 6 , and Orai1-YFP cluster boundaries were automatically detected using a localized thresholding method in Matlab . Shown here is a representative image and its corresponding binary cluster mask . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01710 . 7554/eLife . 00802 . 018Figure 6—figure supplement 2 . Quantification of CFP-Stim1 and Orai1-YFP intensities in Stim1-Orai1 clusters of RNAi treated cells . Stim1–Orai1 cluster boundaries were detected as described in Figure 6—figure supplement 1 and Figure 6B . Average intensity of each CFP-Stim1 cluster ( Left Panel ) and Orai1-YFP cluster ( Right Panel ) was calculated , pooled from nine cells , and compared between control ( black ) and α-SNAP depleted cells ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01810 . 7554/eLife . 00802 . 019Figure 6—figure supplement 3 . α-SNAP depletion inhibits SOCE in Stim1-Orai1 over-expressing cells . Average Fura-2 ratios of Orai1-Stim1 expressing HEK 293 cells transduced with α-SNAP RNAi ( red ) or scr RNAi ( black ) and stimulated with TG to measure SOCE . ( n = 3 ) DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 01910 . 7554/eLife . 00802 . 020Figure 6—figure supplement 4 . α-SNAP co-expression augments SOCE in Stim1-Orai1 over-expressing cells . Average Fura-2 ratios of Orai1-Stim1 expressing HEK 293 cells transiently transfected with α-SNAP ( red ) or empty vector as a control ( black ) and stimulated with TG to measure SOCE ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 02010 . 7554/eLife . 00802 . 021Figure 6—figure supplement 5 . Quantification of cell surface Orai1 in α-SNAP over-expressing cells . Orai1-BBS-YFP expressing stable line was transiently transfected with α-SNAP ( red ) or empty vector ( black ) . Subsequently , cells were stimulated with TG ( solid lines ) or left unstimulated ( hashed lines ) and incubated with BTX to estimate cell surface Orai1 as described in Figure 3—figure supplement 2 . BTX binding to Orai1-YFP expressing cells was used as a control ( gray line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 02110 . 7554/eLife . 00802 . 022Figure 6—figure supplement 6 . Quantification of Stim1-Orai1 cluster size in α-SNAP over-expressing cells . CFP-Stim1 and Orai1-YFP expressing stable HEK 293 cells co-expressing α-SNAP ( red ) or empty vector control ( black ) from Figure 6E and 6F were quantified as in Figure 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 022 Given this reduction in the CFP ( Stim1 ) to YFP ( Orai1 ) ratio of α-SNAP deficient junctional clusters , we next wondered whether over-expression of α-SNAP would increase the Stim1/Orai1 ratio in clusters and thereby augment SOCE . Remarkably , over-expression of α-SNAP enhanced SOCE beyond what is typically seen in cells stably co-expressing Stim1 and Orai1 ( Figure 6—figure supplement 4 ) . Next , we examined the relative average intensities of CFP and YFP in junctional clusters of cells stably expressing CFP-Stim1 , Orai1-YFP and transiently transfected with α-SNAP or empty vector . Concomitant with an increase in SOCE , over-expression of α-SNAP significantly enhanced the ratio of CFP ( Stim1 ) to YFP ( Orai1 ) across all junctional clusters ( Figure 6E ) as well as majority of individual clusters ( Figure 6F ) . Notably , over-expression of α-SNAP did not affect total surface Orai1 levels in resting or store-depleted cells ( Figure 6—figure supplement 5 ) or average CFP-Stim1 intensity per cell ( data not shown ) . In summary we have shown that α-SNAP directly binds Stim1 and Orai1 and co-localizes with CRAC channel clusters in the junctional ER . Within CRAC channel clusters , α-SNAP enables a crucial and previously unknown molecular re-arrangement that results in optimal Stim1:Orai1 ratios necessary for the physiological activation of SOCE ( Figure 7 ) . 10 . 7554/eLife . 00802 . 023Figure 7 . A hypothetical model of α-SNAP dependent re-arrangement of Stim1-Orai1 molecules within CRAC channel clusters in ER–PM junctions . Calcium ions ( black circles ) , PM localized Orai1 hexamers ( light blue ) ER localized Stim1 ( green ) , α-SNAP ( grey ) . ( Top left panel ) Resting localization of Orai1 hexamers and Stim1 in α-SNAP sufficient cells . ( Top right panel ) α-SNAP sufficient , store-depleted cells form functional CRAC channel clusters with relatively high Stim1/Orai1 ratio . ( Bottom left panel ) Resting localization of Orai1 hexamers and Stim1 in α-SNAP deficient cells is unaffected . ( Bottom right panel ) α-SNAP deficient , store-depleted cells form bigger non-functional Stim1–Orai1 clusters with low Stim1/Orai ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 00802 . 023 We have identified a novel and direct role for α-SNAP in SOCE . Regulation of SOCE by α-SNAP is completely independent of its well-known role in NSF mediated SNARE complex disassembly . We demonstrate that Stim1–Orai1 clustering at ER–PM junctions is insufficient to fully activate SOCE unless followed by a α-SNAP dependent active molecular re-arrangement within CRAC channel clusters . Furthermore , our study opens the possibility that SNAP proteins are involved in organizing other membrane associated macromolecular complexes . Several pieces of data presented here demonstrate a novel mechanism for α-SNAP mediated regulation of CRAC channel activity . Upon store-depletion , α-SNAP but not its usual binding partner , NSF , co-localizes with Stim1–Orai1 clusters . Secondly , dominant inhibition of NSF does not inhibit SOCE . α-SNAP depletion disrupts SOCE without affecting the resting and stimulated subcellular localization of Orai1 and Stim1 . Furthermore , using deletion and point mutations in α-SNAP we show that the domains and residues crucial for facilitating SOCE do not overlap with those known to be essential for binding or activating NSF dependent SNARE complex disassembly ( Barnard et al . , 1997 ) . Finally , the identification of an NSF independent role of α-SNAP is consistent with our previously reported primary genome-wide RNAi screen , where neither silencing of SNARE proteins nor NSF specifically inhibited SOCE in Drosophila S2 cells ( Vig et al . , 2006b ) . A previous domain analysis of α-SNAP suggested that a hydrophobic loop within α-SNAP , although not directly essential for NSF activation , could function as a membrane attachment site to promote its recruitment to the SNARE complex ( Barnard et al . , 1997; Winter et al . , 2009 ) . Indeed , a similar role could explain the inability of the hydrophobic loop mutant of α-SNAP to efficiently bind Stim1 and Orai1 and co-localize with CRAC channel clusters to reconstitute SOCE . Given that Stim1 and Orai1 can co-cluster in the junctional ER in α-SNAP depleted cells , our data suggest that Stim1 and Orai1 binding across ER–PM junctional space per se is not α-SNAP dependent . These data are consistent with many previous studies that have shown a direct interaction between Stim1 and Orai1 ( Yeromin et al . , 2006; Vig et al . , 2006a ) and identified the CAD domain of Stim1 as a critical domain involved in Stim-Orai1 clustering as well as SOCE activation ( Park et al . , 2009; Yuan et al . , 2009 ) . Interestingly , mutagenesis studies within the CAD domain have identified distinct residues that regulate Stim1 clustering vs activation of SOCE ( Park et al . , 2009; Yuan et al . , 2009 ) . Yet , the existence of non-functional Stim1–Orai1 coupling in a physiological context or the dependence of CRAC channel clusters on a third protein for regulating the final steps in SOCE activation has not been previously speculated . While several proteins have been previously identified to bind Stim1 ( Honnappa et al . , 2009; Smyth et al . , 2009; Walsh et al . , 2010b; Krapivinsky et al . , 2011; Lewis , 2011; Singaravelu et al . , 2011; Soboloff et al . , 2012 ) , only CRACR2A and Junctate have been shown to bind both Stim1 as well as Orai1 ( Srikanth et al . , 2010 , 2012 ) . Of these , CRACR2A has been shown to bind CRAC channel clusters in a calcium dependent fashion and its depletion inhibits the formation of Stim1 clusters . Therefore , consistent with the current model of SOCE where Stim1–Orai1 clustering is considered necessary and sufficient for the physiological activation of SOCE , CRACR2A regulates the formation of Stim1–Orai1 clusters ( Srikanth et al . , 2010; Lewis , 2011; Soboloff et al . , 2012 ) . We find that α-SNAP binds distinct and non-overlapping domains of Orai1 and Stim1 when compared to CRACR2A ( Srikanth et al . , 2010 ) . More importantly , in contrast to CRACR2A , α-SNAP depletion inhibits SOCE without decreasing Stim1 clustering , the density of Stim1 in the junctional ER or the ability of Stim1 to bind Orai1 . Therefore , for the first time , our studies have identified the existence of a crucial late step in the molecular sequence of SOCE that involves α-SNAP dependent active re-arrangement of Stim1 and Orai1 molecules within CRAC channel clusters necessary for the physiological activation of SOCE . We place the requirement for α-SNAP just after Stim1–Orai1 clustering but before CAD mediated activation of SOCE . A fraction of α-SNAP binds Stim1 constitutively , however , binding to Orai1 likely stabilizes its interaction with and enhances its recruitment to the CRAC channel clusters since co-expression of the three proteins shows a significant increase in the percentage of cells showing co-localization of α-SNAP with CRAC channel clusters . In turn , co-expression of α-SNAP synergizes with Stim1 and Orai1 to amplify SOCE likely by ensuring that a higher percentage of Stim1–Orai1 clusters form functional CRAC channels . Interestingly , like in the case of α-SNAP depleted cells , we observed an increase in the size of CRAC channel clusters even in cells over-expressing α-SNAP ( Figure 6—figure supplement 6 ) , suggesting that the ratio of Stim1:Orai1 molecules within CRAC channel clusters determines the amplitude of SOCE rather than the size of clusters themselves or the absolute amount of Stim1 in the junctional ER . The increase in size could result from a disrupted stoichiometry between α-SNAP and a fraction of Stim1–Orai1 clusters in α-SNAP depleted as well as over-expressing cells . Alternatively , expansion in cluster size could result from an increase in the total amount of Stim1 in the junctional ER ( Wu et al . , 2006 ) . Future studies employing sub-diffraction , single molecule imaging approaches would help elucidate the dynamics of α-SNAP dependent re-arrangement of Stim1-Orai1 molecules within junctional CRAC channel clusters . Orai1-Myc and Flag-Orai1 , and Stim1-Myc plasmids have been described previously ( Peinelt et al . , 2006; Vig et al . , 2006b ) . To generate Orai1-CFP and Orai1-YFP , human Orai1 was sub cloned into YFP-N1 and CFP-N1 ( Clontech , Mountain View , CA ) . YFP-STIM1 and CFP-STIM1 were a gift from Dr Tobias Meyer’s lab . Full-length human α-SNAP , β-SNAP , and γ-SNAP were amplified from human cDNA libraries and cloned into pcDNA/4TO/Myc-His ( Invitrogen , Grand Island , NY ) and pEYFP-C1 ( Clontech ) vectors . Stim1 cytosolic domains and Orai1 N- and C- tails were cloned into pGEX-4T-2 vector ( GE Healthcare , Pittsburgh , PA ) and named as GST-CT ( 235–685aa ) , GST-CC1-CAD ( 235–448aa ) , GST-CC1 ( 235–344aa ) , and GST-CAD ( 342–448aa ) for Stim1 domains and GST-Orai1-NT ( 1–87aa ) and GST-Orai1-CT ( 228–301aa ) for Orai1 domains . All plasmid DNA transfections were done with Lipofectamine 2000 ( Invitrogen ) or Amaxa nucleofection kit ( Lonza , Basel , Switzerland ) according to manufacturer’s protocol . All lentiviral RNAi transduction experiments were performed in WT HEK 293 , Jurkat , U2OS ( ATCC , Manassas , VA ) , and HEK 293 T-REx ( Invitrogen ) cell lines cultured in high glucose DMEM ( Hyclone , Logan , UT ) or RPMI ( Hyclone ) with 10% fetal bovine serum ( Hyclone ) , 1% penicillin/streptomycin , and GlutaMax ( Gibco , Grand Island , NY ) or stable cell lines generated using these parent lines . CFP-Stim1 and Orai1-YFP were transfected into HEK 293 cells either alone or together , and stable lines were generated by cell sorting and G418 selection . HEK 293 T-REx ( Invitrogen ) cells were used to make double stable lines expressing Orai1-YFP with Tetracycline inducible Stim1-myc or CAD-myc according to the manufacturer’s protocol . Orai1-BBS-YFP and Orai1-YFP was stably transfected into U2OS cells to estimate cell surface Orai1 . U2OS T-REx cells stably expressing inducible WT-NSF or E329Q-NSF mutant were as described ( Dalal et al . , 2004 ) . Anti-Flag antibodies ( F7425 and F3165 ) were from Sigma ( St . Louis , MO ) . Anti-GFP ( A6455 ) and anti-NFATc1 antibodies were from Invitrogen and Biolegend ( San Diego , CA ) respectively . Supernatants from 9E10 hybridoma cultures were used to detect c-Myc . Anti-α-SNAP ( clone 77 . 2 ) was from SySy ( Goettingen , Germany ) , anti-calreticulin ( PA1-24 , 485 ) from Affinity BioReagents ( ABR , Golden , CO ) , and anti-giantin ( PRB-114C ) from Covance ( Princeton , NJ ) . Anti-mouse IgG AF488 or anti-rabbit IgG Cy3 secondary antibodies were purchased from Jackson Immunoresearch ( West Grove , PA ) . Transferrin-alexa 555 was from Invitrogen . All other reagents were from Sigma Aldrich . shRNA cloned into pLKO . 1-puro vectors were purchased from the RNAi Core at Washington University and co-transfected with psPAX packaging and VSV-G envelope plasmids into HEK 293-FT ( Invitrogen ) cells for generating supernatants containing infectious viral particles . Viral supernatants were collected at 48 hr , filtered and used to transduce target cells along with 4 μg/ml polybrene . Cells transduced with viral supernatant were selected using 2 μg/ml puromycin . Cells plated on coverslips were loaded with 1 μM Fura-2 AM in Ringer’s buffer ( 135 mM NaCl , 5 mM KCl , 1 mM CaCl2 , 1 mM MgCl2 , 5 . 6 mM Glucose , and 10 mM Hepes , pH 7 . 4 ) for 40 min in the dark , washed , and used for imaging . An Olympus IX-71 inverted microscope equipped with a Lamda-LS illuminator ( Sutter Instrument , Novato , CA ) , Fura-2 ( 340/380 ) filter set ( Chroma , Bellows Falls , VT ) , a 10X 0 . 3NA objective lens ( Olympus , UPLFLN , Japan ) , and a Photometrics Coolsnap HQ2 CCD camera was used to capture images at a frequency of ∼1 image pair every 2 s . Data were acquired and analyzed using MetaFluor ( Molecular Devices , Sunnyvale , CA ) , Microsoft Excel , and Origin software . At least 40–50 cells were imaged per group in each experiment . Cells ( ∼100 , 000/well ) plated in 96-well plates were loaded with Fura-2 AM as described above . Fura-2 excitation ratios were measured by alternatively exciting the dye at 340 and 380 nm , at a frequency of ∼1 image pair every 4 s and collecting emission at 510 nm using Flexstation III ( Molecular Devices ) equipped with SoftMax Pro 5 software ( Molecular Devices ) . Cells were lysed in each well at the end of the run to compare the efficiency of dye loading across different groups . Data were analyzed using Softmax Pro 5 , Microsoft Excel , and Origin software . Resting and stimulated cells were washed in cold Ringer’s buffer , resuspended in 100 μl of chilled hypotonic buffer ( 10 mM HEPES [pH 7 . 9] , 10 mM KCL , 0 . 1 mM EDTA , 1 mM DTT , and protease inhibitors ) and swelled on ice for 15 min . Subsequently , 0 . 6% NP-40 was added and cells vortexed and centrifuged . The nuclear pellet was washed three times and resuspended in 50 μl cold protein extraction buffer ( 20 mM HEPES [pH 7 . 9] , 0 . 4 M NaCl , 1 mM EDTA , 1 mM DTT , and protease inhibitors ) , vortexed , centrifuged , and extracts were subjected to Western Blot analysis . HEK 293 cells were lysed using buffer containing 1% CHAPS or 1% NP-40 , 150 mM NaCl , 50 mM Tris , pH 8 . 0 , and protease inhibitors . Unless otherwise mentioned , 1/10th of the whole cell lysate ( WCLs ) were used in the lysate input lanes . The remaining WCLs were pre cleared for 1 hr and incubated with appropriate primary antibodies overnight at 4°C followed by Protein A Sepharose beads for 1 hr at 4°C . The immunoprecipitates were boiled and subjected to western blotting using appropriate primary and anti-mouse IgG or anti-rabbit IgG secondary antibodies . For in vitro binding to full length proteins , purified recombinant α-SNAP was incubated with Flag- or myc-Orai-1 or Stim1-myc immunoprecipitated from HEK 293 cells and bound to Protein A Sepharose beads for 1 hr at 4°C . Post-incubation , beads were washed three times and protein complexes eluted by boiling in SDS containing sample buffer and subjected to SDS-PAGE and western blotting . For in vitro binding to GST-fused Stim1and Orai1 fragments , proteins expressed in Lemo21 E . coli ( NEB C2528H , Ipswich , MA ) were purified using glutathione sepharose and incubated with purified His-α-SNAP ( 10–100 nM ) for 1 hr . After washing , the proteins were eluted by boiling beads with SDS-sample buffer and subjected to SDS-PAGE and western blotting . Cells were plated on poly-Lysine coated coverslips ( VWR , Radnor , PA and Fisher , Pittsburgh , PA ) or 35-mm glass bottom culture dishes ( MatTek , Ashland , MA ) , fixed using 2% PFA for 15 min , permeabilized and blocked in 0 . 1% saponin , 3% BSA in PBS for 45 min , and then stained with appropriate primary and secondary antibodies and imaged using a confocal or a wide field microscope . Cells plated onto coverslips were washed three times with serum free DMEM . Transferrin-alexa 555 ( Invitrogen ) was diluted at 25 μg/ml in serum free DMEM and incubated with cells at 37°C for 30 min , followed by three quick washes with serum free DMEM , pH 5 . 5 , to remove surface bound transferrin . Subsequently cells were fixed with 2% PFA and imaged using IX-71 microscope with a 60X 1 . 25 NA objective lens ( Olympus , UPFLN ) . Cell boundaries were identified using YFP fluorescence and total alexa 555 fluorescence per cell was quantified using ImageJ software . Confocal images were captured on an inverted Olympus Fluoview FV1000 microscope using a 60X 1 . 4 NA objective . Thin optical sections of CFP and YFP were obtained by sequentially scanning 440 nm and 515 nm lasers respectively . Image analysis was performed using ImageJ ( NIH ) and Adobe Photoshop . Wide-field epifluorescence images were captured using the Olympus IX-71 microscope described above with a 60X 1 . 25 NA objective lens ( Olympus , UPFLN ) , FITC/YFP/Cy3 dichroics ( Brightline , Semrock , Rochester , NY ) , and the Micro-Manager software . Image analysis was performed using ImageJ ( NIH ) and Adobe Photoshop . TIRF images were captured on a Nikon Eclipse TiE microscope fitted with the Nikon perfect focus system ( TiE-PFS ) for focus stabilization , a motorized stage ( Marzhauser ) , and using an objective type TIRF geometry . Illumination lasers , 405 nm ( Coherent , Cube ) , and 488 nm ( Coherent , Sapphire ) were individually shuttered ( Uniblitz , Vincent Associates ) , combined , expanded , collimated , and focused at the back focal plane of a 100X 1 . 4 NA objective ( Olympus UPLSAPO ) . CFP or YFP images were sequentially acquired by the same objective and separated using a dichroic set ZT405/488/561/640rpc-ZET405/488/561/640m and filtered using ET450/40m or ET525/50m emission filters ( Chroma ) . Images were captured using custom software on an EM-CCD camera ( Andor , iXon DU-897 ) . Orai1-YFP and CFP-STIM1 expressing cells were first imaged in the resting stage and their positions marked . After addition of 1 μM Thapsigargin ( TG ) , the same cells were imaged to avoid any bias in selection of cells based on cluster brightness or morphology . TIRF images were uniformly scaled to generate a binary image mask ( See Figure 6—figure supplement 1 ) and Orai-Stim cluster boundaries were detected by adapting Otsu’s thresholding method to perform localized thresholding in Matlab ( http://www . mathworks . com/help/images/ref/graythresh . html ) . The cluster mask was applied to corresponding Orai1-YFP and CFP-Stim1 images to obtain the size and average CFP or YFP intensity per cluster . Cluster size and intensity data were pooled from roughly nine cells per group and plotted as a histogram . A 13 amino acid bungarotoxin binding site ( BBS ) WRYYESSLEPYPD ( Sekine-Aizawa and Huganir , 2004 ) was inserted into Orai1 extracellular loop between G207 and Q208 . Orai1-BBS-YFP was stably expressed in U2OS cells . After RNAi treatment or transient over-expression of α-SNAP , cells were labeled with Alexa-647 conjugated α-bungarotoxin ( BTX ) ( Invitrogen ) for 30 min on ice . Samples were analyzed using FACS Caliber and FlowJo software .	Calcium is an essential element for many biological functions . In particular , the movement of calcium ions through the cell membrane has a central role in many of the signalling pathways that cells use to communicate with other cells . Signals are produced by calcium ions both entering and leaving the cell , with information being contained in the rate , location , and duration of the flow of ions . Calcium is stored inside cells in a structure called the endoplasmic reticulum , and when stores of calcium are low , special channels in the cell membrane called CRAC ( calcium release activated calcium ) channels are used to ferry more calcium ions into the cell . This process , known as store-operated calcium entry , relies on two important groups of proteins: the Stim proteins that sense when calcium stores are low; and , the Orai structural proteins that form the actual channel . Previous work has shown that when the calcium stores are low , the Stim proteins—which reside in the endoplasmic reticulum—form clusters and these clusters then move to a part of the endoplasmic reticulum that is next to the cell membrane , where they join the Orai1 proteins to form larger clusters . However , to date it has been unclear whether Stim-Orai clustering at the cell membrane is sufficient for CRAC channels to open , or if additional steps are involved . Miao et al . now show that another protein is involved in the formation of functional CRAC channels . Working with fruit fly cells , Miao et al . used genetic techniques to prevent the expression of various proteins that were thought to have a role in the movement of calcium ions through the cell membrane . One of these candidates , a protein called α-SNAP that is found in the internal fluid of the cell , was identified as having a central role in the import of calcium ions into the cell . Further work showed that α-SNAP re-organizes the Stim and Orai proteins to produce working CRAC channels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2013	An essential and NSF independent role for α-SNAP in store-operated calcium entry
The brain displays a remarkable ability to adapt following injury by altering its connections through neural plasticity . Many of the biological mechanisms that underlie plasticity are known , but there is little knowledge as to when , or where in the brain plasticity will occur following injury . This knowledge could guide plasticity-promoting interventions and create a more accurate roadmap of the recovery process following injury . We causally investigated the time-course of plasticity after hippocampal lesions using multi-modal MRI in monkeys . We show that post-injury plasticity is highly dynamic , but also largely predictable on the basis of the functional connectivity of the lesioned region , gradients of cell densities across the cortex and the pre-lesion network structure of the brain . The ability to predict which brain areas will plastically adapt their functional connectivity following injury may allow us to decipher why some brain lesions lead to permanent loss of cognitive function , while others do not . Lesions to the brain set off a cascade of degenerative and protective plasticity-related processes . Distant grey matter degeneration , and a loss of anatomical connectivity of grey matter areas not directly affected by the lesion are common anatomical consequences of a lesion ( Catani and ffytche , 2005; Zaczek et al . , 1980 ) . In addition , the extent of functional disconnection of intact regions is associated with the degree of behavioral impairment following a lesion ( Corbetta et al . , 2005; He et al . , 2007 ) even if the areas remain structurally connected ( van Meer et al . , 2010 ) . Conversely , some brain areas adapt by altering their connectivity patterns and increasing their connections with other , often unaffected areas ( Yogarajah et al . , 2010 ) . It is thus important to be able to predict the areas that will undergo a relative functional disconnection following a lesion , and to predict which areas may functionally adapt in order to identify potential avenues for guiding adaptive plasticity . Currently , there is no quantitative way of predicting how unlesioned brain areas will adapt to injury elsewhere . Recovery following brain injury is highly variable and occurs in stages . Much recovery occurs in the first few weeks following an injury , but functional improvements may continue until much later ( Berthier et al . , 2011; Smania et al . , 2010 ) . Studies in rodents have shown that the microstructural consequences of brain injury can vary dramatically at different times following injury , which could have serious implications for potential treatment strategies ( Hoskison et al . , 2009 ) . However , small animal models of brain injury are not optimal for investigating chronic plastic changes , due to the short lifespan of rodents leading to a conflation of lesion- and neurodevelopmental- or aging-related plasticity . In human studies , pre-lesion scans are rare and are mostly available in patients with pre-existing brain abnormalities , such as patients with epilepsy . In studies of humans with brain lesions , the presence of possible pre-lesion pathology , combined with the non-specific nature of naturally occurring lesions , complicates interpretation . Consequently , little is known about how plasticity that occurs in the chronic stage following injury may differ from that occurring in the acute stage , and when particular functional and structural adaptations may take place . The rapid advance in tools for measuring brain structure and function has lead to a great increase in the number of potentially informative predictors of plasticity following injury . It has recently been proposed that mapping a lesion onto an atlas of connections could predict the remote areas affected and perhaps the behavioral consequences of a lesion ( Kuceyeski et al . , 2014; Thiebaut de Schotten et al . , 2015 ) . While this approach could be greatly informative , it is not yet clear which remote areas may suffer the permanent negative consequences of an injury , and which may adapt and recover . Other studies have suggested that the role of brain regions within the whole brain architecture may be informative for the vulnerability to injury , with hub regions seemingly more likely to be affected in a variety of brain disorders ( Crossley et al . , 2014 ) . This suggests the hypothesis that hub regions may distribute resources following a brain injury in order to aid recovery in areas that are primarily affected by the injury ( Achard et al . , 2012 ) . How this may occur at a microstructural level is unclear . Recently , there has been a resurgence in interest in large-scale gradients in cortical organization ( Beul et al . , 2017; Burt et al . , 2018; Goulas et al . , 2018; Margulies et al . , 2016; Markov et al . , 2014; Sanides , 1962 ) , and how this may enable cortical areas to specialize for distinct cognitive functions ( Chaudhuri et al . , 2015 ) . However , little attention has been paid to whether cortical gradients of microstructural quantities , such as neuronal densities , or glial densities may also impose critical limits on the ability of an area to adapt to injury . Neuron densities vary smoothly across the cortical surface , with prefrontal cortex having less than half the neuron density of V1 ( Collins et al . , 2010 ) . Non-neuronal cells such as astrocytes and microglia can have both beneficial and detrimental effects on post-injury plasticity ( Anderson et al . , 2003; Loane and Kumar , 2016 ) , and the exact distribution of these cells throughout the brain may also constrain or modulate the response of a region to injury . We set out to investigate whether it is possible to predict plastic changes following a discrete , specific lesion , using a bilateral excitotoxic lesion of the hippocampus . The hippocampus is a key part of the episodic memory circuit , but the impact of lesions restricted to the hippocampus itself is not always large ( Malkova and Mishkin , 2003; Zola-Morgan and Squire , 1986 ) . Because of the widespread nature of the episodic memory circuit ( Aggleton and Brown , 1999 ) , we hypothesized that this may be due to functional plasticity in the form of intact brain regions compensating for the damaged area ( a process we previously showed to be critically dependent on cholinergic inputs to inferior temporal cortex following hippocampal disconnection ( Browning et al . , 2010; Croxson et al . , 2012 ) . We acquired MRI scans in macaque monkeys before and at two time points after bilateral excitotoxic hippocampal lesions and found that the brain reacts to injury in a highly dynamic way , which is in part predictable on the basis of the pre-lesion functional connectivity and micro- and macro-structural anatomy . Areas that were most connected to the hippocampus before the lesion reduced their functional connectivity with areas in other modules in the acute stage , and showed a greater loss of grey matter volume during the chronic stage . Nonetheless , they increased their functional connectivity with other areas in the same module during the chronic stage , suggesting that highly dynamic processes of degeneration and plasticity occur in parallel over the year following the lesion . In contrast , hub regions suffered a general loss of functional connectivity during both the acute and chronic stages . Areas with a higher density of neurons lost connectivity with areas within the same module over the chronic period , while those with a higher density of non-neuronal cells ( including glia and cardiovascular support cells ) significantly increased their between-module functional connectivity over the same period , suggesting that a high density of these cells may be important to the plastic recovery process . This is the first study to demonstrate quantitatively a relationship between pre-lesion functional connectivity and the dynamic course of plasticity following a lesion and shows that information across a range of spatial scales can aid in prediction of the plastic recovery process following a lesion . There was a significant reduction in hippocampal volume bilaterally during the acute stage measured by T2-weighted scans ( Figure 1A ) , histologically ( Figure 1B ) and deformation-based morphometry of T1-weighted structural MRI scans ( Figure 1C–D ) . All three analysis methods gave consistent results . Lesions were mostly bilateral and extensive , although there was some apparent sparing of the right hippocampus posteriorly across the five monkeys ( Figure 1; Table 1 ) . However , this is likely an under-estimation of the amount of damage in the posterior part of the hippocampus , which is narrower and therefore more susceptible to partial volume effects with neighbouring tissue . We measured structural and functional changes across the whole brain using high-resolution MRI at three time points: pre lesion , 3 months-post lesion and 12 months post-lesion in five macaque monkeys . The pre-lesion scans also included data from three additional control animals that did not go on to receive lesions . For clarity , we refer to the following stages: acute ( pre-lesion vs . 3 months post-lesion ) and chronic ( 3 months vs . 12 months post-lesion ) . We do not make any claims as to different rates of behavioral recovery during these stages , and acknowledge that cognitive recovery can occur in either stage following a brain insult ( Berthier et al . , 2011; Lazar and Antoniello , 2008 ) . Across all pairwise connections between brain regions , there was an overall increase in the functional connectivity strength over the acute stage ( t6318 = 9 . 37 , p = 1×10−22 ) , and a decrease over the chronic stage ( t6318 = −16 . 85 , p = 2×10−62 ) . In order to understand the specific regional changes that were driving these global effects , we first divided the brain into multiple ‘modules’ , based on the resting-state functional connectivity data using the Louvain algorithm ( Blondel et al . , 2008 ) . Here a ‘module’ is a set of brain regions that have higher functional connectivity with the other brain regions within the set than with brain regions outside the set . We investigated plastic changes to the mean within-module functional connectivity for each brain area , and to the network participation coefficient , which is a measure of how evenly the connections of a brain area are distributed across all of the modules in the brain . Thus brain regions that have a low proportion of their connections with brain regions outside the local module have a low network participation coefficient , whereas brain regions that are strongly connected with regions outside the local module have a high network participation coefficient . On this basis , the network participation coefficient has been proposed as a marker of connector hubs ( Power et al . , 2013 ) . By analyzing within-module functional connectivity and the network participation coefficient , we can build a picture of the changes in processing within and between functional modules over time . As these methods depend on the definition of the modules , which in turn depends on a rather arbitrary choice of a resolution parameter ( lambda ) , we report results that were robust across the entire range tested ( minimum gamma = 0 . 8 , corresponding to two brain modules , maximum gamma = 1 . 4 , corresponding to just one brain region per module ) . Additionally , using deformation-based morphometry , we assessed changes to grey-matter volume over the acute and chronic stages . We identified four factors that we hypothesized to be potential predictors of plasticity following the lesion . First and second , at a cellular level , post-lesion plasticity depends on the ability of neurons to form novel synaptic connections , and on glial cells ( particularly astrocytes and microglia ) and other cardiovascular support cells to aid in the creation and maintenance of such synapses . We thus investigated whether the gradients of neuronal and non-neuronal cell densities across the cortex were associated with plasticity patterns in the acute and chronic stages . To do this , we mapped neuronal and non-neuronal cell densities from a macaque anatomical study ( Collins et al . , 2010 ) onto the Regional Map macroscopic template ( Kötter and Wanke , 2005 ) ( Figure 2A , B ) . Third , studies in humans have suggested that hub regions are strongly affected following a range of neurological and psychiatric disorders , and that these regions are radically reorganized following injury ( Achard et al . , 2012; Crossley et al . , 2014 ) . We therefore investigated whether the hub-like properties of an area could predict its plastic alterations following hippocampal injury . We created a continuous measure of the degree to which brain areas were hubs ( a . k . a . ‘hubness’ ) using the following method . As both network participation coefficient and node strength are proposed measures of hubness , and are positively correlated , we performed a principal components analysis on the strength and participation coefficient data , and took the first principal component , which explained 73 . 19% of the variance in strength and participation coefficient to be our estimate of hubness ( Figure 2C ) . Fourth , we reasoned that the strength of pre-lesion functional connectivity with the hippocampus ( the lesioned region ) should affect the degree to which other regions in the brain plastically reorganize their functional connectivity following the lesion , with regions that were highly functionally connected with the hippocampus likely being most highly affected by the lesion , and consequently most in need of plastic reorganization . We assessed pre-lesion hippocampal functional connectivity with all other cortical regions on the basis of the pre-lesion resting-state fMRI scans and averaged between left and right hippocampus . The average hippocampal functional connectivity is shown in Figure 2D . The strongest functional connectivity was with medial and ventral temporal regions that are in close proximity to the hippocampus . In contrast , dorsal frontal regions showed a slight negative correlation with the hippocampus . In order to test the anatomical validity of these functional connectivity patterns , we compared them to the anatomical connectivity measures from the original ( Stephan et al . , 2001 ) and ‘enhanced’ ( Deco et al . , 2014 ) versions of the CoCoMac tract-tracing atlas . The enhanced version has previously been a better fit to functional connectivity measures than the discrete-valued version of the tract-tracing atlas ( Deco et al . , 2014; Grayson et al . , 2016 ) . Hippocampal functional connectivity measured in the current study was highly correlated with anatomical connectivity measures in both the original ( r = 0 . 54 , p = 2 . 2×10−7 ) and enhanced versions of the CoCoMac Atlas ( r = 0 . 60 , p = 3 . 6×10−9 ) . We entered the neuronal density , non-neuronal cell density , hubness and pre-lesion hippocampal functional connectivity as predictors of acute changes in network participation coefficient in a stepwise regression ( Figure 3A ) . The model significantly predicted the cortex-wide pattern of acute changes in network participation coefficient , explaining over half of the variance ( F2 , 75 = 42 . 24 , p = 5×10−13 , r2 = 0 . 53 , Figure 3B ) . Hubness ( t73 = −5 . 70 , p = 2×10−7 ) , and pre-lesion hippocampal functional connectivity ( t73 = −5 . 25 , p = 1×10−6 ) were significantly associated with a drop in network participation coefficient over the acute stage ( Figure 3C ) . Neither neuron density ( t73 = −1 . 04 , p = 0 . 29 ) nor non-neuronal cell density showed a significant association ( t73 = −0 . 55 , p = 0 . 59 ) . As both the calculation of the acute and chronic stage changes contained the three-month timepoint , they were not independent . In order to identify the chronic stage changes that were independent of the acute stage changes , we constructed a general linear model , using the acute stage changes to predict the chronic stage changes . The relationship between the acute and chronic stage changes to the network participation coefficient did not differ from chance ( p = 0 . 66 , corrected for the shared timepoint , see Materials and methods ) , suggesting that distinct degenerative and plastic processes affected network participation at the two stages . The residuals of this model were taken to be the chronic stage changes that were independent of the acute stage changes . We used the same predictors as during the acute stage to predict the chronic stage changes in the network participation coefficient ( Figure 3D ) . The model significantly predicted the cortex-wide pattern of chronic stage changes in network participation coefficient ( F2 , 75 = 24 . 3 , p = 7×10−9 , r2 = 0 . 39 , Figure 3E ) . Non-neuronal cell density was significantly associated with a rise in network participation coefficient over the chronic stage ( t73 = 4 . 70 , p = 1×10−5 ) . As in the acute phase , hubness was significantly associated with a drop in network participation coefficient during the chronic stage ( t73 = −4 . 93 , p – 4 × 10−6 ) ( Figure 3F ) . Neither neuron density ( t73 = 1 . 57 , p = 0 . 12 ) nor hippocampal functional connectivity ( t73 = 1 . 24 , p = 0 . 22 ) were significant predictors of chronic stage network participation changes . A model with hubness as the lone predictor ( identified with stepwise regression ) significantly predicted the cortex-wide pattern of acute stage changes in within-module functional connectivity ( Figure 4B; Figure 4C , F1 , 76 = 10 . 25 , p = 0 . 002 , r2 = 0 . 12 , hubness t73 = −3 . 20 , p = 0 . 002 , neuron density t73 = −1 . 39 , p = 0 . 168 , non-neuronal cell density t73 = −0 . 92 , p = 0 . 359 , pre-lesion hippocampal functional connectivity t73 = −0 . 28 , p = 0 . 782 ) . Acute ( Figure 4A ) and chronic stage ( Figure 4D ) changes to within-module connectivity were more strongly positively associated than expected by chance ( p < 0 . 001 , corrected for the shared timepoint , see Materials and methods ) , suggesting that there may have been a continuation of degenerative or plastic processes from the acute to chronic stage . The residuals of this model were used to identify the independent chronic stage changes to within-module connectivity . The stepwise regression model significantly predicted the cortex-wide pattern of chronic stage changes in within-module functional connectivity ( F3 , 74 = 7 . 68 , p = 0 . 0002 , r2 = 0 . 24 , Figure 4E ) . Neuron density ( t73 = −2 . 54 , p = 0 . 013 ) was a significant predictor of a drop in within-module functional connectivity during the chronic stage . In contrast , pre-lesion hippocampal functional connectivity was associated with a chronic stage increase in within-module functional connectivity ( t73 = 4 . 04 , p = 0 . 0001 ) . Non-neuronal cell density was not included in the final model ( t73 = −1 . 33 , p = 0 . 19 ) . Hubness ( t73 = −2 . 08 , p = 0 . 04 ) was a significant predictor of a chronic stage decrease in within-module functional connectivity across most ( lambda = 0 . 8–1 and 1 . 3–1 . 4 ) , but not all ( lambda = 1 . 1 , p = 0 . 053 , lambda = 1 . 2 , p = 0 . 075 ) of the repetitions of the analysis with different resolution parameters ( lambda ) , and thus may be viewed as a marginal result ( Figure 4F ) . None of the four predictors significantly predicted acute stage changes in grey matter volume ( neuron density: t73 = 1 . 10 , p = 0 . 27 , non-neuronal cell density: t73 = 0 . 40 , p = 0 . 69 , hubness: t73 = 1 . 27 , p = 0 . 21 , hippocampal functional connectivity: t73 = −0 . 11 , p = 0 . 91 ) ( Figure 5A–C ) . Acute and chronic stage changes to cortical grey matter volume were more strongly positively associated than expected by chance ( p = 0 . 01 , corrected for the shared timepoint , see Materials and methods ) , suggesting that there may have been a continuation of degenerative grey matter loss from the acute to chronic stage . The residuals of this model were used to identify the independent chronic stage changes to cortical grey matter volume . The stepwise regression model for chronic stage changes to grey matter volume included only pre-lesion hippocampal functional connectivity in the final model ( F1 , 76 = 16 . 39 , r2 = 0 . 17 , t73 = −4 . 05 , p = 0 . 0001 ) . Neuron density ( t73 = 1 . 58 , p = 0 . 118 ) , non-neuronal cell density ( t73 = 0 . 28 , p = 0 . 779 ) and hubness ( t73 = −1 . 92 , p = 0 . 058 ) did not make the cut-off for inclusion in the model ( Figure 5E–G ) . In order to investigate grey-matter volume changes to the whole-brain ( not restricted to cortex ) , we performed a deformation-based morphometry analysis of the grey-matter volume changes using a linear mixed model . Results are shown in Figure 5D , H , thresholded at p < 0 . 005 and a minimum cluster size of 5 mm3 ( Sallet et al . , 2011 ) . During the acute stage , there were very limited volumetric decreases in the medial septum , amygdala and dorsal premotor cortex . No increases survived thresholding ( Figure 5D ) . At the chronic stage , we still saw decreases in the medial septum , but also a larger range of decreases . Some of these were also in areas that are monosynaptically connected with the hippocampus: the medial orbitofrontal cortex , posterior cingulate cortex and posterior parahippocampal cortex . There were also more extensive volumetric decreases , in the anterior prefrontal cortex ( medial and lateral ) , ventrolateral prefrontal cortex , dorsal striatum , visual cortex and superior temporal cortex . There were also volumetric increases in the cerebellum , midbrain and premotor cortex . These results did not survive multiple comparisons correction ( possibly due to our small sample size ) , so these changes should be viewed with this caveat in mind . We investigated the effect that the hippocampal lesions had on the macroconnectivity structure by examining the changes in individual modules , which , along with hubs are considered the canonical forms of integration and segregation , and hallmarks of interareal connectomes ( Rubinov , 2016 ) . We estimated modules based on the pre-lesion data , repeated 10 , 000 times . The most reliable modules are shown in Figure 6A . Four modules were identified , orbitofrontal cortex/anterior temporal lobe , posterior temporal , parieto-occipital and dorsal frontal . The relative functional connectivity of these modules within the pre-lesion network is shown in Figure 6B ( colors as in 6A ) . In these force-directed graph representations , functional connectivity acts as an attractive force between two nodes , so nodes that are closer together are more highly connected . The parieto-occipital module ( orange ) is highly connected to the three other modules . Following the orange parieto-occipital module through time ( Figure 6B ) , three months after the lesion the parieto-occipital module is still relatively closely connected both with itself and with the other three networks ( although its nodes have dispersed a little ) . At 12 months after the lesion however , this module ( orange ) becomes dramatically dispersed . We quantified this dispersion as the mean drop in within-module functional connectivity over the acute and chronic stages ( Figure 6C–D ) . During the acute stage , ( Figure 6C ) , there is not a large drop in within-module functional connectivity , although there is a significant difference between modules ( F3 , 76=19 . 54 , p = 4×10−9 ) , with the parieto-occipital module dispersing somewhat , and the dorsal frontal and OFC/anterior temporal modules increasing their within-module functional connectivity . During the chronic stage ( Figure 6D ) , there is a drop in within-module functional connectivity across all modules , with by far the greatest dispersion occurring in the parieto-occipital module ( F3 , 76=41 . 86 , p = 4×10−16 ) . The dispersion of the parieto-occipital module , which previously acted as a link between other modules , also led to a drop in connectivity between the modules , as seen by an increase in the modularity during the chronic stage ( pre-lesion: 0 . 33 , 3 months post-lesion: 0 . 34 , 12 months post-lesion: 0 . 49; F2 , 297=60230 , p~0 ) . There were always fewer modules at 12 months after the lesion compared to the pre-lesion scan , at all values of lambda tested ( range 0 . 8–1 . 4 ) . In an exploratory analysis we performed a stepwise regression in order to assess the relationship between cell densities and ‘hubness’ . A model containing both neuron and non-neuronal cell density significantly predicted hubness ( F2 , 75 = 6 . 18 , r2 = 0 . 14 , p = 0 . 003 ) . Neuron density was positively associated with hubness ( t = 3 . 49 , p = 0 . 008 ) , while non-neuronal cell density was negatively associated with hubness ( t = −2 . 19 , p = 0 . 032 ) . In order to test whether the above results were due to the fact that , in some cases , scans of different monkeys were used at different timepoints , we repeated all analyses with the data from two monkeys with a complete set of pre- and post-lesion scans . We tested whether the beta values of the four independent variables ( neuron and non-neuronal cell densities , hubness and pre-lesion hippocampal connectivity ) for each of the above regression analyses significantly differed between the two datasets using non-parametric statistics ( see Materials and methods ) . Of the 24 beta-values assessed , the following differences in beta-values between datasets were observed . Hippocampal connectivity as a predictor for chronic changes in participation coefficient ( p = 0 . 034 ) , hubness as a predictor of acute changes in within-module connectivity ( p = 0 . 0004 ) and hubness as a predictor of chronic changes in grey matter volume ( p = 0 . 019 ) . In all three of these cases the significance of the result , and hence the interpretation did not change; that is , hubness remained a significant predictor of acute changes in within-module connectivity in the full and two-monkey datasets , and the two other predictors remained non-significant . Changes in activation in distant brain regions are widely agreed to be among the first adaptations following brain injury before a return to more normal-appearing activation patterns in the spared tissue close to the affected site ( Cramer , 2008 ) . Recently in humans , connectomic information derived from databases of healthy humans has been used to identify remote areas likely to be affected by a lesion ( Kuceyeski et al . , 2014; Thiebaut de Schotten et al . , 2015 ) , but the relationship between the brain’s connectivity profile and the dynamics of plasticity had not previously been investigated . Here , we show that functional connectivity can be highly predictive of dynamic plastic changes in both the acute and chronic stages . Stronger pre-lesion functional connectivity of the hippocampus was associated with a drop in network participation over the acute stage and cortical grey matter volume over the chronic stage . The strength of connectivity with the hippocampus was also associated with an increase in within-module connectivity over the chronic stage . An intriguing possibility that should be investigated in larger future studies is that the functional connectivity loss and recovery may correspond to the timeline of loss and recovery of behavioral function following a lesion . With just three scans over a year-long period , we were able to detect interesting dynamics of local and global plasticity . This begs the question: at which stage over the year following injury are the majority of changes happening ? In a recent study , Grayson and colleagues examined network changes in functional connectivity during reversible chemogenetic suppression of amygdala activity ( Grayson et al . , 2016 ) . In the minutes and hours following amygdala suppression , they were already able to detect some network changes , with strong reduction in functional connectivity of the amygdala and its local modules . On visual comparison of the force-directed graph plots of the current study ( Figure 6 ) and the study of Grayson et al ( their Figure 7 ) , it appears that the global network structure is better preserved following chemogenetic disruption of the amygdala , than following permanent lesions of the hippocampus . This is not particularly surprising as permanent lesions are more likely to induce large-scale plasticity , on the timescales of the present study . In that study they did not explicitly calculate changes in hub functional connectivity or other graph-theory properties presented here , limiting the ability to directly compare results . Nonetheless , future studies combining chemogenetic inactivations and permanent lesions of the same brain regions with multiple scanning timepoints have the potential to uncover the fine-grained timeline of global brain alterations following disruption of brain regions , and disentangle how these two methods for interfering with the function of an area may induce different alterations to brain functional connectivity and plasticity . Our findings also have implications for the understanding of the role of medial temporal lobe structures in memory function . While lesions of the hippocampus have been implicated in human amnesia for decades ( Corkin et al . , 1997; Scoville and Milner , 1957 ) , it is also clear that even focal hippocampal damage has widespread consequences beyond the immediate functional damage , and that memory is a distributed process that contains both segregation and overlap of function ( Gaffan , 2002 ) . This concept of connectional diaschisis ( Carrera and Tononi , 2014 ) has also been identified in human patients with focal hippocampal damage , where the functional alterations to a network extended far beyond the structural damage ( Henson et al . , 2016 ) . We significantly extended these findings by quantitatively predicting changes in whole-brain functional connectivity and grey matter volume from pre-lesion hippocampal functional connectivity , microstructural gradients and network-based brain measures . The processes underlying the mechanism of an excitotoxic lesion are partly overlapping with those involved in human brain injury . The initial phase of an ischemic event , for example , leads to excitotoxic death via activation of glutamate receptors , as in our deliberate NMDA lesion , but this is only one of a cascade of processes ( Cramer , 2008 ) . The mechanisms involved in traumatic brain injury are less similar to our excitotoxic lesions , starting with cerebral edema and increased intracranial pressure , followed by a number of other factors of which glutamate excitotoxicity is just one ( Kinoshita , 2016 ) . The strength of a specific NMDA-induced lesion , which spares fibers of passage within or adjacent to the area ( Coffey et al . , 1988; Köhler and Schwarcz , 1983 ) is that we can study the effect of damage to a specific area on the rest of the brain . None of our factors predicted grey matter loss over the acute stage , but areas that were highly connected with the hippocampus before the lesion suffered a greater loss of grey matter volume over the chronic stage . We observed decreases in the volume of the medial septum , amygdala and posterior parahippocampal cortex . These regions and the white matter tracts connecting them to the hippocampus are also affected in human subjects with developmental amnesia ( Dzieciol et al . , 2017; Olsen et al . , 2013 ) and in people born very preterm ( Ball et al . , 2012; Caldinelli et al . , 2017; Froudist-Walsh et al . , 2017; Salvan et al . , 2014; Tseng et al . , 2017 ) . Although the severe structural abnormalities associated with developmental amnesia lead to seemingly permanent impairments to episodic memory ( Vargha-Khadem et al . , 2001 ) , milder damage to this circuit may enable plastic changes in cortical functional connectivity to partially compensate for damage to the core episodic memory circuit ( Isaacs et al . , 2003; Nosarti and Froudist-Walsh , 2016 ) . In the present study , the incomplete damage to subcortical structures such as the mammillary bodies , fornix and connected thalamic subregions in combination with plastic changes to spared areas may be crucial for the preservation or recovery or anterograde memory abilities ( Baxter , 2013; Froudist-Walsh et al . , 2018; Mitchell et al . , 2008 ) . Nonetheless , we acknowledge a limitation of the study is that our design cannot distinguish between compensatory and maladaptive plasticity . We did not see a significant relationship between neuronal or non-neuronal density and acute post-lesion plasticity . Neuronal cell density was significantly associated with the decrease in within-module functional connectivity during the chronic stage . Dendritic tree size and spine count tend to show opposite gradients to neuron density , with lowest values in early visual cortex and peaking in higher association areas ( Elston et al . , 2010; Scholtens et al . , 2014 ) . Thus the loss of connectivity in areas with higher neuron density may be reflective of other factors , such as a lack of dendritic spines that can be crucial for synaptic plasticity . Indeed , local and distant remodeling of spines and dendritic trees has been observed following stroke ( Brown et al . , 2007; Brown et al . , 2010; Nudo , 2013 ) . We found that non-neuronal cell density was significantly positively associated with the increase in network participation during the chronic stage . Although synaptic plasticity is traditionally thought of as being neuronally initiated , it is now clear that astrocytes and microglia can modify synaptic connectivity in a variety of ways ( Ben Achour and Pascual , 2010; Allen and Barres , 2005; Araque et al . , 1999; Ullian et al . , 2004 ) and can even alter synaptic strength in the absence of neuronal activity ( Clark et al . , 2015 ) . Astrocytes and microglia can have both beneficial and detrimental effects on post-injury plasticity ( Anderson et al . , 2003; Loane and Kumar , 2016 ) and have emerged as promising candidates for treatment following acquired brain injury in humans ( Barreto et al . , 2011; Loane and Kumar , 2016 ) . Our finding that non-neuronal cell density positively correlates with the increase in the network participation coefficient during the chronic stage provides a novel link between the local role of glia at the synapse , and plasticity of large-scale functional connectivity patterns . We found that hub regions were more likely to lose functional connectivity with other regions ( reflected in a drop in both within-module functional connectivity and network participation ) following a lesion . This supports the idea that hubs are generally affected following brain injury or disorder . Crossley et al . ( 2014 ) put forward two hypotheses as to why hub regions are more likely to suffer pathology in brain disorders ( Crossley et al . , 2014 ) . The first hypothesis stated that hub regions are more functionally valuable , and therefore damage to hub regions is more likely to be symptomatic than damage elsewhere . Here we show that hub regions are in fact more likely to suffer a loss of functional connectivity , even if the primary site of injury – the hippocampus – is not itself a hub . This coincides to a greater degree with the second hypothesis of Crossley et al . , namely that hubs are biologically costly , and thus more vulnerable to various pathogenic processes . An extension of that hypothesis is that hubs are more likely to be connected to the site of primary insult ( in this case the hippocampus ) , and more likely to suffer from diaschisis as a result . We showed that ‘hubness’ was an independent predictor of structural and functional losses following a lesion , even after accounting for the effects of functional connectivity to the lesioned area . The degradation of the hubs was also associated with a destruction of the overall network structure in the chronic stage . At 12 months following the lesion , the whole brain network had separated into a smaller number of weakly interconnected modules . This demonstrates effects that focal lesions can have on global brain function . The mechanism underlying the vulnerability of hubs to injury requires further study . Speculatively , in an exploratory analysis , we found a relatively high neuron: non-neuronal cell ratio in hubs , perhaps indicating a lack of glial other support cells per neuron . This may mean that hubs are less able to adapt to injury than non-hub areas . Several studies have recently examined the relationships between cell densities and hub properties yet consistent relationships have yet to emerge , perhaps due to the use of different experimental techniques and definitions of hubs across different species ( Beul et al . , 2015; Beul et al . , 2017; van den Heuvel et al . , 2015; Rubinov et al . , 2015; Scholtens et al . , 2014 ) . Given our finding that plasticity following a lesion is highly dependent on the cellular composition of different brain regions , it may be the case that the specific cellular composition of the hippocampus may also have played a role in the patterns of plasticity we see here . The unique connectivity of the hippocampus relative to other brain regions may also be an important factor . This suggests that we may see very different patterns of plasticity following lesions to other brain regions . Future studies will be needed to determine if the relationships between connectivity patterns , gradients of microstructure and patterns of plasticity following brain injury can be generalised to lesions to other brain regions or are specific to the hippocampus . By combining precise anatomical lesions with multiple , multi-modal scans across a period of a year following a lesion , we were able to make three contributions to the literature . Firstly , we show that functional and structural changes can greatly differ between acute and chronic stages . This highlights the importance of carefully considering the time since injury when studying post-lesion plasticity and behavioral recovery . We advocate , where possible , the collection of data at multiple time points following injury in order to accurately map the dynamic recovery process . Secondly , while it has been known for some time that areas connected to a lesioned brain region are more likely to be affected by the lesion than non-connected areas , to our knowledge this is the first study to show quantitatively that post-lesion plasticity patterns depend on pre-lesion functional connectivity . Lastly , we link across spatial scales , and show how microstructural gradients and macrostructural network measures can provide additional predictive value and insights into the plasticity process . Data are available to download from the INDI PRIMatE Data Exchange ( Milham et al . , 2018 ) : https://www . nitrc . org/account/login . php ? return_to=http://fcon_1000 . projects . nitrc . org/indi/PRIMEdownloads . html: Mount Sinai Philips Achieva 3T dataset . Users will first be prompted to log on to NITRC and will need to register with the 1000 Functional Connectomes Project website on NITRC ( http://fcon_1000 . projects . nitrc . org/indi/PRIME/mssm1 . html ) to gain access to the PRIME-DE datasets . The code used for analysis has been made available on Github: https://github . com/seanfw/froudist-walsh-et-al-elife-2018 ( Froudist-Walsh , 2018; copy archived at https://github . com/elifesciences-publications/froudist-walsh-et-al-elife-2018 ) . Subjects were seven male rhesus macaque monkeys ( Macaca mulatta; mean age at start of experiment 3 . 5 years , range 2 . 9–4 years , mean weight at start of scanning 6 . 0 kg , range 4 . 7–7 . 2 kg ) , and one female cynomolgus macaque monkey ( Macaca fascicularis; 8 years at start of experiment , 4 . 7 kg at start of scanning ) . 4 of the male monkeys and the female monkey received bilateral neurotoxic hippocampal lesions as described below . The animals were young adults at the time of lesion ( mean age for the males , 4 . 4 years , range 3 . 7–4 . 75 years; female age 8 years ) . The other three males acted as unoperated controls , along with pre-lesion data acquired before the lesions in the other monkeys . They were scanned at the same point in the behavioral study as the operated males . Full datasets were not available for every monkey as , due to the difficulty associated with acquiring high-resolution data from monkeys , some datasets were not of sufficient quality . The data acquired for each monkey is shown in Table 2 . All monkeys except the female cynomolgus were tested on a test of episodic memory , the object-in-place scene learning task . The behavioral results from these monkeys M , N , S and T are described elsewhere ( H1-H4 in Froudist-Walsh et al . , 2018 ) . The monkeys had a retrograde memory impairment but no anterograde memory impairment on the episodic memory task . Monkeys received MRI-guided bilateral neurotoxic hippocampal lesions using methods described by Hampton et al ( Hampton et al . , 2004 ) . Neurosurgical procedures were performed in a dedicated operating theatre under aseptic conditions . Briefly , monkeys were sedated with a cocktail of dex-medetomidine ( 0 . 01 mg/kg ) , buprenorphine ( 0 . 01 mg/kg ) and midazolam ( 0 . 1 mg/kg ) given i . m . . Where necessary , top-ups were given of dex-medetomidine ( 0 . 003 mg/kg ) and midazolam ( 0 . 1 mg/kg ) without buprenorphine ( to avoid excessive respiratory depression ) and any further top-ups of dex-medetomidine ( 0 . 003 mg/kg ) only as necessary . This protocol was selected to avoid the use of the NMDA antagonist ketamine , which would potentially counteract the effects of the NMDA used as an excitotoxin ( Hampton et al . , 2004 ) . Monkeys were intubated , an i . v . catheter placed and anesthesia was maintained with sevoflurane ( 1 . 5–4% , to effect , in 100% oxygen ) . Monkeys were given glycopyrrolate ( 0 . 01 mg/kg i . m . ) , antibiotics ( Cefazolin , 25 mg/kg i . m . ) , steroids ( methylprednisolone , 20 mg/kg i . v . ) , non-steroidal anti-inflammatories ( meloxicam , 0 . 2 mg/kg i . v . ) , and a H2 receptor antagonist ( ranitidine , 1 mg/kg , i . v . ) to prevent against gastric ulceration following the administration of both steroids and non-steroidal anti-inflammatories . Atipamezole was used to reverse the α2-adrenergic agonist if necessary , once anesthesia was stabilized . Monkeys received i . v . fluids throughout the procedure ( 5 ml/kg/hr i . v . ) . The monkey was placed in a stereotaxic frame in exactly the same position as for the pre-operative structural MRI scan ( employing a tooth marker; Saunders et al . , 1990 ) . The head was cleaned with antimicrobial cleaner and the skin and underlying galea were opened in layers . Small holes were drilled over the injection entry points: one dorsal and posterior to the long axis of the hippocampus and one dorsal to the uncus in each hemisphere ( see Hampton et al . ( 2004 ) for details ) . Two micromanipulators ( Kopf Instruments , Tujunga , CA ) were fitted with gas-tight syringes ( Hamilton , Reno , NV ) with a 28 ga needle , point style 4 , using measurements obtained from the preoperative T1-weighted scan at the most anterior extent of the hippocampus and injections of N-methyl D-aspartate ( NMDA; 0 . 3 M in sterile saline ) were made from anterior to posterior , spaced 1 . 5 mm apart . Each injection was 3 μl in volume , made at a rate of 0 . 5 μl/min , with 1 min between targets . After the final injection the needle was raised 0 . 5 mm and 10 min elapsed before it was extracted . For the uncus injections two injections per hemisphere were made , 3 μl in volume , made at a rate of 0 . 5 μl/min , with 3 min between targets . Propanolol ( 0 . 5 ml of 1 mg/ml per dose ) was administered immediately prior to the NMDA injections and re-administered as necessary ( up to four times ) to prevent tachycardia during the injections due to nonspecific effects of NMDA . One monkey received propofol ( 4 . 0 ml total in boluses of 0 . 5–1 . 0 ml of a 10 mg/ml solution ) to supplement anesthesia , due to tachypnoea , also likely to be a nonspecific effect of NMDA . Once the lesion was completed the skin and galea were sewn in layers . When the lesion was complete , monkeys received 0 . 2 mg/kg metoclopramide ( i . m . ) to prevent postoperative vomiting . Monkeys also received 0 . 1 mg/kg midazolam ( i . m . ) to prevent seizures . They were extubated when a swallowing reflex was evident , returned to the home cage , and monitored continuously until normal posture was regained . Post-operatively monkeys were treated with antibiotics , steroids and analgesia for 3–5 days . Operated monkeys were returned to their social groups within 3 days of the surgery . Following the first surgery we assessed the lesion extent with a T2-weighted scan ( Málková et al . , 2001 ) and used the result to plan the second surgery , targeting the injection co-ordinates to regions with low hypersignal . All monkeys received two lesion surgeries except monkey E , which only required one . Whole-brain BOLD functional MRI data were collected for 40 min using a three-dimensional sequence with the following parameters: 40 axial slices; dimensions 1 . 5 × 1 . 5×1 . 5 mm; TR , 2600 ms; TE , 19 ms; 988 volumes , acceleration factor = 2 . A structural scan ( three averages ) was acquired for each monkey using a T1-weighted magnetization-prepared rapid-acquisition gradient echo sequence ( 0 . 5 × 0 . 5×0 . 5 mm ) . An additional T1-weighted scan and a T2-weighted scan ( 0 . 5 × 0 . 5×0 . 5 mm ) were acquired 6 days post-operatively to assess lesion extent . For the resting-state fMRI scans , isoflurane levels were kept to a minimum to ensure the preservation of resting-state networks: mean isoflurane 1 . 2% , range 1 . 0–1 . 6% ( Hutchison et al . , 2014; Vincent et al . , 2007 ) . Resting-state fMRI was carried out at least 2 hr after ketamine administration , to reduce detrimental effects of ketamine on resting-state networks ( Bonhomme et al . , 2016 ) . End-tidal CO2 was maintained in a normocapnic range wherever possible , to avoid effects of hypercapnia on the BOLD signal: mean CO2 39 mmHg , range 33–45 mmHg ( Bandettini and Wong , 1997; Kastrup et al . , 1999; Rostrup et al . , 2000 ) . At the end of the study , monkeys were deeply anaesthetized with ketamine ( 10 mg/kg ) , intubated and given sodium barbiturate ( sodium pentobarbital , 100 mg/kg ) intravenously . They were then transcardially perfused with 0 . 9% saline followed by 4% parafomaaldehyde . Brains were post-fixed in paraformaldehyde overnight and then cryoprotected in 30% sucrose solution in 0 . 9% saline and cut into 50 μm sections coronally on a freezing microtome . 1 in five sections was stained with cresyl violet for cell bodies . The sections containing the hippocampus were photographed using a Nikon Eclipse 80i light microscope with a 4x objective . Hippocampal volumetric reduction was carried out in Fiji , a version of the image analysis program ImageJ ( https://imagej . nih . gov/ij/ ) . The volume of the hippocampus was manually delineated on sections of the monkey atlas ‘Red’ ( using criteria from Málková et al . , 2001 ) and the remaining hippocampal volume of the hippocampus was manually delineated on images of the cresyl violet sections . The sections were then nonlinearly warped to the atlas using the function bUnwarpJ and the volume of each hippocampal section calculated as a percentage of normal hippocampal volume ( Table 1 ) . The overlap between the remaining hippocampal volume across all five monkeys and normal hippocampal volume is shown in Figure 1B . The structural data were first analyzed using a VBM-style analysis as employed by Sallet et al . ( Sallet et al . , 2011 ) , using the tools FNIRT and Randomise ( Jenkinson et al . , 2012; Winkler et al . , 2014 ) . First , all brains were warped onto the MNI rhesus macaque atlas template ( Frey et al . , 2011 ) using the affine linear registration tool FLIRT and then the nonlinear registration tool FNIRT to produce a study-specific template image . Because the amount of warping expected from the pre-operative to 3 months and 12 months time points was disproportionately large due to the lesions , we included all off the brains , not just the control data , in the template ( Reuter and Fischl , 2011; Reuter et al . , 2012 ) . The nonlinear warping underwent five iterations , each with a higher resolution warp and increasing refinement of the template , with the final warp using a warp resolution ( knot-spacing of cubic b-splines ) of 1 mm isotropic . The restricted log determinant of the Jacobian of the warp field for each brain to the template was extracted . This is the scalar value of the amount of directional stretching required to align each structural image with the template . Voxelwise analysis was carried out on an area limited by a grey matter mask extracted using automated segmentation with FAST on the rhesus macaque MNI template ( Frey et al . , 2011 ) . Longitudinal changes in grey matter volume were assessed using a linear mixed-effects model , implemented in Matlab with the FreeSurfer function lme_mass_fit_vw for mass-univariate linear mixed model analysis ( Bernal-Rusiel et al . , 2013 ) ( https://surfer . nmr . mgh . harvard . edu/fswiki/LinearMixedEffectsModels ) . For consistency , only time points and monkeys for which we had resting-state data were analysed . Regions were designated as significant if they passed a threshold of p < 0 . 005 , with a cluster extent threshold of 5 mm3 voxels ( Sallet et al . , 2011 ) .	The brain has the ability to adapt after injury , a process known as plasticity . When one area sustains damage , for example following a car accident or stroke , other areas change their activity and structure to compensate . Understanding how this happens is critical to helping people recover from brain injuries . Certain factors may affect how well the brain can repair itself . These include how much the damaged area interacts with other areas , and which cell types different areas of the brain contain . Froudist-Walsh et al . set out to determine how these factors influence recovery from brain injury in monkeys , whose brains are similar to our own . The monkeys had damage to a structure called the hippocampus . This part of the brain has a key role in memory , which is often impaired in patients with brain injuries . The hippocampus cannot repair itself because the brain has only a limited capacity to grow new neurons . Instead , the brain attempts to compensate for disruption to the hippocampus via changes in other , undamaged areas . Using brain imaging , Froudist-Walsh et al . show that the types of changes that occur depend on how much time has passed since the injury . In the first three months , many areas of the brain change how much they coordinate their activity with other areas . Highly connected areas reduce their communication with other areas the most . In the long-term , the responses of brain areas depend more on which cell types they contain . Areas with more support cells known as “glia” – which supply nutrients and energy to neurons – are better able to adapt their connectivity up to a year after the injury . These findings may ultimately benefit people who have suffered brain injuries after accidents or stroke . They suggest that stimulating intact brain areas may be helpful in the months immediately after an injury . By contrast , long-term therapy may need to focus more on structural repair . Future studies must build on these results to discover the best ways to induce successful recovery from brain injury .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Macro-connectomics and microstructure predict dynamic plasticity patterns in the non-human primate brain
In several neurodegenerative diseases and myelin disorders , the degeneration profiles of myelinated axons are compatible with underlying energy deficits . However , it is presently impossible to measure selectively axonal ATP levels in the electrically active nervous system . We combined transgenic expression of an ATP-sensor in neurons of mice with confocal FRET imaging and electrophysiological recordings of acutely isolated optic nerves . This allowed us to monitor dynamic changes and activity-dependent axonal ATP homeostasis at the cellular level and in real time . We find that changes in ATP levels correlate well with compound action potentials . However , this correlation is disrupted when metabolism of lactate is inhibited , suggesting that axonal glycolysis products are not sufficient to maintain mitochondrial energy metabolism of electrically active axons . The combined monitoring of cellular ATP and electrical activity is a novel tool to study neuronal and glial energy metabolism in normal physiology and in models of neurodegenerative disorders . In the vertebrate nervous system , long axons have emerged as the 'bottle neck' of neuronal integrity ( Coleman and Freeman , 2010; Nave , 2010; Hill et al . , 2016 ) . In neurodegenerative diseases , axons often reveal first signs of perturbation , e . g . spheroids , long before the pathological hallmarks of disease are seen in neuronal somata , such as large protein aggregates or neuronal cell death . In white matter tracts , Wallerian degeneration of axons is a carefully regulated program of self-destruction that upon reaching a threshold leads to rapid calcium-dependent proteolysis of axonal proteins . This degeneration can be experimentally triggered in different ways , including acute axonal dissection or by blocking mitochondrial energy metabolism ( Coleman and Freeman , 2010; Hill et al . , 2016 ) . In disease situations , the mechanisms leading to axon loss are much less clear . For example , in spastic paraplegia ( SPG ) , which comprises an entire family of inherited axon degeneration disorders , long myelinated tracts of the spinal cord are progressively lost . However , virtually none of the subforms shows a clear link between the primary defect and the later loss of myelinated axons , with the possible exception of mitochondrial disorders , in which the perturbation of axonal energy metabolism is assumed to underlie a loss of axonal ATP ( Ferreirinha et al . , 2004; Tarrade et al . , 2006 ) . However , a disturbed energy balance has never been shown . Similarly , in the equally heterogeneous group of Charcot-Marie-Tooth ( CMT ) neuropathies of the peripheral nervous system only a small subset is caused by mitochondrial dysfunction . In most patients ( CMT1A ) the primary defect resides in the axon-associated myelinating glia ( Schwann cells ) , and it is unknown whether the axonal energy metabolism is perturbed before the onset of axon degeneration . Axonal ATP consumption in white matter tracts largely depends on the electric spiking activity that differs between regions , and it is virtually impossible to biochemically determine ATP selectively in the axonal compartment and under 'working conditions' with a high temporal and spatial resolution . To solve this problem , we have developed a novel approach to simultaneously study action potential propagation and ATP levels in axons of optic nerves of mice , a white matter tract suitable for studying axonal energy metabolism ( Stys et al . , 1991; Saab et al . , 2016 ) . We established the transgenic mouse line ThyAT , in which the fluorescent ATP-sensor ATeam1 . 03YEMK ( Imamura et al . , 2009 ) shows pan-neuronal expression in vivo . Stimulating electrically optic nerves acutely isolated from those mice , we combined recordings of compound action potentials ( CAP ) with confocal imaging of the ATP-sensor to evaluate and compare both axonal conductivity and ATP levels . We find that ( 1 ) axonal glycolysis is not sufficient to robustly sustain CAPs and physiological ATP levels , but mitochondrial function is needed to provide ATP; ( 2 ) during high-frequency propagation of action potentials , CAPs are a well-suited parameter to estimate axonal ATP levels; ( 3 ) lactate metabolism is essential for the maintenance of axonal ATP levels , and its inhibition severely affects energy homeostasis of myelinated axons . The genetically encoded ATP-sensor ATeam1 . 03YEMK allows FRET-based monitoring of cellular ATP levels close to real-time ( Imamura et al . , 2009 ) . We generated transgenic mice for pan-neuronal in vivo expression of this sensor driven by the murine Thy1 . 2 promoter ( Caroni , 1997 ) . One mouse line with 21 copies of the transgene [B6-Tg ( Thy1 . 2-ATeam1 . 03YEMK ) AJhi , referred to as ThyAT] displayed widespread neuronal expression ( Figure 1 ) . In the retina , fluorescence marked ganglion cells and their axons ( Figure 1G , H ) and expression of the ATP-sensor in myelinated axons appeared robust in all recorded channels ( Figure 1I–K ) . 10 . 7554/eLife . 24241 . 003Figure 1 . Characterization of the expression pattern of the newly generated B6-Tg ( Thy1 . 2-ATeam1 . 03YEMK ) AJhi ( ThyAT ) -mouse line . ( A ) Sagittal section of the brain highlights broad ATeam1 . 03YEMK expression in neurons in almost all brain regions with the exception of the olfactory bulb . Scale bar: 1 mm . ( B , C ) ThyAT expression pattern in coronal brain sections revealing sensor expression e . g . in thalamus , hypothalamus , amygdala , cortex and hippocampus . Scale bar: 1 mm . Abbreviations used in panels A–C are: AV: arbor vitae; BLA: basolateral amygdalar nucleus , anterior; CA1: CA1 region of the hippocampus; cc: corpus callosum; Cf: columns of the fornix; CN: cerebellar nuclei; Cp: cerebral peduncle; CPu: caudate putamen; Fi: fimbria; IC: inferior colliculus; ic: internal capsule; LH: lateral area of the hypothalamus; M: medulla; opt: optic tract; P: pons; S: subiculum; SC: superior colliculus; SMA: somato-motor area ( cortex ) ; SN: substantia nigra; st: stria terminalis; T: thalamus; Zi: zona incerta ( thalamus ) . ( D ) Within the cortex , neurons expressing ATeam1 . 03YEMK are clearly visible including their processes . Note the lack of ATP-sensor localization to the nucleus . Scale bar: 100 μm . ( E ) Also in the hippocampus neurons strongly express ATeam1 . 03YEMK . Scale bar: 100 μm . ( F ) In the cerebellum , Purkinje cells express the ATP-sensor . In addition , incoming mossy fibers strongly express ATeam1 . 03YEMK . Scale bar: 100 μm . Images in panels A , D–F are obtained on brain slices from a four month old animal , images in panels B and C are from mice at the age of two month . ( G ) Expression pattern of the ATP-sensor in the retina . Thy1 . 2 promoter drives the expression of ATeam1 . 03YEMK in ganglion cells . Scale bar: 1 mm . ( H ) Magnified view of neurons and axons in the retina expressing ATeam1 . 03YEMK . Scale bar: 100 μm . ( I–K ) Representative images of optic nerve axons showing the YFP channel ( I ) , FRET channel ( J ) and CFP channel ( K ) . The ATeam1 . 03YEMK expression is present in different axons independent of their diameter . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 003 Optic nerves from adult ThyAT mice were studied ex vivo , assessing axonal ATP levels by confocal microscopy and simultaneously monitoring stimulus-evoked CAPs ( Figure 2 ) . To verify sensor function , nerves were subjected to ATP depletion by blocking mitochondrial respiration with sodium azide ( MB ) and glucose deprivation ( GD ) . An immediate drop of the FRET signal ( Figure 2C , D ) and increase in CFP emission ( Figure 2C , D ) indicated that ATP levels in axons dropped . Changes in pH can modulate YFP-fluorescence ( Nagai et al . , 2004; Zhao et al . , 2011 ) . However , ATeam1 . 03YEMK is almost insensitive to pH within the physiological range ( Imamura et al . , 2009; Surin et al . , 2014 ) and YFP emission upon direct YFP excitation was unchanged ( Figure 2C , D; and data not shown ) , suggesting that pH changes are not the cause of altered FRET signals . 10 . 7554/eLife . 24241 . 004Figure 2 . Imaging of ATP combined with electrophysiology in acutely isolated optic nerves of ThyAT-mice . ( A ) Binding of ATP induces a conformational change in the genetically encoded ATP-sensor ATeam1 . 03YEMK thus increasing the FRET effect ( YFP emission upon CFP excitation ) and simultaneous decreased emission of CFP ( upon CFP excitation ) . The ratio between FRET and CFP can thus be correlated with the concentration of ATP present in the cell . ( B ) Schematic representation of the set-up to acquire evoked CAPs in the optic nerve and to simultaneously investigate relative ATP levels by electrophysiology and confocal imaging , respectively . ( C ) Time course of fluorescence intensity recorded in the YFP , FRET and CFP- channels during application of mitochondrial blockage ( MB ) and glucose deprivation ( GD ) for 2 . 5 min . Values are normalized to YFP intensity prior to application of MB+GD ( n = 3 nerves ) . Time resolution: 10 . 4 s . ( D ) The combination of MB and GD is a fast and reliable way to deplete ATP in axons of the optic nerve . ATP depletion is measured as a decrease in FRET and increase in CFP , calculated as ratio between fluorophore intensity during MB+GD , over fluorophore intensity at baseline condition ( MB+GD/Baseline ) . Notably , YFP emission upon YFP excitation remains unchanged ( n = 5 nerves ) . ( E ) Ratiometric images displaying the FRET/CFP ratio of the ATeam1 . 03YEMK–sensor in the axons of the optic nerve during ATP depletion following MB+GD . The phases before ( Baseline ) and after ( Recovery ) are also shown . Scale bar: 10 μm . ( F ) FRET/CFP ratio values ( not normalized ) during baseline and MB+GD . The boxplots show summarized data of n = 19 nerves , lines in between boxplots show changes in the FRET/CFP ratio of all 19 individual nerves ( **p<0 . 001 ) . ( G ) To assess ATP variations , the ratio of the fluorescence intensities of the FRET and CFP-channel was calculated ( FRET/CFP ratio ) and normalized to baseline ( set as 1 ) and MB+GD ( set as 0 ) . The red dashed line visualizes the slope of ATP drop at the point of maximal velocity of ATP decay during mitochondrial blockage ( MB+GD , n = 3 ) . ( H ) Recording of the evoked compound action potential ( CAP ) , given as the normalized curve integral during mitochondrial blockage and glucose deprivation ( MB+GD ) . The black dashed line represents the slope at the point of maximal velocity of CAP changes during MB+GD treatment ( n = 3 ) . Individual CAP traces are shown in Figure 3—figure supplement 1 . ( I ) Stripe plot describing CAP ( black ) and ATP ( red ) kinetics , expressed as maximal variation per s , during MB+GD ( p=0 . 39 , n = 3 , Welch’s t-test ) . Dots show individual data points , bars and lines represent the mean of all data . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 00410 . 7554/eLife . 24241 . 005Figure 2—source data 1 . Table containing data for Figure 2 . This xlsx-data file contains the data shown in Figure 2D , F and I . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 005 The ratio of FRET/CFP fluorescence ( F/C-ratio ) was calculated as a measure for the cytosolic ATP concentration ( Imamura et al . , 2009; Figure 2E , F ) . The baseline F/C-ratio ( control condition; 10 mM glucose ) was similar and stable between different nerves analyzed ( Figure 2F ) , while after 2 min of MB+GD the F/C-ratio was reduced by 35 . 1 ± 1 . 7% ( p<0 . 001; n = 19 nerves; Figure 2F ) . In further experiments , F/C-ratios were normalized between one ( 10 mM glucose ) and zero ( MB+GD; Figure 2G ) . As expected , during MB+GD the decrease in axonal ATP was mirrored by a decay of CAPs , reflecting conduction blocks ( Figure 2H ) . Strikingly , also the decline rates of axonal ATP and conductivity were similar ( Figure 2I ) . Next , we compared ATP and CAP dynamics in response to different modes of energy deprivation ( Figure 3; Figure 3—figure supplement 1 ) , including glucose deprivation ( GD ) , blockage of mitochondrial respiration by azide ( MB ) , and a combination of both ( MB+GD ) . During GD both ATP and CAPs decayed after 13 min ( Figure 3A ) . Inhibition of respiration resulted in a drop after about 3 min with a steeper decline ( Figure 3B ) . MB+GD induced an immediate and fast decline of ATP , followed by an equally fast decline in CAPs ( Figure 3C–E ) . To reduce permanent damage and to study recovery , MB and MB+GD were limited to 5 min . The observed change of the ATP sensor signal is most likely not caused by major changes in pH as YFP fluorescence remained almost unchanged under all conditions of energy deprivation ( Figure 2D and Figure 3—figure supplement 2 ) . These data suggest that optic nerves continue ATP production under GD , presumably by metabolizing glycogen of astrocytes ( Brown et al . , 2005 ) . However , axonal ATP production strongly depends on mitochondrial respiration , and axonal glycolysis alone is insufficient to maintain ATP levels , even in the virtual absence of electrical activity ( baseline recording conditions with 0 . 033 Hz stimulation frequency ) . 10 . 7554/eLife . 24241 . 006Figure 3 . Impairment of axonal ATP and CAP by glucose deprivation and/or inhibition of mitochondrial respiration . ( A ) Removal of glucose from the aCSF ( glucose deprivation , GD , 45 min ) induces similar ATP ( red ) and CAP ( black ) decays starting at around 13 min after onset of the treatment . Red and black dashed , straight lines represent the maximum velocity of ATP and CAP decay ( also applies to panels B and C ) . When 10 mM glucose is restored , CAP recovery precedes ATP restoration ( n = 4 nerves ) . ( B ) Blockade of mitochondrial respiration by azide ( MB , 5 min ) produces a fast decay in ATP ( red ) and CAP ( black ) starting at 1 . 8 min after beginning of treatment . When azide is removed , CAP and ATP are promptly restored , with CAP recovery preceding the ATP increase ( n = 4 nerves ) . ( C ) Simultaneous removal of glucose and blockade of mitochondrial respiration with azide ( MB+GD , 5 min ) produces a fast decay in ATP ( red ) preceding CAP decay ( black ) , starting already 0 . 5 min after onset of treatment . Following azide removal and replenishment of glucose , CAP and ATP are restored ( n = 4 nerves ) . ( D ) Time of onset of the ATP or CAP decay . The slowest decay induction was observed during glucose deprivation . ( E ) Velocity of signal decay for ATP and CAP during each of the three treatments: glucose deprivation ( GD ) , mitochondrial blockage ( MB ) and the combination of both ( MB+GD ) . ( F ) Time of onset of ATP or CAP recovery after reperfusion with control aCSF containing 10 mM glucose . ( G ) Rate of recovery of both ATP and CAP during reperfusion of the nerves with aCSF containing 10 mM glucose after the treatments indicated . ( H ) Comparison of ATP and CAP area overall recovery after individual treatments . Data in D–H is presented as stripe plots , with dots representing individual data points , bars and lines showing the mean . Hash signs indicate statistically significant differences between ATP and CAP under the same condition ( #p<0 . 05 , ##p<0 . 01 , paired t-test ) ; asterisks on red ( ATP ) and black ( CAP ) lines indicate statistically significant differences between different conditions ( p<0 . 05 , p<0 . 01 , p<0 . 001; one-way ANOVA with Newman-Keuls post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 00610 . 7554/eLife . 24241 . 007Figure 3—source data 1 . Table containing data for Figure 3 . This xlsx-data file contains the data shown in Figure 3D–H and Figure 3—Figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 00710 . 7554/eLife . 24241 . 008Figure 3—figure supplement 1 . Example of progression of CAP traces’ decay during energy deprivation . Shown are single traces obtained under control conditions ( baseline , dashed line ) as well as during application of GD ( A , total 45 min ) , MB ( B , total 5 min ) or MB+GD ( C , total 5 min ) . Single traces are separated by 330 s ( A ) or 30 s ( B , C ) . The shaded area indicates the area under the CAP wave form used for CAP quantification for the baseline condition , the same time window was used for analysis of CAPs at later time points . Under all conditions , no CAP is elicited by electrical stimulation anymore at the end of the treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 00810 . 7554/eLife . 24241 . 009Figure 3—figure supplement 2 . Analysis of fluorescence changes of the ATP sensor during application of different models of energy deprivation to optic nerves . Changes of the fluorescence signal relative to the baseline signal during glucose deprivation ( A; GD ) or mitochondrial blockage ( B; MB ) . Fluorescence intensities of the three recorded channels between 44 . 75 min and 45 min or 4 . 75 min and 5 min of incubation with GD and MB , respectively , were averaged ( GD: n = 4 nerves; MB: n = 4 nerves ) . Compare Figure 2C , D for data on GD+MB . In all cases , fluorescence in the CFP channel increased , while fluorescence in the FRET channel decreased . Of note , YFP emission upon direct YFP excitation remains stable . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 009 When nerves were reperfused ( aCSF , 10 mM glucose ) after GD , CAPs recovered with a delay of about 3 min , and the recovery of CAPs preceded the visible recovery of ATP levels by several minutes ( Figure 3A , F ) . Interestingly , following MB+GD , CAPs recovered significantly sooner , concomitantly with axonal ATP , and faster ( Figure 3F , G ) . We presume that , under MB+GD , lactate/pyruvate was not depleted from the optic nerve and is readily available for ATP generation by mitochondria after cessation of MB+GD . Maintaining axonal conductivity ex vivo depends on energy-rich substrates and high- frequency stimulation of axons causes CAPs to decrease ( Brown et al . , 2001; Tekkök et al . , 2005 ) . However , it has never been possible to study the reverse , i . e . the impact of electrical activity and spiking frequency on axonal ATP levels . In ThyAT optic nerves incubated in aCSF containing 10 mM glucose and subjected to electrical stimulation at 16 Hz , 50 Hz or 100 Hz for 2 . 5 min , CAPs dropped in a stimulation frequency dependent manner ( 16 Hz: 94 . 9 ± 1 . 5%; 50 Hz: 79 . 6 ± 0 . 4%; 100 Hz: 66 . 4 ± 1 . 0%; n = 5 nerves; Figure 4A , C , Figure 4—figure supplement 1 ) . At the same time , axonal ATP levels decreased as stimulation frequencies increased ( 16 Hz: 91 . 6 ± 1 . 0%; 50 Hz: 82 . 3 ± 1 . 4%; 100 Hz: 68 . 9 ± 2 . 8%; n = 5 nerves; Figure 4B , D ) indicating higher axonal ATP consumption . Even maximal ATP changes and maximal loss of conductivity correlated over stimulation frequencies ( Figure 4E , F ) . Similar results were obtained with a different stimulation paradigm with continuously increasing frequencies ( Figure 4—figure supplement 2 ) ; however , the causal relations of the correlation of CAP and ATP changes are most likely more complex ( see discussion below ) . Finally , both the rates of ATP drop and recovery correlated with the amplitude of ATP changes ( Figure 4—figure supplement 3 ) . 10 . 7554/eLife . 24241 . 010Figure 4 . Comparison of ATP and CAP dynamics during high frequency stimulation . ( A ) The CAP area decreases over time during high-frequency stimulation ( HFS ) . The decay amplitude deviates from the absence of HFS , indicated by the dashed line ( 0 . 1 Hz , used for normalization to 1 . 0 ) , and increases progressively with the increase in stimulation frequency ( 16 Hz , 50 Hz , 100 Hz ) . Traces from one representative nerve incubated in aCSF containing 10 mM glucose are shown . ( B ) Axonal ATP levels also decrease with increasing stimulation frequency , reaching a new steady state level which depends on the stimulation frequency . Same experiment as in panel A . ( C ) Remaining CAP area at the end of the HFS ( overall decay amplitude ) during incubation of nerves in different glucose concentrations quantified during the last 30 s of HFS . The stripe plot shows summarized data from n = 5 , 5 , or 4 nerves for 10 mM , 3 . 3 mM and 2 mM glucose , respectively . The dashed line at 1 shows CAP size at 0 . 1 Hz stimulation frequency , which was used for normalization . ( D ) Quantification of ATP decay amplitude during incubation of the same nerves as in ( C ) in different glucose concentrations . The dashed line at 1 shows ATP levels at 0 . 1 Hz stimulation frequency . ( E ) Correlation of the amplitude of ATP and CAP decay during HFS of nerves bathed in aCSF containing the glucose concentrations indicated . Data points are very close to the diagonal of the graph indicating that ATP and CAP change by similar factors . ( F ) Ratio of ATP and CAP drop during HFS in the presence of glucose in the concentrations indicated . If both parameters change by the same factor , this ratio remains equal to one . Data in ( C–D ) is presented as stripe plots , with dots representing individual data points and bars and lines showing the mean . Asterisks indicate statistically significant differences between glucose concentrations ( p<0 . 05 , **p<0 . 001; Welch’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01010 . 7554/eLife . 24241 . 011Figure 4—source data 1 . Table containing data for Figure 4 . This xlsx-data file contains the data shown in Figure 4C , D , F and Figure 4—Figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01110 . 7554/eLife . 24241 . 012Figure 4—figure supplement 1 . Example of progression of CAP traces’ decay during high frequency stimulation ( HFS ) . ( A ) The three peaks recognizable in the baseline trace ( dashed line ) are differently affected by increasing stimulation frequency and for CAP analysis only the first two are considered . Shown are single traces obtained prior to stimulation ( baseline ) as well as at the end of a 2 . 5 min stimulation period at different stimulation frequencies ( 16 Hz , 50 Hz , 100 Hz ) of a nerve incubated in aCSF containing 10 mM glucose . ( B ) Example of progression of CAP from baseline ( dashed line ) during stimulation of an optic nerve at 100 Hz incubated in aCSF containing 10 mM glucose for a total stimulation time of 2 . 5 min . Single traces are separated by 37 . 5 s . The shaded area indicates the area under the CAP wave form used for CAP quantification for the baseline condition ( green ) and after 150 s of HFS ( red ) . Grey shading results from the overlay of green and red shading . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01210 . 7554/eLife . 24241 . 013Figure 4—figure supplement 2 . Stimulation of optic nerves with progressively increasing frequencies . Frequency-dependent changes in relative signal amplitude of ATP and CAP , during progressively increasing stimulation frequencies ( 1 Hz to 100 Hz ) and following recovery . Nerves incubated in aCSF containing 10 mM glucose were stimulated for 45 s each with the indicated frequencies , directly followed by stimulation with the next higher frequency . The dashed line at 1 . 0 shows ATP and CAP values at 0 . 1 Hz stimulation frequency , which are used for respective normalization ( n = 4 nerves ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01310 . 7554/eLife . 24241 . 014Figure 4—figure supplement 3 . Correlation of the rates and amplitudes of CAP and ATP changes during HFS in different glucose concentrations . ( A ) Correlation of the velocity of the initial decay of CAP at the beginning of HFS and the amplitude of CAP decay at the end of HFS . The faster the CAP drops , the larger the CAP amplitude is . ( B ) Same analysis for ATP as for CAP area in panel A . Also a faster ATP consumption at the beginning of HFS coincides with a larger decrease in ATP signal amplitude . ( C ) Correlation of the rates of CAP area recovery after the cessation of stimulation and the amplitudes of CAP changes at the end of the stimulation at different glucose concentrations . The velocity of CAP recovery increases with larger amplitude of CAP decay during stimulation . ( D ) Same analysis as in C for ATP . ATP recovery rates are strongly depending on the amplitude of ATP decrease during HFS at 10 mM and 3 . 3 mM glucose , but much less in the presence of 2 mM glucose . The graphs summarize data from n = 5 , 5 , or 4 nerves for 10 mM , 3 . 3 mM and 2 mM glucose , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01410 . 7554/eLife . 24241 . 015Figure 4—figure supplement 4 . Example of CAP traces before and after high-frequency stimulation ( HFS ) of optic nerves incubated in aCSF with different concentrations of glucose . Shown are mean CAP wave forms ( n = 3 nerves for each condition ) incubated in aCSF containing 10 mM glucose ( A ) , 3 . 3 mM glucose ( B ) and 2 mM glucose ( C ) prior to high- frequency stimulation ( ‘baseline’; dashed lines ) or at the end of the 2 . 5 min HFS ( 100 Hz ) period ( solid lines ) . Grey areas indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01510 . 7554/eLife . 24241 . 016Figure 4—figure supplement 5 . Analysis of fluorescence changes of the ATP sensor during high-frequency stimulation ( HFS ) of optic nerves incubated in aCSF with different concentrations of glucose . Changes of the fluorescence signal at the end of the 2 . 5 min HFS ( 100 Hz ) period relative to the baseline signal prior to stimulation of optic nerves incubated in aCSF containing 10 mM glucose ( A ) , 3 . 3 mM glucose ( B ) and 2 mM glucose ( C ) . During HFS , fluorescence in the CFP channel increased , while fluorescence in the FRET channel decreased . Of note , YFP emission upon direct YFP excitation remains stable . n = 5 , 5 , 4 nerves for 10 mM , 3 . 3 mM and 2 mM glucose , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 016 Next , we asked whether lower ( physiological ) glucose concentrations have an impact on nerve conduction and axonal ATP levels when nerves were challenged with different spiking frequencies ( Figure 4C , D; Figure 4—figure supplement 4 ) . In presence of 10 mM and 3 . 3 mM glucose , CAP performance and ATP levels dropped similarly with increasing stimulation frequencies . However , when only 2 mM glucose was applied both CAP and ATP levels were substantially decreased ( Figure 4C , D ) and the ATP/CAP ratio was more variable at 2 mM glucose ( Figure 4F ) . Analysis of the fluorescence intensity of single channels indicated that the observed changes of the ATP sensor signal are indeed caused by ATP changes because YFP fluorescence remained rather unaffected ( Figure 4—figure supplement 5 ) . Taken together , these data suggest that 2 mM glucose is below the threshold required for maintaining the axonal energy balance of the optic nerve in the ex vivo experimental setting . At high-frequency stimulation and under aerobic conditions , optic nerve CAPs were maintained equally well with either 10 mM glucose , 10 mM lactate or 10 mM pyruvate as extracellular energy substrate ( Figure 5A ) , in agreement with earlier findings ( Brown et al . , 2001; Tekkök et al . , 2005 ) presumably because absolute axonal ATP consumption is unaffected by the type of metabolic support . However , at 100 Hz the relative decrease in axonal ATP levels was larger with lactate ( or pyruvate ) than with glucose ( glucose: 68 . 9 ± 2 . 8%; lactate: 56 . 3 ± 2 . 8%; pyruvate: 47 . 4 ± 2 . 8%; p=0 . 05 and p=0 . 01 , respectively; n = 3 nerves; Figure 5B , Figure 5—figure supplement 1 ) , showing that 10 mM glucose maintains axonal ATP production better than 10 mM lactate or pyruvate if applied exogenously . 10 . 7554/eLife . 24241 . 017Figure 5 . Energy metabolism of optic nerves depends on the type and concentration of substrates and involves lactate metabolism . ( A ) The comparison between CAP area decay of optic nerves incubated in 10 mM glucose aCSF ( n = 5 nerves ) versus optic nerves incubated either in 10 mM lactate or 10 mM pyruvate ( n = 3 nerves ) during HFS , shows no significant differences among the three substrates ( p>0 . 05 , Welch’s t test ) . ( B ) In contrast , analysis of axonal ATP levels shows that at higher frequencies glucose is a better substrate to maintain axonal ATP levels . Same experiments as in panel A . ( C ) In the presence of glucose ( 3 . 3 mM; n = 5 nerves ) as exogenous energy substrate , inhibition of lactate metabolism by D-lactate ( 20 mM; competitive inhibitor of endogenous L-lactate metabolism at MCTs and LDH , n = 6 ) or AR-C155858 ( 10 µM; MCT1 and MCT2 selective inhibitor , n = 6 ) does not significantly affect CAPs . ( D ) Analysis of ATP under the same conditions as in ( C ) : ATP levels undergo a strong decrease at higher frequencies in the presence of D-lactate or AR-C155858 . ( n = 5 nerves for all conditions ) . The dashed lines in panels A–D at 1 show CAP size or ATP levels at 0 . 1 Hz stimulation frequency used for normalization . ( E ) Inhibition of metabolism of endogenously produced L-lactate in the presence of glucose as the sole exogenous energy substrate shifts the correlation of ATP and CAP to the upper left showing that ATP changes more strongly than CAP . ( F ) The ratio of ATP and CAP drop decreases significantly in the presence of inhibitors of lactate metabolism , confirming that ATP changes more strongly than CAP . Asterisks in ( A–D and F ) indicate significant differences among conditions: p<0 . 05 , p<0 . 01 , p<0 . 001; Welch’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01710 . 7554/eLife . 24241 . 018Figure 5—source data 1 . Table containing data for Figure 5 . This xlsx-data file contains the data shown in Figure 5A–D , F and Figure 5—Figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01810 . 7554/eLife . 24241 . 019Figure 5—figure supplement 1 . Correlation of the amplitudes of CAP and ATP changes in the presence of lactate and pyruvate as exogenous energy substrates . ( A ) Correlation of ATP and CAP decay amplitude during HFS of nerves bathed in aCSF containing glucose , lactate or pyruvate ( each 10 mM ) as energy substrates . n = 5 , 3 , 3 nerves for glucose , lactate and pyruvate , respectively . ( B ) ATP to CAP amplitude ratio , calculated for nerves bathed in aCSF containing lactate ( 10 mM ) or pyruvate ( 10 mM ) as energy substrates during HFS . The ratio remains almost equal to one for all conditions supporting the notions that both ATP and CAP change by a similar factor . n = 5 , 3 , 3 for glucose , lactate and pyruvate , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 01910 . 7554/eLife . 24241 . 020Figure 5—figure supplement 2 . Example of CAP traces before and after high-frequency stimulation ( HFS ) of optic nerves in the presence of inhibitors of lactate metabolism . Optic nerves were incubated in aCSF containing 3 . 3 mM glucose in the absence of inhibitors ( A ) or in the presence of either 20 mM D-lactate ( B ) or 10 µM AR-C155858 ( C ) . Shown are mean CAP wave forms ( n = 3 nerves for each condition ) prior to high- frequency stimulation ( ‘baseline’; dashed lines ) or at the end of the 2 . 5 min HFS ( 100 Hz ) period ( solid lines ) . Grey areas indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 24241 . 020 Do axons require pyruvate/lactate metabolism also in the presence of glucose as energy substrate ? To address this question , we studied optic nerves conduction and ATP levels in 3 . 3 mM glucose plus 20 mM D-lactate , which competitively inhibits transport through MCTs as well as lactate dehydrogenase ( LDH; D-lactate is a stereoisomer of L-lactate , which is the physiological molecule in energy metabolism ) . D-lactate did not affect CAPs ( Figure 5C , Figure 5—figure supplement 2 ) but had a strong effect on axonal ATP ( at 100 Hz: Glc: 59 . 7 ± 2 . 9%; Glc+D-Lac: 29 . 2 ± 2 . 2%; p=0 . 0001; n = 5 nerves; Figure 5D ) . Similarly , when MCT1/MCT2-mediated lactate transport was inhibited using the specific inhibitor AR-C155858 ( Ovens et al . , 2010 ) , relative ATP levels were strongly reduced ( at 100 Hz: 24 . 8 ± 4 . 1%; p=0 . 0004; n = 5 nerves ) , but with no effect on CAP performance ( at 100 Hz: Glc: 69 . 5 ± 4 . 2%; Glc+AR-C155858: 62 . 8 ± 3 . 3%; p=0 . 37; n = 6 nerves; Figure 5C , D; Figure 5—figure supplement 2 ) , indicating the presence of a ‘safety window’ . When lactate metabolism was impaired , ATP/CAP ratios strongly deviated from the normal ‘1:1 ratio’ observed in the presence of glucose alone ( Figure 5E , F ) . Collectively , these results strengthen the conclusion that pyruvate/lactate metabolism is required to maintain ATP levels in fast spiking axons . Targeting the energy producing biochemical pathways by the withdrawal of glucose ( GD ) , blocking mitochondrial respiration ( MB ) , or both ( MB+GD ) , rapidly reduced axonal ATP content , as expected . In GD , ATP fell after 13 min , consistent with the utilization of astroglial glycogen as a reserve fuel ( Brown and Ransom , 2007 ) . In contrast , MB caused an immediate drop in ATP . In all conditions , ATP decreases preceded or coincided with CAP decay , confirming that energy depletion causes axonal failure ( but see caveat above ) . Interestingly , after GD in the subsequent recovery phase with glucose , the restoration of ATP levels appeared delayed compared to the recovery of conduction . We therefore suggest that at the beginning of reperfusion most newly generated ATP is consumed by Na+/K+-ATPases for reestablishing ion gradients and conduction before it accumulates in the axon and binds to the ATP-sensor . Astrocytes can mobilize glycogen and generate lactate to support axonal energy homeostasis ( Brown et al . , 2001 , 2003 , 2005 ) . However , it has been a long-standing debate whether neuronal compartments prefer glucose or lactate as energy-rich substrates ( Pellerin and Magistretti , 2012; Dienel , 2012; Nave , 2010; Hirrlinger and Nave , 2014 ) . In the optic nerve , glucose and pyruvate/lactate can sustain CAP propagation . We found that axonal ATP levels were significantly lower during HFS when nerves were incubated with lactate or pyruvate as the sole energy substrate . This suggests that lactate/pyruvate alone does not support ATP production as well as glucose and confirms observations in cultured cerebellar granule cells with repetitive activation of NMDA-receptors ( Lange et al . , 2015 ) . Nevertheless , lactate metabolism contributes to axonal energy homeostasis , even in the presence of glucose , as evidenced by blocking lactate metabolism with D-lactate or AR-C155858 . In the presence of these inhibitors , ATP decreased more strongly than CAP ( Figure 5E , F ) , suggesting the presence of a ‘safety window’ . It also breaks the striking correlation of ATP and CAP observed under control conditions ( Figures 4E , F and 5E , F ) . One possible explanation is that changes in CAP are related to ATP consumption , while inhibition of lactate transport affects ATP production . Since the axonal stimulation protocol is unchanged , also the consumption of ATP should be similar in the presence of inhibitors of lactate metabolism . Therefore , the inhibition of lactate metabolism impairs ATP production and causes a lower steady-state level of ATP , but the consumption rates during HFS remain the same . Furthermore , this finding suggests that ‘endogenous’ lactate ( i . e . produced from glucose ) is more efficient than lactate provided in the medium . One possibility is that glucose enters the nerve more rapidly than lactate , because glucose transporter expression is optimized in comparison to MCT1 . Moreover , glucose and glycolysis intermediates are readily shuttled via gap junctional coupling between astrocytes and oligodendrocytes , with lactate being generated also close to the periaxonal space , whereas exogenous lactate has a more complex path of diffusion involving mostly extracellular space ( Hirrlinger and Nave , 2014 ) . We note that from an evolutionary point of view , such optimization of glucose utilization is of advantage for the brain , because the liver always aims at generating a constant blood glucose level , even under starvation conditions . Taken together , combining ATP imaging and electrophysiology , we demonstrate that metabolic imaging is a very sensitive analysis tool for real-time monitoring of axonal energy homeostasis and the underlying neuron-glia interactions in electrically active fiber tracts . Applied to models of diseases this technology will be able to detect subtle perturbations of axonal energy homeostasis and allow addressing the hypothesis that energy deficits are an early and most likely causal event for neurodegeneration . Animals were treated in accordance with the German Protection of Animals Act ( TSchG §4 Abs . 3 ) , with the guidelines for the welfare of experimental animals issued by the European Communities Council Directive 2010/63/EU as well as the regulation of the institutional ‘Tierschutzkommission’ and the local authorities ( T04/13 , T20/16; Landesdirektion Leipzig , LAVES Niedersachsen ) . Animals were bred in the animal facility of the Max-Planck-Institute for Experimental Medicine as well as the Medical Faculty of the University of Leipzig . Mice were housed in a 12 hr/12 hr light dark cycle with access to food and water ad libitum . The transgene construct was assembled in pTSC ( Hirrlinger et al . , 2005 ) containing Thy1 . 2 promoter sequences . The plasmid pDR-GW AT1 . 03YEMK ( Bermejo et al . , 2010 ) containing the open reading frame of the ATP-sensor ATeam1 . 03YEMK ( Imamura et al . , 2009 ) was obtained from Wolf Frommer ( via Addgene; plasmid #28004 ) . The open reading frame of ATeam1 . 03YEMK was subcloned into the XhoI restriction site of pTSC using PCR . The linearized transgene was injected into fertilized mouse oocytes of the C57BL/6J mouse strain . Transgenic founders were identified using PCR-based genotyping on genomic DNA isolated from tail tips ( primer 1: 5’-CGCTGAACTTGTGGCCGTTTACG-3’; primer 2: 5’-TCTGAGTGGCAAAGGACCTTAGG-3’ ) . The mouse line has been registered at the RRID Portal ( https://scicrunch . org/resources; RRID:MGI:5882597 ) . Mouse genomic DNA was isolated from tail biopsies of four C57BL/6J and six ThyAT-mice with Invisorb Spin Tissue Mini Kit according to the manufacturer’s instructions ( Stratec Biomedical , Birkenfeld , Germany ) . Short FAM-labeled hydrolysis probes ( UPL ) were used for qPCR reactions ( Roche Diagnostics GmbH , Mannheim , Germany ) . Primers and the UPL-probe were designed by ProbeFinder version 2 . 50 for Mouse ( Roche Diagnostics; Primer Thy1s: TGCCGGTGTGTTGAGCTA; Thy1as: TGGTCCTGTGTTCATTGCTG; UPL 60; amplicon 73 bp ) . The genomic sequence of Nrg1 was used to calibrate for the amount of DNA ( Primer Nrg1s: GGCTATAATGCTAACACAGTCCAA; Nrg1as: AGTGGATCGTAACAACACTGTCA; UPL 38; amplicon 61 bp ) as described ( Besser et al . , 2015 ) . 10 to 25 ng of genomic DNA was subjected to qPCR amplification to measure the amount of Thy1 . 2 on a Light Cycler 480 system ( Roche Diagnostics ) according to the manufacturer’s instructions . The copy number of the Thy1 . 2-ATeam1 . 03YEMK transgene was calculated using the ΔΔCt method compared to wild type mice carrying only the two endogenous alleles of the Thy1 gene ( Besser et al . , 2015 ) . Adult mice were transcardially perfused with 4% formaldehyde solution ( PFA , in phosphate buffered saline: 137 mM NaCl , 2 . 7 mM KCl , 8 mM Na2HPO4 , 1 . 5 mM KH2PO4 , pH 7 . 4 ) under deep anesthesia . The brain and the eyes were removed and post-fixed for 24 hr in the same fixative . 45 μm thick sections were cut on a vibratome ( Leica VT1000 S , Nussloch , Germany ) and slices were mounted directly after cutting with Vectashield embedding medium ( Vectashield HardSet Mounting Medium , Vector Laboratories , Burlingame , CA , USA ) . From the eyes , retinal whole mounts were prepared . For imaging of fixed brain slices and retina , confocal images were acquired on a LSM Olympus IX71 inverted microscope using an UPlanFL 10x/0 . 3 objective ( Olympus , Hamburg , Germany ) for overview images ( Figure 1A–C , G ) or an UApo/340 40x/1 . 35 oil objective ( Olympus ) for detailed images ( Figure 1D–F , H ) , respectively . Microscopic images were acquired and processed using Olympus Software Fluoview v5 . 0 . ATeam1 . 03YEMK sensor fluorescence was excited with a 488 nm argon laser and detected through a BA 510–540 nm emission filter ( AHF Analysentechnik AG , Tübingen , Germany ) . For all images , a Kalman filter of two was used for denoising and images were acquired with 1024 × 1024 resolution ( pixel size for overview images: 0 . 9 µm; pixel size for detail images: 0 . 35 µm ) . Z-stacks comprise 12–38 singles z-planes for overview images and 30–85 z-planes for detailed images , respectively , with distances between each z-plane of 2 µm ( overview ) , 0 . 5 µm ( Figure 1D–F ) and 0 . 35 µm ( Figure 1H ) . Single z-stacks were converted to maximum intensity projections ( by using Fiji macro ‘Flattening V2f . ijm’; generously provided by Jens Eilers; Jens-Karl . Eilers@medizin . uni-leipzig . de ) and for overview images maximum intensity projections of different positions ( 210 for Figure 1A; 154 for Figure 1B; 160 for Figure 1C; 49 for Figure 1G ) were stitched by using Fiji software and a Fiji stitching plugin ( Preibisch et al . , 2009 ) . For optic nerve experiments , mice were used in an age of 8 to 12 weeks . Optic nerves were excised from decapitated mice , placed into an interface perfusion chamber ( Harvard Apparatus , Holliston , MA ) and continuously superfused with artificial cerebrospinal fluid ( aCSF ) . The perfusion chamber was continuously aerated by a humidified gas mixture of 95% O2/5% CO2 and experiments were performed at 37°C . Custom-made suction electrodes back-filled with aCSF were used for stimulation and recording as described ( Stys et al . , 1991; Saab et al . , 2016 ) . The stimulating electrode , connected to a battery ( Stimulus Isolator 385; WPI , Berlin , Germany ) delivered a supramaximal stimulus of 0 . 75 mA to the nerve evoking compound action potentials ( CAP ) . The recording electrode was connected to an EPC9 amplifier ( Heka Elektronik , Lambrecht/Pfalz , Germany ) . The signal was amplified 500 times , filtered at 30 kHz , and acquired at 20 kHz or 100 kHz . Before recording , optic nerves were equilibrated for at least 30 min in the chamber . The CAP , elicited by the maximum stimulation of 0 . 75 mA , was recorded at baseline stimulation frequency at 0 . 1 Hz . During HFS a burst-stimulation was applied consisting of 100 stimuli at the given frequency ( 16 , 50 or 100 Hz ) , separated by 460 ms , during which the CAP was recorded; HFS overall duration was 150 s , independent of the frequency . Live imaging of optic nerves was performed using an up-right confocal laser scanning microscope ( Zeiss LSM 510 META/NLO , Zeiss , Oberkochen , Germany ) equipped with an Argon laser and a 63x objective ( Zeiss 63x IR-Achroplan 0 . 9 W ) . The objective was immersed into the aCSF superfusing the optic nerve . Theoretical optical sections of 1 . 7 μm over a total scanned area of 66 . 7 μm x 66 . 7 μm ( 512 × 512 pixels ) of the optic nerve were obtained every 10 . 4 s in three channels , referred as CFP ( excitation 458 nm; emission 470–500 nm ) , FRET ( Ex 458 nm; Em long pass 530 nm ) and YFP ( Ex 514 nm; Em long pass 530 nm ) . Optic nerves were superfused by aCSF containing ( in mM ) : 124 NaCl , 3 KCl , 2 CaCl2 , 2 MgSO4 , 1 . 25 NaH2PO4 , and 23 NaHCO3 , which was continuously bubbled with carbogen ( 95% O2/5% CO2 ) . This solution was substituted with the appropriate energy substrates as indicated for which the solution containing 10 mM glucose served as the standard solution in respect to pH and osmolarity . For glucose deprivation ( GD ) , glucose was removed from the aCSF and substitute by sucrose ( Merck Millipore , Darmstadt , Germany ) to maintain the correct osmolarity . For mitochondrial blockage ( MB ) , aCSF containing 10 mM glucose was supplemented with 5 mM sodium azide ( Merck Millipore ) . For the combination of MB+GD glucose was substituted by sucrose and 5 mM sodium azide was added . For HFS , aCSF was supplemented with different substrates . Glucose ( Fluka BioChemika , Munich Germany ) was used at three concentrations: 10 mM , 3 . 3 mM and 2 mM . L-lactate and pyruvate ( Sigma-Adrich , Munich , Germany ) were used at 10 mM . Inhibitors of lactate metabolism were added to a basal concentration of glucose of 3 . 3 mM: the MCT1 inhibitor AR-C155858 ( Med Chem Express , Sollentuna , Sweden ) was used at 10 μM; sodium D-lactate ( Sigma-Aldrich ) was used at 20 mM . All aCSF-based solutions were adjusted for the same pH , sodium concentration and osmolarity . All data are presented either as mean ± s . e . m . or as stripe plots showing all data points and the mean ( line within the plot ) . The number of optic nerves analyzed for each condition is given as n . As for no condition both optic nerves of one animal were used , the number of nerves is equal to the number of animals analyzed for each condition . If not indicated otherwise in the figure legends , data were statistically evaluated using Welch’s t-test and assuming a normal distribution ( p<0 . 05; p<0 . 01; *p<0 . 001 ) .	The brain contains an intricate network of nerve cells that receive , process , send and store information . This information travels as electrical impulses along a long , thin part of each nerve cell known as the nerve fiber or axon . The act of sending these electrical signals requires a lot of energy , and energy in cells is most often stored within molecules of adenosine triphosphate ( called ATP for short ) . Importantly , a better understanding of how the production and consumption of ATP in nerve cells relates to electrical activity would help scientists to better understand how a shortage of energy in the brain contributes to diseases like multiple sclerosis . However , to date , it has been challenging to study the dynamics of ATP in nerve cells that are active . Now , Trevisiol et al . describe a new system that allows changes in ATP levels to be seen within active nerve cells . First , mice were genetically engineered to produce a molecule that works like an ATP sensor only in their nerve cells . This made it possible to visualize the amount of ATP inside the axons in real-time using a microscope . Measuring ATP levels and recording the electrical signals moving along an axon at the same time allowed Trevisiol et al . to see how ATP content and electrical activity correlate and regulate each other . The experiments reveal that strong electrical activity reduces the ATP content of the axon . Trevisiol et al . also discovered that nerve cells are unable to generate enough energy on their own to sustain their electrical activity . These results provide evidence that other cells in the brain – most likely non-nerve cells called oligodendrocytes – play an active role in delivering energy-rich substances to the axons of nerve cells . In the future , the same tools and approaches could be used to monitor ATP levels and electrical activity in mice that model neurological disorders . Such experiments could tell scientists more about how disturbing energy production in nerve cells affects these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Monitoring ATP dynamics in electrically active white matter tracts
Over 170 different mutations in the gene encoding SOD1 all cause amyotrophic lateral sclerosis ( ALS ) . Available studies have been primarily focused on the mechanisms underlying mutant SOD1 cytotoxicity . How cells defend against the cytotoxicity remains largely unknown . Here , we show that misfolding of ALS-linked SOD1 mutants and wild-type ( wt ) SOD1 exposes a normally buried nuclear export signal ( NES ) -like sequence . The nuclear export carrier protein CRM1 recognizes this NES-like sequence and exports misfolded SOD1 to the cytoplasm . Antibodies against the NES-like sequence recognize misfolded SOD1 , but not native wt SOD1 both in vitro and in vivo . Disruption of the NES consensus sequence relocalizes mutant SOD1 to the nucleus , resulting in higher toxicity in cells , and severer impairments in locomotion , egg-laying , and survival in Caenorhabditis elegans . Our data suggest that SOD1 mutants are removed from the nucleus by CRM1 as a defense mechanism against proteotoxicity of misfolded SOD1 in the nucleus . Amyotrophic lateral sclerosis ( ALS ) is a fatal neurodegenerative disease caused by progressive loss of upper and lower motor neurons in the brain and spinal cord . Most patients die from respiratory failure in 3–5 years after disease onset ( Ajroud-Driss and Siddique , 2015 ) . About 10% of ALS cases have positive family history and are classified as familial ALS ( fALS ) . The remaining 90% of ALS cases are classified as sporadic ALS ( sALS ) . Mutations in over 40 genes have been associated with fALS ( Ajroud-Driss and Siddique , 2015 ) , including the first identified gene encoding Cu/Zn superoxide dismutase ( SOD1 ) ( Deng et al . , 1993; Rosen et al . , 1993 ) . To date , over 170 mutations involving 88 of the 154 amino acids of SOD1 have been identified in patients diagnosed with ALS ( Taylor et al . , 2016 ) ( alsod . iop . kcl . uk ) . Mutations in SOD1 account for about 20% of familial and 2–7% of sporadic ALS cases ( alsod . iop . kcl . uk ) , which makes SOD1 the second most frequently mutated gene after C9ORF72 in affected Caucasians ( Ajroud-Driss and Siddique , 2015; Renton et al . , 2014 ) . SOD1 is a highly conserved and ubiquitously expressed free-radical scavenging enzyme . It catalyzes the dismutation of superoxide radical ( O2 . − ) into oxygen and hydrogen peroxide ( H2O2 ) , which in turn is reduced to water and oxygen by catalase . SOD1 is highly abundant , comprising 1–2% of the total soluble protein in the nervous system ( Pardo et al . , 1995 ) . Despite its important physiological function , SOD1 knockout mice failed to develop an ALS-like phenotype ( Reaume et al . , 1996 ) . Increasing evidence indicates that a toxic gain-of-function in SOD1 mutants is a fundamental mechanism for the pathogenesis of ALS ( Williamson et al . , 2000; Deng et al . , 2006; Sau et al . , 2007; Rotunno and Bosco , 2013; Bunton-Stasyshyn et al . , 2015; Silverman et al . , 2016 ) . Although the exact mechanisms by which diverse SOD1 mutations all cause ALS remains unclear , many toxic effects caused by SOD1 mutants have been reported , including excitotoxicity , oxidative stress , endoplasmic reticulum ( ER ) stress , mitochondrial dysfunction , axonal transport disruption , prion-like propagation , and non-cell autonomous toxicity of neuroglia ( Hayashi et al . , 2016 ) . These toxic effects may contribute to mutant SOD1-induced neuronal degeneration . Aggregation of SOD1 mutants in motor neurons is a common pathological feature . Therefore , coaggregation of an unidentified essential component or components or aberrant catalysis by misfolded mutants may underlie part of the mutant-mediated toxicity ( Bruijn et al . , 1998 ) . Disease pathogenesis and progression is not only determined by the toxic effects of SOD1 mutants but also modulated by intrinsic cellular defenses against the toxic proteins , such as protein quality control . Indeed , it has been reported that removal of SOD1 mutants through degradation by the ubiquitin-proteasome system and autophagy , or through induction of heat shock proteins reduces their cytotoxicity ( Kabuta et al . , 2006; Tan et al . , 2008; Ying et al . , 2009; Crippa et al . , 2010; Kalmar et al . , 2014 ) . However , available studies have been primarily focused on how mutant SOD1 exerts its toxicity , and much less is known about how cells defend against the toxic effects . In this study , we identified a novel mechanism that reduces mutant SOD1 cytotoxicity through limiting proteotoxicity in the nucleus . SOD1 is normally localized in both the cytoplasm and the nucleus of cells ( Crapo et al . , 1992 ) . We found that when GFP-tagged wt-SOD1 ( GFP-SOD1wt ) and two mutants ( GFP-SOD1G85R and GFP-SOD1G93A ) were expressed in HeLa cells , both mutants showed predominantly cytoplasmic distribution while GFP-SOD1wt was evenly distributed in the cytoplasm and the nucleus as expected ( Figure 1A ) . Impairment of the CRM1-dependent nuclear export pathway with a CRM1 inhibitor , leptomycin B ( LMB ) , caused the mutants to accumulate in the nucleus ( Figure 1A ) . Similarly , knockdown of CRM1 expression by RNAi or co-expression of a Q69L mutant of Ran GTPase , a dominant-negative inhibitor of CRM1 ( Xu et al . , 2010; Kehlenbach et al . , 1999 ) , also led to accumulation of the mutants in the nucleus ( Figure 1B , C ) . To rule out the contribution of proteasomal degradation to the nuclear clearance of SOD1 mutants ( Kabuta et al . , 2006; Niwa et al . , 2002 ) , we inhibited the proteasome activity with a proteasome inhibitor MG132 along with inhibition of protein synthesis by cycloheximide ( CHX ) . Proteasome inhibition for 60 min did not significantly affect the nuclear levels of GFP-SOD1G85R but again LMB treatment did ( Figure 1D ) . These results suggest that the lack of nuclear distribution of SOD1 mutants is mediated by CRM1-dependent nuclear export . 10 . 7554/eLife . 23759 . 003Figure 1 . ALS-linked SOD1 mutants are exported from the nucleus by CRM1 . ( A ) Inhibition of CRM1-dependent nuclear export increases nuclear distribution of SOD1G85R and SOD1G93A . HeLa cells expressing GFP-tagged SOD1wt , SOD1G85R or SOD1G93A were treated with LMB ( 20 nM ) for 1 hr . ( B ) Knockdown of CRM1 increases nuclear distribution of GFP-SOD1G85R . ( C ) Overexpression of FLAG-tagged RanQ69L but not wt Ran increases nuclear distribution of GFP-SOD1G85R . ( D ) Time-lapse imaging of SOD1G85R . HeLa cells expressing GFP-SOD1G85R were treated with cycloheximide ( CHX , 100 nM ) , in combination with MG132 ( 30 μM ) or LMB ( 20 nM ) as indicated . Images were acquired at indicated time points . The nuclear ( N ) and total ( T ) GFP fluorescence was measured for each cell with ImageJ . The N/T ratio expressed as mean ± SEM was calculated from 30 cells for each group and plotted . The mean N/T ratios at start point ( 0 min ) were arbitrary set as 1 . Paired t-test , CHX vs CHX+LMB: p=0 . 008; CHX+LMB vs CHX+MG132: p=0 . 006; CHX vs CHX+MG132: p=0 . 494 . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 003 It is well known that CRM1 exports its physiological cargo proteins through the recognition of the leucine-rich nuclear export signal ( NES ) ( Kosugi et al . , 2008; Xu et al . , 2012 ) . Selective export of mutant but not wt SOD1 suggests that SOD1 does not have a physiological NES . We hypothesized that misfolding of SOD1 mutants reveals a normally buried hydrophobic peptide that contains a NES consensus sequence . To test this hypothesis , we analyzed SOD1 protein sequence and located two potential NES-like sequence-containing peptides ( Figure 2A ) . The peptides were each fused to mCherry and expressed in HeLa cells . Fusion with peptide P1 corresponding to residues 24–55 of SOD1 , but not with P2 resulted in the nuclear clearance of mCherry , which could be restored by LMB treatment ( Figure 2B ) . GST pull-down experiments showed that P1 and the NES from protein kinase inhibitor ( PKI ) ( Güttler et al . , 2010 ) as a positive control , but not P2 , interact with CRM1 protein ( Figure 2C ) . These results suggest that peptide P1 in human SOD1 contains a functional NES consensus sequence . 10 . 7554/eLife . 23759 . 004Figure 2 . A nuclear export signal ( NES ) -like sequence is essential for the nuclear export of SOD1 mutants . ( A ) Two putative regions containing NES consensus sequences in SOD1 ( P1 and P2 ) . ( B ) P1 or P2 fused with mCherry was expressed in HeLa cells and treated with LMB for 1 hr . ( C ) GST pull-down . Immobilized GST or GST-tagged PKI-NES ( NES ) , P1 or P2 was incubated with recombinant CRM1 and RanGTP . ( D ) Identification of the key hydrophobic residues required for nuclear export activity in P1 . Each of eight hydrophobic residues in P1 was mutated to Arginine in GFP-SOD1G93A . The mutants were expressed in HeLa cells . ( E ) Alignment of key residues in P1 with corresponding residues in human SOD1 orthologs and PKI-class NES consensus ( Güttler et al . , 2010 ) . Note that residue Leu42 in human ( Φ2 ) is not conserved in several other species , including mouse . ( F ) Mutation of Leu42 in GFP-tagged human SOD1G93A to mouse corresponding residue ( Gln ) abolishes nuclear export of the mutant SOD1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 004 Next , we asked if the NES activity of P1 sequence is responsible for the nuclear export of SOD1 mutants . Eight hydrophobic residues in P1 are among the candidates for the NES consensus residues , and therefore , each was mutated to Arginine ( R ) in GFP-SOD1G93A . Five of the resulting mutants , including I35R , L38R , L42R , F45R , and V47R , showed even distribution in the cytoplasm and the nucleus , suggesting these five hydrophobic residues are essential for G93A nuclear export ( Figure 2D ) . These essential residues are in good agreement with the PKI-class NES consensus Φ0-X2-Φ1-X3-Φ2-X2–3-Φ3-X-Φ4 ( Güttler et al . , 2010 ) ( Figure 2E ) , where Φ0-Φ4 are five key hydrophobic residues spaced by different number of amino acids ( denoted as Xn ) that are preferentially charged , polar or small . Importantly , sequence alignment revealed that one of the five essential hydrophobic residues , Leu42 ( Φ2 ) , is not conserved in orthologs for human SOD1 ( Figure 2E ) . Mutation of this residue to Glutamine ( Q ) , as is naturally present in the mouse ortholog , abolished the nuclear export ability of GFP-SOD1G93A ( Figure 2F ) . These results indicate that P1 is not an evolutionarily conserved physiological NES but may be exposed by misfolding of SOD1 mutants . To determine whether the NES-like sequence is normally buried , we analyzed the available structure of wt-SOD1 ( PDB: 2c9v ) . The NES-like consensus sequence ( aa35-47 ) forms part of strand β3 , the crossover loop , and strand β4 , and is almost completely buried in the structure ( Figure 3A ) . In fact , Leu42 is the only one of the five essential hydrophobic residues exposed on the protein surface ( Figure 3A ) , which may explain why natural substitution of Leu42 by Gln in mouse SOD1 does not affect its structure ( Figure 2E ) . Next , we assessed the accessibility of the NES-like sequence to CRM1 in SOD1 mutants using pull-down assays with recombinant GST-tagged full-length wt or mutant SOD1 . CRM1 was precipitated by all four ALS-linked mutants tested , including Q22L , G85R , G93A , and L144S , but not by wt-SOD1 ( Figure 3B ) . Disruption of the NES consensus sequence by creating an L38R mutation in G85R or G93A largely diminished their binding to CRM1 ( Figure 3B , Lane 7 vs 4 , Lane 8 vs 5 ) . Moreover , CRM1-NES-like sequence interaction was dependent on the presence of RanGTP . The interaction was diminished when RanGTP was omitted from the pull-down assay ( Figure 3C ) , indicating a receptor-cargo relationship for CRM1-NES-like sequence of SOD1 . The exposure of the NES-like sequence in cells was determined by native immunoprecipitation ( IP ) . To do this , we generated rabbit polyclonal antibodies against the NES-like sequence-containing peptide ( NLP ) . Under native condition , anti-NLP efficiently precipitated GFP-SOD1G93A and six other mutants but not GFP-SOD1wt from lysates prepared from HEK293T cells transiently expressing the respective protein ( Figure 3D , E ) . These results indicate that the NES-like sequence of SOD1 is normally buried in the wild-type protein but is exposed in the mutants possibly due to protein misfolding . 10 . 7554/eLife . 23759 . 005Figure 3 . The NES-like sequence is exposed only in SOD1 mutants but not in wt SOD1 . ( A ) Structural localization of the NES-like peptide in wt SOD1 . The X-ray crystallographic structure of wt SOD1 ( PDB: 2c9v , only chain A of the dimer was shown ) is analyzed in UCSF Chimera 1 . 8 . Key residues in the NES-consensus sequence are labeled in red , whereas other residues are labeled in blue . Note that only Leu42 is exposed in the surface . Upper panel: ribbon structure; lower panel: surface structure . ( B ) GST pull-down . Immobilized GST or GST-tagged protein as indicated was incubated with recombinant CRM1 and RanGTP . ( C ) RanGTP-dependent interaction between GST-SOD1G93A and CRM1 . Recombinant CRM1 was precipitated by immobilized GST , GST-SOD1wt or GST-SOD1G93A with or without RanGTP . ( D ) Native immunoprecipitation ( IP ) . GFP-SOD1wt and GFP-SOD1G93A were each expressed in HEK293T cells . The cell lysates prepared under native condition were immunoprecipitated with either the preimmune serum ( control ) or antibodies against a peptide containing the NES-like sequence of SOD1 ( α-NLP ) . ( E ) Native IP . HEK293T cells expressing different GFP-tagged SOD1 proteins were lysed under native condition and immunoprecipitated with α-NLP . WCL: whole cell lysates . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 005 The GST pull-down and IP results also suggest that the NES-like sequence is exposed in multiple ALS-linked SOD1 mutants . We then asked whether exposure of the NES-like sequence is a common feature for ALS-linked SOD1 mutants . We tested 17 representative ALS-linked SOD1 mutants , including those deficient in copper binding ( H46R , H48R , and H120L ) or zinc binding ( H80R and D83G ) and one that cannot form disulfide bond ( C57R ) ( Valentine et al . , 2005 ) . These mutants were tagged with GFP and expressed in HeLa cells . GFP-SOD1wt was expressed as a negative control . As described above , exposure of the NES-like sequence can result in predominantly cytoplasmic distribution ( nuclear < cytoplasmic: N<C ) of some SOD1 mutants ( Figure 1A ) . Therefore , we first evaluated the exposure of the NES-like sequence in these GFP-tagged mutants by directly observing GFP localization . We found that cells expressing GFP-tagged SOD1 mutants exhibited either a N<C distribution or a distribution pattern similar to GFP-SOD1wt ( wt-SOD1-like ) . While N<C localization was the predominant pattern found in over half of the mutants tested , this localization pattern was found only in a fraction of cells expressing the remaining mutants , including L38R ( 0% ) , H46R ( 3 . 9% ) , C57R ( 62 . 6% ) , H80R ( 53 . 7% ) , D90A ( 59 . 5% ) , H120L ( 55 . 5% ) , L126S ( 66 . 5% ) and L144S ( 59 . 8% ) ( Figure 4A and Figure 4—figure supplement 1 ) . Importantly , disruption of the NES-consensus sequence by adding an L38R mutation to eight tested mutants resulted in wt-SOD1-like GFP distribution in all transfected cells ( Figure 4—figure supplement 2 ) . We then examined the exposure of the NES-like sequence by anti-NLP immunofluorescence . Positive anti-NLP staining was found in 90–100% of cells for all mutants except for L126S ( 81% ) and L144S ( 77% ) ( Figure 4A , B and Figure 4—figure supplement 1 ) . Consistent with the notion that exposure of the NES-like sequence exports misfolded SOD1 to the cytoplasm , anti-NLP immunofluorescence exhibited predominantly cytoplasmic localization , even in cells exhibiting wt-SOD1-like GFP localization , such as those expressing H46R ( Figure 4A , B ) . The L38R mutant that has a disrupted NES-consensus sequence , as an exception , was stained both in the nucleus and the cytoplasm ( Figure 4B ) . These results suggest that all SOD1 mutants tested expose the NES-like sequence , but the propensities to expose it vary among the mutants . Native IP with anti-NLP antibodies further confirmed that mutants with less N<C GFP distribution ( H46R , D90A , L126S and L144S ) were also less efficiently precipitated by anti-NLP , compared to mutants showing predominant N<C GFP distribution ( A4V , G85R , G93A and G108V ) ( Figure 4C ) . Interestingly , available reports indicate that ALS patients with H46R , D90A , L126S or L144S mutation have slow disease progression , whereas disease progress faster in patients with A4V , G85R , G93A or G108V mutation ( Bali et al . , 2017; Byström et al . , 2010 ) . It is possible that SOD1 mutants can assume more than one conformations in cells , among which their tendencies to expose the NES-like sequence correlate with disease progression . 10 . 7554/eLife . 23759 . 006Figure 4 . The NES-like sequence is exposed in ALS-linked SOD1 mutants and misfolded wt-SOD1 . ( A ) HeLa cells expressing GFP-tagged wt-SOD1 or indicated mutants were stained with α-NLP . Percentages of cells exhibiting predominantly cytoplasmic GFP distribution ( N<C% ) and cells positively stained by α-NLP were plotted . n = 392 ( WT ) , n = 395 ( A4V ) , n = 340 ( Q22L ) , n = 431 ( L38R ) , n = 389 ( H46R ) , n = 411 ( H48R ) , n = 423 ( C57R ) , n = 376 ( P66A ) , n = 393 ( H80R ) , n = 402 ( D83G ) , n = 450 ( G85R ) , n = 435 ( D90A ) , n = 408 ( G93A ) , n = 495 ( G108V ) , n = 355 ( H120L ) , n = 367 ( L126S ) , n = 462 ( L126Z ) , n = 410 ( L144S ) . ( B ) Anti-NLP specifically stained SOD1 mutants but not wt-SOD1 as detected by immunofluorescence microscopy . Asterisks indicate SOD1-expressing cells not being stained by α-NLP . ( C ) Native IP . HEK293T cells expressing different GFP-tagged SOD1 proteins were lysed under native conditions and immunoprecipitated with α-NLP . Immunoprecipitates and whole cell lysates ( WCL ) were blotted with α-GFP antibody . Relative IP efficiency for each SOD1 protein was calculated as the ratio of the band density from IP sample over that from corresponding WCL sample and was plotted as fold-difference relative to wt-SOD1 . ( D ) Demetallation induces exposure of the NES-like sequence in GFP-SOD1wt . HeLa cells expressing GFP-SOD1wt were treated with TPEN ( 10 μM ) and then stained with α-NLP . n = 392 ( 0h ) , n = 404 ( 2h ) , n = 527 ( 4h ) , n = 505 ( 6h ) . ( E ) Control HeLa cells or HeLa cells transfected with SOD1 siRNA ( KD: SOD1 ) were treated with TPEN ( 4 μM ) for 20 hr . Then the cells were stained with α-NLP and α-SOD1 antibodies and imaged for immunofluorescence . ( F ) DTT induces exposure of the NES-like sequence in endogenous SOD1 . HeLa cells were treated with DTT ( 0 . 4 mM ) and LMB ( 5 nM ) as indicated for 20 hr . Then the cells were stained with α-NLP . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 00610 . 7554/eLife . 23759 . 007Figure 4—source data 1 . Data for Figure 4A , C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 00710 . 7554/eLife . 23759 . 008Figure 4—figure supplement 1 . Anti-NLP specifically recognizes various ALS-linked SOD1 mutants but not wt SOD1 . HeLa cells expressing GFP-SOD1wt or indicated mutants were stained with α-NLP . Asterisks indicate GFP-SOD1wt and mutant expressing cells negative for α-NLP staining . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 00810 . 7554/eLife . 23759 . 009Figure 4—figure supplement 2 . A L38R mutation restores the nuclear distribution of various ALS-linked SOD1 mutants . GFP-SOD1wt and indicated mutants were expressed in HeLa cells and imaged by fluorescence microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 00910 . 7554/eLife . 23759 . 010Figure 4—figure supplement 3 . Oxidative stress induces exposure of the NES-like sequence in endogenous SOD1 . Control HeLa cells or HeLa cells transfected with SOD1 siRNA ( KD: SOD1 ) were treated with NaAsO2 ( 10 μM ) for 20 hr . Then , the cells were stained with α-NLP and commercial α-SOD1 antibodies and examined by immunofluorescence microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 010 We found that about 2% of the cells expressing GFP-SOD1wt were positive for anti-NLP staining , suggesting that the NES-like sequence is exposed in misfolded wt SOD1 ( Figure 4A ) . In support for this observation , previous studies have shown that wt-SOD1 can become misfolded under certain stress conditions and the misfolded wt-SOD1 adopts a ‘toxic conformation’ that is similar to ALS-linked SOD1 mutants ( Rotunno and Bosco , 2013 ) . We , therefore , further investigated whether the NES-like sequence is exposed in misfolded wt-SOD1 . It was reported that zinc binding stabilizes wt-SOD1 , whereas chelation of zinc leads SOD1 to adopt a ‘mutant-like conformation’ ( Homma et al . , 2013 ) . We , therefore , treated HeLa cells expressing GFP-SOD1wt with the zinc chelator , N , N , N' , N'-tetrakis ( 2-pyridylmethyl ) ethylenediamine ( TPEN ) followed by anti-NLP staining . More cells lost nuclear distribution of GFP-SOD1wt following TPEN treatment , and these cells were specifically stained by anti-NLP ( Figure 4D ) . Anti-NLP also stained endogenous SOD1 in TPEN-treated cells , whereas a non-conformation-specific anti-SOD1 antibody ( 71G8 ) stained both control and TPEN-treated cells ( Figure 4E ) . Moreover , in TPEN-treated cells , both 71G8 and anti-NLP showed predominantly cytoplasmic staining ( Figure 4E ) , suggesting that misfolded endogenous SOD1 is also exported from the nucleus . The specificity of anti-NLP staining of endogenous SOD1 was demonstrated by siRNA-mediated knockdown of SOD1 expression . RNAi dramatically decreased the staining by both 71G8 and anti-NLP staining ( Figure 4E ) . Similarly , cells treated with sodium arsenite , an oxidative stress inducer , were also specifically stained by anti-NLP , and the staining was significantly reduced when SOD1 was knocked down ( Figure 4—figure supplement 3 ) . Treatment of HeLa cells with dithiothreitol ( DTT ) , a strong reducing agent that may break the disulfide bond in SOD1 , also resulted in predominantly cytoplasmic staining of anti-NLP ( Figure 4F ) . LMB treatment of DTT-treated cells increased anti-NLP staining in the nucleus ( Figure 4F ) , indicating that the reduced nuclear staining is due to CRM1-dependent nuclear export of misfolded SOD1 . Taken together , these results suggest that certain environmental stresses can induce wt-SOD1 to adopt ‘mutant-like’ conformation with exposed NES-like sequence . To determine the clinical relevance of this novel conformation , we performed anti-NLP immunostaining of spinal cord sections from human ALS patients . The specificity of the anti-NLP immunohistochemistry was first demonstrated by a lack of staining in spinal cord sections from non-transgenic mice ( Figure 5A ) . Conversely , the motor neurons and neurites in the spinal cord sections from SOD1L126Z transgenic mice were anti-NLP positive ( Figure 5B ) ( Deng et al . , 2011 ) . In addition , the staining was predominantly cytoplasmic in most anterior horn neurons , which is consistent with the distribution pattern we have shown in HeLa cells for GFP-SOD1L126Z ( Figure 4—figure supplement 1 ) . The spinal cord sections from three ALS patients harboring A4V or G85R mutation were also stained with anti-NLP . The anti-NLP staining appeared cytoplasmic , both defuse and as aggregates , in anterior horn neurons and neurites ( Figure 5C–E ) . 10 . 7554/eLife . 23759 . 011Figure 5 . Exposure of the NES-like sequence in spinal cord sections from transgenic mice and ALS cases . ( A ) . Spinal cord sections from non-transgenic mice were stained with α-NLP . Representative anterior horn neurons are indicated by arrows and also showed in the inset . Scale bar: 100 m . ( B ) Spinal cord sections from SOD1L126Z transgenic mice were stained with anti-NLP . Representative SOD1 aggregates in anterior horn neurons and neuritis are indicated by arrows and arrowheads , respectively . Inset shows a representative anterior horn neuron with diffused α-NLP staining predominantly in the cytoplasm . Scale bar: 300 m . ( C ) to ( E ) Autopsy spinal cord sections from two SOD1A4V ALS patients ( C and D ) and a SOD1G85R ALS patient ( E ) were stained with anti-NLP . Representative SOD1 aggregates in anterior horn neurons and neurites are indicated by arrows and arrowheads , respectively . Scale bar: 100 m . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 011 Misfolded SOD1 mutants are toxic to neurons . When present in the nuclei of neurons , SOD1 mutants are likely to cause proteotoxicity in the nucleus ( Shibata and Morimoto , 2014; Gallagher et al . , 2014 ) . Therefore , nuclear clearance may modulate toxicity of SOD1 mutants through reducing proteotoxicity in the nucleus . To test this possibility , we proposed to restore the nuclear localization of SOD1 mutants by disrupting the NES-like consensus sequence using G85R as an example . As shown earlier in Figures 2E and 3A , Leu42 , one of the NES consensus residue in human SOD1 , normally exists as Gln ( Q ) in mouse and is exposed on the SOD1 protein surface . Therefore , a L42Q substitution is expected to disrupt the NES consensus sequence without affecting SOD1 structure . We generated lentiviral plasmids encoding GFP-SOD1wt , GFP-SOD1L42Q , GFP-SOD1G85R and GFP-SOD1G85R/L42Q , and expressed the variants in NSC34 motor neuron-like cells . As expected , GFP-SOD1wt , GFP-SOD1L42Q , and GFP-SOD1G85R/L42Q were localized in both the nucleus and the cytoplasm , whereas GFP-SOD1G85R was primarily in the cytoplasm ( Figure 6A ) . Importantly , native IP showed that GFP-SOD1L42Q was not recognized by anti-NLP , whereas about same amounts of GFP-SOD1G85R/L42Q and GFP-SOD1G85R were precipitated ( Figure 6B ) , supporting that L42Q mutation has little or no effects on SOD1 folding . The cytotoxic effects of GFP-SOD1G85R , GFP-SOD1L42Q , and GFP-SOD1G85R/L42Q were examined by a WST-1 assay for cell viability , and GFP-SOD1wt-expressing cells were used as a control . As previously reported ( Kabashi et al . , 2012; Kitamura et al . , 2014 ) , proteostress was imposed through inhibition of proteasomal degradation . Viability was similar for cells expressing GFP-SOD1wt and GFP-SOD1L42Q but significantly decreased in cells expressing GFP-SOD1G85R ( Figure 6C ) . Cells expressing GFP-SOD1G85R/L42Q showed lower viability than those expressing GFP-SOD1G85R , suggesting that disruption of the NES by L42Q mutation increases the cytotoxicity of GFP-SOD1G85R ( Figure 6C ) . These results suggest that nuclear export of SOD1 mutants plays a defensive role against proteotoxicity in the nucleus . 10 . 7554/eLife . 23759 . 012Figure 6 . Disruption of the NES by L42Q mutation in SOD1G85R mutant results in higher cytotoxicity in NSC34 cells . NSC34 cells were infected with lentiviruses expressing GFP-tagged WT , L42Q , G85R or G85R/L42Q SOD1 proteins . ( A ) SOD1L42Q and GFP-SOD1G85R/L42Q have similar subcellular localizations as GFP-SOD1wt in NSC34 cells . ( B ) Native IP . NSC34 cells expressing different GFP-tagged SOD1 proteins were lysed under native condition and immunoprecipitated with α-NLP . Immunoprecipitates and whole cell lysates ( WCL ) were blotted with α-GFP antibody . Relative IP efficiency for each SOD1 protein was calculated as the ratio of the band density from IP sample over that from corresponding WCL sample and was plotted as fold-difference relative to wt-SOD1 . ( C ) Cell viability ( WST-1 assay ) . NSC34 cells expressing GFP-tagged SOD1 proteins as indicated were treated with MG132 ( 5 μM ) for 24 hr . Data represent means ± SEM , n = 4 . Unpaired t-test , G85R/L42Q vs G85R: p=0 . 005; WT vs L42Q: p=0 . 564 . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01210 . 7554/eLife . 23759 . 013Figure 6—source data 1 . Data for Figure 6B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 013 We further determined the effects of restoring SOD1G85R nuclear localization in C . elegans . A C . elegans model of ALS has been established previously by expression of SOD1G85R-YFP in neurons ( Wang et al . , 2009 ) . For comparison , we generated new transgenic C . elegans lines expressing SOD1L42Q-YFP or SOD1G85R/L42Q-YFP . Three independent C . elegans lines each for SOD1L42Q-YFP and SOD1G85R/L42Q-YFP were used in this study . As observed in cultured cells , SOD1wt-YFP and SOD1L42Q-YFP were localized in both the cytoplasm and the nucleus of the neurons throughout the lifespan of C . elegans ( Figure 7A , B ) . SOD1G85R-YFP was cleared from the nuclei of motor neurons in both L1 and adult C . elegans ( Figure 7A , B ) . In contrast , SOD1G85R/L42Q-YFP was localized in both the cytoplasm and the nuclei in motor neurons in L1 C . elegans ( Figure 7B ) . In adult C . elegans , the nuclear localization of SOD1G85R/L42Q-YFP was only observed in a fraction of the neurons and no visible aggregate was observed in the nucleus . ( Figure 7B ) . A possible explanation is that cytoplasmic aggregation of the mutant protein prevents its entry into the nucleus . Indeed , we found that in adult C . elegans , like SOD1G85R-YFP , SOD1G85R/L42Q-YFP formed large cytoplasmic aggregates ( Figure 7B ) . Immunoblotting revealed that these new lines expressed similar levels of the SOD1 transgenes as the previously generated SOD1wt-YFP and SOD1G85R-YFP lines , with the exception of the line L42Q-1 ( Figure 7C ) . Similar amount of SOD1G85R-YFP and SOD1G85R/L42Q-YFP , but not SOD1wt-YFP or SOD1L42Q-YFP were detected in the insoluble fractions ( Figure 7C , upper panel , lanes 2 , 6–8 ) , indicating the formation of aggregates for the variants in these C . elegans lines . To assess the effects of SOD1G85R-YFP and SOD1G85R/L42Q-YFP on locomotion , we measured the body bending rates for L4 larvae of all these lines . As controls , the lines expressing SOD1wt-YFP and SOD1L42Q-YFP showed significantly higher bending rates than SOD1G85R-YFP and SOD1G85R/L42Q-YFP lines ( Figure 7D ) . Importantly , all three SOD1G85R/L42Q-YFP lines showed significantly lower bending rates than the SOD1G85R-YFP line ( Figure 7D ) . Next , we compared the survival rates of the transgenic lines . SOD1L42Q-YFP and SOD1wt-YFP worms showed similar survival rates , whereas the SOD1G85R-YFP line had significantly decreased survival rate , in agreement with previous reports . Remarkably , about 80% of all worms expressing SOD1G85R/L42Q-YFP died by day 3 ( Figure 8A , B ) . A potential cause for the decreased survival of SOD1G85R/L42Q-YFP worms was identified , as these animals exhibited an egg-laying defect where eggs hatched inside mothers , a phenotype known as ‘bagging’ . About 70% of SOD1G85R/L42Q-YFP worms showed the bagging phenotype , with an average of 4 . 2 eggs hatched per mother . This was significantly worse than observed for SOD1G85R-YFP animals , where about 17% of worms showed a bagging phenotype , with an average of 2 . 4 eggs hatched per mother ( Figure 9A , B , C ) . Remarkably , less eggs were laid by SOD1G85R/L42Q-YFP animals in 48 hr compared to other lines ( Figure 9D ) . These results support increased toxicity of SOD1G85R/L42Q-YFP compared to SOD1G85R-YFP when expressed in neurons of C . elegans . 10 . 7554/eLife . 23759 . 014Figure 7 . Disruption of the NES by L42Q mutation in SOD1G85R mutant causes severe defects in locomotion in transgenic C . elegans . Transgenic worm lines neuronal specifically expressing SOD1wt-YFP ( WT ) , SOD1L42Q-YFP ( L42Q ) , SOD1G85R-YFP ( G85R ) or SOD1G85R/L42Q-YFP ( G85R/L42Q ) in neurons were generated previously or in this study . ( A ) , Expression of YFP-tagged SOD1 proteins in L1 of transgenic worm lines . Insets show higher magnification of motor neurons . The dot lines depict the nuclear profiles of the enlarged neurons ( also in B ) . ( B ) Microscopy of adult animals . DAPI ( pseudo colored red ) was used to stain the nucleus . ( C ) Immunoblotting . L1 larvae were lysed by sonication on ice . The soluble ( S ) and insoluble ( P ) fractions were processed for immunoblotting . ( D ) Body bending rates . L4 animals were transferred to a drop of M9 buffer and counted for their total body bends in 1 min . Data represent means ±SEM , n = 20 . Two-tailed unpaired t-test was used to calculate the p values . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01410 . 7554/eLife . 23759 . 015Figure 7—source data 1 . Data for Figure 7D . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01510 . 7554/eLife . 23759 . 016Figure 8 . Disruption of the NES by L42Q mutation in SOD1G85R mutant decreases survival of transgenic C . elegans . ( A ) Survival curves . Mid-L4 animals were picked and followed for their survival . The surviving worms were transferred to fresh plates every 2 days until all the worms died . n = 50 for each transgenic line . ( B ) Average survival days from ( A ) . Two-tailed unpaired t-test was used to calculate the p values . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01610 . 7554/eLife . 23759 . 017Figure 8—source data 1 . Data for Figure 8A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01710 . 7554/eLife . 23759 . 018Figure 9 . Disruption of the NES by L42Q mutation in SOD1G85R mutant causes severer egg-laying defect in transgenic C . elegans . ( A ) Bagging in transgenic worms . Insets show hatched eggs in bagged worms . ( B ) An example of bagged G85R/L42Q C . elegans . ( C ) Rates of bagging in transgenic worms . L4 animals were transferred to fresh plates , and bagged worms were counted 48 hr later . n = 46 ( WT ) , n = 45 ( L42Q ) , n = 41 ( G85R ) , n = 47 ( G85R/L42Q ) . ( D ) Average hatched eggs in bagged worms from ( C ) . n = 7 ( G85R ) , n = 33 ( G85R/L42Q ) . Data represent means±SEM . G85R/L42Q vs G85R: p=0 . 048 by two-tailed unpaired t-test . ( E ) Eggs laid in 48 hr . Each L4 animal was transferred to one well of 96-well dish containing S-complete medium with OP50 . 48 hr later , larvae ( hatched eggs ) and unhatched eggs were counted for each well . Data represent means ±SEM . n = 12 . G85R vs . G85R/L42Q: p=4 . 07E-05 by Two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 01810 . 7554/eLife . 23759 . 019Figure 9—source data 1 . Data for Figure 9C , D and E . DOI: http://dx . doi . org/10 . 7554/eLife . 23759 . 019 Taken together , results from the cell and the animal models indicate that accumulation of misfolded SOD1 in both the cytoplasm and nucleus is more toxic than accumulation only in the cytoplasm . Therefore , removal of the misfolded SOD1 from the nucleus , mediated by the exposed NES-like sequence and CRM1-dependent nuclear export , is potentially a defense mechanism against the proteotoxic effects of ALS-linked SOD1 mutants in the nucleus . In this study , we found that SOD1 misfolding caused by either mutation or chemical insults induces nuclear clearance of SOD1 protein . The underlying mechanism is the exposure of a normally buried NES-like sequence caused by the misfolding event . CRM1 can recognize the NES-like sequence and facilitates export of the misfolded SOD1 from the nucleus to the cytoplasm . Exposure of the NES-like sequence is a common conformational feature for all 17 representative ALS-linked SOD1 mutants tested , and also for misfolded wt-SOD1 . Moreover , this conformational feature is also present in spinal cord motor neurons of human ALS patients . Our data also suggest that the exposure of the NES-like sequence is a disease-modifying feature for SOD1 mutants . First , this conformation appears to correlate with ALS disease progression . Several mutants that show lower propensities to acquire the NES-like sequence-exposed conformation , including H46R , D90A , L126S and L144S , have been previously reported to associate with slow progression of ALS ( Bali et al . , 2017; Byström et al . , 2010 ) . Second , although the potential toxic effects of this exposed NES-like sequence itself were not evaluated by the present study , data from cells and C . elegans suggest that nuclear export of misfolded SOD1 decreases the general toxicity of G85R mutant , probably due to diminishing the proteotoxicity in the nucleus . Disruption of the NES-consensus in SOD1G85R significantly increases defects in locomotion and egg-laying in C . elegans models . It is likely that loss of muscle function as a result of more severe motor neuron degeneration is responsible for these severe phenotypes . Therefore , the ability of neurons to remove misfolded proteins from the nucleus after exposure of the NES-like sequence may play an important role in pathogenesis of SOD1-linked ALS . Nuclear export not only decreases the levels of SOD1 mutants in the nucleus , but also increases their levels in the cytoplasm . Loss of nuclear SOD1 may lead to loss of function in the nucleus , but it is unlikely a disease mechanism because SOD1 knockout does not lead to ALS phenotype and pathology ( Reaume et al . , 1996 ) . To the contrary , knockdown of SOD1 mutants by siRNA are currently being pursued as a therapeutic approach for ALS ( Ralph et al . , 2005; Thomsen et al . , 2014; van Zundert and Brown , 2017 ) . Nuclear export may facilitate the removal of misfolded SOD1 , which would produce beneficial effects . It is known that the nucleus relies on exclusively the proteasomal degradation pathway to remove misfolded and unwanted proteins , whereas the cytoplasm has both proteasomal and lysosome-dependent autophagy pathways . Thus , nuclear export may facilitate mutant SOD1 degradation by autophagy . On the other hand , nuclear export increases the cytoplasmic concentration of the abnormal SOD1 , which may promote toxic effects on the functions of cytoplasmic organelles , for example , ER and mitochondria . It is likely that NES-like sequence-mediated changes in subcellular localization of SOD1 mutants would have complex effects on neuronal functions and survival , but collectively , may reduce the toxicity of SOD1 mutants . Future careful interrogation of this complex issue is warranted . Our results also indicate that exposure of the NES-like sequence may be a common property of misfolded SOD1 . Although we did not test every previously reported ALS-linked SOD1 mutations , those examined span almost the full length of SOD1 and also included almost all representative mutants . Misfolded wt-SOD1 also shares this feature , which supports the controversial proposition that misfolding of wt-SOD1 is involved in pathogenesis of sporadic ALS ( Rotunno and Bosco , 2013; van Zundert and Brown , 2017; Ayers et al . , 2016; Grad et al . , 2014 ) . The exposure of NES-like sequence can be specifically detected with anti-NLP antibody , which may become a useful tool for research and diagnosis . Interestingly , several other antibodies that recognize regions covered or partially overlapping with the NES-like sequence have been reported , including D3H5 ( epitope: residues 24–55 ) ( Rotunno and Bosco , 2013; Gros-Louis et al . , 2010 ) , AJ10 ( epitope: residues 29–57 ) ( Sábado et al . , 2013 ) , and USOD ( epitope: residues 42–48 ) ( Kerman et al . , 2010 ) . All these antibodies were also reported to selectively recognize mutant SOD1 over wt-SOD1 , suggesting that the epitopes formed by the NES-like sequence are highly specific for misfolded SOD1 . Therefore , the NES-like sequence may serve as a biomarker for the detection of misfolded SOD1 in ALS patients . Moreover , this immunogenic region may be used for immunotherapy for SOD1-linked ALS as was reported for the SOD1 exposed dimer interface ( SEDI ) peptide , that is specifically exposed in monomeric SOD1 ( Rakhit et al . , 2007; Liu et al . , 2012 ) . Our findings also have far-reaching implications for nuclear protein quality control . Proper maintenance of nuclear protein homeostasis is important for preserving cell function ( Shibata and Morimoto , 2014; Gallagher et al . , 2014 ) . Accumulation of misfolded proteins in the nucleus can affect cell function and cause diseases . A number of neurodegenerative diseases are associated with aggregates and inclusions formed by misfolded proteins in the nucleus , such as polyglutamine ( polyQ ) repeat diseases , polyalanine ( polyA ) repeat diseases , the RNA-mediated diseases , neuronal intranuclear inclusion disease ( NIID ) , neuronal intermediate filament inclusion disease ( NIFID ) , multiple system atrophy ( MSA ) , neuroferritinopathy , and inclusion body myopathy with early onset Paget's disease and frontotemporal dementia ( IBMPFD ) ( Gallagher et al . , 2014; Woulfe , 2008 ) . In addition to the known protein quality control mechanisms , such as chaperones and ubiquitin-proteasome system ( Shibata and Morimoto , 2014; Gallagher et al . , 2014 ) , our findings suggest that exposure of a normally buried NES-like sequence leading to CRM1-dependent nuclear export of the misfolded SOD1 may represent a novel mechanism for maintaining nuclear proteostasis . Exposure of normally buried hydrophobic regions is a common feature of misfolded proteins . These hydrophobic regions are commonly recognized by chaperones and quality control E3 ubiquitin ligases , to initiate the refolding and ubiquitination of the misfolded proteins , respectively ( Rosenbaum et al . , 2011; Fredrickson et al . , 2013; Horwich , 2014 ) . Interestingly , proteomic analysis has revealed that NES consensus sequences are frequently observed in hydrophobic regions buried inside of proteins ( Xu et al . , 2012 ) . This suggests that many misfolded proteins may expose buried hydrophobic regions containing NES consensus sequences , which may result in being exported from the nucleus by CRM1 if not efficiently degraded by the nuclear proteasomes or refolded by chaperones . Nuclear export may be a mechanism responsible for cytosolic mislocalization of many disease-causing mutant proteins ( Hung and Link , 2011 ) . Notably , failure in nuclear import is also a cause for mislocalization of some disease proteins . For example , mutations in the NLS of FUS , which also cause ALS , leads to its mislocalization to the cytoplasm ( Dormann et al . , 2010; Gal et al . , 2011 ) . Mutations in another ALS-linked gene , TDP43 , also lead to exclusion of TDP43 protein from the nucleus of the neurons and the formation of cytoplasmic TDP43 aggregates , but the underlying mechanism is not clear ( Mackenzie et al . , 2007; Deng et al . , 2010 ) . Abnormal nucleocytoplasmic transport may represent a convergent pathogenic pathway for some neurodegenerative diseases . Most recent studies have suggested that the nuclear-cytoplasmic transport is disrupted by the hexanucleotide repeat expansion ( HRE ) GGGGCC ( G4C2 ) in C9orf72 , which causes ALS and frontotemporal dementia ( FTD ) ( Zhang et al . , 2015; Freibaum et al . , 2015; Jovičić et al . , 2015 ) . It is possible that the exposure of NES-like sequence in mutant and misfolded SOD1 may also affect the nucleocytoplasmic transport of other cargo , because CRM1-mediated export is saturable and SOD1 is a highly abundant protein in the cell . However , there is an important distinction that nuclear export is deleterious in C9orf72-caused ALS , but it is protective in SOD1-linked ALS . pAcGFP1-SOD1-WT , -A4V , -G93A were acquired from Addgene ( Stevens et al . , 2010 ) . All other mutants used were created by site-directed mutagenesis . To express HIV Rev1 NES , P1 or P2 peptide of SOD1 in a fusion with mCherry , DNA fragments encoding these peptides were individually inserted into Bgl II/ Sal I sites of pmCherry-C1 . To express recombinant GST-tagged SOD1 proteins or peptides , Bgl II/Sal I fragments encoding SOD1 wt or mutants , P1 or P2 peptide of SOD1 , or PKI NES peptide were individually inserted into BamH I/Sal I sites of pGEX-4T-3 . pET3a-CRM1 was kindly provided by Dr . Dirk Görlich ( Paraskeva et al . , 1999 ) . Nde I/BamH I fragment of pET3a-CRM1 encoding CRM1 was inserted into Nde I/BamH I sites of pET28a ( + ) , to create pET28a-CRM1 expressing N-terminal His-tagged CRM1 . pcDNA3-FLAG-Ran WT and Q69L were kindly provided by Dr . Mien-Chie Hung ( Giri et al . , 2005 ) . HEK293T , HeLa , and NSC34 cells were obtained from ATCC . These cells were not independently authenticated . Mycoplasma contamination is monitored frequently in our laboratory by cytoplasmic DAPI staining and by PCR . All cells were cultured in DMEM ( Dulbecco’s modified Eagle’s medium ) supplemented with 10% ( V/V ) fetal bovine serum , 50 units/ml penicillin , 50 µg/ml streptomycin and 2 mM Glutamine under 5% CO2 in a humidified incubator . Mouse monoclonal anti-CRM1 ( C-1 , sc-74454 ) antibody was purchased from Santa Cruz Biotechnology ( Dallas , TX ) . Mouse monoclonal anti-FLAG ( M2 , F3165 ) , anti-α-Tubulin ( B-5-1-2 ) and anti-β-Actin ( AC-74 ) antibodies were purchased from Sigma-Aldrich ( St Louis , MO ) . Mouse monoclonal anti-SOD1 ( 71G8 , #4266 ) antibody was purchased from Cell Signaling Technology ( Danvers , MA ) , respectively . HRP-conjugated anti-GFP ( MA5-15256-HRP ) was purchased from Thermo scientific ( Waltham , MA ) . To produce antibodies against the NES like sequence of SOD1 , a SOD1 peptide corresponding to residues 33–51 ( GSIKGLTEGLHGFHVHEFGC ) was synthesized by Peptide 2 . 0 Inc . ( Chantilly , VA ) . A Cysteine was added to the C-terminus for conjugation of the peptide to keyhole limpet hemocyanin ( KLH ) to improve the immunogenicity . Anti-NLP was purified from rabbit antiserum using the same peptide that was biotinylated at the C-terminus and bound to streptavidin magnetic beads ( Thermo scientific , Waltham , MA ) . plv-AcGFP1-SOD1 ( wt ) was acquired from Addgene . plv-AcGFP1-SOD1-L42Q , G85R and G85R/L42Q were created by site-directed mutagenesis . These backbone plasmids were each co-transfected with helper vectors pDelta8 . 7 and pVSVG in a ratio of 5:5:3 to HEK293T cells to produce lentiviral particle as previously described ( Zhong and Fang , 2012 ) . Negative control #1 , Crm1 and SOD1 siRNAs ( SOD266 ) were previously reported ( Zhong and Fang , 2012; Strunze et al . , 2005; Maxwell et al . , 2004 ) . siRNAs were transfected with Lipofectamine RNAiMAX ( Invitrogen , Carlsbad , CA ) . 48 hr after siRNA transfection , cells were transfected with different plasmids as indicated with Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) . His-tagged RanGTP was prepared as reported ( Zhong et al . , 2011 ) . His-tagged CRM1 was expressed from BL21 ( DE3 ) transformed with pET28a-CRM1 . The bacteria were cultured at 30°C and induced with 0 . 1 mM Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) for 1 hr . Then the bacteria were lysed by sonication in 50 mM Tris-HCl pH7 . 4 , 100 mM NaCl , 250 mM Sucrose . His-CRM1 was purified using Ni- NTA-agarose ( Qiagen , Germantown , MD ) following the user’s manual . GST and GST-SOD1 wt and mutants were expressed from JM109 transformed with pGEX-4T-3 or pGEX- SOD1 constructs , respectively . The bacteria were cultured at 37°C and induced with 0 . 1 mM IPTG for 1 hr and then lysed by sonication in 20 mM Tris-HCl pH7 . 4 , 50 mM NaCl , 0 . 5% Triton X-100 . GST proteins were immobilized on Glutathione-Sepharose 4B beads ( Amersham Biosciences , Pittsburgh , PA ) . Then pre-incubated CRM1 and RanGTP were added to the beads and incubated for 1 hr at room temperature in CRM1 pull-down buffer ( 20 mM Tris-HCl pH7 . 4 , 50 mM NaCl , 0 . 5 mM GTP , and 0 . 15% Nonidet P-40 ) . After washing three times , the beads were boiled in SDS sample buffer and processed for SDS-PAGE and immunoblotting . For microscopy , the cells were grown on chambered coverglass and stained with a cell permeable nuclear dye Hoechst 33342 . When required , the cells were fixed with 2% paraformaldehyde for 5 min at room temperature before microscopy . For immunofluorescence staining , the cells were fixed in 3 . 7% paraformaldehyde for 30 min and permeabilized in 0 . 25% Triton X-100 for 5 min . After blocking in 5% bovine serum albumin ( BSA ) for 30 min , the cells were incubated with primary antibodies as indicated for 1 hr and then labeled with Alexa Fluor 488 or 594 conjugated secondary antibodies for 1 hr . DAPI or Hoechst 33342 was used for nuclear staining . Fluorescent microscopy was performed using a Zeiss Axiovert 200M fluorescent microscope . HeLa cells stably expressing GFP-SOD1G85R were treated with cycloheximide ( CHX , 100 nM ) , in combination with MG132 ( 30 μM ) or LMB ( 20 nM ) as indicated . Live cell images were acquired under a 40x objective lens mounted on a Nikon Eclipse Ti fluorescence microscope equipped with Perfect Focus System , a high-sensitivity CCD camera ( QuantEM 512SC; Photometrics , Tucson , AZ ) , and environment control units . At least 30 cells for each group were selected for time-lapse imaging . Cells were lysed in cell lysis buffer ( 150 mM NaCl , 10 mM Tris-HCl pH7 . 4 , 1 mM EDTA , 1 mM EGTA , 0 . 2% Nonidet P-40 and protease inhibitor cocktail ) . Cell lysate was incubated with antibodies as indicated and protein A Sepharose beads ( Zymed ) for 2 hr at 4°C . The beads were washed three times with cell lysis buffer and processed for immunoblotting . Immunohistochemistry was performed as previously described ( Deng et al . , 2006 , Deng et al . , 2010 ) . Briefly , 6 μm sections were cut from formalin-fixed , paraffin-embedded brain and spinal cord from mouse model or ALS patients with SOD1 mutations . The sections were deparaffinized and rehydrated by passing the slides in serial solutions as described . After antigen retrieval and blocking , the sections were incubated with affinity-purified anti-NLP rabbit polyclonal antibodies for 1 hr at room temperature and then with a biotinylated secondary antibody for 30 min . Positive signals were developed by first incubating the slides with peroxidase-conjugated streptavidin ( BioGenex , San Ramon , CA ) and then with 3-amino-9-ethylcarbazole chromogen ( BioGenex , Fremont , CA ) . The slides were examined and photographed under light microscope . NSC34 cells expressing GFP-tagged SOD1 proteins were seeded to 96-well plates at 2 × 104 cells/well . The next day , cells were treated with MG132 ( 5 μM ) for 24 hr . The cell viability was determined using Cell Proliferation Reagent WST-1 ( Roche , South San Francisco , CA ) according to the manufacturer’s protocol . Stable transgenic N2 Bristol C . elegans lines expressing SOD1wt-YFP or SOD1G85R-YFP were previously reported ( Wang et al . , 2009 ) . Psnb1:L42Q-YFP and Psnb1:G85R/L42Q-YFP were created by site-directed mutagenesis to express SOD1L42Q-YFP and SOD1G85R/L42Q-YFP in C . elegans , respectively ( Wang et al . , 2009 ) . N2 Bristol strain of C . elegans was transformed by DNA injection and the extrachromosomal lines were further treated with gamma ray to generate integrated lines stably expressing SOD1L42Q-YFP or SOD1G85R/L42Q-YFP . Three independent integrants for each construct showing 100% transmission were used for further experiments . To analyze the expression levels and solubility of SOD1 proteins , L1 larvae were suspended in extraction buffer ( PBS , 1 mM EDTA , 1 mM EGTA , 1 mM TCEP ) with protease inhibitor cocktail ( Sigma , St . Louis , MO ) and lysed by sonication on ice . The large debris was removed by spinning at 1000 ×g for 5 min . Then the supernatant was centrifuged at 55 , 000 rpm ( 100 , 000×g ) for 15 min to separate into soluble ( supernatant ) and insoluble ( pellet ) fractions . The insoluble fractions were washed once with the extraction buffer and then solubilized in 4% SDS . Equal amount of both fractions were processed for immunoblotting . The body bending rates and survival curves of transgenic worm lines were analyzed as previously reported ( Wang et al . , 2009 ) . Briefly , L4 animals were transferred to a drop of M9 buffer and counted for their total body bends in 1 min . To analyze the survival curves of transgenic animals , mid-L4 animals were picked and the surviving worms were transferred to fresh plates every 2 days until all the worms died . For microscopy of the worms , L1 or adult worms were frozen in liquid nitrogen for 10 s to crack the cuticles and then fixed in 3 . 7% PFA for 30 min . After permeabilization in 0 . 1% Triton X-100 for 5 min , the worms were stained for nucleus with DAPI for 5 min . All statistical analyses were performed using GraphPad Prism 5 . 0 or Microsoft Excel software . Statistical significance was assessed using paired or unpaired two-tailed Student’s t-test , as indicated in the corresponding figure legends . For all analyses , p<0 . 05 was considered statistically significant . Reported sample sizes indicate the number of biological replicates , with data obtained from individual cells , organisms , or sample wells . All collected data were included in analyses ( no outlier removal ) . For experiments analyzing cytotoxicity , sample units ( n ) are the number of biological replicates . For all other experiments , n represents the number of cells or animals used in the analysis for each group . No statistical method was used to predetermine sample size . However , the sample sizes we used were similar to those reported in previous studies . No samples or animals were excluded from the analyses .	Amyotrophic lateral sclerosis ( ALS ) is a disease that leads to muscle weakness and paralysis . The symptoms become progressively worse over time to the point that patients die because they become unable to breathe . Over 170 different genetic mistakes ( or mutations ) in a gene that encodes a protein called SOD1 are known to cause ALS . These mutations cause the SOD1 protein to form different shapes that are toxic to nerve cells , leading to the gradual loss of the nerve cells that control movement . SOD1 is normally found in a compartment within nerve cells called the nucleus , which is where most of the cell’s genetic information is stored and managed . A nematode worm called Caenorhabditis elegans has often been used as a model to study the role of SOD1 in ALS because its nervous system shares many features in common with ours but is much smaller . Some evidence suggests that cells may be able to defend themselves against the harmful effects of abnormal SOD1 proteins . However , it is not clear how these defences might work . Zhong et al . examined variants of SOD1 proteins from human cells grown in a laboratory . The experiments show that some mutant SOD1 proteins fold in such a way that a small section of the protein that is normally buried within the protein’s structure is exposed on the surface . Mutant SOD1 proteins that expose this “peptide” are removed from the nucleus and are linked with faster progression of ALS in patients . Further experiments show that another protein called CRM1 can recognise this exposed peptide , leading to the removal of the mutant SOD1 proteins from the nucleus . Zhong et al . found that if mutant SOD1 is not removed from the nucleus of nerve cells , the nematode worms developed ALS symptoms even faster . These findings suggest that cells may be able to remove some mutant SOD1 proteins from the nucleus to defend themselves against the proteins’ toxic effects . Future work will reveal whether other cells use this approach to protect themselves against other diseases . The peptide discovered in this work may also have the potential to be used as a marker to predict how individual cases of ALS will progress , or as a target for treatments against the disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Nuclear export of misfolded SOD1 mediated by a normally buried NES-like sequence reduces proteotoxicity in the nucleus
Low-density lipoprotein receptor-related protein 1 ( LRP1 ) is a multifunctional cell surface receptor with diverse physiological roles , ranging from cellular uptake of lipoproteins and other cargo by endocytosis to sensor of the extracellular environment and integrator of a wide range of signaling mechanisms . As a chylomicron remnant receptor , LRP1 controls systemic lipid metabolism in concert with the LDL receptor in the liver , whereas in smooth muscle cells ( SMC ) LRP1 functions as a co-receptor for TGFβ and PDGFRβ in reverse cholesterol transport and the maintenance of vascular wall integrity . Here we used a knockin mouse model to uncover a novel atheroprotective role for LRP1 in macrophages where tyrosine phosphorylation of an NPxY motif in its intracellular domain initiates a signaling cascade along an LRP1/SHC1/PI3K/AKT/PPARγ/LXR axis to regulate and integrate cellular cholesterol homeostasis through the expression of the major cholesterol exporter ABCA1 with apoptotic cell removal and inflammatory responses . Atherosclerosis is a chronic condition characterized by impairments in three major processes: systemic and cellular cholesterol homeostasis , inflammation , and apoptosis/efferocytosis ( Libby et al . , 2011 ) . Essential roles for the multifunctional transmembrane protein LDL receptor-related-protein 1 ( LRP1 ) have been reported for each of these three processes ( Boucher et al . , 2003; Boucher et al . , 2002; El Asmar et al . , 2016; Mantuano et al . , 2016; Subramanian et al . , 2014; Zhou et al . , 2009a; Zurhove et al . , 2008 ) . Together with LDLR , LRP1 regulates the clearance of circulating cholesterol-rich remnant proteins by hepatocytes ( Rohlmann et al . , 1998 ) . LRP1 is also a key regulator of intracellular cholesterol accumulation in macrophages and smooth muscle cells ( SMCs ) ( Boucher et al . , 2003; Lillis et al . , 2015; Zhou et al . , 2009a ) . In SMCs , LRP1 functions as a co-receptor with platelet-derived growth factor receptor ( PDGFRβ ) , which in turn activates inflammatory and pro-thrombotic processes ( Boucher et al . , 2002; Loukinova et al . , 2002 ) . LRP1 itself has also been implicated in limiting cellular inflammatory responses ( Mantuano et al . , 2016; Zurhove et al . , 2008 ) , and the dual role of LRP1 in cholesterol homeostasis and inflammation ( El Asmar et al . , 2016; Woldt et al . , 2011 , 2012 ) was recently demonstrated in an adipocyte-specific LRP1-deficient mouse model ( Konaniah et al . , 2017 ) . Finally , LRP1 is necessary for efficient efferocytosis in macrophages and dendritic cells ( Subramanian et al . , 2014; Yancey et al . , 2010 ) . LRP1 is ubiquitously expressed , but how it affects cellular signaling mechanisms can differ substantially in a cell-type dependent manner . In SMCs , LRP1 regulates ERK1/2 activation , which leads to increased cPLA2 phosphorylation and release of arachidonic acid ( Graves et al . , 1996; Lin et al . , 1992 ) , a suppressor of LXR-driven ATP-binding cassette transporter A1 ( ABCA1 ) expression ( DeBose-Boyd et al . , 2001 ) , thereby increasing SMC intracellular cholesterol accumulation ( Zhou et al . , 2009a ) . ABCA1 , however , mediates not just reverse cholesterol export , it also regulates cellular inflammatory responses ( Ito et al . , 2015; Tang et al . , 2009; Zhu et al . , 2010 ) . This suggests the presence of an LRP1/LXR/ABCA1 axis that controls and integrates cellular cholesterol export and inflammatory responses . One of the atheroprotective roles of LRP1 in the vascular wall is the regulation of the well-established mitogenic pathway mediated by platelet derived growth factor ( PDGF ) , which promotes atherosclerosis ( Ross , 1993 ) . Mitogenic signaling is commonly associated with tyrosine phosphorylation , and stimulation of PDGFRβ with PDGF-BB induces tyrosine phosphorylation of LRP1 by Src-family tyrosine kinases at the distal cytoplasmic NPxY motif ( Barnes et al . , 2001; Boucher et al . , 2002; Loukinova et al . , 2002 ) . This phosphorylation event can either promote or inhibit the association of the LRP1-ICD ( intracellular domain ) with several intracellular adaptor proteins , which in turn modulates down-stream signaling cascades ( Gotthardt et al . , 2000 ) . However , the physiological significance of LRP1 NPxY phosphorylation has remained unclear . To investigate this role of LRP1 tyrosine phosphorylation in atherosclerosis , we developed a knock-in mouse model in which the tyrosine in the distal NPxY motif was replaced with phenylalanine ( Lrp1Y63F ) , thereby disabling Lrp1 NPxY phosphorylation while leaving the interaction with phosphotyrosine-independent PTB-domain containing adaptor proteins intact ( Gotthardt et al . , 2000; Trommsdorff et al . , 1998 ) . We found that LRP1 NPxY phosphorylation induces a molecular switch of associated cytoplasmic adaptor proteins away from endocytosis-promoting DAB2 to the well-known signal-transducer SHC1 . SHC1 is an evolutionarily conserved adaptor protein that mediates mitogenic signaling ( Dieckmann et al . , 2010; Pelicci et al . , 1992; van der Geer , 2002; van der Geer and Pawson , 1995; van der Geer et al . , 1995 ) and is required for LRP1-dependent signal transduction in SMC through activation of PI3K and phosphorylation of AKT ( Gu et al . , 2000; Radhakrishnan et al . , 2008 ) . Surprisingly , lack of LRP1 tyrosine phosphorylation was inconsequential in SMCs . By contrast , in macrophages , SHC1 binding to phosphorylated LRP1 activated PI3K/AKT and PPARγ/LXR driven ABCA1 expression , a major cellular cholesterol exporter , and one of many mechanisms that is impaired in atherogenesis . Consequently , disabling LRP1 tyrosine phosphorylation resulted in enhanced macrophage intracellular lipid accumulation and decreased clearance of apoptotic cells , thus resulting in accelerated atherosclerosis independent of plasma cholesterol levels . To investigate the role of LRP1 tyrosine phosphorylation in atherogenesis , we generated a knock-in mouse that was incapable of tyrosine phosphorylation at the distal cytosolic NPxY motif at cytoplasmic amino acid position 63 ( Lrp1Y63F ) and then backcrossed it into an Ldlr-deficient background ( Lrp1Y63F;Ldlr−/−; Figure 1A ) . Western blot analysis of liver and aortic extracts from Lrp1Y63F mice revealed no effect of the Y63F mutation on Lrp1 expression in the liver or the aorta ( Figure 1B ) . Pull-down experiments in SMC stimulated with PDGF-BB confirmed decreased phosphorylation of the Lrp1 cytoplasmic domain in Lrp1Y63F SMC when compared with wild type SMC ( Figure 1C ) . The LRP1 cytoplasmic domain interacts with numerous intracellular adaptor and scaffold proteins ( Gotthardt et al . , 2000; Herz and Strickland , 2001; Stockinger et al . , 2000 ) . Using co-immunoprecipitation , we observed differential binding of intracellular adaptor proteins Dab2 ( Disabled-2 ) and Shc1 ( SHC-transforming protein 1 ) to Lrp1 depending on the phosphorylation state of the NPxY motif . After stimulation of wild type SMC with PDGF-BB , which induces phosphorylation of the distal NPxY motif in the Lrp1 cytoplasmic domain , Shc1 was bound to Lrp1 , but Dab2 was not . In Lrp1Y63F SMC , in which phosphorylation of the distal NPxY motif is prevented , Shc1 was no longer able to bind to Lrp1 , while Dab2 could now be co-immunoprecipitated with Lrp1 ( Figure 1C ) . Thus , binding of Shc1 to Lrp1 is regulated by tyrosine phosphorylation of the distal NPxY motif of Lrp1 . LDLR and LRP1 have functions in systemic lipoprotein metabolism . We therefore analyzed the effect of high cholesterol/high fat ( HCHF ) diet feeding on Lrp1Y63F;Ldlr−/− mice . Body weight , plasma cholesterol and triglyceride concentrations of Lrp1Y63F;Ldlr−/− mice on chow diet were essentially identical to age and sex-matched Ldlr−/− controls ( Figure 2A ) . However , after 16 weeks on HCHF diet , Lrp1Y63F;Ldlr−/− mice had significantly increased body weight , plasma cholesterol ( 2035 ± 99 . 53 mg/dl vs . 1672 ± 45 . 98 mg/dl , p<0 . 01 ) and triglyceride levels ( 257 . 4 ± 26 . 2 mg/dl vs . 161 . 8 ± 13 . 55 mg/dl , p<0 . 01 ) compared to Ldlr−/− mice ( Figure 2A ) . Western blot analysis revealed increased ApoB48 and ApoE in Lrp1Y63F;Ldlr−/− mice , whereas ApoB100 and ApoAI were decreased compared to Ldlr−/− mice after HCHF diet feeding ( Figure 2B ) . FPLC lipoprotein analysis demonstrated that elevated plasma cholesterol levels in Lrp1Y63F;Ldlr−/− mice fed with HCHF diet were attributed to increased VLDL-cholesterol with slightly reduced LDL- and HDL-cholesterol contents ( Figure 2C ) . VLDL-triglyceride levels were also increased in Lrp1Y63F;Ldlr−/− mice ( Figure 2C ) . Consistent with the results of total plasma apolipoprotein analysis , size-fractionated ApoB100 and ApoAI were significantly decreased , but ApoE was increased in Lrp1Y63F;Ldlr−/− mice ( Figure 2D ) . Interestingly , ApoB48 remnant particles were shifted towards a larger size in Lrp1Y63F;Ldlr−/− mice compared to Ldlr−/− mice on HCHF diet , possibly suggesting impaired clearance through SR-B1 ( Rigotti et al . , 1997 ) . We conclude that the Lrp1Y63F mutation not only promotes the accumulation of ApoB48 remnants , as also seen in the hepatic Lrp1 knockout ( Rohlmann et al . , 1998; Rohlmann et al . , 1996 ) , but also causes decreases in HDL and ApoA1 , which could independently increase atherosclerosis reminiscent of the increased risk of premature atherosclerosis in individuals with Tangier’s disease ( Tall and Wang , 2000 ) . The unexpected , though modest , change in plasma ApoA1 in addition to the accumulation of very large lipoprotein remnants led us to investigate atherosclerosis susceptibility in Lrp1Y63F;Ldlr−/− mice . Compared to Ldlr−/− mice , Lrp1Y63F;Ldlr−/− mice showed a 1 . 7-fold and 2 . 7-fold increase of atherosclerotic lesion area throughout the entire aorta and in the aortic root , respectively ( Figure 3A and B ) . Our previous studies demonstrated that smooth muscle Lrp1 deficiency leads to increased SMC proliferation and migration and retention of cholesterol , leading to profound aortic atherosclerosis . Unlike the SMC-targeted Lrp1 knockout mouse model ( smLrp1−/− ) , which displays prominent aortic wall thickening and increased numbers of SMC even under normolipidemic conditions , morphometric analysis of the thoracic aorta from Lrp1Y63F;Ldlr−/− and Ldlr−/− mice revealed no difference in aortic wall thickness or numbers of SMC ( Figure 4A ) . Furthermore , analysis of whole aorta homogenate ( Figure 4B ) and explanted primary aortic SMC ( Figure 4C ) from the Lrp1Y63F;Ldlr−/− and Ldlr−/− mice showed no significant differences in the activation of known PDGF down-stream signal effectors . Finally , to test the functional properties of the explanted SMC , we used Transwell Migration and BrdU Proliferation assays to compare their migratory and proliferative response to PDGF-BB . No difference in migration or proliferation was observed between wild type and Lrp1Y63F SMC ( Figure 4D ) . We therefore concluded that the Lrp1Y63F mutation does not increase atherosclerosis propensity through a change of SMC phenotype . Aside from SMCs , macrophages are central to atherogenesis . Semi-quantitative , cross-sectional analysis of the composition of atherosclerotic lesions in the aortic sinus revealed a 57% increase in Mac-3–positive macrophages , but a comparable amount of α-actin–positive cells ( corresponding to SMC ) in Lrp1Y63F;Ldlr−/− mice , consistent with a pronounced infiltration of macrophages into the atherosclerotic plaque ( Figure 5A ) . Further semi-quantitative analysis of H&E staining showed a 40% increase in necrotic core area in Lrp1Y63F;Ldlr−/− mice ( Figure 5B ) . The increase in apoptosis was confirmed by TUNEL staining of atherosclerotic plaques in vivo ( Figure 5C ) and oxLDL loaded peritoneal macrophages in vitro ( Figure 5D ) . Deficiency in MER proto-oncogene tyrosine kinase ( MerTK ) is associated with increased macrophage apoptosis and decreased efferocytosis , leading to increased necrotic cores in Apoe−/− mice ( Li et al . , 2006; Scott et al . , 2001; Thorp et al . , 2008 ) . We isolated peritoneal macrophages from Lrp1Y63F;Ldlr−/− and control mice and found a significant decrease in MerTK expression both at baseline and in response to treatment with oxLDL ( Figure 5E ) . Therefore , increased necrotic core size and apoptosis in the Lrp1Y63F;Ldlr−/− mice may be attributed at least in part to MerTK dysfunction associated with the macrophage Lrp1 mutation , in addition to other efferocytosis-related functions that are controlled by LRP1 ( Subramanian et al . , 2014; Yancey et al . , 2010 ) . Based on these results , we suspected that the striking increase in atherosclerosis in Lrp1Y63F;Ldlr−/− mice is due to macrophages and not SMCs . Because of the observed change of the lipoprotein profile in the Lrp1Y63F;Ldlr−/− mice , a trivial explanation for the increased atherosclerosis would be impaired chylomicron clearance and decreased HDL . To exclude the possibility that the atherosclerotic phenotype was secondary to changes in lipid metabolism and to explore the degree to which atherosclerosis was driven by Lrp1 tyrosine phosphorylation in macrophages , we performed bone marrow transplants where Ldlr−/− or Lrp1Y63F;Ldlr−/− bone-marrow was transplanted into irradiated Ldlr−/− mice . After 16 weeks of HCHF diet all irradiated mouse groups that underwent bone marrow transplant exhibited high plasma cholesterol levels ( Figure 6A ) . However , there was no difference in plasma lipid and apolipoprotein levels between Ldlr−/− → Ldlr−/− and Lrp1Y63F; Ldlr−/− → Ldlr−/− mice ( Figure 6B , C and D ) . By comparing Figure 6C to Figure 2C , we demonstrated that bone marrow transplantation eliminated the difference in peripheral lipid profiles , thereby demonstrating that the Lrp1Y63F mutation in macrophages is not responsible for the changes in the lipid profile . Furthermore , the correction of HDL and ApoA1 levels in the bone marrow transplant model suggested a role for Y63 phosphorylation in the liver . Increased catabolism of ApoA1 is a hallmark of the functional impairment of ABCA1 in Tangier’s disease ( Tall and Wang , 2000 ) and in liver Abca1 deficient mice ( Timmins et al . , 2005 ) . We therefore investigated the effect of the Lrp1Y63F mutation in the liver . After a 16 week HCHF diet challenge , Western blot analysis showed decreased expression of Abca1 in the liver from Lrp1Y63F;Ldlr−/− mice ( Figure 7A ) , but no change in Abca1 expression in the liver of Lrp1Y63F;Ldlr−/− → Ldlr−/− mice ( Figure 7B ) . Therefore , as in Tangier’s disease , the significant decrease in hepatic Abca1 likely accounts for the decreased plasma ApoA1 in the Lrp1Y63F mouse . However , even though hepatic Abca1 expression ( and therefore plasma ApoA1 levels ) is normalized in the bone marrow transplantation model , Lrp1Y63F;Ldlr−/− → Ldlr−/− still developed more atherosclerotic lesions when compared to Ldlr−/− → Ldlr−/− controls on HCHF diet ( Figure 8 ) . These results indicate that the relatively modest change in lipoprotein profiles ( increased chylomicrons and decreased HDL ) is insufficient to explain the striking increase in atherosclerosis generated by the Lrp1Y63F mutation . Instead , a cell autonomous effect of Lrp1 tyrosine phosphorylation in macrophages is primarily responsible for increased atherosclerosis . Histological analysis of the atherosclerotic plaque suggests a role for macrophage apoptosis as well as intracellular lipid accumulation . To examine whether the Lrp1Y63F mutation impacts macrophage foam cell formation , we incubated peritoneal macrophages isolated from Ldlr−/− and Lrp1Y63F;Ldlr−/− mice with oxLDL . Intracellular lipid accumulation was significantly higher in cultured Lrp1Y63F;Ldlr−/− macrophages compared to control cells as shown by Oil-Red O staining ( Figure 9A ) . To understand the etiology for increased intracellular lipid accumulation , [3H]-labeled cholesterol efflux experiments in peritoneal macrophages from Ldlr−/− and Lrp1Y63F;Ldlr−/− mice were performed . Compared to Ldlr−/− , macrophages from Lrp1Y63F;Ldlr−/− mice had significantly reduced cholesterol efflux capacity to ApoAI , but not to HDL ( Figure 9B ) . We next examined the expression profiles of lipid transporters involved in cholesterol transport ( Figure 9C ) . In response to oxLDL , Cd36 was increased in Ldlr−/− and Lrp1Y63F;Ldlr−/− macrophages . However , there was decreased expression of the reverse cholesterol transporter Abca1 both at baseline and in response to oxLDL in Lrp1Y63F;Ldlr−/− macrophages , consistent with the cholesterol efflux data . No difference in the expression levels of Abcg1 and Srb1 were observed in the two genotypes after oxLDL stimulation . As both ABCA1 and LRP1 are transmembrane proteins , we used surface biotinylation to investigate the distribution of these proteins . Analysis of surface and total expression of Abca1 and Lrp1 in oxLDL-stimulated macrophages revealed no difference in the ratio of surface/total Abca1 or Lrp1 between Lrp1Y63F;Ldlr−/− and Ldlr−/− macrophages ( Figure 9D ) , suggesting that translocation of Abca1 and Lrp1 to the cell membrane is functioning normally . Finally , cross-sectional analysis of the aortic root confirmed greatly decreased expression of Abca1 in atherosclerotic plaques from HCHF diet fed Lrp1Y63F;Ldlr−/− mice in vivo ( Figure 10 ) . Together , these results demonstrate that macrophage cholesterol efflux through Abca1 is selectively impaired by the Lrp1Y63F mutation . We next explored transcriptional mechanisms that might be regulated by LRP1 tyrosine phosphorylation and impact ABCA1 expression . A major regulator of ABCA1 expression is LXR: both LXRα and LXRβ respond to oxidized sterols and regulate cholesterol transport ( Janowski et al . , 1996 ) . In macrophages , activation of PPARγ induces LXRα transcriptional activation , thereby increasing ABCA1 expression ( Chawla et al . , 2001 ) . We therefore asked whether LRP1 may regulate PPARγ/LXR . We have previously shown in SMC that LRP1 indirectly regulates LXR-mediated ABCA1 expression independent of LXR gene transcription ( Zhou et al . , 2009a ) . An additional precedent is provided by a recent study demonstrating that the LRP1 cytoplasmic domain itself can regulate gene expression by directly interacting with PPARγ ( Mao et al . , 2017 ) . To investigate whether phosphorylation of LRP1 at Y63 affects the PPARγ/LXR pathway in oxLDL-induced ABCA1 expression , we tested the transcription levels of LXR-dependent gene regulation . The Lrp1Y63F mutation diminished oxLDL-induced Abca1 gene transcription , but not Abca7 gene transcription , which by contrast is increased in macrophages lacking LRP1 completely ( Yancey et al . , 2010 ) . Furthermore , the Lrp1Y63F mutation had no effect on Nr1h3 ( a . k . a . Lxrα ) , Nr2h2 ( a . k . a . Lxrβ ) , or Pparg gene expression in macrophages in the presence or absence of oxLDL ( Figure 11A ) , and there was no change in LXRα or PPARγ protein expression ( data not shown ) . Using the PPARγ inhibitor T0070907 , we confirmed that PPARγ activity was necessary for Abca1 expression in both Ldlr−/− and Lrp1Y63F;Ldlr−/− macrophages ( Figure 11B ) . To further investigate the role of PPARγ/LXR activation , isolated peritoneal macrophages were incubated with the PPARγ agonist Rosiglitazone ( Figure 11C ) , the nonspecific LXR agonist T0901317 ( Figure 11D ) , or LXR623 ( a LXRβ full agonist and LXRα partial agonist ) ( Figure 11E ) . Neither LXR or PPARγ agonists , over a range of doses , were able to fully correct the blunted Abca1 levels in the Lrp1Y63F;Ldlr−/− mice ( Figure 11C–E ) . Together these data demonstrate that the entire PPARγ/LXR/ABCA1 axis is severely impaired by the Lrp1Y63F mutation in macrophages and suggest that a membrane proximal event that is dependent on Lrp1 tyrosine phosphorylation is required to activate Abca1 transcription through PPARγ and LXR . We have previously shown in SMCs that LRP1 regulates mitogenic signaling , leading to ERK and AKT phosphorylation ( Boucher et al . , 2003; Zhou et al . , 2009b ) . Demers et al . showed that PI3K/AKT activation can modulate the PPARγ/LXR/ABCA1 axis ( Demers et al . , 2009 ) , thereby raising the possibility that PI3K activation might be dependent upon LRP1 . We thus investigated the effect of oxLDL and Lrp1 phosphorylation on PI3K/Akt signaling in macrophages . Phospho-Akt levels were significantly decreased in Lrp1Y63F;Ldlr−/− macrophages , both at baseline and in response to oxLDL ( Figure 12A ) . Moreover , inhibition of the PI3K/Akt pathway with LY294002 , a PI3K inhibitor , or an Akt inhibitor , in Ldlr−/− macrophages significantly attenuated Abca1 induction in response to oxLDL , such that Abca1 expression levels were comparable to those found in Lrp1Y63F;Ldlr−/− macrophages ( Figure 12B ) . To show that PI3K/Akt is required for full induction of Abca1 expression by PPARγ/LXR , Ldlr−/− macrophages were pre-incubated with PI3K/Akt inhibitors followed by PPARγ/LXR agonists . Inhibition of PI3K/Akt results in incomplete induction of Abca1 by both PPARγ and LXR agonists , similar to what is observed in macrophages carrying the Lrp1Y63F mutation ( Figure 12C and D ) . This suggested the existence of an endogenous mechanism that is dependent upon LRP1 tyrosine phosphorylation to activate PI3K/Akt and subsequently PPARγ/LXR-induced Abca1 expression . To identify the missing link that connects LRP1 to PI3K/Akt , we evaluated the numerous adaptor proteins that interact with LRP1 ( Barnes et al . , 2001; Gotthardt et al . , 2000 ) . SHC1 is such an intracellular adaptor protein with an established role in mitogenic signal transduction ( Pelicci et al . , 1992 ) . Tyrosine phosphorylation of Lrp1 is required for the binding of the intracellular adapter protein Shc1 in SMC , and the Lrp1Y63F mutation prevents binding of Shc1 to Lrp1 ( Figure 1 ) . Using co-immunoprecipitation assays , we confirmed that the interaction between phosphorylated Lrp1 and Shc1 also occurs in macrophages ( Figure 13A ) . To investigate the role of Shc1 in oxLDL-induced Abca1 expression , we employed shRNA-directed gene silencing to specifically reduce the expression levels of Shc1 in Ldlr−/− and Lrp1Y63F;Ldlr−/− macrophages . The induction of Akt phosphorylation and Abca1 expression in response to oxLDL was significantly impaired by the knockdown of Shc1 in Ldlr−/− macrophages , whereas Shc1 knock-down had no effect on Akt phosphorylation and Abca1 expression in Lrp1Y63F;Ldlr−/− macrophages ( Figure 13B ) . We therefore concluded that Akt phosphorylation and Abca1 expression is dependent on the interaction between Shc1 and phosphorylated Lrp1 . Collectively , our data provide a novel mechanistic basis for the role of macrophage LRP1 in atherogenesis . As summarized in Figure 14A , tyrosine phosphorylation of LRP1 , which is mediated by activated Src family tyrosine kinases in a range of inflammatory or proliferative conditions ( van der Geer , 2002 ) , is required for interaction with the adaptor protein SHC1 , which in turn augments PI3K/AKT activation . Phospho-AKT modulates PPARγ/LXR driven gene expression of which ABCA1 is a prominent target with well-documented roles in lipid metabolism and atherosclerosis ( Aiello et al . , 2002; Huang et al . , 2015; Rader et al . , 2009; Shao et al . , 2014; Singaraja et al . , 2002; Su et al . , 2005; Tall and Yvan-Charvet , 2015; Yvan-Charvet et al . , 2007; Zhao et al . , 2010 ) . The activation of this LRP1/SHC1/PI3K/AKT/PPARγ/LXR axis is necessary for the full effect of nuclear hormone receptor ligand-induced and PPARγ/LXR-dependent expression of ABCA1 and the engulfment receptor MerTK . In the absence of LRP1/SHC1 interaction , the decreased expression of atheroprotective LXR target genes in macrophages , as exemplified here by ABCA1 and MerTK , leads to increased intracellular lipid accumulation , macrophage foam cell formation , impaired apoptotic cell clearance and thus increased atherosclerotic plaque formation , independent of systemic lipid levels . We have generated a novel knock-in mouse model to investigate the physiological role of tyrosine phosphorylation at the distal NPxY motif within the cytoplasmic domain of LRP1 and its effect on staving off atherosclerosis . Using bone marrow transplantation , we showed that the accelerated atherosclerosis in Lrp1Y63F;Ldlr−/− mice is caused by a malfunction of macrophages leading to increased intracellular lipid accumulation independent of systemic lipid levels , and deceased apoptotic cell clearance . We further showed that LRP1 , through a phosphotyrosine-dependent interaction with SHC1 and subsequent activation of PI3K/AKT , is required for the efficient stimulation of the anti-inflammatory and cholesterol export-promoting PPARγ and LXR pathway . In this study we have concentrated on ABCA1 , a prominent , though by no means exclusive , target of this novel atheroprotective LRP1/SHC1/PI3K/AKT/PPARγ/LXR axis , which works in aggregate with other LRP1-dependent mechanisms ( Barnes et al . , 2001; Boucher et al . , 2002; El Asmar et al . , 2016; Subramanian et al . , 2014; Zhou et al . , 2009a; Zurhove et al . , 2008 ) to prevent macrophage foam cell formation , suppress inflammation and promote apoptotic cell clearance ( Tall and Yvan-Charvet , 2015 ) . The inability of Shc1 to bind to Lrp1Y63F resulted in decreased activation of PI3K/AKT and reduced expression of the LXR target genes ABCA1 and MerTK , an established inducer of apoptotic cell clearance , in macrophages . Our data further demonstrate that this SHC1-mediated AKT phosphorylation is necessary for the LRP1 dependent induction of ABCA1 . Rescue of Lrp1Y63F macrophages with LXR and PPARγ agonists did not fully correct the diminished Abca1 levels , suggesting either the existence of an alternative mechanism wherein PI3K/AKT regulates ABCA1 expression independent of PPARγ/LXR ( Chen et al . , 2012 ) , or that PI3K/AKT is a necessary upstream regulator of PPARγ/LXRα driven gene expression ( Demers et al . , 2009; Ishikawa et al . , 2013 ) . The model we propose begins with LRP1 tyrosine phosphorylation , an event mediated by Src-family kinases ( SFK ) , which are often localized to lipid rafts ( Simons and Toomre , 2000 ) . LRP1 associates with lipid rafts in the plasma membrane , thus creating transient microdomains that are cell specific ( Wu and Gonias , 2005 ) . Membrane organization into lipid rafts not only explains LRP1 tyrosine phosphorylation , but also underscores the relationship with ABCA1 to maintain the homeostatic balance of cellular membrane composition and signal transduction ( Figure 14B ) . Through cellular cholesterol export ABCA1 regulates cellular plasma membrane cholesterol content and membrane structural organization , specifically the formation of lipid rafts ( Ito et al . , 2015; Sezgin et al . , 2017; Zhu et al . , 2010 ) . In the setting of inflammation or cholesterol loading , increased proximity of LRP1 with activated Src family kinases in lipid rafts favors LRP1 tyrosine phosphorylation . This , in turn , activates a SHC1/PI3K/AKT/PPARγ/LXRα axis that promotes ABCA1 expression and cellular cholesterol export . The induction of ABCA1 leads to reduction of lipid raft cholesterol content ( Ito et al . , 2015 ) and dissociation of LRP1 from lipid rafts , thus creating a self-limiting negative feedback loop and returning homeostatic balance to the plasma membrane . In cases of impaired LRP1 phosphorylation , the dynamic interaction between LRP1 and lipid rafts may be intact , however the downstream signaling events triggered by LRP1 tyrosine phosphorylation cannot be initiated , thus impairing cholesterol homeostasis and plasma membrane composition . LRP1 and ABCA1 , however , play an even larger role by integrating cholesterol homeostasis with inflammatory signaling . By modulating membrane cholesterol content and lipid raft organization , macrophage ABCA1 regulates trafficking of Toll-like-receptor 4 ( TLR4 ) to lipid rafts ( Zhu et al . , 2010 ) . In lipid rafts , TLR4 forms a complex with TLR6 and oxLDL-binding CD36 ( Figure 14B and C ) , which leads to recruitment of the adaptor MyD88 , thus triggering inflammatory signaling ( Stewart et al . , 2010 ) . ABCA1 itself has anti-inflammatory properties , as interaction of ApoA1 with ABCA1 activates JAK2/STAT3 ( Tang et al . , 2009 ) . Finally , as demonstrated by Ito et al , induction of Abca1 by LXRs inhibits inflammatory NF-κB and MAPK signaling pathways downstream of TLRs ( Ito et al . , 2015 ) . Therefore , by modulating LXR-driven ABCA1 gene expression through the SHC1/PI3K/AKT/PPARγ/LXRα axis , LRP1 is presenting itself as a pivotal regulator of the metabolic as well as inflammatory processes underlying atherosclerosis . In addition to the indirect anti-inflammatory effect through ABCA1 , there is also emerging data that LRP1 can directly regulate inflammatory gene transcription in macrophages ( Zurhove et al . , 2008 ) . After cleavage of the LRP1 intracellular domain ( ICD ) by γ-secretase from the plasma membrane , the LRP1 ICD translocates into the nucleus and acts as a transcriptional repressor of genes for inflammatory cytokines ( Figure 14B ) ( Zurhove et al . , 2008 ) . In an analogous manner in endothelial cells , the LRP1 ICD has been shown to directly interact with PPARγ and function as a transcriptional co-regulator of the PPARγ target gene Pdk4 ( Mao et al . , 2017 ) . Together these findings underscore the integrative relationship between LRP1 and transcriptional regulators LXR and PPARγ and their interdependent roles in cholesterol homeostasis and inflammation . A third critical macrophage process underlying atherogenesis is clearance of apoptotic cells , or efferocytosis ( Figure 14C ) . Macrophage-specific deficiency of LRP1 is associated with increased macrophage apoptosis and decreased efferocytosis , leading to larger necrotic cores in atherosclerotic plaque ( Misra and Pizzo , 2005; Yancey et al . , 2010; Yancey et al . , 2011 ) Mechanistically , this has been associated with reduced levels of pAKT ( Yancey et al . , 2010 ) , similar to the findings in our Lrp1Y63F;Ldlr−/− mice . An additional mechanism through which the Lrp1Y63F mutation may affect efferocytosis is through impaired activation of LXR . LXR-deficient macrophages have reduced phagocytic capacity ( A-Gonzalez et al . , 2009 ) , which is associated with decreased expression of Mertk , an LXR-responsive gene . MerTK is critical for efficient macrophage phagocytosis and efferocytosis , and deficiencies in macrophage MerTK expression contribute to increased necrotic cores in atherosclerotic lesions ( Li et al . , 2006; Scott et al . , 2001; Thorp et al . , 2008 ) . Finally , LRP1 itself is necessary for the efficient phagocytosis of apoptotic cells , as it is required for the formation of a multiprotein complex involving LRP1 , the receptor tyrosine kinase AXL , and the scaffolding protein RANBP9 ( Subramanian et al . , 2014 ) . As described by Subramanian et al . , whereas AXL is involved in the recognition and binding of apoptotic cells , LRP1 , by associating with GULP1 ( Su et al . , 2002 ) , is required for the final engulfment of the dead cells . Whether these mechanisms , which are distinct from the Src kinase family-mediated activation of the LRP1/Shc1 signaling cascade , are themselves also modulated by LRP1 tyrosine phosphorylation remains currently unknown , but can now be tested using the knockin mouse model we have reported here . Our results , which show an identical atherogenic lipoprotein profile upon cholesterol-feeding of Ldlr−/− mice transplanted with Ldlr−/− or Lrp1Y63F;Ldlr−/− bone marrow , differ on first glance from those of Bi et al . ( Bi et al . , 2014 ) , who showed reduced plasma lipid levels in mice lacking Abca1 completely in their monocytes/macrophages . This effectively negated the proatherogenic effect of Abca1 deficiency in lesion macrophages . In their model , however , elevated plasma cholesterol levels induced by cholesterol feeding and the resulting cholesterol overload of circulating cells led to the secretion of inflammatory cytokines , which reduced VLDL secretion by the liver and thus resulted in a less atherogenic plasma lipid profile , which subsided on a low cholesterol chow diet . In our case , monocyte/macrophages carrying the Y63F mutation in Lrp1 retain residual , though substantially lower , Abca1 expression , which would protect the circulating cells from cholesterol overload in a hyperlipidemic , cholesterol-fed state . In conclusion , our current studies have revealed a novel mechanism through which macrophage LRP1 not only regulates ABCA1 expression to maintain cholesterol efflux , but also integrates cellular cholesterol homeostasis with inflammation and efferocytosis . In aggregate , these macrophage-specific processes demonstrate the atheroprotective role of LRP1 in mechanisms distinct from those found in SMCs . Together , these results underscore the cell-specific nature of LRP1-mediated signaling and highlight the central role of LRP1 in integrating cellular cholesterol homeostasis with other atheroprotective processes . The PPARγ inhibitor T0070907 ( cat# T8703 ) , LXR activator T0901317 ( cat# T2320 ) , LXR 623 ( ApexBio ) , Rosiglitizone ( cat# R2408 ) , LY294002 ( cat# L9908 ) , and Akt inhibitor 1 , 3-dihydro-1- ( 1- ( ( 4- ( 6-phenyl-1H-imidazo[4 , 5-g]quinoxalin-7-yl ) phenyl ) methyl ) -4-piperidinyl-2H-bezimidazole-2-one trifluoroacetate salt hydrate ( cat# A6730 ) , and In Situ Cell Death Detection Kit , TMR red ( cat# 12156792910 ) were purchased from Sigma-Aldrich ( St . Louis , MO ) . EZ-LinkTM-Sulfo-NHS-SS-Biotin ( cat# 21331 ) , cholesterol kit ( cat# TR13421 ) and triglyceride kit ( cat# TR22421 ) were purchased from Thermo Fisher Scientific ( Rockfold , IL ) . Antibodies against ABCG1 ( cat# ab36969 ) , β-actin ( cat# ab8227 ) , CD68 ( cat# ab955 ) , α-actin ( cat# ab5694 ) , and LXRα ( cat# ab41902 ) were purchased from Abcam ( Cambridge MA ) . Antibodies against ABCA1 ( cat# NB400-164 ) , SR-B1 ( cat#NB400-104 ) , and CD36 ( cat# NB400-144 ) were from Novus Biologicals ( Littleton , CO ) . Antibodies against cPLA2 ( cat# 2832 ) , p-cPLA2 ( cat# 2831 ) , Calnexin ( cat#2433 ) , total Akt ( cat# 9272 s ) , p-Akt ( cat# 9271 ) , PDGFRβ ( cat# 3169 ) , p-PDGFRβ ( cat# 88H8 ) , p-Smad2/3 ( cat# 3108 ) , PPARγ ( cat#2443 ) , Erk 1/2 ( cat# 4695 ) and p-Erk1/2 ( cat# 4370 ) were purchased from Cell Signaling ( Danvers , MA ) . Antibodies against Shc1 ( cat# 610081 ) , Dab2 ( cat# 610465 ) , and Mac-3 ( cat# 550292 ) were purchased from BD Transduction Laboratories ( Franklin Lakes , NJ ) . The antibodies against apoB ( cat# 178467 ) , apoE ( cat# 178479 ) , PDGF-BB ( cat# GF149 ) , and phosphotyrosine 4G10 antibody ( cat# 05–1050 ) and were purchased from EMD Millipore ( Billerica , MA ) . The antibody against MerTK ( cat# AF591 ) was purchased from R and D Systems . Purified human apoAI was a gift from Mingxia Liu and John S . Parks at Wake Forest School of Medicine . A rabbit polyclonal antibody against a C-terminal epitope of LRP1 was synthesized as previously described ( Herz et al . , 1988 ) . The targeting strategy used to generate the mice carrying the Lrp1Y63F substitution is illustrated in Figure 1 and detailed below . 8 week old mice were fed normal rodent Chow diet ( Harlan Laboratories , Indianapolis , IN ) or a high cholesterol/high fat ( HCHF ) diet containing 21% ( w/w ) milk fat , 1 . 25% ( w/w ) cholesterol and 0 . 5% ( w/w ) cholic acid ( TD 02028 , Harlan Laboratories , Indianapolis , IN ) for 16 weeks . A loxP-flanked exon 88-exon 89 fragment followed by exon 88-neomycin-exon 89 cassette was inserted downstream of exon 87 of the murine LRP1 gene ( Figure 1 ) . The TAT sequence in the distal NPxY motif of the second exon 89 encoding for tyrosine 63 was mutated to TTT coding for phenylalanine by site-directed mutagenesis . The construct was electroporated into murine 129Sv/J embryonic stem cells . Homologous recombinant clones were identified by Southern blot analysis . A positive ES cell clone was injected into blastocysts to generate germ line chimaeras . Mutant offspring were generated by inter-crossing with Meox-Cre transgenic mice to remove the first exons of 88 and 89 . Homozygous Lrp1Y63F knock-in mutant mice ( Lrp1ki/ki , Lrp1Y63F ) were subsequently crossed to homozygous LDL receptor knockout mice ( Ldlr−/− ) on a C57Bl/6J background . The resultant Lrp1Y63F;Ldlr−/− and Lrp1loxp/loxpLdlr−/− ( =Ldlr−/− ) and their offspring were used for further experimental analysis . Bone marrow recipients ( Ldlr−/− mice ) were treated twice with 500 rad with a 4 hr break between exposures . Bone marrow was flushed from the femurs , tibia and fibulas of donor mice ( Ldlr−/− or Lrp1Y63F;Ldlr−/− mice ) using PBS with 2% FCS . The bone marrow suspension was passed three times through an 18 g needle to create a single cell suspension and filtered on a 100 uM cell filter . Suspension was centrifuged at 1000 g for 6 min then diluted to 2 × 107 cells/ml in low glucose DMEM . Recipient mice were anesthetized using isoflurane and received a 100 ul injection of the cell dilution via the retro-orbital plexus . Post-transplant recipient mice were housed in sterile caging and administered prophylactic antibiotics in their drinking water for two weeks , followed by the indicated diet for 16 weeks . Primary mouse aortic smooth muscle cells were explanted from the thoracic aorta of 8–10 week old mice with indicated genotype , following the protocol by Clowes et al . ( 1994 ) . Briefly , the aortas were dissected under sterile conditions and the connective tissue and adventitia were removed carefully . The aortas were opened longitudinally and the intimas were scraped on luminal surface . Then the aortas were minced into small pieces and placed into a T25 flask with high glucose ( 4 . 5 g/L ) DMEM containing 15% FCS , 100 U/ml penicillin , 100 mg/ml streptomycin , 20 mM L-glutamine . Both explants and cells were cultured at 37°C in 5% CO2 . Cells were detached by incubation with 0 . 25% trypsin-EDTA solution . Cell of passages 4–6 were used for the experiments under the indicated conditions . Mouse peritoneal macrophages from Ldlr−/− and Lrp1Y63F;Ldlr−/− mice were isolated 3 days after intraperitoneal injection of 3% thioglycollate solution and cultured in RPMI-1640 Medium containing 10% FBS . After overnight starvation , macrophages were treated under the indicated conditions , followed by Oil Red O staining or lysis for western blot analysis . Oxidized LDL ( oxLDL ) was prepared from LDL isolated from human plasma , which was treated with 10 μM CuSO4 for 24 hr at 37°C and then dialyzed in 0 . 09% saline containing 0 . 24 mM EDTA . SMC migration was assayed using a 6 . 5 mm- Transwell with 8 µm polycarbonate pore membrane inserts ( cat# 3422 , Corning ) . 5 × 104 cells in 100 µl were loaded into the top chamber of each well while the lower chambers were filled with 0 . 5% DMEM with or without PDGF-BB ( 10 ng/ml ) . After incubating at 37°C in 5% CO2 for 6 hr , non-migrated cells were scraped from the upper surface of the membrane insert . Cells on the lower surface were fixed with methanol and stained with Harris Modified Hematoxylin ( HHS-16 , Sigma ) . The number of SMC on the lower surface of the membrane insert was determined by counting four continuous high-power ( 200× ) fields of constant area per well . Experiments were performed three times in duplicate wells . For SMC proliferation assays , 5000 cells were seeded into each well of 96-well plates and starved for 24 hr . Cells were incubated with or without 10 ng/ml PDGF-BB for 24 hr and then stained using BrdU Cell Proliferation Assay Kit ( cat# 6813 , Cell Signaling Technology ) , following the instructions . After 16 weeks of HCHF diet feeding , mice were subjected to an overnight fasting period and were then euthanized with isoflurane . Blood samples were collected for future use . For tissue preparation , the heart and entire aorta to the iliac bifurcation were harvested after perfusion with phosphate buffered saline ( PBS ) followed by 4% paraformaldehyde ( Electron Microscopy Sciences , Hatfield , PA ) . After 24 hr in 4% paraformaldehyde , tissues were stored in 30% sucrose solution in PBS until further analysis . Aortas were dissected and opened longitudinally under a dissection microscope ( Model Z30L , Cambridge Instruments ) . Hearts were embedded in Tissue-Tek O . C . T . compound and frozen to −20°C for analysis of atherosclerotic lesions at the aortic root . Three 10 µm thick cryosections spaced 100 µm apart were used for analysis . To evaluate the lesion size , the cryosections and whole aorta were stained with Oil Red O or hematoxylin and eosin . Semi-quantitative immunostaining was performed using primary antibodies against Mac-3 ( 1:200; BD Pharmingen ) , α–smooth muscle actin ( 1:200; Abcam ) , ABCA1 ( 1:50; Novus Biologicals ) or CD68 ( 1:50; Abcam ) and fluorescently labeled secondary antibodies goat anti-rat Alexa Fluor 488 ( Thermo Fisher Scientific , A11006 ) , goat anti-rabbit Alexa Fluor 594 ( Thermo Fisher Scientific , A11012 ) , or goat anti-mouse Alexa Fluor 594 ( Thermo Fisher Scientific , A11032 ) . Nuclei were counterstained with DAPI ( Thermo Fisher Scientific , P36935 ) . Images were obtained with a Zeiss Axiophot microscope , and the percentage of lesion area that was positively stained was determined using Image-Pro v . 6 . 2 ( Media Cybernetics ) . Cryo-sections of aortic roots or mouse peritoneal macrophages were prepared as described above and were stained using an In Situ Cell Death Detection Kit , following the manufacture’s instructions . Briefly , tissues or cells were rinsed in PBS and fixed with 4% paraformaldehyde solution at room temperature for 15 min , followed by washing in PBS for 30 min . Samples were incubated with 0 . 1% Triton X-100 in cold PBS for 2 min and rinsed twice in PBS , then incubated with TUNEL reaction mixture for 1 hr at 37°C in a humidified chamber in the dark . Samples were washed twice in PBS and analyzed directly by fluorescence microscopy . Nuclei were counterstained with DAPI . Total RNA from peritoneal macrophages treated with or without oxLDL was isolated by using TRIzol Reagent ( Life technologies ) and cDNA was prepared with a RT-kit ( Applied Biosystems ) . The expression levels of Abca1 , Pparg , Nr1h3 and Nr2h2 were measured by RT-PCR on 7900HT Fast Real-time PCR system with SYBR Green reagents and the following primer sets: mouse Abca1 ( Forward: 5’-CGTTTCCGGGAAGTGTCCTA-3’ , Reverse: 5’-GCTAGAGATGACAAGGAGGATGGA-3’ ) , mouse Abca7 ( Forward: 5’- ATCCTAGTGGCTGTCCGTCA-3’ , Reverse: 5’-ATGGCTTGTTTGGAAAGTGG-3’ ) , mouse Pparg ( Forward: 5’- CACAATGCCATCAGGTTTGG-3’ , Reverse: 5’-GCTGGTCGATATCACTGGAGATC-3’ ) , mouse Nr1h3 ( Forward; 5’-TCTGGAGACGTCACGGAGGTA-3’ , Reverse: 5’-CCCGGTTGTAACTGAAGTCCTT-3’ ) , Nr2h2 ( Forward: 5’-CTCCCACCCACGCTTACAC-3’ , Reverse: 5’-GCCCTAACCTCTCTCCACTCA-3’ ) and mouse Cyclophilin ( Forward: 5’-TGGAGAGCACCAAGACAGACA-3’ , Reverse: 5’-TGCCGGAGTCGACAATGAT-3’ ) , which was used as the internal standard . Mouse peritoneal macrophages were grown in 6-well culture dishes and cell surface proteins were biotinylated as previously described . ( Ding et al . , 2016 ) After ox-LDL treatment , macrophages were washed with cold PBS buffer and then incubated in PBS buffer containing 1 . 0 mg/ml sulfo-NHS-SS-biotin ( Pierce ) for 30 min at 4°C . Excess reagent was quenched by rinsing in cold PBS containing 100 mM glycine . Cell lysates were prepared in 160 μl of RIPA lysis buffer [50 mM Tris-HCl , 150 mM NaCl , 1% NP-40 , 2 mM EDTA , 2 mM MgCl2 , and protease inhibitor mixture ( Sigma ) , ( pH8 . 0 ) ] . After 20 min incubated at 4°C , lysates were collected and centrifuged at 14 , 000 × rpm for 10 min . 100 μg of total proteins were incubated with 100 μl of NeutrAvidin agarose ( Pierce ) at 4°C for 1 . 5 hr . Agarose pellets were washed three times using washing buffer [500 mM NaCl , 150 mM Tris-HCl , 0 . 5% Triton X-100 ( pH8 . 0 ) ] , biotinylated surface proteins were eluted from agarose beads by boiling in 2x SDS sample loading buffer . Protein were separated by sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) , transferred to nitrocellulose , and blotted with different antibodies . Peritoneal macrophages were seeded in 24-well plates , labeled with 2 μCi/ml [3H]cholesterol ( cat# 8794 , Sigma-Aldrich ) , and loaded with 100 μg/ml oxLDL for 24 hr . Cells were then washed and equilibrated with 0 . 2% BSA-DMEM medium overnight . To determine cholesterol efflux , medium containing 20 μg/ml free apoA-I or 25 μg/ml HDL was added to cells . After 4 hr , aliquots of the medium were collected , and the [3H]cholesterol released was measured by liquid scintillation counting . The [3H]cholesterol remaining in the cells was determined by extracting the cell lipids in 1N NaOH and measured by liquid scintillation counting . The cholesterol efflux capacity ( efflux% ) is expressed as medium count/ ( medium count +cell count ) X 100% . Plasma samples were obtained by cardiac puncture from mice fasted overnight . Total plasma cholesterol and triglyceride levels were determined by commercially available enzymatic kits ( TR13421 and TR22421 , Thermo Scientific , Waltham , MA ) following manufacturer`s instructions . For lipoprotein profile analysis , plasma samples were prepared by using Microvette 500K3E EDTA-columns ( Sarstedt , Germany ) following the same procedure as described for serum . Plasma lipoproteins were fractionated by fast-protein liquid chromatography ( FPLC ) using a Superose six column ( 10/300 GL ) . Cholesterol and triglyceride content of all fractions was measured by using the same method as described for serum lipid levels . Briefly , 20 µg of protein extracted from specified tissues or primary cultured cells or 20 μl of fractionated plasma samples was loaded on SDS-PAGE gel ( BioRad , Hercules , CA ) and transferred to nitrocellulose membranes ( HybondTM-C Extra . RPN303 , Amersham Biosciences , Piscataway , NJ ) . Blots were developed , visualized , and analyzed using the LiCor Odyssey CLX or ECL . Primary smooth muscle cells ( SMC ) were cultured in serum-deprived DMEM medium ( 4500 mg/L glucose , Sigma-Aldrich ) for 24 hr and then treated with 10 ng/ml PDGF-BB for 10 min . To show reduced phosphorylation in Lrp1Y63F SMC , membrane fractions were obtained . Briefly , cell pellets were homogenized in homogenization buffer ( 20 mM TrisCl , 2 mM MgCl2 , 0 . 25M Sucrose , 10 mM EDTA , and 10 mM EGTA ) and spun down at low speed at 800 g for 5 min . Supernatants were transferred to into new tubes for ultracentrifugation at 55 , 000 rpm for 30 min ( MX-120 , Thermo Scientific ) . After removal of the supernatant , the pellet was re-suspended in re-suspension buffer ( 50 mM TrisCl , 2 mM CaCl2 , 80 mM NaCl , and 1% NP-40 ) . 500 µg membrane protein extract was used for immunoprecipitation ( IP ) and pre-cleared with non-immune serum . Non-specific binding was precipitated with protein A DynaBeads ( Invitrogen , Grand Island , NY ) . Supernatant was mixed with mouse anti-phospho-tyrosine antibody at 4°C for 4 hr , and precipitated with DynaBeads at 4°C overnight . The antigen-antibody-protein complex was washed three times with wash buffer ( 500 mM NaCl , 150 mM Tri-HCl , 0 . 5% Triton-X100 , pH8 . 0 ) . The washed antigen-antibody-protein complex was resolved on SDS-PAGE gel and immunoblotted with rabbit anti-LRP1 antibody . To show altered adaptor-protein binding properties in the Lrp1Y63F knock in mutation , SMC were treated with 10 ng/ml PDGF BB for 10 min following a starvation period of 24 hr . Cells were cross-linked using 0 . 05 µM DSP ( Thermo Scientific , Waltham , MA ) for 30 min and lysed in PBS containing 1% TritonX with proteinase inhibitor cocktail and phosphatase inhibitor for 30 min . Immunoprecipitation and Western blots were performed as described for the reduced phosphorylation experiment with secondary antibodies against adaptor proteins Shc1 and Dab2 . For the macrophage co-IP experiments , membrane proteins from macrophages were prepared as described above after oxLDL treatment ( 100 μg/ml ) for 24 hr . A pLKO . 1 lentiviral vector expressing Shc1 shRNA ( TRCN0000055180 ) , with the functional sequence CCGGGCTGAGTATGTTGCCTATGTTCTCGAGAACATAGGCAACATACTCAGCTTTTTG , to target the Shc-1 gene sequence ( GCTGAGTATGTTGCCTATGTT ) , was purchased from Sigma Aldrich ( St . Louis , MO ) . The scramble shRNA was a gift from David Sabatini ( Addgene plasmid #1864 ) . ( Sarbassov et al . , 2005 ) The hairpin sequence was as follows: CCTAAGGTTAAGTCGCCCTCGCTCGAGCGAGGGCGACTTAACCTTAGG . To generate lentiviral particles , HEK293T cells were co-transfected with the Shc1 shRNA lentiviral plasmid pLKO . 1 and Lentiviral packaging mix ( psPAX2 , pMD2 . G; Addgene ) using Fugene 6 Transfection Reagent ( Roche , Hamburg , German ) . Culture supernatants containing lentivirus were harvested 48 hr after transfection . For lentiviral transduction , primary macrophage cells were treated with either Shc1-shRNA lentivirus , or scramble shRNA lentivirus for 24 hr . After recovering from viral infection , cells were incubated in fresh RPMI-1640 Medium with 100 µg/mL oxLDL for 24 hr . Cells were then lysed and protein analysis was performed by western blotting . All values were expressed as mean ±SEM . Unpaired 2-tailed student's t test or two-way ANOVA test were used for statistical analysis . A P value less than 0 . 05 between two groups was considered significant . Animal procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee ( IACUC ) at the University of Texas Southwestern Medical Center at Dallas .	Atherosclerosis is a disease in which “plaques” build up inside the walls of arteries . Plaques consist of a fatty substance called cholesterol , together with immune cells such as macrophages and other material from the blood . Over time , the plaque narrows and hardens the arteries . This restricts the flow of blood to vital parts of the body , which increases the risk of heart attacks , strokes and other severe conditions . Macrophages play an important role in atherosclerosis . At the early stage of the disease , macrophages enter the developing plaques to take up the excess cholesterol . Cholesterol taken up by macrophages needs to be exported out of the cell and sent to the liver for removal . Yet , these processes can go awry . Macrophages can fill up with too much cholesterol and become trapped in the arteries . These cholesterol-laden macrophages can also start dying . These problems enable the plaques to grow and worsen the disease . LRP1 is an important protein present on the surface of many types of cells . In macrophages , LRP1 helps to export excess cholesterol out of the cell , thus lowering the risk of atherosclerosis . LRP1 also reduces cell death in the plaque , which slows the plaques’ progression . Previous research has shown that the region of LRP1 present inside the cell can be modified by the attachment of a phosphate group – a process termed phosphorylation . Whether phosphorylation of LRP1 plays a role in preventing atherosclerosis is not understood . To address this question , Xian , Ding , Dieckmann et al . engineered mice in which LRP1 was unable to get phosphorylated . The results show that phosphorylated LRP1 – but not the non-phosphorylated version – turns on a signaling pathway in macrophages . This pathway increases the expression of a transporter protein that exports cholesterol out of the cell . This reduces the amount of cholesterol that accumulates in macrophages . Lastly , mice with problems with LRP1 phosphorylation developed more severe atherosclerotic plaques with more dying cells present in the affected areas compared to normal mice . These findings show how phosphorylation of LRP1 protects against atherosclerosis . Understanding this process in further detail may help scientists to devise new ways to treat this disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	LRP1 integrates murine macrophage cholesterol homeostasis and inflammatory responses in atherosclerosis
Emerging evidence points to an unexpected diversification of core promoter recognition complexes that serve as important regulators of cell-type specific gene transcription . Here , we report that the orphan TBP-associated factor TAF9B is selectively up-regulated upon in vitro motor neuron differentiation , and is required for the transcriptional induction of specific neuronal genes , while dispensable for global gene expression in murine ES cells . TAF9B binds to both promoters and distal enhancers of neuronal genes , partially co-localizing at binding sites of OLIG2 , a key activator of motor neuron differentiation . Surprisingly , in this neuronal context TAF9B becomes preferentially associated with PCAF rather than the canonical TFIID complex . Analysis of dissected spinal column from Taf9b KO mice confirmed that TAF9B also regulates neuronal gene transcription in vivo . Our findings suggest that alternative core promoter complexes may provide a key mechanism to lock in and maintain specific transcriptional programs in terminally differentiated cell types . Sequence-specific transcription factors are essential for directing cellular differentiation and identity . These factors play key roles in establishing gene expression patterns that dictate cell-type specific functions and characteristics . Cellular reprogramming and de-differentiation leading to the formation of induced pluripotent stem cells offer dramatic examples of how targeted expression of particular sets of transcription factors can control cell fate ( reviewed by Graf , 2011 ) . Sequence-specific transcription factors work in concert with a cohort of co-activator complexes and core promoter recognition factors to execute transcriptional activation . These core promoter factors often called ‘basal’ or ‘general’ transcription factors have traditionally been considered invariant and universally required for the expression of all protein-coding genes . However , recent studies suggest that at least some components of this large and diverse group of factors necessary to form the transcriptional pre-initiation complex ( PIC ) can exhibit both promoter and enhancer targeting activities that are cell-type specific . This specificity depends on the variegated composition of the components in the PIC ( reviewed in D'Alessio et al . , 2009; Goodrich and Tjian , 2010; Müller et al . , 2010 ) . The first evidence of tissue-specific functions of core PIC factors came from studies of TAF ( TBP-associated factors ) subunits in the core promoter recognition factor TFIID . A paralog of TAF4 , called TAF4B , was found to control oocyte-specific activation of transcription in mouse ovaries ( Freiman et al . , 2001 ) . In Drosophila , a group of five TAF paralogs ( No hitter/TAF4; Cannonball/TAF5; Meiois I arrest/TAF6; Spermatocyte arrest/TAF8; and Ryan express/TAF12 ) all play specific roles in spermatogenesis ( Hiller et al . , 2004; Chen et al . , 2005 ) . Similarly , another orphan TAF , TAF7L , cooperates with TBP-related factor 2 ( TRF2 ) to regulate spermatogenesis in mice ( Cheng et al . , 2007; Zhou et al . , 2013a ) . Tissue-specific functions of TAF7L were also found in adipocytes where it acts in conjunction with PPARγ to control the transcription necessary for adipogenesis ( Zhou et al . , 2013b ) . In mouse embryonic stem ( ES ) cells , TAF3 pairs up with CTCF to drive the expression of endoderm specific genes while in myoblasts TAF3 works with TRF3 in the differentiation of myotubes ( Deato and Tjian , 2007; Liu et al . , 2011 ) . Collectively these experiments suggest that combinations of different subunits of the multi-protein core promoter factors can be enlisted to participate in gene- and tissue-specific regulatory functions . Thus , mouse ES cells and other progenitor cells very likely have quite different requirements for such factors compared to terminally differentiated mature cell-types . Dissecting the various diversified mechanisms that control gene transcription in terminally differentiated cells should contribute to our still rudimentary understanding of the gene regulatory processes that modulate homeostasis in somatic cells and those that could lead to degeneration of adult tissue in disease states . A more detailed analysis of these critical molecular mechanisms may also help improve new strategies to achieve efficient cellular reprogramming and stem cell differentiation . Despite emerging evidence for unexpected activities carried out by core promoter factors in various cellular differentiation pathways , little was known about their potential involvement in the formation of neurons during embryogenesis . In this study we explore whether TAFs or other core promoter recognition factors become engaged in neuronal specific functions to regulate the expression of neuronal genes . To address this question we used an in vitro differentiation protocol to induce murine ES cells to form spinal cord motor neurons ( MN ) , which control muscle movement . Loss of motor neurons gives rise to devastating diseases , including amyotrophic lateral sclerosis ( ALS ) ( reviewed by Robberecht and Philips , 2013 ) . Consequently , motor neurons have been the focus of intense study and several key classical sequence-specific DNA-binding transcription factors regulating the expression of motor neuron-specific genes have been identified ( reviewed by di Sanguinetto et al . , 2008; Kanning et al . , 2010 ) . However , there was scant information regarding the role , if any , of core promoter factors in directing the network of gene transcription necessary to form neurons . In this report , we have combined genomics , biochemical assays , and gene knockout strategies to dissect the transcriptional mechanism used to generate motor neurons from murine ES cells in vitro as well as to uncover novel in vivo neuronal-specific changes in core promoter factor involvement and previously undetected co-activator functions . To examine whether the expression of various components of the core promoter recognition complex changes upon neuronal differentiation , we induced ES cells to form motor neurons using retinoic acid ( RA ) and the smoothened agonist SAG as described previously ( Wichterle et al . , 2002 ) . We confirmed the generation of motor neurons in embryoid bodies ( EBs ) by immunostaining for motor neuron-specific markers LHX3 and ISL1/2 ( Figure 1A ) as well as by RNA-seq analysis ( Figure 1—figure supplement 1A ) . To obtain enriched populations of motor neurons , we differentiated a murine ES cell line containing a motor neuron-specific promoter ( Mnx1 ) fused to GFP as a means of isolating post-mitotic motor neurons present in these EBs ( Figure 1B , Figure 1—figure supplement 1B; Wichterle et al . , 2002 ) . Using this highly enriched motor neuron cell population , we compared the mRNA expression levels of genes coding for canonical TAF subunits in TFIID as well as various TAF paralogs ( Figure 1C ) . As seen in other differentiated cell types , several TAFs are down-regulated upon motor neuron differentiation ( Figure 1; Deato and Tjian , 2007; D'Alessio et al . , 2011; Zhou et al . , 2013b; Guermah et al . , 2003 ) . Intriguingly , one TAF paralog ( Taf9b ) was significantly up-regulated in GFP expressing ( GFP+ ) motor neurons but not in the GFP negative ( GFP− ) cells ( Figure 1C ) . We confirmed that the up-regulation of TAF9B and down-regulation of several other TAFs occur at the level of both mRNA and protein ( Figure 1D ) , and follows the kinetics of post-mitotic neuronal marker Tubb3 but not the progenitor cell markers Pax6 and Olig2 ( Figure 1—figure supplement 1C ) . We next dissected spinal cord tissue from newborn mice and performed RNA-seq to measure in vivo Taf expression levels and compare them to those observed for mouse ES cells in culture . As expected , most Taf subunits of TFIID in newborn spinal cord are expressed at lower levels than in mouse ES cells , while Taf9b is up-regulated more than 10-fold , consistent with the results obtained with the in vitro differentiated motor neurons ( Figure 1E ) . Notably , changes in the expression levels of Tafs in newborn spinal cord are more pronounced than what we observed for the in vitro differentiated motor neurons . We also found that many components of the PIC and selected co-activators were down-regulated upon neuronal differentiation ( Figure 1—figure supplement 1D and 1E ) . These results strongly suggest that induction of TAF9B upon neuronal differentiation is specific and distinct from regulated expression of most other subunits of the PIC . 10 . 7554/eLife . 02559 . 003Figure 1 . TAF9B is up-regulated upon neuronal differentiation . ( A ) Mouse embryonic stem ( ES ) cells were differentiated into motor neurons ( MN ) using embryoid body cultures ( EB-MN ) for 6 days and immunostained with MN markers ISL1/2 and LHX3 , and neuronal marker TUJ1 . ( B ) Mouse ES cells containing a MN-specific GFP marker were used to isolate GFP expressing ( GFP+ ) MN cells and to monitor induction of MN under different sonic hedgehog ( SHH ) concentrations . ( C ) TAF expression levels were measured by qRT-PCR in GFP+ and GFP negative ( GFP− ) cells sorted from EB-MN differentiated for 6 days and compared to the levels observed in undifferentiated ES cells . Bars show mean ± SD of three biological replicates . ( D ) Western blot analysis of ES cells and EB-MN differentiated for 6 and 8 days detecting TAFs , the neuronal marker TUBB3 , the ES cell marker OCT4 , and ACTB . ( E ) RNA-seq analysis of TAF expression comparing mouse ES cells and mouse newborn spinal cord tissue . Bars show FPKM values of spinal cord tissue relative to FPKM values of ES cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 00310 . 7554/eLife . 02559 . 004Figure 1—figure supplement 1 . TAF9B is up-regulated upon neuronal differentiation . ( A ) Gene expression levels detected by RNA-seq analysis comparing mouse ES cells and EB-MN cells differentiated for 6 days . Fold inductions are shown as log2 fold change between EB-MN and ES cells . ( B ) Mouse ES cells containing a MN-specific GFP marker were used to isolate GFP expressing ( GFP+ ) and GFP negative ( GFP− ) cells and qRT-PCR analysis was done to compare levels of induction of neuronal markers Tubb3 , Mnx1 , and Isl1 in both populations . Bars show mean ± SD of three biological replicates . ( C ) qRT-PCR analysis detecting Taf9b , neuronal marker Tubb3 , and progenitor markers Pax6 and Olig2 in mouse ES cells differentiated into EB-MN in the presence of SAG and Notch inhibitor DAPT ( ‘Materials and methods’ ) . Graphs represent mean ± SD of three biological replicates . ( D ) RNA-seq analysis comparing the expression levels of components of the core promoter machinery between mouse ES cells and newborn spinal cord tissue . Data points show FPKM values of spinal cord tissue relative to FPKM values of ES cells . MEDs = mediator subunits , GTFs = General transcription factors subunits ( TFIIA-TFIIH , not including TFIID ) . ( E ) RNA-seq analysis comparing the expression levels of the components of the SAGA and ATAC complexes between mouse ES cell and newborn spinal cord tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 004 To test the function of TAF9B in mouse ES cells and during motor neuron differentiation , we used Taf9b knock-out ( KO ) ES cells generated by the trans-NIH Knock-Out Mouse Project ( KOMP ) . In these cells , all protein-coding exons of Taf9b have been replaced with the LacZ-Neo cassette ( Figure 2A ) . As expected , the induction of Taf9b upon motor neuron differentiation was detected only in wild type ( WT ) cells but not in Taf9b KO cells ( Figure 2B , C ) . We found that Taf9b KO cells have a normal mouse ES cell morphology and can be grown in the presence of LIF without displaying any obvious cell growth or cell cycle defects ( Figure 2D and data not shown ) . To determine the role of TAF9B in regulating transcription in murine ES cells , we performed RNA-seq analysis to compare global gene expression patterns in WT and Taf9b KO cells . The results show that TAF9B is largely dispensable for global gene expression with only a small number of genes deregulated under these conditions ( Figure 2E ) . Interestingly , whereas the core TAF components of TFIID have recently been reported to be critical for murine ES cell pluripotency and in controlling genes such as Pou5f1 , Nanog and Klf4 ( Pijnappel et al . , 2013 ) , the absence of TAF9B did not affect expression of pluripotency markers ( Figure 2E , F ) . These results suggest that , unlike other TAFs present in the canonical TFIID complex , TAF9B is dispensable for global gene expression and pluripotency of murine ES cells . 10 . 7554/eLife . 02559 . 005Figure 2 . TAF9B is dispensable for global gene expression in murine ES cells . ( A ) In Taf9b KO murine ES cells a LacZ-Neo cassette replaces all protein coding exons of Taf9b gene on the X chromosome . ( B ) qRT-PCR of in vitro differentiated EB-MN for 6 and 8 days detecting the expression of Taf9b in WT and Taf9b KO cells . Bars show mean ± SEM of three biological replicates . ( C ) Western blot analysis of in vitro differentiated EB-MN for 6 and 8 days detecting the expression of TAF9B in WT and Taf9b KO cells . ( D ) Cell cycle profile of WT and Taf9b KO ES cells based on BrdU incorporation . Graph represents average values of two independent biological duplicates . ( E ) RNA-seq analysis of WT and Taf9b KO ES cells grown in undifferentiated conditions . Scatter plot represents FPKM values for genes expressed in WT and Taf9b KO cells . Red dots are genes whose expression is affected more than twofold with p-values <0 . 05 . Gray dots represent genes not changing more than the selected cutoff . ( F ) qRT-PCR analysis of the ES cell markers Pou5f1 and Nanog in WT and Taf9b KO ES cells grown in undifferentiated conditions . Bars show mean ± SEM of three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 005 We next tested whether TAF9B is required for efficient in vitro motor neuron differentiation . ES cells were differentiated as described previously and motor neuron markers Isl1 , Mnx1 , and Lhx3 were measured by qRT-PCR . We observed a significant reduction in all motor neuron markers in Taf9b KO cells upon differentiation ( Figure 3A ) . Immunostaining for motor neuron marker ISL1/2 as well as western blot analysis for TUBB3 further confirmed that motor neuron differentiation was compromised in Taf9b KO cells ( Figure 3B , C ) . We next measured the down-regulation of ES specific markers Pou5f1 and Nanog upon differentiation and found that Taf9b KO cells failed to efficiently down-regulate these ES cell markers ( Figure 3A ) . Immunostaining of OCT4 and NANOG confirmed that a fraction of cells in these neuralized EBs derived from Taf9b KO cells continue to express pluripotency markers , whereas few or no cells expressing OCT4 and NANOG were detected in WT cells ( Figure 3B ) . Consistent with the maintenance of an undifferentiated state , a higher number of dividing cells persisted in Taf9b KO EBs as determined by the marker for cellular proliferation Ki67 ( Figure 3B ) . We also found an increase in apoptosis in Taf9b KO cells compared to WT controls during early stages of differentiation ( Figure 3D ) . These differences in apoptosis were particularly prominent at day 4 when progenitor cells are most enriched . This is consistent with previous observations implicating TAF9B in cell survival and control of apoptosis in human HeLa cells ( Chen and Manley , 2003; Frontini et al . , 2005 ) . Taken in aggregate , our findings suggest that TAF9B is required for the efficient in vitro differentiation of murine ES cells into motor neurons , and that the absence of TAF9B causes the persistent expression of pluripotency markers under differentiation conditions . 10 . 7554/eLife . 02559 . 006Figure 3 . TAF9B is required for efficient differentiation of murine ES cells into motor neurons in vitro . ( A ) qRT-PCR analysis for the MN markers ( Mnx1 , Lhx3 , Isl1 ) and ES cell markers ( Pou5f1 and Nanog ) in WT and Taf9b KO cells differentiated for 6 days . Values are relative to undifferentiated ES cells . Bars show mean ± SEM of three biological replicates . ( B ) WT and Taf9b KO EB-MN , differentiated for 6 days , were immunostained using antibodies against MN marker ISL1/2 , ES cell markers OCT4 and NANOG , and cellular proliferation marker Ki67 . ( C ) Western blot analysis of WT and Taf9b KO EB-MN samples differentiated for 6 and 8 days detecting neuronal marker TUBB3 and ACTB . ( D ) Apoptosis levels were determined using an annexin V based assay in WT and Taf9b KO cells at different time points during MN differentiation . Graph shows mean ± SEM for the ratio of Taf9b KO to WT cells in three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 006 To obtain a more comprehensive view of the role of TAF9B during motor neuron differentiation , we performed RNA-seq analysis at regular intervals during in vitro motor neuron differentiation of WT and Taf9b KO ES cells . We found that as motor neuron differentiation progressed there was a concomitant increase in the number of genes deregulated by the loss of TAF9B ( Figure 4A ) . Striking deregulation of genes in Taf9b KO cells was observed at the time point at which progenitor cells are most enriched in these cultures ( day 4 ) . Genes known to be important for neuronal progenitor identity such as Olig2 , Pax6 , Sox1 , Nkx6-1 , Nkx6-2 , Gli1 , and genes controlled by the Notch signaling pathway were among the most down-regulated by the loss of TAF9B ( Figure 4B , Figure 4—figure supplement 1A ) . Gene ontology ( GO ) terms analysis of down-regulated genes showed categories such as neuron differentiation and anterior/posterior pattern formation among the most enriched terms ( Figure 4—figure supplement 1B ) . Among up-regulated genes we found genes involved in apoptosis ( Supplementary file 1 ) , which is consistent with the increase in apoptosis levels we observed in Taf9b KO cells at this time point . By day 6 , when post-mitotic motor neurons are normally present in these cultures , approximately 10% of all expressed genes were down-regulated and another 10% were up-regulated in Taf9b KO cells . This finding suggests that TAF9B is specifically required for the appropriate expression of a subset of genes , rather than for global gene expression in these differentiated EBs . The most significant GO term categories of genes down-regulated by the loss of TAF9B in this time point included genes involved in neuron differentiation , axogenesis , and neuron projection ( Figure 4C ) . Moreover , the specific motor neuron markers Lhx3 , Lhx4 , Isl1 , Mnx1 , and pan-neuronal marker Tubb3 were among the most down-regulated genes in Taf9b KO cells ( Figure 4B ) . In contrast , genes up-regulated in Taf9b KO cells were enriched in categories such as vascular development , stem cell maintenance , and other non-neuronal developmental categories ( Figure 4C ) . Contrary to the large number of genes deregulated in neuralized EB samples , we found that samples differentiated for 2 days have a relatively small number of genes affected in Taf9b KO cells . Interestingly , genes involved in differentiation were still found among these early down-regulated genes ( Supplementary file 1 ) . These results indicate that TAF9B controls the expression of specific genes required at least for the in vitro differentiation of murine ES cells into motor neurons . 10 . 7554/eLife . 02559 . 007Figure 4 . TAF9B is specifically required for the efficient activation of neuronal gene expression during in vitro motor neuron differentiation . ( A ) RNA-seq analysis of WT and Taf9b KO cells at different time points during EB-MN differentiation . Scatter plots represent FPKM values for genes expressed in WT and Taf9b KO cells . Red dots are genes whose expression is affected more than twofold with p-values <0 . 05 . Gray dots represent genes not changing more than the selected cutoff . ( B ) Selected genes from RNA-seq analysis are shown as log2 fold change between Taf9b KO and WT cells . Examples are given for key genes expressed in ES cells , progenitor cells , and motor neurons . ( C ) GO term analysis of genes affected by the loss of TAF9B in RNA-seq analysis of EB-MN differentiated for 6 days . Lists show the top eight GO term Biological Process categories obtained ranked by p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 00710 . 7554/eLife . 02559 . 008Figure 4—figure supplement 1 . TAF9B is specifically required for the efficient activation of neuronal gene expression during in vitro motor neuron differentiation . ( A ) qRT-PCR analysis of progenitor markers ( Sox1 , Pax6 , Nkx6-2 ) during differentiation of WT and Taf9b KO ES cells into EB-MN . qRT-PCR analysis of MN markers ( Isl1 , Mnx1 , Lhx3 ) and ES cell markers ( Pou5f1 , Nanog ) from Figure 3 as well as Krt18 are shown as full time course . Graphs show mean ± SEM of three biological replicates . ( B ) GO term analysis of genes affected by the loss of TAF9B in RNA-seq analysis of EB-MN differentiated for 4 days . Lists show the top eight GO term Biological Process categories obtained ranked by p-value . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 008 To determine whether TAF9B acts directly to regulate neuronal genes we mapped its binding sites genome-wide by ChIP-seq . Because most TAFs are expected to bind to promoters as part of TFIID in the PIC , we compared ChIP-seq peaks of TAF9B with those of RNA POL2 . Approximately a third of all detected TAF9B binding regions overlapped with RNA POL2 occupancy ( TAF9B-POL2 ) and mapped generally near transcription start sites ( TSS ) of protein coding genes ( Figure 5A , Figure 5—figure supplement 1B , C ) . However , TAF9B peaks located at TSS were not enriched at promoters of neuronal specific genes ( Figure 5B ) . Intriguingly , approximately two thirds of all TAF9B binding regions showed little or no overlap with regions of high RNA POL2 occupancy ( TAF9B-only ) and were generally located distal to annotated TSS ( Figure 5A , Figure 5—figure supplement 1B , 1C ) . Importantly , genes with TAF9B bound to regions distal to their TSS were significantly enriched in neuronal genes categories , including GO terms such as midbrain-hindbrain boundary development , rostrocaudal neural tube patterning and dorsal spinal cord development ( Figure 5B ) . Several of the most down-regulated genes in the absence of TAF9B detected by RNA-seq ( e . g . , Neurog2 and Lhx4 ) were associated with TAF9B distal binding sites ( Figure 5—figure supplement 1A , E ) . Moreover , genes bearing distal TAF9B peaks ( e . g . , 500–5000 bp from TSS ) showed a higher percentage of genes down-regulated in Taf9b KO neuralized EBs than control gene sets ( Figure 5—figure supplement 1D ) . Collectively , these results suggest that distal TAF9B binding sites are likely functional regulatory elements controlling neuronal genes . We subsequently compared the binding sites of TAF9B to the binding sites of OLIG2 , a key transcriptional activator involved in motor neuron differentiation mapped previously ( Mazzoni et al . , 2011 ) . We found co-localization between TAF9B and OLIG2 at both the TAF9B distal peaks and at the TAF9B TSS peaks , suggesting that TAF9B and OLIG2 act in concert at a subset of neuronal genes to regulate gene expression during motor neuron development ( Figure 5C ) . The incomplete overlap of TAF9B and OLIG2 peaks may in part represent intrinsic differences between progenitor cells and post-mitotic neurons , since the OLIG2 binding was performed with neuralized EBs enriched for progenitor cells while TAF9B mapping was done with EBs enriched for post-mitotic motor neurons . In addition , it is likely that TAF9B co-localizes with activators other than OLIG2 in this developmental pathway . 10 . 7554/eLife . 02559 . 009Figure 5 . TAF9B binds promoter and distal regions of neuronal genes . ( A ) ChIP-seq analysis of TAF9B and RNA POL2 binding sites in EB-MN samples differentiated for 8 days . TAF9B and RNA POL2 ChIP-seq peaks examples are given for Tubb3 and Hes6 genes . ( B ) List of top GO terms Biological Process of genes associated to TAF9B distal peaks and TAF9B peaks at annotated transcription start sites ( TSS ) ranked by p-value . Association of ChIP-seq peak to annotated genes and GO analysis was performed using GREAT . ( C ) Co-localization of TAF9B distal peaks and TAF9B peaks at TSS with OLIG2 ChIP-seq peaks in EB-MN samples . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 00910 . 7554/eLife . 02559 . 010Figure 5—figure supplement 1 . TAF9B binds promoter and distal regions of neuronal genes . ( A ) TAF9B , OLIG2 , HOXC9 and RNA POL2 ChIP-seq peaks examples are given for Tubb3 , Mnx1 , Ngn2 and Hes6 genes . ( B ) Total numbers of peaks obtained by MACS in each ChIP-seq category . ( C ) ChIP-seq analysis of TAF9B and RNA POL2 binding sites in EB-MN samples relative to their distance to annotated transcription start sites ( TSS ) performed using GREAT . ( D ) Percentage of genes affected more than twofold in the absence of TAF9B as determined by RNA-seq analysis in Taf9b KO EB-MN differentiated for 6 days . Genes are grouped in different categories depending on the distance of the TAF9B peaks from their TSS . Genes associated to TAF9B-only peaks were used for the analysis . A gene list randomly generated was used as control . ( E ) Representative ChIP-qPCR analysis of TAF9B and RNA POL2 of selected loci identified by ChIP-seq . D = distal region , P = proximal promoter , NC = negative control . Bars represent mean ± SD of three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 010 Some TAFs are known to be part of not only the TFIID complex but also the SAGA complex ( Spt-Ada-Gcn5-acetyltransferase ) , which can mediate histone acetylation and deubiquitination ( reviewed by Spedale et al . , 2012 ) . In human HeLa cells , TAF9B had previously been reported to be a component of both the TFIID and SAGA complexes ( Frontini et al . , 2005 ) . To test whether TAF9B in mouse neurons is primarily associated with TFIID or SAGA-like complexes , we performed co-immunoprecipitation ( co-IP ) experiments with antibodies against TAF9B using in vitro differentiated motor neurons . We observed little detectable association of TAF9B with TAF4 and TAF7 , core components of the TFIID complex , while TBP antibodies co-IP'ed those TFIID components efficiently . In contrast , we observed an association between TAF9B and PCAF , a histone acetyltransferase present in the mammalian SAGA complex ( Figure 6A ) . Immunoprecipitations of 293T cells transfected with flag-tagged versions of TAF9B or TAF9 confirmed that these IPs are specific for TAF9B and not the closely related TAF9 subunit ( Figure 6B ) . We also performed TAF9B co-IP experiments on dissected spinal column tissue samples from WT and Taf9b KO newborn mice , confirming that the interactions between TAF9B and PCAF occur in vivo ( Figure 6C ) . Thus , TAF9B primarily associates with PCAF likely in a SAGA-like complex rather than the canonical TFIID in murine neurons . 10 . 7554/eLife . 02559 . 011Figure 6 . TAF9B is associated with PCAF . ( A ) TAF9B and TBP were immunoprecipitated from EB-MN cells differentiated for 8 days and co-immunoprecipitated proteins analyzed by western blotting using antibodies against the TFIID subunits TAF4 and TAF7 , and the histone acetyltransferase PCAF . ( B ) 293T cells were transfected with pCMV-3xFLAG-TAF9B or -TAF9 , immunoprecipitated with TAF9B antibodies , and analyzed by western blotting using anti-Flag antibodies . ( C ) TAF9B was immunoprecipitated from spinal column extracts from WT and Taf9b KO newborn mice and co-immunoprecipitated proteins were analyzed by western blotting detecting TAF4 , TBP and PCAF . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 011 To test the in vivo role of TAF9B during neuronal development , we generated a KO mouse using the Taf9b KO ES cells obtained from KOMP ( Figure 7—figure supplement 1A ) . Taf9b KO mice were viable and fertile , though the number of pups and birth weights were reduced in Taf9b KO matings compared to WT controls ( Figure 7—figure supplement 1C ) . As expected , spinal cord isolated from newborn KO animals did not express TAF9B ( Figure 7A , B , Figure 7—figure supplement 1D ) . Given that Taf9b KO cells in vitro showed clear de-regulation of genes in both progenitor cells and post-mitotic motor neurons ( Figure 4 ) , we surmised that loss of TAF9B in vivo may affect not only motor neuron formation but perhaps also proper differentiation of other cell types present in the spinal cord . To test whether neuronal gene expression in vivo is specifically affected in the absence of TAF9B , we carried out RNA-seq analysis to compare whole lumbar spinal column tissue dissected from newborn WT and KO littermates ( Figure 7C ) . Our data indicate that genes down-regulated in the absence of TAF9B are enriched for neuronal genes including gene categories such as neuron projection , synapse and axonogenesis ( Figure 7D ) . Among the down-regulated genes identified by RNA-seq were neuronal genes such as Map1b , L1cam , Nefm , Nefh , and Isl2 . Gene expression analysis by qRT-PCR of lumbar spinal column tissues confirmed that those neuronal genes are consistently down-regulated in KO compared to WT littermate controls ( Figure 7E ) . Several different markers of specific neuronal populations , as well as the general marker Tubb3 , were also affected in the absence of TAF9B , suggesting a global defect in neuronal gene expression in the spinal cord ( Figure 7—figure supplement 1E; Supplementary file 2 ) . These results are also consistent with a global reduction in neuronal tissue compared to surrounding non-neuronal tissue in the dissected spinal column preparations . Our data suggest that the observed pan-neuronal down-regulation and loss of neuronal tissue may be due to problems in the control of gene expression that coordinates the precise differentiation of progenitor cells during neuronal development . These in vivo results taken together with our in vitro differentiation studies strongly support the notion that TAF9B is specifically required for the regulation of neuronal genes during neuronal development in the spinal cord . 10 . 7554/eLife . 02559 . 012Figure 7 . TAF9B controls neuronal gene expression in vivo . ( A ) Western blot analysis of spinal cord tissue from WT and Taf9b KO newborn mice detecting TAF9B and ACTB as control . ( B ) Spinal cord tissue from WT and Taf9b KO newborn mice were dissected and analyzed by qRT-PCR for Taf9b expression . Graph shows mean ± SEM of three biological replicates . ( C ) RNA-seq analysis of dissected lumbar spinal column tissue from WT and Taf9b KO newborn mice . Scatter plot represents FPKM values of genes expressed in WT and Taf9b KO samples . Red dots are genes whose expression is affected more than twofold with p-values <0 . 05 . Gray dots represent genes not changing more than the selected cutoff . ( D ) GO term analysis of genes affected by the loss of TAF9B in RNA-seq analysis . List shows the top eight GO term Biological Process categories obtained ranked by p-value . ( E ) Gene expression analysis by qRT-PCR of lumbar spinal columns tissues of newborn mice . Graphs represent mean ± SEM of littermate comparisons ( WT n = 6 , KO n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 01210 . 7554/eLife . 02559 . 013Figure 7—figure supplement 1 . TAF9B controls neuronal gene expression in vivo . ( A ) Taf9b KO ES cells were used to generate a Taf9b chimeric mouse which was subsequently mated with WT to test for germline transmission and obtain Taf9b heterozygous mice . ( B ) Representative PCR-based DNA genotyping of Taf9b KO mice . Primers are described in Supplementary file 3 . ( C ) Number of pups and weight of newborn mice from WT × WT and KO × KO matings . ( D ) RNA-seq data of spinal cord sample for Taf9b locus in WT and Taf9b KO animals . ( E ) Gene expression analysis by qRT-PCR of lumbar spinal columns tissues of newborn mice . Graphs represent mean ± SEM of littermate comparisons ( WT n = 6 , KO n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02559 . 013 The control of tissue-specific gene expression has been largely attributed to the combinatorial action of transcriptional activators and repressors . These regulatory factors are thought to interact in concert with a highly conserved and prototypic set of core promoter recognition proteins and co-activators that form the PIC at gene promoters . However , this model has been challenged by different studies demonstrating changes that occur in the core promoter machinery in a tissue-specific manner ( reviewed by D'Alessio et al . , 2009; Goodrich and Tjian , 2010 ) . Despite the emerging roles that these core promoter components play in regulating cell-type specific gene expression , relatively little was known about possible changes that may occur in the composition of core promoter factors during neuronal differentiation . In this study , we screened for changes in the expression of core promoter factors including TAFs , general transcript factors ( GTFs ) , and mediator subunits ( MEDs ) upon neuronal differentiation . We found that most of the components in the mammalian PIC are expressed at relatively low levels in dissected spinal cord tissue compared to ES cells . Likewise , we observed down-regulation of several of these core promoter recognition components during in vitro differentiation of murine ES cells into motor neurons . By contrast , one orphan TAF , TAF9B , becomes highly up-regulated during motor neuron formation . Using a loss-of-function approach , we found that TAF9B is required for the efficient expression of neuronal genes upon in vitro motor neuron differentiation . Importantly , we found that TAF9B is dispensable for global gene expression in undifferentiated ES cells in contrast to the critical role that other TAFs in TFIID play in these cells ( Pijnappel et al . , 2013 ) . Using ChIP-seq , we found that TAF9B binds promoter and distal regulatory regions of neuronal genes co-localizing in a subset of regions with OLIG2 , a key regulator of motor neuron differentiation . Strikingly , Taf9b KO mice showed defects in the expression of neuronal genes examined by transcriptome analysis of whole spinal column tissues . Importantly , the mechanism by which TAF9B mediates neuronal-specific transcription programs is unlike all previous TAF involvement in cell specific function: rather than operating through TFIID or distal enhancer complexes ( Freiman et al . , 2001; Liu et al . , 2011; Zhou et al . , 2013b ) , it associates with a SAGA/PCAF co-activator complex . Our results and those of several other studies probing tissue-specific activities of core promoter factors and co-activators suggest a model in which changes in these complexes are integral to the transcriptional program leading to terminal differentiation . Different cell types employ distinct combinations of activators that must engage in specific protein–protein interactions with core promoter factors . We speculate that rather than existing as a static complex awaiting recruitment by these transcription factors , core promoter complexes themselves must undergo compositional changes using cell type-specific components such as TAF9B . These alternative promoter complexes expand the diversity and number of possible protein–protein transactions necessary to achieve proper fine-tuned control of gene expression . Furthermore , we speculate that such diversified core machineries might provide a mechanism to help lock down and maintain the functions and identity of differentiated cell types . We imagine that in a given cell type , if the silencing of an undesirable activator or repressor is imperfect , their specific interacting partners in the targeted core machinery may nevertheless not be present , thus serving as a fail safe mechanism to prevent unscheduled alterations in programs of gene transcription . We suspect that such core promoter complex dependent control and maintenance of cell identity would likely be accompanied by activator selectivity and integrated with changes in chromatin structure and other layers of regulation that must work coordinately to keep expression patterns stable yet responsive to different stimuli in terminally differentiated cells . We anticipate that further analysis of adult neuronal cell-types as well as other differentiated somatic cells will bring new insights regarding the use of alternative core machineries that regulate gene expression in normal and disease states . To start addressing these questions in the context of tissues in vivo , we generated Taf9b KO mice . We found that mice lacking TAF9B are viable , indicating that this orphan TAF is either not critical for normal development of cell-types other than neurons or its function is partly redundant with the closely related TAF9 homolog . We established that indeed Taf9 is highly expressed in both mouse ES cells and neurons ( as detected by RNA-seq ) . It is possible that TAF9 partially compensates for the loss of TAF9B thus blunting further global gene expression defects in Taf9b KO cells . In human HeLa and chicken DT40 cells partially redundant functions have been observed for these paralogs ( Chen and Manley , 2003; Frontini et al . , 2005 ) . It will be interesting to test whether Taf9 KO mouse are viable and whether the double KO of these two related genes ( Taf9 and Taf9b ) results in more pronounced phenotypes . No Taf9 KO mice have been reported to date . Taf4b and Taf7l KO mice are also viable , but unlike Taf9b KO mice , they are infertile ( Freiman et al . , 2001; Falender , 2005; Zhou et al . , 2013a ) . Our results suggest that TAF9B is not critical for gametogenesis , in contrast to what has been observed for these two other orphan TAFs . Instead , our results point to a model where TAF9B is involved in the activation of neuronal genes by binding to distal and promoter proximal DNA regulatory elements associated with the histone acetyltransferase PCAF that , like its paralog GCN5 , are subunits of the SAGA/STAGA/TFTC complex . This complex contains several TAFs shared with TFIID including TAF6 , TAF9 , TAF5 , TAF10 , TAF12 , and a de-ubiquitinase ( DUB ) module that removes ubiquitin from histone H2B ( reviewed by Spedale et al . , 2012; Weake and Workman , 2012 ) . Because TAFs are structural components of both SAGA and TFIID complexes , it is possible that TAF9B is affecting the competitive recruitment or activity of a SAGA-like complex in place of TFIID . Among the genes affected by the loss of TAF9B are several targets of the Notch- and Shh-signaling pathway , both known to play key roles in neuronal development ( reviewed by Louvi and Artavanis-Tsakonas , 2006; Jessell , 2000 ) . Interestingly , several lines of evidence connect the activation of Notch dependent genes with the PCAF-containing complex providing a potential link between PCAF , TAF9B , and activation of these genes . For example , PCAF has been shown to cooperate with another co-activator protein , p300 , in Notch intracellular domain-dependent transcription assays using chromatin templates in vitro , as well as in transfection/reporter assays ( Kurooka and Honjo , 2000; Wallberg et al . , 2002 ) . Other components of the SAGA complex have been linked to the specific control of gene expression in cells of the central nervous system . Mutations in two subunits of the Drosophila DUB complex , Nonstop and Sgf11 , lead to defects in neural development in the optic lobe ( Weake et al . , 2008 ) . In mice , mutations in the HAT domain of GCN5 result in defects in cranial neural tube closure and embryonic lethality ( Bu et al . , 2007 ) . In humans , mutations in the DUB subunit ATXN7 are associated with the development of spinocerebellar ataxia type 7 , a disease characterized by the loss of motor control as well as retinal defects ( reviewed by Koutelou et al . , 2010 ) . Interestingly , we observed that Taf9b KO mice display defects in eye development including microphthalmia and cataracts-like phenotypes albeit with modest penetrance ( data not shown ) . The type of neuronal-specific role performed by TAF9B is reminiscent of the cell-type specific roles of particular subunits of the ATP-dependent chromatin remodeling complex BAF . This complex undergoes changes in composition upon neuronal differentiation as specific BAF subunits are incorporated in neuronal progenitors cells as well as in post-mitotic neurons ( Olave et al . , 2002; Lessard et al . , 2007 ) . In this study , we have focused on characterizing the molecular phenotypes and have documented defects in gene regulation due to the absence of TAF9B during in vitro differentiation of motor neurons and in vivo in spinal column tissues . It is also possible that the down-regulation of neuronal genes observed in whole spinal column tissues in Taf9b KO mice may represent a specific loss of neuronal tissue relative to the surrounding vertebrae due perhaps to defects in neuronal progenitor cells during embryonic development . In future experiments , we hope to address the extent of potential neurological defects in Taf9b KO mice , including deficits in motor skills as well as other potential neurological defects due to the role that TAF9B may be playing in other neuronal populations besides motor neurons . Preliminary experiments using Taf9b KO mice carrying the LacZ reporter gene suggest that Taf9b is highly expressed in the developing hypothalamus at mid-gestation ( data now shown ) . In a recent study describing neuronal activity-dependent ribosome profiling in the adult mouse hypothalamus ( Knight et al . , 2012 ) , Taf9b can be identified in the genomic data set as highly expressed in this tissue compared to several other TAFs . Moreover , the levels of Taf9b mRNA associated with ribosomes increased when neurons in the hypothalamus were activated , while several other TAFs were expressed at low levels or not enriched in a neuronal activity-dependent manner . These results suggest that TAF9B may be involved in setting up transcriptional responses in at least certain neuronal circuits in the hypothalamus . In the future , it will be interesting to directly test the role of TAF9B in the adult hypothalamus and to extend the analysis to other adult neuronal types . Murine embryonic stem cells were grown in embryoMAX DMEM ( EMD Millipore , Billerica , MA ) supplemented with LIF and 15% FBS . Cells were differentiated into embryoid bodies enriched in motor neurons ( EB-MN ) as described previously ( Wichterle et al . , 2002 ) . 2 × 105 cells/ml were incubated in differentiation media ( 25% embryoMAX DMEM , 25% F12 media , 25% neurobasal media , 1x B27 supplement , 1 . 5 mM L-glutamine and 0 . 1 mM β-mercaptoethanol ) supplemented with 1 μM retinoic acid and 0 . 8 μM smoothened agonist SAG ( N-Methyl-N′-[3-pyridinylbenzyl]-N′-[3-chlorobenzo{b}thiophene-2-carbonyl]-1 , 4-diaminocyclohexane , VWR , Radnor , PA ) for up to 8 days . Samples were collected at different times as indicated . SHH conditioned media was generated by transfecting 293T cells with a vector expressing Shh-N and used when noted . The Notch signaling pathway inhibitor DAPT ( N-[N-{3 , 5-difluorophen- acetyl}-l-alanyl]-S-phenylglycine t-butyl ester , Tocris Bioscience , United Kingdom ) was added when noted to deplete progenitor cells and enrich for post-mitotic neurons in the culture ( Crawford and Roelink , 2007 ) . The Taf9b KO mouse ES cells ( Taf9btm1 ( KOMP ) Vlcg ) were generated by the trans-NIH Knock-Out Mouse Project ( KOMP ) and obtained from the KOMP Repository ( www . komp . org ) . NIH grants to Velocigene at Regeneron Inc ( U01HG004085 ) and the CSD Consortium ( U01HG004080 ) funded the generation of gene-targeted ES cells for 8500 genes in the KOMP Program and archived and distributed by the KOMP Repository at UC Davis and CHORI ( U42RR024244 ) . We confirmed the described deletion by DNA sequencing ( data not shown ) . Since the Taf9b gene is located in the X chromosome this male KO ES cells are a complete null for this gene . JM8 WT murine ES cells are from the same genetic background as Taf9b KO ES cells ( C57BL/6 ) . JM8 and Taf9b KO ES cells were grown on MEF feeder layers inactivated with mitomycin C ( Sigma-Aldrich , St . Louis , MO ) . Murine ES cells carrying a GFP reporter under Mnx1 promoter ( HBG3 ) were described previously ( Wichterle et al . , 2002 ) . Total RNA was isolated using RNeasy kit ( Qiagen , Germany ) and cDNA was generated using Superscript III RT system using manufacture's instructions ( Life Technologies , Carlsbad , CA ) . Primers used in these assays are described in Supplementary file 3 . Relative expression levels in in vitro time course differentiation experiments were normalized to total RNA . Relative levels of expression of spinal column tissues were normalized to Gapdh . PCR reactions were performed using SYBR Green PCR Master Mix according to the manufacturer's instructions in an ABI 7300 real time PCR machine ( Applied Biosystems , Grand Island , NY ) . Taf9b and Taf9 cDNAs were obtained from Open Biosystems clones ( TAF9B clone ID#30468965 , TAF9 clone ID#5006430 ) and cloned into p3xFLAG-CMV-10 vector to obtain p3xFLAG-CMV-TAF9B and -TAF9 . Plasmids were transfected into 293T cells by lipofectamine 2000 ( Life Technologies ) . Cells were collected 48 hr after transfection . Transfected 293T cells were lysed and the protein concentration was measured by the Bradford method . EB-MN differentiated for 8 days were collected and nuclear extract was prepared as previously described ( Pugh , 1995 ) . Spinal column tissues were dissected from newborn mice , immediately frozen in liquid nitrogen , and stored at −80°C . Tissue samples were ground into powder in liquid nitrogen and cell lysis buffer was added to extract proteins . After centrifugation , the protein concentration was measured by the Bradford method . Indicated antibodies were mixed with protein A sepharose ( GE Healthcare life science , Pittsburgh , PA ) or protein G Dynabeads ( Life technologies ) for 1 hr at 4°C . Protein extracts ( 1 mg ) were mixed with the antibodies/beads complexes for overnight at 4°C on a rotating wheel . The beads were washed three times with washing buffer containing 300 mM KCl and 0 . 05% NP-40 and once with washing buffer containing 100 mM KCl . Samples were boiled in SDS loading buffer for 5 min and analyzed by Western blotting using indicated antibodies . Western blots were performed using whole cell extracts and SDS-PAGE . Immunostainings of EB-MN were performed after fixing samples in 4% paraformaldehyde for 30 min at room temperature . Antibodies were incubated in PBS , 0 . 1% Triton X-100 and washed three times with PBS , 0 . 1% Triton X-100 for 10 min each time . Antibodies used are anti-TBP ( cat#51841; Abcam , Cambridge , Massachusetts ) , anti-TAF4 ( cat#612054; BD Biosciences , San Jose , CA ) , anti-TAF7 ( cat#H00006879-M01; Abnova , Taiwan ) , anti-TAF9B ( cat#A303-810A; #G2306 and Bethyl Laboratories , Montgomery , TX ) , anti-TAF10 ( Santa Cruz cat#102125 ) , anti-OCT4 ( Santa Cruz cat#8628 , Dallas , Texas ) , anti-TUBB3 ( cat#MMS435P; Covance , Princeton , New Jersey ) , anti-ACTB ( cat#A2228; Sigma-Aldrich ) , anti-RNA POL2 ( 8WG16; Clone ) , anti-PCAF ( cat#13124; Santa Cruz ) , anti-ISL1/2 ( Gift from the Jessell Lab , Columbia University ) , anti-NANOG ( cat#A300-397A; Bethyl Laboratories ) , anti-Ki67 ( cat#16667; Abcam ) , anti-FLAG ( cat#F3165; Sigma-Aldrich ) and Rabbit IgG control ( cat#46540; Abcam ) . TAF9B antibodies #G2306 were produced by injecting a peptide corresponding to residues 226–245 of TAF9B into rabbit by OpenBiosystems ( GE Healthcare ) . Antiserum was tested for specificity using extracts from 293T cells transfected with p3xFLAG-CMV-TAF9B and -TAF9 . Cell cycle was measured using BrdU Flow kit following manufacturer's instructions ( cat#552598; BD Biosciences ) . Apoptosis was measured using PE annexin V apoptosis detection kit following manufacturer's instructions ( cat#559763; BD Biosciences ) . Total RNA was isolated from cells or tissue extracts using RNeasy Kit ( Qiagen ) and 1 . 5 μg of total RNA was used to prepared Poly-A RNA-seq libraries following Illumina protocols . Samples were sequenced using Illumina Sequencers at the QB3 Vincent J Coates Genomics Sequencing Laboratory , UC Berkeley . Sequenced reads were mapped against RefSeq genes using TopHat ( Trapnell et al . , 2009 ) and differentially expressed genes were determined using Cuffdiff ( Trapnell et al . , 2010 ) . Genes expressed more than 1 FPKM and with >twofold difference and p-value < 0 . 05 as calculated by Cuffdiff were considered as differentially expressed for scatter plots representations and for GO term analysis using DAVID Bioinformatic resources ( Huang et al . , 2009 ) . ChIP-seq was performed using 1 mg of chromatin for each IP reaction using antibodies against TAF9B ( #G2306 ) , RNA POL2 ( 8WG16 ) , and IgG control with EB-MN differentiated for 8 days . EB-MN samples were fixed with 0 . 5% formaldehyde for 10 min at room temperature and ChIPs were performed as described previously ( Liu et al . , 2011; Zhou et al . , 2013b ) . ChIP-qPCR experiments were performed using EB-MN samples differentiated for 6–8 days . ChIP-seq libraries were prepared following Illumina protocols and sequenced at the QB3 Vincent J Coates Genomics Sequencing Laboratory , UC Berkeley . Reads were mapped to UCSC version mm9 using Bowtie ( Langmead et al . , 2009 ) . ChIP-seq peaks were determined using MACS ( Zhang et al . , 2008 ) and significant peaks were associated to genes using GREAT default parameters ( McLean et al . , 2010 ) . All animal experiments were performed in strict accordance with the recommendations in the guide for the care and use of laboratory animals of the National Institutes of Health and following the animal use protocol ( #R007 ) approved by the Animal Care and Use Committee ( ACUC ) of the University of California , Berkeley . Taf9b KO mouse was generated in the UC Berkeley transgenic facility by injecting Taf9b KO murine ES cells ( TAF9btm1 ( KOMP ) Vlcg ) into albino C57BL/6 background ( C57BL/6J-Tyrc-2J ) and chimeric mice were obtained . Taf9b heterozygous animals were backcrossed to WT C57BL/6 mice to obtain Taf9b heterozygous mice without the albino phenotype ( Tyrc-2J ) . Mice were genotyped by PCR using primers detecting WT and KO sequences . Primers sequences are available in Supplementary file 3 .	Almost all the cells in an organism contain the same genetic information , but they develop into many different types of cells that perform a variety of specialized functions in the body . Brain cells , for example , have a very different shape and function from red blood cells . A small group of proteins act inside cells to switch on the expression of genes it needs to carry out the specific functions of a given cell-type , and switch off the genes that are only needed in other cell types . Some of these regulatory proteins called ‘core promoter factors’ bind to the DNA near the start of genes . These core factors are known to work in combination with various other proteins to switch genes on or off in specific cell types . However , the specific core promoter factors and partner proteins that guide a cell into becoming a neuron have not been well characterized . Now , Herrera et al . have identified a core promoter factor called TAF9B that is produced at higher levels when mouse stem cells are coaxed into becoming the motor neurons that carry nerve impulses to muscles . The TAF9B protein works together with an enzyme ( called PCAF ) to help to switch on the genes that control the development of these cells . Without this regulatory protein , mouse stem cells grown in the lab fail to properly switch on the genes that are necessary to become motor neurons . These mutant stem cells also fail to efficiently switch off genes that stop stem cells from becoming more specialized . High levels of TAF9B were also found in the spinal cord of newborn mice and when Herrera et al . engineered mice that lack TAF9B , these mice did not properly regulate the expression of neuronal genes in their spines . These new findings might , in the future , improve our ability to guide stem cells into forming neurons , or to reprogram other types of specialized cells into becoming motor neurons . This new information could also prove useful for researchers interested in better understanding neuronal development and might aid in the design of therapies to treat neuronal injuries or diseases , such as motor neuron disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2014	Core promoter factor TAF9B regulates neuronal gene expression
Gene transcription can be activated by decreasing the duration of RNA polymerase II pausing in the promoter-proximal region , but how this is achieved remains unclear . Here we use a ‘multi-omics’ approach to demonstrate that the duration of polymerase pausing generally limits the productive frequency of transcription initiation in human cells ( ‘pause-initiation limit’ ) . We further engineer a human cell line to allow for specific and rapid inhibition of the P-TEFb kinase CDK9 , which is implicated in polymerase pause release . CDK9 activity decreases the pause duration but also increases the productive initiation frequency . This shows that CDK9 stimulates release of paused polymerase and activates transcription by increasing the number of transcribing polymerases and thus the amount of mRNA synthesized per time . CDK9 activity is also associated with long-range chromatin interactions , suggesting that enhancers can influence the pause-initiation limit to regulate transcription . Transcription in metazoan cells is often regulated at the level of promoter-proximal pausing ( Core et al . , 2008; Day et al . , 2016; Henriques et al . , 2013; Nechaev et al . , 2010; Rougvie and Lis , 1988; Strobl and Eick , 1992 ) , which can be detected by measuring the occupancy with paused Pol II by ChIP-seq ( Johnson et al . , 2007 ) , GRO-seq ( Core et al . , 2008 ) , ( m ) NET-seq ( Mayer et al . , 2015; Nojima et al . , 2015 ) , or PRO-seq ( Kwak et al . , 2013 ) . Genes with paused Pol II are conserved across mammalian cell types and states ( Day et al . , 2016 ) . The mechanisms underlying how Pol II pausing can regulate RNA transcript synthesis remain unclear . Transcription of a human protein-coding gene of average length takes at least half an hour to be completed . The duration of pausing however lies in the range of minutes ( Jonkers et al . , 2014 ) and does not considerably change the overall time it takes to complete a transcript . Thus , how can changes in the pause duration lead to synthesis of a different number of RNA transcripts per time ? It has been suggested that a decreased pause duration goes along with a higher initiation frequency , because occupancy peaks for promoter-proximal Pol II can increase upon gene activation ( Boehm et al . , 2003 ) or can remain high even when pausing is impaired ( Henriques et al . , 2013 ) . The height of Pol II occupancy peaks however cannot directly inform on initiation frequency or pause duration because it depends not only on the number of polymerases that pass the pause site but also on their residence time ( Ehrensberger et al . , 2013 ) . A kinetic model of transcription predicted that pause duration delimits the initiation frequency and suggested that paused Pol II sterically interferes with initiation ( Ehrensberger et al . , 2013 ) . Indeed , modeling reveals that a paused polymerase positioned up to around 50 bp downstream of the TSS could sterically interfere with formation of the Pol II initiation complex ( Figure 1—figure supplement 1 ) . Even if a paused polymerase is located further downstream , it may still interfere with initiation if one or more additional elongating polymerases line up behind it . The critical relationship between pausing and initiation could thus far not be tested experimentally , as no methods were available to measure initiation frequencies . A recently developed method , transient transcriptome sequencing ( TT-seq ) ( Schwalb et al . , 2016 ) , now allows to unveil the flow of polymerases as it measures local RNA synthesis rates genome-wide at nucleotide resolution . Here we investigate whether changes in pause duration alter initiation frequency in living cells . We specifically inhibit the kinase CDK9 , which facilitates Pol II pause release ( Laitem et al . , 2015; Marshall and Price , 1992; Peterlin and Price , 2006 ) , and monitor RNA synthesis and initiation frequencies by TT-seq . A combination of TT-seq data with mNET-seq data allows us to derive pause durations for active genes . We conclude that the duration of pausing can control transcription initiation at human genes , and derived determinants for CDK9-dependent pause release and initiation activation . To specifically inhibit CDK9 , we used a chemical biology approach ( Lopez et al . , 2014 ) that circumvents off-target effects of standard CDK9 inhibitors ( Morales and Giordano , 2016 ) . We introduced a CDK9 analog sensitive mutation ( CDK9as ) into human Raji B cells by CRISPR-Cas9 ( Materials and methods , Figure 1—figure supplement 2A–B ) . This allows for rapid and highly specific CDK9 inhibition with the adenine analog 1-NA-PP1 ( Lopez et al . , 2014 ) , which does not have any effect on wild type cells ( Figure 1—figure supplement 2C ) . CDK9 protein levels were unchanged in CDK9as mutant cells compared to wild type cells ( Figure 1—figure supplement 2D ) . After 72 hr of incubation with 1-NA-PP1 , growth of CDK9as cells ceased , whereas wild type cells grew normally ( Figure 1—figure supplement 2E ) . We treated CDK9as cells with 5 μM of 1-NA-PP1 for 10 min and monitored changes in RNA synthesis by TT-seq ( Schwalb et al . , 2016 ) , using a RNA labeling time of 5 min ( Figure 1A ) . TT-seq data were highly reproducible ( Spearman correlation coefficient 1 ) and monitored transcription activity before and after CDK9 inhibition ( Figure 1B ) . CDK9 inhibition resulted in reduced TT-seq signals at the beginning of genes , indicating that less Pol II was released into gene bodies ( Figure 1B , Figure 2—figure supplement 1A–B ) . This gave rise to a ‘response window’ revealing the distance traveled by Pol II during 10 min inhibitor treatment ( Figure 1C ) . Downstream of the response window , the TT-seq signal was largely unchanged , indicating continued RNA synthesis from Pol II elongation complexes that had been released before CDK9 inhibition . To determine the relative response of genes to CDK9 inhibition , we calculated response ratios for those transcribed units ( TUs , Materials and methods ) that synthesized RNA , harbored a single TSS , and exceeded 10 kbp in length ( 2 , 538 TUs ) . The response ratio of TUs varied between 0% to 100% ( fully responding TUs ) with a median of 58% ( Figure 1C–E ) . A remaining TT-seq signal in the response window likely reflects the proportion of polymerases that move to productive elongation without CDK9 kinase activity , but we cannot exclude that it stems from incomplete CDK9 inhibition . However , based on the assumption that the inhibitor is evenly distributed across cells and within , the portion of CDK9 that has not been fully inhibited must be very low . The width of the response window differs between TUs ( Figure 1D ) and informs on Pol II elongation velocity ( Materials and methods ) . The average width of the response window was 23 kbp , and thus the average elongation velocity was 2 . 3 kbp/min ( Figure 2A–B ) , which agrees with previous estimates ( Fuchs et al . , 2014; Jonkers et al . , 2014; Saponaro et al . , 2014; Veloso et al . , 2014 ) . Gene-specific elongation velocities ( Figure 2C , Figure 2—figure supplement 1A–B ) were significantly higher in TUs with longer first introns ( Figure 2D , Wilcoxon rank sum test , p-value<1 . 916·10−11 ) , consistent with faster transcription of introns ( Jonkers et al . , 2014 ) . Elongation velocity correlated positively with nucleosome density , and negatively with the stability of the DNA-RNA hybrid , CpG density and topoisomerase occupancy ( Figure 2—figure supplement 1C ) . To study the kinetics of CDK9-dependent Pol II pause release , we generated mNET-seq data that map the RNA 3’-end of engaged Pol II and extracted the position of paused polymerases ( Materials and methods ) . mNET-seq data were highly reproducible ( Spearman correlation coefficient 0 . 93 ) . Of the above TUs , 2135 ( 84 % ) showed mNET-seq signal peaks above background ( Materials and methods ) . The called pause sites were distributed around a maximum located ~84 bp downstream of the TSS ( Figure 3A , Figure 3—figure supplement 1A ) . At these sites we detected an enrichment for G/C-C/G dinucleotides ( Figure 3—figure supplement 1B ) with a strongly conserved cytosine at the RNA 3’-end ( Figure 3B ) . We also observed a minimum of the predicted melting temperature of the DNA-RNA hybrid ( Materials and methods ) immediately downstream of the pause site ( Figure 3C ) . A weak DNA-RNA hybrid in the active center of Pol II is known to destabilize the elongation complex ( Kireeva et al . , 2000 ) , and could be a major determinant for establishing the paused state . To quantify pausing , we defined the pause duration d as the time a polymerase needs to pass through a 200 bp ‘pause window’ located ±100 bp around the pause site . The pause duration d can now be derived from a combination of mNET-seq and TT-seq data . In particular , the mNET-seq signal corresponds to the number of polymerases in the pause window , which is determined by d and by the initiation frequency I ( Figure 4A ) ( Ehrensberger et al . , 2013 ) . Thus , d is proportional to the ratio of the mNET-seq signal over I . To calculate I we integrated TT-seq signals over exons , excluding the first exon ( Materials and methods ) . This provides the ‘productive initiation frequency’ , that is the number of polymerases that initiate and successfully exit from the pause window . We use the term ‘productive’ because we do not know whether there is a small fraction of polymerases terminating within the pause window . Finally , to derive absolute values of d , we scaled the reciprocal of d ( the elongation velocity in the pause window ) according to the elongation velocity obtained from CDK9 inhibition ( Materials and methods ) . We obtained a mean productive initiation frequency of 2 . 7 polymerases cell−1min−1 , and pause durations in the range of minutes , with strong variations between TUs . The pause durations are generally consistent with reported half-lives of paused Pol II in mouse ( Jonkers et al . , 2014 ) and Drosophila cells ( Buckley et al . , 2014; Henriques et al . , 2013 ) but slightly shorter . Pause durations were also consistent with kinetic modeling of TT-seq data alone . At TUs with long pause durations we observed less labeled RNA in the short region between the TSS and the pause site ( Figure 4—figure supplement 1 ) . This confirms that indeed initiation frequencies are altered . It also indicates that the fraction of Pol II enzymes that terminate within the pause window is low , in agreement with previous findings ( Henriques et al . , 2013 ) . For strongly CDK9-responding TUs , we obtained a significantly longer pause duration ( Wilcoxon rank sum test , p-value<10−12 ) and lower initiation frequencies ( Figure 4B–C ) . These results prompted us to ask whether the pause duration is generally related to the initiation frequency . We indeed found a robust anti-correlation between I and d in normally growing cells , and an upper boundary for combinations of I and d which we call ‘pause-initiation limit’ . ( Figure 4D , Figure 4—figure supplement 2A ) . Thus , genes with shorter pausing show higher initiation frequencies and more RNA synthesis . This fundamental relationship can be verified by calculating the pause duration d without the initiation frequency I , d^ ( Materials and methods , Figure 4—figure supplement 2B–C , E ) . Repeated random shuffling of mNET-seq signal assignment to TUs abolishes the correlation between d^ and I ( Figure 4—figure supplement 2D ) . It also shows that the observation of impossible combinations of pause duration d^ and initiation frequency I ( points above ‘pause-initiation limit’ ) are minimal ( Figure 4—figure supplement 2F ) . In conclusion , independent mNET-seq and TT-seq data led to independent measures of pause duration and productive initiation frequency for each gene , which were then observed to be globally anti-correlated . These findings now allowed us to test directly whether longer pause durations lead to lower initiation frequencies , by analyzing TT-seq data after CDK9 inhibition . CDK9 inhibition resulted in significantly reduced labeled RNA in the short region between the TSS and the pause site ( Wilcoxon rank sum test , p-value<10−16 ) ( Figure 5A–B ) . Productive initiation frequencies were significantly downregulated after CDK9 inhibition ( Wilcoxon rank sum test , p-value<10−16 ) ( Figure 5C ) . Because CDK9 specifically targets paused Pol II , and not initiating polymerase , these results show that pausing limits initiation , and not the other way around . Thus , human genes have a ‘pause-initiation limit’ . To monitor the occupancy of engaged Pol II we generated mNET-seq data before and after CDK9 inhibition ( Materials and methods ) . CDK9 inhibition resulted in increased mNET-seq signal at the beginning of genes and decreased signal in the gene body , indicating that less Pol II was released from the pause site ( Figure 6A ) . Indeed , calculation of pause durations from mNET-seq and TT-seq data after CDK9 inhibition showed that Pol II resides significantly longer at the pause site after CDK9 inhibition ( Wilcoxon rank sum test , p-value<10−16 ) ( Figure 6B ) . Taken together , CDK9 inhibition increases the pause duration and decreases the initiation frequency at human genes ( Figure 6C–D ) . To investigate possible reasons for polymerase pausing and its consequences , we compared different properties of TUs with long and short pause durations . For the 5’-region of TUs with longer pause durations , the transcript adopts more RNA secondary structure in vivo and in silico ( Wilcoxon rank sum test , p-value<10−16 ) ( Figure 7A , Figure 7—figure supplement 1A ) ( Rouskin et al . , 2014 ) . TUs with longer pause durations were also enriched for hyper-methylated CpG islands ( ENCODE Project Consortium , 2012 ) upstream of the pause site ( Figure 7B ) , consistent with a previous report ( Hendrix et al . , 2008 ) . Comparison of strongly and weakly CDK9-responding TUs around the pause site showed that TUs that responded strongly to CDK9 inhibition showed a higher tendency to establish long-range chromatin interactions ( Figure 7C ) as observed by Hi-C ( Ma et al . , 2015 ) . This is consistent with the idea that interactions of an enhancer with its target promoter can stimulate Pol II pause release ( Ghavi-Helm et al . , 2014; Rahl et al . , 2010 ) . This tendency however seems to be independent of the pause duration as comparing TUs with long and short pause durations leads to no observable difference in Hi-C signal . Finally , we investigated which factors preferentially occupy pause windows with longer pause durations . This is now possible because ChIP-seq signals can be normalized with the productive initiation frequency . Without such normalization , ChIP-seq derived factor occupancies are artificially high in pause windows with long pause durations ( Ehrensberger et al . , 2013 ) . Correlation of such normalized ChIP-seq signals in the pause window with pause durations ( Figure 7—figure supplement 1B–C ) resulted in a positive correlation for Pol II phosphorylation at sites that are associated with elongation , and also for NELF-E , CDK9 , and Brd4 , which are all factors involved in Pol II pausing and release . Taken together , our results show that Pol II pausing can control transcription initiation and demonstrate the central role of CDK9 in controlling pause duration and thereby the productive initiation frequency . Our results have implications for understanding gene regulation . Genes that show initiation frequencies below the pause-initiation limit may be activated by increasing the initiation frequency without changing pause duration . However , activation of genes that are transcribed at the pause-initiation limit requires a decrease in pause duration , that is stimulation of pause release , to enable higher initiation frequencies . We suggest that pause-controlled initiation evolved because mutations in the promoter-proximal region can change pause duration , and thereby limit initiation , but do not compromise a high initiation capacity of the core promoter around the TSS . This may have enabled the evolution of genes that remain highly inducible but can be efficiently downregulated . After our work had been completed , a publication appeared that concluded that polymerase pausing inhibits new transcription initiation ( Shao and Zeitlinger , 2017 ) . The conclusion in this paper is consistent with our general finding of an interdependency of Pol II pausing and transcription initiation , but the two studies differ in three aspects . First , we used human cells whereas the published work was conducted in Drosophila cells . Second , our work uses a multi-omics approach to enable a kinetic description , whereas the published work is based on changes in factor occupancy . Third , we selectively inhibited CDK9 using CRISPR-Cas9-based engineering and chemical biology , whereas the published work used small molecule inhibitors that may target multiple kinases . Despite these differences , the general conclusion that promoter-proximal pausing of Pol II sets a limit to the frequency of transcription initiation holds for both human and Drosophila cells and is likely a general feature of metazoan gene regulation . Raji B cells were obtained from DSMZ ( DSMZ no . : ACC 319; RRID:CVCL_0511 ) . CDK9as Raji B cells were generated in this study by CRISPR-Cas9-based engineering of Raji B cells obtained from DSMZ ( DSMZ no . : ACC 319; RRID:CVCL_0511 ) . Raji B cells and CDK9as Raji B cells were grown in RPMI 1640 medium ( Thermo Fisher Scientific , Waltham , MA USA ) supplemented with 10% foetal calf serum ( bio-sell , Nürnberg , Germany ) , 100 U/mL penicillin and 100 µg/mL streptomycin ( Thermo Fisher Scientific , Waltham , MA USA ) , and 2 mM L-glutamine ( Thermo Fisher Scientific , Waltham , MA USA ) at 37°C and 5% CO2 . Cells were verified to be free of mycoplasma contamination using Plasmo Test Mycoplasma Detection Kit ( InvivoGen , San Diego , CA USA ) . CDK9as contains a point mutation of the so-called gatekeeper residue that enables the kinase active site to accept bulky ATP analogs ( 1-NA-PP1 ) ( 4-Amino-1-tert-butyl-3- ( 1ʹ-naphthyl ) pyrazolo[3 , 4-d]pyrimidine ) . To identify the gatekeeper residue ( Lopez et al . , 2014 ) , the amino acid sequence of the human CDK9 kinase ( UniProt , P50750-1 ) was aligned with sequences of previously characterized kinases carrying analog sensitive mutations . Multiple sequence alignment was performed with the web tool Clustal Omega 1 . 2 . 4 ( Sievers et al . , 2011 ) . For the canonical isoform of CDK9 , phenylalanine ( F ) 103 was identified as the gatekeeper residue and selected for mutation to alanine ( A ) . Mutation of F103 at the CDK9 gene loci in Raji B cells was performed using the CRISPR-Cas9 system ( Doudna and Charpentier , 2014; Hsu et al . , 2014 ) as described ( Mulholland et al . , 2015 ) with minor modifications . Briefly , the single guide RNA ( sgRNA ) for editing CDK9 was designed by using the web tool Optimized CRISPR design ( http://crispr . mit . edu/ ) , and was incorporated to pSpCas9 ( BB ) −2A-GFP ( PX458 ) vector by BpiI restriction sites ( Addgene plasmid # 48138 ) ( Ran et al . , 2013 ) . For nucleotide replacement ( gttc to cgcg ) , 200 nt single-stranded DNA oligonucleotides ( ssODNs ) were synthesized by Integrated DNA Technologies ( IDT , Leuven , Belgium ) and used as homology-directed repair ( HDR ) template . A BstUI cutting site was incorporated into the HDR template for screening . The vector and HDR template were introduced into human Raji B cells using Amaxa Mouse ES Cell Nucleofector Kit ( Lonza , Basel , Switzerland ) according to the manufacturer’s instructions . Two days after transfection , GFP positive cells were single cell sorted into 96 well plates using FACS Aria II instrument ( Becton Dickinson , Franklin Lakes , NJ USA ) . After two weeks , individual colonies were expanded for genomic DNA isolation . The mutant lines were validated by PCR using respective primers , BstUI digestion ( Figure 1—figure supplement 2A ) and DNA sequencing ( Figure 1—figure supplement 2B ) . HDR template ( A103 is underlined , BstUI cutting site in small letters ) : AAAGTGTGTTGGGTGTGGTTTTCTTGACTTTTTCTTCTTTCTATTCCTGCCTCAGCTTCCCCCTATAACCGCTGCAAGGGTAGTATATACCTGGTcgcgGACTTCTGCGAGCATGACCTTGCTGGGCTGTTGAGCAATGTTTTGGTCAAGTTCACGCTGTCTGAGATCAAGAGGGTGATGCAGATGCTGCTTAACGGCCT Primers for sgRNA generation and screening: CDK9-sgRNA-F: 5’-CACCGGCTCGCAGAAGTCGAACACC-3’ CDK9-sgRNA-R: 5’-AAACGGTGTTCGACTTCTGCGAGCC-3’ CDK9-screen-F: 5’-CCCCGTAGCTGGTGCTTCCTCG-3’ CDK9-screen-R: 5’-CCCCAGCAGCCTTCATGTCCCTAT-3’ Proteins equivalent to 1 × 105 Raji B cells were loaded in Laemmli buffer and subjected to SDS-PAGE before transfer to nitrocellulose . Unspecific binding of antibodies was blocked by incubation of the membrane with 5% milk in Tris-buffered saline containing 1% Tween . Primary antibodies were anti-CDK9 ( sc-484 ) ( Santa Cruz , Dallas , TX USA ) and anti-α-tubulin ( DM1A ) ( Sigma-Aldrich , St . Louis , MO USA ) . Fluorophore-coupled secondary antibodies ( Rockland Immunochemicals Inc . , Pottstown , PA USA ) were used and blots were visualized using the Odyssey system ( LI-COR , Lincoln , NE USA ) . Cell proliferation at increasing 1-NA-PP1 inhibitor concentrations was measured in four biological replicates using the CellTiter 96 AQueous One Solution Cell Proliferation Assay System ( Promega , Madison , WI USA ) . Cells were seeded in a 96-well plate and increasing concentrations of 1-NA-PP1 ( Calbiochem , EMD Millipore , Danvers , MA USA ) or DMSO ( Sigma-Aldrich , St . Louis , MO USA ) were added . After 72 hr , MTS tetrazolium compound was added to each well for one hour . Subsequently , the quantity of the MTS formazan product was measured as absorbance at 490 nm with a Sunrise photometer ( TECAN , Männedorf , Switzerland ) that was operated using the Magellan data analysis software ( v7 . 2 , TECAN , Männedorf , Switzerland ) . Relative signals for each concentration were calculated by dividing the signals of the CDK9as inhibitor treated cells by the corresponding signals of the control . Two biological replicates of reactions including RNA spike-ins were performed essentially as described ( Schwalb et al . , 2016 ) . Briefly , 3 . 3 × 107 Raji B ( CDK9as or wild type ) cells were treated for 15 min with solvent DMSO ( control ) or 5 µM of 1-NA-PP1 ( CDK9as inhibitor ) . After 10 min of treatment , labeling was performed by adding 500 µM of 4-thiouracil ( 4sU ) ( Sigma-Aldrich , St . Louis , MO , USA ) for 5 min at 37°C and 5% CO2 . Cells were harvested by centrifugation at 3000 x g for 2 min . Total RNA was extracted using QIAzol according to the manufacturer’s instructions . RNAs were sonicated to generate fragments of <1 . 5 kbp using AFAmicro tubes in a S220 Focused-ultrasonicator ( Covaris Inc . , Woburn , MA USA ) . 4sU-labeled RNA was purified from 150 µg total fragmented RNA . Separation of labeled RNA was achieved with streptavidin beads ( Miltenyi Biotec , Bergisch Gladbach , Germany ) as described in ( Schwalb et al . , 2016 ) . Prior to library preparation , 4sU-labeled RNA was purified and quantified . Enrichment of 4sU-labeled RNA was analyzed by RT-qPCR as described ( Schwalb et al . , 2016 ) . Input RNA was treated with HL-dsDNase ( ArcticZymes , Tromsø , Norway ) and used for strand-specific library preparation according to the Ovation Universal RNA-Seq System ( NuGEN , Leek , The Netherlands ) . The size-selected and pre-amplified fragments were analyzed on a Fragment Analyzer before clustering and sequencing on the Illumina HiSeq 1500 . Paired-end 50 base reads with additional 6 base reads of barcodes were obtained for each of the samples , that is two TT-seq replicates with 1-NA-PP1 ( CDK9as inhibitor ) and two TT-Seq replicates with DMSO ( control ) treatment . Reads were demultiplexed and mapped with STAR 2 . 3 . 0 ( Dobin and Gingeras , 2015 ) to the hg20/hg38 ( GRCh38 ) genome assembly ( Human Genome Reference Consortium ) . Samtools ( Li et al . , 2009 ) was used to quality filter SAM files , whereby alignments with MAPQ smaller than 7 ( -q 7 ) were skipped and only proper pairs ( -f2 ) were selected . Further data processing was carried out using the R/Bioconductor environment . We used a spike-in ( RNAs ) normalization strategy essentially as described ( Schwalb et al . , 2016 ) to allow observation of global shifts and antisense bias determination ( ratio of spurious reads originating from the opposite strand introduced by the RT reactions ) . Read counts for spike-ins were calculated using HTSeq ( Anders et al . , 2015 ) . Sequencing depth calculations did not detect global differences . Antisense bias ratios were calculated for each sample j according tocj=mediani⁡kijantisensekijsense for all available spike-ins i . For each annotated gene , transcription units ( TUs ) were defined as the union of all existing inherent transcript isoforms ( UCSC RefSeq GRCh38 ) . Read counts for all features were calculated using HTSeq ( Anders et al . , 2015 ) and corrected for antisense bias using antisense bias ratios cj calculated as described above . The real number of read counts sij for transcribed unit i in sample j was calculated assij=Sij-cjAij1-cj2 where Sij and Aij are the observed number of read counts on the sense and antisense strand . Read counts per kilobase ( RPK ) were calculated upon bias corrected read counts falling into the region of a transcribed unit divided by it’s length in kilobases . Based on the antisense bias corrected RPKs a subgroup of expressed TUs was defined to comprise all TUs with an RPK of 100 or higher in two summarized replicates of TT-seq without inhibitor treatment . An RPK of 100 corresponds to approximately a coverage of 10 per sample due to an average fragment size of 200 . This subset was used throughout the analysis unless stated otherwise . Aligned duplicated fragments were discarded for each sample . Of the resulting unique fragment isoforms only those were kept that exhibited a positive inner mate distance . The number of transcribed bases ( tbj ) for all samples was calculated as the sum of the coverage of evident ( sequenced ) fragment parts ( read pairs only ) for all fragments smaller than 500 bases in length and with an inner mate interval not entirely overlapping a Refseq annotated intron ( UCSC RefSeq GRCh38 , ~96% of all fragments ) in addition to the sum of the coverage of non-evident fragment parts ( entire fragment ) . We first checked that no significant global shifts were detected in a comparison of two TT-seq replicates with 1-NA-PP1 ( CDK9as inhibitor ) treatment against two TT-seq replicates with DMSO treatment ( control ) in the above described spike-ins normalization strategy . Then all samples were subjected to an alternative , more robust normalization procedure . For each sample j the antisense bias corrected number of transcribed bases tbj was calculated on all expressed TUs i exceeding 125 kbp in length . 50 kbp were truncated from each side of the selected TUs to avoid influence of the response to CDK9as inhibition ( Laitem et al . , 2015 ) . On the resulting intervals , size factors for each sample j were determined asσj=mediani⁡tbij∏v=1mtbij1/m where m denotes the number of samples . This formula has been adapted ( Anders and Huber , 2010 ) and was used to correct for library size and sequencing depth variations . For each condition j ( control or CDK9as inhibited ) the antisense bias corrected number of transcribed bases tbij was calculated on all expressed TUs i exceeding 10 kbp in length . Of all remaining TUs only those were kept harboring one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) . Response ratios were calculated for a window from the TSS to 10 kbp downstream ( excluding the first 200 bp ) for each TU i asri=1- tbi [0 . 2 , 10 kbp]CDK9as inhibited/tbi [0 . 2 , 10 kbp]Control where negative values were set to 0 . For each condition j ( control or CDK9as inhibited ) the antisense bias corrected number of transcribed bases tbij was calculated on all expressed TUs i with a given response ratio ri , excluding the first 200 bp . All TUs were truncated by 5 kbp in length from the 3’ end prior to calculation to avoid influence of some alterations in signal around the pA site after CDK9as inhibition ( Laitem et al . , 2015 ) . A robust common elongation velocity estimate was calculated by finding an optimal fit for all TUs i between 25 to 200 kbp in length Li , that is minimizing the functionloss=mediani ( \|1−tbiCDK9asinhibitedtbiControl−riv ( t∗−t ) Li\| ) on the interval [0 , 10000] with inhibitor treatment duration t=15 [min] and labeling duration t = 5 [min] , given thattbiCDK9as inhibited-tbiControl=ritbiControlLivit-t that is the difference of transcribed bases obtained by the CDK9as inhibitor treatment equals the number of transcribed bases per nucleotide tbiControl/Li times the number of nucleotides traveled vit-t in t-t minutes corrected by the amount of the response ri . For each condition j ( control or CDK9as inhibited ) the antisense bias corrected number of transcribed bases tbij was calculated on all expressed TUs i exceeding 35 kbp in length , excluding the first 200 bp . All TUs were truncated by 5 kbp in length from the 3’ end prior to calculation to avoid influence of some alterations in signal around the pA site after CDK9as inhibition ( Laitem et al . , 2015 ) . Of all remaining TUs only those were kept harboring one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) . For each TU i with ri>0 . 25 the elongation velocity vi [kbp/min] was calculated asvi=tbiControl-tbi CDK9as inhibitedtbiControl∙riLit-t with inhibitor treatment duration t=15 [min] and labeling duration t = 5 [min] . Two biological replicates of reactions including empigen BB detergent treatment during immunoprecipitation ( IP ) were performed essentially as described ( Nojima et al . , 2016; Schlackow et al . , 2017 ) , with minor modifications . Briefly , 1 . 6 × 108 Raji B ( CDK9as ) cells were treated for 15 min with solvent DMSO ( control ) or 5 µM of 1-NA-PP1 ( CDK9as inhibitor ) . Cell fractionation was performed as described ( Conrad and Ørom , 2017 ) . Isolated chromatin was digested with micrococcal nuclease ( MNase ) ( NEB , Ipswich , MA USA ) at 37°C and 1 , 400 rpm for 90 s . To inactivate MNase , EGTA was added to a final concentration of 25 mM . Digested chromatin was collected by centrifugation at 4°C and 13 , 000 rpm for 5 min . The supernatant was diluted tenfold with IP buffer containing 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 05% ( vol/vol ) NP-40 , and 1% ( vol/vol ) empigen BB ( Sigma-Aldrich , St . Louis , MO USA ) . For each IP , 50 µg of Pol II antibody clone MABI0601 ( BIOZOL , Eching , Germany ) was conjugated to Dynabeads M-280 Sheep Anti-Mouse IgG ( Thermo Fisher Scientific , Waltham , MA USA ) . Pol II antibody-conjugated beads were added to diluted sample . IP was performed on a rotating wheel at 4°C for 1 hr . The beads were washed six times with IP buffer ( 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 0 . 05 % NP-40 , and 1% empigen BB ) and once with 500 µL of PNKT buffer containing 1 x T4 polynucleotide kinase ( PNK ) buffer ( NEB , Ipswich , MA USA ) and 0 . 1% ( vol/vol ) Tween-20 ( Sigma-Aldrich , St . Louis , MO USA ) . Beads were incubated in 100 µL of PNK reaction mix containing 1 x PNK buffer , 0 . 1% ( vol/vol ) Tween-20 , 1 mM ATP , and T4 PNK , 3’ phosphatase minus ( NEB , Ipswich , MA USA ) at 37°C for 10 min . Beads were washed once with IP buffer . RNA was extracted with TRIzol reagent . RNA was precipitated with GlycoBlue co-precipitant ( Thermo Fisher Scientific , Waltham , MA USA ) and resolved on 6% denaturing acrylamide containing 7 M urea ( PanReac AppliChem , Darmstadt , Germany ) gel for size purification . Fragments of 35–100 nt were eluted from the gel using elution buffer containing 1 M NaOAc , 1 mM EDTA , and precipitated in ethanol . RNA libraries were prepared according to the TruSeq Small RNA Library Kit ( Illumina , Massachusetts USA ) and as described ( Nojima et al . , 2016 ) . The size-selected and pre-amplified fragments were analyzed on a Fragment Analyzer before clustering and sequencing on an Illumina HiSeq 2500 sequencer . Paired-end 50 base reads with additional 6 base reads of barcodes were obtained for each of the samples , that is mNET-seq samples with 1-NA-PP1 ( CDK9as inhibitor ) and with DMSO ( control ) treatment . Reads were demultiplexed and mapped with STAR 2 . 3 . 0 ( Dobin and Gingeras , 2015 ) to the hg20/hg38 ( GRCh38 ) genome assembly ( Human Genome Reference Consortium ) . Samtools ( Li et al . , 2009 ) was used to quality filter SAM files , whereby alignments with MAPQ smaller than 7 ( -q 7 ) were skipped and only proper pairs ( -f2 ) were selected . Further data processing was carried out using the R/Bioconductor environment . Antisense bias ( ratio of spurious reads originating from the opposite strand introduced by the RT reactions ) was determined using positions in regions without antisense annotation with a coverage of at least 100 according to Refseq annotated genes ( UCSC RefSeq GRCh38 ) . mNET-seq coverage tracks were size factor normalized on 260 TUs that showed a response of less than 5% ( ri<0 . 05 ) in the TT-seq signal upon 1-NA-PP1 ( CDK9as inhibitor ) treatment . The response ratio ri was determined as described above including also TUs with multiple TSS to extend the number of TUs for normalization . Note that variation of the response ratio cutoff and thereby the number of TUs available for normalization does virtually not change the normalization parameters . Coverage tracks for further analysis were restricted to the last nucleotide incorporated by the polymerase in the aligned mNET-seq reads . For all expressed TUs i exceeding 10 kbp in length with one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) the pause site m* was calculated for all bases m in a window from the TSS to the end of the first exon ( excluding the last 5 bases ) via maximizing the functionρi=maxmpim where ρi needed to exceed 5 times the median of the signal strength pim for all non-negative antisense bias corrected mNET-seq coverage values ( Nojima et al . , 2015 ) . Note that all provided coverage tracks were used . The gene-wise mean melting temperature of the DNA-RNA and DNA-DNA hybrid was calculated from subsequent melting temperature estimates of 8-base pair DNA-RNA and DNA-DNA duplexes tiling the respective area according to ( SantaLucia , 1998; Sugimoto et al . , 1995 ) . The known sequence and mixture of the utilized spike-ins allows to calculate a conversion factor to RNA amount per cell [cell−1] given their molecular weight assuming perfect RNA extraction . The number of spike-in molecules per cell N [cell−1] was calculated asN=mMnNA with the number of spike-ins m 25 . 10−9 [g] , the number of cells n 3 . 27 . 107 , the Avogadro constant NA 6 . 02214085774 . 1023 [mol−1] and molar-mass ( molecular weight ) of the spike-ins M [g mol−1] calculated asM = An⋅329 . 2 + ( 1−τ ) ⋅Un⋅306 . 2 + Cn⋅305 . 2 + Gn⋅345 . 2 + τ⋅4sUn⋅322 . 26 + 159 where An , Un , Cn , Gn and 4sUn are the number of each respective nucleotide within each spike-in polynucleotide . τ is set to 0 . 1 in case of a labeled spike-in and 0 otherwise . The addition of 159 to the molecular weight takes into account the molecular weight of a 5' triphosphate . Provided the above the conversion factor to RNA amount per cell κ [cell−1] can be calculated asκ=mean ( mediani ( tbiLi⋅N ) ) for all labeled spike-in species i with length Li . Note that imperfect RNA extraction efficiency would lead to an underestimation of cellular labeled RNA in comparison to the amount of added spike-ins and thus to an underestimation of initiation frequencies . In case of a strong underestimation however the real initiation frequencies would lie above the pause-initiation limit , which is theoretically impossible . Thus we assume this effect to be insignificant . The antisense bias corrected number of transcribed bases tbiControl was calculated on all expressed TUs i exceeding 10 kbp in length . Of all remaining TUs only those were kept harboring one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) . For each TU i the productive initiation frequency Ii [cell−1min−1] , which corresponds to the pause release rate , was calculated asIi=1κ∙tbiControlt∙Li with labeling duration t = 5 [min] and length Li . Note that tbiControl and Li were restricted to regions of non-first constitutive exons ( exonic bases common to all isoforms ) . For all expressed TUs i exceeding 10 kbp in length with one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) the pause duration di [min] was calculated as the residing time of the polymerase in a window ±100 bases m around the pause site ( see above ) asdi=∑+/−100pimIi . mediani ( viIivi ( t∗−t ) /∑response windowpim ) with pause release rate Ii and the number of polymerases pim ( antisense bias corrected mNET-seq coverage values [Nojima et al . , 2015] ) in a window ±100 bases around the pause site . For pause sites below 100 bp downstream of the TSS the first 200 bp of the TU were considered . Note that the right part of the formula is restricted to mNETseq instances above the 50% quantile for robustness and adjusts di to an absolute scale by comparing the CDK9 derived elongation velocities vi with those derived from combining mNET-seq and TT-seq data in the response window 200 , vit-t . The previously derived inequality from ( Ehrensberger et al . , 2013 ) vI≥50 [bp] states that new initiation events into productive elongation are limited by the velocity of the polymerase in the promoter-proximal region and that steric hindrance occurs below a distance of 50 bp between the active sites of the initiating Pol II and the paused Pol II . Given the calculations of pause duration d and ( productive ) initiation frequency I above , we can reformulate this inequality to200 [bp]d∙I≥50 [bp] with 200 [bp] being the above defined pause window . Based on the following model we simulated TT-seq coverage values by providing elongation velocity profiles vt , a labeling duration tlab and a uracil content dependent labeling biaslf=1− ( 1−plab ) #uf plab denotes the labeling probability ( set to 0 . 05 ) and #uf the number of uracil residues of a given fragment f ( set to 0 . 28 times fragment length ) . The elongation velocity profile vt can be used to calculate the number of elongated positions of the polymerase τt at timepoint t asτt=∫0tvtdt Given the transcription start site τ0 the number of elongated positions τt can be used to determine the end of an emerging nascent fragment f . Based on that we determined the start position of a fragment as τ ( max ( t−tlab , 0 ) ) for each labeling duration tlab as the position of the polymerase at the beginning of the labeling process . Subsequently , we used the number of uracil residues present in the RNA fragment #uf to weight the amount of coverage contributed by this fragment as lf . Additionally , we applied a size selection similar to that in the original protocol for fragments below 80 bp in length with a sigmoidal curve that mimics a typical size selection spread . Given a pause position of 80 bp downstream of the TSS and pause duration of 1 or 2 min we adjusted the elongation velocity profile to simulate polymerase pausing . Note that neither reasonable changes in labeling probability , size selection probability nor changes in uracil residue content change the general observation that longer pause durations induce a greater shortage of TT-seq coverage in the region between the TSS and the pause site . For each condition j ( control or CDK9as inhibited ) the antisense bias corrected number of transcribed bases tbij was calculated on all expressed TUs i exceeding 35 kbp in length , excluding the first 200 bp . All TUs were truncated by 5 kb in length from the 3’ end prior to calculation to avoid influence of some alterations in signal around the pA site after CDK9as inhibition ( Laitem et al . , 2015 ) . Of all remaining TUs only those were kept harboring one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) . For each TU i with ri>0 . 25 the cumulative sums of the difference of the number of transcribed bases tbij for each base k was calculated asS0=0 Sn=Sn-1+tbiControl-tbiCDK9as inhibited starting at the unique TSS ( position 0 ) to n=Li the length of the TU . A elongation length estimate Liresponse window was then calculated by finding an optimal fit for n between 0 to Li , that is maximizing the functiongain=nmax ( Sn⋅Limaxn=1 . . LiSn−n+1 ) on the interval [0 , Li] . In words , finding the maximum of the cumulative sums of difference in coverage rotated 45 degrees clockwise . The elongation velocity v^i [kbp/min] was subsequently calculated asv^i=Liresponse window ( t∗−t ) with inhibitor treatment duration t=15 [min] and labeling duration t = 5 [min] . For all expressed TUs i exceeding 10 kb in length with one unique TSS given all Refseq annotated isoforms ( UCSC RefSeq GRCh38 ) the pause duration d^i [min] was calculated as the residing time of the polymerase in a window ±100 bases m around the pause site ( see above ) asdi^=∑+/−100pim . Liresponse window∑response windowpim . v^i with elongation length estimate Liresponse window and the number of polymerases pim ( antisense bias corrected mNET-seq coverage values ) in a window ±100 bases around the pause site . For pause sites below 100 bp downstream of the TSS the first 200 bp of the TU were considered . Note that d^i was adjusted to the height as di by a single proportionality factor for visualization purposes . The gene-wise DMS-seq coverage ( 300 μl in vivo ) for a window of [−15 , –65] bp upstream of the pause site was normalized by subtraction from the respective DMS-seq coverage ( denatured ) allowing for maximal 5% negative values which were set to 0 ( sequencing depth adjustment ) . The gene-wise mean values were subsequently normalized by dividing with the initiation frequency . Note that the latter normalization has an insignificant effect . The gene-wise mean minimum free energy for a window of [−15 , –65] bp upstream of the pause site was calculated from subsequent minimum free energy estimates of 13-base pair RNA fragments tiling the respective area using RNAfold from the ViennaRNA package ( Lorenz et al . , 2011 ) .	Genes can contain the coded instructions to make proteins . These instructions must first be copied , or transcribed , into an intermediate molecule called a messenger RNA by an enzyme known as RNA polymerase II . Shortly after it begins , this enzyme – which is called Pol II for short – pauses , and it only starts again after it recruits other proteins , including one called CDK9 . The number of RNA copies made of a gene depends upon how many Pol II enzymes begin transcription . Pol II pausing also has an effect – if the enzymes pause for longer , less messenger RNA is transcribed . But why does this happen ? One hypothesis is that paused Pol II enzymes interfere with other Pol II enzymes initiating transcription . Yet , until recently it was not possible to measure if this actually happens in living cells . Now , Gressel , Schwalb et al . used a new biochemical method together with a compound that blocks CDK9 to measure pausing and transcription initiation for active genes in living human cells . The CDK9 inhibitor was used to make Pol II enzymes pause for longer than normal . Gressel , Schwalb et al . found that different genes responded differently to CDK9 inhibition , meaning that some remained paused for longer than others . The number of Pol II enzymes that initiated transcription was calculated by measuring how many RNA copies had been made locally at that the site of transcription . These experiments showed that blocking the release of paused Pol II strongly reduced the number of RNA copies made . Gressel , Schwalb et al . conclude that Pol II pausing can control initiation of transcription . Cells may use Pol II pausing to adjust how many copies of an RNA are made , helping to ensure that different cell types make the appropriate number of RNA copies from a gene . Many diseases are associated with gene transcription being incorrectly regulated . This and future studies will help scientists to better understand how Pol II pausing contributes to the control of transcription in both normal and diseased cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2017	CDK9-dependent RNA polymerase II pausing controls transcription initiation
Rodents are emerging as increasingly popular models of visual functions . Yet , evidence that rodent visual cortex is capable of advanced visual processing , such as object recognition , is limited . Here we investigate how neurons located along the progression of extrastriate areas that , in the rat brain , run laterally to primary visual cortex , encode object information . We found a progressive functional specialization of neural responses along these areas , with: ( 1 ) a sharp reduction of the amount of low-level , energy-related visual information encoded by neuronal firing; and ( 2 ) a substantial increase in the ability of both single neurons and neuronal populations to support discrimination of visual objects under identity-preserving transformations ( e . g . , position and size changes ) . These findings strongly argue for the existence of a rat object-processing pathway , and point to the rodents as promising models to dissect the neuronal circuitry underlying transformation-tolerant recognition of visual objects . Converging evidence ( Wang and Burkhalter , 2007; Wang et al . , 2011 , 2012 ) indicates that rodent visual cortex is organized in two clusters of strongly reciprocally connected areas , which resemble , anatomically , the primate ventral and dorsal streams ( i . e . , the cortical pathways specialized for the processing of , respectively , shape and motion information ) . The first cluster includes most of lateral extrastriate areas , while the second encompasses medial and parietal extrastriate cortex . Solid causal evidence confirms the involvement of these modules in ventral-like and dorsal-like computations – lesioning laterotemporal and posterior parietal cortex strongly impairs , respectively , visual pattern discrimination and visuospatial perception ( Gallardo et al . , 1979; McDaniel et al . , 1982; Wörtwein et al . , 1993; Aggleton et al . , 1997; Sánchez et al . , 1997; Tees , 1999 ) . By comparison , functional understanding of visual processing in rodent extrastriate cortex is still limited . While studies employing parametric visual stimuli ( e . g . , drifting gratings ) support the specialization of dorsal areas for motion processing ( Andermann et al . , 2011; Marshel et al . , 2011; Juavinett and Callaway , 2015 ) , the functional signature of ventral-like computations has yet to be found in lateral areas . In fact , parametric stimuli do not allow probing the core property of a ventral-like pathway – i . e . , the ability to support recognition of visual objects despite variation in their appearance , resulting from ( e . g . ) position and size changes . In primates , this function , known as transformation-tolerant ( or invariant ) recognition , is mediated by the gradual reformatting of object representations that takes place along the ventral stream ( DiCarlo et al . , 2012 ) . Recently , a number of behavioral studies have shown that rats too are capable of invariant recognition ( Zoccolan et al . , 2009; Tafazoli et al . , 2012; Vermaercke and Op de Beeck , 2012; Alemi-Neissi et al . , 2013; Vinken et al . , 2014; Rosselli et al . , 2015 ) , thus arguing for the existence of cortical machinery supporting this function also in rodents . This hypothesis is further supported by the preferential reliance of rodents on vision during spatial navigation ( Cushman et al . , 2013; Zoccolan , 2015 ) , and by the strong dependence of head-directional tuning on visual cues in rat hippocampus ( Acharya et al . , 2016 ) . Yet , in spite of a recent attempt at investigating rat visual areas with shape stimuli ( Vermaercke et al . , 2014 ) , functional evidence about how rodent visual cortex may support transformation-tolerant recognition is still sparse . In our study , we compared how visual object information is processed along the anatomical progression of extrastriate areas that , in the rat brain , run laterally to V1: lateromedial ( LM ) , laterointermediate ( LI ) and laterolateral ( LL ) areas ( Espinoza and Thomas , 1983; Montero , 1993; Vermaercke et al . , 2014 ) . By applying specially designed information theoretic and decoding analyses , we found a sharp reduction of the amount of low-level information encoded by neuronal firing along this progression , and a concomitant increase in the ability of neuronal representations to support invariant recognition . Taken together , these findings provide compelling evidence about the existence of a ventral-like , object-processing pathway in rat visual cortex . Under the hypothesis that , along an object-processing pathway , information about low-level image properties should be partially lost ( DiCarlo et al . , 2012 ) , we measured the sensitivity of each recorded neuron to the lowest-level attribute of the visual input – the amount of luminous energy a stimulus impinges on a neuronal receptive field . This was quantified by a metric ( RF luminance; see Materials and methods ) , which , for some neuron , seemed to account for the modulation of the firing rate not only across object transformations , but also across object identities . This was the case for the example V1 neuron shown in Figure 2A , where the objects eliciting the larger responses along the position axis ( red and blue curves ) were the ones that consistently covered larger fractions of the neuron’s RF ( yellow ellipses ) , thus yielding higher RF luminance values ( Figure 2—figure supplement 1A–B ) , as compared to the less effective object ( green curve ) . By contrast , for other units ( e . g . , the example LL neuron shown in Figure 2B ) , RF luminance did not appear to account for the tuning for object identity – similarly bright objects ( Figure 2—figure supplement 1C–D ) yielded very different response magnitudes ( compare the red with the green and blue curves ) . 10 . 7554/eLife . 22794 . 007Figure 2 . Tuning of an example V1 and LL neuron . ( A–B ) The bottom plots show the average firing rates ( AFRs ) of a V1 ( A ) and a LL ( B ) neuron , evoked by three objects , presented at eight different visual field positions ( shown in Figure 1B . 1 ) . The top plots show the peri-stimulus time histograms ( PSTHs ) obtained at two positions , along with the images of the corresponding stimulus conditions . These images also display the RF profile of each neuron , in the guise of three concentric ellipses , corresponding to 1 SD , 2 SD and 3 SD of the two-dimensional Gaussians that were fitted to the raw RFs . The numbers ( white font ) show the luminosity that each object condition impinged on the RF ( referred to as RF luminance; the distributions of RF luminance values produced by the objects across the full set of positions and the full set of transformations are shown in Figure 2—figure supplement 1 ) . The gray patches over the PSTHs show the spike count windows ( 150 ms ) used to compute the AFRs ( their onsets were the response latencies of the neurons ) . Error bars are SEM . ( C–D ) Luminance sensitivity profiles for the two examples neurons shown in ( A ) and ( B ) . For each neuron , the stimulus conditions were grouped in 23 RF luminance bins with 10 stimuli each , and the intensity of firing across the resulting 23 × 10 matrix was color-coded . Within each bin , the stimuli were ranked according to the magnitude of the response they evoked . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 00710 . 7554/eLife . 22794 . 008Figure 2—figure supplement 1 . RF luminance values produced by very effective and poorly effective objects: a comparison between an example V1 and an example LL neuron . ( A ) The arrows show the RF luminance values that were measured , across eight positions , for three objects that were differently effective at driving an example V1 neuron ( same cell of Figure 2A and C ) . The most effective object ( #7; see Figure 2A ) consistently yielded larger RF luminance values ( red arrows ) , compared to the least effective one ( #1; green arrows ) . By comparison , another object ( #9 ) , which was also quite effective at driving the cell ( see Figure 2A ) , produced RF luminance values ( blue arrows ) that substantially overlapped with those of object #7 . ( B ) Same comparison as in ( A ) , but with the distributions of RF luminance values of the three objects , computed across the full set of 23 transformations each object underwent ( same color code as in A ) . ( C ) The RF luminance values produced by three objects , across eight positions , for an example LL neuron ( same cell of Figure 2B and D ) . In this case , the most effective object ( #6; see Figure 2B ) did not yield the largest RF luminance values ( red arrows ) . The other two objects ( #9 and #8 ) , which were both ineffective at driving the cell ( see Figure 2B ) , produced either very similar ( #9; green arrows ) or considerably larger ( #8; blue arrows ) RF luminance values , as compared to object #6 . ( D ) Same comparison as in ( C ) , but with the distributions of RF luminance values of the three objects , computed across the full set of 23 transformations each object underwent ( same color code as in C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 008 To better appreciate the sensitivity of each neuron to stimulus energy , we considered a subset of the stimulus conditions , consisting of 23 transformations of each object , for a total of 230 stimuli ( see Materials and methods ) . We then grouped these stimuli in 23 equi-populated RF luminance bins and we color-coded the intensity of the neuronal response across the resulting matrix of 10 stimuli per 23 bins ( the stimuli within each bin were ranked according to the magnitude of the response they evoked ) . Many units , as the example V1 neuron of Figure 2C ( same cell of Figure 2A ) , showed a gradual increase of activity across consecutive bins of progressively larger luminance , with little variation of firing within each bin . Other units , as the example LL neuron of Figure 2D ( same cell of Figure 2B ) , displayed no systematic variation of firing across the luminance axis , but were strongly modulated by the conditions within each luminance bin , thus suggesting a tuning for higher-level stimulus properties . Note that , although the example LL neuron of Figure 2D fired more sparsely than the example V1 neuron of Figure 2C , the sparseness of neuronal firing across the 230 stimulus conditions , measured as defined in ( Vinje and Gallant , 2000 ) , was not statistically different between the LL and the V1 populations ( p>0 . 05; Mann-Whitney U-test ) , with the median sparseness being ~0 . 13 in both areas . Critically , this does not imply that the two areas do not differ in the way they encode visual objects , because sparseness is a combined measure of object selectivity and tolerance to changes in object appearance , which is positively correlated with the former and negatively correlated with the latter . As such , a concomitant increase of both selectivity and tolerance can lead to no appreciable change of sparseness across an object-processing hierarchy ( Rust and DiCarlo , 2012 ) . This suggests that other approaches are necessary to compare visual object representations along a putative ventral-like pathway . In our study , we first quantified the relative sensitivity of a neuron to stimulus luminance and higher-level features by using information theory , because of two main advantages this approach offers in investigating neuronal coding . First , computing mutual information between a stimulus’ feature and the evoked neuronal response provides an upper bound to how well we can reconstruct the stimulus’ feature from observing the neural response on a single trial , without committing to the choice of any specific decoding algorithm ( Rieke et al . , 1997; Borst and Theunissen , 1999; Quiroga and Panzeri , 2009; Rolls and Treves , 2011 ) . Second , information theory provides a solid mathematical framework to disentangle the ability of a neuron to encode a given stimulus’ feature from its ability to encode another feature , even if these two features are not independently distributed across the stimuli , i . e . , even if they are correlated in an arbitrarily complex , non-linear way ( Ince et al . , 2012 ) . This property was crucial to allow estimating the relative contribution of luminance and higher-level features to the tuning of rat visual neurons , given that luminance did co-vary , in general , with other stimulus properties , such as object identity , position , size , etc . ( see Figure 2A–B ) . In our analysis , for each neuron , we first computed Shannon’s mutual information between stimulus identity S and neuronal response R , formulated as: ( 1 ) I ( R;S ) =∑sP ( s ) ∑rP ( r\|s ) log2P ( r\|s ) P ( r ) , where P ( s ) is the probability of presentation of stimulus s , P ( r\|s ) is the probability of observing a response r following presentation of stimulus s , and P ( r ) is the probability of observing a response r across all stimulus presentations . The response R was quantified as the number of spikes fired by the neuron in a 150 ms-wide spike count window ( e . g . , see the gray patches in Figure 2A–B ) , while the stimulus conditions S included all the 23 transformations of the 10 objects , previously used to produce Figure 2C–D , for a total of 230 different stimuli ( see Materials and methods for details ) . As graphically illustrated in Figure 3A , I ( R;S ) measured the discriminability of these 230 different stimuli , given the spike count distributions they evoked in a single neuron . 10 . 7554/eLife . 22794 . 009Figure 3 . Information conveyed by the neuronal response about stimulus luminance and luminance-independent visual features . ( A ) Illustration of how total stimulus information per neuron was computed . All the views of all the objects ( total of 230 stimuli ) were considered as different stimulus conditions , each giving rise to its own response distribution ( colored curves; only a few examples are shown here ) . Mutual information between stimulus and response measured the discriminability of the stimuli , given the overlap of their response distributions . ( B ) Mutual information ( median over the units recorded in each area ± SE ) between stimulus and response in each visual area ( full bars; p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The white portion of the bars shows the median information that each area carried about stimulus luminance , while the colored portion is the median information about higher-order , luminance-independent visual features . The number of cells in each area is written on the corresponding bar . ( C ) Median fraction of luminance-independent stimulus information ( fhigh; see Results ) that neurons carried in each area . Error bars and significance levels/test as in ( B ) . The mutual information metrics obtained for neurons sampled from cortical layers II-IV and V-VI are reported in Figure 3—figure supplement 1 . The fhigh values obtained for neuronal subpopulations with matched spike isolation quality are shown in Figure 3—figure supplement 2 . The sensitivity of rat visual neurons to luminance variations of the same object is shown in Figure 3—figure supplement 3 . The information carried by rat visual neurons about stimulus contrast and contrast-independent visual features is reported in Figure 3—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 00910 . 7554/eLife . 22794 . 010Figure 3—figure supplement 1 . Information conveyed by the neuronal response about stimulus luminance and luminance-independent visual features: a comparison between superficial and deep layers . Same mutual information analysis as the one shown in Figure 3 , but considering separately the neuronal populations recorded in cortical layers II-IV ( left plots ) and V-VI ( right plots ) . Colors , symbols , significance levels and statistical tests as in Figure 3 . ( A ) To check whether the drop of stimulus information ( full bars ) was similarly sharp in superficial and deep layers , a two-way ANOVA , with visual area and layer as factors , was carried out . The test yielded a significant main effect for area ( p<0 . 001 , F3 , 687 = 15 . 87 ) and a significant interaction between area and layer ( p<0 . 001 , F3 , 687 = 5 . 61 ) , but not a main effect for layer alone ( p>0 . 9 , F1 , 687 = 0 . 01 ) . This indicates that the information loss was sharper in deep layers . As for the case of the entire populations ( see Figure 3B ) , the drop of information was mainly due to a reduction of the information about stimulus luminosity ( white bars ) , while the information about higher-order visual attributes ( colored bars ) remained more stable across the areas . This was especially noticeable for deep layers , where the information about luminosity in LM and LI became almost half as large as in V1 , further dropping in LL to about one third of what observed in V1 . ( B ) The fraction of luminance-independent stimulus information ( fhigh ) that neurons carried in each area increased more gradually and was , overall , larger in deep than in superficial layers . A two-way ANOVA , with visual area and layer as factors , confirmed this observation , yielding a significant main effect for both area ( p<0 . 001 , F3 , 687 = 63 . 58 ) and layer ( p<0 . 001 , F1 , 687 = 14 . 7 ) and also a significant interaction ( p<0 . 01 , F3 , 687 = 4 . 06 ) . In terms of pairwise comparisons , this resulted in a very large and significant increase of fhigh in LL , as compared to all other areas , in both superficial and deep layers , and in a significant difference also between LI and V1/LM , but in deep layers only . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01010 . 7554/eLife . 22794 . 011Figure 3—figure supplement 2 . Independence of the fraction of luminance-independent stimulus information from the quality of spike isolation . ( A–B ) Same mutual information analysis as the one shown in Figure 3C , but restricted to units within a given tirtile of the SNR and RV metrics , used to asses the quality of spike isolation ( see Materials and methods ) . Note that the quality of spike isolation increases as function of SNR , while it decreases as a function of RV . The number of units in each area per tirtile is superimposed to the corresponding bar . ( C ) Same mutual information analysis , considering only the neuronal subpopulations with good spike isolation quality ( i . e . , with both SNR >10 and RV <2% ) – these constraints also equated the neuronal subpopulations in terms of firing rate ( Figure 6—figure supplement 3B ) . In ( A–C ) , symbols , significance levels , and statistical tests are as in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01110 . 7554/eLife . 22794 . 012Figure 3—figure supplement 3 . Sensitivity of rat visual neurons to luminance variations of the same object . ( A ) The thin black curves show the sensitivity to luminance variations of the same object for the neurons recorded in each area . Each curve was obtained by taking , for each neuron , the responses across the four luminance levels that were tested for each object ( i . e . , 12 . 5% , 25% , 50% and full luminance; see the example images reported below the abscissa in the leftmost panel ) , and then averaging these responses across the 10 object identities . This yielded a curve showing the average response of the neuron to each luminance level . Each curve was then normalized to its maximum , so as to allow a better comparison among the different neurons in a given area . The colored curves show the averages of these luminance-sensitivity curves across all the neurons recorded in each area . Note that the luminance levels on the abscissa are reported on a logarithmic scale . ( B ) For each neuron , the sharpness of its sensitivity to luminance changes of the same object was quantified by computing the sparseness of the black curves shown in ( A ) . The bars show how the sparseness decreased across the four visual areas ( median over the units recorded in each area ± SE; p<0 . 01 , 1-tailed U-test , Holm-Bonferroni corrected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01210 . 7554/eLife . 22794 . 013Figure 3—figure supplement 4 . Information conveyed by the neuronal response about stimulus contrast and contrast-independent visual features . ( A ) The full bars show the mutual information ( median over the units recorded in each area ± SE ) between stimulus and response in each visual area . The white portion of the bars shows I ( R;C ) , the information that each area carried about stimulus contrast C ( as measured by the RF contrast metric; see Materials and methods ) , while the colored portion shows I ( R;S'\|C ) , the information about stimulus identity ( S’ ) , as defined by any possible visual attribute with the exception of the RF contrast . The number of cells in each area is written on the corresponding bar . Note these numbers are lower than the total numbers of neurons recorded in each area ( e . g . , compare to Figure 3B ) , because only a fraction of units met the criterion to be included in this analysis ( see Materials and methods ) . Also note that , because only a subset of neurons and object conditions contributed to this analysis ( see Materials and methods ) , the total stimulus information reported here for V1 , LM and LI is lower than the total stimulus information shown in Figure 3B . In particular , this is due to the fact that only stimuli that covered at least 10% of each RF were included in this analysis ( see Materials and methods ) . This reduced the range of luminance values spanned by the stimulus conditions – a fact that , given the strong luminance sensitivity of V1 , LM and LI neurons ( see Figure 3B ) , produced the drop of stimulus information carried by individual neurons in these areas . No appreciable reduction of total stimulus information was found in LL ( compare the fourth full bar in this figure to the matching bar in Figure 3B ) , thus confirming once more the much lower sensitive of LL neurons to stimulus luminance . ( B ) Median fraction of contrast-independent stimulus information that neurons carried in each area . Error bars are SEs of the medians ( p<0 . 05 , p<0 . 01 , *p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 013 We then decomposed this overall stimulus information I ( R;S ) into the sum of the information about stimulus luminance and the information about luminance-independent , higher-level features , using the following mathematical identity ( Ince et al . , 2012 ) : ( 2 ) I ( R;S ) ≡I ( R;S′&L ) =I ( R;S′\|L ) +I ( R;L ) . Here , L is the RF luminance of the visual stimuli; I ( R;L ) is the information that R conveys about L; and I ( R;S'\|L ) measures how much information R carries about a variable S′ that denotes the identity of each stimulus condition S , as defined by any possible visual attribute with the exception of the RF luminance L ( i . e . , S=S′& L ) . The overall amount of visual information decreased gradually and significantly along the areas’ progression ( full bars in Figure 3B ) , with the median I ( R;S ) being about half in LL ( ~0 . 06 bits ) than in V1 ( ~0 . 12 bits; 1-tailed , Mann-Whitney U-test , Holm-Bonferroni corrected for multiple comparisons; hereafter , unless otherwise stated , all the between-area comparisons have been statistically assessed with this test; see Material and methods ) . This decline was due to a loss of the energy-related information I ( R;L ) ( white portion of the bars ) , rather than to a drop of the higher-level , energy-independent information I ( R;S′\|L ) ( colored portion of the bars ) , which changed little across the areas . As a result , the fraction of total information that neurons carried about higher-level visual attributes , i . e . , fhigh= I ( R;S′\|L ) /I ( R;S ) , became about twice as large in LL ( median ~0 . 5 ) as in V1 , LM and LI , and such differences were all highly significant ( Figure 3C ) . All these trends were largely preserved when neurons in superficial and deep layers were considered separately ( Figure 3—figure supplement 1 ) and did not depend on the quality of spike isolation ( Figure 3—figure supplement 2; see Discussion ) . The decrease of sensitivity to stimulus luminance along the areas’ progression was confirmed by measuring the tuning of rat visual neurons across the four luminance changes each object underwent ( i . e . , 12 . 5% , 25% , 50% and 100% luminance; see Figure 1B . 5 ) . In all the areas , the luminance-sensitivity curves showed a tendency of the firing rate to increase as a function of object luminance ( Figure 3—figure supplement 3A ) . However , such a growth was steeper in V1 and LM , as compared to LI and LL , where several neurons had a relatively flat tuning , with a peak , in some cases , at intermediate luminance levels . These trends were quantified by computing the sparseness of the response of each neuron over the luminance axis , which decreased monotonically along the areas’ progression ( Figure 3—figure supplement 3B ) , thus confirming the drop of sensitivity to luminance from V1 to LL . To further explore whether rat lateral visual areas were differentially sensitive to other low-level properties , we defined a metric ( RF contrast; see Materials and methods ) that quantified the variability of the luminance pattern impinged by any given stimulus on a neuronal RF . We then measured how much information rat visual neurons carried about RF contrast , and how much information they carried about contrast-independent visual features ( i . e . , we applied Equation 2 , but with RF contrast instead of RF luminance ) . The information about RF contrast decreased monotonically along the areas’ progression ( white portion of the bars in Figure 3—figure supplement 4A ) , while the contrast-independent information peaked in LL . As a consequence , the fraction of contrast-independent information carried by neuronal firing grew sharply and significantly from V1 to LI ( Figure 3—figure supplement 4B ) . Taken together , the results presented in this section show a clear tendency for low-level visual information to be substantially pruned along rat lateral extrastriate areas . Next , we explored whether neurons along lateral extrastriate areas also become gradually more capable of coding the identity of visual objects in spite of variation in their appearance ( DiCarlo et al . , 2012 ) . To this aim , we relied on both information theoretic and linear decoding analyses . Both approaches have been extensively used to investigate the primate visual system , with mutual information yielding estimates of the capability of single neurons to code both low-level ( e . g . , contrast and orientation ) and higher-level ( e . g . , faces ) visual features at different time resolutions and during different time epochs of the response ( Optican and Richmond , 1987; Tovee et al . , 1994; Rolls and Tovee , 1995; Sugase et al . , 1999; Montemurro et al . , 2008; Ince et al . , 2012 ) , and linear decoders probing the suitability of neuronal populations to support easy readout of object identity ( Hung et al . , 2005; Li et al . , 2009; Rust and Dicarlo , 2010; Pagan et al . , 2013; Baldassi et al . , 2013; Hong et al . , 2016 ) . In our study , we used both methods because of their complementary advantages . By computing mutual information , we estimated the overall amount of transformation-invariant information that single neurons conveyed about object identity . By applying linear decoders , we measured what fraction of such invariant information was formatted in a convenient , easy-to-read-out way , both at the level of single neurons and neuronal populations . In the information theoretic analysis , we defined every stimulus condition S as a combination of object identity O and transformation T ( i . e . , S = O and T ) and we expressed the overall stimulus information I ( R;S ) as follows: ( 3 ) I ( R;S ) ≡I ( R;O&T ) =I ( R;T\|O ) +I ( R;O ) . Here , I ( R;O ) is the amount of view-invariant object information carried by neuronal firing , i . e . , the information that R conveys about object identity , when the responses produced by the 23 transformations of an object ( across repeated presentations ) are considered together , so as to give rise to an overall response distribution ( see illustration in Figure 4A , bottom ) . The other term , I ( R;T\|O ) , is the information that R carries about the specific transformation of an object , once its identity has been fixed . 10 . 7554/eLife . 22794 . 014Figure 4 . Comparing total visual information and view-invariant object information per neuron . ( A ) Illustration of how total visual information and view-invariant object information per neuron were computed , given an object pair . In the first case , all the views of the two objects were considered as different stimulus conditions , each giving rise to its own response distribution ( colored curves ) . In the second case , the response distributions produced by different views of the same object were merged into a single , overall distribution ( shown in blue and red , respectively , for the two objects ) . ( B ) Total visual information per neuron ( median over the units recorded in each area ± SE ) as a function of the similarity between the RF luminance of the objects in each pair , as defined by ThLumRatio ( see Results ) . ( C ) Left: view-invariant object information per neuron ( median ± SE ) as a function of ThLumRatio . Right: view-invariant object information per neuron ( median ± SE ) for ThLumRatio=0 . 9 ( p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area is written on the corresponding bar . ( D ) Ratio between view-invariant object information and total information per neuron ( median ± SE ) , for ThLumRatio=0 . 9 . Significance levels/test as in ( C ) . The invariant object information carried by RF luminance as a function of ThLumRatio is reported in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01410 . 7554/eLife . 22794 . 015Figure 4—figure supplement 1 . Invariant object information carried by RF luminance as a function of ThLumRatio . For each neuron and object pair , we computed I ( L;O ) , the mutual information between object identity ( with all the views of each object in the pair taken into account ) and RF luminance . The colored curves show the median of I ( L;O ) ( ± SE ) across all the units recorded in each area as a function of the similarity between the RF luminance of the objects in each pair , as defined by ThLumRatio ( see Results ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 015 Given a neuron , we computed these information metrics for all possible pairs of object identities , while , at the same time , measuring the similarity between the objects in each pair in terms of RF luminance . Such similarity was evaluated by computing the ratio between the mean luminance of the dimmer object ( across its 23 views ) and the mean luminance of the brighter one – the resulting luminosity ratio ranged from zero ( dimmer object fully dark ) to one ( both objects with the same luminance ) . By considering object pairs with a luminosity ratio larger than a given threshold ThLumRatio , and allowing ThLumRatio to range from zero to one , we could restrict the computation of the information metrics to object pairs that were progressively less discriminable based on luminance differences , thus probing to what extent the ability of a neuron to code invariantly object identity depended on its luminance sensitivity . The overall stimulus information per neuron I ( R;S ) followed the same trend already shown in Figure 3B ( i . e . , it decreased gradually along the areas’ progression ) and such trend remained largely unchanged as a function of ThLumRatio ( Figure 4B ) . By contrast , the amount of view-invariant object information per neuron I ( R;O ) strongly depended on ThLumRatio ( Figure 4C , left ) . When all object pairs were considered ( ThLumRatio=0 ) , areas LM and LI conveyed the largest amount of invariant information , followed by V1 and LL . This trend remained stable until ThLumRatio reached 0 . 3 , after which the amount of invariant information in V1 , LM and LI dropped sharply . By contrast , the invariant information in LL remained stable until ThLumRatio approached 0 . 7 , after which it started to increase . As a result , when only pairs of objects with very similar luminosity were considered ( ThLumRatio=0 . 9 ) , a clear gradient emerged across the four areas , with the invariant information in LI and LL being significantly larger than in V1 and LM ( Figure 4C , right ) . These results indicate that neurons in V1 , LM and LI were able to rely on their sharp sensitivity to luminous energy ( Figure 3B ) and use luminance as a cue to convey relatively large amount of invariant object information , when such cue was available ( i . e . , for small values of ThLumRatio ) . The example V1 neuron shown in Figure 2A is one of such units . This cell successfully supported the discrimination of some object pairs only because of its strong sensitivity to stimulus luminance ( Figure 2C ) , which , for those pairs , happened to co-vary with object identity , in spite of the position changes and the other transformations that the objects underwent ( Figure 2—figure supplement 1A–B ) . In cases like this , observing a large amount of view-invariant information would be an artifact , because luminance would not at all be diagnostic of object identity , if these neurons were probed with a variety of object appearances as large as the one experienced during natural vision ( where each object can project thousands of different images on the retina ) . It is only because of the limited range of transformations that are testable in a neurophysiology experiment that luminance can possibly serve as a transformation-invariant cue of object identity . To verify that luminance could indeed act as a transformation-invariant cue , we measured I ( L;O ) – the amount of view-invariant object information conveyed by RF luminance alone ( Figure 4—figure supplement 1 ) . As expected , when no restriction was applied to the luminance difference of the objects to discriminate ( i . e . , for small values of ThLumRatio ) , I ( L;O ) was very large in all the areas . This confirmed that RF luminance , by itself , was able to convey a substantial amount of invariant object information . When ThLumRatio was allowed to increase , I ( L;O ) dropped sharply , eventually reaching zero in all the areas for ThLumRatio=0 . 9 . Interestingly , this decrease was the same found for I ( R;O ) in V1 , LM and LI ( see Figure 4C ) , thus showing that a large fraction of the invariant information observed in these areas ( but not in LL ) at low ThLumRatio was indeed accounted for by luminance differences between the objects in the pairs . Hence , the need of setting ThLumRatio=0 . 9 , thus considering only pairs of objects with very similar luminance to nullify the luminance confound , when comparing the areas in terms of their ability to support invariant recognition . Our analysis shows that , when this restriction was applied , a clear gradient emerged along the areas’ progression , with LL conveying the largest amount of invariant information per neuron , followed by LI and then by V1/LM ( Figure 4C , right ) . A similar , but sharper trend was observed when the relative contribution of the view-invariant information to the total information was measured , i . e . , when the ratio between I ( R;O ) and I ( R;S ) at ThLumRatio=0 . 9 was computed ( Figure 4D ) . The fraction of invariant information increased very steeply and significantly along the areas’ progression , being almost four times larger in LL than in V1/LM , and ~1 . 7 times larger in LL than in LI . Overall , these results indicate that the information that single neurons are able to convey about the identity of visual objects , in spite of variation in their appearance , becomes gradually larger along rat lateral visual areas ( see Discussion for further implications of these findings ) . I ( R;O ) ( Figure 4C ) provides an upper bound to the amount of transformation-invariant information that single neurons can encode about object identity , but it does not quantifies how easily this information can be read out by simple linear decoders ( see illustration in Figure 5A ) . Assessing this property , known as linear separability of object representations , is crucial , because attaining progressively larger levels of linear separability is considered the key computational goal of the ventral stream ( DiCarlo et al . , 2012 ) . In our study , we first measured linear separability at the single-cell level – i . e . , given a neuron and a pair of objects , we tested the ability of a binary linear decoder to correctly label the responses produced by the 23 views of each object ( this was done using the cross-validation procedure described in Materials and methods ) . For consistency with the previous analyses , the discrimination performance of the decoders was computed as the mutual information between the actual and the predicted object labels from the decoding outcomes ( Quiroga and Panzeri , 2009 ) . 10 . 7554/eLife . 22794 . 016Figure 5 . Linear separability of object representations at the single-neuron level . ( A ) Illustration of how a similar amount of view-invariant information per neuron can be encoded by object representations with a different degree of linear separability . The overlap between the response distributions produced by two objects across multiple views is similarly small in the top and bottom examples; hence , the view-invariant object information per neuron is similarly large . However , only for the distributions shown at the bottom , a single discrimination boundary can be found ( dashed line ) that allows discriminating the objects regardless of their specific view . ( B ) Linear separability of object representations at the single-neuron level ( median over the units recorded in each area ± SE; p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area is written on the corresponding bar . ( C ) Ratio between linear separability , as computed in ( B ) , and view-invariant object information , as computed in Figure 4C , right ( median ± SE ) . Significance levels/test as in ( B ) . In both ( B ) and ( C ) , ThLumRatio=0 . 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 016 The linear separability of object representations at the single-neuron level increased monotonically and significantly along the areas’ progression ( Figure 5B ) , being ~4 times larger in LL as in V1 and LM , and reaching an intermediate value in LI . This increase was steeper than the growth of the view-invariant information I ( R;O ) ( Figure 4C , right ) , thus suggesting that not only LL neurons encoded more invariant information than neurons in the other areas , but also that a lager fraction of this information was linearly decodable in LL . To quantify this observation we computed , for each cell , the ratio between linear separability ( Figure 5B ) and amount of invariant information ( Figure 4C , right ) . The resulting fraction of invariant information that was linearly decodable increased from ~55% in V1 to ~70% in LL , being significantly larger in LL than in any other area ( Figure 5C ) . An alternative ( more stringent ) way to assess transformation-tolerance is to measure to what extent a linear decoder , trained with a single view per object , is able to discriminate the other ( untrained ) views of the same objects ( see illustration in Figure 6A ) – a property known as generalization across transformations . Since , at the population level , large linear separability does not necessarily imply large generalization performance ( Rust and Dicarlo , 2010 ) ( see also Figure 8A ) , it was important to test rat visual neurons with regard to the latter property too ( see Materials and methods ) . Our analysis revealed that , when assessed at the single-cell level , the ability of rat visual neurons to support generalization to novel object views increased as significantly and steeply as linear separability ( Figure 6B ) . Interestingly , this trend was widespread across the whole cortical thickness and equally sharp in superficial and deep layers , although the generalization performances were larger in the latter ( Figure 6C ) . These conclusions were supported by a two-way ANOVA with visual area and layer as factors , yielding a significant main effect for both area and layer ( p<0 . 001 , F3 , 687 = 9 . 9 and F1 , 687 = 6 . 93 , respectively ) but no significant interaction ( p>0 . 15 , F3 , 687 = 1 . 74 ) . 10 . 7554/eLife . 22794 . 017Figure 6 . Ability of single neurons to support generalization to novel object views . ( A ) Illustration of how the ability of single neurons to discriminate novel views of previously trained objects was assessed . The blue and red curves refer to the hypothetical response distributions evoked by different views of two objects . A binary decoder is trained to discriminate two specific views ( darker curves , left ) , and then tested for its ability to correctly recognize two different views ( darker curves , right ) , using the previously learned discrimination boundary ( dashed lines ) . ( B ) Generalization performance achieved by single neurons with novel object views ( median over the units recorded in each area ± SE; p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area is written on the corresponding bar . ( C ) Median generalization performances ( ± SE ) achieved by single neurons for the neuronal subpopulations sampled from cortical layers II-IV ( left ) and V-VI ( right ) . Significance levels/test as in ( B ) . Note that the laminar location was not retrieved for all the recorded units ( see Materials and methods ) . ( D ) Median generalization performances ( ± SE ) achieved by single neurons , computed along individual transformation axes . In ( C–D ) , significance levels/test are as in ( B ) . ThLumRatio=0 . 9 in ( B–D ) . The generalization performances achieved by single neurons across parametric position and size changes are reported in Figure 6—figure supplement 1 . The generalization performances obtained for neuronal subpopulations with matched spike isolation quality are shown in Figure 6—figure supplement 2 . The firing rate magnitude measured in the four areas ( before and after matching the neuronal populations in terms of spike isolation quality ) is reported in Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01710 . 7554/eLife . 22794 . 018Figure 6—figure supplement 1 . Generalization achieved by single neurons across parametric position and size changes . ( A ) The curves show the generalization performances of binary decoders ( median over the units recorded in each area ± SE ) that were trained to discriminate two objects presented at the same position in the visual field , and then were required to discriminate those same objects across increasingly wider positions changes , with respect to the training location ( see Materials and methods for details ) . Differences among the four areas were quantified by a two-way ANOVA with visual area and distance from the training position as factors . The main effects of area ( F3 , 595 = 4 . 078 ) and distance ( F3 , 1785 = 2 . 706 ) were both significant ( p<0 . 01 and p<0 . 05 , respectively ) , thus confirming the existence of a gradient in the amount of position tolerance along the areas’ progression and along the distance axis . The interaction term was also significant ( p<0 . 001 , F9 , 1785 = 3 . 239 ) , indicating that the four decoding performances dropped at a different pace along the distance axis . ( B ) Discrimination performances of binary decoders ( median over the units recorded in each area ± SE ) that were required to generalize from a given training size to either a smaller ( left ) or larger ( right ) test size ( see Materials and methods for details; p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01810 . 7554/eLife . 22794 . 019Figure 6—figure supplement 2 . Independence of the generalization performances yielded by single neurons from the quality of spike isolation . ( A–B ) Same decoding analysis as the one shown in Figure 6B , but restricted to units within a given tirtile of the SNR and RV metrics , used to asses the quality of spike isolation ( see Materials and methods and the legend of Figure 3—figure supplement 2 ) . The number of units in each area per tirtile is the same reported in Figure 3—figure supplement 2A–B . ( C ) Same decoding analysis , considering only the neuronal subpopulations with good spike isolation quality ( i . e . , with both SNR >10 and RV <2% ) – these constraints also equated the neuronal subpopulations in terms of firing rate ( Figure 6—figure supplement 3B ) . In ( A–C ) , symbols , significance levels , and statistical tests are as in Figure 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 01910 . 7554/eLife . 22794 . 020Figure 6—figure supplement 3 . Magnitude of the firing rates . ( A ) Median number of spikes per second fired by the neurons in each visual area , as a response to: ( 1 ) the most effective object condition ( left ) ; ( 2 ) the 10 most effective object conditions ( middle ) ; and ( 3 ) all the 230 object conditions used in the single-cell information theoretic and decoding analyses . Error bars are SE of the medians . Stars indicate the statistical significance of pairwise comparisons ( p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area is superimposed to the corresponding bar . ( B ) Same plots as in ( A ) , but after considering only the neuronal subpopulations with good spike isolation quality ( i . e . , same subpopulations as in Figure 3—figure supplement 2C and Figure 6—figure supplement 2C ) . As a result of this selection , the median firing rates became statistically indistinguishable across all the four visual areas . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 020 The growth of transformation tolerance from V1 to LL was also observed when the individual transformation axes were considered separately in the decoding analysis ( Figure 6D ) . In this case , the training and testing views of the objects to be decoded were all randomly sampled from the same axis of variation: either position , size , rotation ( both in-plane and in-depth , pooled together ) or luminance . In all four cases , LL yielded the largest decoding performance , typically followed by LI and then by the more medial areas ( V1 and LM ) . The difference between LL and V1/LM was statistically significant for the position , size and rotation changes . Other comparisons were also significant , such as LL vs . LI for the position and size transformations , and LI vs . V1 and LM for the rotations . To further compare the position tolerance afforded by the four visual areas , we also measured the ability of single neurons to support generalization across increasingly wider positions changes – a decoder was trained to discriminate two objects presented at the same position in the visual field , and then tested for its ability to discriminate those same objects , when presented at positions that were increasingly distant from the training location ( Materials and methods ) . The neurons in LL consistently yielded the largest generalization performance , which remained very stable ( invariant ) as a function of the distance from the training location ( Figure 6—figure supplement 1A ) . A much lower performance was observed for LI and LM , while V1 displayed the steepest decrease along the distance axis ( these observations were all statistically significant , as assessed by a two-way ANOVA with visual area and distance from the training position as factors; see the legend of Figure 6—figure supplement 1A for details ) . A similar analysis was carried out to quantify the tolerance to size changes . In this case , we trained binary decoders to discriminate two objects at a given size , and then tested how well they generalized when those same objects were shown at either smaller or larger sizes ( Materials and methods ) . The resulting patterns of generalization performances ( Figure 6—figure supplement 1B ) confirmed once more the significantly larger tolerance afforded by LL neurons , as compared to the other visual areas . Taken together , the results of Figure 6D and Figure 6—figure supplement 1 indicate that the growth of tolerance across the areas’ progression was widespread across all tested transformations axes , ultimately yielding to the emergence , in LL , of a general-purpose representation that tolerated a wide spectrum of image-level variation ( see Discussion ) . Inspired by previous studies of the ventral stream ( Li et al . , 2009; Rust and Dicarlo , 2010 ) , we also tested to what extent RF size played a role in determining the growth of invariance across rat lateral visual areas ( intuitively , neurons with large RFs should respond to their preferred visual features over a wide span of the visual field , thus displaying high position and size tolerance ) . Earlier investigations of rat visual cortex have shown that RF size increases along the V1-LM-LI-LL progression ( Espinoza and Thomas , 1983; Vermaercke et al . , 2014 ) . Our recordings confirmed and expanded these previous observations , showing that RF size ( defined in Materials and methods ) grew significantly at each step along the areas’ progression ( Figure 7A ) , with the median in LL ( ~30° of visual angle ) being about twice as large as in V1 . At the same time , RF size varied widely within each area , resulting in a large overlap among the distributions obtained for the four populations . This allowed sampling the largest possible V1 , LM , LI and LL subpopulations with matched RF size ranges ( gray patches in Figure 7B ) , and computing the generalization performances yielded by these subpopulations ( Figure 7C ) – again , LL and LI afforded significantly larger tolerance than V1 and LM . A similar result was found for the fraction of energy-independent stimulus information fhigh carried by these RF-matched subpopulations ( Figure 7D ) – fhigh followed the same trend observed for the whole populations ( Figure 3C ) , with a sharp , significant increase from the more medial areas to LL . 10 . 7554/eLife . 22794 . 021Figure 7 . Single-neuron decoding and mutual information metrics , compared across neuronal subpopulations with matched RF size . ( A ) Spreads ( left ) and medians ± SE ( right ) of the RF sizes measured in each visual area ( p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area for which RF size could be estimated is reported on the corresponding bar of the right chart . Spreads and medians of the response latencies are reported in Figure 7—figure supplement 1 . ( B ) For each visual area , the generalization performances achieved by single neurons ( same data of Figure 6B ) are plotted against the RF sizes ( same data of panel A ) . In the case of LI and LL , these two metrics were significantly anti-correlated ( p<0 . 01; 2-tailed t-test ) . ( C ) Median generalization performances ( ± SE ) achieved by single neurons , as in Figure 6B , but considering only neuronal subpopulations with matched RF size ranges , indicated by the gray patches in ( B ) ( p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of neurons in each area fulfilling this constraint is reported on the corresponding bar . ( D ) Median fraction of luminance-independent stimulus information ( ± SE ) conveyed by neuronal firing as in Figure 3C , but including only the RF-matched subpopulations used in ( C ) . Significance levels/test as in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 02110 . 7554/eLife . 22794 . 022Figure 7—figure supplement 1 . Response latencies . Spreads ( left ) and medians ± SE ( right ) of the response latencies measured in each visual area ( p<0 . 05 , p<0 . 01 , p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) . The number of cells in each area is reported on the corresponding bar of the right chart . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 022 Interestingly , the decoding performances obtained for the RF-matched subpopulations were larger than those obtained for the whole populations , especially in the more lateral areas ( compare Figure 7C to Figure 6B ) . This means that the range of RF sizes that was common to the four populations contained the neurons that , in each area , yielded the largest decoding performances . This was the result of the specific relationship that was observed , within each neuronal population , between performance and RF size ( Figure 7B ) . While in V1 performance slightly increased as a function of RF size ( although not significantly; p>0 . 05; two-tailed t-test ) , performance and RF size were significantly anti-correlated in LI and LL ( p<0 . 01 ) . As explained in the Discussion , these findings suggest a striking similarity with the tuning properties of neurons at the highest stages of the monkey ventral stream ( DiCarlo et al . , 2012 ) . The growth of transformation tolerance reported in the previous section was highly significant and quite substantial , in relative terms . In LL , the discrimination performances were typically 2–4 times larger than in V1 , yet , in absolute terms , their magnitude was in the order of a few thousandths of a bit . This raised the question of whether such single-unit variations would translate into macroscopic differences among the areas , at the neuronal population level . To address this issue , we performed a population decoding analysis , in which we trained binary linear classifiers to read out visual object identity from the activity of neuronal populations of increasing size in V1 , LM , LI and LL . Random subpopulations of N units , with N={6 , 12 , 24 , 48 , 96} , were sampled from the full sets of neurons in each area . Given a subpopulation , the response axes of the sampled units formed a vector space , where each object condition was represented by the cloud of population response vectors produced by the repeated presentation of the condition across multiple trials ( see illustration in Figure 8A ) . The binary classifiers were required to correctly label the population vectors produced by different pairs of objects across transformations , thus testing to what extent the underlying object representations were linearly separable ( Figure 8A , left ) and generalizable ( Figure 8A , right ) . To avoid the luminance tuning confound , a pair of objects was included in the analysis , only if at least 96 neurons in each area could be found for which the luminance ratio of the pair was larger than a given threshold ThLumRatio . We set ThLumRatio to the largest value ( 0 . 8 ) that yielded at least three object pairs ( Figure 8B show the pairs that met this criterion , while Figure 8C and E show the mean classification performances over these three pairs ) . 10 . 7554/eLife . 22794 . 023Figure 8 . Linear separability and generalization of object representations , tested at the level of neuronal populations . ( A ) Illustration of the population decoding analyses used to test linear separability and generalization . The clouds of dots show the sets of response population vectors produced by different views of two objects . Left: for the test of linear separability , a binary linear decoder is trained with a fraction of the response vectors ( filled dots ) to all the views of both objects , and then tested with the left-out response vectors ( empty dots ) , using the previously learned discrimination boundary ( dashed line ) . The cartoon depicts the ideal case of two object representations that are perfectly separable . Right: for the test of generalization , a binary linear decoder is trained with all the response vectors ( filled dots ) produced by a single view per object , and then tested for its ability to correctly discriminate the response vectors ( empty dots ) produced by the other views , using the previously learned discrimination boundary ( dashed line ) . As illustrated here , perfect linearly separability does not guarantee perfect generalization to untrained object views ( see the black-filled , mislabeled response vectors in the right panel ) . ( B ) The three pairs of visual objects that were selected for the population decoding analyses shown in C-F , based on the fact that their luminance ratio fulfilled the constraint of being larger than ThLumRatio=0 . 8 for at least 96 neurons in each area . ( C ) Classification performance of the binary linear decoders in the test for linear separability , as a function of the number of neurons N used to build the population vector space . Performances were computed for the three pairs of objects shown in ( B ) . Each dot shows the mean of the performances obtained for the three pairs ( ± SE ) . The performances are reported as the mutual information between the actual and the predicted object labels ( left ) . In addition , for N = 96 , they are also shown in terms of classification accuracy ( right ) . The dashed lines ( left ) and the horizontal marks ( right ) show the linear separability of arbitrary groups of views of two objects ( same three pairs used in the main analysis; see Results ) . ( D ) The statistical significance of each pairwise area comparison , in terms of linear separability , is reported for each individual object pair ( 1-tailed U-test , Holm-Bonferroni corrected ) . In the pie charts , a black slice indicates that the test was significant ( p<0 . 001 ) for the corresponding pairs of objects and areas ( e . g . , LL > LI ) . ( E ) Classification performance of the binary linear decoders in the test for generalization across transformations . Same description as in ( C ) . ( F ) Statistical significance of each pairwise area comparison , in terms of generalization across transformations . Same description as in ( D ) . The same analyses , performed over a larger set of object pairs , after setting ThLumRatio=0 . 6 , are shown in Figure 8—figure supplement 1 . The dependence of linear separability and generalization from ThLumRatio is shown in Figure 8—figure supplement 2 . The statistical comparison between the performances achieved by a population of 48 LL neurons and V1 , LM and LI populations of 96 neurons is reported in Figure 8—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 02310 . 7554/eLife . 22794 . 024Figure 8—figure supplement 1 . Linear separability and generalization of object representations , tested at the level of neuronal populations , using a larger set of object pairs . ( A ) Classification performance of binary linear decoders in the test for linear separability ( see Figure 8A , left ) , as a function of the size of the neuronal subpopulations used to build the population vector space . This plot is equivalent to the one shown in Figure 8C , with the difference that , here , the objects that the decoders had to discriminate were allowed to differ more in terms of luminosity ( this was achieved by setting ThLumRatio=0 . 6 ) . As a result , much more object pairs ( 23 ) could be tested , compared to the analysis shown in Figure 8C . Each dot shows the mean of the 23 performances obtained for these object pairs and the error bar shows its SE . The larger number of object pairs allowed applying a 1-tailed , paired t-test ( with Holm-Bonferroni correction ) to assess whether the differences among the average performances in the four areas were statistically significant ( p<0 . 05 , p<0 . 01 , *p<0 . 001 ) . The performances are reported both as mutual information between actual and predicted object labels ( left ) and as classification accuracy ( i . e . , as the percentage of correctly labeled response vectors; right ) . ( B ) Classification performance of binary linear decoders in the test for generalization across transformations ( see Figure 8A , right ) . Same description as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 02410 . 7554/eLife . 22794 . 025Figure 8—figure supplement 2 . Dependence of linear separability and generalization , measured at the neuronal population level , from the luminance difference of the objects to discriminate . ( A ) Classification performance of binary linear decoders in the test for linear separability ( see Figure 8A , left ) as a function of the similarity between the RF luminance of the objects to discriminate , as defined by ThLumRatio ( see Results ) . The curves , which were produced using populations of 96 neurons , report the median performance in each visual area ( ± SE ) over all the object pairs obtained for a given value of ThLumRatio . Note that , for ThLumRatio=0 . 8 and ThLumRatio=0 . 6 , the performances are equivalent to those already shown in the left panels Figure 8C and Figure 8—figure supplement 1A ( rightmost points ) . Also note that , for ThLumRatio=0 . 1 , all the available object pairs contributed to the analysis . As such , the corresponding performances are those yielded by the four visual areas when no restriction was applied to the luminosity of the objects to discriminate . ( B ) Same analysis as in ( A ) , but for the test of generalization across transformations ( see Figure 8A , right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 02510 . 7554/eLife . 22794 . 026Figure 8—figure supplement 3 . Statistical comparison between the performance achieved by a population of 48 LL neurons and the performances yielded by populations of 96 neurons in V1 , LM and LI . ( A ) For each of the three object pairs tested in Figure 8 ( shown in Figure 8B ) , we checked whether a population of 48 LL neurons yielded a significantly larger performance than V1 , LM and LI populations of twice the number of neurons ( i . e . , 96 units ) in the linear discriminability task ( see Figure 8A , left ) . The resulting pie chart shown here should be compared to the rightmost column of the pie chart in Figure 8D , with a black slice indicating that the comparison was significant ( p<0 . 001; 1-tailed U-test , Holm-Bonferroni corrected ) for the corresponding pairs of objects and areas – e . g . , LL ( 48 units ) > LI ( 96 units ) . ( B ) Same analysis as in ( A ) , but for the test of generalization across transformations ( see Figure 8A , right ) . The pie chart shown here should be compared to the rightmost column of the pie chart in Figure 8F . DOI: http://dx . doi . org/10 . 7554/eLife . 22794 . 026 The ability of the classifiers to linearly discriminate the object pairs increased sharply as a function of N ( Figure 8C , left ) . In LL , the performance grew of ~500% , when N increased from 6 to 96 , reaching ~0 . 22 bits , which was nearly twice as large as the performance obtained in LI . More in general , linear separability grew considerably along the areas’ progression , following the same trend observed at the single-cell level ( see Figure 5B ) , but with performances that were about two orders of magnitude larger ( for N=96 ) . As a result , the differences among the areas became macroscopic – when measured in terms of classification accuracy ( Figure 8C , right ) , LL performance ( ~76% correct discrimination ) was about eight percentage points above LI , 10 above V1 and 17 above LM . Given the small number of object pairs that could be tested in this analysis ( 3 ) , the significance of each pairwise area comparison was assessed at the level of every single pair – e . g . , we tested if the objects belonging to a pair were better separable in the LL than in the LI representation ( this test was performed for N=96 , using , as a source of variability for the performances , the 50 resampling iterations carried out for each object pair; see Materials and methods ) . For all object pairs , the performances yielded by LL were significantly larger than in any other area ( last column of Figure 8D ) . In the case of LI , the performances were significantly larger than in LM and V1 , respectively , for all and two out of three pairs ( middle column of Figure 8D ) . Following the same rationale of a recent primate study ( Rust and Dicarlo , 2010 ) , we also checked how well binary linear classifiers could separate the representations of two arbitrary groups of object views – i . e . , with each group containing half randomly-chosen views of one of the objects in the pair , and half randomly-chosen views of the other object ( Materials and methods ) . For all the areas , the resulting discrimination performances were barely above the chance level ( i . e . , 0 bits and 50% correct discrimination; see dashed lines in Figure 8C ) . This means that rat lateral visual areas progressively reformat object representations , so as to make them more suitable to support specifically the discrimination of visual objects across view changes , and not generically the discrimination of arbitrary image collections . In the test of generalization to novel object views , the classification performances ( Figure 8E , left ) were about one order of magnitude smaller than those obtained in the test of linear separability ( Figure 8C , left ) . Still , for N=96 , they were about one order of magnitude larger than those obtained in the generalization task for single neurons ( compare to Figure 6B ) . In LL , the performance increased very steeply as a function of N , reaching ~0 . 032 bits for N=96 , which was more than three times larger than what obtained in LI . This implies a macroscopic advantage of LL , over the other areas , in terms of generalization ability – when measured in terms of accuracy ( Figure 8E , right ) , the performances in V1 , LM and LI were barely above the 50% chance level , while , in LL , they approached 60% correct discrimination . This achievement is far from trivial , given how challenging the discrimination task was , requiring generalization from a single view per object to many other views , spread across five different variation axes . Again , the statistical significance of each pairwise area comparison was assessed at the level of the individual object pairs . For all the pairs , the performances yielded by LL were significantly larger than in any other area ( last column of Figure 8F ) , while , for LI , the performances were significantly larger than in LM and V1 for two out of three pairs ( middle column of Figure 8F ) . To check the generality of our conclusions , we repeated these decoding analyses after loosening the constraint on the luminance ratio , i . e . , after lowering ThLumRatio to 0 . 6 , which yielded a larger number of object pairs ( 23 ) . Linear separability and generalization still largely followed the trends shown in Figure 8C and E , with LL yielding the largest performances , and LM the lowest ones ( Figure 8—figure supplement 1 ) . The main difference was that the performances in V1 were larger , reaching the same level of LI . This was expected , since lowering ThLumRatio made it easier for V1 , given its sharp tuning for luminosity , to discriminate the object pairs based on their luminance difference . In fact , it should be noticed that whatever cue is available to single neurons to succeed in an invariant discrimination task , that same cue will also be effective at the population level , because the population will inherit the sensitivity to the cue of its constituent neurons . This was the case of the luminance difference between the objects in a pair , which , unless constrained to be minimal , acted as a transformation-invariant cue for the V1 , LM and LI populations . This is shown by the dependence of linear separability and generalization , in these areas , on ThLumRatio , while , for the LL population , both metrics remained virtually unchanged as a function of ThLumRatio ( Figure 8—figure supplement 2 ) . Hence , the need of matching as closely as possible the luminance of the objects , while assessing transformation tolerance , also at the population level . Finally , we asked to what extent the decoding performances observed at the population level could be explained by the single-neuron properties illustrated in the previous sections . One possibility was that the superior tolerance achieved by the LL population resulted from the larger view-invariant information carried by the individual LL neurons , as reported in Figure 4C ( bar plot ) . To test whether this was the case , we matched the LL population and the other populations in terms of the total information they carried , based on the per-neuron invariant information observed in the four areas . Specifically , since in LL the invariant information per neuron was about twice as large as in V1 and LM and about 50% larger than in LI , we compared an LL population of N units to V1 , LM and LI populations of 2N units . This ensured that the latter would approximately have an overall view-invariant information that was either equal to or larger than the one of the LL population . We carried out this comparison by counting how many object pairs were still better discriminated by a population of 48 LL neurons , as compared to V1 , LM and LI populations of 96 units ( this is equivalent to compare the second-last red point to the last green , cyan and violet points in Figure 8C and E ) . We found that , consistently with what reported when comparing populations of equal size ( Figure 8C and E ) , also in this case LL yielded significantly larger performances than the other areas in all comparisons but one ( see Figure 8—figure supplement 3 ) . This indicates that the larger view-invariant information per neuron observed in LL is not sufficient , by itself , to explain the extent by which the LL population surpasses the other populations in the linear discriminability and generalization tasks – the better format of the LL representation plays a key role in determining its tolerance . A number of control analyses were performed to verify the solidity of our conclusions . The loss of energy-related stimulus information and the increase of transformation tolerance across the V1-LI-LL progression were found: ( 1 ) across the whole cortical thickness ( Figure 3—figure supplement 1 and Figure 6C ) ; ( 2 ) after matching the RF sizes of the four populations ( Figure 7C–D ) ; ( 3 ) across the whole spectrum of spike isolation quality in our recordings ( Figure 3—figure supplement 2A–B and Figure 6—figure supplement 2A–B ) ; and ( 4 ) when considering only neuronal subpopulations with the best spike isolation quality ( Figure 3—figure supplement 2C and Figure 6—figure supplement 2C ) , which also equated them in terms of firing rate ( Figure 6—figure supplement 3B ) . This means that no inhomogeneity in the sampling from the cortical laminae , in the amount of visual field coverage , in the quality of spike isolation , or in the magnitude of the firing rate could possibly account for our findings . Another issue deserving a discussion is the potential impact of an inaccurate estimate of the neuronal RFs on the calculation of the RF luminance , a metric that played a key role in all our analyses . Two orders of problems emerge when estimating the RF of a neuron . First , the structure and size of the RF depend , in general , on the shape and size of the stimuli used to map it . In our experiments , we used very simple stimuli ( high-contrast drifting bars ) presented over a dense grid of visual field locations . The rationale was to rely on the luminance-driven component of the neuronal response to simply estimate what portion of the visual field each neuron was sensitive too . This approach was very effective , because all the recorded neurons retained some amount of sensitivity to luminance , even those that were tuned to more complex visual features than luminance alone , as the LL neurons ( see the white portion of the bars in Figure 3B ) . As a result , very sharp RF maps were obtained in all the areas ( see examples in Figure 1C ) . Another problem , when estimating RFs , is that they do not always have an elliptical shape , although , in many instances , they are very well approximated by 2-dimensional Gaussians ( Op De Beeck and Vogels , 2000; Brincat and Connor , 2004; Niell and Stryker , 2008; Rust and Dicarlo , 2010 ) . In our study , we took two measures to prevent poor elliptical fits from possibly affecting our conclusions . In the RF size analysis ( Figure 7 ) , we only included data from neurons with RFs that were well fitted by 2-dimensional Gaussians ( see Materials and methods ) . More importantly , for the computation of the RF luminance , we did not use the fitted RFs , but we directly used the raw RF maps ( see Materials and methods and Figure 1—figure supplement 2C ) . This allowed weighting the luminance of the stimulus images using the real shapes of the RFs , thus reliably computing the RF luminance for all the recorded neurons . Finally , it is worth considering the implications of having studied object representations in anesthetized rats , passively exposed to visual stimuli . Two motivations are at the base of this choice . First , the need of probing visual neurons with the repeated presentation ( tens of trials; see Materials and methods ) of hundreds of different stimulus conditions , which are essential to properly investigate invariant object representations . The second motivation was the need of excluding potential effects of top-down signals , task- and state-dependence , learning and memory ( Gavornik and Bear , 2014; Cooke and Bear , 2015; Burgess et al . , 2016 ) , which are all detrimental when the goal is to understand the initial , largely reflexive , feed-forward sweep of activation through a visual processing hierarchy ( DiCarlo et al . , 2012 ) . For these reasons , many primate studies have investigated ventral stream functions in anesthetized monkeys [e . g . , see ( Kobatake and Tanaka , 1994; Ito et al . , 1995; Logothetis et al . , 1999; Tsunoda et al . , 2001; Sato et al . , 2013; Chen et al . , 2015 ) ] or , if awake animals were used , under passive viewing conditions [e . g . , see ( Pasupathy and Connor , 2002; Brincat and Connor , 2004; Hung et al . , 2005; Kiani et al . , 2007; Willmore et al . , 2010; Rust and Dicarlo , 2010; Hong et al . , 2016; El-Shamayleh and Pasupathy , 2016 ) ] . In the face of these advantages , anesthesia has several drawbacks . It can depress cortical activity , especially in high-order areas ( Heinke and Schwarzbauer , 2001 ) , and put the cortex in a highly synchronized state ( Steriade et al . , 1993 ) . In our study , we took inspiration from previous work on the visual cortex of anesthetized rodents ( Zhu and Yao , 2013; Froudarakis et al . , 2014; Pecka et al . , 2014 ) , cats ( Busse et al . , 2009 ) and monkeys ( Logothetis et al . , 1999; Sato et al . , 2013 ) , and we limited the impact of these issues by combining a light anesthetic with fentanyl-based sedation ( Materials and methods ) . This yielded robust visually-evoked responses both in V1 and extrastriate areas ( see PSTHs in Figure 2A–B , top ) . Still , we observed a gradual reduction of firing rate intensity along the areas’ progression ( Figure 6—figure supplement 3A ) , but such decrease , as mentioned above , did not account for our findings ( see Figure 6—figure supplement 3B , Figure 3—figure supplement 2C and Figure 6—figure supplement 2C ) . Obviously , this does exclude the impact of anesthesia on more subtle aspects of neuronal processing . For instance , isoflurane and urethane anesthesia have been reported to alter the excitation-inhibition balance that is typical of wakefulness , thus resulting in very time persistent stimulus-evoked responses , broad RFs and reduced strength of surround suppression ( Haider et al . , 2013; Vaiceliunaite et al . , 2013 ) . However , under fentanyl anesthesia , surround suppression was found to be very robust in mouse V1 ( Pecka et al . , 2014 ) , and our own recordings show that responses and RFs were far from sluggish and very similar to those obtained , from the same cortical areas , in awake rats ( Vermaercke et al . , 2014 ) – sharp tuning was observed in both the time and space domains , with transient responses , rarely lasting longer than 150 ms ( see examples in Figure 2A–B , top ) , and well-defined RFs ( see examples in Figure 1C ) , some as small as 5° of visual angle ( Figure 7A ) . Finally , neuronal activity in mouse visual cortex during active wakefulness has been shown to be very similar to that in the anesthetized state , with regard to a number of key tuning properties . These include the sharpness of orientation tuning in V1 ( Niell and Stryker , 2008 , 2010 ) , the sparseness and discriminability of natural scene representations in V1 , LM and anterolateral area AL ( Froudarakis et al . , 2014 ) , and the integration of global motion signals in rostrolateral area RL ( Juavinett and Callaway , 2015 ) . Taken together , the evidence reviewed above is highly reassuring with regard to the validity and generality of our findings . Obviously , being our data collected in passively viewing rats , the increase of transformation tolerance observed along the V1-LI-LL progression is not the result of a supervised learning process . Rather , our findings suggest that lateral extrastriate areas act as banks of general-purpose feature detectors , each endowed with an intrinsic degree of transformation tolerance . By virtue of their larger invariance , the detectors at the highest stages are able to automatically support transformation-tolerant recognition , without the need of explicitly learning the associative relations among all the views of an object . These conclusions are in full agreement with a recent behavioral study , showing that rats are capable of spontaneously generalize their recognition to previously unseen views of an object , without the need of any training ( Tafazoli et al . , 2012 ) . Another important implication of our study concerns the increase of the invariant object information per neuron found across rat lateral visual areas ( Figure 4C , bar plot ) . Critically , this finding does not imply that the overall invariant information per area also increases from V1 to LL . In fact , if the areas’ progression acted as a purely feed-forward processing chain , a per area increase would mean that new object information is created from one processing step to the next , a fact that would violate the data processing inequality ( Cover and Thomas , 2006 ) . But in rat visual cortex , there is a strong reduction of the number of neurons from V1 through the succession of lateral extrastriate areas . LM , LI and LL occupy a cortical surface that is , respectively , 31% , 3 . 5% and 2 . 1% of the surface of V1 ( Espinoza and Thomas , 1983 ) . Therefore , the object information per neuron can increase along the areas’ progression , without this implying that the total object information per area also increases . In addition , the connectivity among rat lateral visual areas is far from being strictly feed-forward . In both rats ( Sanderson et al . , 1991; Montero , 1993; Coogan and Burkhalter , 1993 ) and mice ( Wang et al . , 2012 ) , many corticortical and thalamocortical ‘bypass’ routes reach higher-level visual areas , in addition to the main anatomical route that connects consecutive processing stages step-by-step ( e . g . , V1 directly projects to LI and LL , and LL receives direct projections from the thalamus ) . Thus , the concentration of more object information in each individual neuron , while the visual representation is reformatted across consecutive processing stages , should not be interpreted as an indication that total object information increases across areas . Rather , our interpretation is that the increase of invariant object information per neuron is likely an essential step to make information about object identity gradually more explicit , and more easily readable by downstream neurons that only have access to a limited number of presynaptic units ( as confirmed by the linear decoding analyses shown in Figures 5–8 ) . To conclude , we believe that these results pave the way for the exploration of the neuronal mechanisms underlying invariant object representations in an animal model that is amenable to a large variety of experimental approaches ( Zoccolan , 2015 ) . In addition , the remarkable similarity between the anatomical organization of rat and mouse visual cortex suggests that mice too can serve as powerful models to dissect ventral stream computations , given the battery of genetic and molecular tools that this species affords ( Luo et al . , 2008; Huberman and Niell , 2011; Katzner and Weigelt , 2013 ) . All animal procedures were in agreement with international and institutional standards for the care and use of animals in research and were approved by the Italian Ministry of Health: project N . DGSAF 22791-A ( submitted on Sep . 7 , 2015 ) was approved on Dec . 10 , 2015 ( approval N . 1254/2015-PR ) ; project N . 5388-III/14 ( submitted on Aug . 23 , 2012 ) and project N . 3612-III/12 ( submitted on Sep . 15 , 2009 ) were approved according to the legislative decree 116/92 , article 7 . We used 26 naïve Long-Evans male rats ( Charles River Laboratories ) , with age 3–12 months and weight 300–600 grams . The rats were anesthetized with an intraperitoneal ( IP ) injection of a solution of fentanyl ( Fentanest: 0 , 3 mg/kg; Pfizer ) and medetomidin ( Domitor: 0 , 3 mg/kg; Orion Pharma ) . Body temperature was maintained at 37 . 5°C by a feedback-controlled heating pad ( Panlab , Harvard Apparatus ) . Heart rate and oxygen level were monitored through a pulse oximeter ( Pulsesense-VET , Nonin ) , and a constant flow of oxygen was delivered to the rat to prevent hypoxia . The anesthetized animal was placed in a stereotaxic apparatus ( Narishige , SR-5R ) . Following a scalp incision , a craniotomy was performed over the left hemisphere ( ~1 . 5 mm wide in diameter ) and the dura was removed to allow the insertion of the electrode array . Stereotaxic coordinates for V1 recordings ranged from −5 . 16 to −7 . 56 mm anteroposterior ( AP ) , with reference to bregma; for extrastriate areas ( LM , LI and LL ) , they ranged from −6 . 42 to −7 . 68 mm AP . The exposed brain surface was covered with saline to prevent drying . The eyes were protected from direct light and prevented from drying by application of the ophthalmic solution Epigel ( Ceva Vetem ) . Once the surgical procedure was completed , the rat was maintained in the anesthetized state by continuous IP infusion of the fentanyl/medetomidin solution ( 0 , 1 mg/kg/h ) . The level of anesthesia was periodically monitored by checking the absence of tail , ear and paw reflex . The right eye of the animal was immobilized using an eye-ring anchored to the stereotaxic apparatus ( the left eye was covered with black tape ) , with the pupil’s orientation set at 0° elevation and 65° azimuth . The stereotax was positioned , so as to align the eye with the center of the stimulus display , and was rotated leftward of 45° , so as to bring the binocular field of the right eye to cover the left side of the display . Recordings were performed with different configurations of 32-channel silicon probes ( NeuroNexus Technologies ) . To maximize the coverage of the monitor by V1 RFs , neurons in this area were recorded using 8-shank arrays with either 177 µm2 site area and 100 µm spacing ( model A8 × 4-2mm100-200-177 ) or 413 µm2 site area and 50 µm spacing ( model A8 × 4-2mm50-200-413 ) , and 4-shank arrays with 177 µm2 site area and 100 µm spacing ( model A4 × 8–5 mm-100-200-177 ) . To map the retinotopy along extrastriate areas ( Figure 1C ) , recordings from LM , LI and LL were performed using single-shank probes with either 177 µm2 site area and 25 µm spacing ( model A1 × 32-5mm25-177 ) or 413 µm2 site area and 50 µm spacing ( model A1 × 32-5mm50-413 ) . For V1 , the probe was inserted perpendicularly to the cortex ( Figure 1—figure supplement 1C ) , while , for the lateral areas , it was tilted with an angle of ~30° ( Figure 1A and Figure 1—figure supplement 1A–B ) . For the probes with 25 µm spacing , only half of the channels ( i . e . , either odd or even ) were used , to avoid considering as different units the same neuron recorded by adjacent sites . To allow the histological reconstruction of the electrode insertion track , the electrode was coated , before insertion , with Vybrant DiI cell-labeling solution ( Life Technologies ) , and , at the end of the recording session , an electrolytic lesion was performed , by passing a 5 µA current for 2 s through the last 2 ( multi-shank probes ) or 4 ( single-shank probe ) channels at the tip of each shank ( see below for a detailed description of the histological procedures ) . To decide how many neurons to record in each area , we took inspiration from previous population coding studies that have compared different ventral stream areas in terms of their ability to support object recognition ( Rust and Dicarlo , 2010; Pagan et al . , 2013 ) . These studies show that pairwise area comparisons become macroscopic when the size of the neuronal population in each area approaches 100 units . Therefore , in our experiments , we aimed at recording more than 100 units for each of the four visual areas under investigation . The final number of units obtained per area depended on the yield of each individual recording session ( reported in Figure 1—source data 1 ) and on how accessible to recording any given area was – e . g . , recordings from the deepest area ( LL ) were the most challenging and , since LI and LL were typically recorded simultaneously ( see Figure 1—source data 1 ) , we collected a large number of units from LI in the attempt of adequately sampling LL . The final number of units recorded in each area ranged from 131 ( LM ) to 260 ( LI ) ( note that these numbers refer to the visually driven and stimulus informative units; see below for an explanation ) Extracellular signals were acquired using a system three workstation ( Tucker-Davis Technologies ) with a sampling rate of 25 kHz and were filtered from 0 . 3 to 5 kHz . Action potentials ( spikes ) were detected and sorted for each recording site separately , using Wave Clus ( Quiroga et al . , 2004 ) in MATLAB ( The MathWorks ) . Spike isolation quality was assessed using two apposite metrics ( see below ) . Neuronal responses were quantified by counting spikes in neuron-specific spike-count windows ( e . g . , see gray patches in Figure 2A–B ) , with a fixed duration of 150 ms and an onset that was equal to the latency of the neuronal response , so as to capture most of the stimulus-evoked activity . The latency was estimated as the time , relative to the stimulus onset , when the response reached 20% of its peak value [for a full description see ( Zoccolan et al . , 2007 ) ] . At the end of the recording session , each animal was deeply anesthetized with an overdose of urethane ( 1 . 5 gr/kg ) and perfused transcardially with phosphate buffer saline ( PBS ) 0 . 1 M , followed by 4% paraformaldehyde ( PFA ) in PBS 0 . 1 M , pH 7 . 4 . The brain was extracted from the skull and postfixed overnight in 4% PFA at 4°C . After postfixation , the tissue was cryoprotected by immersion in 15% w/v sucrose in PBS 0 . 1 M for at least 24 hr at 4°C , and then kept in 30% w/v sucrose in PBS 0 . 1 M , until it was sectioned coronally at either 20 or 30 µm thickness on a freezing microtome ( Leica SM2000R , Nussloch , Germany ) . Sections were mounted immediately on Superfrost Plus slides and let dry at room temperature overnight . A brief wash in distilled water was performed , to remove the excess of crystal salt sedimented on the slices , before inspecting them at the microscope . For each slice , we acquired three different kinds of images , using a digital camera adapted to a Leica microscope ( Leica DM6000B-CTR6000 , Nussloch , Germany ) . First , various bright field photographs were taken at 2 . 5X and 10X magnification , so as to fully tile the region of the slice ( left hemisphere ) where the visual cortical areas were located . Second , for each of such bright field pictures , a matching fluorescence image was also acquired ( using a red filter with emission at 700 nm ) , to visualize the red fluorescence track left by the insertion of the probe ( that had been coated with Vybrant DiI cell-labeling solution before the insertion ) ( DiCarlo et al . , 1996; Blanche et al . , 2005 ) . Following the acquisition of this set of images , the slices were further stained for Nissl substance ( using the Cresyl Violet method ) and pictures were taken at 2 . 5X and 10X magnification . In addition , to better visualize the anatomic structures within the slice , a lower magnification picture of the entire left hemisphere was also taken , using a Canon 6D digital camera with a Tamron 90 mm f/2 . 8 macro lens . The fluorescence , bright field , and Nissl-stained images were processed using Inkscape ( an open source SVG editor; http://www . inkscape . org ) , so as to reconstruct the anteroposterior coordinate of the probe insertion track and , when possible , the laminar location of the recording sites along the probe . This was achieved in three steps . First , each pair of matching fluorescence and bright field images were superimposed to produce a single picture , showing both the insertion track of the probe and the anatomical structure of the section . Then , these pictures and the Nissl-stained images were aligned by matching anatomical landmarks ( e . g . , the margins of the section ) . Finally , for each section , the mosaic of matched fluorescence , bright field and Nissl images were stitched together ( by relying , again , on anatomical landmark ) , and then superimposed to the low-magnification image taken with the Canon digital camera . In the resulting image ( see examples in Figure 1—figure supplement 1 ) , the position of the shank ( s ) of the probe was drawn ( thick black lines in Figure 1—figure supplement 1 ) , by tracing the outline of the fluorescent track , and taking into account , when available , the location of the electrolytic lesion performed at the end of the recording session . Based on the known geometry of the silicon probes , the location of each recording site was also drawn over the shank ( s ) ( yellow dots over the black lines in Figure 1—figure supplement 1 ) . In the final image , the boundaries between the cortical layers ( red lines in Figure 1—figure supplement 1 ) were identified , based on the difference in size , morphology and density of the Nissl-labeled cells across the cortical thickness . This allowed estimating the laminar location of the recording sites . Since the number of neurons sampled from each individual cortical layer was not very large , in our analyses , we grouped the units recorded from layers II/III and IV and those recorded from layers V and VI ( see Figure 6C and Figure 3—figure supplement 1 ) . In some sessions , multiple recordings blocks were performed at different depths along the same probe insertion . In these cases , it was not always possible to histologically reconstruct the probe location within each block , given that a single fluorescent track was often observed , without clean-cut interruptions that could serve as landmarks for the individual blocks . In all such cases , the laminar position of the recorded units was not assigned . This explains why , in each area , the sum of the layer-labeled cells reported in our analyses is not equal to the total number of recorded units ( e . g . , compare the numbers of neurons reported in Figure 6B to those reported in Figure 6C ) . Since , in multi-block recording sessions , the probe reconstruction was especially difficult for the initial ( more superficial ) block , LM ( the first area to be recorded during oblique penetrations ) was the area in which , for a relative large fraction of neurons , we did not assign a laminar position . Our stimulus set consisted of 380 visual object conditions , obtained by producing 38 distinct views ( or transformations ) of 10 different objects , using the ray tracer POV-Ray ( http://www . povray . org/ ) . The objects were chosen , so as to span a range of shape features and low-level properties ( e . g . , luminance ) . Six of them ( #1–6 ) were computer-graphics reproductions of real-world objects , while the remaining four ( #7–10 ) were artificial shapes , originally designed for a behavioral study assessing invariant visual object recognition in rats ( Tafazoli et al . , 2012 ) . Figure 1B ( top ) shows these objects , as they appeared in the pose that we defined as default , with regard to the rotation parameters ( i . e . , 0° in-plane and in-depth rotation ) , and at their default luminosity ( i . e . , 100% luminance ) . Other default parameters were the default size ( 35° of visual angle ) and two default azimuth positions ( −15° and +15° of visual angle ) over the stimulus display . Each of these default parameters was held constant , when some other parameter was changed to generate the object transformations , as detailed below . The elevation of the stimuli was fixed at 0° of visual angle and never varied . Size was computed as the diameter of an object’s bounding circle . The transformations that each object underwent were the following . Positions changes: each object was horizontally shifted across the stimulus display , from −22 . 5° to +30° of visual angle ( azimuth ) , in steps of 7 . 5° ( Figure 1B . 1 ) . These numbers refer to the center of the display ( 0° ) , which was aligned to the position of the right eye of the rat ( i . e . , a hypothetical straight line passing through the eye and perpendicular to the monitor would hit it exactly in the middle ) . The pose , luminance and size of the objects were the default ones . Size changes: each object was scaled from 15° to 55° of visual angle , in steps of 10° ( Figure 1B . 2 ) . The pose , luminance and positions ( −15° and +15° ) of the objects were the default ones . In-plane rotations: each object was in-plane rotated from −40° to +40° , in steps of 20° ( Figure 1B . 3 ) . The in-depth rotation , luminance , size and positions of the objects were the default ones . In-depth rotations: each object was presented at five different in-depth ( azimuth ) rotation angles: −60° , −40° , 0° , + 40° , and +60° ( Figure 1B . 4 ) . The in-plane rotation , luminance , size and positions of the objects were the default ones . Luminance changes: each object was presented at four different luminance levels: 100% ( i . e . , default luminance ) , 50% , 25% and 12 . 5% ( Figure 1B . 5 ) . The pose , size and positions of the objects were the default ones . The object conditions were presented in rapid sequence ( 250 ms stimulus on , followed by 250 ms of blank screen ) , randomly interleaved with the drifting bars used to map the neuronal RFs ( Figure 1—figure supplement 2A–B ) . During a recording session , the presentation of each condition was repeated a large number of times . For V1 neurons , each conditions was presented , on average , 26 ± 2 times; for LM neurons , 25 ± 4 times; for LI neurons , 27 ± 3 times; and for LL , neurons 28 ± 2 times . This allowed obtaining a good estimate of the conditional probability P ( R\|O & T ) of measuring a given response R to any combination of object identity O and transformation T . The stimuli were displayed on a 47 inch LCD monitor ( Sharp PN-E471R ) , with 1920 × 1080 pixel resolution , 60 Hz refresh rate , 9 ms response time , 700 cd/m2 maximum brightness , 1 . 200:1 contrast ratio , positioned at a distance of 30 cm from the right eye , spanning a visual field of 120° azimuth and 90° elevation . To avoid distortions of the objects’ shape , the stimuli were presented under a tangent screen projection ( see next section and Figure 1—figure supplement 2D–F ) . As explained above , the stimulus display was positioned at a distance of 30 cm from the rat eye . This number was chosen , because the optimal viewing distance for rats has been reported to range between 20 and 30 cm ( Wiesenfeld and Branchek , 1976 ) . In addition , earlier behavioral studies from our group have shown that rats are capable of discriminating complex visual objects under a variety of identity-preserving transformations , when viewing the objects at a distance of 30 cm ( Zoccolan et al . , 2009; Tafazoli et al . , 2012; Alemi-Neissi et al . , 2013; Rosselli et al . , 2015 ) . However , because of such a short viewing distance , objects will appear distorted , when they undergo large translations over the stimulus display . Because of this , position changes will also result in size changes and distortions of the objects’ shape , unless appropriate corrections are applied . To address this issue , we displayed the stimuli under a tangent screen projection . This projection allows presenting the stimuli as they would appear , if they were shown on virtual screens that are tangent to a circle centered on the rat’s eye , with a radius equal to the distance from the eye to the point on the display just in front of it ( i . e . , 30 cm ) . Thanks to this projection , the shape , size and aspect ratio of each stimulus were preserved across all the eight azimuth positions tested in our experiment ( see previous section ) . The tangent projection is explained in Figure 1—figure supplement 2D–F . Panels D and E show , respectively , a top view and a side view of the rat eye and the stimulus display ( O1 ) . R is the distance between the eye and the center of the display . O2 is an example ( virtual ) tangent screen , where an object would be shown , if its center was translated of an azimuth angle θ to the left of the center of the stimulus display ( while maintaining the default elevation of 0° ) . The coordinate x0 indicates the projection of the object’s center over O1 , following this azimuth shift θ . The Cartesian coordinates ( x2 , y2 ) indicate the position of a pixel of the stimulus image , relative to the object’s center , over the virtual screen O2 , while ( x1 , y1 ) indicate the projection of this point over the display O1 , i . e . , its coordinates , relative to the center of the object ( x0 , 0 ) in O1 . The red line drawn over O2 shows the distance of this pixel from the object’s center over the tangent screen , while the red line drawn over O1 is the projection of this distance over the stimulus display . The projection of any point ( x2 , y2 ) in the virtual tangent screen O2 to a point ( x1 , y1 ) in the stimulus display O1 can be computed using simple trigonometric relationships . Figure 1—figure supplement 2D shows how x1 can be expressed as a function of x2 and θ:x1=R tan⁡ ( ϑ+φ ) −x0 , where:x0=R tan⁡ ( ϑ ) andφ=tan−1⁡ ( x2/R ) . Note that φ is the azimuth position of the point ( x2 , y2 ) , when expressed in spherical coordinates . Similarly , Figure 1—figure supplement 2E shows how y1 can be expressed as a function of y2 and θ:y1=y2R1R2 where:R2=Rcos⁡φandR1=Rcos⁡ ( φ+ϑ ) . To better illustrate the effect of the tangent screen projection , Figure 1—figure supplement 2D shows how two points at the same distance from ( but on opposite sides of ) the object’s center in the tangent screen ( see , respectively , the red and green lines over O2 ) would be projected over the stimulus display ( see , respectively , the red and green lines over O1 ) . As shown by the drawing , these two points would not be equidistant any longer from the object’s center , in the projection over O1 . This can be better appreciated by looking , in Figure 1—figure supplement 2F ( top ) , at the images of an example object ( object #8 ) shown at positions −22 . 5° , −15° , −7 . 5° and 0° of visual angle ( azimuth ) over the stimulus display ( these are a subset of the eight different positions tested for each object in our experiment; see above ) . The distortion applied to the object by the tangent screen projection becomes progressively larger and more asymmetrical ( with respect to the vertical axis of the object ) , the larger is the distance of the object’s center from the center of the display ( 0° azimuth ) . Critically , this distortion was designed to exactly compensate the one produced by the perspective projection to the retina , so that , regardless of the magnitude of the azimuth displacement , the resulting projection of an object over the retina will have the same shape , size and aspect ratio as the ones produced by the object shown at center of the display , right in front of the rat’s eye ( Figure 1—figure supplement 2F , bottom ) . Throughout our study , data analysis was restricted to those neurons that met two criteria: ( 1 ) being visually driven; and ( 2 ) being stimulus informative . The first criterion was implemented as following . Given a neuron , its response to each of the 380 object conditions and its background firing rate were computed . A 2-tailed t-test was then applied to check if at least one of these conditions evoked a response that was significantly different ( either larger or lower ) than background ( p<0 . 05 with Bonferroni correction for multiple comparison ) . The second criterion was based on the computation of the mutual information between stimulus and response I ( R;S ) ( defined below ) . Note that , in this case , only 230 object conditions S were used ( see next section ) . To be included in the analysis , a neuron had to carry a significant amount of information I ( R;S ) ( p<0 . 05; permutation test; described below ) . The number of neurons that met these two criteria from each area per rat is reported in Figure 1—source data 1 . As explained above , 38 different transformations of each object were presented during the experiment . These included eight positions across the horizontal extent ( azimuth ) of the stimulus display , plus four sizes , four in-plane rotations , 4-in-depth rotations and three luminance levels , presented at two of the eight positions ( i . e . , −15° and +15° azimuth ) . For any given neuron , all the eight positions were included in the single-cell mutual information and decoding analyses ( Figures 3–7 ) , while , of the remaining transformations , only those shown at the position that was closer to the neuron’s RF center ( i . e . , either −15° or +15° ) were used . This yielded 23 different transformations per object , for a total of 230 stimuli S . The motivation of choosing this subset of transformations was to maximize the coverage of the object conditions by the RFs of the recorded neurons . In the case of the population decoding analyses ( Figure 8 ) , 19 of these 23 transformations per object were used ( see below ) . Given a neuron , we obtained an estimate of its receptive field ( RF ) profile , using a procedure adapted from Niell and Stryker ( 2008 ) . We measured the neuronal response to 10° long drifting bars with four different orientations ( 0° , 45° , 90° , and 135° ) , presented over a grid of 6 × 11 cells on the stimulus display ( Figure 1—figure supplement 2A ) . The responses to the four orientations in each cell were averaged to obtain a two-dimensional map , showing the firing intensity at each location over the display ( Figure 1—figure supplement 2B , left ) . This raw RF map was fitted with a two-dimensional Gaussian , with independent widths ( SDs ) σx and σy ( Figure 1—figure supplement 2B , right ) . The RF size ( diameter ) was computed as the average of σx and σy ( see the black ellipse in Figure 1—figure supplement 2B , left ) . The goodness of the fit was measured by the coefficient of determination R2 , defined as:R2=1−SSresidSStotal where SSresid is the sum of the squared residuals of the fit and SStotal is the sum of the squared differences from the mean of the raw RF values . A fit was considered acceptable if R2 was larger than 0 . 5 . To estimate how much luminous energy any given stimulus ( i . e . , object condition ) impinged on a neuronal RF , we defined a metric ( which we called RF luminance ) , resulting from computing the dot product between the raw RF map of the neuron ( normalized to its maximal value , so as to range between 0 and 1 ) and the luminance profile of the stimulus over the image plane . This is equivalent to weight the stimulus image by the RF profile of the neuron ( Figure 1—figure supplement 2C provides a graphical description of this procedure ) . This quantity was then normalized by the maximal luminance that could possibly fall into the neuron’s RF ( corresponding to the case of a full-field stimulus at maximal brightness ) , hence the percentage values reported in Figure 2A–B ( white font ) and Figure 2—figure supplement 1 . A similar approach was used to measure the variability of the pattern of luminance impinged by a stimulus on a neuronal RF , using a metric that we called RF contrast . This metric was defined as the standard deviation of the RF-weighted luminance values produced by the stimulus that were contained within the RF itself ( see the rightmost plot in Figure 1—figure supplement 2C ) . Note that , in general , since the neuronal RFs did not fully cover all the object conditions ( see examples in Figure 2A–B ) , also the background contributed to the RF contrast metric . Therefore , in the analysis in which we assessed the amount of information carried by single neurons about RF contrast ( Figure 3—figure supplement 4 ) , we only considered object conditions that covered at least 10% of the RFs . This restriction was applied because we wanted to measure how neurons coded the contrast over a large enough surface of the visual objects . In addition , since the number of stimuli that met this criterion was different for the various neurons , we included in the analysis only neurons for which at least 23 object conditions fulfilled this constraint ( this explains why the number of neurons reported in Figure 3—figure supplement 4A is lower than the total number of units recorded in each area; e . g . , compare to Figure 3B ) . This insured that the information about RF contrast was estimated with enough visual stimuli . The quality of spike isolation was assessed through two widely used benchmarks: signal-to-noise ratio ( SNR ) and refractory violations ( RV ) ( Quiroga et al . , 2005 , 2008; Gelbard-Sagiv et al . , 2008; Hill et al . , 2011 ) . These metrics were defined as following . SNR=AsignalAnoise , where Asignal is the average peak-to-peak amplitude of all the spikes detected as a single cluster by the sorting algorithm , and Anoise is an estimate of the variability of the background noise computed from the rest of the filtered signal ( in our application , this was the median of the absolute value of the filtered signal , divided by 0 . 6745 [Quiroga et al . , 2004] ) . RV was defined as the fraction of inter spike intervals ( ISI ) that were shorter than 2 ms , the rationale being that a large RV indicates a substantial violation of the neuron’s refractory period and , therefore , a contamination from other units ( Hill et al . , 2011 ) . These metrics were used in two ways . First , the neurons recorded from the four areas were pooled together and the resulting distributions of SNR and RV values were divided in three equi-populated bins ( tertiles ) . Within each tertile , the mutual information and generalization metrics were re-computed , so as to compare neuronal subpopulations within the same range of isolation quality ( Figure 3—figure supplement 2A–B and Figure 6—figure supplement 2A–B ) . Second , we selected from the four areas only neurons with good isolation quality , as defined by imposing both SNR >10 and RV <2% , and , again , we restricted the computation of the mutual information and generalization metrics to these subpopulations ( Figure 3—figure supplement 2C and Figure 6—figure supplement 2C ) . These constraints also equated the neuronal subpopulations in terms of firing rate ( Figure 6—figure supplement 3B ) , which , otherwise , would significantly decrease from V1 to the more lateral areas ( Figure 6—figure supplement 3A ) . All the single-cell properties measured for the visual areas have been reported in terms of medians ± Standard Error ( SE ) of the median ( estimated by bootstrap ) . In all these cases , the statistical significance of each pairwise , between-area comparison was assessed with a 1-tailed , Mann-Whitney U-test , with Holm-Bonferroni correction for multiple comparisons . The choice of the 1-tailed test was motivated by the fact that , in any comparison , we had a clear hypothesis about the rank of each visual area , with respect to the measured property , based on previous anatomical , lesion and functional studies ( see Introduction and Discussion ) . The mutual information I ( R;S ) between stimulus S and response R was computed according to eq . 1 ( see Results ) , and its meaning is graphically illustrated in Figure 3A . Here , we report some more technical details about how the calculation of the information metrics used through the study ( i . e . , see Equation 1 , 2 and 3 ) was carried out . In all the information theoretic analyses , the response R was quantified as the number of spikes fired by a neuron in a 150 ms-wide spike count window ( see above ) , discretized into three equi-populated bins ( whose boundaries were computed , for each neuron independently , on the whole set of responses collected across repeated presentations of all the stimuli ) . Given that we recorded an average of 26 . 5 trials per stimulus ( see above ) , this discretization yielded about nine trials per stimulus per response bin . Earlier studies ( Panzeri and Treves , 1996; Panzeri et al . , 2007 ) showed that the limited sampling bias in information calculations can be effectively corrected out – leading to precise information estimations – if there are at least four repetitions per stimulus and per response bin . Quantizing responses in three bins was thus appropriate to obtain highly conservative and unbiased information estimates . Yet , we explored a range of different bin numbers within the extent for which the bias could reliably be corrected , and we found similar patterns of mutual information values across the four visual areas . As shown in the Results ( see Equation 2 and Figure 3B ) , I ( R;S ) can be rewritten , using conditional mutual information ( Ince et al . , 2012 ) , as the sum of the luminance information I ( R;L ) and the higher-level , luminance-independent information I ( R;S′\|L ) . Since , for each object and neuron , the maximum number of possible luminance values L is equal to the number of unique transformations the objet underwent ( i . e . , 23 ) , L was discretized in 23 bins ( see Figure 2C–D ) . This number of luminance bins fully covered the luminance variation for each object , without loss of luminance information , and it yielded , on average , 100 trials per response and stimulus bin , which was comfortably sufficient to control for the sampling bias in the computation of the mutual information ( Panzeri et al . , 2007 ) . The same arguments apply to the analysis in which I ( R;S ) was decomposed as the sum of the contrast information I ( R;C ) and the contrast-independent information I ( R;S′\|C ) ( see Figure 3—figure supplement 4 ) . To evaluate the statistical significance of mutual information values we used a random permutation method ( Ince et al . , 2012 ) . In this method , separately for each considered information calculation and for each neuron , the association between stimulus and response was randomly permuted 100 times , to generate a null-hypothesis distribution of information values in the absence of any association between stimulus and response . The significance levels ( p values ) of the mutual information values were obtained from this null-hypothesis distribution , as explained in ( Ince et al . , 2012 ) . All information measures were computed using the Information Breakdown Toolbox ( Magri et al . , 2009 ) and were corrected for limited sampling bias using the Panzeri-Treves method ( Panzeri and Treves , 1996; Panzeri et al . , 2007 ) . This method uses an asymptotic large-N expansion ( where N is the total number of trials available across all stimuli ) to estimate and then subtract out the limited sampling bias from the raw information estimate . The asymptotic estimation of the bias has the following expression: BIAS=12Nln2[∑s ( Rs−1 ) − ( R−1 ) ] where Rs and R are the number of bins with non-zero stimulus-specific and stimulus-unspecific response probabilities , respectively . These numbers of bins are estimated from the data with a Bayesian procedure , as described in Panzeri and Treves ( 1996 ) . This analysis was meant to assess the ability of single neuronal responses to support the discrimination of a pair of objects in spite of changes in their appearance . Given a neuron , we defined the following variables . R is the neuron’s response , measured by counting the number of spikes fired in a given spike count window , following stimulus presentation ( see above for the choice of the spike count window ) . O denotes the identity of a visual object , among the set of 10 available objects , i . e . , O={ o1 , o2 , … , o10 } . T denotes the specific transformation an objet underwent , among the set of 23 transformations ( i . e . , views ) available for that neuron ( see above for an explanation of how these conditions were chosen for each neuron ) , i . e . , T={ t1 , t2 , … , t23 } . For each neuron , we selected all possible pairs of objects oi and oj , and , for each pair , we performed two kinds of analysis: ( 1 ) a test of linear separability of object representations across transformations ( Figure 5 ) ; and ( 2 ) various tests of generalization of the discrimination of the two objects across transformations ( Figure 6 ) . The results obtained for each object pair were then averaged , to estimate the linear separability and generalizability afforded by each neuron . Both procedures are detailed below . This analysis was carried out following a 5-fold cross-validation procedure . Given a neuron and a pair of objects oi and oj , we randomly partitioned the set of responses of the neuron to the repeated presentation of each object ( across all 23 tested views ) in five subsets . In each cross-validation loop , only four of these subsets were used as the training data to build the decoder , while the remaining subset was used as the test data to measure the decoder’s performance at discriminating the two objects . More formally , we considered the conditional probabilities of observing a response R to the presentation of any view of each object , i . e . , Ptrain ( R=r\|O=oi & T={t1 , t2 , … , t23} ) and Ptrain ( R=r\|O=oj & T={t1 , t2 , … , t23} ) , where the ‘train’ subscript indicates that only 4/5 of the responses ( those belonging to the training data set ) were used to compute the conditional probabilities . We then used these probabilities to find a boundary , along the spike count axis , that allowed the discrimination of all the views of an object from all the views of the other object . To avoid overfitting the decoder to the training data , we minimized the assumptions on the shape of the conditional probabilities , and we simply parameterized them by computing their means μi and μj and variances σi and σj ( with μi≥μj and σi≠σj ) . We then used these parameters to find the discrimination boundary , by applying quadratic discriminant analysis ( Hastie et al . , 2009 ) . That is , the discrimination boundary was defined as the threshold response rth , obtained by setting to zero the log-ratio of the posterior probabilities , i . e . : ( 4 ) log⁡Ptrain ( O=oi & T={t1 , t2 , … , t23}\|R=rth ) Ptrain ( O=oj & T={t1 , t2 , … , t23}\|R=rth ) =0 Having fixed this decision boundary , the decoder was then tested for its ability to correctly classify the remaining 1/5 of the responses belonging to the test set , according to the following binary classification rule: ( 5 ) O={oi , if R>rthoj , if R<rth This procedure was repeated for all possible combinations of the five subsets of responses in four train sets and one test set , with each combination yielding a distinct cross-validation loop . The outcomes of the classification , obtained across the resulting five cross-validation loops , were collected in a confusion matrix , which was then used to compute the performance of the decoder with that specific object pair . As explained in the Results , the performance was computed as the mutual information between the actual and the predicted object labels from the decoding outcomes [i . e . , as the mutual information between the rows and columns of the confusion matrix ( Quiroga and Panzeri , 2009 ) ] . This computation was performed for all possible object pairs . The resulting performances were then averaged to obtain the linear separability afforded by the neuron . Note that , to prevent large decoding performances from being trivially achieved only based on luminance differences , only those pairs with objects having a luminosity ratio larger than a given threshold ThLumRatio ( see Results ) were included in the final average . Given a neuron and a pair of objects oi and oj , we first trained a binary classifier to partition the spike count axis with a boundary that allowed the discrimination of two specific transformations tx and ty of the two objects . To this aim , we considered the conditional probabilities of observing a response R to the presentation of the two object views , i . e . , P ( R=r\|O=oi & T=tx ) and P ( R=r\|O=oj & T=ty ) . Note that these conditional probabilities were obtained by taking into account all the responses of the neuron to the repeated presentation of that specific combination of object identity O and transformation T . Similarly to what done in the linear separability analysis , the conditional probabilities were parameterized by computing their means μix and μjy and variances σix and σjy ( with μix≥μjy and σix≠σjy ) , and these parameters were used to find the discrimination boundary , by applying quadratic discriminant analysis ( Hastie et al . , 2009 ) . That is , the discrimination boundary was defined as the threshold response rth , obtained by setting to zero the log-ratio of the posterior probabilities , i . e . : ( 6 ) log⁡P ( O=oi & T=tx\|R=rth ) P ( O=oj & T=ty\|R=rth ) =0 Having fixed this decision boundary , the decoder was then tested for its ability to correctly classify the responses of the neuron to different transformations ( i . e . , T≠tx and T≠ty ) of the same two objects oi and oj , according to the following binary classification rule: ( 7 ) O={oi , if R>rthoj , if R<rth Following this general scheme , the cross-validation procedure worked as following . Given a neuron , and given a pair of objects , we randomly sampled , independently for each object , one of the 23 transformations available for that neuron . The spike count distributions produced by these transformations were used to build the decision boundary rth , as defined in Equation ( 6 ) . We then used the responses to the repeated presentation of the remaining 22 transformations of each object as the test set to measure the generalization performance of the decoder , according to the classification rule defined in Equation ( 7 ) . For each object pair , this procedure was repeated 1000 times . In each of these runs , the training transformations of the two objects were randomly sampled , so as to span a wide range of training and testing object views . This yielded an average generalization performance per object pair . This computation was performed for all possible pairs . The resulting performances were then averaged to obtain the generalization performance afforded by the neuron . As for the linear separability analysis , the pairs included in the final average were only those with objects having a luminosity ratio larger than a given threshold ThLumRatio . Two different versions of this analysis were carried out . In the first one , the training and testing views were randomly sampled from the full set of 23 transformations available for each neuron ( Figure 6B–C ) . In the second one , the training and testing views were randomly sampled from the individual variation axes: position , size , in-plane and in-depth rotations and luminance ( Figure 6D ) . In the case of the transformations along the position axis , a second kind of analysis was also performed . This analysis was meant to assess how sharply the generalization performance of the binary decoders decayed as a function of increasingly large position changes , thus providing an estimate of the spatial extent of the invariance afforded by a neuron over the visual field ( Figure 6—figure supplement 1A ) . Given a neuron , and given a pair of objects , we selected one of the eight azimuth positions tested during the experiment ( Figure 1B , bottom ) . Note that here , differently from the cross-validation procedure described above , the same transformation ( i . e . , the same azimuth position ) was selected for both objects in the pair . We used the spike count distributions produced by the two objects at this specific position to build the decision boundary rth , as defined in Equation 6 . We then tested the ability of this decision boundary to correctly discriminate the two objects presented at testing positions that were increasingly further apart from the training location . That is , for each testing position , we took the responses of the neuron to the repeated presentation of the two objects at that position , and we measured the generalization performance of the decoder , according to the classification rule defined in Equation 7 . The testing positions were chosen either to the left or to the right of the selected training position , depending what direction allowed spanning at least 30° of the visual field . Specifically , for training positions between −22 . 5° and 0° of visual angle , four testing positions were taken to the right of the training location; vice versa , for training positions between +7 . 5° and +30° of visual angle , four testing positions were taken to the left of the training location . For any training position , this procedure yielded a curve , showing how stable the discriminability of the objects in the pair was across a span of 30° over the visual field . For any given object pair , this procedure was repeated across all possible training positions , starting from the leftmost one ( at −22 . 5° ) , up to the rightmost one ( +30° ) . The resulting generalization performances were averaged to yield the mean generalization curve over the position axis for that specific object pair . This computation was then repeated for all possible pairs , and the resulting performances were averaged to obtain the generalization curve over the position axis afforded by the neuron . The pairs included in this final average were those with objects having a luminosity ratio larger than 0 . 9 ( i . e . , with ThLumRatio=0 . 9 ) . In addition , to obtain a cleaner estimate of the dependence of position tolerance from the magnitude of the translation , the analysis was restricted to the neurons with unimodal and elliptical RF profiles , which were well fitted by 2-dimensional Gaussians ( i . e . , those neurons contributing to the RF statistics shown in Figure 7A ) . The resulting median generalization curves obtained for the four visual areas are shown in Figure 6—figure supplement 1A . A similar kind of analysis was performed for the transformations along the size axis ( see Figure 6—figure supplement 1B ) . The cross-validation procedure was similar to the one described in the previous paragraph for the position changes . However , since the size conditions were fewer than the position ones ( Figure 1B , bottom ) , and lent themselves to be analyzed in terms of generalization from small to larger sizes ( and vice versa ) , we grouped them into three classes: ( 1 ) the default size ( i . e . , 35° ) ; ( 2 ) the two sizes than were smaller than 35° ( i . e . , 15° and 25° ) ; and 3 ) the two sizes than were larger than 35° ( i . e . , 45° and 55° ) . Given an object pair , we tested all possible generalizations from the default size to the smaller and larger sizes and , vice versa . In the first case , we used the spike count distributions produced by presenting the two objects at size 35° to build the decision boundary rth , as defined in Equation 6 . We then tested the ability of this decision boundary to correctly discriminate the two objects presented at each of the other sizes . That is , for each testing size , we took the responses of the neuron to the repeated presentation of the two objects at that size , and we measured the generalization performance of the decoder , according to the classification rule defined in Equation 7 . This procedure yielded four generalization performances: two from size 35° to larger sizes , and two from size 35° to smaller sizes . A similar approach was used to compute the generalization from the other sizes to the default one . Also in this case , four generalization performances were obtained: two from smaller sizes to size 35° , and two from larger sizes to size 35° . Overall , the full procedure yielded eight generalization performances per object pair . These performances were divided into two groups . The first group included the generalization from size 35° to smaller sizes and from larger sizes to size 35° . The second group included the generalization from size 35° to larger sizes and from smaller sizes to size 35° . The four performances within each group were averaged to yield the final large→small and small→large generalization performances for the object pair . This procedure was repeated for each pair and the resulting performances were averaged to estimate the generalization afforded by the neuron , when object size changed from large to small and vice versa . The pairs included in these final averages were those with objects having a luminosity ratio larger than 0 . 9 ( i . e . , with ThLumRatio=0 . 9 ) . The resulting median generalization performances obtained for the four visual areas are shown in Figure 6—figure supplement 1B . In the cross-validation procedures described above ( for both the tests of linear separability and generalization ) , each step yielding a classification performance also yielded a chance generalization performance . This was computed by randomly permuting the object identity labels associated to the responses of the neuron before building the decoder . While carrying out this permutation , care was taken to randomly shuffle only the label O , which specifies object identity , and not the label T , which specifies the transformation the objet underwent . This means that , given a pair of objects oi and oj , the identity labels oi and oj were randomly shuffled among the responses produced by matching views of the two objects ( i . e . , the shuffling was performed , separately , for each of the 23 transformations the two objects underwent ) . This was done to isolate the contribution of the only variable of interest ( i . e . , object identity ) to the performances attained by the classifiers . For any given neuron , chance performances were obtained for all object pairs , and were then averaged , according to the same procedures described above , to yield the final chance decoding performance of the neuron . These chance performances were used in two ways . First , to obtain conservative estimates of the ability of the decoders to generalize to new object views , the chance performances were subtracted from the actual ( i . e . , un-shuffled ) classification performances . This allowed measuring and reporting , in all the single-cell decoding analyses ( Figures 5–7 ) , the net amount of transformation-tolerance afforded by single neurons in the four visual areas ( in general , these chance performances obtained by shuffling were all very close to the theoretical 0 bit chance level of the binary classification task ) . Second , the distribution of classification performances obtained for a neuronal population was compared to the matching distribution of chance performances , to assess whether the former was significantly larger than the latter ( according to a 1-tailed , Mann-Whitney U-test ) . This was the case for all the classification performances reported in Figures 5–7 . The population decoding analysis followed the same rationale of the single-cell decoding analysis . Binary linear decoders were built to assess the ability of a population of neurons in a visual area to support the discrimination of a pair of objects in spite of changes in their appearance . Again , two kinds of analyses were preformed – a test of linear separability of the object representations , and a test of generalization of the discrimination of the two objects across transformations . Following the design of previous population decoding studies of the monkey ventral stream ( Hung et al . , 2005; Rust and Dicarlo , 2010; Pagan et al . , 2013 ) , we measured whether ( and how sharply ) the decoders’ performance grew as a function of the size of the neuronal populations , by randomly sampling subpopulations of 6 , 12 , 24 , 48 and 96 units from the full sets of neurons recorded in each area . More specifically , for a given subpopulation size N , with N={ 6 , 12 , 24 , 48 , 96 } , we run 50 subpopulation resampling iterations . In each iteration , N different units were randomly chosen from the full population , and , for each stimulus condition S ( i . e . , each combination of object identity and transformation S = O&T ) , M pseudo-population response vectors were built , drawing from the responses of the N units to the repeated presentation of S ( in the cartoons of Figure 8A , a set of pseudo-population vectors corresponding to a given stimulus S is graphically illustrated as a cloud of dots with a specific color ) . Since the neurons belonging to each area were recorded across different sessions ( and , therefore , not all the neurons in a given area were recorded simultaneously; see Figure 1—source data 1 ) , the pseudo-population response vectors to stimulus S were built by assigning to each component of the vector the response of the corresponding neuron in a randomly sampled presentation of S . For each component , M of such responses were sampled without replacement , to obtain the final set of M pseudo-population vectors to be used in a given subpopulation resampling iteration . In each resampling iteration , linear separability and generalization were computed as detailed below . As in the case of the single-cell analysis , we applied a 5-fold cross-validation procedure . Given a neuronal subpopulation of size N ( obtained in a specific resampling iteration ) and a pair of objects oi and oj , we randomly partitioned the set of pseudo-population response vectors associated to the presentation of each object ( across all 23 tested views ) in five subsets . In each cross-validation loop , only four of these subsets were used as the train data to build the decoder , while the remaining subset was used as the test data to measure the decoder’s performance at discriminating the two objects . Based on the train data , a linear decoder learned to find a hyperplane that partitioned the population vector space into two semi-spaces , each corresponding to a specific object label ( i . e . , either oi or oj , ) . As a decoder , we used a Support Vector Machine ( SVM ) ( Cristianini and Shawe-Taylor , 2000 ) in its Matlab implementation , with a linear kernel . The specific method used to find the hyperplane was Sequential Minimal Optimization ( SMO ) ( Schölkopf and Smola , 2001 ) . The soft-margin parameter C was set to its default value of 1 for all the trainings . Having found the hyperplane using the train data , the decoder was tested for its ability to correctly classify the remaining 1/5 of the response vectors ( belonging to the test set ) , depending on the semi-space in which each vector was located . This procedure was repeated for all possible combinations of the five subsets of response vectors in four train sets and one test set , with each combination yielding a distinct cross-validation loop . The outcomes of the classification , obtained across the resulting five cross-validation loops , were collected in a confusion matrix , which was then used to compute the performance of the decoder with that specific object pair in that specific subpopulation resampling iteration . The performance was computed both as the mutual information between the actual and the predicted object labels from the decoding outcomes and as the classification accuracy . As mentioned above , 50 resampling iterations were run for each object pair . The resulting 50 decoding performances were then averaged to obtain the linear separability of the object pair . Finally , the performances obtained for different pairs were averaged to estimate the linear separability afforded by neuronal subpopulation . As in the case of the single-cell analysis , only those pairs with objects having a luminosity ratio larger than a given threshold ThLumRatio ( see Results ) were included in the final average . Note that , in the case of the population analysis , this constraints had to be satisfied by all the neurons included in a given subpopulation , for all the areas . Obviously , the larger the subpopulation was ( i . e . , the larger N ) , the harder was to find pairs of objects that satisfied the constraint . For this reason , it was impossible to set ThLumRatio=0 . 9 , as done in the single-cell analysis , because this choice would have yielded only a single object pair for the largest subpopulation ( i . e . , for N = 96 ) . Therefore , for the analysis shown in the main text ( Figure 8C–F ) , we set ThLumRatio=0 . 8 , which yielded three object pairs for N = 96 , and still allowed assessing linear separability in a regime where the luminance confound was kept under control . In addition , to check the generality of our conclusions , we also recomputed our analysis for ThLumRatio=0 . 6 , which yielded 23 object pairs ( Figure 8—figure supplement 1A ) . The overall scheme of the decoding procedure was similar to the one described above ( again , binary SVMs with linear kernels were used ) , with the key difference that , in any given subpopulation resampling iteration , only the response vectors of two randomly chosen transformations tx and ty of the two objects oi and oj were used to train the decoder . Following this training ( i . e . , once found the hyperplane ) , the decoder was tested for its ability to correctly classify ( i . e . , generalize to ) the response vectors of all the remaining transformations ( i . e . , T≠tx and T≠ty ) of the same two objects , and the outcomes of this classification were collected in a confusion matrix . This procedure was repeated in such a way to choose exhaustively all possible combinations of train views tx and ty . Since , in the population decoding analysis , 19 views per object were used ( i . e . , for each object , T={ t1 , t2 , … , t19 }; see below for an explanation ) , this yielded 192 confusion matrixes . The data from these matrixes were merged into a global confusion matrix , which was then used to compute the performance of the decoder with that specific object pair in that specific subpopulation resampling iteration . Again , the performance was computed both as the mutual information between the actual and the predicted object labels from the decoding outcomes and as the classification accuracy . The final generalization performance obtained for a given object pair was computed as the mean of the performances resulting from the 50 resampling iterations . Finally , the performances obtained for different pairs were averaged to estimate the generalization afforded by neuronal subpopulation . As in the case of the linear separability analysis ( see previous section ) , only those pairs with objects having a luminosity ratio larger than a given threshold ThLumRatio ( see Results ) were included in the final average . ThLumRatio was set to 0 . 8 ( for the results shown in Figure 8E–F ) and to 0 . 6 ( for the results shown in Figure 8—figure supplement 1B ) , following the same rationale explained in the previous section . As in the case of the single-cell analyses , we also computed chance classification performances ( for both the tests of linear separability and generalization ) . These were obtained by running the same decoding analyses described above , but with the key difference of randomly shuffling first the association between object identity labels and pseudo-population response vectors . This was done following the same rationale of the single-cell analysis ( see description above ) , i . e . , by randomly shuffling only the label O , which specifies object identity , and not the label T , which specifies the transformation the objet underwent . For any given neuronal subpopulation and any given object pair , chance performances were subtracted from the actual ( i . e . , un-shuffled ) classification performances . This allowed measuring and reporting the net amount of transformation-tolerance afforded by the neuronal populations in the four visual areas . This subtraction was done only for the performances computed as the mutual information between the actual and the predicted object labels from the decoding outcomes ( i . e . , results shown in Figure 8C and E , left and Figure 8—figure supplement 1A and B , left ) . For the results reported in terms of classification accuracy ( Figure 8C and E , right and Figure 8—figure supplement 1A and B , right ) , the chance performances were not subtracted from the actual performances , so as to better highlight that the theoretical chance level was 50% correct discrimination . These chance performances were all between 49 . 4% and 50 . 1% correct discrimination for the linear separability test , and between 49 . 98% and 50 . 02% correct discrimination for the generalization test . Finally , for each pair of objects oi and oj , the linear separability analysis was also repeated for arbitrary groups of views of the two objects . To perform this test , in any subpopulation resampling iteration , out of the 19 views of object oi , nine were randomly chosen and assigned to the first group , while the remaining 10 were assigned to the second group . Similarly , 10 randomly sampled views of object oj were assigned to the first group and the remaining nine to the second group . The decoding procedure was then performed as previously described . Only , in this case , the linear SVM decoders were trained to find a hyperplane in the response vector space that discriminated the set of views of the first group from the set of views of the second one . Fifty resampling iterations were carried out , each yielding a classification performance . The resulting 50 decoding performances were then averaged to obtain the linear separability of two arbitrary groups of object views , taken from that specific object pair . Finally , the performances obtained for different pairs were averaged to estimate the linear separability of arbitrary groups of object views afforded by neuronal subpopulation ( shown as dashed lines in Figure 8C and Figure 8—figure supplement 1A ) . As explained in Materials and methods , out of the 38 transformations tested for each object , 23 were chosen , for each neuron , to be used in the single-cell decoding analyses . These included all the eight position changes , plus the remaining transformations ( size and luminance changes , as well as in-plane and in-depth rotations ) shown at the position that was closer to the neuron’s RF center ( i . e . , either −15° or +15° of visual angle ) . Out of these 23 transformations , 19 were used in the population decoding analysis . In fact , for the neurons with RF center that was closer to −15° , the rightmost four positions ( i . e . , from +7 . 5° to +30° of visual angle ) were excluded . Similarly , for the neurons with RF center that was closer to +15° , the leftmost four positions ( i . e . , from −22 . 5° to 0° of visual angle ) were excluded . As a result , for the group of neurons with RFs closer to the −15° position , the 19 chosen transformations were the complementary set of the 19 transformations chosen for the group of neurons with RFs closer to the +15° position . This allowed aligning the two groups of neuronal RFs , by mapping the [−22 . 5° , 0°] range of positions of the first group onto the [ + 7 . 5° , +30°] range of positions of the second one , thus obtaining a single neuronal population , from which the neurons in each resampling iteration could be drawn .	Everyday , we see thousands of different objects with many different shapes , colors , sizes and textures . Even an individual object – for example , a face – can present us with a virtually infinite number of different images , depending on from where we view it . In spite of this extraordinary variability , our brain can recognize objects in a fraction of a second and without any apparent effort . Our closest relatives in the animal kingdom , the non-human primates , share our ability to effortlessly recognize objects . For many decades , they have served as invaluable models to investigate the circuits of neurons in the brain that underlie object recognition . In recent years , mice and rats have also emerged as useful models for studying some aspects of vision . However , it was not clear whether these rodents’ brains could also perform complex visual processes like recognizing objects . Tafazoli , Safaai et al . have now recorded the responses of visual neurons in rats to a set of objects , each presented across a range of positions , sizes , rotations and brightness levels . Applying computational and mathematical tools to these responses revealed that visual information progresses through a number of brain regions . The identity of the visual objects is gradually extracted as the information travels along this pathway , in a way that becomes more and more robust to changes in how the object appears . Overall , Tafazoli , Safaai et al . suggest that rodents share with primates some of the key computations that underlie the recognition of visual objects . Therefore , the powerful sets of experimental approaches that can be used to study rats and mice – for example , genetic and molecular tools – could now be used to study the circuits of neurons that enable object recognition . Gaining a better understanding of such circuits can , in turn , inspire the design of more powerful artificial vision systems and help to develop visual prosthetics . Achieving these goals will require further work to understand how different classes of neurons in different brain regions interact as rodents perform complex visual discrimination tasks .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Emergence of transformation-tolerant representations of visual objects in rat lateral extrastriate cortex
Economic theories posit reward probability as one of the factors defining reward value . Individuals learn the value of cues that predict probabilistic rewards from experienced reward frequencies . Building on the notion that responses of dopamine neurons increase with reward probability and expected value , we asked how dopamine neurons in monkeys acquire this value signal that may represent an economic decision variable . We found in a Pavlovian learning task that reward probability-dependent value signals arose from experienced reward frequencies . We then assessed neuronal response acquisition during choices among probabilistic rewards . Here , dopamine responses became sensitive to the value of both chosen and unchosen options . Both experiments showed also the novelty responses of dopamine neurones that decreased as learning advanced . These results show that dopamine neurons acquire predictive value signals from the frequency of experienced rewards . This flexible and fast signal reflects a specific decision variable and could update neuronal decision mechanisms . Individuals frequently make predictions about the value of future rewards and update these predictions by comparison with experienced outcomes . A fundamental determinant of reward value is reward probability ( Pascal , 2005 ) . When an environmental cue predicts reward in a probabilistic fashion , the brain needs to learn the value of such a cue from the frequency of experienced rewards . Such learning enables individuals to compute the economic value of environmental cues and thus allows forpa efficient decision making . The phasic activity of dopamine neurons encodes reward prediction error ( Schultz et al . , 1997; Bayer and Glimcher , 2005; Enomoto et al . , 2011; Cohen et al . , 2012 ) . These prediction error responses increase monotonically with the expected value of reward , including reward probability ( Fiorillo et al . , 2003; Tobler et al . , 2005 ) . Cues that predict reward with high probability evoke larger responses than cues predicting the same reward with lower probability ( Fiorillo et al . , 2003 ) . Moreover , during an economic choice task , responses of dopamine neurons and striatal dopamine concentration reflect the reward probability of the cue the animal has chosen ( Morris et al . , 2006; Saddoris et al . , 2015 ) . In these studies , neuronal responses to reward predicting cues were examined only after the animals received substantial training with the same reward-predicting cues . The responses of dopamine neurons have been also examined during learning ( Mirenowicz and Schultz , 1994; Hollerman and Schultz , 1998 ) . These studies primarily focused on how dopamine responses to rewards develop during learning of cue-reward association . This neuronal acquisition happens gradually ( Hollerman and Schultz , 1998 ) , and is well-approximated by reinforcement learning ( RL ) models ( Pan et al . , 2005 ) . Similarly , striatal dopamine concentration reflects values of probabilistically delivered rewards during learning ( Hamid et al . , 2016 ) . However , it remains unknown how learning about probabilistic rewards shapes responses of dopamine neurons to reward predicting cues , and how this neuronal learning participates in decision making . We addressed these questions by recording the activity of dopamine neurons in monkeys during the learning of novel cues predicting specific reward probabilities . We studied dopamine responses during both simple Pavlovian conditioning and during risky choices . In both tasks , dopamine responses to cues showed two distinct response components: an early component reflecting novelty , and a later component that developed during learning to encode the value of probabilistic rewards acquired from experienced reward frequencies . Reinforcement learning models served to separate these two components more formally . During choice , the acquired dopamine responses reflected the value of the chosen option relative to the unchosen option . Pavlovian conditioning is the most basic mechanism by which an organism can learn to predict rewards . This behavioural paradigm provides a straightforward platform for monitoring neuronal correlates of learning . To investigate how responses of dopamine neurons develop during learning to reflect the value of probabilistically delivered rewards , we monitored two monkeys during a Pavlovian conditioning task ( Figure 1A ) . Visual cues ( fractal pictures , never seen before ) predicted gambles between a large ( 0 . 4 ml ) and a small ( 0 . 1 ml ) juice reward , delivered 2 s after cue onset . The probability of receiving the large reward was p=0 . 25 , p=0 . 50 , or p=0 . 75; the probability of small reward was correspondingly 1 - p . In each learning block , three novel cues were differentially associated with three different reward probabilities , but only one cue was presented to the animal on each trial . This situation conforms to a learning set in which the animals learned to rapidly assign in each learning block one of the three possible probabilities to each cue ( Harlow , 1949 ) . 10 . 7554/eLife . 18044 . 003Figure 1 . Monkeys rapidly learn the value of cues that predict rewards with different probabilities . ( A ) Pavlovian task . Left: example of novel visual cues ( fractal images ) presented to monkeys . In each trial , animals were presented with a visual cue and received a large ( 0 . 4 ml ) or small ( 0 . 1 ml ) drop of juice reward 2s after cue onset . Specific cues predicted the large reward with probabilities of p=0 . 25 , p=0 . 5 and p=0 . 75 , together with small reward at 1–p . In each session of the experiment ( lasting 90–120 trials ) , three novel cues were differentially associated with the three tested reward probabilities . Over consecutive trials , cues with different reward probabilities were presented to animals pseudorandomly . Trials were separated by inter-trial intervals of 2–5 s . Animals had no specific behavioural requirements throughout this task . ( B ) Monkeys’ lick responses during Pavlovian learning . The lick responses were measured from cue onset to onset of reward delivery . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 003 The animals gradually developed differential anticipatory lick responses , measured between cue onset and reward delivery , over about 10 trials for each cue ( Figure 1B , p<0 . 01 in both animals , one-way ANOVA ) . This result suggests that the monkeys learned the value of novel sensory cues that predicted rewards with different probabilities . We recorded the responses of 38 and 32 dopamine neurons in monkeys A and B , respectively , during this Pavlovian learning task . The neuronal responses to cues showed two components ( Figure 2A ) , analogous to previous studies ( Nomoto et al . , 2010; Stauffer et al . , 2014; Schultz , 2016 ) . Specifically , an early activation at 0 . 1–0 . 2 s after cue onset most likely reflected the previously observed novelty signals ( Ljungberg et al . , 1992; Horvitz et al . , 1997; Schultz , 1998; Costa et al . , 2014; Gunaydin et al . , 2014 ) . It decreased progressively during learning blocks ( Figure 2B , Figure 2—figure supplement 1A; and 55/70 neurons; p<0 . 05 power function fit to trial-by-trial responses ) , reflecting cue repetition better than number of consecutive trials ( R2 = 0 . 86 vs . R2 = 0 . 53 , linear regression ) . This response component failed to differentiate between the cues predicting different reward probabilities ( Figure 2B , p=0 . 61 , one-way ANOVA ) . 10 . 7554/eLife . 18044 . 004Figure 2 . Responses of dopamine neurons acquire predictive value from the frequency of rewards . ( A ) Peri-stimulus time histograms ( PSTHs ) of a dopamine neuron in response to novel cues predicting rewards with different probabilities . Pink ( 0 . 1–0 . 2 s after cue onset ) and grey ( 0 . 2–0 . 6 s after cue onset ) horizontal bars indicate analysis windows used in B and C , respectively . ( B ) Decrease of neuronal population responses , measured at 0 . 1–0 . 2 s after cue onset ( pink inset ) , over consecutive learning trials . Error bars show standard error of mean ( s . e . m . ) across neurons ( n = 70 , pooled from monkeys A and B ) . ( C ) Differentiation of neuronal population responses , measured at 0 . 2–0 . 6 s after cue onset ( grey inset ) , over consecutive learning trials . The following figure supplement is available for Figure 2:DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 00410 . 7554/eLife . 18044 . 005Figure 2—figure supplement 1 . Compound novelty-value responses of dopamine neurons to novel cues associated with different probabilistic rewards . ( A ) PSTHs of dopamine population responses to novel reward predicting cues . Neuronal responses in the first , second , third and fourth trials are plotted separately . ( B ) Neuronal population responses to cues ( measured 0 . 1–0 . 6 s after the cue onset ) over consecutive learning trials . The grey zone shows the analysis window , comprising both novelty and value responses . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 005 In contrast to the initial novelty response , a subsequent response component occurred at 0 . 2–0 . 6 s after cue onset and became differential during the learning of different reward probabilities ( Figure 2C , 26/70 neurons , one-way ANOVA on responses from sixth to last trial , p<0 . 05 ) . These responses became statistically distinct after experiencing each cue six times ( Figure 2C , p<0 . 01 , one-way ANOVA on trial-by-trial neuronal population responses ) . Thus , throughout each short learning session with a new set of fractal images , a considerable fraction of dopamine neurons learned the value of reward predicting cues from the frequency of experienced rewards . Analysis on the whole duration of neuronal response ( 0 . 1–0 . 6 s after cue onset ) showed that the compound novelty-value responses decreased over consecutive learning trials and also reflected the learned value of cues ( Figure 2—figure supplement 1B ) . Taken together , these results demonstrate how dopamine neurons gradually acquire probability-dependent value responses from the frequency of experienced rewards , and how these responses differ from their novelty responses . Examination of dopamine prediction error responses to reward delivery provided further evidence for neuronal acquisition of reward probability . Neuronal responses to reward developed gradually to reflect the values of the cues . Specifically , activating neuronal responses to large reward ( 0 . 4 ml ) were larger after cues that predicted this outcome with lower probability , compared to cues predicting the same outcome with higher probability ( Figure 3A top ) . Conversely , depressant neuronal responses to small reward ( 0 . 1 ml ) were more pronounced after cues predicting large reward with higher probability ( Figure 3A bottom ) . Thus , both the activating and depressant responses were consistent with reward prediction error coding . The neuronal responses to both large and small rewards differentiated gradually over consecutive trials , based on the predicted probability of getting each of those rewards , and reached statistical significance after 9 and 16 trials , respectively ( Figure 3B , p<0 . 02 , one-way ANOVA on neuronal population responses ) . The development of dopamine responses to rewards further suggests that early and late responses to cues convey distinct signals . If early responses to cues contained predictive values signals ( i . e . reflecting an optimistic value initialisation ) , such signals should have contributed to prediction error computations at reward time . However , the pattern of neuronal reward prediction errors ( Figure 3B ) suggests that these responses were computed in relation to late responses to cues , and reflected cue values initialised around the average value of all cues . Accordingly , neuronal responses to rewards were accounted for by the late component of neuronal responses to cues as well as the received reward size , with no significant contribution from the early component of cue responses ( p=0 . 0001 , 0 . 43 and 0 . 021 for reward size , early and late cue responses , respectively; multiple linear regression ) . Thus , the development of the prediction error responses at the time of reward reflect the acquisition of probability-dependent value responses to cues; dopamine neurons learn the value of novel cues and use these learned values to compute prediction errors at the time of the outcome . 10 . 7554/eLife . 18044 . 006Figure 3 . Responses of dopamine neurons to reward delivery develop over trials to reflect the learned value of probabilistic cues . ( A ) PSTHs of example dopamine neurons in response to delivery of large and small juice rewards ( top , bottom ) . Probabilities indicated in colour refer to the occurrence of the large reward in gambles containing one large and one small reward ( 0 . 4 ml and 0 . 1 ml , respectively ) . ( B ) Neuronal population responses to large and small juice rewards over consecutive learning trials . Responses were measured in analysis windows indicated by corresponding grey horizontal bars in A ( top: 0 . 15–0 . 5 s , bottom: 0 . 2–0 . 45 s after reward onset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 006 Dopamine responses to rewards and reward-predicting cues are described well by prediction errors derived from standard reinforcement learning ( RL ) models . These models calculate trial-by-trial prediction errors and use these values , weighted by a learning rate parameter , to update associative strengths . While it is straightforward to see that in the RL framework positive and negative reward prediction errors , encountered upon receiving large and small rewards , can lead to reward probability-dependent cue responses , it is not clear what form of RL model can best account for the development of value and novelty driven dopamine responses during learning . We therefore investigated different variants of RL models to discover which RL variant can best capture the observed development of dopamine responses . We devised models with three different types of learning rates: ( 1 ) a learning rate which was fixed over trials resembling the original Rescorla-Wagner model ( Rescorla and Wagner , 1972 ) , ( 2 ) a learning rate that decayed over trials thus representing the idea that updating should occur faster in early trials , and ( 3 ) a learning rate that was adaptively adjusted on every trial based on past prediction errors , thus capturing the idea that learning is slower when prediction errors are negligible ( Pearce and Hall , 1980; Pearce et al . , 1981; Le Pelley , 2004 ) . For each of the three learning rate-type models , we fit the data with and without the presence of a term for the novelty which decayed over trials ( Figure 4A , see Materials and methods ) ( Kakade and Dayan , 2002 ) . Thus , in total we explored six model variants , and fit the models to the dopamine responses using the rewards actually delivered during the experiments ( Materials and methods ) . 10 . 7554/eLife . 18044 . 007Figure 4 . A reinforcementlearning model with a novelty term and an adaptive learning rate account for dopamine responses during learning . ( A ) Schematic of RL models fitted on neuronal responses . In each trial , the model updates the value of stimulus based on the experienced reward prediction error . Six variants of RL models were tested ( three different learning rates , each with or without novelty term ) . In brief , we optimized the free parameters of each model so that it minimized the difference between dopamine responses to cues ( measured 0 . 1–0 . 6 s after the cue , thus including both novelty and value component ) and model’s estimates of novelty + value . We then examined the relation between value-driven neuronal responses and value estimates of the superior model and also the relation between novelty-driven neuronal responses and novelty estimates of the superior model . For details of model implementation and fitting procedure see Materials and methods . ( B ) Left: Value estimates of the superior model ( i . e . the model with a novelty term and adaptive learning rate ) overlaid on neuronal population responses measured 0 . 2–0 . 6s after the cue onset , ( from Figure 2C ) . For details of parameter estimation and model comparison see Supplementary file 1 . Right: Regression of dopamine responses to cues ( dopamine value responses , i . e . 0 . 2–0 . 6 s after the cue onset ) onto value estimates of the superior RL model . See Figure 4—figure supplement 1 for regression of dopamine novelty signals onto novelty-driven model’s estimates . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 00710 . 7554/eLife . 18044 . 008Figure 4—figure supplement 1 . A reinforcement learning model with a novelty term and an adaptive learning rate account for dopamine responses during learning . ( A ) Novelty + value estimates of the superior model ( i . e . the model with a novelty term and adaptive learning rate ) overlaid on neuronal population responses measured 0 . 1–0 . 6s after the cue onset ( from Figure 2—figure supplement 1B ) . ( B ) Regression of dopamine novelty signals ( measured 0 . 1–0 . 2 s after the cue onset ) onto novelty-driven estimates of the superior model . ( C ) Average of estimated learning rate and estimated novelty term of the superior model . ( D ) Scatter plot of session-by-session estimated learning rates of early and late components of dopamine responses to cues . For this analysis , the model with fixed learning rate that included novelty term was rearranged so that its novelty term followed an error-driven learning ( see Methods and materials ) . Each dot corresponds to a session . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 008 The model that included an adaptive learning rate and novelty term outperformed all other model variants in accounting for dopamine responses ( Figure 4B left and Figure 4—figure supplement 1A , see Supplementary file 1 for details of parameter estimation and model comparisons ) . Consistent with this , regression of second component dopamine responses to cues ( 0 . 2–0 . 6 s after the cue onset ) onto value estimates of the superior model was highly significant ( Figure 4B , right , R2 = 0 . 93 , p=0 . 00003 ) . Moreover , regression of dopamine novelty responses to model-driven novelty estimates was statistically significant ( Figure 4—figure supplement 1B , R2 = 0 . 63 , p=0 . 001 ) . In this simulation , the estimated learning rate decayed over trials , while fluctuating based on past prediction errors ( Figure 4A and Figure 4—figure supplement 1C , Supplementary file 1 ) . The model fittings suggested that the development of early and late responses to cues follows different temporal dynamics . In a model variant that is rearranged to include the novelty decay term as an error-driven learning process ( simulating optimistic value initialisation ) , the recovered learning constant for the first component of cue responses was significantly larger than the recovered learning constant for the late response component ( Figure 4—figure supplement 1D , p<0 . 0001 , Mann-Whitney U test , see Materials and methods ) . This observation suggests that early and late components of dopamine responses follow distinct temporal dynamics . Together , these results suggest that , during learning about probabilistic rewards , the trial-by-trial dopamine responses to cues adjust according to how much learning has advanced . Neuronal responses to cues rapidly develop early during learning and value updating becomes slower as learning progresses and prediction errors become smaller . Our findings so far demonstrate that dopamine neurons acquire the value of probabilistic rewards during a Pavlovian learning paradigm . We next investigated the behavioural and neuronal signatures of learning about probabilistic rewards during a decision task . As before , we used new fractal stimuli for each learning episode , which also prevented a carry-over of learned pictures from the Pavlovian to the choice task . The animals made saccade-guided binary choices between two cues ( Figure 5A ) . The animals had extensive prior experience with one of the cues ( familiar cue ) that predicted a 50% chance of receiving a large reward ( 0 . 4 ml ) and 50% chance of receiving a small reward ( 0 . 1 ml ) . In each block ( typically 50 trials ) the familiar cue was always offered as one choice alternative . The other cue was novel , and its reward probability was unknown to the animal . Similar to the Pavlovian task , the novel cues were associated with reward probabilities of 0 . 25 , 0 . 50 or 0 . 75 of receiving the large ( 0 . 4 ml ) reward and 0 . 1 ml otherwise . This situation resembled a learning set ( Harlow , 1949 ) in which the animals rapidly learned to assign one of three possible values to the novel cue . In this task monkeys had to choose the novel cue in order to learn its reward probability . At the onset of each learning block , both monkeys consistently selected the novel cue in the first few trials ( Figure 5B , p<0 . 01 in both animals , Mann-Whitney U test on choice probabilities in trials 1–4 versus later trials ) . This exploratory behaviour was accompanied by shorter saccadic response times ( measured between cue onset and saccadic acquisition of the chosen option ) , compared to the response times observed during later trials when the highest probability option was usually chosen ( Figure 5C , p<0 . 01 in both animals , Mann-Whitney U test on trials 1–4 versus later trials ) . After five trials , both animals chose the higher option 75% of the time ( p<0 . 001; one-way ANOVA on choice probabilities ) . These results suggest that the animals rapidly learned the value of novel reward predicting cues in the choice task and used these learned values to make efficient economic choices . 10 . 7554/eLife . 18044 . 009Figure 5 . Monkeys rapidly learn to make meaningful choices among probabilistic reward predicting cues . ( A ) Choice task . In each trial , after successful central fixation for 0 . 5 s , the animal was offered a choice between two cues , the familiar cue and the novel cue . The animal indicated its choice by a saccade towards one of the cues . The animal was allowed to saccade as soon as it wanted . The animal had to keep its gaze on the chosen cue for 0 . 5 s to confirm its choice . Reward was delivered 1 . 5 s after the choice confirmation . The animals had extensive prior experience with one of the cues ( familiar cue predicting 50% chance of getting 0 . 4 ml and 50% chance of receiving 0 . 1 ml ) . The alternative cue was a novel cue with the reward probability unknown to the animal . The novel cues were associated with reward probabilities of 0 . 25 , 0 . 50 or 0 . 75 of receiving the large ( 0 . 4 ml ) reward and 0 . 1 ml otherwise . After a block ( of typically 50 trials ) the novel cue was replaced with another novel cue . Trials were separated with inter-trial interval of 2–5 s . Failure to maintain the central fixation or early breaking of fixation on the chosen option resulted in 6 s time-out . ( B ) Monkeys’ choice behaviour . At the onset of each learning session , both animals chose the novel cue over the familiar cue for 4–5 trials . Afterwards , animals preferentially chose the cue that predicted reward with higher probability . ( C ) Saccadic choice response times . Both monkeys showed significantly faster reaction times ( defined as the interval between the cue onset and the time the animal’s saccade acquired the chosen option ) in the first 4–5 trials of each learning block . Error bars are s . e . m across behavioural sessions . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 009 We recorded dopamine activity during the choice task ( 57 and 42 neurons in monkey A and B ) . In order to examine neuronal signatures of probability-dependent value learning , we first focused on trials in which animal choose the novel cues . Neuronal responses immediately after the cue onset ( 0 . 1–0 . 2 s after the cue onset ) decreased over consecutive trials , reflecting the stimulus novelty ( 76/99 neurons , power function fit on trial-by-trial responses , p<0 . 05 ) , but never differentiated to reflect the learned value of cues ( Figure 6A , p>0 . 1 in both animals , one-way ANOVA on population responses ) . In contrast , the later component of the neuronal response ( 0 . 4 to 0 . 65 s after the cue onset ) developed differential responses that reflected the learned value of cues ( Figure 6B , p<0 . 01 from fifth trial onwards , one-way ANOVA on neuronal population responses ) . Neuronal activity between these two windows of analysis reflected a smooth transition from encoding stimulus novelty to encoding the learned value signals ( Figure 6—figure supplement 1 ) . These results indicate that during economic choices , dopamine responses contain two distinct components; the first component of the neuronal responses mainly reflects the stimulus novelty , whereas the second component of neuronal activity differentiates during learning to encode the learned value of cues . 10 . 7554/eLife . 18044 . 010Figure 6 . Dopamine responses to cues differentiate as monkeys learn the value of novel cues in the choice task . ( A ) Neuronal population responses to cues over consecutive trials of the choice task , measured during 0 . 1–0 . 2 s after the cue onset ( Dopamine novelty responses , see inset ) . Only trials in which animal chose the novel cue were shown in all panels of this figure . ( B ) Neuronal population responses to cues over consecutive trials of the choice task , measured during 0 . 4–0 . 65 s after the cue onset ( Dopamine value responses , see inset ) . See Figure 6—figure supplement 1 for more detailed analysis of time course of the neuronal activity . ( C ) Population dopamine responses to the large reward over trials in which the novel cue was chosen and large reward was delivered . ( D ) Population dopamine responses to the reward delivery in trials in which the novel cue was chosen . Each bar demonstrates the mean neuronal response averaged across later ( 30th to last trial ) of each session . Bars on the left represent neuronal activity in response the large reward ( 0 . 4 ml ) . Bars on the right represent neuronal activity in response to the small reward ( 0 . 1 ml ) . Inset illustrates PSTHs of an example neuron in response to small and large rewards . Horizontal bars in the inset indicate the temporal window used for computing bar plots ( large rewards: 0 . 1–0 . 55 s after the reward onset , small rewards: 0 . 2–0 . 45 s after the reward onset ) . Error bars represent s . e . m across neurons ( n = 99 , pooled from monkeys A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 01010 . 7554/eLife . 18044 . 011Figure 6—figure supplement 1 . Neuronal responses to cue in the choice task . The responses were averaged in the time window indicated in each panel . In each panel , only trials in which animal chose the novel cue were shown . Responses very early after cue onset only reflect the novelty of stimuli ( Figure 6A ) . However , later component of dopamine response reflected both novelty signals as well as learned values . Finally the very late part of neuronal responses reflected only learned values ( Figure 6B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 011 We then explored signatures of value learning in neuronal responses to rewards . We analysed the dopamine responses to rewards , focusing again on trials in which animals chose the novel cue . Following the initial trials of each learning block , neuronal responses to reward began to reflect the probability predicted by the chosen cue ( Figure 6C , p<0 . 01 after 36 choice trials; one-way ANOVA on responses to the large reward ) . In accordance with reward prediction error coding , the responses to 0 . 4 ml juice were significantly larger when the chosen cue predicted this outcome with lower compared to higher probability ( Figure 6D left , p<0 . 001 in both animals , one-way ANOVA on neuronal responses averaged from 30th to the last trial of each block ) . Similarly , the negative prediction error responses to the small ( 0 . 1 ml ) rewards were more pronounced ( i . e . stronger depression of activity ) when the chosen cue predicted this outcome with lower probability ( Figure 6D right , p=0 . 02 , one-way ANOVA on neuronal responses averaged from 30th to last trial of the block ) . Together , these results indicate that during a learning task that included economic choice , dopamine neurons learn the value of novel cues from the probabilistic outcomes associated with those cues and compute reward prediction errors by comparing these learned values with the actual trial outcome . To investigate whether dopamine responses depended on the animals’ choice , we divided the trials according to the choice that the animals made ( i . e . lower probability chosen or higher probability chosen ) , and examined the trial-by-trial neuronal responses to cue presentations . The magnitude of the neuronal response depended on the choice ( Figure 7A ) . Larger neuronal responses occurred when the animal chose the higher probability ( more valuable ) option , compared to the lower probability ( less valuable ) option ( Figure 7A , p<0 . 02 , Mann-Whitney U test on population responses during 0 . 25–0 . 65 s after cue onset in both animals and p<0 . 02 in 11 out of 99 single cells , Mann-Whitney U test ) . The early response component ( 0 . 1–0 . 2 s after cue onset ) did not reflect animals’ choice ( p>0 . 1 in both animals , Mann-Whitney U test ) . The choice-sensitivity of neuronal responses developed rapidly during learning; they reached statistical significance after five choice trials ( Figure 7B , p<0 . 01 from fifth trial onwards , one-way ANOVA ) . Within a given trial , choice predictive activity arose as early as 130 ms prior to saccade onset ( Figure 7C , analysis window starting 0 . 2 s before the choice onset , p<0 . 01 from 130 ms before saccade onset , Mann-Whitney U test ) . These results demonstrate that during learning dopamine responses rapidly develop choice sensitivity and reflect the value of the option chosen by the animal ( i . e . chosen value ) . Furthermore , these neurons began encoding this decision variable even before the overt choice ( i . e . onset of saccade ) occurred . 10 . 7554/eLife . 18044 . 012Figure 7 . During learning dopamine neurons acquire choice-sensitive responses which emerge prior to response initiation . ( A ) Population dopamine PSTHs to cues in the choice task . Grey horizontal bar indicates the temporal window used for statistical analysis . In all plots , all trials of learning blocks are included . Note that the results would be similar after excluding initial trials of each learning session . ( B ) Population dopamine responses to cues ( 0 . 4–0 . 65 s after the cue onset ) over consecutive choice trials . Trials are separated based on animal’s choice . ( C ) Population dopamine PSTHs aligned to the saccade initiation ( i . e . the time on which animal terminated the central fixation to make a saccade towards one of the cues ) . Dopamine choice-sensitive responses appeared ~130 ms prior to saccade initiation . ( D ) Averaged neuronal population responses to cues in trials in which animals chose the familiar cue . Despite the fact that animal had extensive experience with the familiar cue ( and hence accurate estimate of its value ) , neuronal responses showed dependency on the value of the unchosen cue . See Figure 7—figure supplement 1 for the time course of this effect over consecutive trials of learning . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 01210 . 7554/eLife . 18044 . 013Figure 7—figure supplement 1 . Population dopamine responses to cues over trials in which animals chose the familiar cue over the novel cues . After nine choice trials , neuronal responses showed dependency to the value of the unchosen cue . Responses to cues at first and second trials are not shown because in these trials animals almost never chose the familiar cue ( see Figure 5B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 013 We next investigated whether neuronal responses during choice could also reflect the value of unchosen option . We inspected trials in which the animals chose the familiar cue , and divided those trials according to the reward probability of the novel cue . Despite the fact that the animals had extensive experience with the familiar cues ( and hence accurate estimates of their value ) , the neuronal responses during choices of the familiar cue were significantly larger when the alternative option predicted low compared to high probability of the same large reward ( Figure 7D , p=0 . 01 , one-way ANOVA , see Figure 7—figure supplement 1 for the time course of the effect ) . Together , these results suggest that during choices among probabilistic rewards , dopamine responses are sensitive to the value of both chosen and unchosen options . We used RL models , similar to those described in the Pavlovian experiment but modified to account for choice , to explore the observed neuronal coding in a trial-by-trial fashion ( Figure 8—figure supplement 1 , see Materials and methods ) . The model with adaptive learning rate and novelty term outperformed all other models in accounting for animals’ choices ( see Supplementary file 2 for parameter estimation and model comparison ) . This model accounted well for animals' trial-by-trial choices throughout learning blocks ( Figure 8A , left ) as well as their session-by-session preferences ( Figure 8A , right , r = 0 . 86 , p<0 . 0001 ) . Similar to the Pavlovian experiments , the estimated learning rate exhibited a decay over trials while maintaining sensitivity to past prediction errors ( Figure 8—figure supplement 2A ) . In all tested models , the estimated learning of the novel cue was larger than the estimated learning rate of the familiar cue , indicating that during learning animals updated the value of novel cue more than the value of familiar cue ( Supplementary file 2 ) . Linear regression of the neuronal responses ( measured 0 . 4–0 . 65 s after the cue onset ) onto model’s value estimates revealed a positive relationship to chosen values and an negative relationship to unchosen value ( Figure 8B , chosen value: R2 = 0 . 65 , p=0 . 005 , unchosen value: R2 = 0 . 84 , p=0 . 0001 , single linear regression ) . However , a relative chosen value variable , defined as chosen value – unchosen value , fit the data far better , compared to the chosen or unchosen value variables ( Figure 8C , R2 = 0 . 91 , p=0 . 00005 , single linear regression , p<0 . 02 in 15 out of 99 single cells ) , confirming earlier results shown in Figure 7 . Similar to the Pavlovian experiment , regression of dopamine novelty responses ( 0 . 1–0 . 2 s after the cue ) onto model’s novelty estimates was significant ( R2 = 0 . 61 , p=0 . 001 , Figure 8—figure supplement 2B ) . Together , these results suggest that when the animals learn to choose among probabilistic rewards , dopamine neurons took the value of both chosen and unchosen options into account and thus reflected relative chosen value . 10 . 7554/eLife . 18044 . 014Figure 8 . Dopamine neurons encode relative chosen values . ( A ) Left: Animals choices were simulated using standard reinforcement learning ( RL ) models ( see Figure 8—figure supplements 1 and 2 and Materials and methods ) . Dotted lines show the performance of the model in predicting monkeys’ choices . Solid lines show monkeys’ choice behaviour ( identical to Figure 5B ) . The parameters of the RL model were separately optimized for each behavioural session ( Supplementary file 2 ) . Right: The RL model’s session-by-session probability of choosing the novel cue , estimated using model’s optimized parameters , versus monkeys’ session-by-session probability of choosing the novel cue . ( B ) Upper panel: Regression of neuronal population responses to cues onto trial-by-trial chosen values estimated from the RL model fitted on monkeys’ choice data . Lower panel: Regression of neuronal population responses to cues onto trial-by-trial unchosen values estimated from the RL model fitted on the choice data . ( C ) Regression of neuronal population responses to cues onto trial-by-trial relative chosen values ( i . e . chosen value – unchosen value ) estimated from the RL model fitted on the choice data . Importantly , the chosen and unchosen value variables were not , on average , strongly correlated ( r = −0 . 039 , Pearson’s correlation ) , and we excluded from this analysis sessions in which the absolute value of the correlation coefficient between the chosen and unchosen variables was larger than 0 . 25 . In B and C , the neuronal responses were measured 0 . 4–0 . 65 s after cue onset ( i . e . dopamine value signals ) and are regressed against value estimates of the superior model . In explaining the neuronal responses , relative chosen value outperformed other variables in all six models tested . See Figure 8—figure supplement 2B for regression of responses measured 0 . 1–0 . 2 s after cue onset ( i . e dopamine novelty responses ) onto model-driven novelty estimates . Regression of whole neuronal responses ( 0 . 1–0 . 65 s after the cue onset ) against value estimates of the RL model further confirmed relative chosen value as the best explanatory variables ( R2 = 0 . 57 , 0 . 61 and 0 . 83 for unchosen , chosen and relative chosen values ) . In all plots , all trials of learning blocks are included ( regression results are similar after excluding initial ( i . e . 5 ) trials of each session ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 01410 . 7554/eLife . 18044 . 015Figure 8—figure supplement 1 . Schematic of the RL model used for simulating monkeys’ choice behaviour . In each trial , the model makes a choice by comparing values associated with familiar and novel cues ( for models with novelty term: value vs novelty + value associated with familiar and novel cues are compared ) . Upon receiving the outcome , the model computes a prediction error , i . e . the difference between received outcome and prior expectation , which is used to update the value of chosen option . We fit six variations of RL models on monkeys’ choices , differing in their learning rate parameter and in having/not having a novelty term . See Materials and methods and Supplementary file 2 for details of the model implementation . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 01510 . 7554/eLife . 18044 . 016Figure 8—figure supplement 2 . Estimated learning rates of the RL model and regression of dopamine novelty responses to model-driven novelty estimates . ( A ) Average estimated learning rates of the superior model for familiar and novel cues . ( B ) Regression of neuronal population responses measured 0 . 1–0 . 2 s after the cue onset onto novelty estimates of the superior model . DOI: http://dx . doi . org/10 . 7554/eLife . 18044 . 016 Building on previous findings that the prediction error responses of dopamine neurons increase monotonically with reward probability and expected value ( Fiorillo et al . , 2003; Tobler et al . , 2005 ) , this study shows how these probability dependent value responses evolve through learning . Dopamine responses showed two distinct response components . Responses immediately after the cues decreased as learning advanced , reflecting novelty . The second response component developed during learning to encode the value of probabilistic rewards acquired from experienced reward frequencies . Correspondingly , the prediction error responses at reward time changed over the course of learning to gradually reflect the learned reward values . Results from previous studies on fully established tasks suggest that the acquired dopamine responses to probabilistic rewards do not code reward probability on its own but rather increase monotonically with the statistical expected value ( Fiorillo et al . , 2003; Tobler et al . , 2005 ) . The present learning data are fully compatible with those results . During choices , the acquired dopamine value signals coded the value of the chosen option relative to the unchosen option . These results are consistent with previous findings that showed chosen value coding in dopamine neurons ( Morris et al . , 2006 ) . However , we provide new evidence in favour of a more nuanced , relative chosen value coding scheme whereby dopamine responses also reflect the value of un-chosen option . Together , our data suggest that dopamine neurons extract predictive reward value from the experienced reward frequency and code this information as relative chosen value . Throughout learning , dopamine responses to cues developed to reflect the value of upcoming rewards , indicating that these neurons extract predictive value signals from experienced reward frequencies . In the learning experiment that involved choices , the neuronal responses rapidly differentiated to reflect animal’s choice . These differential responses , despite appearing more than 100 ms prior to overt behaviour , reflect prediction errors in relation to an already computed choice , and thus might not directly participate in current choice computation . Our modelling results provided further insights into the dynamics of neuronal learning process . First , the development of neuronal responses over trials as well as animals’ choices were best explained by models that adaptively adjusted their learning rate based on past prediction errors , resembling previous studies in human subjects ( Nassar et al . , 2010; Diederen and Schultz , 2015 ) . Second , value-dependent dopamine responses were still updated even after the dopamine novelty responses stabilized , suggesting two distinct time courses for these two components of neuronal activity . Interestingly , in both Pavlovian and choice tasks , behavioural preferences as well as neuronal responses to cues reflected reward probability earlier during learning than the neuronal reward responses . This temporal difference might suggest an origin of behavioural preferences and acquired dopamine cue responses in other brain structures , rather than relying primarily on dopamine reward prediction error signals . We observed that early during learning , dopamine novelty responses were large and they slowly decreased over consecutive trials , due to a decrease in stimulus novelty as suggested previously ( Horvitz et al . , 1997; Schultz , 1998; Kakade and Dayan , 2002; Gunaydin et al . , 2014 ) . In both tasks , the novelty signals were mainly present in initial component of neuronal responses to cues . We used RL models to investigate how these novelty signals affected the neural and behavioural computation of value . In principle , novelty can be incorporated into RL models in two ways: ( 1 ) novelty directly augments the value function , thus increasing the predicted value and distorting future value and prediction error computations , or ( 2 ) novelty promotes exploration ( in a choice setting ) but does not distort value and prediction error computation ( Kakade and Dayan , 2002 ) . If novelty increased value estimates early in the learning session ( i . e . an optimistic value initialisation ) , then positive prediction errors at the reward time should be very small in early trials and should slowly grow over trials , as optimism faded . Similarly , negative prediction errors would appear as strong suppressions which would be mitigated later . However , our results showed the opposite . We observed a clear development of reward prediction errors depending on the learned value of cues ( Figure 3 ) . On the other hand , when dopamine novelty responses were large , i . e . during early trials of choice blocks , monkeys had a strong behavioural tendency to explore the unknown option ( Figure 5 ) . Thus , it appears that novelty increased dopamine responses to cues and was correlated with high levels of exploration , consistent with previous studies ( Costa et al . , 2014 ) , but the neural responses did not reflect optimistic value initiation . Given the substantial projections of dopamine neurons to cortical and subcortical structures involved in decision making ( Lynd-Balta and Haber , 1994; Williams , 1998 ) , dopamine responses to novel situations might set downstream neuronal dynamics to an activity regime that is optimal for learning ( Puig and Miller , 2012 ) . Previous learning studies have shown that dopamine neurons are activated by unpredictable rewards , but not by completely predicted rewards ( Mirenowicz and Schultz , 1994 ) . Accordingly , dopamine neurons respond most strongly to rewards delivered near the start of learning , when rewards are most unpredictable and induce positive prediction errors ( Hollerman and Schultz , 1998 ) . Reward responses steadily decrease as the rewards become progressively more predictable ( Hollerman and Schultz , 1998 ) . However , in that study a small fraction of neurons ( 12% ) responded to fully predicted rewards . Similarly , in studies using rodent models some dopamine responses to fully predicted rewards have remained ( Pan et al . , 2005; Cohen et al . , 2012; Hamid et al . , 2016 ) . Several possible mechanisms can explain dopamine responses to ‘completely predicted’ rewards . With regard to the two cited learning studies in primates , the former task ( in which dopamine neurons did not respond to fully-predicted rewards ) was a simple instrumental task ( Mirenowicz and Schultz , 1994 ) , whereas in the latter task the monkeys had to make a choice before performing the instrumental response ( Hollerman and Schultz , 1998 ) . It is therefore possible that the more complex task context led to less subjective certainty about upcoming reward . In our study cues predicted the reward only probabilistically , not allowing us to study dopamine responses to fully predicted rewards . Nevertheless , both excitation and suppression of dopamine responses to rewards developed over trials , in a manner consistent with prediction error signalling . Dopamine neurons respond to prediction errors elicited by conditioned stimuli , which predict the future delivery of reward ( Schultz et al . , 1997 ) . The dopamine response to the simultaneous onset of choice options is a special case of this responding , because future reward delivery is contingent upon the choice as well as the values that are currently on offer . Previous studies of dopamine activity during choice have shown chosen value coding by dopamine signals ( Morris et al . , 2006; Saddoris et al . , 2015 ) , but other studies have shown coding of the best available option , irrespective of choice ( Roesch et al . , 2007 ) . Our results confirm the chosen value character of this response and indicate that choice-dependent dopamine signals arose very early with respect to both the onset of learning block as well as the onset of choice within each trial ( Figure 7 ) . However , distinct from previous reports , our results indicate that the dopamine signal takes the value of both chosen and unchosen options into account , thus reflecting relative chosen value . The relative value coding nature implies that choosing the exact same option is associated with very different responses in dopamine neurons depending on the value of the alternative option . From this standpoint , our results are fundamentally compatible with a recent report ( Kishida et al . , 2016 ) indicating that striatal dopamine concentration in human subjects reflects standard reward prediction error as well as counterfactual prediction error ( the difference between the actual outcome and outcome of the action not taken ) . Our findings provide a cellular correlate for this phenomenon and indicate that flexible encoding of both choice options already occurs at the level of dopamine action potentials . Dopamine prediction error responses are well-known teaching signals . These signals are transmitted to the striatum and cortex where they would be capable to update stimulus and action values ( Reynolds et al . , 2001; Shen et al . , 2008 ) . Dopamine signals induce value learning ( Steinberg et al . , 2013 ) and are implicated in multiple aspects of goal-directed behaviour ( Schultz , 1998; Bromberg-Martin et al . , 2010; Stauffer et al . , 2016 ) . The results demonstrated in this study advance our knowledge of dopamine function by suggesting that dopamine signals might play a critical role in computing flexible values needed for economic decision making ( Padoa-Schioppa , 2011 ) . The fast and flexible dopamine responses we observed during choice correspond well to recent findings demonstrating the encoding of economic utility by dopamine neurons ( Lak et al . , 2014; Stauffer et al . , 2014 ) and the necessity of phasic dopamine responses for consistent choices ( Zweifel et al . , 2009 ) . Taken together , these data point to a possible function for dopamine neurons in influencing decisions , in form of updating neuronal decision mechanisms in a rapid and flexible manner . Two male rhesus monkeys ( Macaca mulatta ) were used for all experiments ( 13 . 4 and 13 . 1 kg ) . All experimental protocols and procedures were approved by the Home Office of the United Kingdom . A titanium head holder ( Gray Matter Research ) and stainless steel recording chamber ( Crist Instruments and custom made ) were aseptically implanted under general anaesthesia before the experiment . The recording chamber for vertical electrode entry was centered 8 mm anterior to the interaural line . During experiments , animals sat in a primate chair ( Crist Instruments ) positioned 30 cm from a computer monitor . During behavioural training , testing and neuronal recording , eye position was monitored noninvasively using infrared eye tracking ( ETL200; ISCAN ) . Licking was monitored with an infrared optical sensor positioned in front of the juice spout ( V6AP; STM Sensors ) . Eye , lick and digital task event signals were sampled at 2 kHz . The behavioural tasks were controlled using Matlab ( Mathworks Inc . ) running on a Microsoft Windows XP computer . Custom-made , movable , glass-insulated , platinum-plated tungsten microelectrodes were positioned inside a stainless steel guide cannula and advanced by an oil-driven micromanipulator ( Narishige ) . Action potentials from single neurons were amplified , filtered ( band-pass 100 Hz to 3 kHz ) , and converted into digital pulses when passing an adjustable time–amplitude threshold ( Bak Electronics ) . We stored both analog and digitized data on a computer using Matlab ( Mathworks Inc . ) . Dopamine neurons were functionally localized with respect to ( a ) the trigeminal somatosensory thalamus explored in awake animals and under general anaesthesia ( very small perioral and intraoral receptive fields , high proportion of tonic responses , 2–3 mm dorsoventral extent ) , ( b ) tonically position coding ocular motor neurons and ( c ) phasically direction coding ocular premotor neurons in awake animals . Individual dopamine neurons were identified using established criteria of long waveform ( >2 . 5 ms ) and low baseline firing ( <8 impulses/s ) ( Schultz and Romo , 1987 ) . Following the standard sample size used in studies investigating neuronal responses in non-human primates , we recorded extracellular activity from 169 dopamine neurons in two monkeys ( Pavlovian task: 38 and 32 neurons in monkey A and B; Choice task: 57 and 42 neurons in monkey A and B , respectively ) . Most neurons that met these criteria showed the typical phasic activation after unexpected reward , which we used as a fourth criterion for inclusion in data analysis . We constructed Peri-stimulus time histograms ( PSTHs ) by aligning the neuronal impulses to task events and then averaging across multiple trials . The impulse rates were calculated in non-overlapping time bins of 10 ms . PSTHs were smoothed using a moving average of 70 ms for display purposes . The analysis of neuronal data used defined time windows that included the major positive and negative response components following cue onset and juice delivery , as detailed for each analysis and each figure caption . To quantify the development of probability-dependent dopamine responses over trials , we employed one-way ANOVA , which we serially applied to trial-by-trial population responses , i . e . , to responses of all neurons in trial 1 , trial 2 , etc . Likewise , for quantification of the time course that dopamine responses differentiate in relation to animal’s choice , we used a Mann-Whitney U test on the neuronal population responses ( 10 ms non-overlapping window of analysis starting 200 ms before the choice ) . In order to quantify the differences among responses to cues for each cell recorded in the Pavlovian task , we used a one-way ANOVA on neuronal responses from sixth to last trial of each session . In order to quantify the changes of dopamine novelty responses over trials we fitted a power function ( tn , where t represents trial number ) on normalized neuronal responses of each cell . For this fitting , responses of each neuron were normalized to its response on the first trials of the learning block . This fit results in negative or positive values of n for neurons that exhibit a decreasing or increasing cue-evoked response over trials , respectively . We used 95% confidence interval of the fit to acquire statistical significance . In order to test whether dopamine novelty responses ( in the Pavlovian task ) better reflect cue repetition or progress through the block ( i . e . trial number ) , we regressed neuronal responses on number of times the cues were seen and also on the trial number in the block ( for this analysis we only focused on first 10 trials of the block to better dissociate these two variables ) . To examine the contribution of early and late components of dopamine cue responses to prediction error computation at reward time , we employed a multiple linear regression analysis . In order to relate neuronal response in the Pavlovian task to RL model estimates , we used single linear regressions ( see Reinforcement learning models section ) . To relate neuronal response in the choice task to RL model fits on the behavioural choice data , we used single linear regression analysis both on neuronal population response as well as on responses of each dopamine neuron ( see Reinforcement learning models section ) .	Learning to choose the most fulfilling reward from a number of options requires the ability to infer the value of each option from unpredictable and changing environments . Neurons in the brain that produce a chemical called dopamine are critical for this learning process . They broadcast a 'prediction error' signal that alerts other areas of the brain to the difference between the actual reward and the previously predicted reward . Previous studies have shown that when dopamine neurons signal a prediction error , new learning about the value of an option takes place . To understand exactly what happens during this learning process , Lak et al . recorded electrical activity from dopamine neurons in the brains of two monkeys . Over a number of trials , the monkeys were shown one of three different novel images , each of which was associated with a different likelihood of receiving a large amount of a fruit juice reward . The recordings showed that the dopamine response to cues was divided into early and late components . At the start of learning , when the monkeys were unfamiliar with the likelihood that each image would yield a large juice reward , the early part of the dopamine response was large . The size of this part of the response decreased as the monkeys became more familiar with each image . The later part of the dopamine response changed to reflect the rewards the monkeys had received on previous trials . On trials where a reward was delivered , this part of the response grew larger , but diminished if a reward was not given . When the monkeys had to choose between rewards , the dopamine response was larger when the monkey chose the higher valued option over the lesser valued one , and smaller when the opposite choice was made , thus reflecting the animal’s choice . These choice-dependent responses were also sensitive to the value of unchosen option , and therefore , reflected the difference between the value of chosen and unchosen options . Future studies are now required to find out how manipulating the activity of the dopamine neurons influences the way animals learn and make decisions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Dopamine neurons learn relative chosen value from probabilistic rewards
Antifreeze proteins ( AFPs ) protect certain cold-adapted organisms from freezing to death by selectively adsorbing to internal ice crystals and inhibiting ice propagation . The molecular details of AFP adsorption-inhibition is uncertain but is proposed to involve the Gibbs–Thomson effect . Here we show by using unbiased molecular dynamics simulations a protein structure-function mechanism for the spruce budworm Choristoneura fumiferana AFP , including stereo-specific binding and consequential melting and freezing inhibition . The protein binds indirectly to the prism ice face through a linear array of ordered water molecules that are structurally distinct from the ice . Mutation of the ice binding surface disrupts water-ordering and abolishes activity . The adsorption is virtually irreversible , and we confirm the ice growth inhibition is consistent with the Gibbs–Thomson law . Many species of fish , insects , plants and micro-organisms living in cold environments produce antifreeze proteins ( AFPs ) whose main function is to target and modify the growth of regular ice ( Raymond and DeVries , 1977; Davies and Hew , 1990 ) . By binding to ice , AFPs are thought to both lower the freezing point and raise the melting point of ice through the Gibbs–Thomson effect whereby adsorbed AFPs cause increased micro-curvature of the ice front ( Yeh and Feeney , 1996 ) . The resulting difference between the melting and freezing points , known as the thermal hysteresis gap ( TH ) , can range from a few tenths to over 6° , depending on the specific AFP ( Lin et al . , 2011 ) . Regular hexagonal ice ( Ih ) has three equivalent a-axes ( a1–a3 ) which are parallel to three respective prism planes and perpendicular to the c-axis . The ice basal plane lies perpendicular to the c-axis . Remarkably AFPs have evolved independently numerous times to bind stereo-selectively to ice in various orientations , resulting in a structurally diverse protein family ( Hew et al . , 1985; Ko et al . , 2003; Pentelute et al . , 2008; Hakim et al . , 2013; Middleton et al . , 2013 ) . The potency of different AFPs reflects their adsorption characteristics and environmental adaptation . Polar fish , for example , do not normally experience temperatures much lower than the freezing point of sea water ( −1 . 9°C ) , and have AFPs that provide protection to just a few tenths of a degree below this . Fish AFPs often bind oblique to ice prism planes , typically forming hexagonal bipyramid crystals at temperatures within the TH gap which rapidly grow as elongated spicules along the c-axis below the TH gap ( Davies and Hew , 1990 ) . In contrast insects which are exposed to much lower terrestrial temperatures , ( often below −30°C ) , require AFPs with higher activity , possibly by selecting more effective prism and basal plane binding ( Lin et al . , 2011 ) . Despite a wide selection of available structural and experimental data , a complete molecular description of how an AFP binds to its respective ice planes and inhibits crystal growth remains elusive . The difficulty in characterizing the molecular mechanisms of AFPs arises largely due to their unusual relationship with water , which in this case serves as both the protein's solvent and , as ice , its target ( Jia and Davies , 2002 ) . Slowly freezing solutions of dilute AFP will incorporate AFPs into the growing ice phase at roughly their original concentration , while non-AFPs and solutes are largely excluded ( Marshall et al . , 2004 ) . Many structural studies have shown a correlation between the periodicity of the ice-binding surface elements of AFPs and their respective ice-binding planes , though it has not been entirely clear whether AFPs bind directly to ice or through some ordered water intermediate ( Sharp , 2011 ) . Observation of the molecular arrangements of AFPs and surrounding water as they bind and modify the surface curvature of ice is currently technically unattainable . Given these experimental limitations , computer simulations have played a crucial role in expanding our understanding of the AFP phenomena . Early computational simulations of winter-flounder AFP ( wfAFP ) by Wen and Laursen ( 1992 ) ; Jorgensen et al . ( 1993 ) ; Madura et al . ( 1994 ) firmly established the correlation of the distance between regularly spaced polar threonine residues and that of water molecules of the ice lattice of the {2021} pyramidal ice plane in the [112] direction by searching for energetically stable conformations . This was in agreement with previous experimental work of Knight et al which showed the preferential binding orientation of wfAFP on single crystal ice hemisphere ( Knight et al . , 1991 ) and also the later work of Sicheri and Yang who determined the wfAFP crystal structure ( Sicheri and Yang , 1995 ) . Cheng and Merz further expanded this work using molecular dynamics simulations of solvated wfAFP in the bound conformation to the {2021} pyramidal plane to propose key hydrogen bond interactions between polar residues and the ice surface , suggesting that hydrogen bonds were the primary driving force of ice adsorption ( Cheng and Merz , 1997 ) . However , this model was to be short lived as mutational work by the groups of Harding and Laursen ( Zhang and Laursen , 1998; Haymet et al . , 1999 ) raised questions about the nature of the wfAFP-ice interaction; conservative substitution of threonine residues for serine abolished activity , while threonine to valine substitution , which removed the proposed threonine hydroxyl–ice interactions surprisingly retained AFP activity . Spurred by the paradoxical mutant wfAFP results both Dalal and Sönnichsen ( 2000 ) and Jorov et al . ( 2004 ) employed Monte Carlo simulations of the wfAFP on the pyramidal plane , both groups finding that hydrophobic interactions contributed significantly to the adsorption to ice . Further related work by Wierzbicki et al . ( 2007 ) proposed that wfAFP doesn't bind directly to ice but rather accumulates at the ice-water interface with preference for the hydrophobic over the hydrophilic face facing the ice . Additional computational studies of different AFPs have followed , including type II sea-raven AFP ( Wierzbicki et al . , 1997 ) , type III eel pout AFP ( Madura et al . , 1996 ) and an insect AFPs from Tenebrio molitor ( Liu et al . , 2005 ) . Nada and Furukawa ( Nada and Furukawa , 2011 ) were amongst the first to provide a model of the spruce budworm AFP ( sbwAFP ) at the prism ice-water interface . Using rigid models of sbwAFP they showed that two pre-determined binding conformations were able to affect local ice growth kinetics , inferring that reduced ice growth and induced curvature was indicative freezing point depression consistent of the Gibbs–Thomson effect . As highlighted by an excellent review by Nada and Furukawa . ( Nada and Furukawa , 2012 ) , in real world systems , AFPs bind to an ice-water interface rather than ice crystal planes alone , thus earlier AFP simulations are unable to provide complete mechanistic details of the AFP-ice interaction . Even though a number of AFP studies have simulated ice-water interactions ( McDonald et al . , 1995; Dalal et al . , 2001; Wierzbicki et al . , 2007; Nada and Furukawa , 2008 ) , Nada and Furukawa also point out the need for simulations to target a growing ice-water interface . Ideally , simulations would allow unconstrained AFPs to migrate from the water phase to interact freely the growing ice-water interface without introducing prior binding orientation bias . The computational requirements to do so are significant , and only recently became relatively accessible for researchers . The highly active AFP from the spruce budworm , Choristoneura fumiferana , ( sbwAFP ) provides an excellent model for antifreeze simulations and has been used previously in computational studies ( Nutt and Smith , 2008; Nada and Furukawa , 2011 ) . SbwAFP has a β-helix structure with a triangular cross-section forming three parallel β-sheet faces and a hydrophobic core as shown in Figure 1 . The ice binding face has a regular array of repeating threonine-x-threonine ( T-X-T ) motifs whereby the spacing between threonine residues match the repeat spacing parallel to the c-axis of ice of approximately 7 . 4 Å and motifs on adjacent β-strands closely match the a-axis repeating dimensions of the prism face of ice of 4 . 5 Å . The second row of threonine residues of the ice-binding surface is particularly sensitive to mutation , with replacement of the threonines drastically reducing sbwAFP's potency as an AFP ( Graether et al . , 2000 ) . In addition to inducing a non-colligative freezing point depression of ice by up to 6°C , sbwAFP also demonstrates a significant melting inhibition ( super-heating ) of ice of up to 0 . 44°C ( Celik et al . , 2010 ) . Here we employ multiple and extended molecular dynamics simulations of a potent sbwAFP isoform 501 ( Leinala et al . , 2002 ) with a novel ice/water interface model designed to create continual ice growth conditions , providing new insight into molecular specificity and the resultant Gibbs–Thomson induced thermal hysteresis . 10 . 7554/eLife . 05142 . 003Figure 1 . Cartoon representations of the spruce budworm AFP ( sbwAFP ) 501 isoform . This shows the triangular cross section of parallel β-helix structure . The ice binding surface is drawn as sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 003 The TIP4P water model ( Jorgensen et al . , 1983 ) was used for simulations due to its ability to give realistic phase transformation behavior and superior freezing characteristics compared with the default TIP3P model , its support by the NAMD software ( Phillips et al . , 2005 ) , and its compatibility with biological CHARMM parameters ( Glass et al . , 2010 ) . Although the melting point ( Tm ) of the TIP4P water model is known to be 230 . 5 K , ( García Fernández et al . , 2006 ) it is highly unlikely such a water model will spontaneously freeze or remain frozen in simulation , particularly in the few degrees temperature range below Tm where AFPs exhibit their biological activity , due to the equilibrium radius R of the critical ice nucleus governed by the Gibbs–Thomson effect . This effect is a well known phenomenon whereby the chemical potential across an interface varies with curvature . For positive interfacial energies smaller crystals , with their higher curvature , are only able to be in equilibrium with their melt at lower temperatures than larger crystals . AFPs are thought to exploit this property by effectively subdividing a large ice surface of relatively low curvature into smaller surfaces with greater curvature thus giving rise to lower freezing temperatures , as shown in Figure 2 . The critical radius at a given temperature Tr can be approximated with the following expression adapted from Pereyra et al . ( 2011 ) : ( 1 ) R≃MwσLρiAgT0 ( T0−Tr ) , where Mw is the molecular weight of water , σ is the surface tension of the ice-water interface , L is the latent heat of fusion , ρi is the density of ice , T0 is the melting temperature at atmospheric pressure , Ag is a constant of 2 or 1 for spherical or cylindrical symmetry conditions respectively . This relationship implies large ice crystals are required to maintain growth . For example , assuming spherical geometry , for a single degree of supercooling the critical radius of an ice embryo is about 518 Å containing approximately 18 million water molecules , which is currently computationally prohibitive to model ( see ‘Materials and methods’ for calculation ) . To simulate AFP-ice interactions more efficiently a specially arranged model of an ice-water interface was developed to provide a continual ice growth mechanism while mimicking an infinite ice plane by taking advantage of periodic boundary conditions ( PBCs ) . The ice/water system employed here consisted of 17 , 434 water molecules in a 51 . 5 × 126 . 6 × 83 . 9 Å box with PBCs , of which 1692 water molecules were harmonically constrained to an ideal prism ice lattice . This served as the seed crystal while an additional 1049 water molecules below the seed were constrained in a 5 Å thick disordered layer to act as an ice barrier preventing downward ice growth , leaving a small wedge of mobile water between constrained layers as shown in Figure 3 . A perspective view of the model is shown in Figure 3—figure supplement 1 . The seed ice crystal included an additional small prism step to initiate step layer growth and the entire seed was angled such that the primary prism face was approximately 4 . 3° with respect to the long axis . The angle of the seed is such that step layer growth across the prism face reappears , due to PBCs , one layer higher to where it started , thus providing a continual ice growth mechanism . Unimpeded , this mechanism will freeze the area above the seed crystal at any temperature below Tm . This arrangement mimics step growth as might be produced by screw dislocations , as originally proposed by Frank ( 1949 ) and later observed on ice surfaces by Ketcham and Hobbs ( 1968 ) . 10 . 7554/eLife . 05142 . 004Figure 2 . Schematic figure of antifreeze protein ( AFP ) ice inhibition via the Gibbs–Thomson effect . ( A ) A mixture of AFP and non-AFP proteins ahead of a growing ice front . ( B ) AFPs selectively bind to the ice , while non-AFPs are excluded . AFP adsorption to the ice surface subdivides into smaller growth fronts which increases local curvature . If the induced curvature of the new ice fronts is less than the critical radius ( R ) of an ice embryo at a given temperature then the ice remains in equilibrium with the melt . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 00410 . 7554/eLife . 05142 . 005Figure 3 . Section view of the sbwAFP simulation setup . The seed ice crystal is shown in dark blue , constrained disordered water coloured cyan and unrestrained water is coloured red . The sbwAFP is centrally positioned 20 Å above the seed ice . The prism face of the seed ice crystals is angled approximately 4 . 3° with respect to the long axis of the simulation box , such that the periodic image of the seed ice is consistent with itself at the edges . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 00510 . 7554/eLife . 05142 . 006Figure 3—figure supplement 1 . Perspective view of the model showing the seed ice as dark blue spheres and free waters as transparent surface . The blue line is parallel to the a-axis of ice ( prism face ) while the red line is parallel to the c-axis of ice . The green lines represent the relative alignment of the sbwAFP with the ice . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 006 As the simulation proceeds , water molecules join the ice phase through hydrogen bonding and can be easily identified by their tetrahedral arrangement , which appears hexagonal when viewed directly down the c-axis of ice . Within this ice/water box the unconstrained sbwAFP model was initially placed 20 Å above the seed , with its ice binding face directed towards the seed approximately aligned in the expected binding orientation . In our preliminary work the seed ice crystal was arranged with the prism face parallel with the axis of the simulation boundaries . Crystals from these arrangements did not grow well , extending the ice front very slowly if at all . We realized that this was because one had to effectively wait for a two-dimensional nucleation event to occur on the prism plane and reach a critical size before the next layer of ice would grow . To characterize the binding mode of sbwAFP , 32 independent , unconstrained ( apart from the waters constrained in the ice seed and ice-barrier ) simulations of at least 150 nanoseconds ( ns ) to as long as 250 ns duration were carried out at 228 K . Definitive ice binding was observed 26 times with each binding event occurring with equivalent orientation with respect to the ice lattice , though adsorbing at a number of different step heights . The remaining 6 simulations either diffused away from the growing ice front ( 3 times ) , or had partial engagement at alternate angles ( 3 times ) . Of the partial engagements , the AFP still had significant rotational and translational movement with respect to the ice lattice so was not considered truly bound . Each of the definitive ice binding events incorporated a linear array of 6 water molecules bridging between the prism face and a row of conserved threonine residues ( Thr 7 , 23 , 39 , 54 , 69 , 84 and 101 ) as shown in Figure 4 . These water molecules had the same periodicity of the prism face , spaced approximately 4 . 5 Å apart . Interestingly they were distinct from the bulk ice crystal , bridging between two prism face water molecules parallel along the c-axis which would normally be bridged by two water molecules in regular ice as shown in Figure 5 . The sbwAFP-bound water molecules had fully satisfied tetrahedral hydrogen bonding arrangements; two hydrogen bonds to the prism face ice water molecules and two hydrogen bonds to the hydroxyl groups of adjacent threonine residues of the sbwAFP ice binding face . 10 . 7554/eLife . 05142 . 007Figure 4 . The ice-binding surface of sbwAFP . Ordered water molecules are shown in yellow hydrogen bonded between adjacent threonine residues of the top row . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 00710 . 7554/eLife . 05142 . 008Figure 5 . The ice binding arrangement of spruce budworm AFP in relation to ice . ( A ) c-axis view of ice ( cyan spheres ) and the arrangement of the sbwAFP ordered water ( yellow spheres ) . ( B ) a-axis view of the sbwAFP binding orientation showing the relative arrangement of the ordered water to ice . The ice lattice repeats approximately 4 . 5 Å and 7 . 4 Å along the a axis and c axis respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 008 To further investigate the function of the ordered waters of the ice binding surface 16 independent simulations of a sbwAFP mutant ( with 4 Thr to Leu substitutions at positions 7 , 21 , 39 , 69 ) known to disrupt activity ( Graether et al . , 2000 ) , were performed , each of 250 ns length . No ice binding events were observed with all mutant AFPs diffusing ahead of the growing ice surface in distinct contrast to the wild type protein . Figure 6 shows the relative height coordinates of both wild type and mutant sbwAFP throughout the docking simulations clearly demonstrating the former AFP adsorbing and fixing to the ice surface while the latter , unable to bind , diffuses ahead of the growing ice front . The Thr to Leu substitutions disrupt the binding of the ordered water to sbwAFP by removing the stabilizing hydroxyl side chain hydrogen bond interactions of the threonine residues . These results indicate ordered waters are crucial in initiating sbwAFP recognition and binding to ice and are suggestive of longer range protein-water interactions consistent with recent experimental findings by Meinster et al , of a similar beetle AFP from Dendroides canadensis , measured with terahertz spectroscopy ( Meister et al . , 2013 ) . Representative simulations of the binding and non-binding events of the respective wild type and mutant sbwAFP are shown in supplementary Videos 1 , 2 . 10 . 7554/eLife . 05142 . 009Figure 6 . Simulations of ice docking of wildtype and inactive mutant sbwAFP against a growing ice surface . ( A ) Representative side view of final docking orientation of wild type sbwAFP . ( B ) Overlay of 16 independent wild type sbwAFP docking simulations mapping the centre of mass z-axis coordinate of the AFP over time , an unchanging z-coordinate shows ice binding . ( C ) Representative side view of mutant sbwAFP in simulation . ( D ) Overlay of 16 independent mutant sbwAFP docking simulations showing all AFPs diffusing ahead of the ice front and no ice binding events . The mutant form of the sbwAFP consisted of 4 Thr to Leu mutations at positions 7 , 21 , 39 and 69 which disrupts the ice binding surface . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 00910 . 7554/eLife . 05142 . 010Video 1 . Representative simulation of spruce budworm AFP ( sbwAFP ) docking to ice . Side view of unconstrained sbwAFP docking to the prism face of a growing ice surface at 228 K . ( 250 nanosecond total simulation . ) The seed ice crystal is in dark blue . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 01010 . 7554/eLife . 05142 . 011Video 2 . Representative simulation of mutant sbwAFP . Unconstrained sbwAFP mutant simulated at 228 K for 250 nanoseconds . Ice binding ability has been lost by mutation of 4 threonine residues ( 7 , 21 , 39 and 69 ) of the ice binding surface to leucine . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 011 Three contiguous ∼1 μs MD simulations of sbwAFP adsorbed to an active ice front at temperatures of 225 K , 230 K and 232 K were run to test and simulate the adsorption-inhibition mechanism by observing subsequent ice growth as shown in Figure 7 . Representative cross-section snapshots of the simulation at the three temperatures is shown in Figure 7—figure supplement 1 . Though the Gibbs–Thomson phenomena is widely accepted as the basis of AFP activity , to the best of our knowledge it has not been observed directly at the ice-water interface . In the first simulation the sbwAFP adsorbed from the solvent phase to the advancing ice front at 225 K ( 5 K below the model system Tm ) . The sbwAFP adsorbed to the ice front within 40 ns and within 150 ns the ice front displayed marked convex curvature creating cylindrical ice fronts as shown in Figure 8 . By 250 ns the ice front had reached equilibrium with an averaged cylindrical radius of approximately 50 Å . The extent of the inhibited ice front fluctuated , advancing and retreating in the range of 5–8 Å showing the reversible arrangement of water molecules to the crystal ice lattice , ( this is an important indication of reaching equilibration and adequate sampling given the reduced water diffusion at these low temperatures ) . The ice fraction ( defined as a proportion of total ice-like water molecules above the ice seed crystal ) , was measured over the 280 to 1000 ns time frame and found to equilibrate at 0 . 46 ± 0 . 01 ( SD ) . In contrast AFP-free simulations froze almost completely within 150 ns . The vacillating nature of the ice front in simulation makes quantitative measurement of the curvature difficult , however we find good qualitative agreement between theoretical Gibbs–Thomson curvature and the averaged ice front as shown in Figure 8 where the theoretical radius is overlaid on to the averaged simulation ice cross-section . This is good indication that the simulation ice curvature is due to the expected Gibbs–Thomson phenomena . 10 . 7554/eLife . 05142 . 012Figure 7 . Freezing and melting inhibition of ice by adsorbed sbwAFP . Ice fraction of 3 contiguous ∼1 microsecond simulation temperatures 225 K ( approx . 5 K below melting point Tm ) , 230 K ( approx . at Tm ) and 232 K ( approx . 2 K above Tm ) . Linear regression of equilibrated states shows no appreciable growth or melting ( red lines ) . A simulation containing no AFP ( blue line on left ) shows rapid freezing of the entire model . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 01210 . 7554/eLife . 05142 . 013Figure 7—figure supplement 1 . SbwAFP simulation side-profile snapshots equilibrated at different temperatures . ( A ) Representative snapshot of cross-section view of equilibrated simulation ( with adjacent periodic boundary images ) at 225 K showing curvature of ice between adsorbed sbwAFP ( seed ice is blue ) . ( B ) Side view of equilibrated simulation at 230 K showing a relatively flat ice front . ( C ) The simulation equilibrated at 232 K showing still adsorbed sbwAFP stabilizing ice protrusions on the superheated ice . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 01310 . 7554/eLife . 05142 . 014Figure 8 . Side and end views of steady-state equilibrated ice fronts inhibited by adsorbed sbwAFP . Cyan spheres represent the averaged ice front ( defined as water occupation of greater than 85% averaged over approximately 1 microsecond simulation for each temperature ) . Dark blue spheres represent the constrained ice seed . ( A ) At 225 K the ice front shows curvature around the adsorbed sbwAFP . The middle shaded circle segment has a radius of 46 Å equal to the theoretical Gibbs–Thomson cylindrical curvature of TIP4P water at 5 K supercooling . ( B ) At 230 K the ice retreats , but is stabilized directly beneath the bound sbwAFP . ( C ) At 232 K , ( approximately 2°C above Tm ) , the bound sbwAFP remains in place stabilizing an ice protrusion , pinning back further ice retreat . The side profile views are made from two adjacent simulation periodic images . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 014 When the temperature was raised to 230 K , ( the approximate system melting point Tm ) the simulation showed a significant reduction of the ice fraction to an equilibrated value of 0 . 21 ± 0 . 02 , ( calculated over the 1100 to 2130 ns time frame ) shown in Figure 8B . The sbwAFP remained bound to ice , stabilizing an ice protrusion directly beneath the AFP approximately 7 Å high relative to the main ice front . In the final contiguous simulation the temperature was raised to 232 K , ( approximately 2 K above the simulation melting point ) and showed further reduction in the ice fraction which equilibrated to an average of 0 . 15 ± 0 . 01 ( calculated over 2190 to 3160 ns time frame ) . The sbwAFP remained attached to the ice protrusion approximately 8 . 5 Å above the main ice plane which developed slight concave curvature with respect to the adsorbed protein ( Figure 8C ) . A continuation of the simulation at 235 K resulted in the detachment of the sbwAFP from the ice surface within 10 ns allowing further melting of the ice to a residual fraction of 0 . 08 ± 0 . 01 . Steady-state representations of the equilibrated ice fronts for each temperature were generated averaging water density over each simulation segment , by mapping water occupancy of greater than 85% using the ‘volmap’ module of the visualization software VMD ( Humphrey et al . , 1996 ) . Extended simulations showing the induced curvature of the ice surface are shown in Videos 3 , 4 , 5 . These simulations clearly demonstrate the Gibbs–Thomson effect showing how adsorbed protein induces ice-surface curvature and subsequent melting and freezing inhibition of ice . An important aspect of AFP activity that remains to be resolved involves the question of the strength of binding of AFP to ice . Previous estimations of the binding energy of AFPs to ice have varied greatly in the literature . Initial calculations from Cheng and Merz ( 1997 ) put the wfAFP binding energy of −157 kcal/mol while Dalal and Sönnichsen ( 2000 ) put the figure at −67 kcal/mol . Later estimates by Jorov et al . ( 2004 ) and Wierzbicki et al . ( 2007 ) both put the figure for wfAFP much lower at below −5 kcal/mol . Such low values of binding , however , imply an equilibrium with appreciable amounts of exchange of the AFP from the ice surface and its surroundings . Contrary to this , experimental evidence suggests that the binding energy must be relatively high and virtually irreversible . Microfluidic experiments by Celik et al . ( 2013 ) show that ice growth remains inhibited despite the exchange of the solution surrounding AFP-inhibited ice crystals with AFP-free buffer , suggesting there is little or no exchange of AFPs from the ice surface once bound . This finding is further strengthened by recent field observations of Cziko et al . where it was found Antarctic notethenoid fishes internally accumulated ice does not melt as expected over summer warming periods ( Cziko et al . , 2014 ) . Our own observations in this study suggest relatively strong binding of a magnitude similar to that proposed by Dalal and Sönnichsen ( 2000 ) , with the sbwAFP never detaching once bound to ice , ( apart from deliberate melting events ) , even over the course of the extended millisecond range simulations at above melting temperatures . We intend to address the magnitude of the binding energy in our future work . 10 . 7554/eLife . 05142 . 015Video 3 . Ice-bound sbwAFP simulated for 1000 nanoseconds at 232 K , showing stabilization of the ice front by the presence of the bound sbwAFP . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 01510 . 7554/eLife . 05142 . 016Video 4 . Simulation of sbwAFP binding to the prism ice front at 225 K . The adsorption of the sbwAFP to the prism face causes curvature to the ice consistent with the Gibbs–Thomson effect . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 01610 . 7554/eLife . 05142 . 017Video 5 . Continuation of Video 4 showing the ice curvature with periodic boundary conditions applied . The second half of the video highlights the positions of the bound water molecules and shows how they migrate to the ice binding surface from the beginning of the simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 05142 . 017 Our simulations have shown that sbwAFP activity depends on binding and arranging water in a similar periodicity to ice which in turn bind to the prism face of ice , and that the ordered water plays a crucial role in determining ice plane specificity and integrating into the ice lattice . This conclusion is supported by previous work by Nutt and Smith ( 2008 ) whose molecular dynamics simulations of sbwAFP in solution indicated water was more ice-like on its binding face at low temperatures and speculated that this preconfigured water was important for initial recognition and binding events . Recent work by Midya and Bandyopadhyay ( 2014 ) simulating insect AFP from Tenebrio molitor with a similar ice binding surface to sbwAFP in water have also found equivalent arrangements of bound water molecules emphasizing the importance of the hydroxyl groups of the ice-binding threonine residues . Contrary to this , as mentioned previously , it is noteworthy that the putative ice-binding threonine residues of the alpha helical wfAFP can be substituted to structurally similar but hydrophobic valine residue with little effect on activity ( Haymet et al . , 1999 ) . This shows wfAFP ice binding does not depend on the threonine hydroxyl group , even suggesting a hydrophobic mode of interaction . Though tempting to make comparisons with sbwAFP threonine residues , we should be reminded that these proteins are completely different structures with significantly different ice-binding properties . It will be important to characterize both the wfAFP/ice interaction as well as analogous sbwAFP threonine to valine mutations before valid interpretations can be made . Many of the solved AFP crystal structures have been found to contain ordered waters which have been postulated to integrate directly into the ice lattice ( Liou et al . , 2000 ) . In a recently determined structure for a highly active bacterial AFP from Marinomonas primiryensis ( mpAFP ) , bound waters exhibited an extensive ‘clathrate-like’ network that matched ice lattice dimensions ( Garnham et al . , 2011 ) . Interestingly , unlike most AFPs , mpAFP appears to bind to all ice plane orientations , rather than specific ones . This may result from extended ice-like ordering of water from the ice binding surfaces interacting with ice in non-specific orientations , perhaps in a process similar to ice sintering ( Kuroiwa , 1961 ) . As noted in a recent review by Nada and Furukawa ( 2012 ) , previous AFP simulations have focused on modeling prearranged AFPs at the ice-water interface at equilibrium and have not yet tackled free AFP interacting with an actively growing ice interface which would reveal more mechanisms of the molecular process and be less biased to initial positioning . Our freely diffusing MD simulations of AFP spontaneously adsorbing and inhibiting an active ice surface model have demonstrated that simulations can be used to provide detailed insight into molecular processes like crystal inhibition and modification . We have presented simulations of the sbwAFP spontaneously interacting with a growing ice surface and demonstrated key aspects of the AFP mechanisms; namely the process of molecular surface recognition and irreversible adsorption inhibition of the ice surface . We have shown that the Gibbs–Thomson formula agrees closely with the simulated inhibition of the growing ice front , and demonstrated irreversibility of binding by free energy analysis . This is one of the first computer simulations to provide a complete in silico demonstration of a protein's biological function . Molecular simulations were performed using NAMD2 . 9 ( Wierzbicki et al . , 1997 ) with CHARM22 ( MacKerell et al . , 1998 ) forcefield employing a TIP4P water model supported on both IBM BlueGene/P and BlueGene/Q architecture . The sbwAFP model was based on the pdb structure 1M8N ( Leinala et al . , 2002 ) . Simulations were run with PBCs using the NPT ensemble at various temperatures ( 225 , 228 , 230 , 232 and 235 K ) and 1 bar pressure employing Langevin dynamics . The PBCs were constant in the XY dimensions . Long-range Coulomb forces were computed with the Particle Mesh Ewald method with a grid spacing of 1 Å . 2 fs timesteps were used with non-bonded interactions calculated every 2 fs and full electrostatics every 4 fs while hydrogens were constrained with the SHAKE algorithm . The cut-off distance was 12 Å with a switching distance of 10 Å and a pair-list distance of 14 Å . Pressure was controlled to 1 atmosphere using the Nosé-Hoover Langevin piston method employing a piston period of 100 fs and a piston decay of 50 fs . Trajectory frames were captured every 100 picoseconds . Ice fractions were determined by measuring the ratio of mobile to immobile water molecules . Water molecules were determined to be immobile , ( i . e . , ice ) , if the average oxygen position of water had moved less than 0 . 8 Å over three successive trajectory frames ( i . e . , 200 ps ) . The initial pdb model of the sbwAFP at the ice/water interface is included in the supplementary data ( Supplementary file 1 ) . Using Equation 1 for the case of spherical geometry ( Ag = 2 ) , R≃MwσLρi2T0 ( T0−Tr ) , where R is the radius , Mw the molecular weight of water ( 0 . 018 kg/mol ) , L the latent heat of melting of water ice ( 6 . 02 × 103 J/mol ) ( Vega et al . , 2005 ) , σ is the ice-water surface energy 29 . 1 × 10−3 J/m2 ( Handel et al . , 2008 ) , ρi is the density of ice ( 917 kg/m3 ) , T0 the normal melting temperature ( 273 . 15 K ) , and Tr is the melting temperature depressed by curvature ( 272 . 15 K ) . These values give R = 518 Å . A sphere of this radius has a volume of 5 . 82 × 108 Å3 and would contain 1 . 8 × 107 water molecules . Using parameters derived from the TIP4P water model ( Vega et al . , 2005; Handel et al . , 2008 ) , the corresponding values are T0 of 230 . 5 K , σ of 23 × 10−3 J/m2 , ρi of 944 kg/m3 and L of 4 . 4 × 103 J/mol . Using the cylindrical version of this Gibbs–Thomson equation ( Ag = 1 ) and a freezing point depression of 5 K , we find the critical cylindrical radius to be 46 Å .	Water expands as it freezes . If this happens to the water inside plants and animals , the resulting ice crystals can rupture cells . To prevent this , many plants and animals that live in cold climates have evolved ‘antifreeze proteins’ . When a small particle of ice first starts to form , the antifreeze proteins bind to it and prevent the water around it freezing , hence preventing the growth of an ice crystal . There are many different types of antifreeze protein , and some are more active than others . For example , some insects including the spruce budworm are exposed to extremely cold temperatures—sometimes below −30°C—and these insects have antifreeze proteins that are highly active . It is not fully understood how different antifreeze proteins interact with ice and prevent the growth of ice crystals . This is largely because , as yet , there are no experimental techniques that make it possible to see how antifreeze proteins and water molecules arrange themselves at the surface of a growing particle of ice . Instead , scientists have developed computer simulations to investigate this process . While many of these studies have provided valuable information , the computational methods used have only recently become powerful enough to analyze how the antifreeze proteins approach the surface of the ice particle . Kuiper et al . carried out simulations involving a highly active antifreeze protein from the spruce budworm . The results of these simulations revealed that this antifreeze protein does not bind directly to ice; instead , water molecules at the surface of the protein act as a bridge between the protein and the ice . These water molecules are highly ordered and though they have similarities with how water is structured in the ice , they are distinct from the ice lattice itself . Furthermore , this arrangement appears to be important for allowing the spruce budworm antifreeze protein to interact with the ice . This study provides detailed insights as to how a highly active antifreeze protein helps to prevent ice crystals forming . In the future , the computational simulations used here may be extended to study the dynamics of other antifreeze proteins , and also how crystals of other materials form .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2015	The biological function of an insect antifreeze protein simulated by molecular dynamics
Perforin-2 ( MPEG1 ) is a pore-forming , antibacterial protein with broad-spectrum activity . Perforin-2 is expressed constitutively in phagocytes and inducibly in parenchymal , tissue-forming cells . In vitro , Perforin-2 prevents the intracellular replication and proliferation of bacterial pathogens in these cells . Perforin-2 knockout mice are unable to control the systemic dissemination of methicillin-resistant Staphylococcus aureus ( MRSA ) or Salmonella typhimurium and perish shortly after epicutaneous or orogastric infection respectively . In contrast , Perforin-2-sufficient littermates clear the infection . Perforin-2 is a transmembrane protein of cytosolic vesicles -derived from multiple organelles- that translocate to and fuse with bacterium containing vesicles . Subsequently , Perforin-2 polymerizes and forms large clusters of 100 Å pores in the bacterial surface with Perforin-2 cleavage products present in bacteria . Perforin-2 is also required for the bactericidal activity of reactive oxygen and nitrogen species and hydrolytic enzymes . Perforin-2 constitutes a novel and apparently essential bactericidal effector molecule of the innate immune system . Multicellular eukaryotes deploy pore-forming proteins to disrupt the cellular integrity of bacterial pathogens and virally infected cells . The first immunologically relevant discovery of a pore-former was the spontaneous polymerization and refolding of the hydrophilic complement component C9 into a membrane-associated cylindrical complex ( Podack and Tschopp , 1982; Tschopp et al . , 1982 ) . This finding resolved the question of the molecular nature of the membrane attack complex of complement ( MAC ) ( Humphrey and Dourmashkin , 1969; Mayer , 1972; Muller-Eberhard , 1975; Bhakdi and Tranum-Jensen , 1978 ) where C5b-8 complexes , first assembled around membrane-bound C3b , trigger C9 to polymerize and form 100 Å pores in bacterial surfaces ( Schreiber et al . , 1979; Podack and Tschopp , 1982; Tschopp et al . , 1982 ) . The recognition that a single protein species , C9 , was able to form pores by polymerization suggested the possibility that cytotoxic lymphocytes may be equipped with a similar pore-forming protein . Analysis of natural killer ( NK ) cells and cytotoxic T lymphocytes ( CTL ) identified Perforin-1 as the pore-forming killer protein for virus-infected cells and tumor cells ( Dennert and Podack , 1983; Podack and Dennert , 1983; Blumenthal et al . , 1984 ) . Sequence alignment of Perforin-1 and C9 identified a conserved domain , named the Membrane Attack Complex/Perforin ( MACPF ) domain in reference to its founding members ( Lichtenheld et al . , 1988 ) . During polymerization , the MACPF-domains of individual protomers refold and expose an amphipathic helix that inserts into the targeted membranes ( Rosado et al . , 2007; Baran et al . , 2009; Kondos et al . , 2010; Law et al . , 2010 ) . The hydrophilic surface of the membrane-inserted portion of polymerizing MACPF forms the inner , hydrophilic lining of the nascent pore driving the displacement of hydrophobic membrane components . MACPF generated pores disrupt the innate barrier function of membranes and provide access for chemical or enzymatic effectors that finalize destruction of the target ( Schreiber et al . , 1979; Masson and Tschopp , 1987; Trapani et al . , 1988; Shiver et al . , 1992; Smyth et al . , 1994 ) . Macrophage Expressed Gene 1 ( MPEG1 ) is the most recently identified protein with a MACPF-domain ( Spilsbury et al . , 1995 ) . We renamed the new MACPF-containing protein Perforin-2 when we confirmed that it also was a pore forming protein . Evolutionary studies of Perforin-2 , have demonstrated that Perforin-2 is one of the oldest eukaryotic MACPF members , present in early metazoan phyla including Porifera ( sponges ) ( D'Angelo et al . , 2012; Wiens et al . , 2005; McCormack et al . , 2013a; McCormack et al . , 2013b ) . Orthologues of Perforin-2 are highly conserved throughout the animal kingdom ( Mah et al . , 2004; Wiens et al . , 2005; Wang et al . , 2008; He et al . , 2011; Kemp and Coyne , 2011; Green et al . , 2014 ) . Recent studies in vertebrates ( mammalia ) demonstrate that expression of Perforin-2 is not limited to macrophages , as it was also detected in murine embryonic fibroblasts ( MEF ) and human epithelial cells after bacterial infection ( Fields et al . , 2013; McCormack et al . , 2013a ) suggesting that Perforin-2 expression is tied to antibacterial activity . Similarly , in Zebrafish one of its two isoforms , MPEG1 . 2 , is induced following bacterial infection and limits bacterial burden ( Benard et al . , 2015 ) . Here we show that Perforin-2 is a major antibacterial effector protein of the innate immune system in phagocytic and in tissue forming cells . Perforin-2 is an essential innate effector protein that kills gram-positive , gram-negative , and acid-fast bacteria . The absence of Perforin-2 enables survival of pathogenic bacteria in vitro and systemic dissemination in vivo indicating that expression of Perforin-2 in professional phagocytes and in parenchymal cells is required to eliminate pathogenic bacteria in vitro and in vivo . We demonstrate that Perforin-2 can polymerize to form pores visible by negative staining transmission electron microscopy in bacterial surfaces . The presence of Perforin-2 potentiates the antibacterial activity of other known effectors including reactive oxygen and nitrogen species . In our accompanying manuscript we report some of the molecular mechanisms of Perforin-2 activation and describe how a bacterial virulence factor blocks Perforin-2 function . Professional phagocytes avidly ingest and kill bacteria . To elucidate the contribution of Perforin-2 towards their bactericidal activity , we compared professional phagocytes from Perforin-2 deficient mice with Perforin-2 heterozygous and wild-type phagocytes . Perforin-2-deficient murine peritoneal exudate macrophages ( PEM ) , neutrophils , and bone marrow-derived macrophages ( BMDM ) are unable to kill three different species of Mycobacteria ( Mycobacterium smegmatis , Mycobacterium avium , M . tuberculosis ) , as indicated by significant intracellular bacterial replication in MPEG1 ( Perforin-2 ) −/− compared to +/+ or +/− phagocytes ( Figure 1A–C , Figure 1—figure supplement 1 ) . Although BMDM express Perforin-2 constitutively , they must be activated with IFN and LPS in order to mediate Perforin-2-dependent growth inhibition of M . tuberculosis ( Mtb ) ( Figure 1—figure supplement 2 ) . This suggest that the destruction of Mtb requires both the expression and activation of Perforin-2 . 10 . 7554/eLife . 06508 . 003Figure 1 . Perforin-2 deficiency or siRNA knockdown abrogates intracellular killing of pathogenic bacteria . ( A-C ) Perforin-2 knockout , heterozygous , and wild-type macrophages and neutrophils were infected with Mycobacterium species ( A ) PEM infected with Mycobacterium smegmatis , ( B ) Neutrophils infected with Mycobacterium avium , and ( C ) BMDM infected with Mycobacterium tuberculosis . ( D-H ) Perforin-2 knockdown can be complemented in BV2 microglia cells infected with ( D ) M . avium , ( E ) M . smegmatis , ( F ) Salmonella typhimurium , and ( G ) MRSA . ( H ) Western blot demonstrating protein levels after complementation: BV2 transfected with ( Lane 1 ) Perforin-2-RFP and Perforin-2 siRNA , ( Lane 2 ) RFP and Perforin-2 siRNA , ( Lane 3 ) RFP and Perforin-2 scramble siRNA , and ( Lane 4 ) Perforin-2 siRNA alone . In western blots , Perforin-2-RFP is detected as a 105 kD band compared to the 72 kD band seen for endogenous Perforin-2 ( lane 1 and 3 respectively ) . ( I ) Human MDM infection with MRSA . = MPEG1 ( Perforin-2 ) wild-type cells ( +/+ ) , = MPEG1 ( Perforin-2 ) heterozygous cells ( +/− ) , • MPEG1 ( Perforin-2 ) knockout cells ( −/− ) . ■= RFP + Perforin-2 siRNA transfected cells , □= RFP + scramble siRNA transfected cells . ▼= Perforin-2-RFP + Perforin-2 siRNA transfected cells . One-way ANOVA with Tukey's multiple comparisons post-hoc test was used for A–G . ( A–C ) p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type cells; p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type and Perforin-2 knockout:Perforin-2 heterozygous cells . ( D–G ) p < 0 . 05 between RFP + Perforin-2 siRNA:RFP + scramble siRNA and RFP + Perforin-2 siRNA:Perforin-2-RFP + Perforin-2 siRNA . ( I ) p < 0 . 05 multiple t-tests with post-hoc correction for multiple comparisons using the Holm-Sidak method . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00310 . 7554/eLife . 06508 . 004Figure 1—figure supplement 1 . Perforin-2 genotype reflects total amount of Perforin-2 protein produced . Murine peritoneal macrophages ( A , B ) or neutrophils ( C , D ) were isolated from Perforin-2 wild-type ( wt ) , Perforin-2 heterozygous ( het ) , or Perforin-2 homozygous knockout ( ko ) mice . Western blot after probing with mouse Perforin-2 and densitometry analysis demonstrates that Perforin-2 wild-type have the greatest amount of Perforin-2 , heterozygous mice have a moderate level of Perforin-2 , and the knockout animals have no Perforin-2 protein detected . Densitometry analysis includes a minimum of three Western Blots . Statistical analysis was conducted utilizing one-way ANOVA with Tukey's multiple comparisons post-hoc test in B and D p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00410 . 7554/eLife . 06508 . 005Figure 1—figure supplement 2 . Perforin-2 dependent growth inhibition of Mtb in BMDM requires activation by LPS and IFN-γ . BMDM were collected from Perforin-2 wild-type animals ( ) , Perforin-2 heterozygous animals ( ) , or Perforin-2 knockout animals ( • ) . After differentiation to BMDM , macrophages were infected with Mtb . This experiment is representative of four different experiments . Statistical analysis was performed with one-way ANOVA with Tukey Post-hoc Multiple Comparisons . No significant difference was observed at any time point . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00510 . 7554/eLife . 06508 . 006Figure 1—figure supplement 3 . Perforin-2 is present in human MDM and PMN and can be knocked down in RA treated HL60 . ( A ) Western blot analysis of human Perforin-2 expression from two donors for primary monocyte derived macrophages and neutrophils ( PMN ) . In addition , the efficiency of human Perforin-2 knockdown is demonstrated from two separate RA treated HL-60 experiments . ( B ) Densitometry analysis of both HL-60 human Perforin-2 knockdown experiments from ( A ) . Statistical analysis was performed with Students T-test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00610 . 7554/eLife . 06508 . 007Figure 1—figure supplement 4 . Human HL-60-differentiated neutrophils require Perforin-2 to eliminate pathogens . Human HL-60-neutrophils were generated by differentiating HL-60 with RA . One day prior to the experiment , HL-60/PMN cells were transfected with either a pool of scramble ( □ ) or human Perforin-2 specific ( ■ ) siRNA . Cells were infected with ( A ) S . typhimurium , ( B ) MRSA , or ( C ) M . smegmatis 24 hr after transfection , which converged with maximal RA neutrophil differentiation . Statistical analysis was performed utilizing multiple T-tests with correction for multiple comparisons using the Holm-Sidak method . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 007 We also used Perforin-2 siRNA to ablate Perforin-2 in cells in vitro . To exclude the possibility of off-target protein effects by Perforin-2-siRNA , we performed complementation assays in BV2-microglia with C-terminally tagged Perforin-2-RFP ( Figure 1D–G ) . Endogenous Perforin-2 was silenced ( Figure 1H , lane 2 ) with siRNA specific for the 3′-untranslated sequence and the cells were complemented by transfection with siRNA-resistant Perforin-2-RFP ( Figure 1H , lane 1 ) . Only Perforin-2-RFP but not control RFP transfection restored bactericidal activity . The data indicated that RFP-tagged Perforin-2 was fully active and that siRNA ablation of Perforin-2 has negligible off-target effects on bactericidal activity . The results suggested that Perforin-2 is critical for intracellular bacterial killing . We also sought to determine the role of Perforin-2 in human professional phagocytes . Human macrophages and neutrophils express Perforin-2 protein constitutively ( Figure 1—figure supplement 3 ) . Human monocyte derived macrophages ( MDM ) efficiently killed intracellular MRSA . Perforin-2 knockdown by siRNA abrogated killing of MRSA and resulted in MRSA replication ( Figure 1 I ) . In addition , we used the human promyelocytic cell line HL-60 that differentiates into Perforin-2-expressing neutrophils upon treatment with retinoic acid ( RA ) . Perforin-2 siRNA silencing in RA-differentiated HL-60 cells abolished Perforin-2 protein expression ( Figure 1—figure supplement 3 ) and was associated with intracellular replication of Salmonella typhimurium , MRSA , and M . smegmatis ( Figure 1—figure supplement 4 ) . In contrast , scramble-transfected controls continued to express endogenous Perforin-2 and the number of recovered bacteria was reduced over several hours . This result suggested that Perforin-2 was also required for the killing of bacteria in human neutrophils ( Figure 1—figure supplement 4 ) . In summary , the results indicate that professional murine and human phagocytes require Perforin-2 to kill phagocytosed bacteria . This finding raised the question of the function of the other known bactericidal mediators in relation to Perforin-2 . Reactive oxygen ( ROS ) and reactive nitrogen ( RNS ) species are recognized for their bactericidal activity in phagocytic cells . PEM activated by IFN-γ and LPS produce both families of effectors ( Figure 2—figure supplement 1 ) . We used the well-characterized chemical inhibitors N-acetyl-L-cysteine ( NAC ) and L-NG-nitroarginine methyl ester ( L-NAME ) to block ROS and nitric oxide ( NO ) respectively as Perforin-2 and ROS or Perforin-2 and RNS knockout animals are not currently available ( Vazquez-Torres et al . , 2000; Mantena et al . , 2008; Sohn et al . , 2011 ) . First , we established that the addition of the inhibitors reduced levels of ROS and NO produced by activated PEMs ( Figure 2—figure supplement 1 ) . The role of endogenous ROS and NO on cellular bactericidal activity in PEM in the presence and absence endogenous of Perforin-2 was assessed in two complementary ways . First , we assessed the effect of chemical inhibitors of ROS and NO on killing of intracellular wild-type S . typhimurium ( Figure 2A–D ) . ROS is known to be active and produced during the first 4 hr after S . typhimurium infection in PEM ( Mastroeni et al . , 2000 ) . In Perforin-2 deficient PEM , S . typhimurium replicated equally well regardless of ROS inhibition ( Figure 2B ) . This suggests ROS had minimal influence on intracellular replication of S . typhimurium in the absence of Perforin-2 . In contrast , with PEM that express Perforin-2 , the inhibition of endogenous ROS by NAC enables S . typhimurium to replicate significantly more than mock treatment during the first 4 hr after infection , suggesting that ROS in combination with Perforin-2 helps to restrain S . typhimurium replication during this period ( Figure 2A ) . 10 . 7554/eLife . 06508 . 008Figure 2 . Antimicrobial compounds ( ROS and NO ) enhance Perforin-2 mediated killing of S . typhimurium by PEM but have limited activity in the absence of Perforin-2 . ( A–D ) Wild-type S . typhimurium infection of PEMs isolated from either MPEG1 ( Perforin-2 ) +/+ ( A , C ) , or MPEG1 ( Perforin-2 ) −/− mice ( B , D ) . Non-filled symbols indicated MPEG1 ( Perforin-2 ) +/+ PEMs; whereas filled symbols are MPEG1 ( Perforin-2 ) −/− PEMs . Cells were incubated with NAC ( blue line ) , NAME ( green line ) , or mock ( black line ) . To assess bacterial resistance mechanisms against these effectors , ( E ) SodC1 or ( F ) HmpA knockout S . typhimurium were used to infect MPEG1 ( Perforin-2 ) −/− or +/+ PEMs . The above experiments were conducted with six biologic replicates and are representative of four independent experiments . Statistical analysis was performed utilizing multiple T-tests with correction for multiple comparisons using the Holm-Sidak method . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00810 . 7554/eLife . 06508 . 009Figure 2—figure supplement 1 . Nitrite and reactive oxygen production in PEMs following addition of inhibitors . ( A ) PEM nitrite production following stimulation with LPS and IFN-γ and incubation with ROS ( NAC ) or NO ( NAME ) inhibitors as indicated . ( B ) Reactive oxygen production of PEMs after LPS or PMA stimulation with addition of indicated inhibitors . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 00910 . 7554/eLife . 06508 . 010Figure 2—figure supplement 2 . Antimicrobial compounds ( ROS and NO ) enhance Perforin-2 mediated killing of M . smegmatis by PEM . PEMs were isolated from either ( A , C ) wild-type , or ( B , D ) Perforin-2 knockout mice . □= MPEG1 ( Perforin-2 ) +/+ PEMs , ■= MPEG1 ( Perforin-2 ) −/− PEMs . Cells were incubated with NAC ( blue line ) , NAME ( green line ) , or mock ( black line ) . The above graphs were conducted with eight biologic replicates and are representative of three experiments . Statistical analysis was performed utilizing multiple T-tests with correction for multiple comparisons using the Holm-Sidak method . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 010 Inhibition of endogenous NO production with L-NAME in the presence of Perforin-2 also allowed increased intracellular S . typhimurium replication beginning several hours post-infection ( Figure 2C ) coinciding with the known time period required for onset of NO production ( see Figure 2—figure supplement 1 ) ( Mastroeni et al . , 2000 ) . As with ROS , endogenous NO had little effect on S . typhimurium replication in the absence of Perforin-2 ( Figure 2D ) . The results indicate that although ROS and NO contribute towards the intracellular killing of bacterial pathogens , their bactericidal activity is largely dependent upon the presence of Perforin-2 . To validate that the cooperation between Perforin-2 and ROS and NO was not specific to S . typhimurium , we repeated the above with M . smegmatis . The cooperative activity of Perforin-2 with ROS and NO was also evident in killing of M . smegmatis in PEM ( Figure 2—figure supplement 2 ) . Pharmacologic inhibitors significantly increased bacterial survival only when Perforin-2 was present , with little additional effect when Perforin-2 was absent . This data suggested that the critical importance of Perforin-2 in facilitating ROS and NO activity is not specific to only S . typhimurium . To further investigate the dependence of ROS and NO bactericidal activity upon Perforin-2 we utilized S . typhimurium mutants lacking the periplasmic superoxide dismutase ( sodC1 ) or flavohemoglobin ( hmpA ) genes , which neutralize ROS or NO , respectively , in intracellular killing assays . ( Stevanin et al . , 2002; Uzzau et al . , 2002; Krishnakumar et al . , 2004; Prior et al . , 2009 ) . If ROS and NO activity is dependent upon Perforin-2 we hypothesized that bacterial ROS and NO defense mechanisms would be unnecessary in Perforin-2 knockout cells . Unlike other superoxide dismutases that protect bacteria from oxygen radicals produced intracellularly as a byproduct of cellular respiration , SodC1 is a periplasmic superoxide dismutase . In vivo , sodC1 mutants are significantly attenuated relative to wild-type S . typhimurium ( De Groote et al . , 1997; Fang et al . , 1999; Krishnakumar et al . , 2004 ) . Flavohemoglobin acts by either catalyzing an O2-dependent denitrosylase reaction converting NO to a nitrate ion or N2O , or an anoxic reductive reaction forming NO− . As with SodC1 , in vivo and in vitro studies substantiate the role of HmpA with significant attenuation observed with HmpA deficient bacteria ( Stevanin et al . , 2002; Bang et al . , 2006 ) . In the presence of Perforin-2 , SodC1-deficient S . typhimurium were killed more efficiently than wild-type S . typhimurium . However , SodC1-deficient S . typhimurium replicate similar to wild-type S . typhimurium when Perforin-2 was absent ( Figure 2E ) . Similarly , hmpA mutants are more susceptible to killing by NO than wild-type bacteria , but only when Perforin-2 is present . In the absence of Perforin-2 , Flavohemoglobin does not enhance the survival and replication of wild-type S . typhimurium relative to the hmpA mutant . Thus , both chemical and genetic analyses indicate that Perforin-2 is required for the bactericidal activity of ROS and NO in macrophages . The induction of Perforin-2 in certain parenchymal cells was reported previously ( Fields et al . , 2013; McCormack et al . , 2013 ) . We expanded this analysis for many human and murine primary cells and established cell lines , ranging from epithelial to endothelial cells , from astrocytes to myoblasts , and from neural cells to secretory cells ( Table 1 , Table 2 ) . Every cell type derived from ectodermal , neuroectodermal , endodermal , or mesodermal lineage tested to date is able to express Perforin-2 message either constitutively or after type I or II IFN induction . Table 1 ( murine cells ) and Table 2 ( human cells ) summarize these results while their respective supplements ( Supplementary files 1 , 2 ) show the inducibility of Perforin-2′s mRNA and protein ( qPCR of ΔCT of Perforin-2 normalized to GAPDH and western blot analysis ) . Moreover , all cell types analyzed ( 54 out of 54 ) are able to kill bacteria in an in vitro bactericidal assay when Perforin-2 is expressed . When infection occurs prior to Perforin-2 induction or when Perforin-2 is siRNA-ablated or genetically deficient using the above assay , bacteria were not killed by cells and consequently replicate . In contrast , cells that express Perforin-2 were bactericidal . These results suggest that Perforin-2 can be expressed ubiquitously to defend cells against bacterial invasion . 10 . 7554/eLife . 06508 . 011Table 1 . Murine perforin-2 expressionDOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 011Cell type:Perforin-2 expression:Peritoneal macrophageConstitutiveBone marrow derived macrophage ( BMDM ) ConstitutiveBone marrow derived dendritic cell ( BMDC ) ConstitutiveBV-2 microglia cell lineConstitutiveRaw264 . 7 macrophage cell lineConstitutiveJ774A . 1 macrophage cell lineConstitutiveMicrogliaConstitutiveNeutrophil ( peritoneum stimulation ) ConstitutiveNeutrophil ( bone marrow ) ConstitutiveGamma delta ( γδ ) T cell ( from Skin ) ConstitutiveGamma delta ( γδ ) T cell ( from Gut ) ConstitutiveGamma delta ( γδ ) T cell ( from Vagina ) ConstitutiveMarginal zone B cellConstitutiveKeratinocyte ( Back ) ConstitutiveIntestinal epithelial cellsConstitutiveSplenocytesConstitutiveOT1 CD8 T cell induced with TGFβ , RA , and IL2ConstitutiveOT1 CD8 T cellInducibleCD4 T cellInducibleB cellInducibleAstrocyteInducibleNeuronInducibleCath . a neuroblastoma cell lineInducibleNeuro-2A neuroblastoma cell lineInducibleAdult CNS fibroblastInducibleEmbryonic fibroblastInducibleNIH 3T3 fibroblast cell lineInducibleBalb/c 3T3 fibroblast cell lineInducibleC2C12 myoblast cell lineInducibleNeonatal ventricular myocytesInducibleCMT-93 rectal carcinoma cell lineInducibleCT26 colon carcinoma cell lineInducibleB16-F10 melanoma cell lineInducibleB16-F0 melanoma cell lineInducibleMOVCAR 5009 ovarian cancer cell lineInducibleMOVCAR 5447 ovarian cancer cell lineInducibleLL/2 Lewis lung carcinoma cell lineInducibleED-1 lung adenocarcinoma cell lineInducibleItalics: Ex vivo primary cells utilized for analysis . 10 . 7554/eLife . 06508 . 012Table 2 . Human peforin-2 expressionDOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 012Cell type:Perforin-2 expression:Monocyte derived macrophage ( MDM ) ConstitutiveMonocyte derived dendritic cell ( MDC ) ConstitutivePBMC isolated NK cellConstitutivePolymorphonuclear granulocyte ( neutrophil ) ConstitutiveHL-60 promyelocyte cell line RA differentiated to PMNConstitutiveHL-60 cell line PMA differentiated to MacrophageConstitutiveFetal keratinocyteConstitutiveAdult keratinocyteConstitutivePMA differentiated Thp-1 monocyte cell lineConstitutiveNK-92 cell lineConstitutiveNormal colon biopsyConstitutiveNormal skin biopsyConstitutiveUmbilical endothelial cell ( HUVEC ) InducibleHeLa cervical carcinoma cell lineInducibleA2EN endocervical epithelial cell lineInducibleUM-UC-3 bladder cancer cell lineInducibleUM-UC-9 bladder cancer cell lineInducibleCaCo-2 colorectal carcinoma cell lineInducibleHEK293 embryonal kidney cell lineInducibleMIA-PaCa-2 pancreatic cancer cell lineInducibleSkin fibroblastInducibleThp-1 monocyte cell lineInducibleHL-60 promyelocyte cell lineInducibleOVCAR3 ovarian carcinoma cell lineInducibleA549 alveolar adenocarcinoma cell lineInducibleU-1752 bronchiolar epithelial cell lineInducibleJeg-3 placental choriocarcinoma cell lineInducibleItalics: Ex vivo primary cells utilized for analysis . Perforin-2 siRNA knockdown was used to determine its contribution towards intracellular killing of bacteria by IFN-induced murine and human parenchymal cells . Although IFN induces hundreds of antimicrobial genes in addition to Perforin-2 , silencing of Perforin-2 alone was sufficient to cause bacterial replication . Without exception , Perforin-2 expression and function were essential for killing a diverse array of intracellular pathogenic bacteria by parenchymal or phagocytic cells . Examples of bactericidal activity include human vascular endothelial cells ( HUVEC ) ; human pancreatic cells ( MIA-PaCa-2 ) ; human uroepithelial cells ( UM-UC-9 ) ; murine ovarian epithelial cells ( MOVCAR 5009 ) ; murine colon epithelial cells ( CT26 ) ; and murine cardiac myoblasts ( C2C12 ) , respectively ( Figure 3 and Supplementary file 3 ) . Examples of human and murine siRNA-mediated Perforin-2 protein knockdown include HUVECs and myoblasts ( Figure 3 , Supplementary file 4 ) . To certify that Perforin-2 siRNA targeting was specific , siRNA resistant Perforin-2-RFP was utilized to complement Perforin-2 siRNA in parenchymal tissue forming cells . Figure 3—figure supplement 1 highlights representative examples of Perforin-2 complementation in myoblasts , intestinal epithelial cells , and PEM . 10 . 7554/eLife . 06508 . 013Figure 3 . Perforin-2 significantly contributes to intracellular killing in non-hematopoietically derived cells . One day prior to the infection , cells were transfected with either a pool of scramble ( □ ) or Perforin-2 specific ( ■ ) siRNA and 14 hr prior to the infection induced with IFN-γ . ( A ) HUVEC cells infected with M . smegmatis , ( B ) MIA-PaCa-2 cells infected with S . typhimurium , ( C ) UM-UC-9 infected with MRSA , ( D ) Perforin-2 MEF infected with MRSA , ( E ) Human Kc infected with MRSA induced with IFN-γ , ( F ) Human Kc infected with MRSA with no IFN-γ induction . = MPEG1 ( Perforin-2 ) +/+ , = MPEG1 ( Perforin-2 ) +/− , •= MPEG1 ( Perforin-2 ) −/− . ( A–C , E , F ) The above graphs contain 5–9 biologic replicates , and are representative of 3–7 independent experiments . Statistical analysis was performed utilizing multiple T-tests with correction for multiple comparisons using the Holm-Sidak method . p < 0 . 05 . ( D ) One-way ANOVA with Tukey post-hoc multiple comparisons . p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type mice p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type and Perforin-2 knockout:Perforin-2 heterozygous mice . p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type , Perforin-2 knockout:Perforin-2 heterozygous , and Perforin-2 heterozygous:Perforin-2 wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 01310 . 7554/eLife . 06508 . 014Figure 3—figure supplement 1 . Perforin-2 knockdown is complementable and is able to replicate endogenous Perforin-2 bactericidal function . Murine cells were transfected with either a pool of murine Perforin-2 specific siRNA and a RFP vector control plasmid ( ■ ) ; a pool of scramble siRNA and a RFP vector control plasmid ( □ ) ; or a pool of murine Perforin-2 specific siRNA and a siRNA resistant Perforin-2-RFP vector ( ▼ ) and stimulated for 14 hr with IFN-γ . ( A ) C2C12 infected with S . typhimurium , ( B ) CMT93 infected with MRSA , and ( C ) PEM infected with M . smegmatis . The above graphs were conducted with biologic triplicates and are representative of four experiments . Statistical analysis was performed with one-way ANOVA with Tukey Post-hoc Multiple Comparisons . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 01410 . 7554/eLife . 06508 . 015Figure 3—figure supplement 2 . Perforin-2 protein expression in human primary keratinocytes after knockdown compared to human monocyte derived macrophages ( MDM ) . Human primary keratinocytes were transfected with a pool of scramble siRNA or human Perforin-2 specific siRNA 24 hr prior to cell collection for western blot . ( A ) Western blot analysis demonstrating Perforin-2 protein levels in ( Lane 1 ) human MDM , ( Lane 2 ) human keratinocytes with Perforin-2 siRNA ablation , ( Lane 3 ) human keratinocytes transfected with scramble siRNA . ( B ) Densitometry analysis from A analyzing five different experiments . Statistical analysis was performed with one-way ANOVA with Tukey post-hoc Multiple comparisons . p < 0 . 05 , ns = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 015 To further validate the requirement of Perforin-2 for bactericidal activity in non-phagocytic cells , we used genetically deficient MEFs obtained from MPEG1 ( Perforin-2 ) +/+ , +/− , and −/− littermates . After overnight induction with IFN-γ , MPEG1 ( Perforin-2 ) +/+ MEFs eliminate MRSA . In contrast , IFN-γ treated Perforin-2 −/− MEFs enable MRSA to replicate . Heterozygous MEFs had intermediate bactericidal activity ( Figure 3D ) suggesting a gene dose effect of Perforin-2 . Keratinocytes ( Kc ) represent the first cellular barrier to infection in the skin ( Song et al . , 2002; Mempel et al . , 2003; Bernard and Gallo , 2011 ) . Unlike other parenchymal cells that need to be induced , primary human Kc express Perforin-2 constitutively . The identity of human Perforin-2 in Kc was confirmed by western blotting ( Figure 3—figure supplement 2 ) . Densitometry analysis suggests similar Perforin-2 protein levels in MDM and Kc ( Figure 3—figure supplement 2 ) . Expression could also be silenced with human Perforin-2 siRNA ( Figure 3—figure supplement 2 , lane 2 ) , but not with scramble siRNA ( Figure 3—figure supplement 2 , lane 3 ) . The inhibition of Perforin-2 expression abrogated the bactericidal activity of Kc ( Figure 3E ) that were able to kill MRSA , irrespective of prior IFN activation ( Figure 3F ) . Cumulatively , our results suggest that Perforin-2 is an effector for cellular defense against pathogenic bacteria in professional phagocytes and in other cells . The ubiquity of Perforin-2 expression suggests a critical importance in cellular defenses of many , if not all , tissue against pathogenic bacteria . The findings raise the question of the molecular mechanisms by which Perforin-2 exerts its potent bactericidal function . The MACPF domain of Perforin-2 suggests that it is a pore-forming protein similar to the pore-formers of complement ( C9 ) and cytotoxic lymphocytes ( Perforin-1 ) ( Podack and Tschopp , 1982; Dennert and Podack , 1983; Law et al . , 2010 ) . In analogy to C9 and perforin-1 , pore-formation by the MACPF domain of Perforin-2 on the bacterial surface may constitute the lethal hit . However , Perforin-2 , unlike C9 and Perforin-1 , is anchored in membrane vesicles with its MACPF domain predicted to reside inside the vesicle or outside on the plasma membrane ( Figure 4A ) . Therefore we decided to study the cell biology of Perforin-2 in resting cells and following bacterial infection . 10 . 7554/eLife . 06508 . 016Figure 4 . Endogenous Perforin-2 is located in intracellular sites allowing for rapid translocation to bacteria . ( A ) Schematic demonstrating proposed orientation of Perforin-2 in vesicles . ( B ) Fractionation results of endogenous Perforin-2 from human macrophages . ( Lane L ) is a post-nuclear lysate control , ( Lane 1–8 ) are individual fractions corresponding with specific indicated organelles . ( C ) Overexpression of murine Perforin-2-GFP in murine BV2 microglial cells . ( D–F ) Confocal images taken 5 min after S . typhimurium infection in Perforin-2-GFP + Perforin-2 siRNA transfected BV2 cells . White arrows denote extracellular S . typhimurium , red arrows highlight a DNA cloud corresponding with S . typhimurium ( D ) DAPI only , ( E ) Perforin-2-GFP only , ( F ) Merge of DAPI and Perforin-2-GFP . ( G–I ) Confocal images taken 5 min after Escherichia coli-GFP infection in Perforin-2-RFP + Perforin-2 siRNA transfected BV2 cells . Arrows point to extracellular E . coli-GFP that has made contact but is still extracellular with normal bacilli morphology maintained . ( G ) E . coli-GFP only , ( H ) Perforin-2-RFP only , and ( I ) merge E . coli-GFP and Perforin-2-RFP . Fractions in B were probed as follows: Cytoplasm—MEK1/2; Early Endosome—EEA1; Lysosome—Lamp1; ER—calreticulin; Golgi—Golgin-97; Mitochondria—Prohibitin; Peroxisome—Catalase; Plasma Membrane—Cadherin . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 01610 . 7554/eLife . 06508 . 017Figure 4—figure supplement 1 . Perforin-2-RFP colocalizes with ER , Golgi , and early endosomes . ( A-L ) We used RAW264 . 7 macrophages which constitutively express Perforin-2 and studied the subcellular localization of fluorescent Perforin-2 . RAW264 . 7 macrophages were transfected with Perforin-2-RFP and stimulated with IFN-γ and LPS for 14 hr to induce a shift towards M1 macrophages . Cells were fixed and stained as indicated . These images are representative of 3 separate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 01710 . 7554/eLife . 06508 . 018Figure 4—figure supplement 2 . S . typhimurium infection of macrophages with Perforin-2 and DAPI localization through the cell ( multiple Z-sections ) . ( A-F ) BV2 cell line overexpressing Perforin-2-GFP infected with S . typhimurium . Images are collected 5 min post-infection . Arrows indicate extracellular S . typhimurium that has maintained the normal shape of S . typhimurium-likely attributed to not being surrounded by the bactericidal Perforin-2-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 01810 . 7554/eLife . 06508 . 019Figure 4—figure supplement 3 . Perforin-2-RFP colocalizes with E . coli-GFP within minutes of infection . BV2 microglia were transfected with Perforin-2-RFP and stimulated overnight with IFN-γ . Transfected cells were infected with E . coli-GFP for several minutes upon which the cells were fixed and imaged . Arrows point to extracellular E . coli-GFP . ( A ) Perforin-2-RFP only , ( B ) E . coli-GFP only , ( C ) Merge of Perforin-2-RFP and E . Coli-GFP . Yellow in ( C ) corresponds with colocalization of Perforin-2-RFP with E . coli-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 019 In lieu of antibodies capable of detecting native endogenous Perforin-2 , as these reagents are not currently available , we transfected c-terminal tagged Perforin-2–RFP or–GFP into cells in which endogenous Perforin-2 was knocked down . The Perforin-2 fusion proteins were functionally active as shown above in complementation studies ( Figure 1 , Figure 3—figure supplement 1 ) . In a murine macrophage cell line ( RAW264 . 7 ) tagged murine Perforin-2 was localized to ER , Golgi , and early endosomes based on its overlay with ER tracker , GM130 , and EEA1 respectively ( Figure 4 , Figure 4—figure supplement 1 ) . To validate the results obtained thus far with the Perforin-2 fusion proteins , we also analyzed endogenous human Perforin-2 in fractions of post-nuclear homogenates of human monocytic leukemia ( THP-1 ) differentiated to a macrophage phenotype ( Figure 4B ) . Upon density gradient ultracentrifugation , Perforin-2 co-sedimented with membrane vesicles staining with markers for ER , Golgi , early endosome , or plasma membrane . Perforin-2 was not detected in fractions containing mitochondrial , peroxisomal , or lysosomal membranes , nor in the cytoplasm . As Perforin-2 localizes to the same organelles following co-sedimentation of endogenous Perforin-2 or imaging after overexpressed c terminal tagged Perforin-2 , it is unlikely that the tag will disrupt Perforin-2 localization following bacterial challenge . In non-infected murine BV2 microglial cells , Perforin-2-GFP vesicles were distributed in a membrane pattern throughout the cell ( Figure 4B , C ) . Following S . typhimurium infection , Perforin-2-GFP redistributed and accumulated , within minutes , on endosomal/phagosomal bodies ( Figure 4D–F , red arrows ) that contained diffuse DNA as indicated by DAPI staining of phagocytosed S . typhimurium as compared to an intact extracellular S . typhimurium bacillus with its typical rod-like morphology ( white arrow ) . Figure 4—figure supplement 2 demonstrated a series of z-sections through the same cell suggesting that phagocytosis of several S . typhimurium bacilli had occurred . These images suggested that bacterial DNA was released from S . typhimurium killed by Perforin-2-GFP carried in vesicular membranes . To directly demonstrate the presence of bacteria in Perforin-2-RFP-containing endosomes , we infected microglial cells expressing Perforin-2-RFP with Escherichia coli expressing GFP . Fluorescent imaging indicated that Perforin-2-RFP is largely concentrated on phagosomes that contain E . coli-GFP ( Figure 4G–I ) . Phagocytosis and killing of E . coli-GFP ranged from incipient ingestion of largely intact E . coli-GFP to phagosomes containing green fluorescence that may have been released from killed E . coli-GFP . Merging green and red fluorescence shows co-localization . Regions with only green or red fluorescence suggested that a green bacterium being phagocytosed had not yet fused with the Perforin-2-RFP expressing endosome . Figure 4—figure supplement 3 shows lower magnification field views . Overall , these studies demonstrated that Perforin-2 is able to rapidly translocate to endosomal membranes , or in some cells may already be localized in these membranes in order to trap intracellular bacteria ( upon fusion with phagosomes ) . During this process GFP is released from GFP synthesizing bacteria suggesting bacterial demise . Similarly , the appearance of diffuse bacterial DNA inside the vesicle suggests that integrity of the bacteria has been compromised . Owing to Perforin-2′s highly conserved MACPF domain , we next were interested in elucidating if Perforin-2 and its MACPF domain can form pores . To address the hypothesis that Perforin-2 can form pores in membranes akin to C9 of complement and Perforin-1 we used an artificial system by overexpressing Perforin-2 in HEK-293 cells , followed by the isolation of post-nuclear membranes , limited proteolysis , and their analysis by negative staining electron microscopy ( EM ) . The cytoplasmic domain of Perforin-2 has a predicted trypsin cleavage site close to the N-terminus of its transmembrane domain which may be functionally important . We observed clusters of typical membrane pores of ∼100 Å inner diameter in these membranes ( Figure 5A , B ) that were morphologically similar to poly-C9 and poly-Perforin-1 pores on bacterial and cell membranes ( Schreiber et al . , 1979; Podack and Tschopp , 1982; Podack and Dennert , 1983 ) . Pores were not observed on any membranes in the absence of trypsin treatment . These experiments suggest that Perforin-2 , putatively via its MACPF domain , is able to form clusters of regular polymeric pores on membranes akin to C9 of complement and Perforin-1 . Although our methodology facilitated visualization of poly Perforin-2 pores , we are not suggesting that Perforin-2 is typically activated by trypsin nor are we suggesting that eukaryotic membranes are typical targets of Perforin-2 . 10 . 7554/eLife . 06508 . 020Figure 5 . Perforin-2 forms pores in bacterial surfaces after infection and in Perforin-2 overexpressed eukaryotic membranes that are visible by negative stain transmission electron microscopy ( TEM ) . ( A , B ) Electron micrograph of polymerized Perforin-2 membrane lesions from Perforin-2-GFP transfected HEK-293 cells , with Perforin-2 activated to form pores by trypsin digestion to the enriched membrane fraction . Panel A Demonstrates the quantity of pores on the Perforin-2 overexpressed membranes after trypsin activation . Panel B denotes a higher magnification to illustrate the uniform pore structure . ( C-G ) Perforin-2 wild-type MEFs were treated with IFN-γ for 14 hr , and infected with ( C–E ) MRSA or ( F , G ) M . smegmatis . After 5 hr the infected bacteria were isolated and imaged utilizing negative stain TEM . Arrows point to black , stain-filled pores on the bacterial cell wall surrounded by white , stain excluding borders created by polymerized Perforin-2 . Round pores measure 8 . 5–10 nm inner diameter , the size typical for polymerized Perforin-2-pores . Panels E and G are close-up images of the boxed region in C and F . Blinded quantification of pore amount with different conditions is demonstrated in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02010 . 7554/eLife . 06508 . 021Figure 5—figure supplement 1 . Quantification of pores from negative stain transmission electron microscopy . 10 blinded fields were counted from the above eight conditions . Pores are defined as regions containing stain ( corresponding to a ‘hole” in the surface ) that is approximately 10 nm in diameter . Around the “hole” ultrastructure needs to be visible to ensure that the presumed “hole” is actually a pore . Statistical analysis was performed by with one-way ANOVA with Tukey Post-hoc Multiple Comparisons . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 021 The MACPF domain has been shown to be the pore-forming domain of Perforin-1 and of the membrane attack complex of complement ( Hadders et al . , 2007; Rosado et al . , 2007; Slade et al . , 2008; Law et al . , 2010 ) . The physical proximity of the pore-forming MACPF domain of Perforin-2 to endocytosed bacteria suggested that bacteria killed inside the endosomes might also exhibit electron microscopic images typical for Perforin-2-lesions on bacterial surfaces/membranes . To test this hypothesis , we induced Perforin-2 expression in MEF overnight with IFN-γ and infected the following day with M . smegmatis or MRSA . Intracellular bacteria were then re-isolated from MEFs after 5 hr and bacterial surfaces were inspected by negative-staining electron microscopy ( Figure 5C–G ) . Large patches of densely clustered pores of ca . 100 Å inner diameter are visualized on both MRSA and M . smegmatis surfaces with similar irregularities as those observed in complement lesions on E . coli surfaces ( Schreiber et al . , 1979 ) . MRSA bacterial surfaces are more hydrophilic and show more uniform negative staining ( Figure 5C–E ) when compared to M . smegmatis samples that appear washed out with lower detail and contrast . This observed difference is due to the high hydrophobicity of the mycobacterial surface that repels negative stain except where it accumulates in the pore ( Figure 5F , G ) ( Noda and Kanemasa , 1986; Stokes et al . , 2004 ) . Pores are only visible after bacteria were isolated from cells expressing Perforin-2 . Pores were not present on control bacteria or after isolation of bacteria from Perforin-2 deficient cells as determined by blinded quantification of the pore-like structures ( Figure 5—figure supplement 1 ) . There were significantly greater numbers of pore-like structures ( two or more orders of magnitude ) associated with bacteria isolated from Perforin-2 expressing cells compared to Perforin-2 deficient cells . This indicates that there is a strong correlation between pore-formation and Perforin-2 . Although it is not yet technically possible to prove definitively that the visualized pores are poly- Perforin-2 , the existence of Perforin-2 pores is contingent upon a physical association between Perforin-2 and bacterial cells . Since we have a panel of antibodies that recognize denatured Perforin-2 , we reasoned that it is possible to prove or disprove the latter . To accomplish this , we utilized five antibodies raised against peptides from different regions of human Perforin-2 ( Figure 6—figure supplement 1 ) . Perforin-2−/− MEFs were transfected with either GFP or human Perforin-2-GFP plasmids and infected for 1 hr with a gram-negative bacteria ( Enteropathogenic E . coli , EPEC ) or gram-positive bacteria ( MRSA ) . Of note these two species were chosen due to their common extracellular preference in order to improve the isolation of bacteria undergoing Perforin-2 mediated bacteriolysis following infection with eukaryotic cells . In order to determine eukaryotic cellular contamination following bacterial purification , non-infected cells were processed in parallel as a control . Bacterial isolation and enrichment was successful as all fractions and filtrates were below detection for contaminating mammalian membranes and cellular debris as determined by immunoblot with antibodies targeting murine clathrin , actin , and GFP ( data not shown ) . Following the isolation procedure and subsequent differential rounds of centrifugation , Perforin-2 reactive bands with different molecular weights were detected in fractions containing either EPEC or MRSA . Perforin-2 was not detected after these fractions were passed through 0 . 22 μM filters ( Figure 6 ) . Prior to filtration each sample was spiked with goat IgG to account for nonspecific loss of soluble proteins . Unlike Perforin-2 , goat IgG was equally detected in both bacterial fractions and filtrates . This indicates that Perforin-2 was not present as a soluble protein nor associated with cellular debris or microvesicles . Rather , Perforin-2 was associated with particles larger than 0 . 22 microns , such as bacterial cells . This conclusion was supported by the detection of the EPEC transcription factor ADA in the bacterial fractions but not filtrates . Likewise , MRSA Penicillin Binding Protein ( PBP ) was only detected in the unfiltered bacterial fractions . 10 . 7554/eLife . 06508 . 022Figure 6 . Cleaved Perforin-2 recovered after cellular infection with bacteria . Perforin-2 −/− MEFs were transfected with human Perforin-2-GFP or GFP induced with murine IFN-γ and infected with the extracellular bacteria MRSA or EPEC . Following infection , bacteria were isolated , goat IgG was added to assess for nonspecific protein loss , and a portion filtered to distinguish bacteria from debris/soluble proteins . A noninfected control ( first lane ) , demonstrates the selectivity of the differential centrifugations to remove mammalian cells . Figure 6—figure supplement 1 illustrates the recognition domain for human each Perforin-2 specific peptide generated antibody as well as verification . Fractions were probed against peptide-generated antibodies against the Perforin-2 domain ( C93 ) , and peptide generated antibodies against the MACPF domain ( C186 , C252 ) . Commercial antibodies against ADA of EPEC and PBP of MRSA were utilized as bacterial markers . An additional band was observed following PBP immunoblot with a slightly higher molecular weight . This band was unspecific because it occurred in all samples including those derived from the experiments using EPEC . No signal was detected with previously validated peptide derived antibodies targeting the cytoplasmic domain of human Perforin-2 ( C174 ) , or peptide derived antibodies targeting a N-terminal portion of the Perforin-2 domain ( C267 ) ( Data not shown ) . In addition , commercial anti-human Perforin-2 antibody ( detecting the cytoplasmic domain ) , clathrin , actin , and GFP also did not generate any signal ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02210 . 7554/eLife . 06508 . 023Figure 6—figure supplement 1 . Human Perforin-2 peptide antibody validation . ( A ) Six peptide antibodies were generated against different regions of human Perforin-2 . Two antibodies detect in the MACPF domain of Perforin-2: C252 detects in the N-terminal portion; C186 detects in the C terminal portion of the domain . Two antibodies detect in the Perforin-2 domain of Perforin-2: C93 detects in the N terminal portion; C267 detects in the C terminal portion of the domain . Two antibodies detect in the same portion of the cytoplasmic tail of Perfroin-2: C246 and C174 ( B ) Antibodies were screened against either Perforin-2 constitutively expressing human macrophages from PMA differentiated Thp1s or murine macrophages ( BV2 microglia cell line ) . All bands are located at ∼72kD , the predicted weight of Perforin-2 . ( C , D ) Example antibodies C186 and C174 with increasing concentrations of Thp1 or BV2 protein loaded to assess detection sensitivity of the peptide antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 023 Antibodies directed to the amino-terminal domains detected Perforin-2 in bacterial fractions . Two different banding patterns were observed depending upon which region the peptide derived Perforin-2 antibodies recognized ( Figure 6—figure supplement 1 ) . A single ca . 50 kD band was detected after probing with an antibody targeting amino-terminal portion of the Perforin-2 domain . Probing with either carboxy- or amino-terminal MACPF domain antibodies recognized two bands in the bacterial fractions . One of these bands is the same size as the ca . 50kD band observed with the amino-terminal detecting Perforin-2 domain antibody , suggesting that this band is a combination of the MACPF domain and an amino-terminal portion of the Perforin-2 domain . Both MACPF domain antibodies also identified another band of ca . 37kD that is not detected with the Perforin-2 domain antibody , indicative of a MACPF domain fragment . Antibodies raised against the carboxy-terminal region of the Perforin-2 domain , the cytoplasmic domain , or the GFP tag were negative in unfiltered bacterial fractions despite validation of both Perforin-2 peptide antibodies with human macrophage whole cell lysates ( data not shown , Figure 6—figure supplement 1 ) . These results indicate that Perforin-2 makes physical contact with bacteria are supports are interpretation of electron dense poly-Perforin-2 pores . We next sought to extend our in vitro studies to in vivo models using Perforin-2 deficient mice . Perforin-2 was neither vital for developmental processes nor for the control of commensal bacteria under specific pathogen-free conditions . This specifically included the regular development of the innate and adaptive immune system ( Figure 7—figure supplement 1 ) . This permitted us to study the antibacterial activity of Perforin-2 in vivo with a traditionally sub-lethal dose of MRSA and later other pathogens ( S . typhimurium ) . S . aureus is part of the normal bacterial skin flora in humans , but it can also become a major cause of serious skin and systemic infections . S . aureus's pathogenic arsenal , coupled with significant complement and antibiotic resistance , has allowed MRSA to evolve into a life-threatening , antibiotic-resistant pathogen in both community acquired as well as nosocomial settings . To determine the bactericidal role of Perforin-2 against MRSA in vivo , we utilized an epicutaneous MRSA challenge model for mice ( Cheng et al . , 2009; Wanke et al . , 2013 ) in which hair removal is followed by ‘tape-stripping’ ( Wanke et al . , 2013 ) to disrupt the keratin barrier while exposing intact keratinocytes without overt wounding . MPEG1 ( Perforin-2 ) −/− , +/− , and +/+ littermates were challenged on the tape-stripped skin with 109 CFU of MRSA , a PFGE type USA300 clinical isolate . Perforin-2 knockout mice exhibited significantly decreased survival and more weight loss compared to Perforin-2 heterozygous or wild-type littermates ( Figure 7A , B , Figure 7—figure supplement 2 ) . To investigate the rate and route of the bacterial spread , 7 animals from each of the three genotypes were sacrificed 6 and 12 days following infection and the colony forming units ( CFU ) in their spleen , kidney , blood , and skin were determined ( Figure 7 , Figure 7—figure supplement 3 ) . On day 6 , all groups showed signs of systemic MRSA infection ( recoverable CFU from internal organs ) . However , CFUs in Perforin-2 −/− mice were significantly higher than in wild-type mice ( Figure 7—figure supplement 3 ) and by day 12 , Perforin-2 deficient mice continued to have bacteremia with 100 to 100 , 000 fold higher MRSA counts in their organs as compared to Perforin-2 heterozygous or wild-type littermates ( Figure 7B–E ) . The majority of heterozygous animals completely cleared the infection , with only a few mice demonstrating recoverable CFU in internal organs and the skin at day 12 ( Figure 7B–E ) while MRSA of the wild-type animals could only be recovered from the skin of three animals . On the other hand , all Perforin-2 knockout mice failed to eliminate MRSA and recover their weight . These animals eventually succumbed to infection ( Figure 7A , Figure 7—figure supplement 2 ) . Similar trends were observed with 129X1/SVJ congenic animals ( Figure 7B ) indicating that MRSA mortality was controlled by Perforin-2 and not attributable to confounding passenger mutations or genetic background differences between 129X1/SVJ and C57Bl/6 mice ( Vanden Berghe et al . , 2015 ) . 10 . 7554/eLife . 06508 . 024Figure 7 . Perforin-2 is required for in vivo survival after MRSA epicutaneous challenge . ( A ) Aggregated survival curves of 60 C57BL/6 × 129 × 1/SJV mice challenged epicutaneously with 109 MRSA . ( B ) Aggregated survival curves of 75 , 129X1/SVJ mice challenged epicutaneously with 109 MRSA . ( C–F ) Organ load twelve days after MRSA epicutaneous infection in ( C ) blood , ( D ) spleen , ( E ) kidney , and ( F ) skin . ( G ) Perforin-2 ex vivo infection of murine neutrophils with MRSA . = MPEG1 ( Perforin-2 ) wild-type animals ( +/+ ) , = MPEG1 ( Perforin-2 ) heterozygous animals ( +/− ) , •= MPEG1 ( Perforin-2 ) knockout animals ( −/− ) . Log-rank ( Mantel–Cox ) test was performed for A and B with statistical significance p < 0 . 0001 . One-way ANOVA with Tukey post-hoc multiple comparisons was performed in C-G . p < 0 . 05 as indicated . p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type and Perforin-2 knockout:Perforin-2 heterozygous neutrophils . p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type , Perforin-2 knockout:Perforin-2 heterozygous , and Perforin-2 heterozygous:Perforin-2 wild-type neutrophils . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02410 . 7554/eLife . 06508 . 025Figure 7—figure supplement 1 . Characterization of lymphocytes in Perforin-2 knockout mice . Frequency of CD4 , CD8 , and NK cells in the peripheral blood of Perforin-2 knockout and wild-type littermate mice . The phenotype of peripheral blood lymphocytes was determined by 7-color flow cytometer analysis . Each bar ( A–D , G ) represents the Mean ± SD from 10 to 14 mice , numbers represent the percentage of cells among live lymphocyte ( A , D , G ) or CD3 gated cell population ( B , C ) . Plots represent ( A ) CD3+ population , ( B ) CD4 population , ( C ) CD8 population , ( D ) NK population , and ( G ) B cell population . Percentage of memory markers on ( E ) CD3+CD4 and ( F ) CD3+CD8 positive cells was determined by expression of CD44 and CD62L: Naïve CD62L + CD44-; Central memory CD62L + CD44+; and Effector memory CD62L-CD44+ . The data are compiled percentage averages of 10–14 mice per group analyzed in two independent experiments . No statistical differences were observed for any adaptive immune populations by Students T-test p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02510 . 7554/eLife . 06508 . 026Figure 7—figure supplement 2 . Weight loss curves after MRSA epicutaneous infection . Mice were shaved , tape-stripped , weighed for preinfection baseline , and infected with 109 S . aureus . Blinded individuals weighed the animals at indicated time points . = Perforin-2 wild-type animals , = Perforin-2 heterozygous animals , • = Perforin-2 knockout animals . ( A ) Weight loss in 25 mice after infection throughout the first six days post-infection . ( B ) Overall weight loss of 25 animals infected with 109 S . aureus , after 30% weight loss animals were euthanized , causing the apparent rebound in weight of surviving animals at day 10 . The above experiments portray a representative experiment of three repeats . Statistical analysis was preformed with one-way ANOVA with Tukey Post-hoc Multiple Comparisons . indicates p < 0 . 05 between both knockout:wild-type and knockout:heterozygous animals . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02610 . 7554/eLife . 06508 . 027Figure 7—figure supplement 3 . Epicutaneous MRSA infection Day 6 organ load . Mice were shaved , tape-stripped , and infected with 109 S . aureus . = Perforin-2 wild-type animals , = Perforin-2 heterozygous animals , • = Perforin-2 knockout animals . ( A–D ) Organs from seven mice were collected six days after MRSA infection , weighed , homogenized , serially diluted , plated , and enumerated . Samples were normalized in weight to one another . ( A ) Blood was collected from cardiac puncture , ( B ) Spleen , ( C ) Kidney , and ( D ) Skin—site of bacterial inoculation . All samples were analyzed by Kruskal–Wallis test with Dunn's post-hoc multiple comparisons test . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 027 We also investigated the in vitro control of MRSA by neutrophilic granulocytes because they contribute to abscess formation and , with some MRSA strains , are credited with the clearance of infection ( Rigby and DeLeo , 2012 ) . We found that MRSA replicated in Perforin-2 −/− neutrophils , but was killed in the presence of Perforin-2 . Heterozygous neutrophils had intermediate bactericidal activity towards MRSA , suggesting that Perforin-2 was a rate-limiting molecule to control intracellular MRSA after infection ( Figure 7G ) . These in vivo results indicate that Perforin-2 is required to limit the early systemic dissemination of MRSA ( Hahn et al . , 2009; Onunkwo et al . , 2010 ) and , ultimately , to clear epicutaneous MRSA infection . To investigate whether the in vivo protection by Perforin-2 was limited to a particular site of infection or gram-positive pathogens , we also infected Perforin-2 deficient animals with gram-negative S . typhimurium via the orogastric route using well established protocols ( Barthel et al . , 2003 ) . However , owing to the previously observed sensitivity of Perforin-2 deficient animals to MRSA , we decreased the infectious inoculum from the normal LD50 of 108 CFU–105 CFU . As expected for this low infectious inoculum , Perforin-2 wild-type animals only transiently lost weight ( <10% ) whereas Perforin-2 deficient mice progressively lost weight and acquired severe diarrhea . Perforin-2 heterozygous littermates had more severe initial weight loss than wild-type animals , but these animals were able to recover ( Figure 8—figure supplement 1 ) . Equal S . typhimurium inoculation and colonization was confirmed by CFU analysis of the feces 12 hr following S . typhimurium inoculation ( Figure 8—figure supplement 2 ) . Finally , 129X1/SVJ congenic animals were infected with 105 S . typhimurium to address possibly confounding passenger mutations resulting from the different genetic background . As with the MRSA experiments , the observed Perforin-2 phenotype did not result from genetic differences because the aggregated survival curves were similar ( compare Figure 8B with Figure 8A ) . 10 . 7554/eLife . 06508 . 028Figure 8 . Perforin-2 is required for in vivo survival after orogastric S . typhimurium challenge . ( A ) Aggregated survival curves of 70 C57BL/6 × 129 × 1/SJV mice challenged with 105 S . typhimurium . ( B ) Aggregated survival curves of 45 , 129X1/SVJ mice challenged with 105 S . typhimurium . ( C–F ) Organ load five days after 105 S . typhimurium infection in C57BL/6 × 129 × 1/SJV mice in ( C ) blood , ( D ) small intestine , ( E ) liver , and ( F ) spleen . = MPEG1 ( Perforin-2 ) wild-type animals ( +/+ ) , = MPEG1 ( Perforin-2 ) heterozygous animals ( +/− ) , •= MPEG1 ( Perforin-2 ) knockout animals ( −/− ) . Log-rank ( Mantel–Cox ) test was performed for A and B with statistical significance p < 0 . 0001 . One-way ANOVA with Tukey post-hoc multiple comparisons was performed in C-F . p < 0 . 05 as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02810 . 7554/eLife . 06508 . 029Figure 8—figure supplement 1 . Weight loss curves after in vivo orogastric S . typhimurium challenge . Representative weight loss curves of 15 mice in each group challenged with 105 S . typhimurium . = MPEG1 ( Perforin-2 ) wild-type ( +/+ ) animals , = MPEG1 ( Perforin-2 ) heterozygous ( +/− ) animals , •= MPEG1 ( Perforin-2 ) knockout ( −/− ) animals . One-way ANOVA with Tukey post-hoc multiple comparisons was performed with p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type mice p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type and Perforin-2 knockout:Perforin-2 heterozygous mice . p < 0 . 05 between Perforin-2 knockout:Perforin-2 wild-type , Perforin-2 knockout:Perforin-2 heterozygous , and Perforin-2 heterozygous:Perforin-2 wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 02910 . 7554/eLife . 06508 . 030Figure 8—figure supplement 2 . Equal Colonization in feces 12 hr following 105 S . typhimurium inoculation . Feces was collected , homogenized , serially diluted , plated , and enumerated from 30 animals of each genotype 12 hr after S . typhimurium oral-gastric inoculation . No statistical differences were observed in fecal shedding between groups at this time point . Statistical analysis was performed with one-way ANOVA with Tukey post-hoc multiple comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 03010 . 7554/eLife . 06508 . 031Figure 8—figure supplement 3 . Organ load 60 hr after 105 S . typhimurium oral-gastric infection . ( A–D ) Organs from 10 mice were collected 60 hr after S . typhimurium infection , weighed , homogenized , serially diluted , plated , and enumerated . Samples were normalized in weight to one another . ( A ) Blood was collected from cardiac puncture , ( B ) Small Intestine , ( C ) Liver , and ( D ) Spleen . All samples analyzed by Kruskal–Wallis test with Dunn's post-hoc multiple comparisons test . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06508 . 031 Colonization of the animals with S . typhimurium was performed 60 hr following their infection ( Figure 8—figure supplement 3 ) . At this time point all three Perforin-2 groups showed signs of systemic S . typhimurium infection; however , with CFUs in Perforin-2 −/− mice were significantly higher than in wild-type animals . Moreover , only the former animals were septic with high bacterial titers in the blood . 5 days following infection , the bacterial burden of S . typhimurium in Perforin-2 −/− mice continued to be significantly higher in all organs as compared to their MPEG1 ( Perforin-2 ) +/+ littermates ( Figure 8C–F ) with intermediate levels observed in the heterozygous littermates . These in vivo results indicate that Perforin-2 is also required to clear S . typhimurium , and other compensatory bactericidal mechanisms are unable to contain orogastric S . typhimurium contributing to the rapid systemic spread of the pathogen . All results together underscore the importance of Perforin-2 in the defeat of bacterial pathogens with individual cells cumulatively protecting entire organisms . Much of our current understanding of antibacterial defenses is derived from studies of professional phagocytes , in particular their ability to migrate to sites of infection where they deploy antimicrobial effector molecules . Keratinocytes in the skin and the mucosal epithelia in the intestinal tract are known to provide local immune defense at barrier sites through the deployment of antimicrobial compounds and peptides to thwart pathogens while maintaining complex commensal communities . These effectors also , minimize chronic immune activation and inflammation ( Linden et al . , 2008; Turner , 2009 ) . In this study we have characterized an additional effector , Perforin-2 , present in professional phagocytes and barrier forming cells such as keratinocytes and mucosal epithelial cells . In the absence of Perforin-2 , the ability of these cells to destroy bacteria is severely compromised ( Figure 2 , Table 1 , Table 2 ) . These findings suggest that Perforin-2 may act upstream of other antimicrobial effectors or that their potency is enhanced by Perforin-2 . Our study demonstrates that parenchymal , tissue-forming cells can also express Perforin-2 and its expression in these cells is essential for killing of bacteria . Our EM analysis of isolated bacteria suggests that this killing is achieved by polymerization and pore-formation of Perforin-2 . In the absence of Perforin-2 , the other innate defense effectors were unable to prevent the replication and systemic dissemination of bacterial pathogens . As per above , this suggests that Perforin-2 may facilitate or augment the activity of other antimicrobial effectors . The unique structure of Perforin-2 as a membrane protein with a luminal and extracellular MACPF-killer domain is well suited for locally killing membrane-entrapped , extracellular or intracellular bacteria . Unlike Perforin-2 , the other immunologic MACPF proteins do not act within the cell producing the effector protein . Perforin-1 is stored in granules in NK and CTL that are exocytosed into the immunologic synapse , thereby exposing Perforin-1 to extracellular calcium ions that trigger Perforin-1 polymerization in target membranes . The components of the MAC of complement are secreted by the liver into blood to exert their antibacterial effects after binding of C3b to the bacterial surface which initiates the assembly of the pore-forming C5b-8-poly-C9 complex . Seemingly every cell can utilize Perforin-2 to defend itself against bacteria—potentially even before professional phagocytes are recruited–according to the constitutive expression of Perforin-2 in kc . Owing to this innate intrinsic protection of kc , bacteria may be cleared by these barrier cells even before the influx and assistance of professional phagocytes . The lack of barrier protection by kc when Perforin-2 is deficient may explain why Perforin-2 knockout mice rapidly ( within days; Figure 7—figure supplement 2 ) exhibit symptoms to epicutaneous infection with MRSA . Thereafter the combined absence of Perforin-2 in the barrier tissue and in professional phagocytes may overwhelm the host by enhancing the systemic dissemination of the bacteria ( Figure 7—figure supplement 3 ) and , ultimately , leading to multiple organ failure . The speed and severity of infections in Perforin-2 deficient animals , including sepsis and death following traditional non-lethal inocula of bacteria ( Figures 7 , 8 ) , may also be assisted by the seemingly ubiquitous expression and function of Perforin-2 , either inducible or constitutive . We suggest that this ubiquitous mechanism provides heretofore unrealized defenses at all potential sites of infection including barrier cells , phagocytes , parenchyma , and possibly even stroma . Further studies are required to determine the relative contribution of Perforin-2 expressed in each of them . The scenarios above are further complicated by differences in the regulation of Perforin-2 in different cells . It is constitutively expressed in macrophages ( and at least in certain barrier cells like kc ) but requires induction in parenchymal cells . Therefore we speculate that unactivated parenchymal cells offer invasive bacteria a window of opportunity for replication and further dissemination . A second issue is that some pathogens have evolved mechanisms to suppress Perforin-2 ( see accompanying report by McCormack et al . ) . On the other hand , our data also shows that when Perforin-2 is expressed at optimal levels , highly pathogenic and antibiotic-resistant bacteria can be killed by Perforin-2 . This leads us to speculate that supraphysiologic levels of Perforin-2 expression could be an approach to defeat antibiotic-resistant bacteria . Vice versa , serious clinically relevant infections including antibiotic resistant strains of MRSA ( Figure 7 ) , S . typhimurium ( Figure 8 ) , and M . avium as well as M . tuberculosis ( Figure 1 ) are all susceptible to Perforin-2 , as demonstrated by greater bacterial pathogenesis in the absence of Perforin-2 . We undertook several approaches to determine whether Perforin-2 is a pore-forming protein and all approaches supported this concept . First , we isolated eukaryotic membranes from Perforin-2 overexpressing cells following limited trypsin digestion ( to activate poly-Perforin-2 pore formation presumably by cleaving the cytoplasmic domain at a predicted trypsin-sensitive site ) and observed 100 Å pores . The caveats of this approach entail its unphysiological setting especially the use of eukaryotic cell membranes that seem unlikely to compromise Perforin-2′s natural target . However , the visualization of pores on bacterial surfaces ( Figure 5 ) was dependent on the availability of Perforin-2 and the bacterial membranes contained Perforin-2-derived fragments ( Figure 6 ) that were only discernable following infection with Perforin-2 expressing cells . These results strongly imply that the antibacterial properties of Perforin-2 involve polymerization and pore-formation in bacteria . The detection of cleavage proteins suggests that Perforin-2 also could be cleaved physiologically at some point during the bacterial infection; however , the order of cleavage and pore-formation as well as the role of additional proteins requires further investigation . Membrane perforation by the MAC of complement and by poly-Perforin-1 of cytolytic lymphocytes provides a conduit for additional effector molecules to ensure complete target destruction . MAC-polyC9 pores act as entry point for serum lysozyme to digest peptidoglycan resulting in bacterial collapse and lysis ( Schreiber et al . , 1979 ) . Poly-Perforin-1 pores provide access to multiple granzymes that promote apoptosis of virally infected cells ( Dennert and Podack , 1983; Podack and Konigsberg , 1984; Masson and Tschopp , 1987; Jenne and Tschopp , 1988; Trapani et al . , 1988; Shiver et al . , 1992; Smyth et al . , 1994 ) . Likewise , our studies suggest that poly-Perforin-2 pores enhance the delivery and action of endogenously produced ROS and RNS . This is consistent with our previous finding that exogenous lysozyme lysed M . smegmatis and MRSA when they were isolated from Perforin-2 expressing fibroblasts but failed to lyse these pathogens when they were isolated from Perforin-2 deficient fibroblasts ( McCormack et al . , 2013 ) . Although high levels of lysozyme , ROS , or RNS can kill pathogens in the absence of Perforin-2 in vitro , we suggest that physiological endogenous levels of these–and potentially other antimicrobial effectors–may be limiting , requiring assistance by poly-Perforin-2 pore-formation to achieve bacterial killing ( see Figure 3 and its supplements ) . As pathogenic microbes are specialized in their invasive strategies so are the innate strategies of the immune system in the use of the three pore forming proteins . First , interstitial extracellular bacterial pathogens are killed by MAC-polyC9 of the complement system whose components are present in the serum . Second , the viral production factories are destroyed by poly-Perforin-1 delivered by NK cells and CTL . Third , invasive or cell membrane attached bacterial pathogens are eliminated by poly-Perforin-2 expressed in most , if not all , cells . Common to all three pore-formers is that their antimicrobial activity cooperates with additional specialized effectors . Our studies add Perforin-2 as the third MACPF-domain driven pore-forming killer protein to the arsenal of the mammalian immune system for protection against microbial invasion . Given the central role of Perforin-2 in antimicrobial responses , the pathogenicity of bacteria may depend on their ability to subvert or evade Perforin-2′s pore-forming ability . In the accompanying manuscript we describe one such virulence factor ( Cif ) that protects bacterial pathogens by blocking translocation of Perforin-2 ( McCormack et al . , 2015 ) . Determining the mechanism of how other pathogenic bacteria inhibit Perforin-2 will elucidate additional critical steps in Perforin-2 activation and may allow the development of new treatments targeting both antibiotic-sensitive and antibiotic-resistant bacterial pathogens . RAW264 . 7 ( TIB-71 ) , J774A . 1 ( TIB-67 ) , HL-60 ( CCL-240 ) , HeLa 229 ( CCL 21 ) , CATH . a ( CRL-11179 ) , Neuro-2A ( CCL-131 ) , NIH/3T3 ( CRL-1658 ) , Balb/c 3T3 ( CCL-163 ) , C2C12 ( CRL-1772 ) , CMT-93 ( CCL-223 ) , CT26 . WT ( CRL-2638 ) , B16-F10 ( CRL-6475 ) , B16-F0 ( CRL-6322 ) , LL/2 ( CRL-1642 ) , MIA PaCa-2 ( CRL-1420 ) , Thp1 ( TIB-202 ) , NK-92 ( CRL-2407 ) , OVCAR3 ( HTB-161 ) , CaCo-2 ( HTB-37 ) , A549 ( CCL-185 ) , U-1752 , JEG-3 ( HTB-36 ) , and HEK-293 ( CRL-1573 ) cell lines were obtained from American Type Culture Collection , Manassas , VA . HUVECs were a gift from Dr . W Balkan , University of Miami , FL . BV2 microglial cell line was a gift from Dr . J Bethea , University of Miami , FL . ED-1 mouse lung adenocarcinoma cell line derived from the lung tumors of transgenic mice that express cyclin-E driven by the human Surfactant-C promoter was a gift from Dr . D Robbins , University of Miami , FL . MOVCAR 5009 and MOVCAR 5447 cells were a gift from Dr . D Connolly , Fox Chase Cancer Center , PA . A2EN primary-cell-like cervical epithelial cells were provided by Dr . K Fields , University of Kentucky , KY . UM-UC-3 and UM-UC-9 bladder cancer cell lines were a gift from Dr . B Grossman , MD Anderson Cancer Center , TX . Human adult and neonatal keratinocytes were obtained from Lonza . Primary murine astrocytes and primary CNS fibroblasts were a gift from Dr . J Lee , University of Miami , FL . Neonatal ventricular myocytes were a gift from Dr . N Bishopric , University of Miami , FL . All cells were cultured at 37°C in a humidified atmosphere containing 5% CO2 following ATCC recommendations for culture conditions . HL-60 were differentiated to neutrophils using retinoic acid or to macrophages using PMA as previously described ( Meyer and Kleinschnitz , 1990; Daigneault et al . , 2010 ) . Murine primary macrophages were obtained from thioglycolate-elicited peritoneal or differentiated from bone marrow utilizing M-CSF as previously described ( Zhang et al . , 2008 ) . Murine bone marrow derived dendritic cells were differentiated from bone marrow utilizing GM-CSF ( Dearman et al . , 2009 ) . Human monocyte derived macrophages and human monocyte derived dendritic cells were differentiated from monocytes as described previously ( Vijayan , 2012 ) and human PMN isolated as previously described ( Oh et al . , 2008 ) . Primary human NK cells were isolated utilizing RosetteSep Human NK cell Enrichment cocktail from Stemcell Technologies . Murine embryonic fibroblasts ( MEFs ) and murine PMN were isolated as previously described ( Luo and Dorf , 2001; Scheuner et al . , 2001 ) . S . typhimurium LT2 ( ATCC 700720 ) and SL1344 ( gift from Dr . J . Galán , Yale University ) , and E . coli K12 DH5α were grown in Luria–Bertani broth ( LB ) or heart infusion broth ( HIB; Becton , Dickinson and Co . , Sparks , MD , United States ) at 37°C . Staphylococcus aureus CLP148 and CLP153 ( MRSA PFGE type USA300 ) were grown in LB or tryptic soy broth ( Sigma–Aldrich , St . Loius , MO , United States ) at 37°C . M . avium ( gift from Dr . T . Cleary , University of Miami ) , and M . smegmatis ( ATCC 700084 ) were grown in Middlebrook 7H9 broth . S . typhimurium LT2 carrying deletions in hmpA ( codons 35 to 361 ) and sodC1 ( codons 16 to 142 ) were generated via lambda Red-mediated recombination ( Datsenko and Wanner , 2000 ) with modifications as described by Bartra et al ( Bartra et al . , 2008 ) , and cultured in the same manner as wild-type S . typhimurium . For the generation of Perforin-2 knockout mice the targeting vector was linearized and electroporated into RW-4 ES cells originating from the 129X1/SvJ strain , followed by selection in G418 . Targeted clones were screened by PCR . From 90 clones , 2 positive clones were selected that had undergone homologous recombination and were identified through Southern blot analysis . One ES clone was utilized for the generation of chimeric mice by injection using C57Bl/6J blastocysts as the host . The resulting female chimeras were further mated with C57Bl/6J male mice for germ line transmission . The heterozygous mice ( F1 mice ) were interbred to obtain wild-type , heterozygous , and homozygous littermates ( F2 ) . C57Bl/6 × 129 × 1/SvJ animals utilized in these experiments were backcrossed 8–10 times for these experiments . Mouse genotype was determined by PCR using PCR probes MP10 and MP11 . To generate 129X1/SvJ inbred animals without potential ES cell passenger mutations , chimeric mice were mated with 129X1/SvJ animals , to assess for germ line transmission . The heterozygous mice were then interbred to obtain a genetically pure 129X1/SvJ strain . Mouse genotype was determined by PCR utilizing PCR probes MP10 and MP11 . Animals were bred at the University of Miami , Miller School of Medicine Transgenic Core Facility . Mice were allowed to freely access food and water and were housed at an ambient temperature of 23°C on a 12 hr light/dark cycle under specific pathogen-free condition . Animal care and handling were performed as per IACUC guidelines . All animal experiments were performed in accordance with University of Miami Animal Care and Use Committee guidelines . Animal's genotype was blinded prior to the experiment to limit bias . Our methodology was adopted from references ( Cheng et al . , 2009; Wanke et al . , 2013 ) . In brief , all mice were shaved and tape-stripped ( 7 applications ) with Transpore tape ( 3M , Minneapolis , MN , United States ) . This level of tape stripping did not create a wound , but was sufficient to disrupt the epidermal barrier . An inoculum of 109 MRSA strain CLP 153 in 0 . 02 ml of phosphate-buffered saline ( PBS ) or PBS control was added to ∼1 cm2 of skin and the area bandaged with plastic sheet and overwrapped with dressings of Transpore tape and Nexcare waterproof tape ( 3M ) for 6 hr , at which time the bandage was removed . Mice were weighed daily throughout the experiment; animals were euthanized after greater than 30% weight loss . For CFU enumeration , mice were sacrificed either 6 or 12 days after infection , cardiac puncture was performed and organs were harvested , weighed , and homogenized using a potter homogenizer in ddH2O with 0 . 05% Triton X-100 . The homogenates were diluted and plated on TSA II plates ( kanamycin and oxacillin selection ) . All samples were normalized based on weight . All animal experiments were performed in accordance with University of Miami Animal Care and Use Committee guidelines . Animal's genotype was blinded prior to the experiment to limit bias . Mice were infected orogastrically with 105 colony-forming units of Streptomycin resistant S . typhimurium ( SL1344 ) 24 hr after orogastric Streptomycin pretreatment as previously described ( Barthel et al . , 2003 ) . Mice were weighed daily throughout the experiment; the animals were euthanized after greater than 30% weight loss . For CFU enumeration , mice were sacrificed 3 days after 105 S . typhimurium orogastric infection , cardiac puncture was performed and organs were harvested , weighed , and homogenized using a potter homogenizer in ddH2O with 0 . 05% Triton X-100 . The homogenates were diluted and plated on MacConkey agar plates ( streptomycin at 100 μg/ml ) . All samples were normalized based on weight . Mouse bone marrow-derived macrophages were cultured in DMEM containing 15% L929-conditioned medium , 10% fetal bovine serum , and 2 mM glutamate for 7 days . Macrophages were plated in 96 well plates and infected at an MOI of 3:1 with M . tuberculosis CDC 1551 expressing the fluorescent protein mCherry under constitutive promoter ( smyc'::mCherry ) . Bacterial survival and growth was monitored by measuring mCherry fluorescence using a Perkin–Elmer EnVision plate reader . This assay has been previously validated by comparison with colony forming unit ( CFU ) counts ( Lee et al . , 2013 ) . Anti-murine Mpeg1 ( ab25146 ) , anti-human Mpeg1 ( ab176974 ) , anti-calreticulin ( ab22683 ) , anti-GM130 ( ab52649 ) , anti-clathrin ( ab2731 ) , anti-Methicillin Resistant Staphylococcus Aureus ( ab73263 ) , anti-Ada ( ab18104 ) , anti-pan Cadherin ( ab140338 ) , and anti-catalase ( ab16731 ) ( Abcam , Cambridge , MA ) ; anti-GFP ( sc9996 ) , and anti-EEA1 ( sc-6415 ) ( Santa Cruz Biotechnology , Dallas , TX ) ; anti-Prohibitin ( Poly6031 ) , anti-Lamp1 ( 1D4B ) , and anti-β-actin ( 2F1-1 ) ( Biolegend , San Diego , CA , United States ) ; anti-Golgin-97 ( Thermo Fisher Scientific , Waltham , MA , United States ) ; MEK1/2 ( D1A5 ) ( Cell Signaling Technology , Danvers , MA , United States ) ; peptide synthesized cytoplasmic Perforin-2 antibody ( 21st Century Biochemicals , Marlborough , MA , United States ) ; and peptide synthesized human Perforin-2 antibodies against each domain ( Abmart , Berkeley Heights , NJ , United States ) were utilized for immunoblots as indicated . Densitometry analysis was performed where indicated utilizing ImageJ software . Antibodies directed against murine CD3 ( clone 145-2C11 , Biolegend ) , CD4 ( clone RM4-5 , Biolegend ) , CD8 ( clone 53-6 . 7 , Biolegend ) , NK1 . 1 ( clone PK136 , Biolegend ) , CD62L ( clone MEL-14 , Biolegend ) , CD44 ( clone IM7 , Biolegend ) , and CD19 ( clone 6D5 , Biolegend ) were used in multicolor FACS analysis . Samples were washed and resuspended in cold flow cytometry staining buffer ( 1% BSA/PBS ) ; stained for 30 min in the dark before a final wash and acquisition . All samples were acquired on a BD Fortessa Flow cytometer running FACS DIVA software . Analysis was performed using FlowJo X software ( TreeStar; OR , United States ) . Subcellular fractionation was performed on human THP-1 cell line induced with PMA for 16 hr and then allowed to rest without PMA stimulation for 48 hr . Samples were isolated following Axis-Shield Density Gradient methods ( Axis-Shield , Norway ) for exocytosis analysis: resolution of plasma membrane from TGN/endosomes and cytosolic proteins in self-generated gradient ( S45 ) . In brief , cells were harvested and homogenized utilizing a Dounce homogenizer . The homogenate was then centrifuged to pellet nuclei and other cell debris . The post-nuclear supernatant was then loaded equally on decreasing concentrations of iodixanol ( 30% , 20% , 10% ) and centrifuged at 353 , 000g for 3 hr . The gradient was then collected in 0 . 1 ml fractions by aspiration from the meniscus . For better resolution of clathrin-coated vesicles , endosomes , and lysosomes protocol S43 was utilized . In brief , cells were homogenized utilizing a Dounce homogenizer , and centrifuged to remove nuclei and other cell debris . The supernatant was then mixed with 12 . 5% iodixanol underlaid with 20% iodixanol . The mixed and layered tube was then centrifuged at 350 , 000g for 1 . 5 hr . The gradient was collected in 0 . 1 ml fractions by aspiration from the meniscus . For better resolution of peroxisomes , protocol S13 was utilized—Purification of mammalian peroxisomes in a self-generated gradient . After collection , fractions were analyzed for protein content by DC Protein Assay ( Bio-Rad , Hercules , CA , United States ) , and screened for localization of subcellular fractions by Western blot analysis . The intracellular gentamicin protection assays were conducted as previously described ( Lutwyche et al . , 1998; Laroux et al . , 2005; McCormack et al . , 2013 ) . Briefly , 100 ng/ml of species specific human or murine IFN-γ was added 14 hr before infection where necessary to uniformly induce Perforin-2 expression . In all graphs bacteria were added as indicated , and after 30–60 min to allow for uptake/invasion , the extracellular bacteria were washed and re-plated in gentamicin supplemented medium . For gentamicin protection assays , the multiplicity of infection was between 20 and 50 bacteria per cell to allow for sufficient uptake of bacteria . For gentamicin-free intracellular assessment of bacterial load , the gentamicin protection assay was modified as follows: achieve >90% eukaryotic cell confluence on infection , decrease multiplicity of infection from between 20–50 to between 0 . 5–5 , and trypsinize eukaryotic cells after initial wash steps to help eliminate attached extracellular bacteria . Every 4 hr , the medium was removed and checked for extracellular bacterial growth , washed twice with PBS , and replaced with fresh medium . The gentamicin-free infection assay described above was modified as follows to increase extracellular bacterial recovery: utilization of pathogens that do not facilitate uptake ( enteropathogenic E . coli and MRSA ) , increased inoculation to a MOI from 0 . 5–5 to 10–30 , and decreased invasion/attachment time from 60 min to 40 min . After the attachment/early infection phase was complete , cells were washed 5x with prewarmed media to remove non-adherent bacteria , and infection was allowed to continue for an additional 30 min . After this was complete , cells were washed an additional 3x with PBS and trypsinized to remove both cells as well as extracellular bacteria . Enrichment of extracellular bacteria was performed by successive low-speed differential centrifugations to remove intact mammalian cells from extracellular bacteria . Cell/bacteria from above were spun at 200g for 15′ with collection of the supernatant , the supernatant was collected and respun for a total for 7 spins . Non-infected mammalian cells were also treated in the same fashion to quantify for mammalian cellular contamination . Following the last spin from above , the supernatant was spun at 20 , 000g for 15′ to pellet all bacteria and cellular debris . Goat IgG was added to the bacterial enriched pellet following resuspension and part of the post-infection bacterial/goat IgG solution was then filtered through prewashed 0 . 22 μm filters . After filtration , the eluent was collected and fractions were mixed and boiled with reducing laemmli sample buffer and loaded on SDS-PAGE for Immunoblotting . For murine cells , RNA interference and transfection were conducted as previously described ( McCormack et al . , 2013 ) . For human cells , the aforementioned murine system was modified through utilizing three human Perforin-2-specific silencer select siRNAs purchased from Thermo Fisher Scientific Silencer Select #s61053 , s47810 , s61054 . Silencer select negative control #1 and 2 from ) Thermo Fisher Scientific were also utilized as a negative control . Murine or human RNA extraction , cDNA synthesis and analysis was performed as previously described ( Fields et al . , 2013; McCormack et al . , 2013 ) . Message for the housekeeping gene GAPDH was amplified as an internal normalization control . All assays were performed in technical triplicate for each RNA sample . For nitrite detection , adherent PEM were stimulated overnight with IFN-γ ( 100 ng/ml ) and stimulated with LPS ( 100 ng/ml ) in the presence of N-Acetyl Cysteine ( NAC ) ( Sigma-Aldrich , St . Louis , MO ) or NG-nitro-L-arginine methyl ester ( L-NAME ) ( Sigma-Aldrich ) for 48 hr . 50 μl of supernatant was collected for analysis of NO2- production using Griess reagents . 50 μl of 1% sulfanilamide in 3% H3PO4 was added to 50 μl of supernatant followed by 50 μl of 0 . 1% napthylethylene dihydrochloride in 3% H3PO4 and the wells were read on a spectrophotometer at 550 nM . Sodium nitrite was used a standard at concentrations ranging from 1 μM to 125 μM . For ROS detection , adherent PEM were stimulated overnight with IFN-γ ( 100 ng/ml ) and then labeled with 10 μM CM-H2DCFDA ( Thermo Fisher Scientific ) in PBS for 30 min at 37°C , followed by washing and addition of complete media . Inhibitors were added 30 min prior to addition of LPS , PMA ( 1 μM ) , or H2O2 ( 200 μM ) . 30 minutes later , cells were scraped and immediately analyzed by flow cytometry . For live cell imaging , RAW264 . 7 or BV-2 cells were nucleofected with Perforin-2-GFP and stimulated overnight with LPS ( 1 ng/ml ) and IFN-γ ( 100U/ml ) in glass bottom dishes with No . 1 . 5 coverglass ( MatTek Corp , Ashland , MA , United States ) . Cells were washed once with PBS and organelles were labeled . For endoplasmic reticulum ( ER ) labeling , ER-Tracker™ Blue-White DPX ( Thermo Fisher Scientific ) was used at a working concentration of 1 μM for 30 min at 37°C . For all other stains , transfected cells were fixed with 3% paraformaldehyde ( PFA ) for 15 min at room temperature , permeabilized with 0 . 5% saponin , blocked with 10% normal goat serum and incubated with primary and secondary antibodies . Anti-CD107a ( LAMP-1 ) ( BD Pharmingen , San Jose , CA , United States ) , anti-EEA1 ( EMD Millipore ) , and anti-GM130 ( BD Biosciences ) were used to identify cellular organelles . Secondary antibodies were all raised in goats . Images were taken on a Leica SP5 inverted confocal microscope with a motorized stage and analyzed using Leica application suite advanced fluorescence software . Students t-test , multiple t-test with Holm-Sidak multiple comparisons correction , one-way ANOVA with Bonferroni multiple comparisons test , or Kruskal–Wallis non-parametric test with Dunn's multiple comparison test was used for comparisons ( GraphPad Prism Version 6 . 0b and SPSS 21 . 0 were utilized for statistical analysis ) .	An effective defense against foreign invaders is fundamental to an organism's survival . It is likely that immunity began to develop shortly after the emergence of Earth's first single-celled organisms and a remnant of that distant past still exists in our present day immune system in the form of Perforin-2 . This ancient protein has been highly conserved throughout evolution from sea sponges to humans . Some studies have suggested that Perforin-2 may have an antimicrobial role in invertebrates ( including clams , mussels , and snails ) and fish . However , its mechanism of killing and its role in the mammalian immune systems has remained largely unknown . McCormack et al . now report that Perforin-2 is a crucial component of host defense against a wide spectrum of infectious bacteria in both mice and humans . This was shown when mice lacking Perforin-2 died from bacterial infections that are not normally lethal . Somewhat unexpectedly , other bactericidal molecules were also found to be less effective in the absence of Perforin-2 . This indicates that Perforin-2 is required for the activity of multiple aspects of the mammalian immune system . McCormack et al . demonstrated that Perforin-2 kills by punching holes in bacteria . Unlike other pore-forming proteins that are only present in specific cells , all mammalian cells can express Perforin-2 . McCormack et al . also showed that when Perforin-2 is produced at optimal levels , cells are able to combat otherwise lethal , drug-resistant bacteria , including methicillin resistant Staphylococcus aureus ( MRSA ) . This means that Perforin-2 provides a rapid self-defense mechanism for cells against bacterial invaders . The protein's dual role as a pore-forming protein and a supporter of other antibacterial molecules is unprecedented . In the future , these findings could inform the development of treatments that activate and optimize Perforin-2 production to target and eradicate bacterial infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2015	Perforin-2 is essential for intracellular defense of parenchymal cells and phagocytes against pathogenic bacteria
The functional communication of neurons in cortical networks underlies higher cognitive processes . Yet , little is known about the organization of the single neuron network or its relationship to the synchronization processes that are essential for its formation . Here , we show that the functional single neuron network of three fronto-parietal areas during active behavior of macaque monkeys is highly complex . The network was closely connected ( small-world ) and consisted of functional modules spanning these areas . Surprisingly , the importance of different neurons to the network was highly heterogeneous with a small number of neurons contributing strongly to the network function ( hubs ) , which were in turn strongly inter-connected ( rich-club ) . Examination of the network synchronization revealed that the identified rich-club consisted of neurons that were synchronized in the beta or low frequency range , whereas other neurons were mostly non-oscillatory synchronized . Therefore , oscillatory synchrony may be a central communication mechanism for highly organized functional spiking networks . Perception , cognition , and movement are generated by the functional interaction of neuronal circuits . In order to understand the basis of these processes , especially in highly complex networks such as the primate brain , it is essential to know their network structure , termed topology . Graph theoretical approaches have enabled analysis of the brain’s network topology ( Watts and Strogatz , 1998; Bullmore and Sporns , 2009 ) . Using such approaches in EEG , MEG , DTI or fMRI studies , anatomical regions have been grouped into functional and anatomically strongly connected modules , which are segregated from each other ( Bullmore and Sporns , 2009 ) . Still , every region can be reached by bypassing a few others ( small-world ) , a topology which is robust and allows efficient information processing ( Hilgetag et al . , 2000; Stephan et al . , 2000; Bullmore and Sporns , 2009 ) . A few regions of the brain are highly connected and centrally located within the network ( van den Heuvel and Sporns , 2013a ) ( hubs ) as well as strongly connected to each other ( van den Heuvel et al . , 2012 ) ( rich-club ) . This rich-club forms a global communication pathway across the network , thereby cross-linking segregated modules ( van den Heuvel and Sporns , 2013b ) . However , single neurons and their functional network topology are the fundamental computational structure of the primate brain . While neuronal modules , hubs , and rich-club organization have been shown in organotypic slices of rats ( Bonifazi et al . , 2009; Shimono and Beggs , 2014; Schroeter et al . , 2015 ) , hardly anything is known about single neuron network topology in the intact brain during behavior . Limitations in recording high number of single neurons in parallel , incorporating distance-dependent connectivity , and addressing subsampling and firing rate biases makes it difficult to assess these networks . Only small-world topology has been debated ( Yu et al . , 2008; Gerhard et al . , 2011 ) and rich-club topology has been shown recently in mice ( Nigam et al . , 2016 ) . Equally important to topology is the mechanism which coordinates and synchronizes neurons during cognitive or perceptual processes . Previous research has revealed oscillatory synchrony in time as a crucial feature of functional coordination ( Fries , 2009; Buzsáki and Wang , 2012; Womelsdorf et al . , 2014 ) . Different distinct frequency bands for information transmission and functional network coordination have been identified , such as gamma ( 40–100 Hz ) and theta ( 4–8 Hz ) in the visual areas and up to frontal cortex for coordinated attention selection ( Roelfsema et al . , 1997; Bosman et al . , 2012; Gregoriou et al . , 2012 ) , and beta ( 18–35 Hz ) and delta ( 1–4 Hz ) in fronto-parietal regions for network coordination during decision and working memory processes ( Brovelli et al . , 2004; Pesaran et al . , 2008; Haegens et al . , 2011; Salazar et al . , 2012; Nácher et al . , 2013 ) . Recently , gamma and theta oscillations have been proposed as feedforward communication frequencies across large parts of the visual network , while beta oscillations have been proposed for feedback communication ( Bastos et al . , 2015 ) . However , firing rate correlations have also been found , independent of oscillatory synchronization , to be of importance for communication in the behaving brain ( Fujisawa et al . , 2008; Smith and Kohn , 2008 ) . Yet , how functional network topology , described by graph theoretical approaches , relates to oscillatory and non-oscillatory synchronization remains unclear . This question must be answered at the level of single neurons , where oscillatory synchrony can be distinguished from non-oscillatory synchrony . Here , we recorded in parallel and assessed functional connectivity and network topology from a large number of single neurons ( 48 to 149 per session ) from the primate grasping circuit ( Luppino et al . , 1999 ) , including the ventral premotor ( F5 ) , primary motor ( M1 ) , and anterior intraparietal ( AIP ) cortex of three behaving macaque monkeys . Across the three cortical areas we found modular , small-world topology with a clear presence of hubs that were organized as a rich-club . Moreover , rich-club hub neurons predominantly spiked and communicated by oscillatory synchrony in the beta and low frequency range , while the remainder of the network predominately communicated by non-oscillatory synchrony , suggesting that oscillatory synchrony is a central coordination mechanism for functional network topology . The functional connectivity between all simultaneously recorded units of the grasping network was estimated by calculating cross-correlation histograms ( CCHs ) ( Figure 2A , Figure 2—figure supplements 1 , 2; see Materials and methods ) , one of the few methods also allowing analyses of the frequency domain ( Bastos and Schoffelen , 2016 ) ( see below ) . It is important to stress that the functional connections we describe here do not necessarily represent monosynaptic connections , but merely the influence of one unit onto another . For each neuron pairing one single CCH was estimated over all task epochs and grasp types , since we were interested in the general network interaction and not grasp type or time specific modulations of the network . A general problem of all connectivity measures is common drive to the network , such as stimulus- or movement-locked , but not pairwise , correlations , causing an overestimations of connections . We corrected these biases by subtracting surrogate CCHs ( Figure 2—figure supplement 1A ) . 10 . 7554/eLife . 15719 . 006Figure 2 . Cross- and auto-correlation histograms and frequency spectra . ( A ) Example crosscorrelation histograms ( CCHs ) for five example neuron pairs . Displayed amplitude is limited to ±2 . 5x10−3 coincidences per spike for better comparison . CCHs are color-coded based on their oscillatory synchronization frequency ( red: beta band; blue: low frequencies; magenta: beta and low frequencies; black: no underlying frequency ) . ( B ) Corresponding frequency spectra of CCHs in a , frequency displayed on logarithmic scale ( for better comparison limited to a power of 8x10−5 ) and color-coded as in A . ( C ) Same as in A , but for auto-correlation histograms ( ACHs ) . ( D ) Same as in B , but for the frequency spectra of the ACHs in C . ( E ) Illustration of different kinds of CCHs to a reference unit and the inferred connectivity . Upper left: No peak is present in the CCH so the unit is not connected to the reference unit . Upper right: A peak at positive time lags indicates a connection from the reference to the target unit . Lower right: A peak is present straddling the 0 time lag with a maximum peak at 0 , indicating a bidirectional connection . Lower left: Several peaks and troughs are present with a clear underlying frequency and a maximum peak at a negative time lag , indicating an oscillatory connection from the target to the reference unit . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 00610 . 7554/eLife . 15719 . 007Figure 2—figure supplement 1 . CCH processing and statistics , and all connections of an example unit oscillatory synchronized in the low frequency range . ( A ) Processing steps of three example CCHs . From left to right: illustration of the processing steps involving surrogate subtraction , smoothing , and cluster statistics to evaluate if a peak or trough in a CCHs was significant . From top to bottom: A CCH with one significant peak , a CCH with multiple significant peaks and troughs having an underlying frequency in the beta range , and a CCH with no significant peak or trough . ( B ) An examples of all CCHs ( small panels ) and the ACH of one unit with all other units of one dataset of a unit communicating and oscillating in the low frequency range . The ACH is boldly framed and displayed in red , significant connections are indicated by dark lines in CCHs and not significant connections as transparent lines . Directionality information , which is also derived from the CCHs , is not represented . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 00710 . 7554/eLife . 15719 . 008Figure 2—figure supplement 2 . All connections of two example units , one non-oscillatory synchronized and one oscillatory synchronized in the beta range . ( A ) Same as in Figure 2—figure supplement1B , but for a non-oscillatory synchronized unit . ( B ) Same as in Figure 2—figure supplement1B , but for a unit communicating and oscillating in the beta range . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 00810 . 7554/eLife . 15719 . 009Figure 2—figure supplement 3 . Detectability of directed functional connections using equal rate model simulations . ( A ) Transfer kernels of one modeled dataset . Gamma functions with different maxima and lengths were used as temporal transfer kernels . The area under the curve was always normalized to 0 . 02 . ( B ) Histogram of detectability of directed connections . Average number of correct rejections and hits are shown for 10 simulated simple networks ( SN ) and 10 simulated complex networks . Error bars show the standard error across simulated networks . ( C ) Same as in B , but for detectability of connections . Any directional information was ignored and it was just estimated if a connection between two units was detected or not . ( D ) Same as in B , but for detectability of directionality for detected connections . The percent of correct rejections and hits is only for the correctly detected connections as displayed in B , thus a pure evaluation of directionality detectability unbiased by connection detectability . ( E ) Average CCHs for bidirectional connections and common drive pairs of all 20 simulations . The data was pooled , since no considerable difference between the two types of simulations was found . All simulated pairs of both groups are included irrespective of whether they were detected as significant . Error bars show the standard error across CCHs . Note that even though the average peak is at the zero time lag , many pairs had peaks on either side of the zero time lag . ( F ) Maximum peak count of bidirectional and common drive pairs ( for each ms bin ) displayed in E . In case CCHs had two peaks or just showed noise fluctuations , only the time lag of the maximum value was considered in order to avoid preselection biases . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 00910 . 7554/eLife . 15719 . 010Figure 2—figure supplement 4 . Maximum peak or trough time and phase lag distributions . ( A ) Maximum peak or trough time lag distribution of all significant connections relative to the zero time lag . In case that more than one significant cluster was detected , only the cluster with the highest absolute value was considered . For bidirectional connections time lags were considered for both directions . Line shadings show standard error across datasets . ( B ) Maximum peak or trough phase relative to the zero time lag for all connections with significant underlying oscillation classified by a significant peak in their corresponding frequency spectra . Results are shown separately for beta at 20 Hz ( red ) and low frequency at 4 Hz ( blue ) oscillations . Note that 4pi ( two cycles ) corresponds to 100 ms for beta and to 500 ms for low frequency oscillations . Line shadings show standard error across datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 010 Connections indicated by significant peaks or troughs in CCHs were identified by a cluster-based surrogate test ( Maris et al . , 2007 ) to all CCHs ( see Materials and methods ) , testing against surrogate CCHs . To control the family-wise error for the entire network , false discovery rate ( FDR ) correction was applied across all significant connections ( Benjamini and Hochberg , 1995 ) . For later topological analyses of oscillatory synchrony in the network , we applied Fourier transformations ( Figure 2B–D; see Materials and methods ) to all CCHs and auto-correlation histograms ( ACHs ) . The latter detected periodicity in the spiking of individual units , ( Figure 2C ) , allowing classifying them as oscillators or non-oscillators . Directional interaction between pairs of units was inferred from the time delay of significant peaks or troughs in the CCHs ( Figure 2E ) . In early studies , a peak or trough in a CCH with a non-zero time lag was classified as a unidirectional connection from one neuron to another while a peak or trough with a zero time lag was classified as common drive to both neurons ( Moore et al . , 1970 ) . However , recent studies based on complex models rather suggest that zero time lag peaks or troughs in CCHs mainly represent bidirectional connections , which can be explained by the dynamical relaying mechanism , and only rarely reflect a common drive ( Vicente et al . , 2008; Gollo et al . , 2014 ) . For this reason , we defined zero time lag peaks and troughs in the CCHs as bidirectional connections . For additional validation of how well we could recover directed functional connectivity , we modeled two sets of 'ground truth' networks with the same distribution of firing rates as recorded single units , one simple network ( SN ) and one complex network ( CN ) set ( Equal rate model , see Materials and methods ) . We could detect directed functional connections reasonably well ( hits: 62% for SN , and 69% for CN ) and hardly detected any false connections ( correct rejections ( CR ) > 99% for SN and CN ) , independent of the underlying topology ( Figure 2—figure supplement 3B ) . To clarify if the missed connections were due to not detecting an existing interaction of a pair of neurons , or due to incorrect classification of directionality , we analyzed the detectability of connections independent of their direction ( Figure 2—figure supplement 3C ) , revealing similar results to the detect directed functional connections ( hits: 58% for SN , and 69% for CN; CR: > 99% for both ) . These findings suggest that the missed connections were due to not detecting an existing connection , in accordance with a high accuracy for extracting directionality of only detected connections ( Figure 2—figure supplement 3D; hits: 97% for SN , and 90% for CN; CR: 75% for SN , and 73% for CN ) . Our simulated networks also allowed for a closer evaluation of zero time lag peaks as a result of either common drive or bidirectional connections . In direct comparison , the average common drive CCH as well as the average bidirectional CCH had a maximum at the zero time lag , but with the average bidirectional CCH having a 24 times higher peak ( 10 . 89 SD surrogate for bidirectional connections , and 0 . 45 SD surrogate for common drive; Figure 2—figure supplement 3E ) , which is well in line with around 1% of all common drive pairs were detected as significant . When analyzing the distribution of maximum peaks in more detail , we found more than 7 times more bidirectional connections having a peak at the 0 time lag than common drive pairs ( Figure 2—figure supplement 3F ) , in line with the results from the models described above ( Vicente et al . , 2008; Gollo et al . , 2014 ) . Taken together , all results from the modeled networks show an accurate detectability of directed functional interactions estimated from CCHs . For a physiological classification of all significantly detected connections , we also analyzed their maximum peak or trough time lag distribution ( Figure 2—figure supplement 4A ) . Interestingly , the maximum peak or trough time lag distribution showed an exponential decay , with most of the peaks or troughs having a very short time lag ( 45 . 67% < 10 ms , and 85 . 12% < 100 ms ) , indicating predominantly direct influences of the units on each other . In case of oscillatory synchronized single units , as strongly present in the data , the classification of the maximum peak or trough time lags was more complex . Given that the maximum peak or trough time lag could be greater than half a cycle of the underlying frequency , it became unclear which unit is leading and which lagging , due to the presence of side lobes ( e . g . , see Figure 2A top panel ) . Since we found high numbers of oscillatory synchronized single units , predominantly in the beta ( 20 Hz ) and in the low frequency range ( 4 Hz ) , as described in detail below , we analyzed the distribution of maximum peaks or troughs phase with respect to the underlying oscillatory frequency ( Figure 2—figure supplement 4B ) , and also found an exponential decay , similar to the maximum time lag peak or trough distribution . The majority of phase lags were within half a cycle around the zero time lag for both frequencies ( beta connections: 77 . 70% < π , low frequency connections: 87 . 66% < π ) , suggesting that for most oscillatory synchronized connections we could accurately determine which unit was leading and which unit was lagging . For analyzing the functional network topology , all units not connected to the largest inter-connected component were first discarded ( mean number of units dropped: 17 . 75 , SD: 9 . 56; mean percentage: 23 . 5% , SD: 13 . 3%; Table 1 ) and binary directional connectivity matrices were created for every dataset ( Figure 3A ) . We did not quantify the connection strength , since it has been shown to be biased by different firing rates ( Cohen and Kohn , 2011 ) . 10 . 7554/eLife . 15719 . 011Figure 3 . Connectivity characteristics and modular topology . ( A ) Connectivity matrix of one dataset from monkey M . Each dot represents a significant connection ( Online Methods ) . Units are ordered by channel number of the recording system . ( B ) Distance dependent connectivity . From left to right: 56 , 7% , 11 , 5% , 5 , 6% , 5 , 5% , 2 , 6% , and 1 , 7% . Note the clear distance dependent decay . ( C ) The same matrix as in A , but with nodes ordered according to an optimal modularity partition . Colored rectangles surround different network modules . ( D ) Anatomical network representation of the connectivity matrix in A . The brain is viewed as in Figure 1B . Single units and connections are color coded by module . ( E ) Schematic illustration of modular topology . Modules ( dashed regions ) consist mainly of single units of one cortical area , but also include small fractions of units from other areas . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01110 . 7554/eLife . 15719 . 012Figure 3—figure supplement 1 . Example anatomical networks from Monkey S and Z . Since no data were recorded from area M1 for these monkeys , the F5 and AIP arrays are presented closer together than in reality for better illustration ( dashed line marks anatomical discontinuity ) . ( A ) Each node colored based on the module , as in Figure 3C . ( B ) Nodes and connections colored based on rich-clubness , as in Figure 4E . ( C ) Nodes and connections colored based on oscillatory components in the ACHs and CCHs , respectively , as in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01210 . 7554/eLife . 15719 . 013Figure 3—figure supplement 2 . Functional network connectivity of an exemplar data set displayed as a web where the locations of all neurons were determined using the visualization of similarities ( VOS ) approach ( Van Eck and Waltman , 2007 ) . ( A ) Each node is colored based on the area it was recorded . ( B ) Each node colored based on its module . ( C ) Nodes and connections colored based on oscillatory components in the ACHs and CCHs , respectively . ( D ) Nodes and connections colored based on rich-clubness . Each circle represents a single neuron and is scaled based on the degree of connectivity . VOS aims to find locations in a low-dimensional space ( in this case 2D ) in such a way that the distance between each node reflects the similarity between these nodes . Similarity is typically found by calculating the association strength ( also known as proximity index ) on the co-occurrence matrix of items , which is in this case the weighted network connectivity matrix . Association strength is simply the co-occurrence of two items divided by the product of the number of occurrences of each item . The location of each node is then found by minimizing the sum of the squared distance between all nodes , weighted by the computed similarity between each node . To avoid trivial solutions in which all nodes are assigned the same location , there is an additional constraint that the average distance between all pairs of items must be equal to one . Mathematically , VOS bares much similarity to the method of multi-dimensional scaling ( Van Eck et al . , 2010 ) . All implementations of VOS were performed using the freely available software , Pajek ( http://mrvar . fdv . uni-lj . si/pajek/ ) , and then plotted in Matlab . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 013 First , we tested if the networks could be subdivided into modules , such that the number of connections was maximized within and minimized between modules . To properly evaluate modular topology , the fact that connectivity decays with distance has to be considered ( Smith and Kohn , 2008; Gerhard et al . , 2011 ) . Figure 3B shows the distance-dependent decay of connectivity of our networks according to different subgroups: on the same electrode , on the same array , in the same area , between AIP and F5 , between F5 and M1 , and between AIP and M1 . Connection density was not significantly different within all subgroups ( Kruskal-Wallis test , p>0 . 05 ) . Modular topology can be quantified by the modularity index Q . If a network can be completely subdivided into modules , Q will be 1 . In contrast , if there is no modular structure present at all , Q will be close to 0 . We found significant modular topology present in most of the networks ( Mean Q: 0 . 405 , SD: 0 . 087; permutation test , p<0 . 05 , sig . 10/12 datasets ) , taking the distance-dependent decay of connectivity into account . Modules were significantly predominated by units from a single area ( mean largest proportion: 81 . 4% , SD: 14%; permutation test , p<0 . 001 ) , but 84% of all modules also included units from other areas , as became apparent when visualized as anatomical networks ( Figure 3D , and Figure 3—figure supplement 1A ) or when displayed as a web where the locations of all units are determined by visualization of similarities ( VOS ) ( Van Eck and Waltman , 2007 ) ( Figure 3—figure supplement 2A , B ) . These results reveal a functional modular topology partially not related to the anatomical boundaries between the different areas ( Figure 3E ) . Having shown that a modular topology is present , what is the detailed structure of how individual units are connected within the network ? For this , we calculated the cluster coefficient C ( with C = 1 corresponding to every neighbor of every unit being interconnected , and C = 0 indicating no interconnections between neighbors ) and the average path length , L ( defined as the average minimum number of units connecting one unit with another , across all pairs of nodes of the network; see Materials and methods section ) . If units have dense local clustering ( large cluster coefficient C ) and can be reached from all other units via a short average path length , L , similar to random networks , the network is considered small-world ( SW ) ( Watts and Strogatz , 1998; Bullmore and Sporns , 2009 ) . Here , a value of SW >> 1 indicates a small-world topology , whereas SW = 1 corresponds to no small-world effect . We found significantly higher average cluster coefficients C in comparison to surrogate networks ( mean: 0 . 266 , SD: 0 . 068; permutation test , p<0 . 001 , sig . 12/12 datasets ) and on average similar path lengths L ( mean: 3 . 451 , SD: 0 . 823; mean difference to surrogate networks: −0 . 007; permutation test , p<0 . 05 , sig . higher 5/12 , sig . smaller 5/12 datasets ) . Consequently , all networks had a significant SW-coefficient ( mean: 3 . 05 , SD: 0 . 66; permutation test , p<0 . 001 , sig . 12/12 datasets ) , suggesting that despite a modular structure the neuronal network is efficiently processing and transmitting information ( Watts and Strogatz , 1998 ) . Some networks , have been shown to exhibit heavy-tailed centrality distributions , with a small number of nodes strongly embedded in the network ( hubs ) , which make a strong contribution to the network function ( van den Heuvel and Sporns , 2013a ) . A simple and robust measure of centrality is degree centrality ( k ) , which is the number of connections per unit . On average 6 . 27% ( SD: 2 . 29% ) of all possible connections were realized . The degree distribution ( Figure 4A ) was heavy-tailed and best described by an exponential truncated power law model ( P ( k ) ~kγ−1ek/kc , γ = 0 . 6839; cutoff degree of kc = 8 . 657; EXPTPL: adjusted R2 = 0 . 9891 , including a penalty for number of fitted variables ) , compared to a power law ( P ( k ) ~k−γ; PL: adjusted R2 = 0 . 9177 ) , exponential ( EXP: adjusted R2 = 0 . 9742 ) , or Gaussian ( GAUS: adjusted R2 = 0 . 6826 ) model . In contrast , surrogate networks with the same distance-dependent connectivity were not heavy-tailed and were best described by a GAUS model ( GAUS: adjusted R2 = 0 . 9655; PL: adjusted R2 = 0 . 3061; EXPTPL: adjusted R2 = 0 . 5006; EXP: adjusted R2 = 0 . 6419 ) . In agreement with the EXPTPL model , networks had significantly more single units within the low , less within the intermediate , and especially more in the high degree range , than surrogate networks ( cluster-based permutation test , p<0 . 05 ) , clear evidence of hubs , independent of distance-dependent connectivity . 10 . 7554/eLife . 15719 . 014Figure 4 . Centrality measures , hubs , and rich-club topology . ( A ) Average degree centrality distribution of all networks ( blue ) and corresponding surrogate networks ( red ) . Black lines reflect different models fitted to the data ( see legend in B ) . The degree distribution of each dataset was normalized to the possible maximum number of connections per network . The area under the curve was normalized to 100% before averaging . Line shadings show standard error across datasets . Asterisks represent significant differences to surrogate networks . Inlay shows the same distribution and models on a log-log scale . ( B ) Same as in A , but for the betweenness centrality distribution . Note that the slopes for the EXPTPL and PL model are identical , since the exponential coefficient of the EXPTPL model was zero . ( C ) Schematic view of a rich-club topology connecting highly clustered modules . ( D ) Average rich-club level of all datasets relative to surrogate datasets . Asterisks represent significant differences of rich-club level to surrogate networks . ( E ) Anatomical network representation , as in Figure 3D , with connections and units color-coded based on rich-club membership ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01410 . 7554/eLife . 15719 . 015Figure 4—figure supplement 1 . Detectability of the underlying network topology using equal rate model simulations . ( A ) Average degree centrality distribution of all networks simulated with the equal rate model ( blue ) and the corresponding detected networks with the described method for detecting directed functional connectivity ( red ) . Results are shown for the same 10 simulated simple networks and 10 simulated complex networks as in Figure 2—figure supplement 3 . Error bars show the standard error across simulated networks . ( B ) Same as in A , but for the betweenness centrality distributions . ( C ) Same as in A , but for the rich-club level relative to surrogate datasets . Asterisks represent significant difference of rich-club level to surrogate networks . Two different sets of surrogate networks were calculated per dataset , one for the simulated network and one for the detected network . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01510 . 7554/eLife . 15719 . 016Figure 4—figure supplement 2 . Subsampling model . ( A ) Average degree centrality distribution of the modeled neuronal plane ( 32000 neurons , 2 areas , each divided into 5 subregions coverable by an array , 160 possible electrode position , and a maximum of 20 single units per electrode ) with distant dependent random connectivity ( Figure 3B ) . The distribution could be best described by a Gaussian model ( adjusted R2 = 0 . 98 ) . ( B ) Average degree centrality distribution of 12 different subsamplings of the modeled neuronal plane with exactly the same number of neurons as in the real datasets . Line shadings show standard error across subsamplings . Datasets were processed as in Figure 4A . Average degree distribution could be best described by a Gaussian model ( adjusted R2 = 1 ) and only poorly by a power law model ( adjusted R2 = 0 . 17 ) . ( C ) Dependency of goodness of power law fit , the size of the largest component relative to the whole network , and the level of compartmentalization on average degree k . Different average degrees were generated by varying the distance-dependent connectivity density of the empirically gained data ( Figure 3B ) by factors of 1/5 , 1/4 , 1/3 , 1/2 , 1 , 2 , 3 , 4 , and 5 times to create a neuronal plane . Goodness of power law fit was highly correlated with the size of the largest component ( adjusted R2 = 0 . 93 ) and the compartmentalization ( adjusted R2 = 0 . 93 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 016 A more global aspect of centrality is captured by betweenness centrality ( g ) , an index of the number of shortest paths from all single units to all others that pass through that single unit , normalized by the number of all shortest paths ( van den Heuvel and Sporns , 2013a ) . Similar to degree centrality , the betweenness centrality distribution ( Figure 4B ) was heavy-tailed and best described by a PL model , with an estimated exponent of γ = 2 . 212 ( PL: adjusted R2 = 0 . 9753; EXPTPL: adjusted R2 = 0 . 9745; EXP: adjusted R2 = 0 . 9593; GAUS: adjusted R2 = −0 . 1509 ) . The betweenness centrality distribution of surrogate networks was also heavy-tailed and was best described by an EXPTPL model ( EXPTPL: adjusted R2 = 0 . 99; PL: adjusted R2 = 0 . 9771; EXP: adjusted R2 = 0 . 9061; GAUS: R2 = −0 . 5511 ) . Still , in contrast to the PL model , the EXPTPL model had smaller values in the high and low betweenness centrality range . Statistically networks showed a significantly higher number of single units in the low and fewer units in the intermediate betweenness range than surrogate networks ( cluster-based permutation test , p<0 . 05 ) . These findings confirm the presence of hub neurons for betweenness centrality . Units with high degree centrality also tended to have high betweenness centrality ( r = 0 . 75 , p<0 . 001 , Spearman correlation ) , suggesting a coherent group of hub units . We found no significant differences in number of hubs per area ( normalized k ≥ 9 , g ≥ 0 . 03; Tukey's honest significant difference test on average group ranks , p<0 . 05 ) , indicating a distributed hub topology with no area acting as a network center . Together , we have shown that centrality of single units is strongly heterogeneous in the network , with a large group of units being marginally involved in the network and a small group of spatial distributed hub units being extremely central . The presence of hubs provides further evidence of a complex network topology at the single unit level . However , it has been shown that detectability of functional connections decreases with lower firing rates ( Cohen and Kohn , 2011 ) . Since the detected firing rates varied approximately across two orders of magnitude ( Figure 1—figure supplement 1B ) , this could lead to an underestimation of degree for low spiking units and an overrepresentation of high firing units as hubs . Therefore , we performed a careful examination of the influence of firing rates on degree and betweenness centrality based on our equal rate model ( see Materials and methods ) . Two sets of networks were tested , simple networks ( SNs ) and complex networks ( CNs ) , as mentioned previously . SNs had normally distributed connectivity based on the best fitting Gaussian model for the surrogate network degree centrality distribution , while connectivity for CNs were set to precisely resemble the EXPTPL model for the average degree centrality distribution of the measured networks . CNs additionally had a small-world and rich-club topology , as described in the following section . Differences in firing rate and any possible biases due to the applied method to estimate directed functional connectivity had no effect on the shape of the degree centrality distribution for both kind of networks ( Figure 4—figure supplement 1A ) . The betweenness centrality distribution for CNs was also unchanged and only slightly impaired for the SN ( Figure 4—figure supplement 1B ) . Nevertheless , the best fitting model for the betweenness centrality distribution of SNs was in neither case ( modeled or detected ) a PL , as it were for the measured data and the CNs , suggesting no distorting effect by differences in firing rate and the applied method to estimate directed functional connectivity . Importantly , also the average C , average L , and SW-coefficient were correctly detected for both kind of networks . It is also possible that subsampling , a natural limitation in electrophysiological recordings , could artificially cause a heavy tailed degree centrality distribution even if the underlying connectivity is random ( Han et al . , 2005; Gerhard et al . , 2011 ) . We simulated a neuronal layer of 32 , 000 neurons with the same distance-dependent connectivity density as detected in our data ( Figure 3B ) , but with Poisson distributed connectivity ( Figure 4—figure supplement 2A; see Materials and methods ) . Subsampling was performed in correspondence with our array configuration down to the number of neurons we recorded for real datasets , showing no change to the shape of the degree distribution ( Figure 4—figure supplement 2B ) . Only when we decreased the connection density of the model below the detected connectivity in our data was a false heavy-tailed degree distribution apparent ( Figure 4—figure supplement 2C ) , which was highly correlated with the networks breaking apart into unconnected components ( R2 = 0 . 93 ) . Additionally , this effect could not be present in our analyzed data since we only analyzed the largest component of the single unit networks . Theses controls suggest that the existence of hubs can neither be explained by distance-dependent connectivity , differences in firing rates , or subsampling . In some networks hubs exhibit a strong tendency to link to each other , forming a rich-club ( Colizza et al . , 2006 ) , which can be measured by a rich-club coefficient that expresses the tendency of highly connected hub nodes to show above-random levels of interconnectivity ( Figure 4C ) . Hub units showed a significantly higher level of interconnectivity than surrogate networks , with up to 15% more connections ( Figure 4D; cluster-based permutation test , p<0 . 05 ) . For our equal rate model , we tested if differences in firing rate and the applied method to estimate directed functional connectivity could cause a false rich-club effect . The present rich-club topology of CNs could be correctly detected , as well as no false rich-club topology was detected for SNs ( Figure 4—figure supplement 1C ) . Although the slope of the rich-club coefficient was changed for CNs , rich-club topology was only significant if present ( cluster-based permutation test , p<0 . 05 ) , suggesting a correct representation of rich-club topology for the measured networks . The rich-club contained neurons from all areas with a rich-club level set to k ≥ 9% ( Figure 4E , Figure 3—figure supplement 1B and 2C; mean rich-club neurons: 27% , SD: 18%; similar results with k set to other levels ) . A rich-club that spans multiple areas , as described here , has been proposed as a robust structure facilitating efficient communication ( van den Heuvel and Sporns , 2013a ) . Oscillatory synchronization has been proposed as a mechanism for efficient communication ( Fries , 2009 ) . As demonstrated above , oscillatory and non-oscillatory synchronized spike patterns for communication could be identified ( Figure 2 , Figure 2—figure supplement 1B , 2 ) . We therefore investigated if specific relationships between distinct frequencies and network topology emerged . Frequency spectra of ACHs of all units and of CCHs that had a significant connection were tested for significant frequency bins above chance ( cluster-based surrogate test , p<0 . 05 ) . We found beta ( 18–35 Hz ) and low frequency ( 3–7 Hz ) oscillations predominantly present in the spiking patterns of all datasets ( Figure 5A , and Figure 5—figure supplement 1C–E ) . Oscillatory synchrony in both frequency ranges was present more often in CCHs ( mean beta: 38 . 3% , low: 44 . 3% ) than in ACHs ( mean beta: 22 . 5% , low: 31 . 7% ) , suggesting that the group of oscillating single units ( oscillators; Table 2 ) communicates in their underlying frequency to a larger group of units . 10 . 7554/eLife . 15719 . 017Figure 5 . Low frequency and beta oscillators and their network topology . ( A ) Average number of significant frequency bins of all ACHs and CCHs over all datasets . Frequencies displayed on a logarithmic scale . Line shadings bars represent standard error across datasets . ( B ) Anatomical network representation as in Figure 3D with connections and units color-coded by underlying oscillations ( see legend in C ) . ( C ) Degree centrality distribution of all datasets separately for beta and low frequency oscillators , non-oscillators , and single units oscillating in both frequency ranges . Upper panel , summed degree centrality distribution of all single units . Median degree is represented by arrows in corresponding color: beta units: 7 . 5 , low frequency units: 6 . 3 , beta and low frequency units: 8 . 9 , and for non-oscillators: 2 . 7 . ( D ) Same as in C but for the betweenness centrality distribution . Median for beta units: 0 . 023 , low frequency units: 0 . 016 , beta and low frequency units: 0 . 026 , and for non-oscillators: 0 . 001 . ( E ) Schematic view of the found network topology of oscillators . Oscillators form a rich-club spanning all areas . ( F ) Distribution of oscillators across areas . The number of single units is normalized to 100% per area . F5 has significantly less beta ( red ) and significantly more low frequency oscillators ( blue ) than M1 and AIP . Note that units oscillating in both frequency ranges are counted in both . Non-oscillators ( black ) still remain the largest group in all areas . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01710 . 7554/eLife . 15719 . 018Figure 5—figure supplement 1 . Frequency dependent Hanning windows used for discrete Fourier transform . ( A ) Hanning windows used for discrete Fourier transform of all CCHs . All windows were aligned to the zero bin and span four times the frequency of interest period ( with a maximum of 1000 ms and a minimum of 150 ms ) . Frequencies of interest were scaled logarithmically ( 100 frequencies from 3 to 100 Hz ) . ( B ) Hanning windows used for discrete Fourier transform of all ACHs . All windows were aligned to the zero bin and span two times the frequency of interest period ( with a maximum of 500 ms and a minimum of 75 ms ) . ( C ) Significant frequency bins of power spectra of all ACHs of one example dataset per monkey . Frequencies were calculated and displayed on a logarithmic scale . ( D ) Significant frequency bins of power spectra of all CCHs of the same example datasets as in C . ( E ) Average number of significant frequency bins of all ACHs and CCHs of the same example datasets as in C and D . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01810 . 7554/eLife . 15719 . 019Figure 5—figure supplement 2 . Sensitivity of CCHs in detecting oscillatory synchrony and non-oscillatory synchrony . ( A ) CCHs for pairs of simulated neurons with an average firing rate around 5 Hz , either firing in an oscillatory ( 20 Hz , red curve ) or non-oscillatory manner ( black curve ) . By jittering their trial-wise temporal offset in firing , we simulated different levels of coupling strength , without disturbing the firing pattern of the individual neurons nor the similarity in firing between the two neurons . Results are shown for a trial-wise jitter of 0 ms ( perfect synchronization ) , 25 ms , and 50 ms ( hardly synchronized ) . ( B ) Maximum CCH peak heights of oscillatory and non-oscillatory neurons with a systematical trial-offset-jitter from 0 to 50 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 01910 . 7554/eLife . 15719 . 020Figure 5—figure supplement 3 . Differences in degree centrality and rich-club level for high and low oscillatory state . ( A ) Average power spectra of population ACHs of trials with high power in both the beta ( 18–35 Hz ) and low frequency ( 3–7 Hz ) band ( red curve ) , and of trials with low power in both frequency bands ( blue curve ) . Due to a limited amount of available trials , data is shown only for the two datasets ( M1 and M2 ) with more than 900 trials recorded . ( B ) Unit-wise degree centrality similarity for networks calculated on low and high oscillatory trials . Degree centrality is normalized by the maximum possible number of connections of all neurons detected . ( C ) Average degree centrality distribution of the same networks as in B . Degree centrality is normalized by the maximum possible number of connections of all neurons which were interconnected , excluding isolated neurons . Note that this normalization is slightly different between the high and low oscillatory state network and slightly different to B . ( D ) Same as in C , but for the average rich-club coefficient relative to surrogate datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 02010 . 7554/eLife . 15719 . 021Table 2 . Number of oscillators in all networks analyzed . Marked datasets correspond to the displayed example networks in Figure 5 and Figure 3—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 021DatasetsOscillators totalNon-OscillatorsBeta OscillatorsLow Frequency oscillatorsOscillators in both frequency rangesM 18365376014M 2 607728375M 3344512253S 1312614269S 2323214224S 3313315204S 4263814197S 5 403822257S 6212614103Z 113205102Z 21317692Z 3 *182310113Average33 . 536 . 715 . 922 . 85 . 3SD19 . 417 . 48 . 713 . 83 . 4 Interestingly , there was also a significant group of oscillating single units present in the gamma range ( 45–80 Hz ) , which was not mirrored in the CCHs . One possible explanation could be that that these units communicate via long-range gamma synchronization with topographically distant areas we did not record , such as the visual cortex ( Gregoriou et al . , 2009 ) . Oscillators and oscillatory connections were widely distributed and seemed to be very central across all areas ( Figure 5B , Figure 3—figure supplement 1C and 2D ) , giving rise to the idea that oscillators could be the hubs of the networks . Figure 5C shows the average degree centrality distribution for all networks , as in Figure 4A , but separately for beta and low frequency oscillators , non-oscillators , and units oscillating in both frequencies . There was a clear dominance ( high percentage ) of oscillators in the high degree range , whereas non-oscillators dominated in the low degree range . The degrees of all three oscillator groups were significantly higher than for non-oscillators ( Tukey-Kramer test for rank , p<0 . 001 ) . Betweenness centrality was also significantly higher for oscillators than for non-oscillators , similar to degree centrality ( Figure 5D; Tukey-Kramer test for rank , p<0 . 001 ) . The number of units oscillating in both frequencies was not higher than expected by coincidental overlap of the two frequency bands ( permutation test , p>0 . 05 ) . Nevertheless , it could be possible that CCHs are more sensitive to oscillatory synchrony than to non-oscillatory synchrony , which would induce a bias when comparing these two groups . At this point , it is important to emphasize that we first tested for significant connectivity independent of oscillatory behavior and only in a second step these connections were tested for their oscillatory behavior as described in the Materials and methods section . This ensured that any detected connection is based on a significant amount ( or suppression ) of coincidental spikes without any selective sensitivity for oscillatory coupling . As an additional test , we simulated pairs of neurons either with an oscillatory or non-oscillatory firing pattern ( see Materials and methods ) . Since peaks and troughs in CCHs reflect a systematic time lag in spiking between units across trials we simulated different degrees of coupling strengths by systematically varying the trial-wise time offset in spiking for both firing pattern types . Synchronization strength was simply a function of the variation in spike timing offsets between the two neurons and not whether the firing pattern was oscillatory or not ( Figure 5—figure supplement 2 ) , confirming that oscillatory coupling is not a priori more detectable than non-oscillatory coupling . Besides these methodological issues already addressed , it is possible that higher firing rates introduce a bias in the statistical detection of significant frequency bins , To control for this possibility , we applied thresholds for the detection of beta and low frequency oscillations . Thresholds were chosen to give , as closely as possible , the same number of beta and low frequency oscillators as statistical methods . Using this method all three groups had a higher degree and betweenness centrality than non-oscillators , similar to statistical detection ( Tukey-Kramer test for rank , p<0 . 001 ) . To rule out that firing rate dependent detectability of functional connections could cause a spurious inter-dependence of high centrality and detection of oscillatory synchrony , we repeated testing for differences in centrality only with units having a firing rate of 10 Hz and above , confirming that oscillators had significantly higher centrality values ( Tukey-Kramer test for rank , p<0 . 001 ) . Similar results were obtained when we tested the data of each monkeys individually ( Tukey-Kramer test for rank , p<0 . 01 ) . To our knowledge , the current results represent the first evidence that oscillators have a higher centrality in the single unit network than non-oscillators . Consequently , the rich-club of all networks overlapped significantly with oscillating single units ( permutation test , p<0 . 05 ) , highlighting oscillators as the backbone ( van den Heuvel et al . , 2012 ) of single unit functional connectivity ( Figure 5E ) . The number of oscillators did not differ between areas ( Tukey-Kramer test for rank , p<0 . 05 ) , in agreement with the distribution of hubs as well as rich-club units across areas . Closer examination of oscillator types revealed significantly more beta oscillators in AIP and M1 than in F5 , and more low frequency oscillators in F5 than in M1 and AIP ( Figure 5F; Tukey-Kramer test for rank , p<0 . 05 ) , reinforcing the notion that different cortical areas operate more strongly in some frequency ranges than others ( Brovelli et al . , 2004 ) . A further unresolved question is whether a direct relationship exists between oscillatory synchronization and functional rich-club topology . It is well known that oscillatory synchrony in frontal and motor areas appears in short bursts of only a couple of cycles with variable length and amplitude ( Murthy and Fetz , 1996; Lundqvist et al . , 2016 ) . We used this property of oscillatory synchrony to split up our data into two equal blocks with high oscillatory and low oscillatory synchrony to investigate the effect on rich-club topology . Since a minimum number of trials are required to properly estimate the functional connectivity for topological analyses , we used the two datasets from monkey M were we recorded more than 900 trials ( Table 1 ) . The data was split into two blocks with equal number of trials per condition to prevent any biases by different epochs or conditions . Instead of calculating unit-wise ACHs we pooled the activity of all units and estimated single trial population ACHs spectra , reflecting the trial-wise level of oscillatory synchronization . Single trial population ACHs calculations and frequency analyses were performed the same way as for single unit ACHs ( see Materials and methods ) and divided by their average power in the beta ( 18–35 Hz ) and low frequency ( 3–7 Hz ) band ( Figure 5—figure supplement 1C ) . After separation into two blocks , the estimation of functional connectivity and network topological analyses were repeated as if they were two separate datasets . For a valid statement about changes in rich-club topology , the network structure and in particular the degree distribution , should not be changed . For both datasets the unit-wise degree as well as the degree distribution were very similar ( Figure 5—figure supplement 3B , C ) , as well as the betweenness centrality distribution ( data not shown ) . However , when comparing the rich-club level there was a striking difference for higher rich-club levels ( Figure 5—figure supplement 3D ) . In both datasets , the high oscillatory state network showed a clear rich-club topology , whereas the low oscillatory state network hardly showed any rich-club effect . These results suggest that a rich-cub topology is only present when there is a high level of oscillatory synchrony in the network . Utilizing the identified network topology , the firing rate of individual units can be predicted by the firing rate of input units , providing an estimate on how much of the single unit activity can be explained by functional network connectivity . Each CCH can be understood as a transfer function of spike rates between two units , describing the coincidences per spike at every time point relative to each other . Negative time bins bin reflect input from the reference unit to the target unit while positive time bins reflect the output . To predict the firing rate of a unit , we convolved the spike trains of all units having a significant connection to the corresponding unit with their respective CCHs ( output part ) . Assuming single units to be simple linear integrators , we summed up the individual convolved spike trains ( Figure 6A , B ) and correlated these estimated signals with the original spike trains of the target units smoothed with a Gaussian kernel ( SD: 3 . 66 ms ) , identical to the CCH smoothing . Ninety-nine percent of predicted firing rate curves were positively correlated with the real firing rates of the corresponding target units ( Figure 6C ) . 10 . 7554/eLife . 15719 . 022Figure 6 . Prediction of firing rates based on network topology . ( A ) Average firing rate of one example single unit recorded in F5 in monkey S for the four conditions used in this study during the fixation ( Fix ) , cue ( Cue ) , memory ( Mem ) , and movement period ( Mov ) . The complex tuning patterns for the different task conditions ( grip types; free-choice vs . instructed trials ) are clearly visible . ( B ) Predicted firing rate of the same unit as in A based on the population activity of the connected neurons . Curves in ( A–B ) were smoothed with an additional Gaussian kernel ( SD: 40 ms ) . ( C ) Histogram of correlation coefficients between the true and predicted spike trains of all single units of all datasets . Significant correlations are marked in red . Note that hardly any correlation coefficient were negative . ( D ) Histogram of correlation coefficients of condition averaged firing rates . Coloring as in C . DOI: http://dx . doi . org/10 . 7554/eLife . 15719 . 022 However , these correlations could also be due to synchronous up and down states of the brain ( Gilbert and Sigman , 2007 ) , which makes proper statistical testing obligatory . Three different permutation tests were applied: shuffling of trials , shuffling of the output parts of CCHs , and shuffling of input units . Only if the correlation coefficient significantly exceeded all three permutation distributions ( p<0 . 05 ) , the correlation was considered significant . Remarkably , 45% of the firing rate patterns of our single units could be significantly predicted by their inputs . The differences between grasp types and decision conditions could be significantly predicted in 9% of all cases ( Figure 6D; positive correlation: 79%; shuffling of the transfer kernels and input units , p<0 . 05 ) , even using this simple approach that involved no parameter fitting . The functional network topology presented here allows a surprisingly accurate prediction of temporal firing dynamics , suggesting that the network captured in our recordings , despite being a small subset of the entire network , accurately represents a large portion of the relevant communication in the fronto-parietal grasping network . We analyzed single unit functional network topology across several cortical regions of three monkeys performing a delayed grasping task . The network was structured as a complex network ( Bullmore and Sporns , 2009 ) with a modular SW topology , and highly central hub-units localized in all three areas forming a rich-club . The advantage of such a topology is that it allows for fast and dynamical information processing combined with high robustness against errors ( Barabási and Oltvai , 2004; Bassett and Bullmore , 2006; Bullmore and Sporns , 2009; van den Heuvel et al . , 2012 ) . More detailed analyses of the kind of synchronization processes within the network revealed that the population of single units could be divided into two groups: oscillatory spiking and synchronized units in the low frequency range or in the beta range , and a group of non-oscillatory spiking units . Importantly , the hubs and therefore the rich-club consisted predominantly of oscillators , while the peripheral neurons were predominantly non-oscillators . Why is oscillatory synchrony such a central element of functional network topology ? More and more evidence supports the hypothesis that information is propagated not only as a simple rate code , but by feed-forward coincidence detection accomplished by oscillatory synchrony ( Fries , 2009 ) , meaning that phase-synchronization of neurons with one another is used as a selection mechanism for information transmission . The advantage of this mechanism is not only a reduction of energy cost , but also rhythmic gain modulation . By changing the phase of a synchronous neural population , such as in high-order areas , the input of one group of neurons can be selectively amplified as inputs to another group of neurons , allowing for high selectivity and high flexibility , which are exactly the requirements a hub has to fulfill ( van den Heuvel and Sporns , 2013a ) . While feed-forward coincidence detection can theoretically also be accomplished by non-oscillatory processes ( Fries , 2009 ) , the coordination of a network spanning different areas requires a larger group of neurons to fire in a coherent manner ( Buzsáki and Wang , 2012 ) . A rich-club of oscillating neurons is exactly that , a coherent structure cross-linking functionally segregated modules ( van den Heuvel and Sporns , 2013b ) , suggesting oscillators act as a backbone promoting and coordinating functional communication across different cortical areas ( van den Heuvel et al . , 2012 ) . This hypothesis is also in accordance with the finding that synchronization over larger distances ( >2 mm ) is almost always oscillatory , whereas synchronization over short distances occurs also in the absence of oscillations ( König et al . , 1995 ) . What are the roles of the two different distinct frequency bands present in this network ? Parietal and motor areas have been found to communicate via ~20 Hz beta synchronization ( Pesaran et al . , 2002; Brovelli et al . , 2004; Pesaran et al . , 2008; Dean et al . , 2012 ) and an increment in beta band activity seems related to the maintenance of the current sensorimotor or cognitive state , in agreement with findings in the basal ganglia ( Engel and Fries , 2010 ) . Oscillatory synchrony in the low frequency range ( 1–4 Hz ) has been shown to be important for communication within and between the prefrontal and motor areas ( Siegel et al . , 2009; Nácher et al . , 2013 ) and as a potential population mechanism of movement generation in motor and premotor cortex during reach initiation ( Churchland et al . , 2012 ) . Therefore , beta seems to be a stabilizing signal , low frequencies a global coordination signal , and both are involved in movement initiation with opposing roles . One possibility is that a function of the rich-club , composed of beta and low frequency oscillators spanning parietal and prefrontal cortex , is coordinating movement generation and initiation . Another possible explanation is that the power of fast oscillations is modulated by the phase of slow oscillations , termed cross-frequency phase-amplitude coupling , which could serve as a neuronal syntax for information transmission ( Buzsáki , 2010; Buzsáki and Mizuseki , 2014 ) . Our observation of oscillators in both frequency ranges simultaneously ( third row of Figure 2C , D , and Figure 5C , D ) supports this concept . Interestingly , we found that beta oscillators were present most frequently in AIP , followed by M1 , hardly in F5 , and in reverse order for low frequency oscillators ( Figure 5F ) . This is in line with the previous findings that information via beta band is primarily transmitted from the parietal to the frontal regions and not vice versa ( Brovelli et al . , 2004 ) . In areas that are hierarchically lower than the parietal lobe , such as the visual system , beta was identified as a top-down communication frequency ( Bastos et al . , 2015 ) . Therefore , the parietal lobe might be a center of beta generation . Low frequency oscillatory synchrony during active behavior has been found predominantly in prefrontal areas ( Siegel et al . , 2009; Nácher et al . , 2013 ) . We speculate that the center of low frequency oscillation could be in the prefrontal cortex , suggesting that different anatomical regions generate and communicate with different frequencies . The exact reason for the presence of distinct frequency bands for communication and their detailed interplay needs to be addressed in future studies . The single unit network topology was highly similar to the regional network of the brain measured by EEG , MEG , DTI or fMRI ( Bullmore and Sporns , 2009; Rubinov and Sporns , 2010; van den Heuvel et al . , 2012; van den Heuvel and Sporns , 2013a ) , which strongly suggests that the observed topological properties are scale-invariant ( Bullmore and Sporns , 2009 ) . Oscillatory synchrony may therefore act as a global coordination mechanism across the whole cortex . The modules of the network were primarily composed of the individual areas themselves . Yet , most modules also consisted of a small , but significant , proportion of units from other areas , indicating that the anatomical distance does not necessarily reflect the functional distance . This finding is in line with a recent study showing that the population of neurons within one area can be split up into 'choristers , ' which are strongly coupled to the rate of the whole population , and 'soloists , ' which are not ( Okun et al . , 2015 ) . We speculate that 'soloists' could be part of functional circuits centered in other brain areas , in accordance with the present modular topology . Since we recorded only from a subpopulation of the actual network , it was important to evaluate whether the observed network topology sufficiently represented the fronto-parietal grasping network . We demonstrated that a significant amount of the firing rate of single units could be predicted using only their network inputs , even for complex tuning patterns , suggesting that even a small fraction of the network is enough to characterize a reasonable amount of the spatio-temporal spiking dynamics . Furthermore , we demonstrated on a model that subsampling from a huge network with the same distance-dependent connectivity density as detected in our data did not affect the shape of the degree distribution ( Figure 4—figure supplement 2 ) . For these reasons , we are confident that our analyzed single unit network constitutes a significant representation of the underlying network dynamics . One possible point of misinterpretation of the functional network structure could be common drive , resulting in an overestimation of connectivity . Our method to detect functional connectivity corrects for common drive due to stimulus- and movement-locked inputs as well as for trial-wise fluctuations in spiking . Nevertheless , there are two possible additional sources of common drive . The first is the possibility that two neurons receive input from a third neuron while themselves being functionally uncoupled , resulting in a significant peak in the CCHs due to their input similarity . We investigated this possibility using our equal rate model , which included physiologically plausible firing rates and pairwise correlations . Common drive pairs of simulated simple or complex networks were detected as being significant in only around 1% of all cases , suggesting that , irrespective of the underlying topology , our method for detecting functional connectivity is hardly biased by pairwise common drive . The second possibility is that cortical columns or areas could receive common drive input that would cause these neurons to fire in a synchronized fashion even if they were functionally uncoupled . In such a scenario two things would be expected: first , units on the same electrode , as well as units in the same area , should show a similar connectivity pattern . Second , all neurons in the network should show a similar number of functional connections , since they are synchronized by common drive , resulting in a uniform degree centrality distribution . However , we found 43% of all neurons on the same electrode to be not connected , and only sparse connectivity was found in the same area with strongly connected pairs of neurons next to unconnected pairs ( e . g . , Figure 2—figure supplement 1B , 2 ) . Most importantly , the degree distribution of the measured networks was highly heterogeneous and heavy-tailed in contradiction to what would be expected by a strong influence of column- or area-specific common drive . Therefore , it is unlikely that event unrelated common drive can account for a significant amount of the detected functional connections . Further evidence arises from the fact that we found beta , low frequency , and non-oscillatory synchronization with different maximum peak or trough time time and phase lags ( Figure 2—figure supplement 3 ) , present simultaneously across all areas , also not consistent with a global common drive bias . To our knowledge , these results provide the first evidence of oscillatory synchrony as a central coordinating mechanism for the formation of functional network topology at the single neuron level . The combination of communication properties of oscillating single units and their functional topology adds an essential dimension to the understanding of neural circuits . By demonstrating that oscillating neurons form a backbone for functional connectivity , spanning several areas , we provide a unified basis for understanding the neuronal computations coordinating and generating behavior at the network level . Neural activity was recorded simultaneously from many channels in two female and one male rhesus macaque monkey ( Animals S , Z , and M; body weight 9 , 7 , and 10 kg , respectively ) . Detailed experimental procedures have been described previously ( Michaels et al . , 2015 ) . All procedures and animal care were in accordance with German and European law and were in agreement with the Guidelines for the Care and Use of Mammals in Neuroscience and Behavioral Research ( National Research Council , 2003 ) . Figure 1A illustrates the time course of the behavioral task as described previously ( Michaels et al . , 2015 ) . Trials started after the monkey placed both hands on the resting positions and fixated a red fixation disk ( fixation period ) . After 600 to 1000 ms , cues in the form of disks were shown next to the fixation disk for 300 ms to instruct the monkey about the required grip type ( power or precision; cue period ) . During this epoch the grasp target , a handle , was also illuminated . In the instructed task one disk was shown , while in the free-choice task both disks were turned on , indicating that the monkey was free to choose between the two grip types . The monkey then had to memorize the instruction for 1100 to 1500 ms ( memory period ) . The switching off of the fixation light cued the monkey to reach and grasp the target ( movement period ) in order to receive a liquid reward . Importantly , during free choice trials the reward was iteratively reduced every time the monkey repeatedly chose the same grip type . All trials were randomly interleaved and executed in darkness . The behavioral task also contained delayed instructed trials , which were not analyzed in this study . Surgical procedures have been described previously ( Michaels et al . , 2015 ) . In short , each animal was implanted with two floating microelectrode arrays per area ( FMAs; Microprobes for Life Sciences; 32 electrodes; spacing between electrodes: 400 μm; length: 1 . 5 to 7 . 1 mm monotonically increasing to target grey matter along the sulcus ) . Animal S and Z were implanted with four FMAs in area AIP and F5 in the left and the right hemisphere , respectively . Animal M was implanted with a total of six FMAs in the same cortical areas and two additional arrays in area M1 , in the left hemisphere ( Figure 1B ) . Neural signals from the implanted arrays were amplified and digitally stored using a 128 channel recording system ( Cerebus , Blackrock Microsystems; sampling rate 30 kS/s; 0 . 6–7500 Hz band-pass hardware filter; for monkey S and Z ) or a 256 channel Tucker-Davis system ( TDT RZ2; sampling rate 24 . 414 kS/s; 0 . 6–10 , 000 Hz band-pass hardware filter; monkey M ) . For spike detection , data were first low-pass filtered with a median filter ( window length 3 ms ) and the result subtracted from the raw signal , corresponding to a nonlinear high-pass filter . Afterwards the signal was low-pass filtered with a non-causal Butterworth filter ( 5000 Hz; fourth order ) . To eliminate common noise-sources principal component ( PC ) artifact cancellation was applied for all electrodes of each array as described previously ( Musial et al . , 2002 ) . To ensure that no individual channels were eliminated , PCs with any coefficient greater than 0 . 36 ( conservatively chosen and with respect to normalized data ) were retained . Spike waveforms were detected and semi-automatically sorted using a modified version of the offline spike sorter Wave_clus ( Quiroga et al . , 2004; Kraskov et al . , 2009 ) . Units were classified as single- or non-single unit based on five criteria: ( 1 ) , the absence of short ( 1–2 ms ) intervals in the inter-spike interval histogram for single units; ( 2 ) , the homogeneity and SD of the detected spike waveforms; ( 3 ) , the separation of waveform clusters in the projection of the first 17 features ( a combination for optimal discriminability of PCs , single values of the wavelet decomposition , and samples of spike waveforms ) detected by Wave_clus; ( 4 ) , the presence of well-known waveform shapes characteristics for single units; and ( 5 ) , the shape of the inter-spike interval distribution . After the semiautomatic sorting process , redetection of the different average waveforms ( templates ) was done to detect overlaid waveforms ( Gozani and Miller , 1994 ) . To achieve this , filtered signals were convolved with the templates starting with the biggest waveform . Independently for each template , redetection and resorting was run automatically using a linear discriminate analysis for classification of waveforms . After spike identification , the target template was subtracted from the filtered signal of the corresponding channel to reduce artifacts during the detection of the next template . This procedure allowed us to detect spikes with a temporal overlap up to 0 . 2 ms . Unit isolation was evaluated again , based on the five criteria mentioned above , to determine the final classification of all units into single or non-single units . Stationarity of firing rate was checked for all units and in case it was not stable over the entire recording session ( more than 30% change in firing rate between the first 10 min and the last 10 min of recording ) the unit was excluded from further analyses ( ~3% of all single units ) . Only single units fulfilling all of these criteria , and no multi-units , were further used in this study . After sorting , spike events were binned in non-overlapping 1-ms windows to produce a continuous firing rate signal ( 1 kHz ) and aligned to cue and movement onset . Two time windows were chosen for further analysis ( Cue onset: −700 to 1500 ms; Movement onset: −300 to 500 ms ) , since neuronal activity was locked to both events , with a variable memory period between them . Note that all three monkeys had very consistent movement times ( mean SD across datasets = 39 ms ) . The functional network topology of single-unit populations was derived from analyses of pairwise correlations ( Yu et al . , 2008 ) . We calculated cross-correlation histograms ( CCHs; time lags: −500 ms to 500 ms ) between all pairs of single units of each dataset ( Bair et al . , 2001 ) : ( 1 ) CCHn1 , n2 ( τ ) =1M∑i=1M∑t=1Nxn1i ( t ) xn2i ( t+τ ) ( N−\|τ\| ) λ1λ2 where M is the number of trials , t is time , N is the number of time bins in the trial , xn1i and xn2i are the spike trains of single units n1 and n2 on trial i , τ is the time lag , and λ1 and λ2 are the mean firing rates of the two single units across the entire time interval M . The denominator is normalizing for the degree of overlap ( N−\|τ\| ) in the CCH and the geometric mean spike rate λ1λ2 , which is the most common normalization used for CCHs ( Bair et al . , 2001; Smith and Kohn , 2008 ) . The normalized CCHs were then averaged across all time periods and task conditions ( e . g . , see Figure 2—figure supplement 1A ) . Subsequently , all CCHs were corrected for correlations induced by common stimulus drive or global state changes , such as arm and hand movements , as well as for trial-wise fluctuation in spiking , by simulating and subsequently subtracting surrogate CCHs . Surrogate CCHs contain the same stimulus locked correlation , but no pairwise temporal correlation . To this end , peri-stimulus time histograms ( PSTH ) were calculated for the same two time windows and alignments ( Cue and Movement onset ) as mentioned above , separately for each single unit and task condition ( smoothed with a Gaussian kernel , SD: 3 . 66 ms ) . Artificial spike trains were generated from an inhomogeneous Poisson process using the PSTHs as the rate function ( Ramalingam et al . , 2013 ) . These artificial spike trains preserved the number of trials and the number of spikes per trial , but varied in the timing of individual spikes ( surrogate data; e . g . , Figure 2—figure supplement 1A ) . Since the number of spikes per trial was preserved for all units recorded simultaneously , any trial-wise common drive is equally present and therefore accounted for in the surrogate data ( Smith and Kohn , 2008 ) . From these surrogate data , surrogate CCHs were calculated by replacing xni with the trials of the artificial spike trains for the corresponding single unit ( surrogate CCHs ) . This procedure was repeated 1000 times . The resulting surrogate CCHs reflected the level of correlation when both units are statistically independent . Finally , average surrogate CCHs were subtracted from the CCHs to yield the corrected CCHs . Auto-correlation histograms ( ACHs ) were generated by setting xn1i=xn2i in Equation 1 for all i , and corrected by generating artificial spike trains and substituting them for xn1i and xn2i in Equation 1 for the calculation of surrogate ACHs . For statistical purposes , all surrogate CCHs were corrected by their own average to achieve an equally processed set compared to the corrected CCHs , containing just the chance level of correlation ( corrected surrogate CCHs ) . These 1000 corrected surrogate CCHs were then used to run a nonparametric cluster-based surrogate test , a variation of the cluster-based permutation test ( Maris and Oostenveld , 2007 ) , to deal with the multiple comparison problem of testing all time lags . Cluster-based tests are tests for dependent variables , which consider contiguous values fulfilling a certain criterion as a cluster . Instead of calculating a test statistic for individual values , the accumulated values of clusters are tested against a null distribution of accumulated cluster values by chance . In our case , adjacent time lags are not independent , since functional coupling of neurons does not follow millisecond precision . We checked significance for a time window of −200 ms to 200 ms . Calculation of this test statistic involved the following steps: This procedure was repeated for every CCH . A critical alpha-level of 0 . 05 was selected . Nevertheless , at this processing step we still have a total alpha-error equal to our set criterion times the number of single unit pairs tested . For complete multiple comparison correction , false discovery rate correction was applied on all found clusters across all compared pairs of single units ( Benjamini and Hochberg , 1995 ) to yield ( 2 ) P ( k ) ≤ kmq where q is our set criterion of 0 . 05 false positives , m the total number of clusters , k = 1 , … , m , and P ( k ) are the p-values of all clusters in increasing order . All clusters whose p-values did not fulfill Equation 2 were rejected . By doing so we achieved a total alpha-level of 0 . 05 for each dataset . For every pair of neurons it was evaluated if there were significant troughs or peaks in their CCHs . If there was only a trough or peak with negative ( or positive ) time lags , this pair was denoted as having a connection from the input to the target ( or the target to the input ) unit ( Figure 2E ) . In case there were several clusters on both sides of the zero time lag , or a cluster straddling the zero time lag , we checked the unsigned maximum peak of the corresponding CCH . If the maximum peak was shifted more than 2 ms to either side , the connection was considered unidirectional , as described before . Otherwise , the connection between the two single units was considered functional bidirectional ( Figure 2E ) , since the units are driven by the circuit at the same time . We systematically varied the maximum peak shift ( 0–5 ms ) for bidirectional classification with little to no change to the results . Repeating this procedure for all pairs of single units led to a binary directed connectivity matrix ( Figure 3A ) . To characterize brain networks on every scale , network measures from the multidisciplinary field of graph theory were utilized ( Rubinov and Sporns , 2010 ) . A network is defined by the nodes ( N ) and connections between pairs of nodes . In our network nodes represented single units . For all following network measures , n is the number of nodes and l the number of connections . aij is the connection between nodes i and j: aij=1 if the link ( i , j ) exists and aij=0 otherwise ( aii=0 for all i ) . Furthermore , we define: Degree centrality , ki , is the number of connections to a node i . ( 3 ) ki= ∑j∈Naij Shortest path length , di , j , is the minimum number of nodes connecting nodes i and ( 4 ) dij= ∑auv∈gi↔j auv j . where gi ↔j is the shortest path between i and j . Characteristic path length , L , is the average shortest path length between all pairs of nodes of the network . ( 5 ) L= 1n ( n−1 ) ∑i , j∈Ni≠jdij Betweenness centrality , gi , is the average fraction of shortest paths that pass through node i . ( 6 ) gi= 1 ( n−1 ) ( n−2 ) ∑h , j∈Nh≠j , h≠i , j≠iρhj ( i ) ρhj where ρhj is the number of shortest paths between h and j , and ρhj ( i ) is the number of shortest paths between h and j that pass through i . Clustering coefficient of the network , C , is the average fraction of existing to maximal possible interconnections between all directly connected nodes to node i . ( 7 ) C= 1n∑i∈N2tiki ( ki−1 ) Where ki are all connected neighbors to node i and ti is the number of links between them . Small-worldness , SW , is the ratio of C and L each normalized by the same measurements for a size matched random network . ( 8 ) SW= C/CrandL/Lrand Small-world networks are formally defined as networks that are significantly more clustered than random networks , yet have approximately the same characteristic path length as random networks ( Watts and Strogatz , 1998 ) . Modularity , Q , is the proportion of all links within modules M with links between modules , when the network is fully subdivided into non-overlapping modules in a way that maximizes the number of within-group connections and minimizes the number of between-group connections . ( 9 ) Q= ∑u∈M [euu− ( ∑v∈M euv ) 2] where euv is the fraction of all links that connect nodes in module u with nodes in module v . Rich-club coefficient , R , at degree k is the fraction of connections between all nodes of degree k or higher , with respect to the maximum possible number of such connections . ( 10 ) R ( k ) = 2E>kN>k ( N>k−1 ) where E>k is the number of connections among the N>k nodes having degree of k or higher ( Colizza et al . , 2006 ) . To reduce inaccuracy for large degrees we calculated the rich-club coefficient only in degree bins containing at least 5 single units ( Nk≥5 ) . For statistical purposes we created two types of surrogate network sets per dataset ( 1000 partitions each ) . All surrogate networks were created by shuffling the connectivity matrix . Since connectivity is a function of distance ( Smith and Kohn , 2008; Gerhard et al . , 2011 ) , distance dependency was reflected in our surrogate data . During shuffling , the number of connections for single units on the same electrode , the same array , the same cortical area , and the different inter-area connections were always held constant ( Figure 3B ) . For all surrogate networks , the total number of single units , number of connections , and the distance-dependent ratio of bi- and uni-directional connections were kept as similar as possible to the original connectivity matrix with only the required network parameter shuffled . We used these sets of surrogate networks to test the small-world coefficient , the degree centrality distribution , and the betweenness centrality distribution . Statistical testing of the rich-club coefficient and conservative testing of modularity requires surrogate networks with a matched degree centrality distribution . To this end , we generated a second set of surrogates networks with the degree distribution preserved . One issue that could arise due to shuffling is that the connectivity matrix of some units or groups of units could become disconnected from the main part of the network , since the calculation of most network measures requires a fully connected , not segregated , network . For this purpose , each surrogate network was tested for segregation into different components . If a network was segregated , it was discarded and the process repeated until 1000 non-segregated networks were generated . To determine if the degree , the betweenness centrality distribution , or the rich-club level were significantly different to surrogate networks , we used a nonparametric cluster-based permutation test ( Maris and Oostenveld , 2007 ) . Briefly , this test evaluates the t-statistic ( independent samples ) between centrality or rich-club distributions and their surrogate distributions over all data points exceeding a critical alpha-level set to 0 . 05 . In a second step , adjacent degree , betweenness values , or rich-club coefficients exceeding the set alpha-level are considered as clusters , extracted , and their t-value summed . A test distribution was generated by randomly permuting the centrality or rich-club distributions across recording days and monkeys with the corresponding surrogate distributions by randomly reassigning them to one of the two groups while maintaining the group size . For each partition ( 1000 partitions ) the t-statistics and clustering was repeated . From every partition the largest cluster-level statistic was used to generate a largest chance cluster distribution . For each real cluster-level statistic a nonparametric statistical test was performed by calculating a p-value under the largest chance cluster distribution . Thus , the multiple comparisons for each sample are replaced by a single comparison , replacing the need to make multiple comparisons . Since some electrode pairs between F5 and M1 are closer than some other pairs within M1 for monkey M , we repeated statistics for network measures for all datasets from monkey M with physical distance dependent shuffling instead of the above mentioned categories such as 'same electrode' , 'same array , ' and 'same area' . To this end , we calculated the pairwise physical distance between all pairs of electrodes based on an anatomical diagram ( Figure 1B ) and defined distance groups with a stepsize of 3 . 6 mm including 0 mm as one group . The physical distance between AIP and the two other areas is misleading , since the neuronal axons have to pass the central sulcus . Therefore , we set all distances between AIP and the two other areas as a separate maximum distance group . Note that we had to define groups to be able to shuffle connections . Nevertheless , the categorical distance dependent shuffling was subdivided into 8 groups , which is more conservative than the 6 groups defined in the original analysis . All statistics for network measures gave nearly identical results , with no case where a measure was significant when it was not for categorical distance dependent shuffling , and vice versa for non-significant measures . In addition , the normalized rich-club coefficient , which depends on the surrogate networks , was highly correlated ( r = 0 . 98 ) between the two different ways of distance dependent shuffling . For validation of the estimates of directed functional connectivity , as well as to check for a possible bias in the detected network topology obtained using CCHs , we modeled artificial directed neuronal networks with the same firing rate distribution as the recorded single units . Two sets of networks were generated , one simple network ( SN ) set with normally distributed connectivity and one complex network ( CN ) set with heterogeneously distributed connectivity , and in agreement with previous studies both with weak connection strength between neuronal pairs ( Cohen and Kohn , 2011 ) . For each simulated neuron , artificial spike trains were generated with Poisson distributed firing and an average rate randomly drawn from the real firing rate distribution . For the SN set , the number of connections from each neuron to other neurons was drawn randomly from a Gaussian distribution ( mean: 5 . 22 , SD: 3 . 214 ) , mirroring the average degree centrality distribution of surrogate networks . For the complex network set ( CN ) , the number of connections followed precisely the EXPTPL model for the average degree centrality distribution of the measured networks ( Figure 4A ) , with a weak rich-club and small-world topology . In case one neuron was connected to another , spikes were added in a probabilistic manner for a certain amount of time , starting with time point t+1 in ms relative to the spike event , reflecting the axonal delay . The network was updated every millisecond , allowing for multiple interactions . Gamma functions were used as temporal transfer kernels , given by ( 11 ) f ( t\|a , b ) = 1baΓ ( a ) ta−1e−tb where f is the probability of an additional spike appearing , t is time in ms , a is a constant set to 5 and b is randomly varied between 0 and 3 ( Figure 2—figure supplement 3A ) . The integral of each gamma kernel was set to 0 . 02 , reflecting the connection strength . Since we added spikes to the network , which increases the average firing rates , we lowered the starting rates by a factor and repeated the process until the average rate resembled the rate before adding the connections . As a criterion for similarity we correlated the randomly drawn rates with the network rates and stopped when the residual error was below 0 . 005 . For the results in Figure 2—figure supplement 3 and Figure 4—figure supplement 1 we did not vary the connection strength in order to avoid interaction effects between connection strength and firing rate . However , we varied connection strength randomly between 0 . 005 and 0 . 035 with no detectible change to the results . Alternatively , we used a Boxcar kernel ( 20 ms , integral: 0 . 02 ) instead of gamma functions as transfer kernel , which did not degrade the results of this model . For both sets of networks ( SN and CN ) , ten artificial networks with 100 neurons were calculated and processed identically to the real data . Signal detection theory was used to evaluate detectability of connections based on significant CCH peaks or troughs with the originally modeled networks as a reference . Each pairing was classified into one of four categories: 'Hit' , if a connection was correctly detected , 'Miss' , if a connection was not detected , 'Correct rejection' ( CR ) , if a non-existing connection was detected as no connection , and 'False Alarm' ( FA ) , if a non-existing connection was detected as a connection . We generated an artificial neuronal plane with random ( Poisson distributed ) , distance-dependent connectivity density based on our empirically collected data ( Figure 3B ) . We modeled 2 cortical areas , each divided into 5 sub-regions coverable by an array , each sub-region covered with 160 electrode positions , and 20 single units per electrode , giving a total of 32 , 000 neurons . Figure 4—figure supplement 2A shows the degree centrality distribution of the full network with an average degree of 3000 and a standard deviation of 70 . Next , we randomly selected 12 subsamples from the neuronal plane with exactly the number of neurons detected as in the real datasets . Subsampling was done with the restriction that always both areas were chosen , with 2 array sub-regions per area and 32 electrode positions per sub-region , reflecting the real recording configuration in most of the datasets . Subsampled networks were then analyzed with the same complex network measures as the real data . To address the problem that subsampling could artificially cause a heavy tailed degree centrality distribution , even if the underlying connectivity is random , as described in Han et al . ( 2005 ) , we had a closer look at the parameters mentioned in this study . The average degree of their analyzed networks was 2 . 19 ( SD = 0 . 45 , min = 1 . 84 , max = 2 . 98 ) , in contrast to our average ( non-normalized ) degree of 8 . 28 ( SD = 5 . 73 , min = 3 . 87 , max = 25 . 59 ) . Note that the highest average degree of their analyzed networks was smaller than the lowest average degree of our analyzed networks . More importantly , the underlying networks of their study were strongly fragmented into components ( min = 70 , max = 591 components ) , while we excluded all single units which were not part of the largest component , resulting in one component for analysis , while their largest average component size was 20 . 2 . Our network analysis was done on average on 70 single units ( min 30 , max 148 single units ) . Based on these different network parameters we concluded that the detected topology , in particular falsely detected power law degree distribution , could be due to the fragmentation into different components . To evaluate this , we created neuronal planes with distance dependent connection density of 1/5 , 1/4 , 1/3 , 1/2 , 1 , 2 , 3 , 4 , and 5 times of the empirically collected data . After subsampling , we estimated the goodness of fit for the power law model to the degree centrality distribution , the size of the largest component relative to the whole network , and the level of compartmentalization , described by ( 12 ) Compartmentalization=P−1N−1 where N is the number of neurons in the network and P the number of separate components ( Figure 4—figure supplement 2C ) . We estimated the oscillatory behavior of significant connections of single units ( according to CCHs ) and the spiking of single units themselves ( Bair et al . , 1994; Mureşan et al . , 2008 ) ( according to ACHs ) . Since different oscillation frequencies could be present , we computed power spectra of all corrected CCHs and ACHs ( Mureşan et al . , 2008 ) . The power spectrum gives the magnitude of a signal as a function of frequency . To avoid distortions by sharp peaks with small delays that are occasionally present in CCHs ( Fujisawa et al . , 2008 ) , which cause a broad band increase in power due to their impulse like properties , we cut out the time range from −5 ms to 5 ms and interpolated the segment linearly . Importantly , sharp peaks were only removed for spectral analyses and not for functional connectivity analyses . Frequency spectra were computed using a discrete Fourier transform algorithm ( Siegel et al . , 2009 ) ( 100 logarithmically scaled frequencies from 3 to 100 Hz ) . Note that computing power spectra of CCHs and ACHs instead of raw spike trains reduced the influence of firing rate on the power spectrum as well as the problem of frequency leakage due to the binary properties of the spike train ( Bair et al . , 1994 ) . In analyzing such a large range of frequencies we had to take the specific characteristics of CCHs into account . Underlying oscillation frequencies in physiology are not phase stable , which leads to a limited number of side lobes in the CCH or ACH . The number of side lobes are also strongly frequency dependent , which makes the ideal window length for Fourier transformation around the 0 time lag frequency dependent . We used Hanning windows of four times the frequency of interest period ( with a maximum of 1000 ms and a minimum of 150 ms ) aligned on the 0 time bin of the CCHs ( Figure 5—figure supplement 1A ) , resulting in approximately 1/frequency and half octave spectro-temporal bandwidth . Each frequency bin was divided by its window length for correct scaling of all frequency bins . To determine significance , we repeated spectral analysis on the corrected surrogate CCHs and ACHs , subtracted their mean spectra from the corresponding spectra of real data and used a cluster-based surrogate test as described before to evaluate the significance of the underlying frequencies in the CCHs . Spectral analysis of the ACHs differed in one point . Hanning windows covering only one half of the ACHs ( with a maximum of 500 ms and a minimum of 75 ms ) aligned on the 0 time lag were used ( Figure 5—figure supplement 1B ) . By doing so , an accurate measure of the full frequency range with little distortion of refractory effects present in ACHs ( Mureşan et al . , 2008 ) was obtained . We generated pairs of neurons with 600 trials and a trial length of 3 . 1 s , similar to our recorded data . Spike trains of neurons were generated as a probabilistic process . In case of oscillatory firing neurons , the probability function was a 20 Hz sinusoid . For non-oscillating neurons , we first randomized the 20 Hz sinusoid , in a second step filtered it with a non-causal 50 Hz low-pass filter ( Butterworth filter , fourth order ) in order to produce a similar decay in spiking probability , and in a last step the filtered probability vector was variance matched with the 20 Hz sinusoid to have a maximum degree matching between the two kinds of probability functions . For each trial the same probability function was used for both neurons with a spiking probability of 0 . 05 per ms to stay in a physiological range . Independent Poisson distributed noise was added to both neurons representing background stochastic firing , resulting in an average rate of around 5 Hz per neuron . Varying the different parameters within physiological ranges did not alter the results . To simulate different degrees of coupling strengths we systematically varied the trial-wise time offset in spiking of the pair of neurons to each other from completely synchronized to a jitter of a complete cycle ( 50 ms ) in steps of 1 ms .	The network of neurons in our brain generates all of our actions , yet it is not well understood how these neurons coordinate their activity with each other . Rhythmic electrical activity that happens at the same time across many different neurons is thought to be crucial for allowing different areas of the brain to communicate . However , it is still unclear what purpose rhythmic activity serves for communication . Are there groups of ‘hub’ neurons in different brain regions that coordinate overall activity by rhythmically synchronizing the network of neurons ? Or is rhythmic activity insignificant for network coordination ? Dann et al . trained three monkeys to follow specific instructions to grasp a handle in different ways . While the monkeys performed the task , the activity of about 100 neurons was recorded simultaneously in three brain regions that are involved in planning and carrying out grasping movements . This revealed that the activity of the neurons was coordinated by a group of strongly connected hub neurons , which were distributed across all three of the brain regions . Nearly all of the hub neurons were rhythmically synchronized with each other , and also communicated with other neurons using rhythmic electrical activity . Overall , the results presented by Dann et al . suggest that rhythmically synchronized activity is essential for neurons to coordinate how information is processed across the brain . Further studies into this method of communicating information will help to reveal how the primate brain can generate an immense range of behaviors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Uniting functional network topology and oscillations in the fronto-parietal single unit network of behaving primates
The circadian clock interacts with other regulatory pathways to tune physiology to predictable daily changes and unexpected environmental fluctuations . However , the complexity of circadian clocks in higher organisms has prevented a clear understanding of how natural environmental conditions affect circadian clocks and their physiological outputs . Here , we dissect the interaction between circadian regulation and responses to fluctuating light in the cyanobacterium Synechococcus elongatus . We demonstrate that natural changes in light intensity substantially affect the expression of hundreds of circadian-clock-controlled genes , many of which are involved in key steps of metabolism . These changes in expression arise from circadian and light-responsive control of RNA polymerase recruitment to promoters by a network of transcription factors including RpaA and RpaB . Using phenomenological modeling constrained by our data , we reveal simple principles that underlie the small number of stereotyped responses of dusk circadian genes to changes in light . Circadian clocks allow organisms from almost all branches of life to alter physiology in anticipation of diurnal changes in the environment . Circadian clocks are autonomous core oscillators that keep time even in the absence of environmental cues ( Dunlap et al . , 2004 ) . Output pathways interpret timing information from the core oscillator to generate oscillating outputs , such as oscillations in the mRNA levels ( expression ) of genes and higher order behaviors ( Dunlap et al . , 2004; Wijnen and Young , 2006 ) . Laboratory studies of the outputs of circadian clocks have been primarily performed under constant conditions to isolate circadian regulation from environmental responses . In nature , however , organisms with circadian clocks must also cope with unexpected fluctuations in the environment . Thus a major challenge in chronobiology is to understand circadian regulation in dynamic environments . Previous studies suggest that circadian clock output pathways interact with responses to the environment to tailor physiology to both the time of day and the current state of the environment . For example , sleep/wake cycles in Drosophila melanogaster and photosynthesis in Arabidopsis thaliana are controlled by both the circadian clock and environmental variables like day length or light ( Lamaze et al . , 2017; Millar and Kay , 1996 ) . Further , circadian clocks can modulate responses to the environment based on the time-of-day in a process called circadian gating ( Hotta et al . , 2007; Greenham and McClung , 2015 ) . However , the complexity of higher organisms has prevented a detailed understanding of the interaction between circadian timing information and environmental responses . In contrast , the circadian clock in the cyanobacterium Synechococcus elongatus PCC7942 , an obligate photoautotroph , has a simple architecture which controls gene expression oscillations ( Figure 1A ) to influence metabolism and growth . S . elongatus must carefully monitor its environment , as the sunlight required for photosynthesis fluctuates on the minute , day , and seasonal timescales ( Figure 1B , [Petty and Weidner , 2017] ) . While it is well understood how the circadian clock in S . elongatus behaves under constant conditions , it is unclear how this system changes in natural , fluctuating light . In S . elongatus grown under ‘Constant Light’ conditions ( Figure 1A , dashed navy blue line ) , genes which show oscillatory expression ( circadian genes ) can be divided into two groups , the dawn and the dusk genes , which peak at subjective dawn and subjective dusk ( Ito et al . , 2009; Vijayan et al . , 2009 ) ( Figure 1A ) . Subjective dawn and subjective dusk refer to the times at which dark-to-light or light-to-dark transitions would occur in a 12 hr light-12 hr dark environmental cycle . The dawn genes consist of the core metabolic and growth genes for S . elongatus , including the photosystems , ATP synthase , carbon fixation/Calvin-Benson-Bassham cycle enzymes , and ribosomal proteins ( Vijayan et al . , 2009; Ito et al . , 2009; Diamond et al . , 2015 ) . In the absence of regulation by the circadian clock under Constant Light , S . elongatus constantly expresses dawn genes ( Markson et al . , 2013 ) . The clock primarily regulates the expression of dusk genes ( Markson et al . , 2013 ) , which include the genes required to utilize glycogen as an energy source in the absence of sunlight , such as glycogen phosphorylase and cytochrome c oxidase . As such , the circadian clock serves a critical function in switching S . elongatus from a daytime state of photosynthesis to a nighttime state of carbon metabolism through glycogen breakdown ( Diamond et al . , 2015; Diamond et al . , 2017; Pattanayak et al . , 2014; Puszynska and O'Shea , 2017 ) . In Constant Light conditions , the dusk and dawn genes show oscillatory expression with a 24 hr period , resulting in broad peaks of maximal expression ( Figure 1A , solid green line and dashed maroon line ) ( Vijayan et al . , 2009; Ito et al . , 2009 ) . Recent whole-cell modeling of metabolism , protein levels , and growth predict that this picture of circadian gene expression should change under the dynamic light conditions of a natural , clear day ( Figure 1B , navy blue line ) ( Reimers et al . , 2017 ) . The modeling suggests that making and using glycogen is a major cost to cell growth and thus the expression of genes required to switch metabolism from photosynthesis to glycogen breakdown should be delayed until absolutely necessary ( Reimers et al . , 2017 ) . However , gene expression in natural light conditions has not been measured in S . elongatus . Consistent with predictions of light-dependent changes in circadian gene expression , current evidence suggests interaction between the circadian and light regulatory pathways . The cyanobacterial clock keeps track of the time of day using a core post-translational oscillator ( PTO ) that consists of three proteins , KaiA , KaiB , and KaiC , whose enzymatic activities result in 24 hr oscillations in the phosphorylation state of KaiC ( Nakajima et al . , 2005; Rust et al . , 2007; Johnson et al . , 2011 ) . In vivo under Constant Light conditions the Kai PTO modulates circadian gene expression by controlling oscillations in phosphorylation state of the master OmpR-type transcription factor RpaA ( Markson et al . , 2013; Takai et al . , 2006 ) to peak at subjective dusk ( Figure 1A , dotted black line; Figure 1C ) ( Gutu and O'Shea , 2013; Takai et al . , 2006 ) . Phosphorylated RpaA ( RpaA∼P ) binds to the promoters of some dusk genes to activate their expression , leading to indirect activation of other dusk genes and repression of dawn genes ( Figure 1C ) ( Markson et al . , 2013 ) . As kaiBC is a dusk gene target of RpaA , the Kai PTO directs its own expression , resulting in a transcription-translation feedback loop that stabilizes the phase of the clock ( Qin et al . , 2010; Teng et al . , 2013; Zwicker et al . , 2010 ) . Exposure to complete darkness at specific times of day causes phase shifts in the PTO to align clock output with the external day/night cycle ( Rust et al . , 2011 ) , in a process called entrainment . However , it is not understood whether any aspect of this model , such as the dynamics of RpaA activity or the transcription-translation feedback loop , changes in the presence of more subtle natural light changes during the day ( Figure 1C ) . Meanwhile the OmpR-type transcription factor RpaB binds to some circadian gene promoters ( Hanaoka et al . , 2012 ) , and the phosphorylation state and DNA binding activity of this protein decreases in response to high light exposure ( Figure 1C ) ( López-Redondo et al . , 2010; Moronta-Barrios et al . , 2012 ) . However , it is not clear how natural light changes like sunset or shade pulses affect RpaB activity ( Figure 1C ) . RpaB clearly plays some role in altering circadian gene expression in response to light ( Espinosa et al . , 2015 ) , but it is unclear how ( Figure 1C ) . While light likely exerts global , growth-rate-dependent regulation of protein levels ( Scott et al . , 2010; Du et al . , 2016; Burnap , 2015 ) , the interaction between circadian and light regulation to control the activities of RpaA and RpaB represents a particularly tractable scenario for dissecting the mechanisms underlying interaction between clock and environment to control circadian gene expression . Here we measure and model circadian gene expression and several layers of regulation in cyanobacteria grown under the fluctuating light intensities typically experienced in nature . We find that fluctuations in light alter the expression patterns of almost all circadian genes . We identify key regulatory steps at which information about changes in light interact with clock output pathways to control gene expression , and reveal a complex regulatory network underlying circadian gene expression in natural conditions . Finally , we show that phenomenological models effectively describe the integration of the circadian clock with responses to environmental fluctuations . To grow and assay cyanobacteria in natural light conditions , we custom-built a culturing setup with a light source that can be programmed to mimic natural fluctuations in sunlight . On a cloudless ‘Clear Day , ’ light intensity varies in a parabolic manner due to the rotation of the Earth , ending with a gradual ramp down of light intensity prior to dusk ( ‘Sunset’ , Figure 1B ) . Rapid changes in cloud cover cause abrupt increases ( ‘High Light pulse’ ) and decreases ( ‘Shade pulse’ ) in sunlight ( Petty and Weidner , 2017 ) ( Figure 1B ) . Using a set of programmable warm white LED arrays ( Materials and methods , Construction of light apparatus and Calibrating light conditions ) for illumination , in all experiments we grew cells for 12 hr in either a Clear Day condition that peaked at 600 μmol photons m-2 s-1 or a continuous Low Light condition of 50 μmol photons m-2 s-1 ( Figure 2A , top panel ) followed by 12 hr of darkness for at least two days to acclimate and synchronize the cells before measurement . Note that the Low Light condition used here differs from the Constant Light condition ( often denoted as LL in the literature; Figure 1A , dashed navy blue line ) in that the cells are exposed to more naturally-relevant 12 hr light-12 h dark days ( LD ) . Cultures grown under the Clear Day condition adjusted their pigment content after two days of exposure to the Clear Day condition ( Figure 2—figure supplement 1 ) . Further , cells acclimated to the Clear Day conditions grew approximately twice as fast as Low Light acclimated cultures at midday , 6 hr after dawn ( Figure 2—figure supplement 1 ) . These data indicate that S . elongatus PCC7942 is capable of acclimating to the higher light intensities of the Clear Day condition and thus that the intensities used in our measurements are relevant for this strain . To determine whether a natural light profile affects circadian output , we compared genome-wide gene expression in Clear Day conditions versus Low Light conditions using RNA sequencing ( Figure 2A , Setup , arrows indicate sampling ) . We acclimated cultures in their respective condition for 2 light/dark cycles , and sampled them ( arrows ) over the next ( third ) light period ( Figure 2A , Setup ) . We focused our analysis on a set of high amplitude circadian genes that show oscillatory expression under Constant Light conditions ( Figure 2—figure supplement 2; see Materials and methods , Definition of circadian genes ) . The Low Light condition ( Figure 2B , upper panel ) reproduces the expression profile previously observed under Constant Light conditions ( Figure 2—figure supplement 2 ) . However , in the Clear Day condition 159 of the 281 dusk genes were expressed at least two fold higher after midday compared to Low Light , demonstrating light-dependent expression . Dawn genes show the opposite behavior — they have higher expression at midday under Clear Day conditions , although this trend is less pronounced ( Figure 2—figure supplement 3 ) . Taken together , Clear Day conditions significantly influence the expression dynamics of almost all circadian genes , with the strongest effects on dusk genes . To look more closely at how the Clear Day condition affects the dusk genes , which are the primary regulatory targets of the clock , we analyze the gene expression dynamics of the representative dusk gene Synpcc7942_1567 . Under Low Light conditions , Synpcc7942_1567 exhibits an increase in expression from dawn to dusk , reaching a plateau by 8 hr after dawn ( Figure 2C , solid black line ) . Under Clear Day conditions , however , the expression of this gene remains low through the midday peak of light intensity ( Figure 2C , solid magenta line; 4–8 hr after dawn ) , and its expression sharply increases just prior to dusk as light intensity decreases , reaching maximal expression just as the dark period begins . This delayed pattern of gene expression can be seen in almost all dusk genes ( Figure 2B; Synpcc7942_1567 indicated with arrows ) . Thus Clear Day conditions significantly alter the dynamics and amplitude of dusk gene expression to peak just before dusk . The delay of dusk gene expression likely enables cyanobacteria to switch to glycogen breakdown only when absolutely necessary so that they can survive the extended darkness of night . The two glycogen breakdown genes , glgP and glgX , are both light-dependent dusk genes that strongly peak in Clear Day at dusk , while glgC , which codes for the rate limiting enzyme of glycogen synthesis , is a dawn gene whose expression is higher in Clear Day conditions compared to Low Light ( Figure 2—figure supplement 4 ) . These gene expression dynamics would favor both the maintenance of glycogen synthesis until the end of the day and a delay in the activation of glycogen breakdown until just before it is required at nighttime , in agreement with predictions from metabolic modeling during the same Clear Day conditions used here ( Reimers et al . , 2017 ) . Thus , environmental conditions are integrated into the output of the circadian clock to potentially optimize resource allocation in naturally-relevant diurnal cycles , as recently suggested ( Reimers et al . , 2017 ) . Remarkably , though in both light conditions the cells experience 50 μmol photons m-2 s-1 at the end of the day just before night , light-dependent dusk genes have substantially higher expression in the Clear Day conditions relative to the Low Light conditions ( Figure 2B–C ) . Indeed , 95/281 dusk genes were expressed at least three fold higher in Clear Day relative to Low Light at 12 hr after dawn . This strong activation of dusk genes occurs concomitant with the decrease in light intensity during Clear Day that mimics Sunset , which hinted that changes in light intensity affect activation of dusk genes as opposed to absolute light intensity levels . Dusk gene expression could thus happen ‘just-in-time’ before the sustained darkness of nighttime regardless of the seasonal timing of Sunset . To test whether changes in light intensity are a key factor controlling the expression of circadian genes , we exposed cells to a High Light pulse or a Shade pulse and measured genome-wide gene expression using RNA sequencing . We grew cultures in either Low Light or Clear Day conditions for three days ( Figure 3A–B , Setup ) . On the fourth day at 8 hr after dawn , when RpaA is most active , we exposed the cells to a High Light pulse ( Figure 3A ) or a Shade pulse ( Figure 3B ) for 1 hr before returning to the original condition . We sampled the cells before , during , and after the perturbation ( Figure 3A–B , Setup , arrows ) . The expression of dusk genes rapidly changed in a direction opposite to the change in light intensity ( Figure 3C , all dusk genes; Figure 3E , example dusk gene; Figure 3D , all dusk genes; Figure 3F , example dusk gene ) , as expected from the effects of the decrease in light intensity at Sunset of the Clear Day condition on circadian gene expression ( Figure 2B–C ) . A large subset of dusk genes were affected by the light pulses , with 105/281 repressed by at least three fold by the High Light condition , and 136/281 induced by at least three fold by the Shade condition . Further , many genes responded rapidly and changed in expression at least three fold after just 15 min into the pulse ( 75/281 repressed by High Light , 79/281 induced by Shade ) . When cultures were restored to their original condition ( High Light to Low Light , Figure 3C , E; Shade to Clear Day , Figure 3D , F ) , dusk gene expression quickly reverted to a level comparable to that before the pulse . Thus , light-induced changes in dusk gene expression are reversible and responsive to successive shifts in light availability . Dawn gene expression showed the opposite behavior of dusk genes , albeit with less dramatic changes ( Figure 3—figure supplement 1 ) . Hence , decreases in light intensity favor the expression of dusk genes ( Sunset in Clear Day , Figure 2; Clear Day to Shade and High Light to Low Light , Figure 3 ) , while increases in light favor the expression of dawn genes ( midday peak in Clear Day , Figure 2—figure supplement 3; Shade to Clear Day and Low Light to High Light , Figure 3—figure supplement 1 ) . Given the more substantial effects of light on dusk gene expression , we focus on these genes for the remainder of the manuscript . To cause these reversible changes in the mRNA levels of dusk genes , changes in light intensity must affect either the transcription and/or the degradation of dusk gene mRNAs . We reasoned that changes in transcription would manifest as differences in the amount of RNA polymerase ( RNAP ) localized to dusk genes . To determine whether changes in light intensity regulate the recruitment of RNAP to dusk gene promoters , we performed chromatin immunoprecipitation followed by high-throughput sequencing ( ChIP-seq ) of RNAP in cells immediately before the High Light or Shade pulse ( 8 hr after dawn in Low Light or Clear Day ) , and then again 15 or 60 min following the start of the pulse . Changes in RNAP enrichment upstream of dusk genes correlated with changes in downstream dusk gene expression ( Figure 3G , H; Figure 3—figure supplement 2 ) . Thus , changes in light affect RNAP recruitment to dusk gene promoters , suggesting that light conditions substantially affect the rates of transcription of dusk gene mRNAs . Because mRNAs in bacteria have very short steady state half lives ( Chen et al . , 2015; Hambraeus et al . , 2003; Salem and van Waasbergen , 2004 ) , we argue that changes in transcription rates of dusk gene mRNAs are sufficient to lead to the rapid changes in dusk gene mRNA levels given a fast basal degradation rate , though we cannot rule out that changes in light may affect the rates of degradation of some mRNAs . These results point to a potential interaction between sunlight and signaling pathways upstream of RNAP . We next explored how the observed changes in dusk gene expression in the presence of natural light fluctuations ( Figures 2 and 3 ) could be achieved via gene regulatory mechanisms . Given the strong dependence of dusk gene expression on RpaA∼P levels under Constant Light conditions ( Figure 1A , [Markson et al . , 2013] ) and the drastic change in dusk gene expression dynamics under our dynamic light conditions ( Figure 2B , C; Figure 3C–F ) , we hypothesized that light conditions alter RpaA∼P dynamics to alter dusk gene expression . However , levels of RpaA∼P increased from dawn to dusk similarly in cells grown in either Low Light or Clear Day conditions , and abrupt changes in light intensity did not affect these dynamics . ( Figure 4A , B; Figure 4—figure supplement 1 ) . Thus , these natural light fluctuations do not affect the phase of the Kai PTO nor the control of RpaA∼P levels by the Kai PTO ( Figure 4E ) . These data demonstrate that the regulation of dusk genes is de-coupled from RpaA∼P levels under dynamic light conditions , and light must affect dusk gene expression downstream of RpaA∼P . Interestingly , ChIP-seq showed that light intensity fluctuations alter RpaA∼P binding upstream of dusk genes ( Figure 4C; Figure 4—figure supplement 2 ) in conjunction with RNAP binding upstream of the same gene ( Figure 4D; Figure 4—figure supplement 3 ) . The binding of RpaA∼P and RNAP correlated with changes in downstream dusk gene expression ( Figure 4C; Figure 4—figure supplement 2 ) . Thus , light fluctuations control the binding of RpaA∼P and RNAP to promoters , suggesting that light-induced changes in the binding of these factors may modulate the activation of dusk gene expression ( Figure 4E ) . Interestingly , RpaA regulation at a small number ( ∼10 ) of promoters including that of kaiBC is not substantially affected by light intensity ( Figure 4C , D - points around origin; Figure 4—figure supplement 4 ) , demonstrating that the light-dependent regulation of RpaA binding is locus-specific . The KaiABC clock regulates RpaA∼P levels independent of changes in light intensity ( Figure 4A , B ) , and kaiBC gene expression dynamics do not substantially change in the Clear Day conditions compared to the Low Light condition ( Figure 4—figure supplement 4H–K ) . Hence , the stabilizing PTO/transcription-translation feedback loop circadian circuit is robust to natural fluctuations in sunlight . The circadian clock can thus control gene expression independent of environmental changes at select promoters . It is possible that RpaA∼P binding to some promoters is dependent on the association of RNAP with that promoter . As such , regulation that affects RNAP binding to a specific promoter , such as that by sigma factor activity ( Gruber and Gross , 2003 ) , could affect RpaA∼P binding to select promoters . Our analysis so far has established that the previous model for the regulation and expression of circadian genes in Constant Light conditions ( Figure 1A ) becomes more complex in natural environmental conditions , suggesting the involvement of other pathways . Thus , we next asked whether RpaB plays a role in controlling light-dependent expression of circadian genes . We observed that levels of RpaB∼P changed rapidly in a direction opposite to the change in light ( Figure 5A , B; Figure 5—figure supplement 1 ) , suggesting that light affects RpaB activity through its phosphorylation state ( Figure 5E ) . Levels of RpaB∼P decreased ∼3 . 1 fold after 15 min in the High Light pulse , and increased ∼1 . 9 fold after 15 min in the Shade pulse , concomitant with the rapid repression and induction of many dusk genes ( Figure 3 ) . Further , RpaB∼P levels increased ∼1 . 7 fold between 10 and 12 hr after dawn in the Clear Day condition concomitant with the decrease in light during Sunset and the strong induction of many dusk genes ( Figure 2 ) . This strong correlation between RpaB∼P levels and the expression of dusk genes under dynamic light conditions ( also compare Figure 3E , F to Figure 5A , B ) suggests that RpaB∼P acts as an activator of dusk gene expression . Indeed , using ChIP-seq we found that RpaB binds upstream of a large subset of dusk genes ( 42/281 dusk genes , Figure 5—source data 2 ) . RpaB binding upstream of these genes shifts after rapid changes in light ( Figure 5C; Figure 5—figure supplement 2 ) , correlating with changes in RpaB∼P levels ( Figure 5A , B ) , RNAP binding upstream of the same gene ( Figure 5D; Figure 5—figure supplement 3 ) , and downstream dusk gene expression ( Figure 5C; Figure 5—figure supplement 2 ) . These results suggest that RpaB∼P directly activates the expression of many dusk genes by binding to promoters with RNAP ( Figure 5E ) . Thus , changes in sunlight can regulate dusk genes by adjusting RpaB∼P levels ( Figure 5E ) . Because RpaA and RpaB bind only a subset of light-responsive dusk genes ( Figure 6A , B ) , additional factors must be involved in controlling light-responsive dusk gene expression . Sigma factors are sequence-specific RNAP subunits which regulate gene expression in bacteria ( Gruber and Gross , 2003 ) . Interestingly , RpaA , RpaB , and RNAP bind to the promoters of three sigma factor genes ( Figure 6C; Figure 6—figure supplement 1A–C ) . The binding of RpaA , RpaB , and RNAP to these promoters shifts in conjunction after abrupt changes in light intensity , correlating with light-responsive changes in expression of these genes ( Figure 6D; Figure 6—figure supplement 1D–F ) . These sigma factor genes show light-dependent dusk gene expression patterns ( Figure 6—figure supplement 1G–L ) that mirror those of the larger group of dusk genes ( Figures 2 and 3 ) , suggesting that these sigma factors could regulate the expression of other dusk genes . Thus , RpaA and RpaB may indirectly regulate the expression of non-target dusk genes by controlling the circadian and light-responsive expression of sigma factor genes ( Hanaoka et al . , 2012 ) , similar to how RpaA drives all dusk gene expression in Constant Light conditions by binding to a subset of dusk genes ( Markson et al . , 2013 ) . It is also possible that changes in light intensity affect dusk gene expression in a manner independent of RpaA∼P and RpaB∼P regulation . For instance , global growth-rate-dependent gene regulatory mechanisms such as the stringent response ( Scott et al . , 2010; Burnap , 2015; Ryals et al . , 1982; Hood et al . , 2016 ) likely cause some of the light-dependent changes in circadian gene expression due to unavoidable differences in the growth rate in different light conditions ( Figure 2—figure supplement 1 ) . We have defined a regulatory picture in which changes in light intensity affect the activity of RpaA and RpaB to control the expression of dusk genes . However , light affects RpaA activity in complex and promoter-specific ways . Additionally , light-dependent regulation in addition to that mediated by RpaA and RpaB may control dusk gene expression in response to environmental perturbations . Still , despite the apparent complexity of regulation of dusk genes in response to light fluctuations , the expression of almost all dusk genes show strikingly regular dynamics ( Figures 2 and 3 ) . Furthermore , the activity of RpaA and RpaB at a subset of promoters ( especially those of sigma factor genes ) could lead to pervasive and coordinated changes in the expression of other dusk genes . Hence , we reasoned that mathematical models ( Alon , 2006 ) of RpaA and RpaB activity might effectively describe the regulatory circuits underlying the dynamics of large groups of dusk genes . Such an approach would enable an understanding of the basic principles of interaction between circadian gene expression regulation with light-dependent regulation without needing to describe all underlying molecular mechanisms . We find that dusk genes collectively display a small number of responses to changes in environmental light intensity . Using k-means clustering of the gene expression dynamics from our different light profiles ( Figures 2 and 3 ) , as well as from perturbations of RpaA ( Figure 2—figure supplement 2 [Markson et al . , 2013] ) , we identify three major groups of dusk genes ( 35–80 genes , see Figure 7—source data 1 for full lists ) which show distinct and coordinated changes in gene expression over circadian time and in response to changes in light intensity ( Figure 7; Figure 7—figure supplement 1 ) . Under Constant Light conditions , all three clusters are activated by RpaA∼P but display distinct activation dynamics from dawn to dusk ( Figure 7—figure supplement 1B ) that are mirrored under our Low Light conditions ( Figure 7—figure supplement 1A ) . We named the clusters the Early , Middle , and Late dusk genes based on the order of activation . The Shade pulse and Sunset in the Clear Day condition have differing effects on the expression of each of the major dusk gene clusters . Early dusk gene expression rapidly increases in response to Shade , but during Sunset plateaus at ∼1/2 of the maximal gene expression reached in Shade ( Figure 7A ) . Conversely , the Late gene cluster responds most strongly to Sunset in Clear Day conditions but has a mild increase in expression in Shade relative to the Early and Middle dusk genes ( Figure 7C ) . In contrast , the Middle gene cluster is induced to a similar magnitude by both Shade and Sunset ( Figure 7B ) . Shade and Sunset represent similar light changes that occur at different times of day ( afternoon and dusk , respectively ) . As such , the Early and Late dusk gene clusters are differentially induced by a decrease in light intensity depending on the time of day in which it occurs . This circadian effect on the intensity and dynamics of a response to environmental change is a signature of circadian gating ( Hotta et al . , 2007; Greenham and McClung , 2015 ) . Though circadian gating has been observed ( e . g . , [Belbin et al . , 2017] ) and modeled without any knowledge of the transcriptional regulation ( Dalchau et al . , 2010 ) in plants , it remains unclear what gene regulatory circuits are sufficient to explain such behavior . At present there is no mechanistic model to explain the differential response of these clusters to circadian regulation and changes in sunlight . Given that there are unknown regulators involved in circadian gene expression ( Figure 6A , B ) , and because it is not possible to exhaustively test all possible models of regulation of dusk gene expression , we sought to construct the simplest models that can describe the expression dynamics of these clusters using a phenomenological modeling approach . Such models can be used to highlight regulatory architectures that are sufficient to recapitulate the observed gene expression dynamics , as well as direct further mechanistic studies to reveal the underlying molecular details of regulation . Given the clear roles for RpaA∼P and RpaB∼P in activating dusk genes , we asked whether the dynamic expression of the major dusk gene clusters in naturally-relevant light conditions could be described by these variables . We constructed phenomenological models that describe the kinetics of the synthesis and breakdown of an average gene in each of the dusk gene clusters ( Mangan and Alon , 2003 ) ( see Materials and methods , Mathematical modeling ) . The rate of synthesis was the sum of a baseline rate of transcription and a maximal adjustable rate of transcription that could be modulated by the activity of one or more regulators . We described the the effects of a regulator such as RpaA∼P or RpaB∼P using a Hill function , whose shape is determined by the Hill coefficient and the coefficient of activation . We determined how well a model could describe the dynamics of a cluster by fitting it to the Clear Day and Shade pulse data and assuming all parameters could vary freely ( see Materials and methods , Mathematical modeling; Table 1 ) . We began by asking whether levels of RpaA∼P or RpaB∼P ( Figure 8A ) can describe the gene expression dynamics of the major dusk clusters in natural light conditions . We first constructed models in which dusk cluster gene expression is solely dependent on RpaA∼P . Activation by RpaA∼P can recapitulate the ordered activation of the dusk gene clusters through differential coefficients of activation for RpaA∼P , but cannot describe the light-responsive expression of these genes ( RpaA-only models , Figure 8—figure supplement 1A–C; Table 2 ) . Further , activation by RpaB∼P alone cannot describe the dusk gene expression patterns of the clusters ( RpaB-only models , Figure 8—figure supplement 1D–F; Table 2 ) . However , models in which dusk gene expression is a function of BOTH RpaA∼P and RpaB∼P can recapitulate much of the time-of-day and light intensity dependent expression of the Early and Late clusters and nearly all of the expression dynamics of the Middle clusters ( RpaA and RpaB models , Figure 8B–E; Table 2 ) . This suggests that RpaB∼P is a variable which can capture the effects of dynamic light conditions on RpaA∼P activity . The fit parameters for simple joint activation can accommodate indirect activation through downstream regulators like sigma factors and thus do not require direct RpaA/B binding to all genes . Conceptually , our results suggest that transcription factors whose activity track the measured dynamics of both RpaA∼P and RpaB∼P can describe the circadian and light-responsive expression of dusk genes . However , joint activation by RpaA∼P and RpaB∼P predicts that the Early and Late clusters will respond similarly to Shade and Sunset in Clear Day conditions ( Figure 8C , E ) , and thus cannot capture well the circadian gating of these clusters . We reasoned that additional regulatory interactions downstream of RpaA and RpaB , or ‘network motifs’ ( Alon , 2006 ) , could account for the observed gating of the Early and Late clusters . Thus , we constructed models in which dusk cluster gene expression is positively or negatively dependent on the expression of another cluster alongside activation by RpaA∼P and RpaB∼P ( Feedback models , Figure 8F–I; Figure 8—figure supplements 2–4; Table 2 ) . Interestingly , the gating of the Early cluster is recapitulated by a model incorporating an incoherent feedforward loop in which the Late cluster represses Early cluster expression downstream of RpaA∼P and RpaB∼P activation ( Figure 8G; Figure 8—figure supplement 2; Table 2 ) . Further , the gating of the Late cluster is well described by a coherent feedforward loop in which Late cluster expression is dependent on RpaA∼P , RpaB∼P , AND Middle cluster expression levels ( Figure 8I; Figure 8—figure supplement 4; Table 2 ) . Thus , we highlight regulatory schemes downstream of RpaA and RpaB which can generate large time-of-day differences , or circadian gating , in the response to a decrease in light intensity . Our results highlight that the measured dynamics of RpaA∼P and RpaB∼P can account for the dynamics of large groups of clock-controlled genes after environmental changes and suggest regulatory schemes that can diversify gene expression responses downstream of RpaA and RpaB . The models suggested here offer constraints and testable hypotheses to guide future studies of the molecular mechanisms underlying these responses . We show that natural fluctuations in light intensity significantly affect the dynamics of circadian gene expression ( Figures 2 and 3 ) . While previous studies have measured genome-wide gene expression in a single natural light condition ( Waldbauer et al . , 2012 ) , here we compare genome-wide circadian gene expression in several physiologically-relevant conditions , including Clear Day , High Light pulse , Shade pulse , and Low Light , to carefully dissect the effects of light on clock output . Natural light changes most greatly affected a large fraction of the dusk genes ( Figures 2B and 3C , D ) , possibly because most of the direct targets of RpaA are dusk genes ( Markson et al . , 2013 ) . We speculate that the opposing trends we observe in dawn gene expression ( Figure 2—figure supplement 3 and Figure 3—figure supplement 1 ) may in part be due to competition for RNAP between the dusk and dawn genes ( Gruber and Gross , 2003; Mauri and Klumpp , 2014 ) or by growth-rate-dependent mechanisms ( Scott et al . , 2010 ) , as this group of genes contains the primary growth genes . A systematic exploration of the effects of light on circadian genes will be necessary to fully elaborate the contributions of light , clock , and growth rate on circadian gene dynamics . We find that large groups of light-responsive dusk genes are activated by diminished light conditions to different extents depending on the time of day the stimulus is applied . These differences in activation may serve to optimally change metabolism for a given light condition and time of day . The light-responsive dusk genes grouped into three clusters - Early , Middle , and Late - with different activation dynamics during Sunset at the end of the Clear Day versus the Shade pulse in the afternoon ( Figure 7 , see Figure 7—source data 1 for full lists of genes in each cluster ) . Glycogen breakdown genes and the central carbon metabolism genes glyceraldehyde-3-phosphate dehydrogenase and oxalate decarboxylase belong to the Middle dusk genes , which are activated to similar levels by Shade and Sunset ( Figure 7B ) . This suggests that cyanobacteria delay the activation of glycogen breakdown pathways ( Reimers et al . , 2017 ) until just before dusk when grown under Clear Day conditions , but can transiently activate these genes in response to Shade to access alternate energy reserves if necessary . Interestingly , genes encoding pyridine nucleotide transhydrogenase , which reversibly converts NADH to the NADPH required for electron transport , belong to the Late cluster and are strongly activated only by Sunset and not afternoon Shade ( Figure 7C ) . Such a response might delay the adjustment of the relative levels of NADH/NADPH until only when absolutely needed at night , when NADPH is potentially important for defense against reactive oxygen species ( Diamond et al . , 2017 ) . The cytochrome c oxidase genes belong to the Early cluster , which respond more intensely to Shade than to Sunset ( Figure 7A ) . This enzyme is essential for preventing photodamage in response to rapid changes in light intensity ( Lea-Smith et al . , 2013 ) ; such changes are not expected to occur during the night , where it serves solely as the terminal electron acceptor for respiration . More generally , the genome-wide gene expression dynamics measured here qualitatively agree with predictions from a whole-cell model of S . elongatus that assumed optimization of growth ( Reimers et al . , 2017 ) . To resolve how the circadian and light-dependent transcriptional changes effect these metabolic changes , future studies must measure enzyme levels and metabolic fluxes under fluctuating light conditions . While light does not alter the post-translational oscillator/transcription-translation feedback loop circadian circuit , it regulates the activation of dusk genes via RpaA∼P promoter binding ( Figure 4 ) and RpaB promoter binding through its phosphorylation state ( Figure 5 ) at a subset of dusk genes . RpaA binding upstream of its target genes under dynamic light conditions ( Figure 4C ) correlates with the changes in expression of non-RpaA target genes ( Figure 6B ) . Thus , RpaA∼P may remain the ‘master regulator’ of circadian gene expression whose promoter binding activity is altered by other molecular factors that encode information about the environment , such as RpaB . Previous work suggested that changes in RpaB∼P phosphorylation would alter RpaA∼P levels through competition with the enzymes that control RpaA∼P levels ( Espinosa et al . , 2015 ) . However , we find that RpaA∼P levels remain constant ( Figure 4A , B ) under conditions in which RpaB∼P levels change substantially ( Figure 5A , B ) , arguing that RpaB∼P does not influence RpaA∼P levels . RpaB∼P might influence RpaA∼P binding at promoters where both proteins bind ( Figure 6—figure supplement 1 ) as previously suggested ( Hanaoka et al . , 2012 ) , and joint control of sigma factors by RpaA and RpaB could feedback to affect RpaA binding at select promoters . Still , the question of how light changes RpaA∼P binding in a promoter-specific way remains unclear . We define a clear role for the stress-responsive transcription factor RpaB as a transcriptional activator of a large subset of dusk genes ( Figure 5E ) . Further , we demonstrate that decreases in light intensity like a Shade Pulse lead to increases in RpaB∼P levels to allow RpaB to activate the expression of genes . This result shows that RpaB acts in scenarios beyond its previously appreciated role in High Light stress ( Kato et al . , 2011; Seki et al . , 2007; Hanaoka and Tanaka , 2008; López-Redondo et al . , 2010 ) . RpaA∼P and RpaB∼P might cooperate to indirectly regulate the expression of most light-responsive dusk genes by jointly controlling the expression levels of multiple sigma factors ( Figure 6—figure supplement 1 ) ( Hanaoka et al . , 2012 ) . However , our attempts to cleanly perturb RpaB activity to further explore its role as a regulator of dusk genes were unsuccessful , in part because the rpaB gene is essential ( López-Redondo et al . , 2010 ) . The role of sigma factors in this network of regulation , while strongly implied , remains ambiguous and attempts to assess this role using genetic deletion of sigma factors yielded inconclusive results . More subtle approaches such as anchors away ( Haruki et al . , 2008 ) might allow perturbation experiments that clearly explicate the roles of the sigma factors and RpaB in mediating circadian gene expression . Although complex molecular mechanisms underlie the light-responsive expression of dusk genes , we demonstrate that phenomenological models effectively describe the differential activation of large groups of dusk genes to afternoon Shade and Sunset . These models suggest that transcription factors with the dynamics of RpaA∼P and RpaB∼P ( Figure 8A ) are sufficient to reproduce much of the activation of the Early , Middle , and Late clusters in response to a Shade pulse in the afternoon or Sunset just before night ( Figure 8C–E ) . Our models suggest that additional feedback from the other gene clusters may be necessary to achieve the extent of circadian gating observed for the Early and Late clusters ( Figure 8G–I ) . Our models suggest that interactions between the major dusk clusters can diversify the responses of these clusters to signals from RpaA and RpaB . Regulatory interactions between the sigma factors RpoD6 , RpoD5 , and SigF2 ( Figure 6—figure supplement 1 ) , which belong to the Early , Middle , and Late clusters , respectively , could generate feedback downstream of RpaA and RpaB similar to that in our models ( Figure 8—figure supplements 2–4 ) to generate the diverse responses of the dusk clusters to light conditions . However , feedback could also come from other sources with similar dynamics to the cluster expression levels . Indeed we could not simultaneously fit our models to all four light conditions , likely because of global growth-rate-dependent differences between the Low Light and Clear Day conditions . Thus , complete modeling of transcription dynamics of light-dependent dusk genes likely requires explicitly including the effects of metabolism and growth on gene expression ( Reimers et al . , 2017; Burnap , 2015; Scott et al . , 2010 ) . RpaB and its cognate upstream histidine kinase NblS ( van Waasbergen et al . , 2002 ) have been implicated in a variety of stress responses ( Marin et al . , 2003; Mikami et al . , 2002; Shoumskaya et al . , 2005 ) , which suggests that the mechanisms and regulatory circuits defined here may apply to other environmental changes such as temperature or osmolarity . The requirement of RpaB for mediating the environmental response of circadian genes suggests that the circadian circuit coevolved with RpaB to optimize responses to predictable and unpredictable changes in the environment and motivates the further exploration of the interaction between light and circadian rhythms in S . elongatus . Resolution of this interaction and subsequent integration into whole cell models of cyanobacterial growth ( Burnap , 2015; Westermark and Steuer , 2016 ) will help to explain the fitness benefits of the circadian clock ( Johnson and Egli , 2014 ) and optimize synthetic biology efforts to engineer cyanobacteria to produce useful compounds ( Ducat et al . , 2011 ) from the constantly changing sunlight in nature . All high throughput sequencing data is available from the Gene Expression Omnibus with the accession number GSE104204 . Most experiments were conducted in a pure wildtype background of Synechococcus elongatus PCC7942 ( ATCC catalog number 33912 , RRID:SCR_001672 ) . For RNAP ChIP experiments , we used a strain in which the β′ subunit of RNA polymerase ( Synpcc7942_1524 , gene info available through Cyanobase , RRID:SCR_007615 ) was C-terminally tagged with a 3x FLAG epitope ( a gift from Ania Puszynska ) . To make this strain , wildtype S . elongatus was transformed with a plasmid encoding the Synpcc7942_1524 gene with sequence encoding a 3X GS linker and a 3X FLAG epitope inserted before the stop codon , targeted to insert at the native locus of the gene . A downstream kanamycin resistance cassette was used for selection . This plasmid is available through Addgene with the ID 102337 . Two different clones of this strain , EOC398 and EOC399 , were confirmed by sequencing colony PCR fragments that amplified the modified regions of the gene , and the presence of the tagged subunit was confirmed with Western blotting . To grow the cyanobacteria in different light profiles , we constructed an apparatus to control the intensity of four high powered LED arrays ( parts list in Table 3 , p . 2 ) . ‘Warm white’ LED arrays ( ∼1 in . x 1 in . , Bridgelux ) were chosen because of maximal overlap with the phycobilisome absorption spectrum . An LED array was mounted on a heatsink ( Nuventix ) and powered by a Flexblock LED driver ( LEDdynamics ) wired in the ‘boost only’ configuration ( Table 4 , p . 3 ) . The intensity of the LEDs was controlled by varying the voltage input into the DIM line of the Flexblock between 0 and 10 V . We used a digital potentiometer ( AD7376 , Analog Devices ) as a controllable 10 V source . The voltage output of the digipot was controlled via serial peripheral interface with an Arduino Uno board ( Arduino ) ( see Table 5 , p . 4 ) . Each LED array was controlled separately , and a single array was sufficient to grow a single 750 mL culture of S . elongatus . All wires carrying substantial currents from the main power supply to the LED arrays were rated 18 AWG , and all other wires were rated 22 AWG . The relatively low voltage of the main power supply ( 18 V ) is essential for being able to turn off the LED arrays completely . A single LED was mounted to shine perpendicular to the ground and isolated from other light sources . A single 750 mL cyanobacterial culture in a 150 cm2 BD Falcon Tissue culture flask ( Fisher Scientific ) was placed beneath the LED , tilted such that the broad face of the culture was almost perpendicular to the incoming light . Each LED was calibrated by passing a known voltage input to the LEDs and recording the intensity of the light in μmol photons m-2 s-1 at the position of the surface of the culture directly beneath the LED using a LI-COR LI-250A light meter equipped a quantum sensor . To access a greater dynamic range of light intensity values , we calibrated the lights to give light intensity values at either of two distances from the light source — raised towards the lights to access higher light intensities , or lowered away from the lights to access lower light intensities . To define the Clear Day conditions , we used light intensity values measured by the Ground-based Atmospheric Monitoring Instrument Suite , Rooftop Instrument Group on March 23rd , 2013 ( Figure 1B , dark blue line , [Petty and Weidner , 2017] ) . We used this light intensity profile to define the rate of change of light intensity in our Clear Day condition , with a maximal light intensity of 600 μmol photons m-2 s-1 . This intensity is consistent with measurements of light intensity in aquatic environments ( Waldbauer et al . , 2012 ) , while also offering an order of magnitude difference in intensity compared to the Low Light condition , which was a constant 50 μmol photons m-2 s-1 . The Shade pulse condition was defined by dividing the intensity value of our Clear Day profile by 10 fold between 8 and 9 hr after dawn . The High Light pulse was defined as the intensity of the Clear Day condition between 8 and 9 hr after dawn . Low Light cultures were grown continuously at 50 μmol photons m-2 s-1 . We generated the dynamic changes in light intensity of our conditions by changing the intensity of the LED every three minutes by passing the calibrated voltage value corresponding to the appropriate light intensity of our defined profile . The light intensity values of the Low Light and Clear Day conditions are listed in Figure 2—source data 1 , and the High Light and Shade pulse values are listed in Figure 3—source data 1 . After the 12 hr light profile , the LEDs were turned off for 12 hr during the dark period . Cultures were grown semi-turbidostatically ( OD750 maintained at 0 . 3 ) with periodic dilution in BG-11M media supplemented with 10 mM HEPES pH 8 . 0 at 30 ∘C , continuously bubbled with 1% CO2 in air , and shaken at 25 rpm in an enclosure impermeable to room lighting . Cells were not grown with antibiotics during the course of the experiment . Recombinant RpaA was purified as previously described ( Takai et al . , 2006 ) . To purify recombinant RpaB , we cloned the rpaB gene ( Synpcc7942_1453 , gene info available through Cyanobase , RRID:SCR_007615 ) into the pET48-b + plasmid ( Novagen ) and overexpressed Trx-His-tagged RpaB in Novagen Tuner ( DE3 ) competent cells carrying this plasmid by adding 300 μM IPTG to mid-log phase cultures . RpaB was purified from cell lysate using Ni-NTA chromatography as described previously ( Gutu and O'Shea , 2013 ) . The Trx-His tag was cleaved from RpaB and removed using a subsequent Ni-NTA step as described ( Gutu and O'Shea , 2013 ) . Purified , cleaved RpaB was dialyzed into a buffer containing 20 mM HEPES-KOH , pH 8 . 0 , 150 mM KCl , 10% w/v glycerol , and 1 mM DTT . Protein concentration was measured with the Pierce BCA assay , and aliquots were flash frozen and stored at −80°C . Anti-RpaB serum was generated by immunization of two rabbits with purified RpaB by Cocalico Biologicals ( Reamstown , PA ) . RpaA- and RpaB-conjugated Affigel 10/15 resin ( Bio-Rad ) was prepared following manufacturer’s instructions as described previously ( Gutu and O'Shea , 2013 ) . Anti-RpaB serum was first passed over an RpaA-conjugated resin and the flowthrough collected to subtract cross-reacting antibodies . Anti-RpaB antibodies were then purified from the flowthrough using an RpaB-conjugated resin as described previously ( Gutu and O'Shea , 2013 ) . The same process was repeated to purify anti-RpaA antibodies using rabbit serum described previously ( Markson et al . , 2013 ) , passing the serum over an RpaB-conjugated resin and purifying with an RpaA-conjugated resin . No cross reactivity of the purified anti-RpaA and anti-RpaB antibodies for the opposite regulator was detected via ELISA assay . Ten mL of cyanobacterial culture with OD=7500 . 3 were collected on cellulose acetate filters and flash frozen prior to storage at −80 ∘C . Cell lysates for Western blotting were prepared from the collected cells as described previously ( Markson et al . , 2013 ) . Equal amounts of cell lysate ( 10–15 μg ) were resolved on Phos-tag acrylamide gels ( Wako Laboratory Chemicals ) and transferred to nitrocellulose membranes as described previously ( Gutu and O'Shea , 2013 ) . Membranes were probed with 1/5000 dilution of purified anti-RpaA and anti-RpaB antibody . RpaA blots were then incubated with goat anti-rabbit HRP-conjugated secondary antibody and developed using the Pierce Femto chemiluminescence kit . The exposed blots were imaged with an Alpha Innotech Imaging station . RpaB blots were incubated with Goat anti-Rabbit Westerndot 585 antibody ( RRID:AB_2556786 ) and imaged with a Typhoon Imager . The intensities of the bands corresponding to unphosphorylated and phosphorylated RpaA/B were quantified using Imagequant software ( GE Healthcare Life Sciences , RRID:SCR_014246 ) using rubber band background subtraction . The percent of RpaA ( or RpaB ) phosphorylated was quantified as the intensity of the RpaA∼P band divided by the sum of the intensities of the RpaA and RpaA∼P bands , multiplied by 100 . Values reported in Figures 4A , B and and 5A , B represent the average of two separate measurements from replicate Western blots , with error bars displaying the range of the measured values ( See Figure 4—source data 1 , and Figure 5—source data 1 for raw data from the replicate experiments ) . The trends seen were reproducibly observed between separate biological replicates of the light condition time courses . Twenty-five mL of cyanobacterial culture with OD=7500 . 3 were collected on cellulose acetate filters and flash frozen prior to storage at −80 ∘C . Cells were resuspended in RNAprotect Bacteria reagent ( Qiagen ) , and 1/3 of the cells were resuspended in a buffer containing 15 mg/mL lysozyme , 10 mM Tris-Cl , 1 mM EDTA pH 8 , and 50 mM NaCl and incubated for 10 min . RNA was purified from the lysed cells using the Qiagen RNeasy Mini Kit . Ribosomal RNA was depleted from 1 . 25 μg of purified RNA using the Ribo-Zero bacteria rRNA removal kit ( Illumina ) . Strand-specific RNAseq libraries were prepared from the depleted RNA using the Truseq Stranded mRNA Sample prep kit ( Illumina ) and sequenced on an Illumina HiSeq 2500 machine by the Bauer Core Facility at the Harvard FAS Center for Systems Biology . Sequencing reads were aligned to the S . elongatus genome using Bowtie ( RRID:SCR_005476 ) as described previously ( Markson et al . , 2013 ) , with samples averaging 8 million aligned reads . We quantified expression of a gene by counting the number of aligned sequencing reads corresponding to the appropriate strand between the start and stop of each gene ( gene info obtained from Cyanobase , RRID:SCR_007615 ) , and normalized these values between all samples from the light conditions in this work using median normalization , followed by dividing the median normalized read count value by the length of the open reading frame of the gene , as described previously ( Anders and Huber , 2010; Markson et al . , 2013 ) . The time course and RNA sequencing was repeated twice for two biological replicates ( data available in Figure 2—source data 1 and Figure 3—source data 1 ) . The data plotted in this work are from replicate 2 , and the trends observed are reproduced in both biological replicates . We defined a subset of previously identified circadian genes on which to focus our analysis . We began with a list of 856 previously described reproducibly circadian genes ( Markson et al . , 2013; Vijayan et al . , 2009 ) . We next required that these genes have a Cosiner amplitude ( Kucho et al . , 2005 ) of greater that 0 . 15 under Constant Light conditions ( Vijayan et al . , 2009 ) . We also required that the gene display expression of at least one read per nucleotide in at least one time point of the RNA sequencing experiments in this study . These filters produce a list of 450 high confidence circadian genes . We noted that genes classified as dawn ( class 2 ) and dusk ( class 1 ) genes under Constant Light conditions ( Vijayan et al . , 2009 ) showed maximal expression at a different time of day under our Low Light conditions , while the relative ordering of genes by Cosiner phase ( Kucho et al . , 2005 ) from Constant Light conditions ( Vijayan et al . , 2009 ) was preserved . As such , we redefined dawn genes as those genes with a phase of 40∘ to 189∘ under Constant Light conditions ( Vijayan et al . , 2009 ) , and dusk genes as those with a phase of 190∘ to 360∘ and 0∘ to 39∘ , as determined by the Cosiner algorithm ( Kucho et al . , 2005 ) . These definitions produce a list of 169 high confidence dawn genes , and 281 high confidence dusk genes . The expression of our redefined circadian genes under Constant Light conditions is plotted in Figure 2—figure supplement 2 . The list of high confidence circadian genes and high confidence class assignments is available in Figure 2—source data 1 and Figure 3—source data 1 . One hundred and twenty mL of OD750 0 . 3 cyanobacterial culture were removed and crosslinked with 1% formaldehyde at 30 ∘C for 5 min in front of a light source . Crosslinking was quenched with 125 mM glycine . Crosslinked cells were washed twice with phosphate buffered saline , pelleted , and flash frozen prior to storage at −80 ∘C . Pellets were resuspended in 1 mL of BG-11M supplemented with 500 mM L-proline and 1 mg/mL lysozyme and incubated at 30 ∘C for 1 hr to digest the cell wall . Cells were collected and resuspended in a Lysis buffer ( 50 mM HEPES pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% sodium deoxycholate , and 1x Roche Complete EDTA-free Protease Inhibitor Cocktail ) prior to shearing in a Covaris E220 Adaptive Focus System ( Peak Incident Power = 175; Duty Factor = 10%; Cycles per burst = 200; Time = 160 s ) . The lysates were cleared via centrifugation , and concentration was determined via the Pierce BCA Assay . For a given pulldown , 800 μg of lysate was incubated overnight at 4 ∘C in 500 μL of lysis buffer with 8 μg of anti-RpaA , anti-RpaB , or FLAG M2 mouse monoclonal antibody ( Sigma-Aldrich ) for RNAP pulldowns . A mock pulldown was carried out in which equal amounts of lysate from every time point of the time course ( Shade 0 , 15 , 60 min , High Light 0 , 15 , 60 min ) in a total of 800 μg was incubated with 8 μg of rabbit Igg . Next , 35 μL of Dynabeads protein G ( Thermo Fischer Scientific ) equilibrated in lysis buffer were added and the sample was incubated with mixing for 2 hr at 4 ∘C . The beads were washed and DNA was eluted and purified as described previously ( Markson et al . , 2013 ) . Sequencing libraries were prepared from the purified ChIP DNA using the NEBNext Ultra II DNA Library Prep Kit ( New England Biolabs , Ipswich , MA ) . Libraries were sequenced on an Illumina HiSeq 2500 instrument by the by the Bauer Core Facility at the Harvard FAS Center for Systems Biology . We created sequencing libraries of ChIP experiments from two separate biological repeats of the time course experiment . Reads were aligned to the S . elongatus genome using Bowtie ( RRID:SCR_005476 ) as described previously ( Markson et al . , 2013 ) , resulting in an average of 3 million aligned reads for replicate 1 , and 5 million aligned reads for replicate 2 . The aligned read data per genomic position was smoothed with a Gaussian filter ( window size = 400 base pairs , standard deviation = 50 ) . Each data set was normalized to the Mock ChIP-seq experiment and peaks which were significantly enriched above the Mock were identified in each data set using a previously described ( Markson et al . , 2013 ) custom-coded form of the Peak-seq algorithm ( Rozowsky et al . , 2009 ) . Within each replicate time course for a given protein , we compiled a list of peaks which were enriched at least 3 . 5 fold over the Mock experiment at the position of highest ChIP signal . Finally , we required that a peak be detected in both replicates for it to be considered . This analysis generated 114 RpaA peaks , 218 RpaB peaks , and 451 RNAP peaks . To calculate enrichment for a peak , we determined the ChIP signal at a given time point at the genomic position of the highest ChIP signal detected for that peak and divided this by the value of the Mock experiment at that position . The data plotted in this manuscript are from replicate 2 , but all trends hold in replicate 1 . We assigned a gene as a target of a peak if: ( i ) the start codon of the gene was within 500 bp of the position of maximal ChIP signal within a peak; ( ii ) the peak resided upstream of the gene; ( iii ) The gene was the closest gene to that peak on the same strand . Lists of RNAP , RpaA , and RpaB peaks and gene targets are found in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . For Figures 3G , 4C and 5C , we identified all RNAP , RpaA , or RpaB peaks with dusk gene targets based on the above criteria , respectively . 82 dusk genes are targets of RNAP peaks , 56 dusk genes were targets of RpaA peaks , and 42 dusk genes are targets of RpaB peaks . Then , for each peak - dusk gene pair , we calculated the change in gene expression of the dusk gene after 60 min , and the change in ChIP enrichment of the upstream peak over the mock pulldown ( described above ) after 60 min in High light , each compared to their respective values at Low light at 8 hr since dawn . We plotted these data on the x- and y-axes , respectively , with orange triangles . We repeated this process , comparing gene expression and ChIP enrichment values after 60 min in Shade compared to 8 hr since dawn in Clear Day conditions , and plotted the data as gray circles . We calculated the correlation coefficient between the change in gene expression and the change in ChIP enrichment for all peak-gene pairs of the relevant factor in the High Light pulse , and then calculated the same correlation in Shade pulse conditions separately . We calculated the correlation coefficients comparing changes after 15 min in either the High Light or Shade pulse conditions , and list these values in the legends of Figure 3—figure supplement 2 , Figure 4—figure supplement 2 , and Figure 5—figure supplement 2 . The data used for these plots for RNAP , RpaA , and RpaB are available in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . We plot data from replicate 2 , and the trends are reproduced in replicate 1 . For Figure 3—figure supplement 2 , Figure 4—figure supplement 2 , and Figure 5—figure supplement 2 , we took the lists of RNAP/RpaA/RpaB peaks with dusk gene targets from above . For each peak - gene pair , we calculated the log2 fold change in ChIP enrichment of the peak and the change in expression of the downstream gene in 15 or 60 min in the High Light pulse compared to the value at 8 hr since dawn in Low Light conditions . We repeated these calculations for each peak-gene pair in 15 or 60 min in Shade pulse compared to 8 hr since dawn in Clear Day conditions . We used hierarchical clustering on the collective ChIP and gene expression data from both conditions to determine the plotting order of the peak-gene pairs in the heat maps , and then plotted the log2 change in ChIP enrichment and dusk target gene expression in the two conditions in separate heat maps . The change in enrichment of a peak and the change in expression of its target dusk gene are aligned horizontally in their respective heat maps . The leftmost column of each heat map is white , because this column compares the time 0 data to itself and thus has a log2 value of 0 . One RpaA peak resides upstream of two dusk genes , and two RpaB peaks reside upstream of two dusk genes each , and thus the listed number of RpaA and RpaB peaks is smaller than the number of RpaA and RpaB target dusk genes . The data used for these plots for RNAP , RpaA , and RpaB are available in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . We plot data from replicate 2 , and the trends are reproduced in replicate 1 . For Figures 4D and 5D we identified all dusk genes that were targets of both RpaA and RNAP ( for Figure 4D ) or both RpaB and RNAP ( for Figure 5D ) . 33 dusk genes are targets of both RpaA and RNAP peaks , and 27 dusk genes are targets of both RpaB and RNAP . Then , for each pair of RpaA/B - RNAP peaks , we calculated the change in ChIP enrichment of the RpaA/B peak after 60 min , and the change in ChIP enrichment of the RNAP peak upstream of the same dusk gene over the mock pulldown ( described above ) after 60 min in High light , each compared to their respective values at Low light at 8 hr since dawn . We plotted these data on the x- and y-axes , respectively , with orange triangles . We repeated this process , comparing RpaA/B ChIP enrichment and RNAP ChIP enrichment values after 60 min in Shade compared to 8 hr since dawn in Clear Day conditions , and plotted the data as gray circles . We calculated the correlation coefficient between the change in RpaA/B ChIP enrichment and the change in RNAP ChIP enrichment for all RpaA/B - RNAP peak pairs of the relevant factor in the High Light pulse , and then calculated the same correlation in Shade pulse conditions separately . We calculated the correlation coefficients comparing changes after 15 min in either the High Light or Shade pulse conditions , and list these values in the legends of Figure 4—figure supplement 3 , and Figure 5—figure supplement 3 . The RNAP , RpaA , and RpaB peaks associated with each dusk gene are listed in Figure 2—source data 1 and Figure 3—source data 1 , and the enrichment values for these peaks are listed in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . The data plotted here are from replicate 2 , and the trends are reproduced in replicate 1 . For Figure 4—figure supplement 3 and Figure 5—figure supplement 3 , we took the lists of RpaA/RpaB - RNAP peaks pairs upstream of the same dusk gene from above . For each RpaA/B - RNAP peak , we calculated the log2 fold change in ChIP enrichment of the RpaA/B peak and the change in ChIP enrichment of the RNAP peak upstream of the same dusk gene in 15 or 60 min in the High Light pulse compared to the value at 8 hr since dawn in Low Light conditions . We repeated these calculations for each peak-gene pair in 15 or 60 min in Shade pulse compared to 8 hr since dawn in Clear Day conditions . We used hierarchical clustering on the collective RpaA/B and RNAP ChIP data from both conditions to determine the plotting order of the RpaA/RpaB - RNAP peak pairs in the heat maps , and then plotted the log2 change in RpaA/B ChIP enrichment and RNAP ChIP enrichment in the two conditions in separate heat maps . The change in enrichment of an RpaA/B peak and the change in enrichment of the RNAP peak upstream of the same dusk gene are aligned horizontally in their respective heat maps . The leftmost column of each heat map is white , because this column compares the time 0 data to itself and thus has a log2 value of 0 . The RNAP , RpaA , and RpaB peaks associated with each dusk gene are listed in Figure 2—source data 1 and Figure 3—source data 1 , and the enrichment values for these peaks are listed in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . The data plotted here are from replicate 2 , and the trends are reproduced in replicate 1 . For Figure 4—figure supplement 4D–F , Figure 6D , and Figure 6—figure supplement 1D–F , we identified all RpaA , RpaB , and RNAP peaks that targeted the specified gene , as described above . Then , we calculated the log2 change in RpaA ( dashed red line ) , RpaB ( dotted blue line ) , RNAP ( dashed green line ) ChIP enrichment or expression of the downstream gene ( solid black lines ) in the High Light pulse compared to 8 hr since dawn in the Low Light condition , and plotted these values with downward triangles . We repeated these calculations , comparing enrichment and gene expression in the Shade pulse to the data at 8 hr since dawn in the Clear Day condition , and plotted these values with circles . The RNAP , RpaA , and RpaB peaks associated with each dusk gene are listed in Figure 2—source data 1 and Figure 3—source data 1 , and the enrichment values for these peaks are listed in Figure 3—source data 2 , Figure 4—source data 2 , and Figure 5—source data 2 , respectively . The data plotted here are from replicate 2 , and the trends are reproduced in replicate 1 . We calculated normalized expression values of high confidence dusk genes under our dynamic light conditions , as well as in previously described RpaA perturbations in Constant Light ( Markson et al . , 2013 ) . We separately normalized the data from set of dynamic light conditions ( Low Light , Clear Day , High Light pulse , Shade pulse ) and the Constant Light data ( Wildtype , OX-D53E cells — rpaA- , kaiBC- , Ptrc::rpaA ( D53E ) — without inducer , OX-D53E with inducer , [Markson et al . , 2013] ) using z-score normalization , and used this data to separate the dusk genes into eight groups with k-means clustering in MATLAB ( RRID:SCR_001622 ) using Pearson correlation as the distance metric . We focused our analysis on the three largest clusters which accounted for most of the dusk genes ( 187/281 genes ) . The lists of genes belonging the three major clusters are found in Figure 7—source data 1 . We observed very regular and systematic changes in the expression of large clusters of dusk genes in natural light conditions ( Figures 2 , 3 and 7 ) that correlated with RpaA/B recruitment of RNAP ( Figures 4–6 ) . Thus , our goal was to determine whether simple phenomenological models similar to that inspired by Alon ( Alon , 2006 ) could reproduce these observations and offer some intuition into how they might arise . While most of the dusk genes underwent systematic changes , a small group of ∼20 genes including kaiBC was relatively insensitive to changes in light intensity ( Figure 4—figure supplement 4 ) , and we do not model those genes’ expression dynamics . Our model treats the activation or repression of the expression of a dusk gene cluster by RpaA∼P , RpaB∼P , or another cluster using effective Hill kinetics . We coarse-grained each of the three groups of circadian dusk genes ( the Early , Middle , and Late clusters in Figure 7 ) to a single effective gene with the average dynamics of the group ( Figure 7 , solid lines ) . We modeled the dynamics of a gene cluster X using a simple kinetic model of an AND gate at a promoter ( Mangan and Alon , 2003 ) , ( 1 ) dX/dt=BX+βXf ( RpaA∼P , KAX , HAX ) f ( RpaB∼P , KBX , HBX ) f ( Y , KYX , HYX ) −αXXwhere BX is the basal transcription rate; f is a function of the interaction of X with RpaA∼P , RpaA∼P , or another cluster Y; βX is the max transcription rate; and αX is the decay/dilution rate . Activating interactions were treated using a simple Hill function , ( 2 ) f ( u , K , H ) = ( u/K ) H/ ( 1+ ( u/K ) H ) , where u is the concentration of the active transcription factor , H is the Hill coefficient of interaction , and K is the coefficient of activation . Bacteria can easily tune the interactions between proteins and between transcription factors and promoters to adjust H and K for different clusters ( Buchler et al . , 2003 ) . RpaA∼P and RpaB∼P , were treated as activators , consistent with the results from Figures 4–6 . Repressive interactions between clusters were treated using ( 3 ) f ( u , K , H ) =1/ ( 1+ ( u/K ) H ) , where K is now the coefficient of repression . In Equation 1 , RpaA∼P , RpaB∼P , and Y were measured experimentally; the remainder of the parameters were left free . We determined the sufficiency of a model to describe the data by fitting the parameters using the range of values shown in Table 1 . Time propagation of the differential Equation 1 was performed using the ode45 solver in MATLAB ( RRID:SCR_001622 ) , with X ( t=0 ) set as the observed expression level at the beginning of the simulated time period . Model fitting was performed in MATLAB using the non-linear least squares solver lsqnonlin . The Akaike Information Criterion ( AIC ) and the Chi-squared test are typically used to quantify whether a model with more parameters fits the data better than another with fewer parameters simply because it is more complex . However , both approaches are for statistical models in which little to no information is used to construct the model and are not strictly applicable to the model constructed here , which is based on our understanding of transcription . If we do use AIC to compare the models , the feedback models are predicted to be most probable . In our model , H and K are effective constants that represent the overall ability of RpaA∼P , RpaB∼P , or another gene cluster Y to affect gene expression . These constants include potential indirect activation through the sigma factors , which is may be why joint activation by RpaA∼P and RpaB∼P describe the dynamics of the Middle cluster reasonably well . However , circadian gating of the Early and Late dusk genes requires further interactions that cannot be described by Hill functions of measured RpaA∼P and RpaB∼P levels . Clearly there may be more complex networks at play than those we have considered here , and much more needs to be done to fully model gene expression in S . elongatus . Here we have constructed a first model to suggest simple principles underlying the interaction of circadian and light regulation of dusk genes and offer directions for further exploration .	Living things face daily , predictable challenges due to the regular day and night cycle imposed by the Earth’s rotation . Many of them have evolved an internal ‘circadian’ clock to anticipate daily changes in the environment . However , nature can also change in unpredictable ways , and in order to survive , organisms must account for both the time of day stipulated by their clocks and changes in their present environment . For example , cyanobacteria depend on the sun for survival and must cope with light variations throughout the day and the absence of light at nighttime . Circadian clocks are made up of specific genes and their proteins . Most of what we know about how these clocks control the behavior of an organism comes from experiments performed under constant conditions . Previous research has shown that under such circumstances , the circadian clock of cyanobacteria periodically turns on a set of genes every 24 hours via a protein called RpaA . However , to understand how cyanobacteria use this clock , we must know how it works in a fluctuating environment . To test this , Piechura , Amarnath and O’Shea measured the activation of genes in cyanobacteria that had been exposed to changes in light mimicking those in nature . Compared to constant conditions , fluctuating light drastically changed the timing of activation of circadian genes . When light decreased – as it would in nature during sunset or if a cloud blocks the sun – the circadian genes were activated . Changes in light did not change the ‘ticking’ of the clock , but did affect the ability of RpaA to turn on circadian genes . Moreover , the activity of a second protein called RpaB increased when light decreased and the genes were activated . Thus , cyanobacteria switch on circadian genes as the sun is setting or during unexpected shade , likely through RpaA and RpaB , to help them survive without light . This study shows that circadian clocks activate genes differently in the real world compared to unnatural , constant conditions . This may prompt scientists to think carefully about how an organism’s natural environment can affect its inner workings . A next step will be to see how else light affects circadian gene levels . A deeper understanding of how cyanobacteria control their genes in a natural environment will be useful for scientists who engineer these organisms to produce biofuels from sunlight .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "computational", "and", "systems", "biology" ]	2017	Natural changes in light interact with circadian regulation at promoters to control gene expression in cyanobacteria
The genetic code has been proposed to be a ‘frozen accident , ’ but the discovery of alternative genetic codes over the past four decades has shown that it can evolve to some degree . Since most examples were found anecdotally , it is difficult to draw general conclusions about the evolutionary trajectories of codon reassignment and why some codons are affected more frequently . To fill in the diversity of genetic codes , we developed Codetta , a computational method to predict the amino acid decoding of each codon from nucleotide sequence data . We surveyed the genetic code usage of over 250 , 000 bacterial and archaeal genome sequences in GenBank and discovered five new reassignments of arginine codons ( AGG , CGA , and CGG ) , representing the first sense codon changes in bacteria . In a clade of uncultivated Bacilli , the reassignment of AGG to become the dominant methionine codon likely evolved by a change in the amino acid charging of an arginine tRNA . The reassignments of CGA and/or CGG were found in genomes with low GC content , an evolutionary force that likely helped drive these codons to low frequency and enable their reassignment . The genetic code defines how mRNA sequences are decoded into proteins . The ancient origin of the standard genetic code is reflected in its near-universal usage , once proposed to be a ‘frozen accident’ that is too integral to the translation of all proteins to change ( Crick , 1968 ) . However , the discovery of alternative genetic codes in over 30 different lineages of bacteria , eukaryotes , and mitochondria over the past four decades has made it clear that the genetic code is capable of evolving to some degree ( Knight et al . , 2001a; Kollmar and Mühlhausen , 2017 ) . The first alternative genetic codes were discovered by comparing newly sequenced genomes to amino acid sequences obtained by direct protein sequencing . Nonstandard codon translations were found this way in human mitochondria ( Barrell et al . , 1979 ) , Candida yeasts ( Kawaguchi et al . , 1989 ) , green algae ( Schneider et al . , 1989 ) , and Euplotes ciliates ( Meyer et al . , 1991 ) . Some reassignments of stop codons to amino acids were detected from DNA sequence alone , based on the appearance of in-frame stop codons in critical genes ( Yamao et al . , 1985; Caron and Meyer , 1985; Cupples and Pearlman , 1986; Keeling and Doolittle , 1996; McCutcheon et al . , 2009; Campbell et al . , 2013; Záhonová et al . , 2016 ) . As DNA sequence data have accumulated faster than direct protein sequences , computational methods have been developed to predict the genetic code from DNA sequence . The core principle of most methods is to align genomic coding regions to homologous sequences in other organisms ( creating multiple sequence alignments ) and then to tally the most frequent amino acid aligned to each of the 64 codons . This approach led to the discovery of new genetic codes in screens of ciliates ( Swart et al . , 2016; Heaphy et al . , 2016 ) , yeasts ( Riley et al . , 2016; Krassowski et al . , 2018 ) , green algal mitochondria ( Noutahi et al . , 2019; Žihala and Eliáš , 2019 ) , invertebrate mitochondria ( Telford et al . , 2000; Abascal et al . , 2006a; Li et al . , 2019 ) , and stop codon reassignments in metagenomic data ( Ivanova et al . , 2014 ) and the development of software for specific phylogenetic groups ( Abascal et al . , 2006b; Mühlhausen and Kollmar , 2014; Noutahi et al . , 2017 ) . Some approaches , such as FACIL ( Dutilh et al . , 2011 ) , have expanded phylogenetic breadth by using profile hidden Markov model ( HMM ) representations of conserved proteins from phylogenetically diverse databases such as Pfam ( El-Gebali et al . , 2019 ) . However , a systematic survey of genetic code usage across the tree of life has not yet been possible . Existing methods are generally either ( 1 ) phylogenetically restricted to clades where multiple sequence alignments can be built for a predetermined set of proteins or ( 2 ) lacking sufficiently robust and objective statistical footing to enable a large-scale screen with high accuracy . A potentially incomplete set of alternative genetic codes limits our ability to understand the evolutionary processes behind codon reassignment . One open question is why some codon reassignments reappear independently . Reassignment of the stop codons UAA and UAG to glutamine is the most common change in eukaryotic nuclear genomes , appearing at least five independent times ( Schneider et al . , 1989; Keeling and Doolittle , 1996; Keeling and Leander , 2003; Karpov et al . , 2013; Swart et al . , 2016 ) . In bacteria , all of the known changes reassign the stop codon UGA to either glycine in the Absconditabacteria and Gracilibacteria ( Campbell et al . , 2013; Rinke et al . , 2013 ) or tryptophan in the Mycoplasmatales , Entomoplasmatales ( Bové , 1993 ) , and several insect endosymbiotic bacteria ( McCutcheon et al . , 2009; McCutcheon and Moran , 2010; Bennett and Moran , 2013; Salem et al . , 2017 ) . These recurring changes may reflect constraints imposed by the existing translational machinery . The mechanism of codon reassignment may involve changes to tRNA anticodons or tRNA wobble nucleotide modifications ( which together dictate anticodon-codon pairing ) , aminoacyl-tRNA synthetase recognition of cognate tRNAs , release factor binding of stop codons , among others , each of which may bias which reassignments are easier to evolve . However , without a complete picture of genetic code diversity , it is hard to disentangle patterns of codon reassignment from observation bias . For instance , in-frame stop codons caused by a stop codon reassignment may be more easily detectable than a subtle change in amino acid conservation indicative of a sense codon reassignment . Another open question is how a new codon meaning can evolve without disrupting the translation of most proteins . Reassigning a codon leads to the incorporation of the incorrect amino acid at all preexisting codon positions ( Crick , 1968 ) . Three evolutionary models differ in the pressure that drives substitutions to remove the codon from positions that cannot tolerate the new translation . In the ‘codon capture’ model , the codon is first driven to near extinction by pressures unrelated to reassignment , such as biased genomic GC content or genome reduction , which then minimizes the impact of reassignment on protein translation ( Osawa and Jukes , 1989 ) . This model was first proposed for the reassignment of the stop codon UGA to tryptophan in Mycoplasma capricolum , whose low genomic GC content ( 25% GC ) in combination with small genome size ( 1 Mb ) was thought to have driven the stop codon UGA to extremely low usage in favor of UAA and allowed ‘capture’ of UGA by a tryptophan tRNA ( Bové , 1993; Osawa and Jukes , 1989 ) . For larger nuclear genomes , other models have been proposed where codon usage changes occur concurrently with , and are driven by , changes in decoding capability . In the ‘ambiguous intermediate’ model , a codon is decoded stochastically as two different meanings in an intermediate step of codon reassignment , and this translational pressure induces codon substitutions at positions where ambiguity is deleterious ( Schultz and Yarus , 1994; Massey et al . , 2003 ) . Extant examples of ambiguous translation support the plausibility of this model , such as yeasts that translate the codon CUG as both leucine and serine by stochastic tRNA charging ( Gomes et al . , 2007 ) or by competing tRNA species ( Mühlhausen et al . , 2018 ) . Alternatively , the ‘tRNA loss-driven reassignment’ model proposes an intermediate stage where a codon cannot be translated efficiently , perhaps due to tRNA gene loss or mutation , creating pressure for synonymous substitutions specifically away from that codon , allowing it to be captured later by a different tRNA ( Mühlhausen et al . , 2016; Sengupta and Higgs , 2005 ) . These three models are not mutually exclusive , and substitutions at the reassigned codon can occur due to a combination of these pressures . Here , we describe Codetta , a computational method for predicting the genetic code that can scale to analyze thousands of genomes . We perform the first survey of genetic code usage in all bacterial and archaeal genomes , reidentifying all known codes in the dataset and discovering the first examples of sense codon changes in bacteria . All five reassignments affect arginine codons ( AGG , CGA , and CGG ) and provide clues to help us understand how alternative genetic codes evolve . We developed Codetta , a computational method that takes DNA or RNA sequences from a single organism and predicts an amino acid translation for each of the 64 codons . Codetta can analyze sequences from all domains of life , including bacteria , archaea , eukaryotes , organelles , and viruses , and the ability to confidently predict codon decodings depends on having protein-coding regions with recognizable homology . The general idea is to align the input nucleotide sequence to probabilistic profiles of conserved protein domains in order to obtain , for each of the 64 codons , a set of profile positions aligned to that codon . Each profile position has 20 probabilities describing the expected amino acid . For each of the 64 codons , we aggregate over the set of aligned profile positions to infer the single most likely amino acid decoding of the codon . Most previous approaches for genetic code prediction use the same basic idea ( Telford et al . , 2000; Abascal et al . , 2006b; Dutilh et al . , 2011; Mühlhausen and Kollmar , 2014; Swart et al . , 2016; Heaphy et al . , 2016; Riley et al . , 2016; Krassowski et al . , 2018; Noutahi et al . , 2019 ) , typically aligning the input sequence to multiple sequence alignments and using a simple rule to select the best amino acid for each codon . With Codetta , we extend this idea to systematic high-throughput analysis by using a probabilistic modeling approach to infer codon decodings and by taking advantage of the large collection of probabilistic profiles of conserved protein domains ( profile HMMs ) in the Pfam database ( El-Gebali et al . , 2019 ) . Profile HMMs are built from multiple sequence alignments , and the emission probabilities at each consensus column are estimates of the expected amino acid frequencies . The Pfam database contains over 17 , 000 profile HMMs of conserved protein domains from all three domains of life , which are expected to align to about 50% of coding regions in a genome ( El-Gebali et al . , 2019 ) . We align Pfam profile HMMs to a six-frame standard genetic code translation of the input DNA/RNA sequence using the HMMER hmmscan program ( Eddy , 2011; Figure 1A ) . Since we rely on a preliminary standard code translation , conserved protein domains could fail to align in organisms using radically different genetic codes . In the set of statistically significant hmmscan alignments ( E-value §lt;10-10 ) , we make the simplifying approximation of considering each aligned consensus column independently , so the alignments are viewed as a set of pairwise associations between a codon Z ( 64 possibilities ) and a consensus column of a Pfam domain profile ( denoted C , an index identifying a Pfam consensus column ) . From these data , we infer each of the 64 codons one at a time ( Figure 1B ) . For a codon Z ( e . g . , UGA ) , the observed data C→Z are a set of N consensus columns CiZ ( i=1 . . . N ) that associate to Z in the provisional alignments . We model the main data-generative process abstractly , imagining that each column CiZ was drawn from the pool of all possible consensus columns by codon Z , which is translated as an unknown amino acid A . Each column has an affinity for codon Z proportional to the column’s emission probability for the amino acid A , P⁢ ( A\|C ) . A consensus column strongly conserved for a particular amino acid A will tend to only associate with codons that translate to A; moreover , consensus columns weakly conserved for A may also associate with probability proportional to their conservation for A . Thus , this abstract-matching process generates an observed CiZ column association with the codon Z ( translated as amino acid A ) with probabilityP⁢ ( CiZ\|A ) =P⁢ ( A\|CiZ ) ⁢P⁢ ( CiZ ) P⁢ ( A ) . Here , P⁢ ( A\|CiZ ) is the emission probability for amino acid A at the Pfam consensus column CiZ . P ( A ) is the average emission probability for amino acid A over the pool of all possible consensus columns C , which we take to be all columns aligned to the target genome in order to better reflect genome-specific biases in amino acid usage . Given the data C→Z and this abstract generative model , we infer the most likely decoding M for codon Z out of 21 possibilities M∈{Ala , Cys , … , Tyr , ? } ( Figure 1B ) . The M= ? model of nonspecific translation draws columns randomly and serves to catch codons that do not encode a specific amino acid , such as stop codons and ambiguously translated codons . For a given decoding M , the probability of the observed columns C→Z is thenP⁢ ( C→Z\|M ) ={∏i=1NP ( A=M\|CiZ ) P ( CiZ ) P ( A=M ) if ⁢M∈{Ala , Cys , … , Tyr}∏i=1NP⁢ ( CiZ ) if ⁢M= ? Setting the prior probability of each decoding , P⁢ ( M ) , to be uniform , we compute the probability of the decoding M asP ( M\|C→Z ) =P ( C→Z\|M ) ΣM′P ( C→Z\|M′ ) We assign an amino acid translation to a codon if it attains a decoding probability above some threshold ( typically 0 . 9999 ) . We assign a ‘ ? ’ if no amino acid decoding satisfies the probability threshold ( including the case where ‘ ? ’ itself has high probability ) . A ‘ ? ’ assignment tends to happen if the codon is rare , with few aligned Pfam consensus columns on which to base the inference , or if the codon is ambiguously translated such that no single amino acid model reaches high probability . Because we do not model stop codons explicitly , we expect ‘ ? ’ to be the inferred meaning since stop codons ideally would have few or no aligned Pfam consensus columns . To assess how many columns in C→Z are needed for reliable codon assignment , we constructed synthetic C→Z datasets ranging from 1 to 500 consensus columns by subsampling the Pfam consensus columns aligned to each of the 61 sense codons in the Escherichia coli genome . We calculated the per-codon error rate ( fraction of samples predicting the incorrect amino acid ) and the per-codon power ( fraction of samples predicting the correct amino acid ) using a probability threshold of 0 . 9999 ( Figure 1—figure supplement 1 ) . Lack of an amino acid inference ( ‘ ? ’ ) was considered neither an error nor a correct prediction . Per-codon error rates were <0 . 00002 for all sizes of C→Z . Depending on the codon , we found that about 8–34 aligned consensus columns suffice for >95% power to infer the correct amino acid . Accuracy may differ in real genomes for various biological reasons , but these results gave us confidence to proceed . We further validated Codetta on the budding yeasts ( Saccharomycetes , 462 sequenced species ) that vary in their translation of CUG as either serine , leucine , or alanine depending on the species ( Mühlhausen et al . , 2016; Krassowski et al . , 2018; Mühlhausen et al . , 2018 ) . In some CUG-Ser clade species , such as Candida albicans , CUG codons are stochastically decoded as a mix of serine ( 97% ) and leucine ( 3% ) because the CUG-decoding tRNACAG is aminoacylated by both the seryl- and leucyl-tRNA synthetases ( Suzuki et al . , 1997; Gomes et al . , 2007 ) . Codetta is not designed to predict ambiguous decoding and is expected to assign either the dominant amino acid or a ‘ ? ’ in cases like C . albicans . For 453 species , the predicted CUG translation was consistent with the known phylogenetic distribution of CUG reassignments ( Figure 2A ) . This includes C . albicans , which was predicted to use the predominant serine translation ( Gomes et al . , 2007 ) . For the remaining nine species , Codetta did not put a high probability on any amino acid decoding of CUG ( inferred meaning of ‘ ? ’ ) . Two of these species – Babjeviella inositovora and Cephaloascus fragrans – are basal members of the CUG-Ser clade . Both of these genomes contain a CUG-decoding tRNACAG gene with features of serine identity ( see Materials and methods ) and B . inositovora has previously been shown to translate CUG codons primarily as serine by whole proteome mass spectrometry ( Krassowski et al . , 2018; Mühlhausen et al . , 2018 ) , suggesting that CUG is decoded as serine in these species . Codetta did not infer an amino acid for CUG because the aligned Pfam consensus columns were not consistently conserved for a single amino acid ( Figure 2—figure supplement 1 ) . The other seven species without an inferred amino acid for CUG all belong to the closely related genera Ascoidea and Saccharomycopsis ( four additional species in these clades were predicted to translate CUG as serine ) . Analysis of tRNA genes revealed that 10 out of 11 species in this clade encode two types of tRNACAG genes , one predicted to be serine-type and one leucine-type , suggesting that CUG may be ambiguously translated as both serine and leucine via competing tRNAs in some of these species ( Figure 2B ) . We used northern blotting to assay the expression of both tRNACAG genes in some of these species under a variety of growth conditions ( data not shown ) , but could detect reliable expression of both serine- and leucine-type tRNACAG genes only in Saccharomycopsis malanga ( only the serine tRNACAG could be detected in other species ) ( Figure 2C ) . To determine whether both tRNAs are aminoacylated , we performed acid urea PAGE northern blotting that separates aminoacylated and deacylated tRNAs but does not identify the charged amino acid . We found that both serine and leucine S . malanga tRNACAG are predominantly charged in cells ( Figure 2C ) , likely partaking in the translation of CUG codons . If CUG is indeed translated ambiguously in this clade , it would explain why Codetta did not place a high probability on any single amino acid decoding for some species . The existence of serine and leucine tRNACAG genes in some Ascoidea and Saccharomycopsis yeasts was reported by Krassowski et al . , 2018 and Mühlhausen et al . , 2018 while we were conducting experiments . Ambiguous translation of CUG was demonstrated in Ascoidea asiatica ( Mühlhausen et al . , 2018 ) ; however , for S . malanga only expression of the serine tRNACAG could be detected ( Krassowski et al . , 2018 ) and incorporation of predominantly serine at protein positions encoded by CUG ( Mühlhausen et al . , 2018 ) . In contrast to these studies , we used a saturated growth condition where the leucine tRNACAG seems to be more highly expressed . While we did not quantify the relative expression of the two tRNACAG in S . malanga , a visual comparison of the band intensities in Figure 2C suggests that the expression of the leucine tRNACAG is at least 10 times less than the serine tRNACAG even in the saturated growth condition . These results show that Codetta can correctly infer canonical and non-canonical codon translations and can flag unusual situations such as ambiguous translation even though it assumes unambiguous translation . All of the remaining 63 codons were inferred to use the expected translation in all species , with the following exceptions . In three species belonging to a lineage of Hanseniaspora with low genomic GC content ( Steenwyk et al . , 2019 ) , the arginine codons CGC and/or CGG had a ‘ ? ’ inference due to few ( <20 ) aligned Pfam consensus columns presumably due to rare usage of those codons . In eight other species , either the stop codon UAG or UGA was inferred to code for tryptophan due to some ( <23 ) aligned Pfam consensus columns . We could not find any nuclear suppressor tRNA genes , and we believe that these inferences are due to the erroneous alignment of Pfam domains to in-frame stop codons in pseudogenes . In-frame stop codons do not appear randomly within pseudogenes but instead are most likely to result from single-nucleotide transitions from certain codons ( such as the UGG tryptophan codon ) . To explore the diversity of genetic codes in bacterial and archaeal genomes , we used Codetta to analyze 251 , 571 assembled genomes from GenBank , including partial assemblies and those derived from single-cell genomics and metagenomic assembly . Summaries of our analysis ( Table 1 and Table 2 ) are shown for a subset of the results , dereplicated to reduce the over-representation of frequently sequenced organisms by selecting a single assembly for each species-level NCBI taxonomic ID ( 48 , 693 unique species: 46 , 384 bacteria , 2309 archaea ) . Results for the full dataset and the dereplicated subset are available in Supplementary file 1 . To see if our screen recovered known alternative genetic codes , we collated a comprehensive literature summary of all bacterial and archaeal clades known to use alternative genetic codes ( Table 1 ) and layered it over the NCBI taxonomy , annotating all remaining organisms with the standard genetic code . This resulted in a genetic code annotation for each species . For most species using known alternative genetic codes in our dataset , our predictions at the reassigned codon agreed with the expected amino acid translation ( Table 1 ) . There were no instances of reassigned codons predicted to translate as an unexpected amino acid , but there were a few cases of reassigned UGA codons that had no amino acid meaning inferred ( ‘ ? ’ inference ) . Since the uninferred codons could represent a lack of sensitivity by Codetta , we looked more closely at these examples . In the Mycoplasmatales and Entomoplasmatales , which are believed to translate the canonical stop codon UGA as tryptophan ( Bové , 1993 ) , eight species had no inferred amino acid meaning for UGA due to fewer than four aligned Pfam consensus columns . All of these genomes lack a UGA-decoding tRNA UCATrp gene and all but one instead contain a release factor 2 gene ( which terminates translation at UGA ) . Five of these species are included in the Genome Taxonomy Database ( GTDB ) ( Parks et al . , 2020 ) , a comprehensive phylogeny of over 190 , 000 bacterial and archaeal genomes , where they are grouped into a different order ( GTDB order RF39 ) . We therefore attribute at least five ( and perhaps all eight ) as a taxonomic misannotation in the NCBI database , and we believe that UGA is a stop codon in these species . In the Gracilibacteria , which are believed to translate the stop codon UGA as glycine ( Rinke et al . , 2013 ) , two species had no inferred amino acid meaning for UGA . Neither genome contained the expected UGA-decoding tRNA UCAGly gene and both instead encoded a release factor 2 gene , supporting that UGA is a stop codon and not a glycine codon in these species . Indeed , one of these species is included in the GTDB and is grouped in a different order than the other UGA-reassigned Gracilibacteria and Absconditabacteria . Across the 48 , 693 genomes ( dereplicated to one assembly per species ) , we predicted the amino acid translation of a total of 2 , 970 , 497 individual sense codons ( roughly 61 times the number of genomes ) , with 99 . 79% of the predictions consistent with the expected amino acid ( similar proportion across bacteria and archaea ) ( Table 2 ) . About 0 . 19% of sense codons had a ‘ ? ’ inference , demonstrating that entire genomes contain more than enough information to infer the amino acid translation of most sense codons . Unexpected amino acid meanings were predicted for 612 sense codons . These are candidates for new codon reassignments , but could also include inference errors . For stop codons , 99 . 80% out of a total of 145 , 855 stop codons across the dereplicated bacterial and archaeal genomes had no inferred amino acid meaning , as expected . 290 bacterial stop codons and 9 archaeal stop codons were inferred to translate as an amino acid , adding to our list of candidate new genetic codes . To prioritize high-confidence novel genetic codes , we gathered additional evidence by examining ( 1 ) the translational components ( tRNA and/or release factor genes ) involved in the reassignment , ( 2 ) the usage of the reassigned codon , including manual examination of alignments of highly conserved single-copy genes , and ( 3 ) the phylogenetic extent of the proposed reassignment . Since many candidate genetic codes were found in uncultivated clades with only rough taxonomic classification on NCBI , we explored phylogenetic relationships using the GTDB ( Parks et al . , 2020 ) . The GTDB is a phylogeny of over 190 , 000 archaeal and bacterial genomes , providing provisional domain-to-species phylogenetic classifications for uncultivated as well as established clades . A list of all candidate novel genetic codes can be found in Supplementary file 2 . We focused on the candidate codon reassignments with the highest degree of additional evidence and attempted to characterize common sources of error . The set of lower-confidence candidates may still include additional real codon reassignments requiring further validation . The most common error was the inference of AGA and/or AGG arginine codons as coding for lysine , occurring in 567 bacterial species . Almost all of the AGA- and AGG-decoding tRNAs found in these genomes were consistent with arginine identity ( based on the arginine identity elements A/G73 and A20 ) , supporting that AGA and AGG are arginine codons in the majority of these species . The unusually high GC content of these genomes ( ranging between 0 . 52 and 0 . 77 , median 0 . 68 ) suggests that the source of the lysine inference may come from high GC content-driven nonsynonymous substitutions of the AAA and AAG lysine codons to AGA and AGG arginine codons at protein residues that can tolerate either positively charged amino acid . As a result , AGA and AGG codons would consistently appear at residues conserved for lysine in other species , which Codetta would mistake for the signature of codon reassignment . Genomic GC content has long been correlated with greater arginine usage and lower lysine usage ( Sueoka , 1961; Lightfield et al . , 2011 ) , possibly due to substitutions between the aforementioned lysine and arginine codons ( Knight et al . , 2001b ) . This error could be mitigated in future analyses by using profile HMMs built from sequences that match the analyzed genome in GC content or amino acid composition . Some erroneous stop codon inferences resulted from genome contamination by organisms with known stop codon reassignments . We suspected contamination when the Pfam consensus columns aligned to a stop codon were only present in a limited part of the genome and confirmed the origin of these regions by homology search of the genes containing the in-frame stop codons . We have found examples of predicted stop reassignments in Sulfolobus assemblies caused by contamination with UGA-recoding Mycoplasma contigs , in an alphaproteobacteria assembly caused by contamination with UAA- and UAG-recoding ciliate contigs , in Chloroflexi assemblies caused by contamination with UGA-recoding Absconditabacteria contigs , and in others . We found five clades using candidate novel alternative genetic codes with additional computational evidence , such as tRNA genes consistent with the new translation . All five new genetic codes involve the reassignment of arginine codons , representing the first sense codon reassignments in bacteria . Eight bacterial genomes were inferred to translate AGG , a canonical arginine codon , as methionine . All eight genomes were assembled from fecal metagenomes of baboons or humans ( Parks et al . , 2017; Almeida et al . , 2019 ) and have only coarse-grained NCBI genome classification as uncultured Bacillales or Mollicutes bacteria . The GTDB assigns these eight genomes to a three species clade within the placeholder genus UBA7642 ( family CAG-288 , order RFN20 , class Bacilli ) , of which all other species were inferred to translate AGG as arginine ( Figure 3A ) . In each of the reassigned genomes , the AGG inference by Codetta is based on a sufficiently large number of aligned Pfam consensus columns ( over 2200 compared to an average of about 1800 for each of the other 60 sense codons ) from over 480 different Pfam domains . Figure 3B shows an example multiple sequence alignment of uridylate kinase , a single-copy conserved bacterial gene , from the reassigned species , outgroup genomes , and several more distantly related bacteria . In the alignment , AGG codons are used interchangeably with AUG methionine codons within the reassigned clade and tend to occur at positions conserved for methionine and other nonpolar amino acids in the other species . In the reassigned clade , AGG is the dominant methionine codon with a usage of 209–235 per 10 , 000 codons in Pfam alignments , outnumbering the canonical methionine codon AUG ( 59–69 per 10 , 000 codons ) ( Figure 3A ) . The process of codon reassignment involves genome-wide codon substitutions to remove the reassigned codon from positions that cannot tolerate the new amino acid , leading to depressed codon usage . High usage of AGG in the reassigned clade suggests that this is an established codon reassignment that has had time to rebound in frequency through synonymous substitutions with the standard AUG methionine codon . In many outgroup genomes , AGG is a rare arginine codon ( Figure 3A ) . Escape from viral infection has been put forth as a potential selective pressure for the evolution of alternative genetic codes , although viruses are also known to infect some alternative genetic code organisms such as Mycoplasma and mitochondria ( Shackelton and Holmes , 2008 ) . We inferred the genetic code of phage genomes assembled by Al-Shayeb et al . , 2020 from the same baboon fecal metagenomic dataset as some reassigned Bacilli genomes . Two phage assemblies were predicted to translate AGG as methionine ( assemblies GCA_902730795 . 1 and GCA_902730815 . 1 ) . The assemblies do not contain genes for the AGG-decoding tRNACCU , so the phage presumably rely on the host tRNAs for translation . Thus , some phage may have adapted to the AGG translation as methionine in the reassigned Bacilli . We used tRNAscan-SE 2 . 0 ( Chan et al . , 2021 ) to determine which tRNAs are available to decode AGG in the reassigned and outgroup genomes ( Figure 3A ) . Some tRNA genes are missing , possibly due to the incomplete nature of some metagenome-assembled genomes as indicated by low genome completeness estimates . The cognate tRNA for the AGG codon , tRNACCU , from the reassigned clade has features of methionine identity ( including an A73 discriminator base and G2:C71 and C3:G70 base pairs in the acceptor stem ) and lacks the important arginine identity element A20 in the D-loop ( Meinnel et al . , 1993; Giegé et al . , 1998 ) , supporting translation of AGG as methionine ( Figure 3C ) . In vitro experiments have shown that anticodon mutations to tRNA CAUMet disrupt recognition by the E . coli methionyl-tRNA synthetase ( MetRS ) ; however , the C35 change necessary to decode the AGG codon affects the least critical anticodon nucleotide ( Schulman and Pelka , 1983 ) . To see if any compensatory changes have occurred in the MetRS from the reassigned clade to accommodate recognition of a new anticodon , we compared the predicted MetRS sequence to that of Aquifex aeolicus ( crystal structure in complex with tRNACAU , PDB 2CSX/2CT8 , Nakanishi et al . , 2005 ) . The three residues that contact the anticodon nucleotides in A . aeolicus , which are conserved through all domains of life ( Nakanishi et al . , 2005 ) , also remain unchanged in the reassigned clade MetRS sequence ( Figure 3—figure supplement 1 ) . The outgroup genomes contain a tRNACCU with features of arginine identity ( including a G73 discriminator base and A20 in the D-loop ) . The genomic context of the tRNACCU is similar in many reassigned clade and outgroup genomes , flanked by a tRNA CCGArg immediately downstream and a homolog of GenBank protein CDA36808 . 1 upstream ( Figure 3D ) . This implies that the reassigned and outgroup tRNACCU evolved from the same ancestral tRNA gene , and the reassigned methionine tRNACCU likely emerged through a change in aminoacylation of an arginine tRNACCU rather than through duplication and anticodon mutation of a methionine tRNA . The reassigned genomes use an arginine-type tRNAUCU to decode the unaffected AGA arginine codon . Depending on the post-transcriptional modification of the U34 anticodon nucleotide , the arginine tRNAUCU could recognize AGG via wobble and potentially cause ambiguous translation . In E . coli , the U34 of tRNAUCU is modified to 5-methylaminomethyluridine ( Sakamoto et al . , 1993 ) , which primarily decodes the AGA codon with a low level of AGG recognition ( Spanjaard et al . , 1990 ) . Mukai et al . , 2015 demonstrated that it is possible to engineer separate decodings for AGA and AGG in E . coli by reducing expression level of the tRNAUCU to the point where decoding of AGG by tRNAUCU is presumably insignificant in competition with the cognate tRNACCU . In most outgroup genomes , AGA is the dominant arginine codon , while in the reassigned clade the preferred arginine codon is CGU ( Figure 3A ) , which may indicate reduced demand and expression of tRNAUCU to avoid ambiguous translation of AGG . A similar potential for ambiguous translation due to U34 wobble exists with the previously known decoding of UGA as glycine and UGG as tryptophan in Absconditabacteria and Gracilibacteria ( Campbell et al . , 2013; Rinke et al . , 2013 ) . The remaining four clades with codon reassignments supported by additional computational evidence all affect the arginine codons CGA and/or CGG ( Figure 4 ) . Three clades are in the phylum Firmicutes: the genus Peptacetobacter is predicted to translate CGG as glutamine ( Figure 4—figure supplement 1 ) , a clade of uncultivated Bacilli in the GTDB order RFN20 ( same as the AGG-reassigned Bacilli ) is predicted to translate CGG as tryptophan ( Figure 4—figure supplement 2 ) , and members of the genus Anaerococcus are also predicted to translate CGG as tryptophan ( Figure 4—figure supplement 3 ) . The fourth clade is Absconditabacteria ( also known as Candidate Division SR1 , part of the Candidate Phyla Radiation ) , which is predicted to have reassigned CGA and CGG both to tryptophan ( Figure 4—figure supplement 4 ) , in addition to the already known reassignment of UGA from stop to glycine . In contrast to the reassignment of AGG to become the dominant methionine codon ( described in the previous section ) , these CGA/CGG reassignments resemble earlier stages of codon reassignment where the reassigned codon has not yet rebounded in frequency through synonymous substitutions with the new amino acid . Due to the rarity of the reassigned CGA/CGG codons , these predictions are based on fewer aligned Pfam consensus columns and may be more prone to error . As a check for each reassignment , we looked for examples of the reassigned codon in conserved regions of single-copy gene alignments ( Figure 4—figure supplement 1B , Figure 4—figure supplement 2B , Figure 4—figure supplement 3B and Figure 4—figure supplement 4B ) and found multiple supporting positions for all reassigned codons except the extremely rare CGG codon in Anaerococcus . We also looked for tRNA genes with an anticodon and amino acid identity elements consistent with the reassignment ( Figure 4—figure supplement 1 , Figure 4—figure supplement 2 , Figure 4—figure supplement 3 and Figure 4—figure supplement 4 ) , and found consistent tRNAs for all clades except for Peptacetobacter whose CGG-decoding tRNACCG resembles neither an arginine or glutamine isotype . While amino acid conservation at the reassigned codon and sequence-based prediction of tRNA charging may lend support to a predicted codon reassignment , only experimental confirmation can establish how the reassigned codons are translated in vivo and whether there is ambiguous translation . In particular , Anaerococcus and Peptacetobacter include culturable species and may be experimentally confirmed in the future . The four CGA/CGG candidate reassignments share several features that suggest common evolutionary forces at play . Most notable is the very low genomic GC content of the reassigned clades ( 0 . 26–0 . 38 , Figure 4A ) . In all four clades , the usage of GC-rich CGN-box codons – including CGA and CGG – is depressed and arginine residues are primarily encoded by AGA codons ( Figure 4B ) . In the three Firmicute CGG reassignments , CGG is an extremely rare codon ( codon usage <6 per 10 , 000 in aligned Pfam domains for all species ) . In the Absconditabacteria , CGG also tends to be quite rare ( <7 per 10 , 000 in all but one species ) with CGA slightly more abundant ( <37 per 10 , 000 in all species ) . In one Absconditabacteria ( assembly GCA_002791215 . 1 ) , the frequency of both CGA and CGG approaches the frequency of the canonical tryptophan codon UGG , consistent with a more advanced stage of codon reassignment ( usage of CGA and CGG is 30 and 24 per 10 , 000 , compared to 35 for UGG ) . Low genomic GC content is thought to be created by mutational bias in favor of AT nucleotides , causing a gradual shift towards synonymous codons with lower GC compositions ( Knight et al . , 2001b; Muto and Osawa , 1987 ) . This may have helped disfavor usage of CGA and/or CGG prior to reassignment , lessening the impact of changing the codon meaning . The tRNAs used to decode the CGN codon box may have also influenced the reassignment of CGA and CGG codons . A shared feature of the three Firmicute CGG reassignments is that the tRNAUCG is missing ( Figure 4B ) , presumably lost prior to or during the reassignment of CGG . If the tRNAUCG were present , it would likely recognize both CGA and CGG via wobble that would complicate assigning different amino acid meanings to those two codons . In the absence of tRNAUCG , CGA ( along with CGU and CGC ) is presumably decoded by a tRNA ICGArg ( derived by deamination of tRNA ACGArg , the only widespread instance of inosine tRNA wobble in bacteria ) . This leaves CGG to be decoded solely by a tRNACCG ( Figure 4B ) . In this situation , CGG is one of a few codons in the genetic code decoded by a single dedicated tRNA , potentially facilitating codon reassignment since the translational meaning of CGG can now be altered independently of neighboring codons . The inosine wobble modification is not used by some deeply branching bacteria ( Rafels-Ybern et al . , 2019 ) , and the tRNA ACGArg gene appears to be lacking in the Candidate Phyla Radiation , including Absconditabacteria . Instead , these bacteria use a tRNA GCGArg to decode CGU and CGC , and rely on a tRNAUCG and tRNACCG to recognize CGA and CGG ( Figure 4B ) . Since the ability of tRNAUCG to decode CGA and CGG makes it difficult to split the translational meanings of the two codons , it may explain why both CGA and CGG are reassigned to tryptophan together in the Absconditabacteria . For some of these reassignments , close outgroup species may shed light on potential intermediate stages of codon reassignment . The CGG reassignment in the Absconditabacteria may extend to members of the sister clade Gracilibacteria – some Gracilibacteria were predicted to translate CGG as tryptophan , while others translate CGG as arginine ( Figure 4—figure supplement 4 ) . This may reflect a complicated history of CGG reassignment and possible reversion to arginine translation . For the CGG reassignment in Peptacetobacter , the closest sister group ( which includes the pathogen Clostridioides difficile ) has extremely rare usage of CGG ( <1 per 10 , 000 in aligned Pfam domains in all but two species ) and appears to lack any tRNA capable of decoding CGG by standard codon-anticodon pairing rules ( Figure 4—figure supplement 1 ) . This may resemble an intermediate stage in codon reassignment before the ability to translate CGG as a new amino acid is gained , similar to the unassigned CGG codon in M . capricolum ( Oba et al . , 1991 ) . In Anaerococcus , all species contain a CGG-decoding tRNACCG with features of tryptophan identity ( Figure 4—figure supplement 3 ) . Unexpectedly , members of an outgroup genus Finegoldia also have a tRNACCG with features of tryptophan identity ( CGG inferred to be ‘ ? ’ by Codetta ) . It is unclear if the tRNACCG genes in these two clades share an evolutionary history or represent independent events . We present a method for computationally inferring the genetic code that can scale to analyze hundreds of thousands of genomes that we call Codetta . We validate Codetta on the well-studied reassignments of CUG in yeasts and rediscover the likely ambiguous translation of CUG as serine and leucine in some Ascoidea and Saccharomycopsis species . We conduct the first systematic survey of genetic code usage across the majority of sequenced organisms , analyzing all sequenced bacteria and archaea ( over 250 , 000 assemblies ) . The five new alternative genetic codes described here substantially expand the known diversity of codon reassignments in bacteria . Now , in addition to reassignments of the stop codon UGA to tryptophan or glycine , we have the first sense codon reassignments in bacteria , affecting the arginine codons AGG , CGA , and CGG . Two reassignments occur in culturable bacteria – in Anaerococcus and Peptacetobacter – and could be experimentally confirmed in the future , for example by proteomic mass spectrometry . Since Codetta selects the most likely amino acid translation among the 20 canonical amino acids , some types of codon reassignments may be missed . We cannot predict reassignment to a noncanonical amino acid – for such codons , Codetta would pick the nonspecific model or an amino acid that is used similarly in other species . We also cannot directly detect ambiguous translation , which may represent an important stage in codon reassignment . However , the failure to infer an amino acid translation despite a significant number of aligned Pfam consensus columns may hint at ambiguous translation , as was the case for CUG in Ascoidea and Saccharomycopsis . Since we do not model translational initiation and termination , we cannot detect the use of new start and stop codons or context-dependent stop codons that also possess an amino acid meaning , known to occur in some eukaryotes ( Swart et al . , 2016; Heaphy et al . , 2016; Záhonová et al . , 2016 ) . Expanding our analysis to eukaryotic , organellar , and viral genomes will help fill in the diversity of alternative genetic codes , but poses additional challenges . Since we align profile HMMs to a six-frame translation of the entire genome , the pervasive pseudogenes in many eukaryotic genomes will likely increase the rate of incorrect codon inferences by having sufficient homology for alignment but enough accumulated mutations to cause incorrect pairing of codons to consensus columns . Smaller scale surveys of eukaryotic genetic code diversity have focused on transcriptomic datasets ( Swart et al . , 2016; Heaphy et al . , 2016 ) , which may alleviate this problem . Some viral and organellar genomes have very few protein-coding genes that may limit the ability to confidently infer the entire genetic code . One strategy is to improve sensitivity at the cost of generalizability by using clade-specific profile HMMs instead of Pfam , which may increase the proportion of aligned coding sequence . Another challenge in some organellar genomes is extensive mRNA editing ( Gray , 1996; Alfonzo et al . , 1997 ) , which violates our assumption that the genomic codon sequence represents the mRNA sequence and may require analyzing the edited transcriptome to ensure correct correspondence of codons to profile HMM positions . In the ‘codon capture’ model of codon reassignment , genome-wide pressures such as biased GC content or genome reduction drive a codon to near extinction such that the codon can acquire a new tRNA decoding with a minimal effect on translation ( Osawa and Jukes , 1989 ) . Most UGA reassignments in bacteria occur in clades with very low genomic GC content , which is thought to have reduced UGA to very low usage in favor of the stop codon UAA . This includes the Mycoplasmatales and Entomoplamatales ( 0 . 24–0 . 39 GC ) ( Jukes , 1985; Bové , 1993 ) , Absconditabacteria and Gracilibacteria ( 0 . 21–0 . 53 GC , Figure 4—source data 1; Campbell et al . , 2013; Rinke et al . , 2013 ) , and most insect endosymbiotic bacterial reassignments ( 0 . 13–0 . 17 GC , except for Hodgkinia cicadicola with 0 . 28–0 . 58 GC ) ( McCutcheon and Moran , 2010; Bennett and Moran , 2013; Salem et al . , 2017; Campbell et al . , 2015 ) . The CGA and/or CGG reassignments described here similarly exhibit low genomic GC content ( 0 . 26–0 . 38 ) and very rare usage of GC-rich codons including CGA and CGG . A codon does not need to completely disappear for reassignment to be facilitated by rare codon usage , and it is likely that a brief period of translational ambiguity or inefficiency helps drive the remaining codon substitutions . We posit that , in bacteria , reduction in codon usage driven by genome-wide processes plays a major role in enabling codon reassignment and may explain why codon reassignments repeatedly evolve in clades such as Firmicutes ( known for their low genomic GC content ) and lifestyles such as endosymbiosis ( which is often accompanied by genome reduction and low GC content ) ( McCutcheon and Moran , 2011 ) . All five of the new reassignments affect arginine codons ( AGG , CGA , and CGG ) . While these are the first instances of arginine codon reassignment in non-organellar genomes , several arginine reassignments are known in mitochondria: in various metazoan mitochondria , the codons AGA and AGG have been reassigned to serine , glycine , and possibly stop and AGG has been reassigned to lysine ( Knight et al . , 2001a; Abascal et al . , 2006a ) , and in various green algal mitochondria AGG has been reassigned to alanine and methionine and CGG to leucine ( Noutahi et al . , 2019 ) . Arginine codons have several unique features that may predispose them to codon reassignment . First , across the tree of life , arginine has an over-representation of codons in the genetic code relative to usage in proteins ( Jukes et al . , 1975; King and Jukes , 1969 ) , contributing to rare usage of the least favored arginine codon . Second , the six arginine codons range from 1 to 3 GC nucleotides in composition ( only equaled by leucine ) , which may create greater bias in codon usage in response to genomic GC content than for amino acids with less GC variability in their codons . In organisms with small genomes , these features alone might make the rarest arginine codon very low in number and more susceptible than other codons to reassignment . The arginine codon CGG may be even more of a target for reassignment because , in most bacteria , the only widespread instance of inosine tRNA wobble is used to decode the CGU , CGC , and CGA arginine codons ( Grosjean et al . , 2010 ) . In the absence of a tRNAUCG , CGG is decoded by a dedicated tRNACCG and can be reassigned without affecting the translation of other codons . Some codon reassignments have convergently reappeared across the tree of life: CGG to tryptophan in three bacterial clades described here , AGG to methionine in a clade of Bacilli described here and in green algal mitochondria ( Noutahi et al . , 2019 ) , UGA to tryptophan in multiple bacterial , mitochondrial , and eukaryotic lineages ( Knight et al . , 2001a ) , and others . Recurrent changes could reflect ( 1 ) a common evolutionary process , for example , low GC content-driven reassignments disproportionately affecting codons sensitive to GC fluctuations , or ( 2 ) shared constraints imposed by conserved translational machinery , including tRNAs and aminoacyl-tRNA synthetases . For example , the tRNA anticodon-codon pairing rules dictate that U- and C-ending codons cannot be assigned separate meanings , and indeed this has not been observed in any known genetic codes . This may explain why in low GC content genomes we see reassignments of the arginine codon CGG but not the arginine codon CGC , which would have to be reassigned together with CGU . The selection of amino acid changes in the codon reassignments described here is not clearly explained by biochemical similarity ( except possibly for the reassignment of CGG from arginine to glutamine ) . The amino acid choice may be related to the constraints on evolving new tRNA anticodons . Most of the changes described here ( and indeed all of the changes known in bacteria ) involve a single nucleotide difference from cognate anticodons: tRNACCU in addition to tRNA CAUMet for the AGG to methionine reassignment , tRNACCG in addition to tRNA CCATrp for the CGG to tryptophan reassignments , and tRNACCG in addition to tRNA CUGGln for the CGG to glutamine reassignment . Evolving a new anticodon through a single mutation may be more probable than through multiple mutations . However , the methionine tRNACCU involved in the reassignment of AGG in a clade of Bacilli appears to have evolved from an arginine tRNACCU through mutations that altered aminoacyl-tRNA synthetase recognition , rather than by an anticodon mutation to a methionine tRNACAU gene . Alternatively , this pattern could result from a limitation on the new anticodons that an aminoacyl-tRNA synthetase could accept since most aminoacyl-tRNA synthetases use the anticodon in part to distinguish cognate and non-cognate tRNAs ( Giegé et al . , 1998 ) . Upon characterizing the diversity of genetic codes in other parts of the tree of life , we may discover that the general patterns and evolutionary pressures differ from bacteria , reflecting differences in translational machinery , lifestyle , or genome characteristics . A preliminary translation of the input nucleotide sequences is produced by first breaking any long sequences into nonoverlapping 100 kb pieces ( because of a limit on input protein sequence length for hmmscan ) , then translating into all six frames ( as six polypeptide sequences ) using the standard genetic code with stop codons translated as ‘X . ’ A custom version of Pfam 32 . 0 profiles was produced from Pfam seed alignments using hmmbuild --enone , which turns off entropy weighting , resulting in emission probability parameters closer to the original amino acid frequencies in the input alignments . Significant homologous alignments were identified by searching each translated polypeptide against the custom Pfam database using hmmscan from HMMER 3 . 1b2 ( Eddy , 2011 ) for domain hits with E-value §lt;10-10 . Alignments were further filtered to remove uncertainly aligned consensus columns ( posterior probabilities of alignment < 95% ) . By default , no single Pfam consensus columns was allowed to account for more than 1% of total aligned consensus columns for a codon in order to mitigate some artifacts such as repetitive pseudogene families in some genomes; when this happened , the number of codon positions aligned to that specific consensus column was downsampled to 1% of the total ( if a codon was aligned to fewer than 100 Pfam consensus columns total , then each unique consensus columns was downsampled to one occurrence ) . We excluded hits to five classes of Pfam models including mitochondrial proteins , viral proteins , selenoproteins , pyrrolysine-containing proteins , and proteins belonging to transposons and other mobile genetic elements . These filtered sets of aligned consensus columns defined the input C→ sets for each codon . The equations from the main text are then used ( in log-probability calculations for numerical stability ) to infer P⁢ ( M\|C→Z ) for each codon , with a default decoding probability threshold of 0 . 9999 . The computational requirements are dominated by the hmmscan step , which takes about an hour on a single CPU core for an ~12 Maa six-frame translation of a typical 6 Mb bacterial genome . We ran different genomes in parallel on a 30 , 000 core computing resource , the Harvard Cannon cluster . We implemented this method as Codetta v1 . 0 , a Python 3 program that can be found at https://github . com/kshulgina/codetta/releases/tag/v1 . 0 , ( copy archived at swh:1:rev:4f5f31a33beed19bc3e10745154705ad002273df , Yekaterina , 2021 ) . A six-frame translation of the E . coli O157:H7 str . Sakai genome ( GCA_000008865 . 2 ) was searched against the custom --enone Pfam 32 . 0 profile database as described above . We generated 1000 random subsamples each of 1 through 50 , 100 , and 500 aligned consensus columns per sense codon and inferred the most likely decoding as described above . A codon inference was considered ‘true’ ( T ) if the correct amino acid meaning was inferred , ‘false’ ( F ) if an incorrect amino acid meaning was inferred , and ‘uninferred’ ( U ) if the nonspecific decoding was most probable or if no model surpassed the model probability threshold . For a given model probability threshold , per-codon error rate is the fraction of samples with a false inference ( F/ ( T + F + U ) ) . Per-codon power is the fraction of samples with a true inference ( T/ ( T + F + U ) ) . Both values were evaluated individually for each sense codon and also aggregated across all sense codons . Assembly identifiers for all archaeal and bacterial genomes were downloaded from the NCBI Genome database on June 4 , 2020 , and Codetta analysis was performed on all archaeal and bacterial genome assemblies . Genetic code inference results for all analyzed genomes can be found in Supplementary file 1 . A variety of additional files supporting new genetic codes are available at https://github . com/kshulgina/ShulginaEddy_21_genetic_codes , ( Shulgina , 2021 , copy archived at swh:1:rev:a2bf2a1ef0bcd5ea0319354ab6e9cbba89f9e934 ) . We used the NCBI taxonomy database ( downloaded on July 15 , 2020 ) to cross-reference all assemblies with taxonomic identifiers . All analyzed genome assemblies from GenBank are associated with an NCBI taxonomic ID ( taxid ) . Because some of these taxids correspond to subspecies or strain-level designations , we assigned a species-level taxid to each assembly by iteratively stepping up the NCBI taxonomy until a species-level node was reached . To create a dereplicated dataset , we picked one genome assembly per NCBI species-level taxid . If multiple genome assemblies were associated with an NCBI species-level taxid , assemblies were sorted based on RefSeq category ( reference , representative , or neither ) and then genome completeness level and a single-genome assembly was randomly selected from the highest ranked category . Phage assemblies derived from the same metagenomic samples as the AGG-recoding Bacilli were obtained by identifying the phage assemblies from Al-Shayeb et al . , 2020 whose sample accessions were linked to the metagenomic sequencing experiments SRX834619 , SRX834622 , SRX834629 , SRX834636 , SRX834653 , SRX834655 , or SRX834666 . Codetta analysis of the phage genomes was performed as described above . A complete list of bacterial clades previously known to use alternative genetic codes was collated with corresponding references for genetic code discovery and taxonomic distribution ( Table 1 ) . For each clade , we determined a set of NCBI taxids best defining the phylogenetic extent of each reassigned clade . We used this to generate a curated genetic code annotation for all NCBI species-level taxids: for the taxids defining each reassigned clade , all species-level child nodes were annotated with the alternative genetic code; all remaining species-level taxids were annotated with the standard genetic code . We used the GTDB ( version R05-RS95; Parks et al . , 2020 ) to determine the phylogenetic placement of species that use candidate new genetic codes and to identify the most closely related outgroup species . The tRNA gene content of genomes was determined by running tRNAscan-SE 2 . 0 ( Chan et al . , 2021 ) with default settings and a tRNA model appropriate for the domain of life ( i . e . , option -E for eukaryotes , -B for bacteria , -A for archaea ) . To help ensure that tRNAs of interest were not missed , we also ran a low-stringency search with the general tRNA model and no cutoff score ( options -G -X 0 ) and a search with previous version of tRNAscan-SE ( option -L ) and manually examined the outputs . We searched bacterial genomes for release factor genes using the TIGRFAM 15 . 0 ( Haft et al . , 2013 ) release factor 2 model ( TIGR00020 ) and release factor 1 model ( TIGR00019 ) and for the methionyl-tRNA synthetase gene using the TIGRFAM 15 . 0 model ( TIGR00399 ) with hmmscan against a six-frame translation of the entire genome with default settings . Since the release factors are homologs , if the two models hit overlapping genomic coordinates , we kept the hit with the more significant E-value . Methionyl-tRNA synthetase alignments were generated using MAFFT v7 . 429 ( Katoh and Standley , 2013 ) with default settings . For the AGG arginine to methionine reassignment in a clade of Bacilli , we classified tRNACCU genes as being primarily arginine acceptors if the tRNA had A20 in the D-loop and a A/G73 discriminator base , and primarily methionine acceptors if the tRNA had an A73 discriminator base and not A20 in the D-loop ( Giegé et al . , 1998 ) . The weaker methionine identity elements G2:C70 , C3:G69 in the acceptor stem were used to support the assignment ( Meinnel et al . , 1993 ) . In the reassignment of CGG to glutamine in Peptacetobacter , we classified tRNAs as arginine-type using the rules above , and as glutamine-type if the tRNA was missing arginine identity element A20 and contained the set of glutamine identity elements consisting of a weak 1:72 basepair , A37 , and A/G73 ( Jahn et al . , 1991 ) . We took the additional glutamine identity elements G2:C71 and G3:C70 in the acceptor stem , G38 in the anticodon loop , and G10 in the D-stem as support for glutamine identity ( Jahn et al . , 1991; Hayase et al . , 1992 ) . For the reassignments of CGA and/or CGG arginine to tryptophan , we classified tRNAs as primarily arginine acceptors using the rules above , and provisionally as tryptophan acceptors if the tRNA had a G73 discriminator base and not A20 in the D-loop ( Giegé et al . , 1998 ) . We considered the weak tryptophan identity element A/G1:U72 in the acceptor stem as support for tryptophan identity but did not require it ( Himeno et al . , 1991 ) . In the Absconditabacteria and Gracilibacteria , we classified tRNAUCA genes as glycine acceptors if the tRNA had G1:C72 , C2:G71 , G3:C70 in the acceptor stem and U73 discriminator base ( Giegé et al . , 1998 ) . We refrained from assigning identity if the tRNA did not fit the above patterns or if the D-loop sequence was unusual such that it was unclear which nucleotide is N20 . D-loop and variable loop insertions were placed at positions following the convention of Sprinzl et al . , 1998 . For some candidate novel alternative genetic codes , we constructed multiple sequence alignments of conserved single-copy bacterial genes from the BUSCO database v3 ( Waterhouse et al . , 2018 ) . To identify orthologs of a BUSCO gene in a particular genome , we first created a dataset of putative protein sequences by translating all open reading frames longer than 50 codons using the inferred genetic code ( assuming standard stop codons unless reassigned ) , with candidate reassigned codons translated as ‘X . ’ Then , we queried each of the 148 bacterial BUSCO profile HMMs against all putative proteins using hmmsearch from HMMER 3 . 1b2 with default settings and an E-value cutoff of 10-13 , and picked the most significant hit if it also yielded a reciprocal best hit against the entire BUSCO profile HMM database using hmmscan with the same E-value cutoff . Multiple sequence alignments were generated using MAFFT v7 . 429 ( Katoh and Standley , 2013 ) with default settings . For the described novel genetic codes , BUSCO alignments containing the reassigned codon in the reassigned clade were individually inspected and alignments containing the reassigned codon at conserved positions in well-aligned regions were preferentially selected as example alignments . To determine the genomic context surrounding the tRNACCU gene in the uncultivated Bacilli predicted to have reassigned AGG to methionine and in close outgroup genomes , we predicted tRNA and protein coding genes in the whole genome as described above . We annotated each putative protein coding gene with the reciprocal best hit homolog among annotated protein-coding genes in the outgroup assembly GCA_000434395 . 1 using phmmer from HMMER 3 . 1b2 with a 10-10 E-value cutoff . For analysis of CUG translation in budding yeasts , we selected all genomes belonging to the class Saccharomycetes ( NCBI taxid 4891 ) , which represent 463 unique NCBI species taxids with at least one genome . The genomes were dereplicated to one assembly per species-level taxid as described above . Yeast species were split into six taxonomic categories based on the ‘major clade’ annotation from the phylogenetic analysis by Shen et al . , 2018 as follows: outgroups ( major clades: Lipomycetaceae , Trigonopsidaceae , Dipodascaceae/Trichomonascaceae , Alloascoideaceae , Sporopachydermia ) , CUG-Leu clade 1 ( major clades: Phaffomycetaceae , Saccharomycodaceae , Saccharomycetaceae ) , CUG-Leu clade 2 ( major clade: Pichiaceae ) , CUG-Ser ( major clade: CUG-Ser1 ) , CUG-Ala ( major clade: CUG-Ala ) , and CUG-Ser/Leu ( major clade: CUG-Ser2 ) . Species that were not included in the analysis by Shen et al . , 2018 were sorted into the same major clade as other members of their annotated genus on NCBI . A single species ( Candida sp . JCM 15000 ) could not be placed into a category and was excluded from the analysis . The expected CUG translation for each clade follows Shen et al . , 2018 and is consistent with other studies of CUG translation ( Riley et al . , 2016; Krassowski et al . , 2018; Mühlhausen et al . , 2018 ) . Genetic codes were predicted by Codetta as described above . A table describing all yeast genomes analyzed can be found in Figure 2—source data 1 . tRNA gene content of yeast genomes was determined using tRNAscan-SE 2 . 0 as described above . In eukaryotes , only leucine- and serine-tRNAs have a long ( >12 nucleotide ) variable loop so we used this feature to confirm the tRNACAG identity as serine or leucine . In yeast , serine tRNAs typically have a conserved G73 discriminator base but can tolerate any nucleotide ( Himeno et al . , 1997 ) , while leucine tRNA identity is conferred by a A73 discriminator base and A35 and G37 in anticodon loop ( Soma et al . , 1996 ) . We categorized tRNACAG genes as either serine-acceptors or leucine-acceptors based on the presence of these features . In some CUG-Ser clade species , serine CAG-tRNAs containing a G37 have been found to be charged with leucine at a low level ( 3% ) ( Suzuki et al . , 1997 ) ; for categorization purposes , we would consider these tRNAs to be primarily serine-acceptors . S . malanga ( NRRL Y-7175 ) was obtained from the Agricultural Research Service Culture Collection ( Peoria , IL ) . Cells were inoculated into 5 mL of YPD liquid media ( containing 1% yeast extract , 2% peptone , and 2% dextrose ) from a colony on a YPD agar plate and grown to saturation for 4 days at 25°C on rotating wheel . Total RNA was extracted in acidic conditions to preserve tRNA charging , following the steps outlined in Varshney et al . , 1991 with the following modifications . Cells were harvested by centrifugation ( 5 min at 4000 rpm at 4°C ) , resuspended in 500 µL ice-cold buffer containing 0 . 3 M NaOAc pH 4 . 5 and 10 mM EDTA and added to 500 µL ice-cold phenol:chloroform ( pH 4 . 5 ) and 500 µL of 0 . 4–0 . 5µm acid-washed glass beads for cell lysis . All RNA extraction steps were performed at 4°C . In the first round of extraction , cells were vortexed for 30 min , rested on ice for 3 min , centrifuged for 15 min at 20 , 000×g , and the aqueous layer was transferred to 500 µL of phenol:chloroform ( pH 4 . 5 ) , which was subject to a second round of extraction ( identical , except for 3 min vortex ) . A last round of extraction was performed in 500 µL of chloroform with a 15 s vortex and 2 min centrifugation . RNA in the aqueous phase was precipitated and resuspended in buffer containing 10 mM NaOAc pH 4 . 5 and 1 mM EDTA . The single-stranded DNA probes used for detection of S . malanga tRNA CAGSer ( 5′ GAAATCCCAGCGCCTTCTGTGGGCGGCGCCTTAACCAAACTCGGC 3′ ) and S . malanga tRNA CAGLeu ( 5′ TTGACAATGAGACTCGAACTCATACCTCCTAG 3′ ) were 5′ end-labeled with [γ-P32]-ATP by T4 polynucleotide kinase ( New England Biosciences ) and purified using ProbeQuant G-50 Micro Columns ( GE Healthcare Life Sciences ) . In vitro transcribed tRNAs were used as controls for probe specificity . For the S . malanga tRNA CAGSer probe , an in vitro transcribed version of the target tRNA CAGSer was used as a positive control and an in vitro transcribed version of tRNA CGUSer was used as a control for cross-hybridization . For the tRNA CAGLeu probe , an in vitro transcribed version of the target tRNA CAGLeu was used as a positive control and an in vitro transcribed version of tRNA CAALeu was used as a control for cross-hybridization . Cross-hybridization controls were selected by aligning the reverse complement of the probe sequence using MAFFT v7 . 429 ( Katoh and Standley , 2013 ) with default settings to all tRNA genes in the S . malanga genome ( found by tRNAscan-SE 2 . 0 ) and selecting the non-target tRNA with the highest pairwise alignment score . In vitro transcribed tRNAs were produced using the MAXIscript T7 Transcription Kit ( Thermo ) from a DNA template composed of a T7 promoter ( 5′ GATCTAATACGACTCACTATAGGGAGA 3′ ) followed by the tRNA sequence ( Figure 2—source data 3 ) . The resulting tRNA transcript has an additional six nucleotides of the promoter included at the 5′ end . CCA-tails were not included in the in vitro transcribed tRNA sequences . Total RNA and in vitro transcribed controls for probe specificity were denatured in formamide buffer ( Gel Loading Buffer II , Thermo ) at 90°C for 5 min and electrophoretically separated on a 10% TBE urea gel ( Novex ) . Gels were rinsed in 0 . 5× TBE and RNA was transferred onto a Hybond N + membrane ( GE Healthcare ) in 0 . 5× TBE by semi-dry transfer ( Bio-Rad Transblot ) at 3 mA/cm2 for 1 hr . Blots were crosslinked on each side using a Stratalinker UV crosslinker on the ‘auto-crosslink’ setting . Blots were prehybridized in PerfectHyb Plus Hybridization buffer ( Sigma ) at 64°C for 1 hr prior to incubation with the radiolabeled DNA probe overnight . Blots were washed at 64°C twice in low-stringency buffer ( 0 . 1% SDS , 2× SSC ) for 15 min and once in high-stringency buffer ( 0 . 1% SDS , 0 . 1× SSC ) for 10 min , exposed on storage phosphor screens , and scanned using a Typhoon imager . For the partial deacylation control , total RNA was treated in 100 mM Tris pH 7 . 0 at 37°C for 30 min , quenched with an equal volume of buffer containing 50 mM NaOAc and 100 mM NaCl , and precipitated . Electrophoresis on acid urea polyacrylamide gels was performed as described in Varshney et al . , 1991 . 4 µg of total RNA and partial deacylation control in acid urea sample buffer ( 0 . 1 NaOAc pH 4 . 5 , 8 M urea , 0 . 05% bromophenol blue , 0 . 05% xylene cyanol ) were loaded onto a 0 . 4 mm thick 6 . 5% polyacrylamide gel ( SequaGel ) containing 8 M urea and 100 mM NaOAc pH 4 . 5 and run for 18 hr at 450 V in 4°C with 100 mM NaOAc pH 4 . 5 running buffer . The region between the two dyes corresponds to the tRNA size range , and was cut out and transferred onto a blot for probing following the same steps as above for northern blotting .	All life forms rely on a ‘code’ to translate their genetic information into proteins . This code relies on limited permutations of three nucleotides – the building blocks that form DNA and other types of genetic information . Each ‘triplet’ of nucleotides – or codon – encodes a specific amino acid , the basic component of proteins . Reading the sequence of codons in the right order will let the cell know which amino acid to assemble next on a growing protein . For instance , the codon CGG – formed of the nucleotides guanine ( G ) and cytosine ( C ) – codes for the amino acid arginine . From bacteria to humans , most life forms rely on the same genetic code . Yet certain organisms have evolved to use slightly different codes , where one or several codons have an altered meaning . To better understand how alternative genetic codes have evolved , Shulgina and Eddy set out to find more organisms featuring these altered codons , creating a new software called Codetta that can analyze the genome of a microorganism and predict the genetic code it uses . Codetta was then used to sift through the genetic information of 250 , 000 microorganisms . This was made possible by the sequencing , in recent years , of the genomes of hundreds of thousands of bacteria and other microorganisms – including many never studied before . These analyses revealed five groups of bacteria with alternative genetic codes , all of which had changes in the codons that code for arginine . Amongst these , four had genomes with a low proportion of guanine and cytosine nucleotides . This may have made some guanine and cytosine-rich arginine codons very rare in these organisms and , therefore , easier to be reassigned to encode another amino acid . The work by Shulgina and Eddy demonstrates that Codetta is a new , useful tool that scientists can use to understand how genetic codes evolve . In addition , it can also help to ensure the accuracy of widely used protein databases , which assume which genetic code organisms use to predict protein sequences from their genomes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "genetics", "and", "genomics" ]	2021	A computational screen for alternative genetic codes in over 250,000 genomes
Brainstem and cerebellar neurons implement an internal model to accurately estimate self-motion during externally generated ( ‘passive’ ) movements . However , these neurons show reduced responses during self-generated ( ‘active’ ) movements , indicating that predicted sensory consequences of motor commands cancel sensory signals . Remarkably , the computational processes underlying sensory prediction during active motion and their relationship to internal model computations during passive movements remain unknown . We construct a Kalman filter that incorporates motor commands into a previously established model of optimal passive self-motion estimation . The simulated sensory error and feedback signals match experimentally measured neuronal responses during active and passive head and trunk rotations and translations . We conclude that a single sensory internal model can combine motor commands with vestibular and proprioceptive signals optimally . Thus , although neurons carrying sensory prediction error or feedback signals show attenuated modulation , the sensory cues and internal model are both engaged and critically important for accurate self-motion estimation during active head movements . For many decades , research on vestibular function has used passive motion stimuli generated by rotating chairs , motion platforms or centrifuges to characterize the responses of the vestibular motion sensors in the inner ear and the subsequent stages of neuronal processing . This research has revealed elegant computations by which the brain uses an internal model to overcome the dynamic limitations and ambiguities of the vestibular sensors ( Figure 1A; Mayne , 1974; Oman , 1982; Borah et al . , 1988; Glasauer , 1992; Merfeld , 1995; Glasauer and Merfeld , 1997; Bos et al . , 2001; Zupan et al . , 2002; Laurens , 2006; Laurens and Droulez , 2007; Laurens and Droulez , 2008; Laurens and Angelaki , 2011; Karmali and Merfeld , 2012; Lim et al . , 2017 ) . These computations are closely related to internal model mechanisms that underlie motor control and adaptation ( Wolpert et al . , 1995; Körding and Wolpert , 2004; Todorov , 2004; Chen-Harris et al . , 2008; Berniker et al . , 2010; Berniker and Kording , 2011; Franklin and Wolpert , 2011; Saglam et al . , 2011; 2014 ) . Neuronal correlates of the internal model of self-motion have been identified in brainstem and cerebellum ( Angelaki et al . , 2004; Shaikh et al . , 2005; Yakusheva et al . , 2007 , 2008 , 2013 , Laurens et al . , 2013a , 2013b ) . In the past decade , a few research groups have also studied how brainstem and cerebellar neurons modulate during active , self-generated head movements . Strikingly , several types of neurons , well-known for responding to vestibular stimuli during passive movement , lose or reduce their sensitivity during self-generated movement ( Gdowski et al . , 2000; Gdowski and McCrea , 1999; Marlinski and McCrea , 2009; McCrea et al . , 1999; McCrea and Luan , 2003; Roy and Cullen , 2001; 2004; Brooks and Cullen , 2009; 2013; 2014; Brooks et al . , 2015; Carriot et al . , 2013 ) . In contrast , vestibular afferents respond indiscriminately for active and passive stimuli ( Cullen and Minor , 2002; Sadeghi et al . , 2007; Jamali et al . , 2009 ) . These properties resemble sensory prediction errors in other sensorimotor functions such as fish electrosensation ( Requarth and Sawtell , 2011; Kennedy et al . , 2014 ) and motor control ( Tseng et al . , 2007; Shadmehr et al . , 2010 ) . Yet , a consistent quantitative take-home message has been lacking . Initial experiments and reviews implicated proprioceptive switches ( Figure 1B; Roy and Cullen , 2004; Cullen et al . , 2011; Cullen , 2012; Carriot et al . , 2013; Brooks and Cullen , 2014 ) . More recently , elegant experiments by Brooks and colleagues ( Brooks and Cullen , 2013; Brooks et al . , 2015 ) started making the suggestion that the brain predicts how self-generated motion activates the vestibular organs and subtracts these predictions from afferent signals to generate sensory prediction errors ( Figure 1C ) . However , the computational processes underlying this sensory prediction have remained unclear . Confronting the findings of studies utilizing passive and active motion stimuli leads to a paradox , in which central vestibular neurons encode self-motion signals computed by feeding vestibular signals through an internal model during passive motion ( Figure 1A ) , but during active motion , efference copies of motor commands , also transformed by an internal model ( Figure 1C ) , attenuate the responses of the same neurons . Thus , a highly influential interpretation is that the elaborate internal model characterized with passive stimuli would only be useful in situations that involve unexpected ( passive ) movements but would be unused during normal activities , because either its input or its output ( Figure 1—figure supplement 1 ) would be suppressed during active movement . Here , we propose an alternative that the internal model that processes vestibular signals ( Figure 1A ) and the internal model that generates sensory predictions during active motion ( Figure 1C ) are identical . In support of this theory , we show that the processing of motor commands must involve an internal model of the physical properties of the vestibular sensors , identical to the computations described during passive motion , otherwise accurate self-motion estimation would be severely compromised during actively generated movements . The essence of the theory developed previously for passive movements is that the brain uses an internal representation of the laws of physics and sensory dynamics ( which has been elegantly modeled as forward internal models of the sensors ) to process vestibular signals . In contrast , although it is understood that transforming head motor commands into sensory predictions is likely to also involve internal models , no explicit mathematical implementation has ever been proposed for explaining the response attenuation in central vestibular areas . A survey of the many studies by Cullen and colleagues even questions the origin and function of the sensory signals canceling vestibular afferent activity , as early studies emphasized a critical role of neck proprioception in gating the cancellation signal ( Figure 1B , Roy and Cullen , 2004 ) , whereas follow-up studies proposed that the brain computes sensory prediction errors , without ever specifying whether the implicated forward internal models involve vestibular or proprioceptive cues ( Figure 1C , Brooks et al . , 2015 ) . This lack of quantitative analysis has obscured the simple solution , which is that transforming motor commands into sensory predictions requires exactly the same forward internal model that has been used to model passive motion . We show that all previous experimental findings during both active and passive movements can be explained by a single sensory internal model that is used to generate optimal estimates of self-motion ( Figure 1D , ‘Kalman filter’ ) . Because we focus on sensory predictions and self-motion estimation , we do not model in detail the motor control aspects of head movements and we consider the proprioception gating mechanism as a switch external to the Kalman filter , similar to previous studies ( Figure 1D , black dashed lines and red switch ) . We use the framework of the Kalman filter ( Figure 1D; Figure 1—figure supplement 2; Kalman , 1960 ) , which represents the simplest and most commonly used mathematical technique to implement statistically optimal dynamic estimation and explicitly computes sensory prediction errors . We build a quantitative Kalman filter that integrates motion signals originating from motor , canal , otolith , vision and neck proprioceptor signals during active and passive rotations , tilts and translations . We show how the same internal model must process both active and passive motion stimuli , and we provide quantitative simulations that reproduce a wide range of behavioral and neuronal responses , while simultaneously demonstrating that the alternative models ( Figure 1—figure supplement 1 ) do not . These simulations also generate testable predictions , in particular which passive stimuli should induce sensory errors and which should not , that may motivate future studies and guide interpretation of experimental findings . Finally , we summarize these internal model computations into a schematic diagram , and we discuss how various populations of brainstem and cerebellar neurons may encode the underlying sensory error or feedback signals . The structure of the Kalman filter in Figure 1D is shown with greater detail in Figure 1—figure supplement 2 and described in Materials and methods . In brief , a Kalman filter ( Kalman , 1960 ) is based on a forward model of a dynamical system , defined by a set of state variables X that are driven by their own dynamics , motor commands and internal or external perturbations . A set of sensors , grouped in a variable S , provide sensory signals that reflect a transformation of the state variables . Note that St may provide ambiguous or incomplete information , since some sensors may measure a mixture of state variables , and some variables may not be measured at all . The Kalman filter uses the available information to track an optimal internal estimate of the state variable X . At each time t , the Kalman filter computes a preliminary estimate ( also called a prediction , X^p ( t ) ) and a corresponding predicted sensory signal S^p . In general , the resulting state estimate X^p and the predicted sensory prediction S^p may differ from the real values X and S . These errors are reduced using sensory information , as follows ( Figure 1—figure supplement 2B ) : First , the prediction S^p and the sensory input S are compared to compute a sensory error δS . Second , sensory errors are transformed into a feedback Xk=K . δS , where K is a matrix of feedback gains , whose dimensionality depends on both the state variable X and the sensory inputs . Thus , an improved estimate at time t is X^ ( t ) = X^p ( t ) +K . δS ( t ) . The feedback gain matrix K determines how sensory errors improve the final estimate X^ ( see Supplementary methods , ‘Kalman filter algorithm’ for details ) . Figure 2 applies this framework to the problem of estimating self-motion ( rotation , tilt and translation ) using vestibular sensors , with two types of motor commands: angular velocity ( Ωu ) and translational acceleration ( Au ) , with corresponding unpredicted inputs , Ωε and Aε ( Figure 2A ) that represent passive motion or motor error ( see Discussion: ‘Role of the vestibular system during active motion: fundamental , ecological and clinical implications’ ) . The sensory signals ( S ) we consider initially encompass the semicircular canals ( rotation sensors that generate a sensory signal V ) and the otoliths organs ( linear acceleration sensors that generate a sensory signal F ) – proprioception is also added in subsequent sections . Each of these sensors has distinct properties , which can be accounted for by the internal model of the sensors . The semicircular canals exhibit high-pass dynamic properties , which are modeled by another state variable C ( see Supplementary methods , ‘Model of head motion and vestibular sensors’ ) . The otolith sensors exhibit negligible dynamics , but are fundamentally ambiguous: they sense gravitational as well as linear acceleration – a fundamental ambiguity resulting from Einstein’s equivalence principle [Einstein , 1907; modeled here as G ( t ) = ∫Ω ( t ) . dt and F ( t ) =G ( t ) +A ( t ) ; note that G and A are expressed in comparable units; see Materials and methods; 'Simulation parameters'] . Thus , in total , the state variable X has 4-degrees of freedom ( Figure 2A ) : angular velocity Ω and linear acceleration A ( which are the input/output variables directly controlled ) , as well as C ( a hidden variable that must be included to model the dynamics of the semicircular canals ) and tilt position G ( another hidden variable that depends on rotations Ω , necessary to model the sensory ambiguity of the otolith organs ) . The Kalman filter computes optimal estimates Ω^ ( t ) , G^ ( t ) , A^ ( t ) and C^ ( t ) based on motor commands and sensory signals . Note that we do not introduce any tilt motor command , as tilt is assumed to be controlled only indirectly though rotation commands ( Ωu ) . For simplicity , we restrict self-motion to a single axis of rotation ( e . g . roll ) and a single axis of translation ( inter-aural ) . The model can simulate either rotations in the absence of head tilt ( e . g . rotations around an earth-vertical axis: EVAR , Figure 2B ) or tilt ( Figure 2C , where tilt is the integral of rotation velocity , G ( t ) = ∫Ω ( t ) . dt ) using a switch ( but see Supplementary methods , ‘Three-dimensional Kalman filter’ for a 3D model ) . Sensory errors are used to correct internal motion estimates using the Kalman gain matrix , such that the Kalman filter as a whole performs optimal estimation . In theory , the Kalman filter includes a total of eight feedback signals , corresponding to the combination of two sensory ( canal and otolith ) errors and four internal states ( Ω^ ( t ) , G^ ( t ) , A^ ( t ) and C^ ( t ) ) . From those eight feedback signals , two are always negligible ( Table 2; see also Supplementary methods , ‘Kalman feedback gains’ ) . We will show how this model performs optimal estimation of self-motion using motor commands and vestibular sensory signals in a series of increasingly complex simulations . We start with a very short ( 0 . 2 s ) EVAR stimulus , where canal dynamics are negligible ( Figure 3 ) , followed by a longer EVAR that highlights the role of an internal model of the canals ( Figure 4 ) . Next , we consider the more complex tilt and translation movements that require all four state variables to demonstrate how canal and otolith errors interact to disambiguate otolith signals ( Figures 5 and 6 ) . Finally , we extend our model to simulate independent movement of the head and trunk by incorporating neck proprioceptive sensory signals ( Figure 7 ) . For each motion paradigm , identical active and passive motion simulations will be shown side by side in order to demonstrate how the internal model integrates sensory information and motor commands . We show that the Kalman feedback plays a preeminent role , which explains why lots of brain machinery is devoted to its implementation ( see Discussion ) . For convenience , all mathematical notations are summarized in Table 1 . For Kalman feedback gain nomenclature and numerical values , see Table 2 . In Figure 3 , we simulate rotations around an earth-vertical axis ( Figure 3A ) with a short duration ( 0 . 2 s , Figure 3B ) , chosen to minimize canal dynamics ( C≈0 , Figure 3B , cyan ) such that the canal response matches the velocity stimulus ( V≈ Ω , compare magenta curve in Figure 3C with blue curve in Figure 3B ) . We simulate active motion ( Figure 3D–K , left panels ) , where Ω=Ωu ( Figure 3D ) and Ωε=0 ( not shown ) , as well as passive motion ( Figure 3D–K , right panels ) , where Ω=Ωε ( Figure 3D ) and Ωu=0 ( not shown ) . The rotation velocity stimulus ( Ω , Figure 3E , blue ) and canal activation ( V , Figure 3F , magenta ) are identical in both active and passive stimulus conditions . As expected , the final velocity estimate Ω^ ( output of the filter , Figure 3G , blue ) is equal to the stimulus Ω ( Figure 3E , blue ) during both passive and active conditions . Thus , this first simulation is meant to emphasize differences in the flow of information within the Kalman filter , rather than differences in performance between passive and active motions ( which is identical ) . The fundamental difference between active and passive motions resides in the prediction of head motion ( Figure 3H ) and sensory canal signals ( Figure 3I ) . During active motion , the motor command Ωu ( Figure 3D ) is converted into a predicted rotation Ω^p=Ωu ( Figure 3H ) by the internal model , and in turn in a predicted canal signal V^p ( Figure 3I ) . Of course , in this case , we have purposely chosen the rotation stimulus to be so short ( 0 . 2 s ) , such that canal afferents reliably encode the rotation stimulus ( V≈Ω; compare Figure 3F and E , left panels ) and the internal model of canals dynamics have a negligible contribution; that is , Ω^p≈V^p ( compare Figure 3I and H , left panels ) . Because the canal sensory error is null , that is δV=V−V^p ≈0 ( Figure 3K , left panel ) , the Kalman feedback pathway remains silent ( not shown ) and the net motion estimate is unchanged compared to the prediction , that is , Ω^= Ω^p=Ωu= Ω . In conclusion , during active rotation ( and in the absence of perturbations , motor or sensory noise ) , motion estimates are generated entirely based on an accurate predictive process , in turn leading to an accurate prediction of canal afferent signals . In the absence of sensory mismatch , these estimates don’t require any further adjustment . In contrast , during passive motion the predicted rotation is null ( Ω^p=0 , Figure 3H , right panel ) , and therefore the predicted canal signal is also null ( V^p=0 , Figure 3I , right panel ) . Therefore , canal signals during passive motion generate a sensory error δV= V-V^p = V ( Figure 3K , right panel ) . This sensory error is converted into a feedback signal Ωk= kδVΩ . δV ( Figure 3J ) with a Kalman gain kδVΩ ( feedback from canal error δV to angular velocity estimate Ω ) that is close to 1 ( Table 2; note that this value represents an optimum and is computed by the Kalman filter algorithm ) . The final motion estimate is generated by this feedback , that is Ω^= kδVΩ . δV=V≈ Ω . These results illustrate the fundamental rules of how active and passive motion signals are processed by the Kalman filter ( and , as hypothesized , the brain ) . During active movements , motion estimates are generated by a predictive mechanism , where motor commands are fed into an internal model of head motion . During passive movement , motion estimates are formed based on feedback signals that are themselves driven by sensory canal signals . In both cases , specific nodes in the network are silent ( e . g . predicted canal signal during passive motion , Figure 3I; canal error signal during active motion , Figure 3K ) , but the same network operates in unison under all stimulus conditions . Thus , depending on whether the neuron recorded by a microelectrode in the brain carries predicted , actual or error sensory signals , differences in neural response modulation are expected between active and passive head motion . For example , if a cell encodes canal error exclusively , it will show maximal modulation during passive rotation , and no modulation at all during active head rotation . If a cell encodes mixtures of canal sensory error and actual canal sensory signals ( e . g . through a direct canal afferent input ) , then there will be non-zero , but attenuated , modulation during active , compared to passive , head rotation . Indeed , a range of response attenuation has been reported in the vestibular nuclei ( see Discussion ) . We emphasize that in Figure 3 we chose a very short-duration ( 0 . 2 s ) motion profile , for which semicircular canal dynamics are negligible and the sensor can accurately follow the rotation velocity stimulus . We now consider more realistic rotation durations , and demonstrate how predictive and feedback mechanisms interact for accurate self-motion estimation . Specifically , canal afferent signals attenuate ( because of their dynamics ) during longer duration rotations – and this attenuation is already sizable for rotations lasting 1 s or longer . We next demonstrate that the internal model of canal dynamics must be engaged for accurate rotation estimation , even during purely actively generated head movements . We now simulate a longer head rotation , lasting 2 s ( Figure 4A , B , blue ) . The difference between the actual head velocity Ω and the average canal signal V is modeled as an internal state variable C , which follows low-pass dynamics ( see Supplementary methods , ‘Model of head motion and vestibular sensors’ ) . At the end of the 2 s rotation , the value of C reaches its peak at ~40% of the rotation velocity ( Figure 4B , cyan ) , modeled to match precisely the afferent canal signal V , which decreases by a corresponding amount ( Figure 4C ) . Note that C persists when the rotation stops , matching the canal aftereffect ( V=−C < 0 after t > 2 s ) . Next , we demonstrate how the Kalman filter uses the internal variable C to compensate for canal dynamics . During active motion , the motor command Ωu ( Figure 4D ) is converted into an accurate prediction of head velocity Ω^p ( Figure 4H , blue ) . Furthermore , Ωu is also fed through the internal model of the canals to predict C^p ( Figure 4H , cyan ) . By combining the predicted internal state variables Ω^p and C^p , the Kalman filter computes a canal prediction V^p that follows the same dynamics as V ( compare Figure 4F and I , left panels ) . Therefore , as in Figure 3 , the resulting sensory mismatch is δV=V−V^p ≈0 and the final estimates ( Figure 4G ) are identical to the predicted estimates ( Figure 4H ) . Thus , the Kalman filter maintains an accurate rotation estimate by feeding motor commands through an internal model of the canal dynamics . Note , however , that because in this case V≠Ω ( compare magenta curve in Figure 4F and blue curve in Figure 4E , left panels ) , V^p≠ Ω^p ( compare magenta curve in Figure 4I and blue curve in Figure 4H , left panels ) . Thus , the sensory mismatch can only be null under the assumption that motor commands have been processed through the internal model of the canals . But before we elaborate on this conclusion , let’s first consider passive stimulus processing . During passive motion , the motor command Ωu is equal to zero . First , note that the final estimate Ω^≈Ω is accurate ( Figure 4G ) , as in Figure 3G , although canal afferent signals don’t encode Ω accurately . Second , note that the internal estimate of canal dynamics C^ ( Figure 4G ) and the corresponding prediction ( C^p; Figure 4H ) are both accurate ( compare with Figure 4E ) . This occurs because the canal error δV ( Figure 4K ) is converted into a second feedback , Ck , ( Figure 4J , cyan ) , which updates the internal estimate C^ ( see Supplementary methods , ‘Velocity Storage’ ) . Finally , in contrast to Figure 3 , the canal sensory error δV ( Figure 4K ) does not follow the same dynamics as V ( Figure 4C , F ) , but is ( as it should ) equal to Ω ( Figure 4B ) . This happens because , though a series of steps ( V^p = -C^p in Figure 4I and δV=V−V^p in Figure 4K ) , C^p is added to the vestibular signal V to compute δV≈Ω . This leads to the final estimate Ω^=Ω^p=δV≈Ω ( Figure 4G ) . Model simulations during even longer duration rotations and visual-vestibular interactions are illustrated in Figure 4—figure supplement 1 . Thus , the internal model of canal dynamics improves the rotation estimate during passive motion . Remarkably , this is important not only during very long duration rotations ( as is often erroneously presumed ) , but also during short stimuli lasting 1–2 s , as illustrated with the simulations in Figure 4 . We now return to the actively generated head rotations to ask the important question: What would happen if the brain didn’t use an internal model of canal dynamics ? We simulated motion estimation where canal dynamics were removed from the internal model used by the Kalman filter ( Figure 4—figure supplement 2 ) . During both active and passive motion , the net estimate Ω^ is inaccurate as it parallels V , exhibiting a decrease over time and an aftereffect . In particular , during active motion , the motor commands provide accurate signals Ω^p , but the internal model of the canals fails to convert them into a correct prediction V^p , resulting in a sensory mismatch . This mismatch is converted into a feedback signal Ωk that degrades the accurate prediction Ω^p such that the final estimate Ω^ is inaccurate . These simulations highlight the role of the internal model of canal dynamics , which continuously integrates rotation information in order to anticipate canal afferent activity during both active and passive movements . Without this sensory internal model , active movements would result in sensory mismatch , and the brain could either transform this mismatch into sensory feedback , resulting in inaccurate motion estimates , or ignore it and lose the ability to detect externally generated motion or movement errors . Note that the impact of canal dynamics is significant even during natural short-duration and high-velocity head rotations ( Figure 4—figure supplement 3 ) . Thus , even though particular nodes ( neurons ) in the circuit ( e . g . vestibular and rostral fastigial nuclei cells presumably reflecting either δV or Ωk in Figures 3 and 4; see Discussion ) are attenuated or silent during active head rotations , efference copies of motor commands must always be processed though the internal model of the canals – motor commands cannot directly drive appropriate sensory prediction errors . This intuition has remained largely unappreciated by studies comparing how central neurons modulate during active and passive rotations – a misunderstanding that has led to a fictitious dichotomy belittling important insights gained by decades of studies using passive motion stimuli ( see Discussion ) . Next , we study the interactions between rotation , tilt and translation perception . We first simulate a short duration ( 0 . 2 s ) roll tilt ( Figure 5A; with a positive tilt velocity Ω , Figure 5B , blue ) . Tilt position ( G , Figure 5B , green ) ramps during the rotation and then remains constant . As in Figure 3 , canal dynamics C are negligible ( V≈Ω; Figure 5F , magenta ) and the final rotation estimate Ω^ is accurate ( Figure 5G , blue ) . Also similar to Figure 3 , Ω^ is carried by the predicted head velocity node during active motion ( Ω^≈ Ω^p; Ωk≈0 ) and by the Kalman feedback node during passive motion ( Ω≈Ωk; Ω^p≈0 ) . That is , the final rotation estimate , which is accurate during both active and passive movements , is carried by different nodes ( thus , likely different cell types; see Discussion ) within the neural network . When rotations change orientation relative to gravity , another internal state ( tilt position G , not included in the simulations of Figures 3 and 4 ) and another sensor ( otolith organs; F=G since A=0 in this simulation; Figure 5F , black ) are engaged . During actively generated tilt movements , the rotation motor command ( Ωu ) is temporally integrated by the internal model ( see Eq . 3c of Supplementary methods , ‘Kalman filter algorithm developed’ ) , generating an accurate prediction of head tilt G^p ( t ) = ∫0tΩu . dt ( Figure 5H , left panel , green ) . This results in a correct prediction of the otolith signal F^p ( Figure 5I , grey ) and therefore , as in previous simulations of active movement , the sensory mismatch for both the canal and otolith signals ( Figure 5L , magenta and gray , respectively ) and feedback signals ( not shown ) are null; and the final estimates , driven exclusively by the prediction , are accurate; G^ ( t ) = G^p ( t ) and Ω^ ( t ) =Ω^p ( t ) . During passive tilt , the canal error , δV , is converted into Kalman feedback that updates Ω^ ( Figure 5K , blue ) and C^ ( not shown here; but see Figure 5—figure supplement 1 for 2 s tilt simulations ) , as well as the two other state variables ( G^ and A^ ) . Specifically , the feedback from δV to G^ ( Gk ) updates the predicted tilt G^p and is temporally integrated by the Kalman filter ( G^ ( t ) = ∫0tGk; see Supplementary methods , ‘Passive Tilt’; Figure 5K , green ) . The feedback signal from δV to A^ has a minimal impact , as illustrated in Figure 5K , red ( see also Supplementary methods , ’ Kalman feedback gains’ and Table 2 ) . Because δV efficiently updates the tilt estimate G^ , the otolith error δF is close to zero during passive tilt ( Figure 5L , gray; see Supplementary methods , ‘Passive Tilt’ ) and therefore all feedback signals originating from δF ( Figure 5J ) play a minimal role ( see Supplementary methods , ‘Passive Tilt’ ) during pure tilt ( this is the case even for longer duration stimuli; Figure 5—figure supplement 1 ) . This simulation highlights that , although tilt is sensed by the otoliths , passive tilt doesn’t induce any sizeable otolith error . Thus , unlike neurons tuned to canal error , the model predicts that those cells tuned to otolith error will not modulate during either passive or actively-generated head tilt . Therefore , cells tuned to otolith error would respond primarily during translation , and not during tilt , thus they would be identified ‘translation-selective’ . Furthermore , the model predicts that those neurons tuned to passive tilt ( e . g . Purkinje cells in the caudal cerebellar vermis; Laurens et al . , 2013b ) likely reflect a canal error that has been transformed into a tilt velocity error ( Figure 5L , magenta ) . Thus , the model predicts that tilt-selective Purkinje cells should encode tilt velocity , and not tilt position , a prediction that remains to be tested experimentally ( see Discussion ) . Next , we simulate a brief translation ( Figure 6 ) . During active translation , we observe , as in previous simulations of active movements , that the predicted head motion matches the sensory ( otolith in this case: F=A ) signals ( A^p=A and F^p=F ) . Therefore , as in previous simulations of active motion , the sensory prediction error is zero ( Figure 6L ) and the final estimate is equal to , and driven by , the prediction ( A^=A^p=A; Figure 6G , red ) . During passive translation , the predicted acceleration is null ( A^p=0 , Figure 6H , red ) , similar as during passive rotation in Figures 3 and 4 ) . However , a sizeable tilt signal ( G^p and G^ , Figure 6G , H , green ) , develops over time . This ( erroneous ) tilt estimate can be explained as follows: soon after translation onset ( vertical dashed lines in Figure 6B–J ) , G^p is close to zero . The corresponding predicted otolith signal is also close to zero ( F^p= A^p+G^p=0 ) , leading to an otolith error δF ≈A ( Figure 6L , right , gray ) . Through the Kalman feedback gain matrix , this otolith error , δF , is converted into: ( 1 ) an acceleration feedback Ak ( Figure 6J , red ) with gain kδFA=0 . 995 ( the close to unity feedback gain indicates that otolith errors are interpreted as acceleration: A^= δF≈A; note however that the otolith error δF vanishes over time , as explained next ) ; and ( 2 ) a tilt feedback Gk ( Figure 6J , green ) , with kδFG=0 . 5 . δt . This tilt feedback , although too weak to have any immediate effect , is integrated over time ( G^ ( t ) = ∫0tGk; see Figure 5 and Supplementary methods , ‘Somatogravic effect’ ) , generating the rising tilt estimate G^ ( Figure 6G , green ) and G^p ( Figure 6H , green ) . The fact that the Kalman gain feedback from the otolith error to the G^ internal state generates the somatogravic effect is illustrated in Figure 6—figure supplement 1 , where a longer acceleration ( 20 s ) is simulated . At the level of final estimates ( perception ) , these simulations predict the occurrence of tilt illusions during sustained translation ( somatogravic illusion; Graybiel , 1952; Paige and Seidman , 1999 ) . Further simulations show how activation of the semicircular canals without a corresponding activation of the otoliths ( e . g . during combination of tilt and translation; Angelaki et al . ( 2004 ) ; Yakusheva et al . , 2007 ) leads to an otolith error ( Figure 6—figure supplement 2 ) and how signals from the otoliths ( that sense indirectly whether or not the head rotates relative to gravity ) can also influence the rotation estimate Ω^ at low frequencies ( Figure 6—figure supplement 3; this property has been extensively evaluated by Laurens and Angelaki , 2011 ) . These simulations demonstrate that the Kalman filter model efficiently simulates all previous properties of both perception and neural responses during passive tilt and translation stimuli ( see Discussion ) . The model analyzed so far has considered only vestibular sensors . Nevertheless , active head rotations often also activate neck proprioceptors , when there is an independent rotation of the head relative to the trunk . Indeed , a number of studies ( Kleine et al . , 2004; Brooks and Cullen , 2009; 2013; Brooks et al . , 2015 ) have identified neurons in the rostral fastigial nuclei that encode the rotation velocity of the trunk . These neurons receive convergent signals from the semicircular canals and neck muscle proprioception and , accordingly , are named ‘bimodal neurons’ , to contrast with ‘unimodal neurons’ , which encode passive head velocity . Because the bimodal neurons do not respond to active head and trunk movements ( Brooks and Cullen , 2013; Brooks et al . , 2015 ) , they likely encode feedback signals related to trunk velocity . We developed a variant of the Kalman filter to model both unimodal and bimodal neuron types ( Figure 7; see also Supplementary methods and Figure 7—figure supplement 1–3 ) . The model tracks the velocity of the trunk in space ΩTS and the velocity of the head on the trunk ΩHT as well as neck position ( N=∫ΩHT . dt ) . Sensory inputs are provided by the canals ( that sense the total head velocity , Ω= ΩTS+ΩHT ) , and proprioceptive signals from the neck musculature ( P ) , which are assumed to encode neck position ( Chan et al . , 1987 ) . In line with the simulations presented above , we find that , during active motion , the predicted sensory signals are accurate . Consequently , the Kalman feedback pathways are silent ( Figure 7—figure supplement 1–3; active motion is not shown in Figure 7 ) . In contrast , passive motion induces sensory errors and Kalman feedback signals . The velocity feedback signals ( elaborated in Figure 7—figure supplement 1–3 ) have been re-plotted in Figure 7 , where we illustrate head in space ( blue ) , trunk in space ( gray ) , and head on trunk ( red ) velocity ( neck position feedback signals are only shown in Figure 7—figure supplement 1–3 ) . During passive whole head and trunk rotation , where the trunk rotates in space ( Figure 7A , Real motion: ΩTS>0 , grey ) and the head moves together with the trunk ( head on trunk velocity ΩHT=0 , red , head in space Ω>0 , blue ) , we find that the resulting feedback signals accurately encode these rotation components ( Figure 7A , Velocity Feedback; see also Figure 7—figure supplement 1 ) . During head on trunk rotation ( Figure 7B , Figure 7—figure supplement 2 ) , the Kalman feedback signals accurately encode the head on trunk ( red ) or in space ( blue ) rotation , and the absence of trunk in space rotation ( gray ) . Finally , during trunk under head rotation that simulates a rotation of the trunk while the head remains fixed in space , resulting in a neck counter-rotation , the various motion components are accurately encoded by Kalman feedback ( Figure 7C , Figure 7—figure supplement 3 ) . We propose that unimodal and bimodal neurons reported in ( Brooks and Cullen , 2009; 2013 ) encode feedback signals about the velocity of the head in space ( Ωk , Figure 7 , blue ) and of the trunk in space ( ΩTSk , Figure 7 , gray ) , respectively . Furthermore , in line with experimental findings ( Brooks and Cullen , 2013 ) , these feedback pathways are silent during self-generated motion . The Kalman filter makes further predictions that are entirely consistent with experimental results . First , it predicts that proprioceptive error signals during passive neck rotation encode velocity ( Figure 7—figure supplement 3L; see Supplementary methods , ‘Feedback signals during neck movement’ ) . Thus , the Kalman filter explains the striking result that the proprioceptive responses of bimodal neurons encode trunk velocity ( Brooks and Cullen , 2009; 2013 ) , even if neck proprioceptors encode neck position . Note that neck proprioceptors likely encode a mixture of neck position and velocity at high frequencies ( Chan et al . , 1987; Mergner et al . , 1991 ) ; and additional simulations ( not shown ) based on this hypothesis yield similar results as those shown here . We used here a model in which neck proprioceptors encode position for simplicity , and in order to demonstrate that Kalman feedback signals encode trunk velocity even when proprioceptive signals encode position . Second , the model predicts another important property of bimodal neurons: their response gains to both vestibular ( during sinusoidal motion of the head and trunk together ) and proprioceptive ( during sinusoidal motion of the trunk when the head is stationary ) stimulation vary identically if a constant rotation of the head relative to the trunk is added , as an offset , to the sinusoidal motion ( Brooks and Cullen , 2009 ) . We propose that this offset head rotation extends or contracts individual neck muscles and affects the signal to noise ratio of neck proprioceptors . Indeed , simulations shown in Figure 7—figure supplement 4 reproduce the effect of head rotation offset on bimodal neurons . In agreement with experimental findings , we also find that simulated unimodal neurons are not affected by these offsets ( Figure 7—figure supplement 4 ) . Finally , the model also predicts the dynamics of trunk and head rotation perception during long-duration rotations ( Figure 7—figure supplement 5 ) , which has been established by behavioral studies ( Mergner et al . , 1991 ) . The theoretical framework of the Kalman filter asserts that the brain uses a single internal model to process copies of motor commands and sensory signals . But could alternative computational schemes , involving distinct internal models for motor and sensory signals , explain neuronal and behavioral responses during active and passive motions ? Here , we consider three possibilities , illustrated in Figure 1—figure supplement 1 . First , that the brain computes head motion based on motor commands only and suppresses vestibular sensory inflow entirely during active motion ( Figure 1—figure supplement 1A ) . Second , that a ‘motor’ internal model and a ‘sensory’ internal model run in parallel , and that central neurons encode the difference between their outputs – which would represent a motion prediction error instead of a sensory prediction error , as proposed by the Kalman filter framework ( Figure 1—figure supplement 1B ) . Third , that the brain computes sensory prediction errors based on sensory signals and the output of the ‘motor’ internal model , and then feeds these errors into the ‘sensory’ internal model ( Figure 1—figure supplement 1C ) . We first consider the possibility that the brain simply suppresses vestibular sensory inflow . Experimental evidence against this alternative comes from recordings performed when passive motion is applied concomitantly to an active movement ( Brooks and Cullen , 2013; 2014; Carriot et al . , 2013 ) . Indeed , neurons that respond during passive but not active motion have been found to encode the passive component of combined passive and active motions , as expected based on the Kalman framework . We present corresponding simulation results in Figure 8 . We simulate a rotation movement ( Figure 8A ) , where an active rotation ( Ωu , Gaussian velocity profile ) is combined with a passive rotation ( Ωε , trapezoidal profile ) , a tilt movement ( Figure 8B; using similar velocity inputs , Ωu and Ωε , where the resulting active and passive tilt components are ∫Ωudt and ∫Ωεdt ) , and a translation movement ( Figure 8C ) . We find that , in all simulations , the final motion estimate ( Figure 8D–F; Ω^ , G^ and A^ , respectively ) matches the combined active and passive motions ( Ω , G and A , respectively ) . In contrast , the Kalman feedback signals ( Figure 8G–I ) specifically encode the passive motion components . Specifically , the rotation feedback ( Ωk , Figure 8G ) is identical to the passive rotation Ωε ( Figure 8A ) . As in Figure 5 , the tilt feedback ( Gk , Figure 8H ) encodes tilt velocity , also equal to Ωε ( Figure 8A ) . Finally , the linear acceleration feedback ( Ak , Figure 8I ) follows the passive acceleration component , although it decreases slightly with time because of the somatogravic effect . Thus , Kalman filter simulations confirm that neurons that encode sensory mismatch or Kalman feedback should selectively follow the passive component of combined passive and active motions . What would happen if , instead of computing sensory prediction errors , the brain simply discarded vestibular sensory ( or feedback ) signals during active motion ? We repeat the simulations of Figure 8A–I after removing the vestibular sensory input signals from the Kalman filter . We find that the net motion estimates encode only the active movement components ( Figure 8J–L; Ω^ , G^ and A^ ) – thus , not accurately estimating the true movement . Furthermore , as a result of the sensory signals being discarded , all sensory errors and Kalman feedback signals are null . These simulations indicate that suppressing vestibular signals during active motion would prevent the brain from detecting passive motion occurring during active movement ( see Discussion , ‘Role of the vestibular system during active motion: ecological , clinical and fundamental implications . ” ) , in contradiction with experimental results . Next , we simulate ( Figure 9 ) the alternative model of Figure 1—figure supplement 1B , where the motor commands are used to predict head motion ( Figure 9 , first row ) while the sensory signals are used to compute a self-motion estimate ( second row ) . According to this model , these two signals would be compared to compute a motion prediction error instead of a sensory prediction error ( third row; presumably represented in the responses of central vestibular neurons ) . We first simulate short active and passive rotations ( Figure 9A , B; same motion as in Figure 3 ) . During active rotation ( Figure 9A ) , both the motor prediction and the sensory self-motion estimate are close to the real motion and therefore the motor prediction is null ( Figure 9A , third row ) . In contrast , the sensory estimate is not cancelled during passive rotation , leading to a non-zero motion prediction error ( Figure 9B , third row ) . Thus , the motion prediction errors in Figure 9A , B resemble the sensory prediction errors predicted by the Kalman filter in Figure 3 and may explain neuronal responses recorded during brief rotations . However , this similarity breaks down when simulating a long-duration active or passive rotation ( Figure 9C , D; same motion as in Figure 4—figure supplement 1A , B ) . The motor prediction of rotation velocity would remain constant during 1 min of active rotation ( Figure 9C , first row ) , whereas the sensory estimate would decrease over time and exhibit an aftereffect ( Figure 9C , second row ) . This would result in a substantial difference between the motor prediction and the sensory estimate ( Figure 9C , third row ) during active motion . This contrasts with Kalman filter simulations , where no sensory prediction errors occur during active motion . A similar difference would also be seen during active translation ( Figure 9E; same motion as in Figure 6 ) . While the motion prediction ( first row ) would encode the active translation , the sensory estimate ( second row ) would be affected by the somatogravic effect ( as in Figure 6 ) , which causes the linear acceleration signal ( red ) to be replaced by a tilt illusion ( green ) , also leading to motion prediction errors ( third row ) . In contrast , the Kalman filter predicts that no sensory prediction error should occur during active translation . These simulations indicate that processing motor and vestibular information independently would lead to prediction errors that would be avoided by the Kalman filter . Beyond theoretical arguments , this scheme may be rejected based on behavioral responses: Both rotation perception and the vestibulo-ocular reflex ( VOR ) decrease during sustained passive rotations , but persist indefinitely during active rotation ( macaques: Solomon and Cohen , 1992 ) ; humans: Guedry and Benson ( 1983 ) ; Howard et al . ( 1998 ) ; Jürgens et al . , 1999 ) . In fact , this scheme cannot account for experimental findings , even if we consider different weighting for how the net self-motion signal is constructed from the independent motor and sensory estimates ( Figure 9 , bottom row ) . For example , if the sensory estimate is weighted 100% , rotation perception would decay during active motion ( Figure 9C , bottom , dark blue ) , inconsistent with experimental results . If the motor prediction is weighted 100% , passive rotations would not be detected at all ( Figure 9B , D , light blue ) . Finally , intermediate solutions ( e . g . 50%/50% ) would result in undershooting of both the steady state active ( Figure 9C ) and passive ( Figure 9B , D ) rotation perception estimates . Note also that , in all cases , the rotation after-effect would be identical during active and passive motion ( Figure 9C , D , bottom ) , in contradiction with experimental findings ( Solomon and Cohen , 1992; Guedry and Benson , 1983; Howard et al . , 1998 ) . Finally , the third alternative scheme ( Figure 1—figure supplement 1C ) , where sensory prediction error is used to cancel the input of a sensory internal model is , in fact , a more complicated version of the Kalman filter . This is because an internal model that processes motor commands to predict sensory signals must necessarily include an internal model of the sensors . Thus , simulations of the model in Figure 1—figure supplement 1C would be identical to the Kalman filter , by merely re-organizing the sequence of operations and uselessly duplicating some of the elements , to ultimately produce the same results . We have developed the first ever model that simulates self-motion estimates during both actively generated and passive head movements . This model , summarized schematically in Figure 10 , transforms motor commands and Kalman filter feedback signals into internal estimates of head motion ( rotation and translation ) and predicted sensory signals . There are two important take-home messages: ( 1 ) Because of the physical properties of the two vestibular sense organs , the predicted motion generated from motor commands is not equal to predicted sensory signals ( for example , the predicted rotation velocity is processed to account for canal dynamics in Figure 4 ) . Instead , the predicted rotation , tilt and translation signals generated by efference copies of motor commands must be processed by the corresponding forward models of the sensors in order to generate accurate sensory predictions . This important insight about the nature of these internal model computations has not been appreciated by the qualitative schematic diagrams of previous studies . ( 2 ) In an environment devoid of externally generated passive motion , motor errors and sensory noise , the resulting sensory predictions would always match sensory afferent signals accurately . In a realistic environment , however , unexpected head motion occurs due to both motor errors and external perturbations ( see ‘Role of the vestibular system during active motion: ecological , clinical and fundamental implications’ ) . Sensory vestibular signals are then used to correct internal motion estimates through the computation of sensory errors and their transformation into Kalman feedback signals . Given two sensory errors ( δV originating from the semicircular canals and δF originating from the otoliths ) and four internal state variables ( rotation , internal canal dynamics , tilt and linear acceleration: Ω^ , C^ , G^ , A^ ) , eight feedback signals must be constructed . However , in practice , two of these signals have negligible influence for all movements ( δV feedback to A^ and δF feedback to Ω^; see Table 2 and Supplementary methods , ‘Kalman Feedback Gains’ ) , thus only six elements are summarized in Figure 10 . The non-negligible feedback signals originating from the canal error δV are as follows ( Figure 10 , left ) : The non-negligible feedback signals originating from the otolith error δF are as follows ( Figure 9 , right ) : The model in Figure 10 is entirely compatible with previous models based on optimal passive self-motion computations ( Oman , 1982; Borah et al . , 1988; Merfeld , 1995; Laurens , 2006; Laurens and Droulez , 2007; Laurens and Droulez , 2008; Laurens and Angelaki , 2011; Karmali and Merfeld , 2012; Lim et al . , 2017; Zupan et al . , 2002 ) . The present model is , however , distinct in two very important aspects: First , it takes into account active motor commands and integrates these commands with the vestibular sensory signals . Second , because it is formulated as a Kalman filter , it makes specific predictions about the feedback error signals , which constitute the most important nodes in understanding the neural computations underlying head motion sensation . Indeed , as will be summarized next , the properties of most cell types in the vestibular and cerebellar nuclei , as well as the vestibulo-cerebellum , appear to represent either sensory error or feedback signals . Multiple studies have reported that vestibular-only ( erroneous term to describe ‘non-eye-movement-sensitive’ ) neurons in the VN encode selectively passive head rotation ( McCrea and Luan , 2003; Roy and Cullen , 2001; 2004; Brooks and Cullen , 2014 ) or passive translation ( Carriot et al . , 2013 ) , but suppress this activity during active head movements . In addition , a group of rostral fastigial nuclei ( unimodal rFN neurons; Brooks and Cullen , 2013; Brooks et al . , 2015 ) also selectively encodes passive ( but not active ) rotations . These rotation-responding VN/rFN neurons likely encode either the semicircular canal error δV itself or its Kalman feedback to the rotation estimate ( blue in Figure 10 , dashed and solid ovals ‘VN , rFN’ , respectively ) . The translation-responding neurons likely encode either the otolith error δF or its feedback to the linear acceleration estimate ( Figure 10 , solid and dashed red lines ‘VN , trans PC’ ) . Because error and feedback signals are proportional to each other in the experimental paradigms considered here , whether VN/rFN encode sensory errors or feedback signals cannot easily be distinguished using vestibular stimuli alone . Nevertheless , it is also important to emphasize that , while the large majority of VN and rFN neurons exhibit reduced responses during active head movements , this suppression is rarely complete ( McCrea et al . , 1999; Roy and Cullen , 2001; Brooks and Cullen , 2013; Carriot et al . , 2013 ) . Thus , neuronal responses likely encode mixtures of error/feedback and sensory motion signals ( e . g . such as those conveyed by direct afferent inputs ) . During large amplitude passive rotations ( Figure 4—figure supplement 3 ) , the rotation estimate persists longer than the vestibular signal ( Figure 4 , blue; a property called velocity storage ) . Because the internal estimate is equal to the canal error , this implies that VN neurons ( that encode the canal error ) should exhibit dynamics that are different from those of canal afferents , having incorporated velocity storage signals . This has indeed been demonstrated in VN neurons during optokinetic stimulation ( Figure 4—figure supplement 1; Waespe and Henn , 1977; Yakushin et al . , 2017 ) and rotation about tilted axes ( Figure 6—figure supplement 3; Reisine and Raphan , 1992; Yakushin et al . , 2017 ) . Based on the work summarized above , the final estimates of rotation ( Figure 4G ) and translation ( Figure 6G ) , which are the desirable signals to drive both perception and spatial navigation , do not appear to be encoded by most VN/rFN cells . Thus , one may assume that they are reconstructed downstream , perhaps in thalamic ( Marlinski and McCrea , 2008; Meng et al . , 2007; Meng and Angelaki , 2010 ) or cortical areas . Interestingly , more than half ( 57% ) of ventral thalamic neurons ( Marlinski and McCrea , 2008 ) and an identical fraction ( 57% ) of neurons of the VN cells projecting to the thalamus ( Marlinski and McCrea , 2009 ) respond similarly during passive and actively-generated head rotations . The authors emphasized that VN neurons with attenuated responses during actively-generated movements constitute only a small fraction ( 14% ) of those projecting to the thalamus . Thus , although abundant in the VN , these passive motion-selective neurons may carry sensory error/feedback signals to the cerebellum , spinal cord or even other VN neurons ( e . g . those coding the final estimates; Marlinski and McCrea , 2009 ) . Note that Dale and Cullen , 2016 , reported contrasting results where a large majority of ventral thalamus neurons exhibit attenuated responses during active motion . Even if not present in the ventral posterior thalamus , this signal should exist in the spatial perception/spatial navigation pathways . Thus , future studies should search for the neural correlates of the final self-motion signal . VN neurons identified physiologically to project to the cervical spinal cord do not to modulate during active rotations , so they could encode either passive head rotation or active and passive trunk rotation ( McCrea et al . , 1999 ) . Furthermore , the dynamics of the thalamus-projecting VN neurons with similar responses to passive and active stimuli were not measured ( Marlinski and McCrea , 2009 ) . Recall that the model predicts that final estimates of rotation differ from canal afferent signals only in their response dynamics ( Figure 4 , compare panels F and G ) . It would make functional sense that these VN neurons projecting to the thalamus follow the final estimate dynamics ( i . e . , they are characterized by a prolonged time constant compared to canal afferents ) – and future experiments should investigate this hypothesis . Another class of rFN neurons ( and possibly VN neurons projecting to the thalamus; Marlinski and McCrea , 2009 , or those projecting to the spinal cord; McCrea et al . , 1999 ) specifically encodes passive trunk velocity in space , independently of head velocity ( bimodal neurons; Brooks and Cullen , 2009; 2013; Brooks et al . , 2015 ) . These neurons likely encode Kalman feedback signals about trunk velocity ( Figure 7 , blue ) . Importantly , these neurons respond equivalently to passive whole trunk rotation when the trunk and the head rotate together ( Figure 7A ) and to passive trunk rotation when the head is space-fixed ( Figure 7C ) . The first protocol activates the semicircular canals and induces a canal error δV , while the later activates neck proprioceptors and generates a proprioceptive error , δP . From a physiological point of view , this indicates that bimodal neurons respond to semicircular canals as well as neck proprioceptors ( hence their name ) . Note that several other studies identified VN ( Anastasopoulos and Mergner , 1982 ) , rFN ( Kleine et al . , 2004 ) and anterior suprasylvian gyrus ( Mergner et al . , 1985 ) neurons that encode trunk velocity during passive motion , but didn’t test their response to active motion . The Kalman filter also predicts that neck proprioceptive signals that encode neck position should be transformed into error signals that encode neck velocity . In line with model predictions , bimodal neurons encode velocity signals that originate from neck proprioception during passive sinusoidal ( 1 Hz , Brooks and Cullen , 2009 ) and transient ( Gaussian velocity profile , Brooks and Cullen , 2013 ) movements . Remarkably , although short-duration rotation of the trunk while the head is stationary in space leads to a veridical perception of trunk rotation , long duration trunk rotation leads to an attenuation of the perceived trunk rotation and a growing illusion of head rotation in the opposite direction ( Mergner et al . , 1991 ) . These experimental findings are also predicted by the Kalman filter model ( Figure 7—figure supplement 5 ) . The simple spike modulation of two distinct types of Purkinje cells in the caudal cerebellar vermis ( lobules IX-X , Uvula and Nodulus ) encodes tilt ( tilt-selective cells ) and translation ( translation-selective cells ) during three-dimensional motion ( Yakusheva et al . , 2007 , 2008 , 2013; Laurens et al . , 2013a; Laurens et al . , 2013b ) . Therefore , it is possible that tilt- and translation selective cells encode tilt and acceleration feedback signals ( Figure 10 , green and red lines , respectively ) . If so , we hypothesize that their responses are suppressed during active motion ( Figures 5 and 6 ) . How Purkinje cells modulate during active motion is currently unknown . However , one study ( Lee et al . , 2015 ) performed when rats learned to balance on a swing indicates that Purkinje cell responses that encode trunk motion are reduced during predictable movements , consistent with the hypothesis that they encode sensory errors or Kalman feedback signals . Model simulations have also revealed that passive tilt does not induce any significant otolith error ( Figure 5J ) . In contrast , passive tilt elicits a significant canal error ( Figure 5K ) . Thus , we hypothesize that the tilt signal present in the responses of Purkinje cells originates from the canal error δV onto the tilt internal state variable . If it is indeed a canal , rather than an otolith , error , it should be proportional to tilt velocity instead of tilt position ( or linear acceleration ) . Accordingly , we observed ( Laurens et al . , 2013b ) that tilt-selective Purkinje cell responses were on average close to velocity ( average phase lag of 36° during sinusoidal tilt at 0 . 5 Hz ) . However , since sinusoidal stimuli are not suited for establishing dynamics ( Laurens et al . , 2017 ) , further experiments are needed to confirm that tilt-selective Purkinje cells indeed encode tilt velocity . Model simulations have also revealed that passive translation , unlike passive tilt , should include an otolith error . This otolith error feeds also into the tilt internal variable ( Figure 9 , somatogravic feedback ) and is responsible for the illusion of tilt during sustained passive linear acceleration ( somatogravic effect; Graybiel , 1952 ) . Therefore , as summarized in Figure 10 ( green lines ) , both canal and otolith errors should feedback onto the tilt internal variable . The canal error should drive modulation during tilt , whereas the otolith error should drive modulation during translation . In support of these predictions , we have demonstrated that tilt-selective Purkinje cells also modulate during translation , with a gain and phase consistent with the simulated otolith-driven feedback ( Laurens et al . , 2013b ) . Thus , both of these feedback error signals might be carried by caudal vermis Purkinje cells – and future experiments should address these predictions . Note that semicircular canal errors must be spatially transformed in order to produce an appropriate tilt feedback . Indeed , converting a rotation into head tilt requires taking into account the angle between the rotation axis and earth-vertical . This transformation is represented by a bloc marked ‘3D’ in Figure 9 ( see also ( eq . 9 ) in Supplemenatry methods , ‘Three-Dimensional Kalman filter’ . Importantly , we have established ( Laurens et al . , 2013b ) that tilt-selective Purkinje cells encode spatially transformed rotation signals , as predicted by theory . In fact , we have demonstrated that tilt-selective Purkinje cells do not simply modulate during vertical canal stimulation , but also carry the tilt signal during off-vertical axis yaw rotations ( Laurens et al . , 2013b ) . In this respect , it is important to emphasize that truly tilt-selective neurons exclusively encode changes in orientation relative to gravity , rather than being generically activated by vertical canal inputs . Thus , it is critical that this distinction is experimentally made using three-dimensional motion ( see Laurens et al . , 2013b; Laurens and Angelaki , 2015 ) . Whereas 3D rotations have indeed been used to identify tilt-selective Purkinje cells in the vermis ( Laurens et al . , 2013b; Yakusheva et al . , 2007 ) , this is not true for other studies . For example , Siebold et al . , 1997 , Siebold et al . , 1999 , 2001 ) , Laurens and Angelaki , 2015 and Zhou et al . ( 2006 ) have reported tilt-modulated cells in the rFN and VN , respectively , but because these neurons were not tested in three dimensions , the signals carried by these neurons remain unclear . As summarized above , the simple spike responses of tilt-selective Purkinje cells during passive motion have already revealed many details of the internal model computations . Thus , we have proposed that tilt- selective Purkinje cells encode the feedback signals about tilt , which includes scaled and processed ( i . e . by a spatial transformation , green ‘3D’ box in Figure 10 ) versions of both canal and otolith sensory errors ( Figure 10 , green oval , ‘tilt PC ? ’ ) . However , there could be alternative implementations of the Kalman filter , where tilt-selective Purkinje cells may not encode only feedback signals , as proposed next: We note that motor commands Ωu must be also be spatially processed ( black ‘3D’ box in Figure 10 ) to contribute to the tilt prediction . One may question whether two distinct neuronal networks transform motor commands and canal errors independently ( resulting in two ‘3D’ boxes in Figure 10 ) . An alternative ( Figure 10—figure supplement 1 ) would be that the brain merges motor commands and canal error to produce a final rotation estimate prior to performing this transformation . From a mathematical point of view , this alternative would only require a re-arrangement of the Kalman filter equations , which would not alter any of the model’s conclusions . However , tilt-selective Purkinje cells , which encode a spatially transformed signal , would then carry a mixture of predictive and feedback signals and would therefore respond identically to active and passive tilt velocity . Therefore , the brain may perform a spatial transformation of predictive and feedback rotation signals independently ( Figure 10 ) ; or may merge them before transforming them ( Figure 10—figure supplement 1 ) . Recordings from tilt-selective Purkinje cells during active movements will distinguish between these alternatives . In summary , many of the response properties described by previous studies for vestibular nuclei and cerebellar neurons can be assigned a functional ‘location’ within the Kalman filter model . Interestingly , most of the central neurons fit well with the properties of sensory errors and/or feedback signals . That an extensive neural machinery has been devoted to feedback signals is not surprising , given their functional importance for self-motion estimation . For many of these signals , a distinction between sensory errors and feedback signals is not easily made . That is , rotation-selective VN and rFN neurons can encode either canal error ( Figure 10 , bottom , dashed blue oval ) or rotation feedback ( Figure 10 , bottom , solid blue oval ) . Similarly , translation-selective VN , rFN and Purkinje cells can encode either otolith error ( Figure 10 , bottom , dashed red oval ) or translation feedback ( Figure 10 , bottom , solid red oval ) . The only feedback that is easily distinguished based on currently available data is the tilt feedback ( Figure 10 , green lines ) . Although the blue , green and red feedback components of Figure 10 can be assigned to specific cell groups , this is not the case with the cyan feedback components . First , note that , like the tilt variable , the canal internal model variable , receives non-negligible feedback contributions from both the canal and otolith sensory errors ( Figure 10 , cyan lines ) . The canal feedback error changes the time constant of the rotation estimate ( Figure 4 and Figure 4—figure supplements 1 and 3 ) , whereas the otolith feedback error may suppress ( post-rotatory tilt ) or create ( horizontal axis rotation ) a rotation estimate ( Figure 6—figure supplement 3 ) . The neuronal implementations of the internal model of the canals ( C^ ) , and of its associated feedback pathways , are currently unknown . However , lesion studies clearly indicate that the caudal cerebellar vermis , lobules X and IX may influence the canal internal model state variable ( Angelaki and Hess , 1995a; Angelaki and Hess , 1995b; Wearne et al . , 1998 ) . In fact , it is possible that the simple-spike output of the translation-selective Purkinje cells also carries the otolith sensory error feedback to the canal internal model state variable ( Figure 10 , bottom , cyan arrow passing though the dashed red ellipse ) . Similarly , the canal error feedback to the canal internal model state variable ( Figure 10 , bottom , cyan arrow originating from the dashed blue ellipse ) can originate from VN or rFN cells that selectively encode passive , not active , head rotation ( Figure 4J , note that the Ck feedback is but a scaled-down version of the Ωk feedback ) . Thus , although the feedback error signals to the canal internal model variable can be linked to known neural correlates , cells coding for the state variable C^ exclusively have not been identified . It is possible that the hidden variable C^ may be coded in a distributed fashion . After all , as already stated above , VN and rFN neurons have also been shown to carry mixed signals - they can respond to both rotation and translation , as well as they may carry both feedback/error and actual sensory signals . Thus , it is important to emphasize that these Kalman variables and error signals may be represented in a multiplexed way , where single neurons manifest mixed selectivity to more than just one internal state and/or feedback signals . This appears to be an organizational principle both in central vestibular areas ( Laurens et al . , 2017 ) and throughout the brain ( Rigotti et al . , 2013; Fusi et al . , 2016 ) . It has been proposed that mixed selectivity has an important computational advantage: high-dimensional representations with mixed selectivity allow a simple linear readout to generate a diverse array of potential responses ( Fusi et al . , 2016 ) . In contrast , representations based on highly specialized neurons are low dimensional and may preclude a linear readout from generating several responses that depend on multiple task-relevant variables . In this treatment , we have considered primarily the importance of the internal models of the sensors to emphasize its necessity for both self-generated motor commands and unpredicted , external perturbations . It is important to point out that self-generated movements involve internal model computations that have been studied extensively in the field of motor control and motor adaptation ( Wolpert et al . , 1995; Körding and Wolpert , 2004; Todorov , 2004; Chen-Harris et al . , 2008; Berniker et al . , 2010; Berniker and Kording , 2011; Franklin and Wolpert , 2011; Saglam et al . , 2011; 2014 ) . While the question of motor adaptation are not addressed directly in the present study , experiments in which resistive or assistive torques are applied to the head ( Brooks et al . , 2015 ) or in which active movements are entirely blocked ( Roy and Cullen , 2004; Carriot et al . , 2013 ) reveal how central vestibular pathways respond in situations that cause motor adaptation . Under these conditions , central neurons have been shown to encode net head motion ( i . e . active and passive indiscriminately ) with a similar gain as during passive motion ( Figure 7—figure supplements 6 and 7 ) . This may be interpreted and modeled by assuming that central vestibular pathways cease to integrate copies of motor commands ( Figure 7—figure supplement 6 ) whenever active head motion is perturbed , until the internal model of the motor plant recalibrates to anticipate this perturbation ( Brooks et al . , 2015 ) . Further analysis of these experimental results ( Figure 7—figure supplement 7 ) indicate that they are fundamentally non-linear and cannot be reproduced by the Kalman filter ( which is limited to linear operations ) and therefore requires the addition of an external gating mechanism ( black pathway in Figure 1D ) . Notably , this nonlinearity is triggered with proprioceptive mismatch , that is , when there is a discrepancy between the intended head position and proprioceptive feedback . Note that perturbing head motion also induces a vestibular mismatch since it causes the head velocity to differ from the motor plan . However , central vestibular neurons still encode specifically passive head movement during vestibular mismatch , as can be shown by superimposing passive whole body rotations to active movements ( Brooks and Cullen , 2013; 2014; Carriot et al . , 2013 ) and illustrated in the model predictions of Figure 8 . Remarkably , the elementary and fundamental difference between these different types of computations has never before been presented in a single theoretical framework . Proprioceptive mismatch is likely a specific indication that the internal model of the motor plant ( necessary for accurate motor control; Figure 1D , red ) needs to be recalibrated . Applying resistive head torques ( Brooks et al . , 2015 ) or increasing head inertia ( Saglam et al . , 2011; 2014 ) does indeed induce motor adaptation which is not modeled in the present study ( but see Berniker and Kording , 2008 ) . Interestingly , the studies by Saglam et al . ( 2011 ) , 2014 ) indicate that healthy subjects use a re-calibrated model of the motor plant to restore optimal motor performance , but that vestibular deficient patients fail to do so , indicating that vestibular error signals participate in motor adaptation ( Figure 1D , broken blue arrow ) . The internal model framework has been widely used to simulate optimal motor control strategies ( Todorov , 2004; Chen-Harris et al . , 2008; Saglam et al . , 2011; 2014 ) and to create Kalman filter models of reaching movements ( Berniker and Kording , 2008 ) and postural control ( van der Kooij et al . , 2001 ) . The present model , however , is to our knowledge the first to apply these principles to optimal head movement perception during active and passive motion . As such , it makes explicit links between sensory dynamics ( i . e . the canals ) , ambiguities ( i . e . the otoliths ) , priors and motor efference copies . Perhaps most importantly , the focus of this study has been to explain neuronal response properties . By simulating and explaining neuronal responses during active and passive self-motion in the light of a quantitative model , this study advances our understanding of how theoretical principles about optimal combinations of motor signals , multiple sensory modalities with distinct dynamic properties and ambiguities and Bayesian priors map onto brainstem and cerebellar circuits . To simplify the main framework and associated predictions , as well as the in-depth mathematical analyses of the model’s dynamics ( Supplementary methods ) , we have presented a linearized one-dimensional model . This model was used to simulate either rotations around an earth-vertical axis or combinations of translation and rotations around an earth-horizontal axis . A more natural and general way to simulate self-motion information processing is to design a three-dimensional Kalman filter model . Such models have been used in previous studies , either by programming Kalman filters explicitly ( Borah et al . , 1988; Lim et al . , 2017 ) , or by building models based on the Kalman filter framework ( Glasauer , 1992; Merfeld , 1995; Glasauer and Merfeld , 1997; Bos et al . , 2001; Zupan et al . , 2002 ) . We show in Supplementary methods , ‘Three-dimensional Kalman filter’ , how to generalize the model to three dimensions . The passive motion components of the model presented here are to a large extent identical to the Particle filter Bayesian model in ( Laurens , 2006; Laurens and Droulez , 2007 , Laurens and Droulez , 2008; Laurens and Angelaki , 2011 ) , which we have re-implemented as a Kalman filter , and into which we incorporated motor commands . One fundamental aspect of previous Bayesian models ( Laurens , 2006; Laurens and Droulez , 2007 , Laurens and Droulez , 2008 ) is the explicit use of two Bayesian priors that prevent sensory noise from accumulating over time . These priors encode the natural statistics of externally generated motion or motion resulting from motor errors and unexpected perturbations . Because , on average , rotation velocities and linear accelerations are close to zero , these Bayesian priors are responsible for the decrease of rotation estimates during sustained rotation ( Figure 4—figure supplement 2 ) and for the somatogravic effect ( Figure 6—figure supplement 2 ) ( see Laurens and Angelaki , 2011 ) for further explanations ) . The influence of the priors is higher when the statistical distributions of externally generated rotation ( Ωε ) and acceleration ( Aε ) are narrower ( Figure 10—figure supplement 2 ) , that is when their standard deviation is smaller . Stronger priors reduce the gain and time constant of rotation and acceleration estimates ( Figure 10—figure supplement 2B , D ) . Importantly , the Kalman filter model predicts that the priors affect only the passive , but not the active , self-motion final estimates . Indeed , the rotation and acceleration estimates last indefinitely during simulated active motion ( Figure 4—figure supplement 2 , Figure 6—figure supplement 2 , Figure 10—figure supplement 2 ) . In this respect , the Kalman filter may explain why the time constant of the vestibulo-ocular reflex is reduced in figure ice skaters ( Tanguy et al . , 2008; Alpini et al . , 2009 ) : The range of head velocities experienced in these activities is wider than normal . In previous Bayesian models , we found that widening the rotation prior should increase the time constant of vestibular responses , apparently in contradiction with these experimental results . However , these models did not consider the difference between active and passive stimuli . The formalism of the Kalman filter reveals that Bayesian priors should reflect the distribution of passive motion or motor errors . In athletes that are highly trained to perform stereotypic movements , this distribution likely narrows , resulting in stronger priors and reduced vestibular responses . One of the predictions of the Kalman filter is that motion illusions , such as the disappearance of rotation perception during long-duration rotation and the ensuing post-rotatory response ( Figure 4—figure supplement 1B ) should not occur during active motion ( Figure 4—figure supplement 1A ) . This has indeed been observed in monkeys ( Solomon and Cohen , 1992 ) and humans , where steady-state per-rotatory responses plateau at 10°/s and post-rotatory responses are decreased by a similar amount ( Guedry and Benson , 1983; Howard et al . , 1998 ) ; see also Brandt et al . , 1977a ) . The fact that post-rotatory responses are reduced following active , as compared to passive , rotations is of particular interest , because it demonstrates that motor commands influence rotation perception even after the movement has stopped . As shown in Figure 4 , The Kalman filter reproduces this effect by feeding motor commands though an internal model of the canals . As shown in Figure 4—figure supplement 1 , this process is equivalent to the concept of ‘velocity storage’ ( Raphan et al . , 1979; see Laurens , 2006 , Laurens and Droulez , 2008 , MacNeilage et al . ( 2008 ) , Laurens and Angelaki ( 2011 ) for a Bayesian interpretation of the concept of velocity storage ) . Therefore , the functional significance of this network , including velocity storage , is found during natural active head movements ( see also Figure 4—figure supplement 3 ) , rather than during passive low-frequency rotations with which it has been traditionally associated with in the past ( but see Laurens and Angelaki , 2011 ) . A recent study ( MacNeilage and Glasauer , 2017 ) evaluated how motor noise varies across locomotor activities and within gait cycles when walking . They found that motor noise peaks shortly before heel strike and after toe off; and is minimal during swing periods . They interpreted experimental findings using principles of sensory fusion , an approach that uses the same principles of optimal cue combination as the Kalman filter but doesn’t include dynamics . Interestingly , this analysis showed that vestibular cues should have a maximal effect when motor noise peaks , in support with experimental observations ( Brandt et al . , 1999; Jahn et al . , 2000 ) . To avoid further complications to the solution to the Kalman filter gains , the presented model does not consider how the brain generates motor commands in response to vestibular stimulation , e . g . to stabilize the head in response to passive motion or to use vestibular signals to correct motor commands . This would require an additional feedback pathway - the reliance of motor command generation on sensory estimates ( Figure 1D , blue broken arrow ) . For example , a passive head movement could result in a stabilizing active motor command . Or an active head movement could be less than desired because of noise , requiring an adjustment of the motor command to compensate . These feedback pathways have been included in previous Kalman filter models ( e . g . van der Kooij et al . , 2001 ) , a study that focused specifically on postural control and reproduced human postural sway under a variety of conditions . Thus , the Kalman filter framework may be extended to model neuronal computations underlying postural control as well as the vestibulo-collic reflex . Neuronal recordings ( Brooks and Cullen , 2013; 2014; Carriot et al . , 2013 ) and the present modeling unambiguously demonstrate that central neurons respond to unexpected motion during active movement ( a result that we reproduced in Figure 8G–I ) . Beyond experimental manipulations , a number of processes may cause unpredictable motion to occur in natural environments . When walking on tree branches , boulders or soft grounds , the support surface may move under the feet , leading to unexpected trunk motion . A more dramatic example of unexpected trunk motion , that requires immediate correction , occurs when slipping or tripping . Complex locomotor activities involve a variety of correction mechanism among which spinal mechanisms and vestibular feedback play preeminent roles ( Keshner et al . , 1987; Black et al . , 1988; Horstmann and Dietz , 1988 ) . The contribution of the vestibular system for stabilizing posture is readily demonstrated by considering the impact of chronic bilateral vestibular deficits . While most patients retain an ability to walk on firm ground and even perform some sports ( Crawford , 1964; Herdman , 1996 ) , vestibular deficit leads to an increased incidence of falls ( Herdman et al . , 2000 ) , difficulties in walking on uneven terrains and deficits in postural responses to perturbations ( Keshner et al . , 1987; Black et al . , 1988; Riley , 2010 ) . This confirms that vestibular signals are important during active motion , especially in challenging environments . In this respect , the Kalman filter framework appears particularly well suited for understanding the effect of vestibular lesions . As mentioned earlier , vestibular sensory errors also occur when the internal model of the motor apparatus is incorrect ( Brooks et al . , 2015 ) and these errors can lead to recalibration of internal models . This suggests that vestibular error signals during self-generated motion may play two fundamental roles: ( 1 ) updating self-motion estimates and driving postural or motor corrections , and ( 2 ) providing teaching signals to internal models of motor control ( Wolpert et al . , 1995 ) and therefore facilitating motor learning . This later point is supported by the finding that patients with vestibular deficits fail to recalibrate their motor strategies to account for changes in head inertia ( Sağlam et al . , 2014 ) . But perhaps most importantly , the model presented here should eliminate the misinterpretation that vestibular signals are ignored during self-generated motion – and that passive motion stimuli are old-fashioned and should no longer be used in experiments . Regarding the former conclusion , the presented simulations highlight the role of the internal models of canal dynamics and otolith ambiguity , which operate continuously to generate the correct sensory prediction during both active and passive movements . Without these internal models , the brain would be unable to correctly predict sensory canal and otolith signals and everyday active movements would lead to sensory mismatch ( e . g . for rotations , see Figure 4—figure supplements 2 and 3 ) . Thus , even though particular nodes ( neurons ) in the circuit show attenuated or no modulation during active head rotations , vestibular processing remains the same - the internal model is both engaged and critically important for accurate self-motion estimation , even during actively-generated head movements . Regarding the latter conclusion , it is important to emphasize that passive motion stimuli have been , and continue to be , extremely valuable in revealing salient computations that would have been amiss if the brain’s intricate wisdom was interrogated only with self-generated movements . Furthermore , a quantitative understanding of how efference copies and vestibular signals interact for accurate self-motion sensation is primordial for our understanding of many other brain functions , including balance and locomotor control . As stated in Berniker and Kording ( 2011 ) : ‘A crucial first step for motor control is therefore to integrate sensory information reliably and accurately’ , and practically any locomotor activity beyond reaching movements in seated subjects will affect posture and therefore recruit the vestibular sensory modality . It is thus important for both motor control and spatial navigation functions ( for which intact vestibular cues appear to be critical; Taube , 2007 ) to correct the misconception of incorrectly interpreting that vestibular signals are cancelled and thus are not useful during actively generated movements . By providing a state-of-the-art model of self-motion processing during active and passive motion , we are bridging several noticeable gaps between the vestibular and motor control/navigation fields . ‘A good model has a delightful way of building connections between phenomena that never would have occurred to one’ ( Robinson , 1977 ) . Four decades later , this beautifully applies here , where the mere act of considering how the brain should process self-generated motion signals in terms of mathematical equations ( instead of schematic diagrams ) immediately revealed a striking similarity with models of passive motion processing and , by motivating this work , opened an avenue to resolve a standing paradox in the field . The internal model framework and the series of quantitative models it has spawned have explained and simulated behavioral and neuronal responses to self-motion using a long list of passive motion paradigms , and with a spectacular degree of accuracy ( Mayne , 1974; Oman , 1982; Borah et al . , 1988; Glasauer , 1992; Merfeld , 1995; Glasauer and Merfeld , 1997; Bos et al . , 2001; Zupan et al . , 2002; Laurens , 2006; Laurens and Droulez , 2007 , Laurens and Droulez , 2008; Laurens and Angelaki , 2011; Karmali and Merfeld , 2012; Lim et al . , 2017 ) . Internal models also represent the predominant theoretical framework for studying motor control ( Wolpert et al . , 1995; Körding and Wolpert , 2004; Todorov , 2004; Chen-Harris et al . , 2008; Berniker et al . , 2010; Berniker and Kording , 2011; Franklin and Wolpert , 2011; Saglam et al . , 2011; 2014 ) . The vestibular system shares many common questions with the motor control field , such as that of 3D coordinate transformations and dynamic Bayesian inference , but , being considerably simpler , can be modeled and studied using relatively few variables . As a result , head movements represent a valuable model system for investigating the neuronal implementation of computational principles that underlie motor control . The present study thus offers the theoretical framework which will likely assist in understanding neuronal computations that are essential to active self-motion perception , spatial navigation , balance and motor activity in everyday life . In a Kalman filter ( Kalman , 1960 ) , state variables X are driven by their own dynamics ( matrix D ) , motor commands Xu and unpredictable perturbations resulting from motor noise and external influence Xε through the equation ( Figure 1—figure supplement 2A ) :Xt=D . Xt-1+M . Xut+E . Xε where matrices M and E reflect the response to motor inputs and perturbations , respectively . A set of sensors , grouped in a variable S , measure state variables transformed by a matrix T , and are modeled as:St=T . Xt+ Sη ( t ) where Sη is Gaussian sensory noise ( Figure 1—figure supplement 2A , right ) . The model assumes that the brain has an exact knowledge of the forward model , that is , of D , M , E and T as well as the variances of Xε and Sη . Furthermore , the brain knows the values of the motor inputs Xu and sensory signals S , but doesn’t have access to the actual values of Xε and Sη . At each time t , the Kalman filter computes a preliminary estimate ( also called a prediction ) X^p ( t ) =D . X^ ( t−1 ) +M . Xu ( t ) and a corresponding predicted sensory signal S^p ( t ) =T . X^p ( t ) ( Figure 1—figure supplement 2B ) . In general , the resulting state estimate X^p ( t ) and the predicted sensory prediction S^p ( t ) may differ from the real values X ( t ) and St because: ( 1 ) Xεt≠0 , but the brain cannot predict the perturbation Xε ( t ) , ( 2 ) the brain does not know the value of the sensory noise Sη ( t ) and ( 3 ) the previous estimate X^ ( t−1 ) used to compute X^p ( t ) could be incorrect . These errors are reduced using sensory information , as follows ( Figure 1—figure supplement 2B ) . First , this prediction S^p ( t ) and the sensory input St are compared to compute a sensory error δSt . Second , sensory errors are then transformed into a feedback Xkt=K . δSt where K is a matrix of feedback gains . Thus , an improved estimate at time t is X^ ( t ) = X^p ( t ) +K . δS ( t ) . The value of the feedback gain matrix K determines how sensory errors ( and therefore sensory signals ) are used to compute the final estimate X^t and is computed based on D , E , T and on the variances of Xε and Sη ( see Supplementary methods , ‘Kalman filter algorithm’ ) . In the case of the self-motion model , the motor commands Ωu and Au are inputs to the Kalman filter ( Figure 2 ) . Note that , while the motor system may actually control other variables ( such as forces or accelerations ) , we consider that these variables are converted into Ωu and Au . We demonstrate in Supplementary methods , ’ Model of motor commands’ that altering these assumptions does not alter our conclusions . In addition to motor commands , a variety of unpredictable factors such as motor noise and external ( passive ) motion also affect Ω and A ( MacNeilage and Glasauer , 2017 ) . The total rotation and acceleration components resulting from these factors are modeled as variables Ωε and Aε . Similar to ( Laurens , 2006; Laurens and Droulez , 2007 , Laurens and Droulez , 2008 ) we modeled the statistical distribution of these variables as Gaussians , with standard deviations σΩ and σA . Excluding vision and proprioception , the brain senses head motion though the semicircular canals ( that generate a signal V ) and the otoliths organs ( that generate a signal F ) . Thus , in initial simulations ( Figures 3–6 ) , the variable S encompasses V and F ( neck proprioceptors are added in Figure 7 ) . The semicircular canals are rotation sensors that , due to their mechanical characteristic , exhibit high-pass filter properties . These dynamics may be neglected for rapid movements of small amplitude ( such as Figure 3 ) but can have significant impact during natural movements ( Figure 4—figure supplement 3 ) . They are modeled using a hidden state variable C . The canals are also subject to sensory noise Vη . Taken both the noise and the dynamics into account , the canals signal is modeled as V= Ω-C+ Vη . The otolith organs are acceleration sensors . They exhibit negligible temporal dynamics in the range of motion considered here , but are fundamentally ambiguous: they sense gravitational as well as linear acceleration – a fundamental ambiguity resulting from Einstein’s equivalence principle ( Einstein , 1907 ) . Gravitational acceleration along the inter-aural axis depends on head roll position , modeled here as G= ∫Ω . dt . The otoliths encode the sum of A and G and is also affected by sensory noise Fη , such that the net otolith signal is F=A+G+Fη . How sensory errors are used to correct motion estimates depends on the Kalman gain matrix , which is computed by the Kalman algorithm such that the Kalman filter as a whole performs optimal estimation . In theory , the Kalman filter includes a total of 8 feedback signals , corresponding to the combination of two sensory errors ( canal and otolith errors ) and four internal states ( see Supplementary methods , ’ Kalman feedback gains’ ) . It is important to emphasize that the Kalman filter model is closely related to previous models of vestibular information processing . Indeed , simulations of long-duration rotation and visuo-vestibular interactions ( Figure 4—figure supplement 2 ) , as well as mathematical analysis ( Laurens , 2006 ) , demonstrate that C^ is equivalent to the ‘velocity storage’ ( Raphan et al . , 1979; Laurens and Angelaki , 2011 ) . These low-frequency dynamics , as well as visuo-vestibular interactions , were previously simulated and interpreted in the light of optimal estimation theory; and accordingly are reproduced by the Kalman filter model . The model presented here is to a large extent identical to the Particle filter Bayesian model in ( Laurens , 2006; Laurens and Droulez , 2007 , Laurens and Droulez , 2008; Laurens and Angelaki , 2011 ) . It should be emphasized that: ( 1 ) transforming the model into a Kalman filter didn’t alter the assumptions upon which the Particle filter was build; ( 2 ) introducing motor commands into the Kalman filter was a textbook process that did not require any additional assumptions or parameters; and ( 3 ) we used exactly the same parameter values as in Laurens , 2006 and Laurens and Droulez , 2008 ( with the exception of σF whose impact , however , is negligible , and of the model of head on trunk rotation that required additional parameters; see next section ) . The parameters of the Kalman filter model are directly adapted from previous studies ( Laurens , 2006; Laurens and Droulez , 2008 ) . Tilt angles are expressed in radians , rotation velocities in rad/s , and accelerations in g ( 1 g = 9 . 81 m/s2 ) . Note that a small linear acceleration A in a direction perpendicular to gravity will rotate the gravito-inertial force vector around the head by an angle α=sin−1 ( A ) ≈A . For this reason , tilt and small amplitude linear accelerations are expressed , in one dimension , in equivalent units that may be added or subtracted . The standard deviations of the unpredictable rotations ( Ωε ) and accelerations ( Aε ) are set to the standard deviations of the Bayesian a priori in Laurens , 2006 and Laurens and Droulez , 2008 , that is , σΩ= 0 . 7 rad/s ( Ωε ) and σA=0 . 3 g ( Aε ) . The standard deviation of the noise affecting the canals ( Vη ) was set to σV=0 . 175 rad/s ( as in Laurens , 2006 and Laurens and Droulez , 2008; see Figure 10—figure supplement 2 for simulations with different parameters ) . The standard deviation of the otolith noise ( Fη ) was set to σF=0 . 002 g ( 2 cm/s2 ) . We verified that values ranging from 0 to 0 . 01 g had no effect on simulation results . The time constant of the canals was set to τc=4s . Simulations used a time step of δt = 0 . 01 s . We verified that changing the value of the time step without altering other parameters had no effect on the results . We ran simulations using a variant of the model that included visual information encoding rotation velocity . The visual velocity signals were affected by sensory noise with a standard deviation σVis = 0 . 12 rad/s , as in Laurens and Droulez , 2008 . Another variant modeled trunk in space velocity ( ΩTS ) and head on trunk velocity ( ΩHT ) independently . The standard deviations of unpredictable rotations were set to σTS = 0 . 7 rad/s ( identical to σΩ ) and σHT = 3 . 5 rad/s . The standard deviation of sensory noise affecting neck afferents was set manually to σP = 0 . 0017 rad . We found that increasing the neck afferent noise reduces the gain of head on trunk and trunk in space velocity estimate ( Figure 7C ) ( e . g . by 60% for a tenfold increase in afferent noise ) . Reducing the value of this noise has little effect on the simulations . For simplicity , all simulations were run without adding the sensory noise Vη and Fη . These noise-free simulations are representative of the results that would be obtained by averaging several simulation runs performed with sensory noise ( e . g . as in Laurens and Droulez , 2007 ) . We chose to present noise-free results here in order to facilitate the comparison between simulations of active and passive motions . A Matlab implementation of the Kalman model is available at: https://github . com/JeanLaurens/Laurens_Angelaki_Kalman_2017 ( Laurens , 2017; copy archived at https://github . com/elifesciences-publications/Laurens_Angelaki_Kalman_2017 ) .	When seated in a car , we can detect when the vehicle begins to move even with our eyes closed . Structures in the inner ear called the vestibular , or balance , organs enable us to sense our own movement . They do this by detecting head rotations , accelerations and gravity . They then pass this information on to specialized vestibular regions of the brain . Experiments using rotating chairs and moving platforms have shown that passive movements – such as car journeys and rollercoaster rides – activate the brain’s vestibular regions . But recent work has revealed that voluntary movements – in which individuals start the movement themselves – activate these regions far less than passive movements . Does this mean that the brain ignores signals from the inner ear during voluntary movements ? Another possibility is that the brain predicts in advance how each movement will affect the vestibular organs in the inner ear . It then compares these predictions with the signals it receives during the movement . Only mismatches between the two activate the brain’s vestibular regions . To test this theory , Laurens and Angelaki created a mathematical model that compares predicted signals with actual signals in the way the theory proposes . The model accurately predicts the patterns of brain activity seen during both active and passive movement . This reconciles the results of previous experiments on active and passive motion . It also suggests that the brain uses similar processes to analyze vestibular signals during both types of movement . These findings can help drive further research into how the brain uses sensory signals to refine our everyday movements . They can also help us understand how people recover from damage to the vestibular system . Most patients with vestibular injuries learn to walk again , but have difficulty walking on uneven ground . They also become disoriented by passive movement . Using the model to study how the brain adapts to loss of vestibular input could lead to new strategies to aid recovery .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	A unified internal model theory to resolve the paradox of active versus passive self-motion sensation
Alternative pre-mRNA splicing expands the complexity of the transcriptome and controls isoform-specific gene expression . Whether alternative splicing contributes to metabolic regulation is largely unknown . Here we investigated the contribution of alternative splicing to the development of diet-induced obesity . We found that obesity-induced changes in adipocyte gene expression include alternative pre-mRNA splicing . Bioinformatics analysis associated part of this alternative splicing program with sequence specific NOVA splicing factors . This conclusion was confirmed by studies of mice with NOVA deficiency in adipocytes . Phenotypic analysis of the NOVA-deficient mice demonstrated increased adipose tissue thermogenesis and improved glycemia . We show that NOVA proteins mediate a splicing program that suppresses adipose tissue thermogenesis . Together , these data provide quantitative analysis of gene expression at exon-level resolution in obesity and identify a novel mechanism that contributes to the regulation of adipose tissue function and the maintenance of normal glycemia . Alternative pre-mRNA splicing is an important mechanism that increases the complexity of the transcriptome ( Pan et al . , 2008; Wang et al . , 2008 ) and expands the diversity and function of the proteome ( Nilsen and Graveley , 2010; Yang et al . , 2016 ) . Indeed , differences in pre-mRNA splicing can contribute to the specialization of different cell types within the body ( Vaquero-Garcia et al . , 2016 ) . The regulation of alternative pre-mRNA splicing is therefore an important aspect of cellular differentiation . Indeed , processes that regulate pre-mRNA splicing represent potential mechanisms that can control cell function . Recent studies have identified changes in pre-mRNA splicing associated with autism ( Irimia et al . , 2014; Weyn-Vanhentenryck et al . , 2014 ) , cardiac hypertrophy ( Mirtschink et al . , 2015 ) , embryonic stem cell re-programming ( Han et al . , 2013a ) , tumorigenesis ( Oltean and Bates , 2014; Hsu et al . , 2015; Koh et al . , 2015 ) , and the regulation of signal transduction pathways ( Gupta et al . , 1996; Tournier et al . , 1999; Maimon et al . , 2014; Martinez et al . , 2015 ) . Moreover , alternative pre-mRNA splicing associated with pathogenesis represents a tractable target for the development of new therapies ( Daguenet et al . , 2015 ) . The role of alternative pre-mRNA splicing in metabolism is unclear . We studied pre-mRNA splicing in adipocytes to investigate whether adipocyte function may be regulated by changes in pre-mRNA splicing . Adipose tissue is critically important for whole body metabolic regulation because it acts as both an endocrine organ and as a storage depot for triglyceride ( Rosen and Spiegelman , 2014 ) . Interestingly , both adipose tissue deficiency ( lipodystrophy ) and adipose tissue accumulation ( obesity ) are associated with the development of metabolic syndrome and pre-diabetes ( Grundy , 2015 ) . Moreover , the widespread incidence of human obesity represents a major risk factor for the development of diabetes and mortality ( Flegal et al . , 2013 ) . It is therefore important that we obtain an understanding of the molecular mechanisms that control adipocyte function . The purpose of this study was to examine alternative pre-mRNA splicing in adipocytes . We report that the consumption of a high fat diet causes differential exon inclusion/exclusion in the transcriptome . Bioinformatics analysis implicated a role for NOVA pre-mRNA splicing factors and this was confirmed by studies of mice with adipocyte-specific NOVA deficiency . Functional studies demonstrated that NOVA acts to suppress adipose tissue thermogenesis . Together , these data demonstrate that alternative pre-mRNA splicing contributes to the regulation of adipocyte biology . We examined gene expression in white epididymal adipocytes by RNA sequencing ( Illumina NextSeq 500 machine , 150 bp paired-end format , approximately 400 million mean reads/sample , n=3 ) ( Table 1 ) . Comparison of adipocyte mRNA isolated from mice fed ( 16 wk ) a chow diet ( CD ) or a high fat diet ( HFD ) demonstrated differential expression of 4941 genes ( q<0 . 05; absolute log2-fold change >0 . 75 ) and differential inclusion/exclusion of 1631 exons ( FDR<0 . 05; absolute change in exon inclusion ( absolute ∆Inc level ) >0 . 1 ) ( corresponding to 1249 genes ) in the transcriptome ( Figure 1A and Figure 1—figure supplements 1 and 2 ) . This differential exon inclusion/exclusion in mRNA most likely represents alternative pre-mRNA splicing in adipocytes . However , it is possible that some of the detected changes in pre-mRNA splicing reflect the differential presence of stromal vascular cells ( Figure 1—figure supplement 3 ) . Only 6 . 4% of the differentially expressed genes were alternatively spliced , but 25% of the alternatively spliced genes were differentially expressed ( Figure 1B ) . These data indicate that the genomic response to the consumption of a HFD causes quantitative changes in both gene expression and alternative pre-mRNA splicing in adipocytes . Analysis of the co-regulated genes demonstrated enrichment for pathways including mRNA processing and multiple signaling pathways ( e . g . insulin signaling and MAPK signaling ) that are known to regulate adipose tissue biology ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 17672 . 003Figure 1 . Diet-induced obesity causes changes in alternative pre-mRNA splicing in adipose tissue . ( A ) RNA-seq analysis demonstrates that the consumption ( 16 wk ) of a HFD , compared with a CD , causes significant changes in mRNA expression genes ( q<0 . 05; absolute log2-fold change >0 . 75 ) and differential exon inclusion/exclusion ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) in the epididymal adipocyte transcriptome . ( B ) Genes that are significantly differentially expressed and genes that are subjected to alternative pre-mRNA splicing are depicted using a Venn diagram . ( C , D ) Classification of HFD-induced alternative splicing events ( C ) and selected examples ( D ) are presented . ( E ) Enrichment of NOVA CLIP-seq tags with significantly ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) alternatively spliced exons ( ± 500 bp of intron/exon junctions ) compared with non-alternatively spliced exons ( FDR≥0 . 05 ) and random genomic sequences . ( F ) Immunoblot analysis of NOVA proteins in lysates prepared from hepatocytes and epididymal adipocytes . ( G ) Nova1 and Nova2 mRNA expression by epididymal adipocytes of wild-type and obese ob/ob mice ( 6 wk old ) was examined by quantitative RT-PCR ( mean ± SEM; n=5; p<0 . 05; *p<0 . 001 ) . The source data are included as Figure 1—source data 1 . ( H ) Wild-type mice were fed a CD or an HFD ( 16 wk ) . Nova1 and Nova2 mRNA expression by epididymal adipocytes was measured by quantitative RT-PCR analysis ( mean ± SEM; n=7; p<0 . 05 ) . The source data are included as Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00310 . 7554/eLife . 17672 . 004Figure 1—source data 1 . Source data for Figure 1G . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00410 . 7554/eLife . 17672 . 005Figure 1—source data 2 . Source data for Figure 1H . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00510 . 7554/eLife . 17672 . 006Figure 1—figure supplement 1 . Alternative pre-mRNA splicing in adipocytes . ( A–D ) Alternatively spliced exons identified by comparison of adipocytes of CD and HFD mice ( Figure 1 ) and comparison of adipocytes of FWT and F∆N1 , 2 mice ( Figure 2 ) are presented . The number of significantly alternatively spliced exons ( ∆Inc level >0 . 01 , >0 . 05 , >0 . 1 , >0 . 25 , >0 . 5 , and >0 . 75 ) is presented with FDR<0 . 05 ( A ) , FDR<0 . 01 ( B ) , FDR<0 . 001 ( C ) , and FDR<0 . 0001 ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00610 . 7554/eLife . 17672 . 007Figure 1—figure supplement 2 . Alternative pre-mRNA splicing in adipocytes . ( A ) Semi-quantitative RT-PCR analysis of epididymal adipocyte mRNA isolated from CD-fed mice , HFD-fed mice , and F∆N1 , 2 mice ( NOVA DKO ) was performed to detect alternative splicing of Adam15 mRNA ( skipping exon 20 ) and Yap1 mRNA ( skipping exon 6 ) . ( B ) Mutually exclusive inclusion of exons 7a and 7b in Fyn mRNA by adipocytes of CD-fed and HFD-fed mice is presented as a sashimi plot . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00710 . 7554/eLife . 17672 . 008Figure 1—figure supplement 3 . Expression of adipocyte and stromal vascular fraction marker genes . Epididymal fat pads of CD-fed and HFD-fed ( 16 wk ) wild-type mice were used to isolate adipocytes and the stromal vascular fraction ( SVF ) . The expression of adipocyte and SVF marker genes were examined by measurement of mRNA by quantitative RT-PCR ( mean ± SEM; n = 4~6; p<0 . 05; *p<0 . 001 ) . These data demonstrate that adipocyte marker genes Adipoq and Leptin were not expressed by the SVF , but low level expression of the SVF marker genes Emr1 ( F4/80 ) , Itgam ( Cd11b ) , and Cd68 was detected the adipocyte fraction . This SVF contamination of the adipocyte fraction may be mediated by lipid-loaded macrophages ( Xu et al . , 2013 ) . Control studies were performed using a gene ( Fabp4 ) that is expressed by both adipocytes and SVF . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00810 . 7554/eLife . 17672 . 009Figure 1—figure supplement 4 . Biological pathway enrichment analysis of differential gene expression and alternative pre-mRNA splicing caused by feeding a HFD . Genes that are differentially expressed and/or alternatively spliced in response to the consumption of a HFD ( Figure 1B ) were examined by pathway analysis to detect significant enrichment for specific biological processes . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 00910 . 7554/eLife . 17672 . 010Figure 1—figure supplement 5 . NOVA expression in human subcutaneous adipose tissue . NOVA1 and NOVA2 mRNA expression in abdominal subcutaneous adipose of obese ( BMI>30 kg/m2 ) and non-obese ( BMI<30 kg/m2 ) matched humans was examined ( mean ± SEM; n = 30 obese , n = 26 non-obese; p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01010 . 7554/eLife . 17672 . 011Table 1 . Summary of RNA-seq data . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 011GEO Accession Subseries/ SuperseriesBiological groupsSample numberPlatformMean read number / sample ( after trimming , if applicable ) Read length ( after trimming , if applicable ) Mean read alignment rateGSE76294/ GSE76134FWT ( 3 ) F∆N1 ( 3 ) F∆N2 ( 3 ) 9Illumina HiSeq 2000/ 2 x 100 bp135 , 500 , 000100 bp89 . 4%GSE76133/ GSE76134CD ( 3 ) HFD ( 3 ) 6Illumina NextSeq 500/ 2 x 150 bp406 , 200 , 00090 bp74 . 9%GSE76317/ GSE76134FWT ( 4 ) F∆N1 , 2 ( 4 ) 8Illumina NextSeq 500/ 2 x 150 bp319 , 700 , 000130 bp92 . 5% The most common form of HFD-induced alternative pre-mRNA splicing was exon skipping ( 1052 exons ) , but we also detected 144 mutually exclusive exon inclusions , 160 retained introns , 123 alternative 5’ splice sites , and 152 alternative 3’ splice sites ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) ( Figure 1C , D ) . In contrast to the extensive changes in alternative pre-mRNA splicing in adipocytes caused by diet-induced obesity ( Figure 1A–C ) few HFD-regulated alternative pre-mRNA splicing events were detected in liver , although we did find 2 skipped exons , 0 mutually exclusive exon inclusions , 13 retained introns , 1 alternative 5’ splice site , and 1 alternative 3’ splice site ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) . This comparative analysis of gene expression indicates that widespread changes in pre-mRNA splicing are not a general response to diet-induced obesity , but a selective response of white adipocytes to the consumption of a HFD . One example of altered splicing is the mutually exclusive inclusion of exons 7a or 7b in the tyrosine kinase FYN that changes the strength of SH3 domain-mediated autoinhibition ( Brignatz et al . , 2009 ) . Increased inclusion of Fyn exon 7b , compared with exon 7a , in response to the consumption of an HFD ( Figure 1—figure supplement 2B ) is anticipated to increase FYN tyrosine kinase activity ( Brignatz et al . , 2009 ) leading to suppression of fatty acid oxidation and promotion of insulin resistance ( Bastie et al . , 2007 ) . Together , these data indicate that alternative pre-mRNA splicing may contribute to the adipocyte response to obesity . To gain insight into the mechanism of HFD-induced pre-mRNA splicing , we examined exons ( plus 500 bp of flanking intron sequence ) to identify potential motifs that were significantly enriched for alternatively spliced exons . This analysis led to the identification of potential binding sites ( YCAY ) for NOVA alternative pre-mRNA splicing factors ( Darnell , 2013 ) . Indeed , we found significant ( p<2 . 2 × 10−16 ) enrichment of UV-mediated cross-linking and immunoprecipitation sequencing ( CLIP-seq ) tags for NOVA proteins identified in brain tissue ( Licatalosi et al . , 2008 ) within the HFD-induced group of alternatively spliced exons in adipocytes ( Figure 1E ) . The NOVA CLIP-seq tags intersected with 56% of the HFD-induced alternatively spliced exons ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) ( Figure 1E ) and were associated with both HFD-induced exon inclusion ( 53% intersection ) and HFD-induced exon exclusion ( 62% intersection ) , consistent with the known role of NOVA proteins to cause context-dependent exon inclusion/exclusion ( Ule et al . , 2006 ) . This analysis implicates a role for NOVA proteins in a sub-set of HFD-induced alternative splicing events . NOVA proteins are expressed in several tissues , including neurons ( Darnell , 2013 ) , vascular endothelial cells ( Giampietro et al . , 2015 ) , and pancreatic β cells ( Villate et al . , 2014 ) . Whether NOVA proteins are expressed in peripheral metabolic tissues is unclear . Indeed , NOVA proteins were not detected in liver , but both NOVA1 and NOVA2 proteins were found in white adipocytes ( Figure 1F ) . Interestingly , Nova gene expression in white adipocytes was partially reduced in obese humans and mice ( Figure 1G and Figure 1—figure supplement 5 ) and in mice fed a HFD ( Figure 1H ) . This decrease in NOVA expression may be relevant to obesity-regulated changes in alternative pre-mRNA splicing in adipocytes . Moreover , the absence of NOVA expression in liver ( Figure 1F ) may contribute to the minimal effect of HFD consumption on alternative splicing of pre-mRNA in the liver . Reduced expression of Nova1 or Nova2 in mice causes developmental defects and neonatal lethality ( Jensen et al . , 2000; Ruggiu et al . , 2009 ) . We therefore established Nova1LoxP/LoxP and Nova2LoxP/LoxP mice to study the role of NOVA proteins in adult mice with tissue-specific NOVA deficiency ( Figure 2—figure supplement 1 ) . Adipoq-Cre−/+ mice were used to selectively ablate Nova1 and Nova2 genes in mature adipocytes . We initially compared control FWT mice ( Adipoq-Cre−/+ ) with compound mutant F∆N1 , 2 mice ( Adipoq-Cre−/+ Nova1LoxP/LoxP Nova2LoxP/LoxP ) . RNA-seq analysis of white adipocytes from HFD-fed FWT and F∆N1 , 2 mice ( 150 bp paired-end format , approximately 320 million mean reads/sample , n = 4 ) was performed ( Table 1 ) . Compound NOVA1/2 deficiency caused only a small change in gene expression ( 55 genes; q<0 . 05; absolute log2-fold change >0 . 75 ) , but NOVA1/2 deficiency caused a large change in differential exon inclusion/exclusion ( 1169 exons; FDR<0 . 05; absolute ∆Inc level >0 . 1 ) in the adipocyte transcriptome ( Figure 2A and Figure 1—figure supplement 1 ) . These data indicate that NOVA deficiency primarily changes pre-mRNA splicing . The most common form of alternative pre-mRNA splicing caused by NOVA deficiency was exon skipping ( 768 exons ) , but we also detected 128 mutually exclusive exon inclusions , 99 intron retentions , 64 alternative 5’ splice sites , and 110 alternative 3’ splice sites ( Figure 2D , E ) . Analysis of white adipocytes with single gene ablations of Nova1 or Nova2 identified fewer changes in alternative pre-mRNA splicing , including 10 & 7 skipped exons , 1 & 4 mutually exclusive exon inclusions , 12 & 16 intron retentions , 1 & 0 alternative 5’ splice sites , and 2 & 4 alternative 3’ splice sites , respectively ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) . These data indicate that NOVA1 and NOVA2 can cause isoform-specific changes in adipocyte pre-mRNA alternative splicing . However , the effect of compound NOVA1 plus NOVA2 deficiency to cause widespread changes in alternative pre-mRNA splicing ( Figure 2 ) indicates that these NOVA proteins exhibit some functional redundancy in adipocytes . 10 . 7554/eLife . 17672 . 012Figure 2 . NOVA proteins contribute to alternative pre-mRNA splicing associated with diet-induced obesity . ( A ) RNA-seq analysis of FWT and F∆N1 , 2 mice fed an HFD ( 16 wk ) identifies significant changes in gene expression ( q<0 . 05; absolute log2-fold change >0 . 75 ) and differential exon inclusion/exclusion ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) in the epididymal adipocyte transcriptome . ( B ) The number of genes with significant differential alternative splicing ( CD-fed vs HFD-fed WT mice and HFD-fed FWT mice vs HFD-fed F∆N1 , 2 mice ) are depicted using a Venn diagram . ( C ) Biological pathway enrichment analysis of the 323 genes co-regulated by alternative pre-mRNA splicing following HFD consumption and NOVA deficiency . ( D , E ) Classification of alternative splicing events caused by NOVA deficiency ( D ) and selected examples ( E ) are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01210 . 7554/eLife . 17672 . 013Figure 2—figure supplement 1 . Establishment of Nova1LoxP/LoxP and Nova2LoxP/LoxP mice . ( A , B ) The methods used to create the floxed Nova1 and floxed Nova2 alleles by homologous recombination are illustrated schematically . ( C , D ) PCR analysis of genomic DNA isolated epididymal white adipose tissue ( eWAT ) and interscapular brown adipose tissue ( BAT ) of Adipoq-cre+ ( FWT ) mice , Adipoq-cre+ Nova1LoxP/LoxP ( F∆N1 ) mice , Adipoq-cre+ Nova2LoxP/LoxP ( F∆N2 ) mice , and Adipoq-cre+ Nova1LoxP/LoxP Nova2LoxP/LoxP ( F∆N1 , 2 ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 013 Comparison of white adipocyte genes with differential alternative splicing ( FDR<0 . 05; absolute ∆Inc level >0 . 1 ) caused by the consumption of a HFD ( Figure 1A ) or NOVA1/2 deficiency ( Figure 2B ) demonstrated 323 co-regulated genes ( Figure 2B ) . These co-regulated genes represent 26% of the 1249 HFD-regulated genes and 34% of the 950 NOVA-regulated genes . Analysis of the co-regulated genes demonstrated enrichment for pathways including mRNA processing and multiple signaling pathways ( e . g . NF-κB signaling , MAPK signaling ) that contribute to the physiological regulation of adipose tissue ( Figure 2C ) . To confirm the observation that NOVA proteins contribute to signaling mechanisms that can mediate metabolic regulation , we examined the expression of the JNK group of MAPK in cells that express NOVA proteins ( white adipocytes ) and cells that do not express NOVA proteins ( hepatocytes ) ( Figure 3A ) . The genes Mapk8 ( encodes the JNK1 protein kinase ) and Mapk9 ( encodes the JNK2 protein kinase ) express pre-mRNA are alternatively spliced by the mutually exclusive inclusion of either exon 7a or 7b to yield the α and β isoforms of the JNK1 and JNK2 protein kinases ( Gupta et al . , 1996 ) . The sequences surrounding exons 7a and 7b contain consensus sites for NOVA binding ( YCAY ) that are established to be NOVA binding sites by CLIP-seq analysis ( Licatalosi et al . , 2008 ) . We designed and validated a Taqman assay to quantitate the inclusion of exon 7a or 7b sequences in Mapk8 mRNA ( Mapk8α and Mapk8β ) and Mapk9 mRNA ( Mapk9α and Mapk9β ) ( Figure 3—figure supplement 1A , B ) . This analysis demonstrated that adipocytes and hepatocytes express different alternatively spliced JNK isoforms ( Figure 3—figure supplement 1C ) . These tissue-specific differences in Mapk8/9 pre-mRNA splicing may be influenced by the selective expression of NOVA proteins in adipocytes , but not hepatocytes ( Figures 1F and 3A ) , although NOVA-independent mechanisms likely also contribute to the observed cell type-specific pattern of JNK isoform expression . 10 . 7554/eLife . 17672 . 014Figure 3 . NOVA promotes signal transduction by JNK in adipose tissue . ( A ) The expression of Nova mRNA in adipocytes and hepatocytes was measured by quantitative RT-PCR ( mean ± SEM; n=7~8; p<0 . 01; p<0 . 001 ) . The source data are included as Figure 3—source data 1 . ( B ) The α / β expression ratios of Mapk8 and Mapk9 mRNA by hepatocytes was measured by quantitative RT-PCR ( mean ± SEM; n=10; p<0 . 05; *p<0 . 01 ) . The effect of hepatic expression of GFP ( Control ) or NOVA2 using adenoviral vectors was examined . The source data are included as Figure 3—source data 2 . ( C ) The expression ratio of the α and β isoforms of Mapk8 and Mapk9 mRNA by FWT and F∆N1 , 2 adipocytes was measured by quantitative RT-PCR ( mean ± SEM; n=5~8; p<0 . 05; *p<0 . 01 ) . The source data are included as Figure 3—source data 3 . ( D ) The mutually exclusive inclusion of exons 7a or 7b in Mapk8 and Mapk9 mRNA is illustrated . NOVA CLIP-seq tags are highlighted in red . ( E ) Mapk8∆/∆ Mapk9-/- MEF transduced with retroviruses expressing JNK1α , JNK1β , JNK2α , JNK2β or empty vector ( - ) were exposed without and with 60 J/m2 UV ( 60 min ) and lysates were examined by immunoblot analysis . ( F ) Adipocytes prepared from FWT and F∆N1 , 2 mice were treated without and with 1 µg/ml anisomycin ( 10 min ) and lysates were examined by immunoblot analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01410 . 7554/eLife . 17672 . 015Figure 3—source data 1 . Source data for Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01510 . 7554/eLife . 17672 . 016Figure 3—source data 2 . Source data for Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01610 . 7554/eLife . 17672 . 017Figure 3—source data 3 . Source data for Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01710 . 7554/eLife . 17672 . 018Figure 3—figure supplement 1 . Design and validation of Taqman assays to detect inclusion of the mutually exclusive exons 7a and 7b in Mapk8 and Mapk9 mRNA . ( A ) Schematic illustration of the design of Taqman assays to detect expression of Mapk8α , Mapk8β , Mapk9α , and Mapk9β mRNA . ( B ) Taqman assays were performed using 1 ng of cDNA template ( Mapk8α , Mapk8β , Mapk9α , and Mapk9β cDNA ) and probes designed to detect Mapk8α , Mapk8β , Mapk9α , and Mapk9β . The data presented are the mean ± SEM ( n = 3 ) . ( C ) The relative expression of Mapk8α , Mapk8β , Mapk9α , and Mapk9β mRNA in epididymal adipocytes and hepatocytes was measured by RT-PCR analysis ( mean ± SEM; n = 3 ) . The numbers above the bars represents the α/β ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01810 . 7554/eLife . 17672 . 019Figure 3—figure supplement 2 . Comparison of JNKα and JNKβ protein kinase activity in vitro . ( A ) Flag-tagged JNK1α1 and JNK1β1 expressed in 293T cells were isolated by immunoprecipitation using anti-Flag ( M2 ) agarose . Protein kinase activity was measured with an in vitro assay using GST-cJun and [γ-32P] ATP as substrates . The amount of JNK in each assay was examined by immunoblot analysis . The amount of phosphorylated cJun was examined by Phosphorimager analysis . ( B ) In vitro kinase assays were performed using Flag-tagged JNK2α2 and JNK2β2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 01910 . 7554/eLife . 17672 . 020Figure 3—figure supplement 3 . Effect of adipocyte-specific JNK-deficiency on thermogenic gene expression . ( A ) Core body temperature of FWT and F∆J1 , 2 mice fed a CD or n HFD ( 16 wk ) was measured by telemetry using an implanted probe ( mean ± SEM; n = 7~10 ) . ( B ) FWT and F∆J1 , 2 mice were fed a CD or an HFD ( 16 wk ) . Gene expression by sub-cutaneous inguinal adipocytes was examined by measurement of mRNA by RT-PCR ( mean ± SEM; n=5~8 ) . Statistically significant differences between FWT and F∆J1 , 2 mice are indicated: p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 020 To test the role of NOVA proteins on JNK isoform expression , we examined the effect of increased and decreased NOVA expression . We found that hepatic expression of NOVA2 caused a decrease in the Mapk8α/β ratio and an increase in the Mapk9α/β ratio , indicating that NOVA can promote the expression of the Mapk8β and Mapk9α alternatively spliced isoforms ( Figure 3B ) . In contrast , adipocyte-specific deficiency of NOVA1 plus NOVA2 caused an increase in the Mapk8α/β ratio and a decrease in the Mapk9α/β ratio , indicating that NOVA-deficiency promotes the expression of the Mapk8α and Mapk9β alternatively spliced isoforms ( Figure 3C ) . These observations are consistent with the finding that HFD consumption causes both decreased NOVA expression ( Figure 1H ) and significant changes in the mutually exclusive inclusion of Mapk8 exons 7a and 7b ( FDR=0 . 0066 ) . Together , these data demonstrate that NOVA proteins can promote the expression of the Mapk8β and Mapk9α isoforms . The sequence differences between the α and β isoforms of JNK1 and JNK2 are located in the substrate binding site ( Figure 3D ) and influence the interaction of JNK with protein substrates ( Davis , 2000 ) . Studies using the substrate cJun demonstrate low activity ( JNK1α and JNK2β ) and high activity ( JNK1β and JNK2α ) groups of JNK protein kinases in vitro ( Figure 3—figure supplement 2 ) and in vivo ( Figure 3E ) . Since NOVA can promote the expression of Mapk8β and Mapk9α mRNA that encode the high activity forms of JNK1 and JNK2 , NOVA-deficiency can be predicted to cause expression of the low activity forms JNK1α and JNK2β in adipocytes . To test this conclusion , we prepared primary adipocytes from FWT and F∆N1 , 2 mice and examined stress-induced JNK activation . No differences in JNK expression or stress-induced activating phosphorylation ( pThr180-Pro-pTyr182 ) of JNK between FWT and F∆N1 , 2 adipocytes were detected by immunoblot analysis ( Figure 3F ) . However , stress-induced phosphorylation of the JNK substrate pSer63 cJun was markedly suppressed in F∆N1 , 2 adipocytes compared with FWT adipocytes ( Figure 3F ) . Together , these data confirm that NOVA proteins can regulate signaling mechanisms in adipocytes . To examine the consequences of reduced JNK signaling in adipocytes , we established F∆J1 , 2 mice with adipocyte-specific JNK-deficiency ( Adipoq-Cre−/+ Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) . These mice exhibited increased core body temperature and increased expression of genes that mediate sympathetic activation ( Adr3b ) and thermogenesis ( Ucp1 ) in adipocytes ( Figure 3—figure supplement 3 ) . We therefore considered the possibility that NOVA-deficiency in adipocytes might also cause increased adipose tissue thermogenesis . To test the physiological role of NOVA proteins in adipocytes , we examined the effect of feeding a CD or a HFD to control FWT and F∆N1 , 2 mice . We found that CD-fed FWT and F∆N1 , 2 mice gained similar body mass , but HFD-fed F∆N1 , 2 mice gained significantly less mass than HFD-fed FWT mice ( Figure 4A and Figure 4—figure supplement 1A ) . Proton magnetic resonance imaging demonstrated that the decreased body mass was caused by reduced fat mass ( Figure 4B ) . Analysis of organs at necropsy demonstrated a reduction in mass of the liver and adipose tissue in HFD-fed F∆N1 , 2 mice compared with HFD-fed FWT mice ( Figure 4—figure supplement 1B ) and analysis of tissue sections demonstrated reduced adipocyte hypertrophy , reduced hepatic steatosis , and reduced hypertrophy of pancreatic islets in the NOVA-deficient mice ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 17672 . 021Figure 4 . NOVA promotes the development of diet-induced obesity . ( A ) The change in body mass of FWT and F∆N1 , 2 mice fed a CD or a HFD is presented ( mean ± SEM; n=8~30; p<0 . 05 ) . ( B ) Body composition was examined by proton magnetic resonance spectroscopy ( mean ± SEM; n=8~22; p<0 . 05; p<0 . 01 ) . The source data are included as Figure 4—source data 1 . ( C ) Blood insulin , leptin , resistin , and glucose in overnight starved CD-fed and HFD-fed ( 12 wks ) FWT and F∆N1 , 2 mice were measured ( mean ± SEM; n=8~16; p<0 . 05; **p<0 . 001 ) . The source data are included as Figure 4—source data 2 . ( D ) CD-fed and HFD-fed ( 12 wks ) FWT and F∆N1 , 2 mice were examined by glucose tolerance tests ( mean ± SEM; n=8~26; p<0 . 05; **p<0 . 001 ) . The source data are included as Figure 4—source data 3 . ( E ) Energy expenditure ( EE ) by HFD-fed ( 4 wks ) FWT mice ( n = 9 ) and F∆N1 , 2 mice ( n = 8 ) was examined using metabolic cages over 3 days ( 12 hr light; 12 hr dark ) . The source data are included as Figure 4—source data 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02110 . 7554/eLife . 17672 . 022Figure 4—source data 1 . Source data for Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02210 . 7554/eLife . 17672 . 023Figure 4—source data 2 . Source data for Figure 4C . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02310 . 7554/eLife . 17672 . 024Figure 4—source data 3 . Source data for Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02410 . 7554/eLife . 17672 . 025Figure 4—source data 4 . Source data for Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02510 . 7554/eLife . 17672 . 026Figure 4—figure supplement 1 . Effect of adipose tissue-specific NOVA1/2-deficiency on organ mass . ( A ) The change in% body mass of FWT and F∆N1 , 2 mice fed a CD or ais presented ( mean ± SEM; n=8~30; p<0 . 05 ) . ( B ) The mass of the liver , epididymal white adipose tissue ( eWAT ) , interscapular brown adipose tissue ( BAT ) , and gastrocnemius muscle ( Gastroc . ) of FWT and F∆N1 , 2 mice fed a CD or a ( 16 wk ) is presented ( mean ± SEM; n=21 ) . Statistically significant differences between FWT and F∆N1 , 2 mice are indicated: p<0 . 05; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02610 . 7554/eLife . 17672 . 027Figure 4—figure supplement 2 . Comparison of adipose tissue , liver , and pancreas in FWT and F∆N1 , 2 mice . ( A ) Analysis of hematoxylin & eosin-stained tissue sections prepared from epididymal adipose tissue of CD-fed and HFD-fed ( 16 wk ) FWT and F∆N1 , 2 mice . The relative area of adipocytes is presented ( mean ± SEM; n=6 mice ) . ( B ) Representative hematoxylin & eosin-stained liver sections are presented ( n=6 ) . ( C ) Sections of the pancreas were stained with an antibody to insulin ( green ) . DNA was stained with DAPI . The relative area of islets is presented ( mean ± SEM; n=6 mice ) . Statistically significant differences between FWT and F∆N1 , 2 mice are indicated: p<0 . 01; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02710 . 7554/eLife . 17672 . 028Figure 4—figure supplement 3 . Effect of adipose tissue-specific ablation of the Nova1 or Nova2 genes . ( A ) The change in body mass of FWT , F∆N1 and F∆N2 mice fed a CD or ais presented ( mean ± SEM; n=8~21 ) . ( B ) Body composition was examined by proton magnetic resonance spectroscopy ( mean ± SEM; n=8~21 ) . ( C ) Blood glucose concentration in overnight starved CD-fed and HFD-fed ( 16 wks ) mice ( mean ± SEM; n=8~21 ) . ( D ) The mass of the liver , epididymal white adipose tissue ( eWAT ) , interscapular brown adipose tissue ( BAT ) , and gastrocnemius muscle ( Musc . ) of FWT and F∆N1 , 2 mice fed a CD or a ( 16 wk ) is presented ( mean ± SEM; n=8~21 ) . ( E ) CD-fed and HFD-fed ( 16 wks ) mice were examined by glucose tolerance tests ( mean ± SEM; n=8~21 ) . ( F ) Sections of the liver , eWAT , and BAT were stained with hematoxylin and eosin ( n=6 ) . Scale bar , 100 µm . ( G ) Blood insulin and leptin in overnight starved mice were measured ( mean ± SEM; n=8~21 ) . Statistically significant differences between Control and NOVA-deficient mice are indicated: p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02810 . 7554/eLife . 17672 . 029Figure 4—figure supplement 4 . Metabolic cage analysis of physical activity and the consumption of food and water . ( A ) Food consumption ( B ) water consumption , and ( C ) physical activity were measured using metabolic cages ( 3 days; 12 hr light and 12 hr dark ) using FWT and F∆N1 , 2 mice fed a CD or a ( 4 wk ) . The data presented are the mean ± SEM ( n=9 mice ) . No statistically significant differences between FWT and F∆N1 , 2 mice were detected ( p>0 . 05 ) . The source data are included as Figure 4—figure supplement 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 02910 . 7554/eLife . 17672 . 030Figure 4—figure supplement 4—source data 1 . Source data for Figure 4—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03010 . 7554/eLife . 17672 . 031Figure 4—figure supplement 5 . Metabolic cage analysis of gas exchange and energy expenditure . ( A ) Energy expenditure ( EE ) , ( B ) VO2 , ( C ) VCO2 , and ( D ) respiratory exchange ratio ( RER ) were measured using metabolic cages ( 3 days; 12 hr light and 12 hr dark ) using FWT and F∆N1 , 2 mice fed a CD or a ( 4 wk ) . The data presented are the mean ± SEM ( n=7~9 mice ) . Statistically significant differences between FWT and F∆N1 , 2 mice are indicated: p<0 . 05; p<0 . 01; ** , 0 . 001 . The source data are included as Figure 4—figure supplement 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03110 . 7554/eLife . 17672 . 032Figure 5—figure supplement 5—source data 1 . Source data for Figure 4—figure supplement 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 032 The reduced obesity of HFD-fed F∆N1 , 2 mice compared with FWT mice suggested that NOVA-deficiency may suppress HFD-induced metabolic syndrome . Indeed , hyperinsulinemia and hyperleptinemia were reduced in F∆N1 , 2 mice compared with FWT mice ( Figure 4C ) . Moreover , HFD-fed FWT mice were found to be more glucose intolerant than F∆N1 , 2 mice ( Figure 4D ) and the HFD-induced hyperglycemia observed in FWT mice was suppressed in F∆N1 , 2 mice ( Figure 4C ) . Similar ( although smaller ) phenotypes were detected in mice with adipocyte-specific single gene ablation of Nova1 or Nova2 ( Figure 4—figure supplement 3 ) . Metabolic cage analysis demonstrated that food/water consumption and physical activity of FWT and F∆N1 , 2 mice were similar ( Figure 4—figure supplement 4 ) , but F∆N1 , 2 mice exhibited greatly increased energy expenditure compared with FWT mice ( Figure 4E and Figure 4—figure supplement 5 ) . This increase in energy expenditure may account for the suppression of HFD-induced obesity caused by NOVA deficiency in adipocytes . These data demonstrate that NOVA proteins in adipocytes suppress energy expenditure and promote obesity-associated metabolic syndrome . We measured core body temperature by telemetry using an implanted probe in mice housed at the ambient temperature of the vivarium ( 21°C ) and following a cold challenge ( 4°C ) . This analysis demonstrated that the core body temperature of F∆N1 , 2 mice was significantly higher than FWT mice during the course of this study ( Figure 5A ) . This was associated with increased expression of genes in sub-cutaneous adipocytes of F∆N1 , 2 mice that are associated with a 'browning' ( beige/brite ) phenotype ( Cidea , Dio2 , Ppargc1a , Ppargc1b , and Ucp1 ) compared with FWT mice ( Figure 5B ) . Together , these data indicate that NOVA proteins in adipocytes can suppress adipose tissue thermogenesis . 10 . 7554/eLife . 17672 . 033Figure 5 . NOVA regulates a thermogenic program in adipose tissue . ( A ) FWT and F∆N1 , 2 mice were subject to cold challenge ( 4°C ) . Core body temperature was measured by telemetry using an implanted probe ( mean ± SEM; n=8; p<0 . 05; p<0 . 01 ) . The source data are included as Figure 5—source data 1 . ( B ) The effect of cold challenge ( 4°C , 6 hr ) on gene expression by inguinal adipocytes ( iWAT ) of FWT and F∆N1 , 2 mice was examined by quantitative RT-PCR ( mean ± SEM; n=6~8; p<0 . 05; *p<0 . 01 ) . The source data are included as Figure 5—source data 2 . ( C ) FWT and F∆N1 , 2 mice were housed under thermal neutral conditions ( 30°C ) . The change in body mass of CD and HFD-fed mice is presented ( mean ± SEM; n=7~18 ) . ( D ) Glucose tolerance tests were performed on FWT and F∆N1 , 2 mice housed under thermoneutral conditions ( 30°C ) . The effect of feeding a CD or a HFD ( 16 wk ) is presented ( mean ± SEM; n=8 ) . The source data are included as Figure 5—source data 3 . ( E ) Sections of brown adipose tissue ( BAT ) and iWAT of HFD-fed ( 16 wk ) mice housed at 21°C and 30°C were stained with hematoxylin & eosin . The data shown are representative of 6 mice per group . ( F ) FWT and F∆N1 , 2 mice housed at 21°C and 30°C were fed a HFD ( 16 wk ) . Gene expression by adipocytes of iWAT and retroperitoneal adipose tissue ( rWAT ) was examined by quantitative RT-PCR ( mean , n=8 ) . The data are presented as a heat map . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03310 . 7554/eLife . 17672 . 034Figure 5—source data 1 . Source data for Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03410 . 7554/eLife . 17672 . 035Figure 5—source data 2 . Source data for Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03510 . 7554/eLife . 17672 . 036Figure 5—source data 3 . Source data for Figure 5D . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03610 . 7554/eLife . 17672 . 037Figure 5—figure supplement 1 . Thermogenic gene expression by primary adipocytes . ( A ) Primary pre-adipocytes were treated with 10 µM forskolin ( 4 hr ) . Gene expression was examined by measurement of mRNA by RT-PCR ( mean ± SEM; n=6 ) . Statistically significant differences between Control and Forskolin-treated adipocytes are indicated: p<0 . 05; p<0 . 01; * , 0 . 001 . ( B ) Primary adipocytes were cultured from FWT and F∆N1 , 2 mice and treated with forskolin ( 4 hr ) . Gene expression was examined by measurement of mRNA by RT-PCR ( mean ± SEM; n=6 ) . Statistically significant differences between FWT and F∆N1 , 2 adipocytes are indicated: p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 03710 . 7554/eLife . 17672 . 038Figure 5—figure supplement 2 . Thermogenic gene expression is increased by shRNA-mediated suppression of Nova1 or Mapk8β/Mapk9α . ( A ) An shNova1 lentiviral vector was used to suppress Nova1 gene expression in 3T3 L1 adipocytes . Nova2 expression was not detected in these cells . Gene expression was examined by measurement of mRNA by RT-PCR ( mean ± SEM; n=6 ) . ( B ) Lentiviral vectors expressing shMapk8β and shMapk9α were used to suppress Mapk8β and Mapk9α mRNA expression in 3T3 L1 adipocytes . Gene expression was examined by measurement of mRNA by RT-PCR ( mean ± SEM; n=6 ) . Statistically significant differences between Control and shRNA knock-down adipocytes are indicated: p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 17672 . 038 To test whether NOVA proteins act by a cell autonomous mechanism to regulate the beige/brite phenotype , we established primary adipocytes in culture . Sympathetic stimulation of adipose tissue increases cAMP and promotes the browning of white adipocytes ( Rosen and Spiegelman , 2014 ) . This process can be studied in vitro by treating adipocytes with drugs that raise intracellular cAMP . Indeed , treatment of primary adipocytes with the drug forskolin increases cAMP concentration and promotes the beige/brite phenotype ( Rosen and Spiegelman , 2014 ) , and causes increased expression of the Adrb3 , Dio2 , Fgf21 , Ppargc1a , and Ucp1 genes ( Figure 5—figure supplement 1A ) . This treatment also causes decreased Nova gene expression and promotes expression of the alternatively spliced isoform Mapk9β ( Figure 5—figure supplement 1A ) . To examine whether NOVA proteins contribute to the beige/brite phenotype , we compared the effects of increased cAMP on FWT and F∆N1 , 2 primary adipocytes . This analysis demonstrated that NOVA-deficiency caused increased expression of genes in adipocytes that mediate sympathetic activation ( Adrb3 ) and thermogenesis ( Ucp1 ) ( Figure 5—figure supplement 1B ) . To further test the cell autonomous role of NOVA proteins on the beige/brite phenotype , we examined 3T3L1 adipocytes that are normally resistant to the effects of cAMP to promote browning . These cells express NOVA1 , but NOVA2 was not detected . Nova1 knock-down using shRNA was associated with increased expression of the beige/brite phenotype genes Prdm16 and Ucp1 ( Figure 5—figure supplement 2A ) . Similarly , knock-down of Mapk8β mRNA plus Mapk9α mRNA using shRNA caused increased expression of the Prdm16 and Ucp1 genes ( Figure 5—figure supplement 2B ) . Together , these data confirm that NOVA-mediated alternative splicing in adipocytes contributes to adipose tissue thermogenesis . To test the role of adipose tissue thermogenesis in the metabolic phenotype of F∆N1 , 2 mice ( Figure 4 ) , we examined the effect of housing mice under thermoneutral conditions ( 30°C ) . We found that F∆N1 , 2 mice and FWT mice gained equal body mass at thermoneutrality ( Figure 5C ) . Moreover , the increased glucose tolerance of F∆N1 , 2 mice compared with FWT mice at 21°C was not observed at 30°C ( Figure 5D ) . These data indicate that the improved glycemic regulation exhibited by F∆N1 , 2 mice compared with FWT mice at 21°C was caused by increased thermogenesis . Consistent with this conclusion , we found that the reduced adipocyte hypertrophy observed in HFD-fed F∆N1 , 2 mice compared with FWT mice at 21°C was not observed at 30°C ( Figure 5E ) . Similarly , the increased expression of thermogenesis-related genes ( Adrb3 , Cidea , Dio2 , Ppargc1a/b Prdm16 , and Ucp1 ) by adipocytes in HFD-fed F∆N1 , 2 mice compared with FWT mice at 21°C was not found at 30°C ( Figure 5F ) . Together , these data demonstrate that adipose tissue thermogenesis contributes to the improved glycemia of mice with NOVA-deficiency in adipocytes . The browning of white adipose tissue is associated with the appearance of beige/brite adipocytes , increased energy expenditure , and improved obesity-induced metabolic syndrome ( Rosen and Spiegelman , 2014 ) . An understanding of molecular mechanisms that account for browning is therefore important for the development of potential therapies for the treatment of metabolic syndrome based on increasing adipose tissue energy expenditure . It is established that beige/brite adipocytes in white adipose tissue depots can arise from specialized progenitor cells ( Wang et al . , 2013b ) . Beige/brite adipocytes may also arise from inactive cells present within white adipose tissue ( Rosenwald et al . , 2013 ) . These two mechanisms may contribute to functional beige/brite cell development ( Rosen and Spiegelman , 2014 ) . The increased adipose tissue browning detected in mice with adipocyte-specific NOVA deficiency was observed using floxed Nova alleles and a Cre driver ( Adipoq-Cre ) that is expressed in mature adipocytes . It is therefore likely that the increased white adipose tissue browning caused by NOVA deficiency does not reflect a change in adipocyte differentiation from specialized progenitor cells , but rather a change in mature adipocyte function . Whether these mature adipocytes represent white adipocytes or inactive beige/brite adipocytes is unclear . However , the observation that housing mice at thermoneutrality prevents the effect of NOVA deficiency to cause 'browning' suggests that NOVA deficiency may act by promoting the activation of inactive beige/brite adipocytes . Our analysis demonstrates that NOVA expression in white adipocytes is reduced in obese humans and mice ( Figure 1 ) . These changes in NOVA expression may cause changes in adipose tissue physiology . Gene knockout studies demonstrate that compound NOVA-deficiency in adipocytes caused increased adipose tissue thermogenesis and improved whole body glycemia in HFD-fed mice ( Figure 4 ) . However , partial NOVA-deficiency in adipocytes ( potentially modeling the changes in NOVA expression caused by feeding a HFD ) caused modest changes in adipocyte physiology ( Figure 4—figure supplement 3 ) . The significance of the decrease in adipocyte NOVA expression in HFD-fed mice is therefore unclear . Nevertheless , our data establish that NOVA proteins in adipocytes function to suppress adipose tissue thermogenesis . The mechanism appears to be mediated by NOVA-regulated changes in pre-mRNA splicing that promote adipocyte thermogenic gene expression ( Figure 5 ) , but we cannot exclude additional contributions caused by other biochemical actions of NOVA proteins , including alternative mRNA polyadenylation ( Licatalosi et al . , 2008 ) and functional regulation of Argonaute/microRNA complexes ( Storchel et al . , 2015 ) . Further studies will be required to examine the relative contributions of NOVA-regulated pre-mRNA splicing , alternative polyadenylation , and microRNA function . However , our analysis demonstrates that NOVA proteins do function as alternative pre-mRNA splicing factors in adipocytes ( Figure 2 ) . The changes in pre-mRNA splicing caused by NOVA proteins include both increased and decreased exon inclusion , consistent with the known context-dependent role of NOVA proteins to promote both exon inclusion and exclusion ( Ule et al . , 2006 ) . Importantly , there are significant gaps in our knowledge concerning mechanisms of alternative pre-mRNA splicing in adipocytes . First , NOVA proteins may act in a combinatorial manner with other splicing factors ( Zhang et al . , 2010 ) , but roles for such factors in adipocytes have not been established . Second , our analysis of mice with NOVA-deficiency in adipocytes implicates a role for NOVA proteins in the alternative splicing of 26% of the genes that are regulated by HFD consumption ( Figure 2B ) . Consequently , NOVA proteins do not contribute to 74% of the genes that exhibit HFD-regulated pre-mRNA splicing and mechanisms that contribute to the regulation of these genes have not been established . Further studies will be required to achieve an understanding of these processes . The regulation of adipose tissue thermogenesis by NOVA proteins may not be entirely dependent upon the classical UCP1 pathway ( Kozak , 2010 ) because of the existence of alternative thermogenic mechanisms ( Kazak et al . , 2015 ) . Nevertheless , our data demonstrate that NOVA proteins can suppress adipocyte thermogenesis . Reduced NOVA expression in adipocytes causes thermogenesis ( Figure 5 ) and may potentially be achieved by drugs targeting NOVA-mediated pre-mRNA splicing . This role of NOVA proteins is most likely mediated by a network response to a pre-mRNA splicing program that collectively regulates adipocyte thermogenesis . In conclusion , we describe a NOVA-dependent alternative pre-mRNA splicing program in white adipocytes that regulates browning of white adipose tissue . These data identify alternative pre-mRNA splicing as a biological process that may be targeted by drugs designed to increase adipocyte thermogenesis and improve metabolic syndrome caused by obesity . C57BL/6J mice ( RRID:IMSR_JAX:000664 ) , B6 . Cg-Lepob/J ( RRID:IMSR_JAX:000632 ) , B6;FVB-Tg ( Adipoq-Cre ) 1Evdr ( RRID:IMSR_JAX:010803 ) ( Eguchi et al . , 2011 ) , B6;129-Gt ( ROSA ) 26Sortm1 ( cre/ERT ) Nat/J mice ( RRID:IMSR_JAX:004847 ) ( Badea et al . , 2003 ) , and B6 . 129S4-Gt ( ROSA ) 26Sortm1 ( FLP1 ) Dym/RainJ mice ( RRID:IMSR_JAX:009086 ) ( Farley et al . , 2000 ) were obtained from The Jackson Laboratory . We have previously reported Mapk8LoxP/LoxP ( Das et al . , 2007 ) , Mapk9LoxP/LoxP mice ( Han et al . , 2013b ) , and Mapk9-/- mice ( RRID:IMSR_JAX:004321 ) ( Yang et al . , 1998 ) . We generated Nova1 and Nova2 conditional mice using ES cells targeted by homologous recombination ( Nova1tm1a ( EUCOMM ) Hmgu and Nova2tm1a ( KOMP ) Wtsi ) , the preparation of chimeric mice , and breeding to obtain germ-line transmission of the disrupted Nova1 and Nova2 genes using standard methods . The Frt-NeoR-Frt cassette was excised by crossing with FLPeR mice to obtain mice with the Nova1LoxP and Nova2LoxP alleles . Mice were genotyped by PCR analysis of genomic DNA using the primers 10F ( 5’-GTCCGTAAGGCATGTC-3’ ) and 2R ( 5’-AGCAAAAAGCCATCCATG-3’ ) to detect the Nova1+ ( 894 bp ) , Nova1LoxP ( 1 , 101 bp ) , and Nova1∆ ( 281 bp ) alleles , the primers 1F ( 5’-CAGAAGAACTGGAGAC-3’ ) and N2-2R ( 5’- GGTTGGGCTGTCAGTG-3’ ) to detect the Nova2+ ( 149 bp ) and Nova2LoxP ( 127 bp ) alleles , or with the primers N2delF1 ( 5’-CAGGCTGGCGCCGGAAC-3’ ) and N2-2R ( 5’-GGTTGGGCTGTCAGTG-3’ ) to detect the Nova2LoxP ( 970 bp ) and Nova2∆ ( 153 bp ) alleles . All mice were studied on the C57BL/6J strain background . Male mice ( 8 wks old ) were fed a chow diet ( Iso Pro 3000 , Purina ) or a HFD ( S3282 , Bioserve ) . Body weight was measured with a scale . Whole body fat and lean mass were non-invasively measured using 1H-MRS ( Echo Medical Systems ) . The mice were housed at 21°C ( alternatively at 4°C or 30°C , where indicated ) in a specific pathogen-free facility accredited by the American Association for Laboratory Animal Care ( AALAC ) . The Institutional Animal Care and Use Committee ( IACUC ) of the University of Massachusetts Medical School approved all studies using animals . The analysis was performed by the Mouse Metabolic Phenotyping Center at the University of Massachusetts Medical School . The mice were housed under controlled temperature and lighting with free access to food and water . The food/water intake , energy expenditure , respiratory exchange ratio , and physical activity were measured using metabolic cages ( TSE Systems ) . Biocompatible and sterile microchip transponders ( IPTT-300 Extended Accuracy Calibration; Bio Medic Data Systems ) were implanted subcutaneously . Cold tolerance tests ( 4°C ) were performed using mice fed a chow diet ad-libitum . Blood glucose was measured with an Ascensia Breeze 2 glucometer ( Bayer ) . Adipokines and insulin in plasma were measured by multiplexed ELISA using a Luminex 200 machine ( Millipore ) . Glucose and insulin tolerance tests were performed by intraperitoneal injection of mice with glucose ( 1 g/kg ) or insulin ( 1 . 5 U/kg ) using methods described previously ( Sabio et al . , 2008 ) . Mice ( 8 wks ) were fed a HFD . At 12 wks , the mice treated with 5 × 109 pfu/mouse Adenovirus-NOVA2 or Adenovirus-GFP ( Applied Biological Materials Inc . ) by tail vein injection . The mice were euthanized at 2 wks post-injection . Human Flag-tagged Mapk8α1 , Mapk8β1 , Mapk9α2 , and Mapk9β2 cDNA cloned in the expression vector pCDNA3 have been described previously ( Gupta et al . , 1996 ) . Murine Mapk8α1 , Mapk8β1 , Mapk9α2 , and Mapk9β2 cDNA were isolated by RT-PCR and cloned by blunt-end ligation in the SnaB1 site of the retroviral expression vector pBABE-puro ( Addgene plasmid #1764 ) ( Morgenstern and Land , 1990 ) . Retroviral stocks were prepared by transfection of Phoenix-ECO cells ( American Type-Culture Collection , ATCC CRL-3214 ) ( Lamb et al . , 2003 ) . The DNA sequences used to generate shRNA vectors were Nova1 ( 5’-CCGGGCTGCTCAGTATTTAATTACTCGAGTAATTAAATACTGAGCAGCTTTTTG-3’ ) , Mapk8β ( 5’-CCGGTCATGGGAGAAATGATCAAAGCTCGAGCTTTGATCATTTCTCCCATGATTTTTG-3’ ) , Mapk9α ( 5’-CCGGGTGAAAGGTTGTGTGATATTCCTCGAGGAATATCACACAACCTTTCACTTTTTG-3’ ) . These sequences were cloned in the Age1/EcoR1 sites of the lentiviral vector pLKO . 1-puro ( Addgene #8453; [Stewart et al . , 2003] ) . Lentiviral stocks were prepared by transfection of 293T cells with the indicated replication-incompetent lentiviral vector ( pLKO1-shRNA ) together with the packaging plasmid psPAX2 and the envelope plasmid pMD2 . G ( Addgene #12259 and #12260; [Naldini et al . , 1996] ) . Primary inguinal adipocytes were prepared from male mice ( 8 wk old ) . The fat pads were minced with a razor blade and incubated ( 40 min , 37°C ) in 12 . 5 mM Hepes pH 7 . 4 , 120 mM NaCl , 6 mM KCl , 1 . 2 mM MgSO4 , 1 mM CaCl2 , 2%BSA , 2 . 5 mM glucose , 1 mg/ml collagenase II ( Sigma ) and 0 . 2 mg/ml DNAse I ( Sigma ) . The digested tissue was filtered through a 100 µm nylon strainer and centrifuged ( 8 min at 250 g ) . The pellet was suspended in Dulbecco’s modified Eagle’s medium ( DMEM ) :Ham’s F12 ( 1:1 ) medium with 8%FBS , 1x Antibiotic-Antimycotic , 2 mM glutamine ( Life Technologies ) . 100 , 000 cells/well were seeded in 12 well plates . The medium was refreshed every 2 days . On day 6 , differentiation was induced using medium further supplemented with 0 . 5 mM IBMX ( Sigma ) , 5 µg/ml insulin ( Sigma ) , 1 µM Troglitazone ( TZD; Calbiochem ) , and 2 . 5 µM dexamethasone ( Sigma ) . On day 8 , the medium was refreshed using medium supplemented with insulin and troglitazone only . On day 11 , the medium was refreshed using medium supplemented with insulin only . Mature adipocytes were observed on day 14 . 3T3-L1 MBX cells ( American Type-Culture Collection , ATCC CRL-3242 ) were cultured in high glucose DMEM supplemented with 10% fetal bovine serum ( FBS ) , sodium pyruvate ( 1 mM ) , glutamine ( 2 mM ) , and 100 units/ml penicillin , and 100 µg/ml streptomycin ( Life Technologies ) . Transduction assays were performed using pLKO1 lentiviruses and selection with 2 µg/ml puromycin . The cells were differentiated to adipocytes by growing to confluence for 48–72 hr . On day 0 , the media were changed to media supplemented with 0 . 5 mM IBMX ( Sigma ) , 5 µg/ml insulin ( Sigma ) , 1 µM Troglitazone ( TZD; Calbiochem ) , and 1 µM dexamethasone ( Sigma ) ) . This medium was refreshed every 2 days . On day 4 , the medium was refreshed using medium supplemented with insulin and TZD only . On day 6 , the medium was refreshed using medium supplemented with 0 . 5 µg/ml insulin only . Mature adipocytes were observed on day 8 . E13 . 5 primary murine fibroblasts ( MEF ) obtained from mice that express 4-hydroxytamoxifen-inducible Cre were established in culture ( Das et al . , 2007 ) . CreERT2-/+ Mapk8+/+ Mapk9+/+ MEF and CreERT2-/+ Mapk8LoxP/LoxP Mapk9-/- MEF ( Das et al . , 2007 ) were cultured in DMEM supplemented with 10% fetal bovine serum , 100 units/ml penicillin , 100 µg/ml streptomycin , 0 . 1 mM 2-mercaptoethanol , and 2 mM L-glutamine ( Life Technologies ) . Transduction assays were performed using pBABE-puro retroviruses and selection with 2 µg/ml puromycin ( Lamb et al . , 2003 ) . Cells were treated with 1 µM 4-hydroxytamoxifen ( 24h ) and subsequently cultured ( 5 days ) . The cells were exposed without or with 60 J/m2 UV-C and incubated ( 60 mins ) prior to harvesting . Lysates were prepared from TNFα-treated ( 10 ng/ml , 10 mins ) and non-treated transfected HEK 293T/17 cells ( American Type-Culture Collection , ATCC CRL-11268 ) expressing Flag-tagged JNK proteins ( Gupta et al . , 1996 ) . JNK proteins were isolated by immunoprecipitation using agarose-bound Flag M2 antibody ( Sigma-Aldrich Cat# A2220 , RRID:AB_10063035 ) ( Gupta et al . , 1996 ) . JNK activity was measured using an in vitro protein kinase assay with the substrates cJun and [γ-32P]ATP as substrates ( Whitmarsh and Davis , 2001 ) . Tissue extracts were prepared using Triton lysis buffer ( 20 mM Tris-pH 7 . 4 , 1% Triton-X100 , 10% glycerol , 137 mM NaCl , 2 mM EDTA , 25 mM β-glycerophosphate , 1 mM sodium orthovanadate , 1 mM PMSF and 10 µg/mL leupeptin plus aprotinin ) . Extracts ( 30–50 µg of protein ) were examined by immunoblot analysis by probing with antibodies to cJUN ( Cell Signaling Technology Cat# 9165L , RRID:AB_2129578 ) , pSer63-cJUN ( Cell Signaling Technology Cat# 9261S , RRID:AB_2130162 ) , Flag M2 ( Sigma-Aldrich Cat# F1804 , RRID:AB_262044 ) , GAPDH ( Santa Cruz Biotechnology Cat# sc-365062 , RRID:AB_10847862 ) , JNK1/2 ( BD Biosciences Cat# 554285 , RRID:AB_395344 ) , pThr180-Pro-pTyr182-JNK ( pJNK ) ( Cell Signaling Technology Cat# 4668P , RRID:AB_10831195 ) , NOVA1 ( Abcam Cat# ab97368 , RRID:AB_10680798 ) , NOVA2 ( Sigma-Aldrich Cat# AV40400 , RRID:AB_1854572 ) , and βTubulin ( BioLegend Cat# 903401 , RRID:AB_2565030 ) . Immunocomplexes were detected by fluorescence using anti-mouse and anti-rabbit secondary IRDye antibodies ( Li-Cor ) and quantitated using the Li-Cor Imaging system Histology was performed using tissue fixed in 10% formalin for 24 hr , dehydrated , and embedded in paraffin . Sections ( 7 µm ) were cut and stained using hematoxylin & eosin ( American Master Tech Scientific ) . Paraffin sections were stained with an antibody to insulin ( Dako Cat# A0564 , RRID:AB_10013624 ) that was detected by incubation with anti-Ig conjugated to Alexa Fluor 488 ( Life Technologies ) . DNA was detected by staining with DAPI ( Life Technologies ) . Fluorescence was visualized using a Leica TCS SP2 confocal microscope equipped with a 405-nm diode laser . Inguinal , retroperitoneal , and epididymal fat pads were surgically removed at necropsy . Adipocytes were isolated after incubation ( 40 min at 37°C with shaking ) of adipose tissue in 12 . 5 mM Hepes pH 7 . 4 , 120 mM NaCl , 6 mM KCl , 1 . 2 mM MgSO4 , 1 mM CaCl2 , 2% BSA , 2 . 5 mM glucose , 1 mg/ml collagenase II ( Sigma #C6885 ) and 0 . 2 mg/ml DNAse I ( Sigma #DN25 ) ) . Larger particles were removed using a 100 µm nylon sieve and the filtrates were centrifuged at 1000 rpm ( 3 min ) . Floating adipocytes were washed twice with 1x PBS and subsequently centrifuged at 1000 rpm ( 3 min ) prior to RNA isolation . The stromal vascular fraction ( SVF ) was collected after centrifugation at 3000 rpm ( 5 min ) prior to RNA isolation . The expression of adipocyte marker genes ( Adipoq & Leptin ) , SVF marker genes ( Emr1 ( F4/80 ) , Itgam ( Cd11b ) , Cd68 ) in adipocytes and the SVF was measured by quantitative RT-PCR analysis ( Figure 1—figure supplement 3 ) . The expression of a gene ( Fabp4 ) that is expressed by both adipocytes and SVF was also examined . RNA was isolated using the RNeasy kit ( Qiagen ) . RNA quality ( RIN > 9 ) was verified using a Bioanalyzer 2100 System ( Agilent Technologies ) . Total RNA ( 10 µg ) was used for the preparation of each RNA-seq library by following the manufacturer’s instructions ( Illumina ) . Three or four independent libraries prepared from different mice were sequenced ( Illumina ) for each condition . Table 1 presents a summary of the adipocyte RNA-seq data and associated GEO accession numbers . The liver RNA-seq data ( CD vs HFD ( n=3 ) ) were previously reported ( Vernia et al . , 2014 ) ( GEO accession number GSE55190 ) . The expression of mRNA was examined by quantitative PCR analysis using a Quantstudio PCR system ( Life Technologies ) . Taqman assays were used to quantitate Adipoq ( Mm00456425_m1 ) , Adrb3 ( Mm02601819_g1 ) , Cd68 ( Mm03047340_m1 ) , Dio2 ( Mm00515664_m1 ) , Emr1 ( F4/80 ) ( Mm00802530_m1 ) , Fabp4 ( Mm00445880_m1 ) , Itgam ( Cd11b ) ( Mm00434455_m1 ) , Leptin ( Mm00434759_m1 ) , Mapk8 ( Mm00489514_m1 ) , Mapk9 ( Mm00444231_m1 ) , Nova1 ( Mm01289097_m1 ) , Nova2 ( Mm01324153_m1 ) , Ppargc1a ( Mm00447183_m1 ) , Ppargc1b ( Mm00504720_m1 ) , Prdm16 ( Mm00712556_m1 ) , Ucp1 ( Mm01244861_m1 ) mRNA and 18S RNA ( 4308329 ) ( Applied Biosystems ) . Standard curves were constructed using the threshold cycle ( Ct ) values for each template dilution plotted as a function of the logarithm of the amount of input template . The number of mRNA copies for each gene-sample combination was calculated using the slope of the standard curve . To obtain a normalized abundance , copy numbers were corrected for the amount of 18S RNA in each sample . The inclusion of exons 7a and 7b in Mapk8 and Mapk9 mRNA was examined by quantitative PCR using the Quantifast probe PCR kit ( Qiagen ) and the following combination of primers and Taqman probes ( Applied Biosystems ) : Mapk8α: Fwd , GGAGAACGTGGACTTATGGTCTGT; Probe: 6FAM-TGCCACAAAATCCT-MGBNFQ; Rev , TGATCAATATAGTCCCTTCCTGGAA . Mapk8β: Fwd , GAACGTTGACATTTGGTCAGTTG; Probe , 6FAM-AGAAATGATCAAAGGTGGTGTT-MGBNFQ; Rev , TCAATATGATCTGTACCTGGGAACA . Mapk9α: Fwd , GGTCAGTGGGTTGCATCATG; Probe , 6FAM-AGCTGGTGAAAGGTT-MGBNFQ; Rev , TGATCAATATGGTCAGTACCTTGGA . Mapk9β: Fwd , ATCTGGTCTGTCGGGTGCAT; Probe: 6FAM-AAATGGTCCTCCATAAAG-MGBNFQ; Rev , GATCAATATAGTCTCTTCCTGGGAACA . Mapk8 and Mapk9 spliced variants were quantitated using the relative quantification method . The alternative splicing is represented as the ratio Mapk8α/Mapk8β and Mapk9α/Mapk9β . RT-PCR analysis was performed using amplimers based on the sequences of Adam15 exons 19 and 21 ( svADAM15F1: 5’-GCGGGCACAGCAGATGAC-3’ and svADAM15R1: 5’- GGGTTGGCAGGCAGTGGC-3’ ) and Yap1 exons 5 and 7 ( svYap1F1: 5-GGAGAAGGAGAGACTG-3’ and svYap1R2: 5’-GTCCCTCCATCCTGCTC-3’ ) . The PCR products were examined by agarose gel electrophoresis and staining with ethidium bromide . Adam15 mRNA and exon 20-skipped Adam15 mRNA were detected as 277 bp and 78 bp DNA fragments . Yap1 mRNA and exon 6-skipped Yap1 mRNA were detected as 144 bp and 96 bp DNA fragments . Fastqc v0 . 10 . 1 ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) and cutadapt ( v1 . 7 ) ( https://pypi . python . org/pypi/cutadapt/1 . 7 . 1 ) were used to generate sequence quality reports and trim/filter the sequences respectively . Reads below a minimum quality PHRED score of 30 at the 3’ end were trimmed ( Table 1 ) . The filtered reads were aligned to the mouse reference genome ( Ensembl GRCm38 ) . Alignments were carried out using Bowtie2 ( v2 . 2 . 1 . 0 ) ( Langmead and Salzberg , 2012 ) and Tophat2 ( v2 . 0 . 9 ) ( Kim et al . , 2013 ) . Samtools ( v0 . 0 . 19 ) ( Li et al . , 2009 ) and IGV ( v2 . 3 . 60 ) ( Thorvaldsdottir et al . , 2013 ) were used for indexing the alignment files and viewing the aligned reads respectively . Gene expression was quantitated as fragments per kilobase of exon model per million mapped fragments ( FPKM ) using Cufflinks ( v2 . 2 . 0 ) ( Trapnell et al . , 2010 ) . Differentially expressed genes were identified using Cufflinks tools ( Cuffmerge and Cuffdiff ) . CummeRbund ( v2 . 4 . 1 ) ( Trapnell et al . , 2012 ) was used to assess biological replicate concordance . Significant changes in gene expression were defined as q<0 . 05 and absolute log2 fold change >0 . 75 . Gene sets were examined using Wikipathways in Webgestalt ( Wang et al . , 2013a ) . Alternative splicing was examined using rMATS software ( v3 . 0 . 9 ) ( Shen et al . , 2014 ) . rMATS was run using default settings to compute p-values and FDRs of splicing events . Significant changes in alternative splicing were defined as FDR<0 . 05 and absolute ∆Inc level >0 . 1 . Rmats2sashimiplot ( https://github . com/Xinglab/rmats2sashimiplot ) and Sashimiplot ( Katz et al . , 2015 ) were used for quantitative visualization of alternative exon expression from rMATS . The microarray analysis of human adipose tissue mRNA expression data ( GEO GSE25402 ) ( Arner et al . , 2012 ) was performed using Genespring . The summarization method was based on RMA16 , the normalization method was based on the median approach on log2 scale , and the fold change was computed from average signal intensity values . NOVA interactions with pre-mRNA were examined using CLIP-seq data obtained from murine brain ( Licatalosi et al . , 2008 ) provided ( in BED format ) by Dr . R . B . Darnell ( Rockefeller University ) . The coordinates were based on the mm9 assembly . The UCSC liftOver utility was used to convert the NOVA CLIP-seq tag coordinates from the mm9-based assembly to the mm10/GRCm38 assembly . The coordinates for the NOVA CLIP-seq tags were sorted based on chromosome and start position . Custom PERL scripts ( Source code 1 and 2 ) were used to extract three groups of genomic regions: ( 1 ) statistically significant differentially expressed alternatively spliced exons ( FDR<0 . 05 and the absolute inclusion level difference >0 . 1 ) ( n = 1631 ) ; ( 2 ) exons not meeting statistical significance ( FDR≥0 . 05 ) ( n = 11 , 490 ) ; and ( 3 ) randomly selected regions of the mouse genome comprising 100 bp ( n = 10 , 000 ) . The coordinates for each genomic region were expanded to include an additional 500 bp of sequence flanking the 5’ and 3’ ends of the genomic region . If an exon belongs to a gene comprising one or more alternatively spliced exons with FDR<0 . 5 and one or more alternatively spliced exons with FDR≥0 . 05 , the exon was excluded from the second group . In each group , the regions were sorted based on the chromosome and the start position . Duplicated regions were removed . The BEDtools ( 2 . 22 . 0 ) intersect command ( Quinlan , 2014 ) was used to determine the intersection between the regions in each group and the NOVA-CLIP-seq tags . The number of intersecting regions between a group and the NOVA binding sites were tallied . For each group , the number and the percentage of regions with and without NOVA binding sites were calculated ( Figure 1E ) . Thus , NOVA CLIP-tags intersecting with the following groups were examined: ( 1 ) statistically significant differentially spliced exons plus 500 bp of flanking sequence both upstream and downstream; ( 2 ) non-alternatively spliced exons plus 500 bp of flanking sequence both upstream and downstream; ( 3 ) random genomic sequences of similar fragment size ( Figure 1E ) . Statistical significance between two groups was determined by the Pearson’s Chi-squared test . Values are given as means ± SEM of at least three independent experiments . Sample sizes were determined by prior experimentation . Differences between groups were examined for statistical significance using the Student´s test or analysis of variance ( ANOVA ) with the Fisher’s test ( p<0 . 05 , p<0 . 01 and *p<0 . 001 ) .	The process of building a protein from the information encoded in a gene begins when the gene is copied to form a pre-messenger RNA molecule . This molecule is then edited to produce a final messenger RNA that is “translated” to form the protein . Different segments of the pre-messenger RNA molecule can be removed to create different messenger RNAs . This “alternative splicing” enables a single gene to produce multiple protein variants , allowing a diverse range of processes to be performed by cells . Fat cells store energy in the form of fats and can release this energy as heat in a process called thermogenesis . This helps to regulate the body’s metabolism and prevent obesity . Vernia et al . now find that that feeding mice a high-fat diet causes widespread changes in alternative splicing in fat cells . Further bioinformatics analysis revealed that the NOVA family of splicing factors – proteins that bind to the pre-messenger RNAs to control alternative splicing – contribute to the alternative splicing of around a quarter of the genes whose splicing changes in response to a fatty diet . Mice whose fat cells were deficient in the NOVA splicing factors displayed increased thermogenesis . As a consequence , when these animals were fed a high-fat diet , they gained less weight than animals in which NOVA proteins were present . Their metabolic activity was also better , meaning they were less likely to show the symptoms of pre-diabetes . Moreover , the activity of certain genes that are known to promote thermogenesis was greater in the fat cells that were deficient in NOVA proteins . Overall , the results presented by Vernia et al . suggest that the normal role of NOVA proteins is to carry out an alternative splicing program that suppresses thermogenesis , which in turn may promote obesity . Drugs that are designed to target NOVA proteins and increase thermogenesis may therefore help to treat metabolic diseases and obesity . The next step is to identify the protein variants that are generated by NOVA proteins and work out how they contribute to thermogenesis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	An alternative splicing program promotes adipose tissue thermogenesis
Venom toxins are invaluable tools for exploring the structure and mechanisms of ion channels . Here , we solve the structure of double-knot toxin ( DkTx ) , a tarantula toxin that activates the heat-activated TRPV1 channel . We also provide improved structures of TRPV1 with and without the toxin bound , and investigate the interactions of DkTx with the channel and membranes . We find that DkTx binds to the outer edge of the external pore of TRPV1 in a counterclockwise configuration , using a limited protein-protein interface and inserting hydrophobic residues into the bilayer . We also show that DkTx partitions naturally into membranes , with the two lobes exhibiting opposing energetics for membrane partitioning and channel activation . Finally , we find that the toxin disrupts a cluster of hydrophobic residues behind the selectivity filter that are critical for channel activation . Collectively , our findings reveal a novel mode of toxin-channel recognition that has important implications for the mechanism of thermosensation . The venom of poisonous animals contains an arsenal of protein toxins that target ion channel proteins to paralyze prey and induce pain by activating nociceptive sensory neurons ( Bohlen and Julius , 2012; Kalia et al . , 2015 ) . These toxins , which typically contain multiple disulfide bonds that stabilize their tertiary structure ( Norton and Pallaghy , 1998; Pallaghy et al . , 1994 ) , activate or inhibit ion channel proteins that open and close in response to membrane voltage ( Cestele et al . , 1998; Milescu et al . , 2013; Swartz and MacKinnon , 1997a ) , binding of neurotransmitters ( Celie et al . , 2005; Chen et al . , 2014; Dellisanti et al . , 2007; Ulens et al . , 2006 ) or other sensory stimuli ( Bohlen et al . , 2011; Bohlen et al . , 2010 ) . Thus , these small proteins are invaluable tools for investigating ion-channel structures and operational mechanisms . Many of these protein toxins have evolved to interact with functionally important domains of ion channels that are freely accessible to water on the external side of the cell membrane , implying that toxin-channel recognition takes place in an aqueous environment . For example , the scorpion toxin charybdotoxin binds to the extracellular side of the pore of potassium channels to physically block the flow of ions ( MacKinnon and Miller , 1988 ) . A recent X-ray structure of a complex between charybdotoxin and the Kv1 . 2-2 . 1 paddle chimera channel revealed that this block is effected by a lysine residue that the toxin inserts into the selectivity filter of the channel , mimicking a K+ ion ( Banerjee et al . , 2013 ) . Another example is the tarantula toxin PcTx1 , which activates acid-sensing ion channels ( ASICs ) by clamping onto helix 5 within the large extracellular domain of the channel , and inserting an Arg finger motif into a subunit interface where protons bind ( Baconguis and Gouaux , 2012; Chen et al . , 2006; Dawson et al . , 2012; Salinas et al . , 2006 ) . The ASIC channel was also crystalized in complex with MitTx ( Baconguis et al . , 2014 ) , a structurally distinct two-subunit snake toxin that binds to an extended region of the extracellular domain of the channel , stabilizing it in an open conformation . The general principle of toxins binding to ion channels within aqueous environments is further illustrated by X-ray structures of a glutamate-activated cation channel in complex with the cone snail toxin con-ikot-ikot ( Chen et al . , 2014 ) , and of the extracellular domains of Cys-loop receptors in complex with cone snail or snake toxins ( Celie et al . , 2005; Dellisanti et al . , 2007; Ulens et al . , 2006 ) . These striking structures not only reveal in great detail the nature of toxin-channel interactions in solution , but also provide valuable insights into the operational mechanisms of an array of ion channels . This notwithstanding , it has been recently suggested that certain types of toxins can target domains within ion channel proteins that are embedded within the lipid membrane . For example , tarantula toxins that modify the gating of voltage-activated ion channels partition into membranes and are thought to bind to the voltage-sensing domains within the membrane environment ( Alabi et al . , 2007; Gupta et al . , 2015; Lee and MacKinnon , 2004; Mihailescu et al . , 2014; Milescu et al . , 2009; Milescu et al . , 2007; Phillips et al . , 2005b ) . Thus far , however , no structures of toxin-channel complexes have been solved for this class of toxins , and therefore our understanding of toxin-channel interactions within membrane environments is limited . Double-knot toxin ( DkTx ) is a tarantula toxin isolated from the venom of a Chinese bird spider ( Bae et al . , 2012; Bohlen et al . , 2010 ) that activates the transient receptor potential vanilloid 1 ( TRPV1 ) channel . TRPV1 is a cation channel expressed in nociceptive sensory neurons that plays important roles in the transduction of noxious stimuli as well as thermosensation ( Julius , 2013 ) , but the molecular mechanism of heat-dependent activation has remained elusive . From a structural perspective , DkTx is intriguing because it has an unusual bivalent architecture , being comprised of two inhibitor-cysteine-knot ( ICK ) motifs that have been designated as the K1 and K2 lobes . DkTx is also quite hydrophobic , requiring the use of detergents to efficiently fold in vitro ( Bae et al . , 2012 ) , which raises the possibility that the toxin might interact with TRPV1 in a membrane environment , similar to toxins targeting voltage-sensing domains of voltage-activated ion channels . In a recent breakthrough , near-atomic resolution structures of the TRPV1 channel were solved using single-particle cryo-electron microscopy ( cryo-EM ) , including a closed state ( apo ) , a capsaicin-bound state , and an open state with both DkTx and the vanilloid resiniferatoxin ( RTx ) bound ( Cao et al . , 2013; Liao et al . , 2013 ) . The structure of the complex between TRPV1 and DkTx/RTx unambiguously reveals that the DkTx lobes bind to sites at the periphery of the external pore of the channel . However , the resolution of electron density maps was insufficient to reveal the structure of DkTx in atomic detail . Therefore , the binding arrangement of the two DkTx knots is unknown , as is the nature of the interactions between toxin and channel , or whether the interaction occurs in aqueous or membrane environments . How DkTx recognition ultimately results in channel opening , and how this process might relate to the mechanism of temperature-sensing , are also open questions . In the present study , we use nuclear magnetic resonance ( NMR ) spectroscopy to solve the solution structure of DkTx , and the molecular-modeling suite ROSETTA together with existing cryo-EM density maps to derive high-quality structural models of the TRPV1 channel with and without DkTx bound . We also investigate the interaction of the toxin with the TRPV1 channel and lipid membranes using fluorescence spectroscopy , electrophysiological recordings and molecular dynamics ( MD ) simulations . Our results demonstrate that DkTx interacts intimately with TRPV1 while inserting hydrophobic residues into the surrounding lipid membrane , and provide the first example of a complex structure between toxin and channel within a membrane environment . Comparison of the improved structures of apo and toxin-bound TRPV1 reveals novel insights into the mechanism by which DkTx induces channel opening , and provides new insight into the mechanism of thermosensation . Our first objective was to solve the structure of DkTx in solution at atomic resolution through solution NMR spectroscopy . Because previous studies have demonstrated that the two lobes ( or knots ) of the toxin , K1 and K2 , can fold independently and can also activate the TRPV1 channel ( Bae et al . , 2012; Bohlen et al . , 2010 ) , we synthesized K1 and K2 individually using solid-phase peptide synthesis . The full-length toxin was also produced recombinantly , with and without 15N-labeling . All constructs were folded in vitro in the presence of detergent , and correctly folded species were purified using reversed-phase HPLC . Our strategy was to first use two-dimensional ( 2D ) 1H-1H nuclear Overhauser effect spectroscopy ( NOESY ) to study each of the lobes separately , because these measurements would yield less crowded spectra and thus facilitate the assignment of a greater number of cross-peaks , which could then be translated into nuclear Overhauser effect ( NOE ) inter-proton distance restraints for structure determination . We then conducted additional NMR experiments using 15N labeled DkTx , taking advantage of the separation of overlapped spectra in 2D 1H-1H NOESY using 15N , and compared the backbone proton resonances of K1 and K2 with the full-length protein . Complete proton resonance assignments for K1 and K2 were made using traditional 2D NMR sequential assignment techniques ( Wüthrich , 1986 ) ( Figure 1—figure supplement 1 ) . Using the proton chemical shift values of isolated K1 and K2 as reference , proton resonances in DkTx were readily identified ( Figure 1—figure supplement 1 , 2 ) . The backbone proton chemical shift values of DkTx were found to be nearly identical to those measured for K1 and K2 separately , except for a few residues in the linker region and in the N- and C-termini of the toxin ( Figure 1—figure supplement 3 ) . We can therefore conclude that the structures of isolated K1 and K2 are highly similar to those in full-length DkTx . The structures of K1 and K2 were determined using the inter-proton distance restraints deduced from the NOESY data ( Figure 1—figure supplement 1 ) , in addition to backbone dihedral angle phi restraints estimated from DQF-COSY spectra ( Kim and Prestegard , 1989 ) , and disulfide-bond restraints inferred from homology with other ICK toxins of known structure . Using the simulated-annealing method and energy function in Xplor-NIH 2 . 37 ( Schwieters et al . , 2006; Schwieters et al . , 2003 ) , we derived a set of 20 energy-minimized structures for each knot that fulfill the NMR restraints , as well as generic protein-geometry and other knowledge-based restraints ( Bermejo et al . , 2012 ) . Statistical analysis of these ensembles ( Figure 1—figure supplement 4 ) indicates that the structures are determined without significant stereochemical violations . A superposition of the 20 energy-minimized structures obtained for each lobe demonstrates that the protein backbone is well defined ( Figure 1A , B ) , except in the inter-connecting linker , for which we obtained relatively few distance restraints . Each lobe consists of two anti-parallel beta-strands ( F21-Y22/K32-K33 in K1 and L61-D62/Y70-C71 in K2 ) , stabilized by three disulfide bonds . 10 . 7554/eLife . 11273 . 003Figure 1 . NMR solution structures of K1 and K2 . ( A , B ) Ensemble of backbone structures of K1 ( A ) , in green , and K2 ( B ) , in cyan . The 20 lowest-energy structures out of 100 energy-minimized structures are shown ( PDB entry 2N9Z for K1 and 2NAJ for K2 ) . The distinct orientations of W11 and W53 , as well as their interactions with residues in loop 4 , are also depicted . The backbone root mean square deviation ( RMSD ) of the structures in these ensembles relative to the average is 0 . 4 Å for both K1 and K2 ( for the most ordered regions of each protein , which include Cys2-Cys31 in K1 and Cys44-Cys71 in K2 ) . ( C , D ) Surface representations of energy-minimized structures of K1 ( C ) and K2 ( D ) . Hydrophobic , basic and acidic residues are colored in green , blue and red , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 00310 . 7554/eLife . 11273 . 004Figure 1—figure supplement 1 . NMR spectra of K1 , K2 and DkTx in solution . ( A , B ) Fingerprinting regions of 1H-1H NOESY ( 298 K ) spectra of K1 ( A ) and K2 ( B ) . Sequential dαN ( i , i+1 ) NOE connectivities are shown in red lines . All intra-residue HN-Hα cross peaks are identified ( except prolines and N-terminal residue of K1/K2 ) and labeled as single-letter amino acid with the residue numbering . ( C ) 1H-15N hetronuclear single quantum coherence ( HSQC ) spectrum of DkTx . All backbone amide groups in the DkTx were labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 00410 . 7554/eLife . 11273 . 005Figure 1—figure supplement 2 . Backbone assignments of DkTx in solution . Strips are derived from 1H-15N NOESY-HSQC spectrum of DkTx , and each strip represents single backbone nitrogen resonance . Residue number and 1HN/15N resonances are labeled above and below the stripes , respectively . Sequential NOE connectivity of DkTx is shown in red-dashed lines . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 00510 . 7554/eLife . 11273 . 006Figure 1—figure supplement 3 . Summary of 3JHNHα coupling constant and proton chemical shifts of DkTx in solution . ( A ) Summary of 3JHNHα coupling constants , secondary structure elements , and chemical shift index ( CSI ) for DkTx . Values of 3JHNHα are represented as ↑ ( >8 Hz ) or ↓ ( <5 . 5Hz ) . ( B ) Chemical shift comparison between K1/K2 lobes and DkTx . Chemical shift difference ( ΔCS in ppm ) between K1/K2 lobe and DkTx was calculated according to ΔCS = [ ( ΔδHα ) 2+ ( ΔδHN ) 2]1/2 , where ΔδHα denotes Hα chemical shift difference between DkTx and K1/K2 , and ΔδHN denotes HN chemical shift difference between DkTx and K1/K2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 00610 . 7554/eLife . 11273 . 007Figure 1—figure supplement 4 . NMR and refinement statistics of K1 and K2aPairwise r . m . s . deviation to the mean was calculated among 20 refined structures for ordered residues , namely 2–31 in K1 and 71 in K2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 007 Comparison of the structures of K1 and K2 reveals that their fold is similar , but not identical; the root mean square difference ( RMSD ) in their backbone , for the core 28 residues , is 1 . 45 Å ( 1 . 56 Å including all heavy-atoms ) . Two notable differences between K1 and K2 are the length of loop 3 , which is longer in K1 than in K2 , and the configuration of a tryptophan side-chain in loop 2 ( W11 in K1 and W53 in K2 ) , which is more solvent accessible in K1 ( 154 ± 16 Å2 ) than in K2 ( 132 ± 7 Å2 ) . The latter stems from actual differences in the NOESY spectra for K1 and K2; the number of NOE cross-peaks between the conserved Trp in loop 2 and the conserved Ile and Pro residues in loop 4 are fewer in K1 than in K2 . The likely cause of this structural difference is a serine ( S10 ) to glycine ( G52 ) substitution at the first position in loop 2 , which seems to allow this loop to bend towards loop 4 in K2 ( Figure 1A , B ) . Surface renderings of K1 and K2 show that both lobes feature a large cluster of exposed hydrophobic residues ( green ) , comprised of W11 , M25 , F27 and I28 in K1 and W53 , L65 , A66 , F67 and I68 in K2 ( Figure 1C , D ) . These extensive and well-conserved hydrophobic surfaces are consistent with our previous finding that detergents are critical for efficient refolding of the toxin in vitro , and with the observation that folded DkTx is more hydrophobic than the linear form ( Bae et al . , 2012 ) . The hydrophobic surfaces in both K1 and K2 are surrounded by basic residues ( blue ) and acidic residues ( red ) , including D1 , E7 , K13 , K14 , H30 and K32 in K1 and E47 , E49 , K56 , E72 , K73 and R75 in K2 . This pronounced amphipathic character is reminiscent of well-characterized voltage-sensor toxins such as hanatoxin ( Takahashi et al . , 2000 ) , SGTx1 ( Lee et al . , 2004 ) , VSTx1 ( Jung et al . , 2005 ) and GxTx-1E ( Lee et al . , 2010 ) , which interact with voltage-sensing domains of Kv channels within a membrane environment ( Gupta et al . , 2015; Lee and MacKinnon , 2004; Milescu et al . , 2009; Milescu et al . , 2007; Phillips et al . , 2005b ) . To begin to discern the mode in which DkTx recognizes and activates TRPV1 , we set out to improve the existing atomic models of the structure of the channel , both in the apo state as well as in complex with DkTx , based on cryo-EM data obtained in a previous study ( Cao et al . , 2013 ) . Notwithstanding the groundbreaking insights provided by these cryo-EM maps , the associated structural models deposited in the Protein Data Bank ( PDB ) are objectively suboptimal in several respects . Aside from the missing backbone fragments and side-chains in unresolved regions , the published structural models show a significant number of stereochemical violations and steric overlaps ( Tables 1 , 2 ) . As an example , the MolProbity score , which is a standardized metric of model quality among structures of comparable resolution , ranks PDB entries 3J5Q and 3J5P in the 29th and 36th percentile , respectively ( with 100th percentile being the best score ) . A specific issue pertaining to the published model of the TRPV1/DkTx/RTx complex is that it is based on a map with imposed fourfold symmetry , despite the fact that the two DkTx lobes are not identical and thus the appropriate symmetry for the complex would be twofold . More importantly , in the published structure of the TRPV1/DkTx/RTx complex , the toxin is modeled as a poly-alanine chain based on the structure of hanatoxin because an atomic-resolution structure of DkTx had not yet been solved . Assignment of the K1 and K2 knots to the EM map was therefore not possible , and so the nature of their interaction with the channel or the membrane has remained unclear . 10 . 7554/eLife . 11273 . 008Table 1 . Evaluation of structural models of apo TRPV1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 008PDB entry 3J5P Number of atoms: 18 , 628Correlation coefficient with cryo-EM map EMD-5778 ( 3 . 4 Å ) : 0 . 71All-atom contactsClashscore , all atomsa76 . 3116th percentilebProtein geometryPoor rotamers52426 . 57%Ideal: <1%Ramachandran outliers00 . 00%Ideal: <0 . 05%Ramachandran favored220494 . 35%Ideal: >98%MolProbity scorec3 . 8236th percentilebCβ deviations >0 . 25 Å00 . 00%Ideal: 0Bad backbone bonds166/190320 . 87%Ideal: 0%Bad backbone angles72/257600 . 28%Ideal: < 0 . 1%Partial ROSETTA modeldNumber of atoms: 18 , 628Correlation coefficient with cryo-EM map EMD-5778 ( 3 . 4 Å ) : 0 . 71All-atom contactsClashscore , all atomsa3 . 4597th percentilebProtein geometryPoor rotamers40 . 20%Ideal: <1%Ramachandran outliers120 . 51%Ideal: <0 . 05%Ramachandran favored220094 . 18%Ideal: >98%MolProbity scorec1 . 5394th percentilebCβ deviations >0 . 25 Å00 . 00%Ideal: 0Bad backbone bonds44/190240 . 23%Ideal: 0%Bad backbone angles56/257440 . 22%Ideal: < 0 . 1%Complete ROSETTA model Number of atoms: 19 , 632Correlation coefficient with cryo-EM map EMD-5778 ( 3 . 4 Å ) : 0 . 71All-atom contactsClashscore , all atomsa3 . 7796th percentilebProtein geometryPoor rotamers80 . 37%Ideal: <1%Ramachandran outliers321 . 33%Ideal: <0 . 05%Ramachandran favored223692 . 70%Ideal: >98%MolProbity scorec1 . 6292nd percentilebCβ deviations >0 . 25 Å00 . 00%Ideal: 0Bad backbone bonds52/200680 . 26%Ideal: 0%Bad backbone angles148/271440 . 55%Ideal: < 0 . 1%aClashscore is the number of serious steric overlaps ( >0 . 4 Å ) per 1 , 000 atoms . b100th percentile is the best score among structures of comparable resolution; 0th percentile is the worst score . For clashscore , the comparative set of structures was selected in 2004 , for MolProbity score in 2006 . cThe MolProbity score combines the clashscore , rotamer and Ramachandran evaluations into a single score , normalized to be in the same scale as X-ray resolution . dThe partial ROSETTA model is identical to the complete ROSETTA model , except that it includes only the set of atoms resolved in PDB entry 3J5P . 10 . 7554/eLife . 11273 . 009Table 2 . Evaluation of structural models of the TRPV1-DkTx complex . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 009PDB entry 3J5Q Number of atoms: 18 , 244Correlation coefficient with cryo-EM map EMD-5776 ( fourfold symmetric , 3 . 8 Å ) : 0 . 78All-atom contactsClashscore , all atomsa101 . 8510th percentilebProtein geometryPoor rotamers45727 . 33%Ideal: <1%Ramachandran outliers200 . 82%Ideal: <0 . 05%Ramachandran favored226492 . 33%Ideal: >98%MolProbity scorec4 . 0429th percentilebCβ deviations >0 . 25 Å00 . 00%Ideal: 0Bad backbone bonds106/180000 . 59%Ideal: 0%Bad backbone angles50/244440 . 20%Ideal: < 0 . 1%Partial ROSETTA modeldNumber of atoms: 18 , 244Correlation coefficient with cryo-EM map relion_ct16_halfSUM_mr ( twofold symmetric , 4 . 3 Å ) : 0 . 77All-atom contactsClashscore , all atomsa4 . 0696th percentilebProtein geometryPoor rotamers60 . 35%Ideal: <1%Ramachandran outliers361 . 48%Ideal: <0 . 05%Ramachandran favored223491 . 56%Ideal: >98%MolProbity scorec1 . 6989th percentilebCβ deviations >0 . 25 Å40 . 17%Ideal: 0Bad backbone bonds43/186100 . 23%Ideal: 0%Bad backbone angles100/252580 . 40%Ideal: < 0 . 1%Complete ROSETTA model Number of atoms: 20 , 814Correlation coefficient with cryo-EM map relion_ct16_halfSUM_mr ( twofold symmetric , 4 . 3 Å ) : 0 . 76All-atom contactsClashscore , all atomsa6 . 3989th percentilebProtein geometryPoor rotamers120 . 53%Ideal: <1%Ramachandran outliers722 . 81%Ideal: <0 . 05%Ramachandran favored228489 . 29%Ideal: >98%MolProbity scorec1 . 9280th percentilebCβ deviations >0 . 25 Å100 . 41%Ideal: 0Bad backbone bonds150/212880 . 70%Ideal: 0%Bad backbone angles261/287920 . 91%Ideal: < 0 . 1%aClashscore is the number of serious steric overlaps ( >0 . 4 Å ) per 1 , 000 atoms . b100th percentile is the best score among structures of comparable resolution; 0th percentile is the worst score . For clashscore , the comparative set of structures was selected in 2004 , for MolProbity score in 2006 . cThe MolProbity score combines the clashscore , rotamer and Ramachandran evaluations into a single score , normalized to be in the same scale as X-ray resolution . dThe partial ROSETTA model is identical to the complete ROSETTA model , except that it includes only the set of atoms resolved in PDB entry 3J5Q . To tackle these methodological issues , we used the molecular-modeling software suite ROSETTA ( Leaver-Fay et al . , 2011 ) ( see Methods ) . This approach enabled us to develop atomic models of apo TRPV1 and the TRPV1-DkTx complex that are considerably improved; the updated MolProbity scores rank these complete models in the 92nd and 80th percentile , respectively , while maintaining the original quality of the fit to the experimental cryo-EM maps ( Tables 1 , 2 ) . To construct the model of the TRPV1-DkTx complex , we used an unpublished cryo-EM map with imposed twofold symmetry , kindly provided by Yifan Cheng and colleagues . Although the resolution of this map is somewhat inferior to that of the fourfold symmetrized map utilized originally ( 4 . 3 Å compared to 3 . 8 Å ) , densities for each of the four DxTk knots ( two per toxin ) are clearly discernable . We initially docked the NMR structures of K1 and K2 into these densities using Xplor-NIH , in either a clockwise ( CW ) or counter-clockwise ( CCW ) configuration ( viewed from the extracellular side ) , optimizing the fit of each lobe to the EM map while also applying NOE distance restraints within each knot ( based on the 2D 1H-1H NMR experiments for K1 and K2 in solution ) . These two partial models were then input into ROSETTA , which was then used to generate a randomized ensemble of 100 models of the complete TRPV1-DkTx complex for each K1-K2 configuration . Analysis of these two ensembles shows that the majority of the 100 CCW models fit to the experimental cryo-EM map significantly better , and also have a better ROSETTA score , than the majority of the 100 CW models ( Figure 2E ) . This result is consistent with the observation that the number of NOE-restraint violations in the CW initial seed model generated with Xplor-NIH was twice as large as for the CCW model , for a comparable fit quality ( Figure 2—figure supplement 1 ) . We can therefore conclude unambiguously that DkTx binds to the outer pore of TRPV1 in a CCW configuration ( Figure 2B , D , F; Video 1 ) . 10 . 7554/eLife . 11273 . 010Figure 2 . Docking of DkTx into the electron density map of DkTx/RTx-bound TRPV1 . ( A , B ) Top-ranking ROSETTA models of the structure of TRPV1 with DkTx bound in a clockwise ( CW , A ) or counter-clockwise ( CCW , B ) configuration . The structures are overlaid on the experimental twofold symmetric cryo-EM map ( unpublished data kindly provided by Yifan Cheng and colleagues ) , and densities close to the toxin molecules are shown at a contour level of 3σ . K1 and K2 are shown in green and cyan , respectively . ( C , D ) View of K1 and K2 from the membrane plane , either in the CW ( C ) or CCW ( D ) configuration . The cryo-EM map is shown as a gray surface . ( E ) Quantitative evaluation of the CW and CCW configurations of DkTx bound to TRPV1 using ROSETTA . The ROSETTA fit-to-density score and the conventional ROSETTA score ( excluding the fit-to-density contribution ) are reported for 100 models generated for either the CW or CCW configuration . ( F ) View of the top-ranking ROSETTA model of the TRPV1-DkTx complex , with the K1 and K2 in a CCW configuration . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01010 . 7554/eLife . 11273 . 011Figure 2—figure supplement 1 . Evaluation of the CW and CCW orientations for docking of K1 and K2 to TRPV1 using Xplor-NIH . aCorrelation coefficient between calculated density maps generated with Xplor-NIH for TRPV1-docked K1 and K2 and the twofold cryo-EM map of the TRPV1-DkTx complex . Shown are the mean values over 10 energy-minimized structures ( ± s . d . ) . bRMSD of violated NOE distance restraints in the same 10 structures ( mean ± s . d . ) . cNumber of violated NOE distance restraints in the same 10 structures ( mean ± s . d . ) . The total number of NOE distance restraints is 548 for K1 and 476 for K2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01110 . 7554/eLife . 11273 . 012Figure 2—figure supplement 2 . Refinement of cryo-EM structures of apo and DkTx-bound TRPV1 using ROSETTA . Score distribution for the 12 , 000 ROSETTA models obtained during the second stage of refinement for the apo ( A ) and DkTx-bound ( B ) TRPV1 channel ( see Methods ) . The ROSETTA fit-to-density score ( based on the experimental cryo-EM maps ) is reported against the total ROSETTA score ( excluding the fit-to-density contribution ) for all the generated models . Note that the fit-to-density score was not optimized in this second stage . The red circles represent the best ROSETTA model obtained after the first refinement stage; the green circles represents the two final models selected , based on clustering analysis of the two 12 , 000-model ensembles . These models are the most representative structure of the most populated cluster in each case . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01210 . 7554/eLife . 11273 . 013Figure 2—figure supplement 3 . Improvements in the newly refined structural models of apo and DkTx-bound TRPV1 . Final ROSETTA models of apo TRPV1 ( A ) and of the TRPV1-DkTX complex ( B ) are shown color-coded according to the change in the calculated ProQM score relative to the initial structures deposited in the PDB ( entries 3J5P and 3J5Q , respectively ) . Blue represents an improved score in the ROSETTA model and/or a region without assigned backbone or side-chain coordinates in the initial structures . Red represents a worse score in the ROSETTA model . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01310 . 7554/eLife . 11273 . 014Video 1 . The top-ranking ROSETTA model of the TRPV1-DkTx complex , with DkTx in a CCW configuration . The K1 knot and linker is colored green and the K2 knot is colored blue . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 014 Further improvements of the ROSETTA models for apo and DkTx-bound TRPV1 focused on the configuration of the side-chains poorly resolved in the EM map ( approximately 80% of the side-chain atoms ) . An ensemble of 12 , 000 models , weighted by the ROSETTA energy function , were generated in each case , and the most representative among these were identified through a clustering analysis . The selected models for apo and DkTx-bound TRPV1 further improve the ROSETTA score without compromising the quality of the fit to the experimental cryo-EM maps ( Figure 2—figure supplement 2 ) . Relative to the structures deposited in the PDB , these optimized models also showed a generalized improvement in ProQM per-residue scores ( Ray et al . , 2010 ) ( Figure 2—figure supplement 3 ) , in addition to the improved MolProbity global scores mentioned above ( Tables 1 , 2 ) . Both models are publicly available upon request to the authors . The model of the TRPV1-DkTx complex reveals that K1 and K2 engage the outer pore of TRPV1 using loop 2 and loop 4 ( Figure 3A , B ) . These two loops straddle the interface between TRPV1 subunits , with loop 2 capping the S6 helix from one subunit , while loop 4 localizes near the N-terminus of the re-entrant pore helix from the adjacent subunit . Scanning mutagenesis previously identified four residues within the outer pore domain of TRPV1 , namely I599A at the top of S5 , F649A in the pore loop , and both A657P and F659A at the top of S6 , where mutations disrupt activation of the channel by DkTx ( Bohlen et al . , 2010 ) ; two additional mutants in this region ( V595A in S5 , and T650A in the pore loop ) also display diminished activation by the toxin , whereas Y631A , near the N-terminus of the pore helix , enhances the effect of DkTx . Our model indicates that among these residues , only F649 , A657 , T650 and Y631 could interact directly with the toxin , suggesting that the other mutations influence toxin activation of the channel through an indirect mechanism ( see below ) . Interestingly , the model also shows that both K1 and K2 drape over the top of S6 at the presumed interface with the surrounding membrane , and position several conserved hydrophobic residues ( W11 , F27 and I28 in K1 , and W53 , F67 and I68 in K2 ) where they would interact directly with lipids in the bilayer . 10 . 7554/eLife . 11273 . 015Figure 3 . Validation of the proposed structural model of the TRPV1-DkTx complex . Close-up of K1 ( A ) and K2 ( B ) in the top-ranking ROSETTA model of DkTx bound to TRPV1 . Side chains of TRPV1 residues where mutation disrupts DkTx-induced channel activation ( I599 , F649 , A657 and F659 ) ( Bohlen et al . , 2010 ) are highlighted in purple . ( C ) Representative time courses of TRPV1 activation at +120 mV by either DkTx ( left panel ) , K1K1 ( middle panel ) or K2K2 ( right panel ) measured from a train of voltage-ramps in the whole-cell configuration with the TRPV1 expressed in HEK cells . In the case of K1K1 , toxin was reapplied to verify that the observed decrease in current upon removal of the toxin was due to toxin dissociation from the channel rather than desensitization . Voltage ramps were from −120 to +140 mV over 1 s and were applied every 3 s from a holding potential of −90 mV . The colored horizontal lines denote the application of toxins . The dotted gray line denotes the zero-current level . Fractional dissociation 1 min after toxin removal was quantified across cells as the current amplitude 1 min after toxin removal divided by the steady-state current amplitude in the presence of the toxin . Fractional dissociation values ( mean ± sem ) were 0 . 8 ± 0 . 05 ( n = 8 ) for DkTx , 0 . 3 ± 0 . 04 ( n = 5 ) for K1K1 and 0 . 96 ± 0 . 03 for K2K2 ( n = 6 ) . ( D ) Mean normalized I-V relations obtained from voltage ramps at steady-state for control solution ( gray ) , toxins ( yellow , green or blue ) and saturating capsaicin ( 10 µM; black ) in the whole-cell configuration using transiently transfected HEK293 cells . The thin colored curves represent the mean and the thicker envelope the standard error ( n = 5–6 for each panel ) . ( E ) Primary sequence of K1 without linker ( K1-ΔL ) , K2 and gain-of-function chimeras between K1 and K2 . Residues of K1-ΔL and K2 are shown in green and cyan , respectively , and cysteines are shown in yellow . Loops between cysteine residues are labeled as L1 , L2 , L3 or L4 . CT denotes C-terminal region . ( F ) Concentration-dependence for activation of TRPV1 by K1 , K2 and chimeras in two-electrode voltage clamp recordings of oocytes expressing TRPV1 . Activation of the full-length TRPV1 channel by the toxin at different concentrations was measured in oocytes at a holding voltage of −60 mV . Toxin-induced currents ( ITx ) were normalized against the current activated by a saturating concentration of capsaicin ( Icap at 10 μM ) , in the same cell . Note that the data for K1 ( D1N ) and K1 ( K2L3 ) are obscured by the data for K1-ΔL . Error bars represent SEM ( n = 3 ) . ( G ) Close-up of the structure of K2 bound to TRPV1 with regions of the toxin colored with the same scheme as in ( F ) . Side chains of K603 and Y627 indicate the region where the pore turret ( residues 604–626 ) would likely reside if not deleted in the construct used for cryo-EM and in our model . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 015 Previous studies have shown that both K1 and K2 can activate TRPV1 , but that responses to K1 are much weaker than K2 ( Bae et al . , 2012; Bohlen et al . , 2010 ) . In the case of the K2 knot , the concentration-dependence for activation of TRPV1 saturates in the low micromolar range , and the extent of maximal activation is comparable to saturating concentrations of DkTx . In contrast , saturation of the concentration-response relation cannot be observed for K1 , and at the highest concentrations that can be tested , K1 produces between 6- and 12-fold weaker activation compared to a saturating concentration of K2 ( see Figure 3F ) . Because we cannot achieve higher concentrations of K1 in aqueous solution due to limited solubility , we investigated whether this knot has lower affinity or efficacy compared to K2 by taking advantage of the fact that bivalency increases the local concentration of each lobe , and therefore will produce higher occupancy of the channel compared to the separate lobes . For these experiments , we produced bivalent versions of K1 ( K1K1 ) and K2 ( K2K2 ) , and tested whether bivalency altered the large difference in the ability of the two lobes to activate TRPV1 . K1K1 should remain a weak activator of TRPV1 if K1 has lower efficacy than K2 , however , bivalency should diminish the differences between K1 and K2 if K1 has lower affinity . Indeed , we found that 1 µM K1K1 produced comparable activation to that produced by the same concentration of DkTx or K2K2 ( Figure 3C , D ) , suggesting that the efficacy of the two knots is comparable , but that the affinity of K1 is much lower than that of K2 . Although both DkTx and K2K2 exhibited slow dissociation , we observed considerably more rapid dissociation for K1K1 ( Figure 3C ) , confirming that K1 has lower binding affinity than K2 even when tested in its bivalent form . The observation that K1 has lower affinity than K2 provides a means to validate our structural model of the TRPV1-DkTx complex . Specifically , chimeras in which we replace either loop 2 and loop 4 in K1 by that in K2 ought to lead to a gain of function for activating the channel , resulting from an enhanced binding affinity; in contrast , substitutions in regions not involved in the channel-toxin interface should not alter the degree of channel activation . We generated such chimeras between K1 without the linker ( K1-ΔL ) and K2 ( Figure 3E ) , and tested their ability to activate full-length TRPV1 channels compared to a saturating concentration of capsaicin ( Figure 3F ) . The weak activation observed with K1-ΔL ( Figure 3F; green ) was not altered in constructs in which either Asn1 in the N-terminus or loop 3 of K2 were transferred into K1-ΔL [i . e . K1 ( D1N ) or K1 ( K2L3 ) ] ( Figure 3F ) . In contrast , chimeras in which loop 2 , loop 4 or the C-terminal part of K2 were transferred into K1-ΔL [i . e . K1 ( K2L2 ) , K1 ( K2L4 ) or K1 ( K2CT ) ] exhibited considerably more robust activation compared to K1-ΔL ( Figure 3F ) . The negative results observed with the N-terminal mutant and the loop 3 transfer , together with the gain of function observed with loop 2 and loop 4 transfer , support the notion that these two loops engage directly with the outer pore of TRPV1 , as observed in our model ( Figure 3G ) . The model also provides a possible explanation for the gain of function observed when the C-terminus of K2 is transferred to K1 . The location of the C-termini of K1 and K2 indicates that they are likely to interact with the pore turret ( residues 604–626 ) of TRPV1 ( Figure 3G ) , a region absent in our model as it was deleted from the TRPV1 construct used for cryo-EM ( Liao et al . , 2013 ) , but preserved in the full-length construct used for our functional studies . To investigate the interaction between the toxin and channel in more detail , we carried out an all-atom MD simulation of the complex embedded in a phospholipid bilayer ( Figure 4A ) . To enhance the exploration of diverse interaction patterns in a limited simulation time ( ~500 ns ) , we coupled the χ1 and χ2 torsion angles of all interfacial side-chains in the toxin and channel to a fictitious high-temperature bath , using an extended-Lagrangian approach ( Iannuzzi et al . , 2003 ) ( see Methods ) . To preclude the dissociation of the complex under this bias , the cryo-EM envelop was used as a three-dimensional restraint . A contact map between residues in DkTx and the outer pore of TRPV1 was then generated from the last 200 ns of simulation , so as to identify the most pronounced interactions ( Figure 4D ) . Two regions of the extracellular surface of the channel stand out as forming the most persistent side-chain contacts with either of the two lobes of DkTx , namely the N-terminus of the pore-helix , primarily via Y631 , and a stretch of the pore-loop and N-terminus of S6 , including N652 , D654 , F655 , K656 , A657 and V658; additional interactions are mediated by K535 and E536 , in S4 ( Figure 4D; Figure 4—figure supplement 1; Video 2 ) . The contacts with the pore and S6 helices are particularly worth noting because the A657P mutation effectively abolishes channel activation by DkTx , while Y631A enhances it ( Bohlen et al . , 2010 ) . Many of the contacts on the toxin are with residues that are equivalent in K1 and K2 , e . g . W11 and W53 , F27 and F67 , K14 and K56 and G12 and G54 , respectively , and a computational alanine-scanning ( Ala-scan ) analysis of the toxin-channel interface , based on the configurations explored during the MD simulation ( see Methods ) , indicate that these are all influential ( Figure 4—figure supplement 2 ) . However , there are also differences between K1 and K2 , which might underlie their different affinity for TRPV1 . The most interesting unique interactions for K2 involve S55 ( K13 in K1 ) , L65 ( M25 in K1 ) and R75 ( K35 in K1 ) . The persistence of these interactions is worth noting because these residues are located in either loop 2 ( S55 ) , loop 4 ( L65 ) or the C-terminus ( R75 ) , and transfer of these segments from K2 into K1 yields a pronounced increase in the binding affinity of isolated K1 ( Figure 3F ) . These residues are also predicted to have a significant stabilizing effect by the computational Ala-scan ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 11273 . 016Figure 4 . DkTx interactions with TRPV1 and the surrounding lipid bilayer from MD simulations . ( A ) Molecular system employed in the simulation of the TRPV1-DkTx complex . The molecular surface of the channel is shown in yellow , while the two bound DkTx molecules are shown in green ( K1 ) and cyan ( K2 ) . Electrolyte ions are shown as silver ( Na+ ) and tan ( Cl- ) spheres , while water molecules are indicated by red dots , and the POPC bilayer in orange . A comparable system was used to simulate the apo channel . ( B , C ) Residues of K1 ( B ) and K2 ( C ) that continue to interact with the membrane after DkTx recognizes TRPV1 . The side-chains in the toxin are color-coded according to the number of contacting atom pairs , Np , that were observed for each of these side-chains and any lipid molecule in the membrane ( see Methods ) . The values plotted are averages over the 200 ns accelerated MD simulation of the TRPV1-DkTx complex . ( D ) Analysis of the side-chain contacts between TRPV1 and DkTx , as observed in the accelerated MD simulation of the complex . The color-coded matrix reports on the contacts between specific side-chain pairs in the toxin and channel . The color-key reflects the number of contacting atom pairs in each case , Np , averaged over the simulated time ( see Methods ) . The bar charts report on the total value of Np for each residue in either the toxin ( top ) or the channel ( left ) , when any residue in the other protein is considered . The data for K1 and K2 are colored differently ( green and cyan , respectively ) in the bar graphs . ( E ) Contacts between DkTx residues and lipid molecules in the surrounding membrane . The bar plot reflects the number of contacting atom pairs between any of the lipid head-groups or tails , and each toxin residue . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01610 . 7554/eLife . 11273 . 017Figure 4—figure supplement 1 . Interactions between DkTx and TRPV1 from MD simulations . The interaction analysis described in Figure 4 is represented graphically for K1 and TRPV1 ( A–D ) , and K2 and TRPV1 ( E–H ) . See also Video 2 . Residues of TRPV1 ( A , C , E , G ) are color-coded according to the number of contacting atom pairs with one or more toxin residues ( color keys are shown between panels A/C and E/G ) . Conversely , residues of K1 ( D ) and K2 ( H ) are color-coded based on the number of contacting atom pairs with one or more channel residues . ( A , E ) TRPV1-bound K1 ( green ) and K2 ( cyan ) are viewed sideways , from the plane of the membrane . ( B , F ) TRPV1-bound K1 ( green ) and K2 ( cyan ) are viewed from the extracellular side . Subunits of TRPV1 are colored in light brown , purple , light yellow and light green . ( C , G ) TRPV1 residues in contact with K1 ( C ) or K2 ( G ) , viewed from the extracellular side . Some of the residues in the channel that form persistent interactions with the toxin are shown as transparent surfaces and labeled . ( D , E ) Surfaces of K1 and K2 that interact with TRPV1 are shown with the color key . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01710 . 7554/eLife . 11273 . 018Figure 4—figure supplement 2 . Computational alanine-scan of the DkTx-TRPV1 interface . The plot reports the estimated change in the association free energy of the toxin-channel complex upon alanine substitution of each of the residues in the toxin , calculated with ROSETTA ( see Methods ) . Positive values indicate a destabilizing effect . The estimates are averages over 200 , 000 snapshots of the complex extracted from a MD simulation trajectory in which the experimental cryo-EM map is used as a three-dimensional restraint . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 01810 . 7554/eLife . 11273 . 019Video 2 . Close up of the interface between TRPV1 and DkTx , color-coded according to the persistence of protein-protein contacts from MD simulations . See also Figure 4—Supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 019 Interestingly , the simulation of the TRPV1-DkTx complex also reveals that interactions between toxin and the membrane occur concurrently with those with the channel ( Figure 4B , C , E ) . Several hydrophobic residues in loop 2 and loop 4 of both K1 and K2 stand out as forming long-lived interactions with lipids , including W11 , F27 and I28 in K1 , and W53 , F67 and I68 in K2 . Both knots also showed some unique interactions with lipid head-groups , for example Y70 ( in K2 ) and H30 ( in K1 ) ( Figure 4B , C , E ) . In summary , therefore , the simulation indicates that K1 and K2 interact similarly with the outer pore of TRPV1 and the surrounding membrane , but also reveals a number of protein-protein and protein-lipid interactions that might explain the higher affinity of K2 for the TRPV1 channel . Motivated by the amphipathic character evident in the structure of DkTx , together with the placement of the toxin at the protein-lipid interface when bound to TRPV1 , we investigated the interaction of DkTx and each of its two lobes with lipid membranes . For these experiments , we took advantage of the presence of a single conserved and solvent accessible Trp residue in loop 2 of both K1 and K2 ( Figure 1C , D; W11 and W53 ) , and used Trp fluorescence to monitor the partitioning of the toxin from the aqueous solution into the membrane environment ( Gupta et al . , 2015; Ladokhin et al . , 2000; Milescu et al . , 2007 ) . When aqueous solutions of DkTx were excited at a wavelength of 280 nm , emission spectra with maxima at 353 nm were obtained ( Figure 5A , left ) , as expected for Trp residues residing in such an environment . In contrast , upon addition of large unilamellar vesicles ( LUVs ) , the fluorescence emission spectra for DkTx shifted to shorter wavelengths ( i . e . a blue shift ) , and the fluorescence intensity increased , suggesting that the toxin partitions into membranes where the environment of the Trp side-chains is more hydrophobic and their dynamics are more constrained ( Ladokhin et al . , 2000 ) . To quantify the extent of partitioning as a function of lipid concentration , we measured the relative fluorescence intensity in a blue-shifted region of the spectra ( e . g . 320 nm ) as a function of lipid concentration , and fit a partition function to the data to obtain a mole-fraction partition coefficient ( Kx ) of ( 2 . 3 ± 0 . 7 ) × 106 , indicating an energetically favorable interaction of DkTx with lipid membranes ( Figure 5A , right ) . This strong interaction of DkTx with membranes would facilitate binding of the toxin to the channel by increasing the local concentration of the toxin near the channel and by a reduction in the dimensionality of diffusion within the membrane ( Axelrod and Wang , 1994 ) . 10 . 7554/eLife . 11273 . 020Figure 5 . Interaction of DkTx and toxin constructs with lipid membranes measured using Trp fluorescence . ( A–E , left panel ) Trp emission spectra of DkTx ( A ) , K1 without linker ( B ) , K2 ( C ) , bivalent K1 ( K1K1 ) ( D ) and bivalent K2 ( K2K2 ) ( E ) in the absence ( black ) and presence ( blue , 1 . 6 mM for DkTx , 2 mM for K1-ΔL and K1K1 , and 3 mM for K2 and K2K2 ) of lipid vesicles ( 1:1 mix of POPC:POPG ) . ( A–E , right panel ) Relative fluorescence intensity at 320 nm ( F/F0 ) as a function of available lipid concentrations ( 60% of total lipid concentration ) . Smooth curves correspond to partition functions with Kx = ( 2 . 3 ± 0 . 7 ) × 106 and F/F0max = 2 . 65 ± 0 . 08 for DkTx , Kx = ( 3 . 9 ± 1 . 1 ) × 105 and F/F0max = 2 . 63 ± 0 . 07 for K1-ΔL , Kx = ( 2 . 0 ± 0 . 3 ) × 104 and F/F0max = 3 . 71 ± 0 . 40 for K2 , Kx = ( 2 . 9 ± 0 . 3 ) × 106 and F/F0max = 2 . 47 ± 0 . 03 for K1K1 and Kx = ( 8 . 1 ± 2 ) × 105 and F/F0max = 2 . 77 ± 0 . 55 for K2K2 . Error bars correspond to SEM ( n = 3 or 4 ) . Partition functions for K1-ΔL and K2 are shown as dotted lines for comparison in D and E , respectively . ( F ) Comparisons of mole-fraction partition coefficient ( Kx ) of toxins . 7G-DkTx denotes a construct of DkTx whose linker ( PYVPVTT ) is replaced with 7 Gly residues . Error bars correspond to SEM ( n = 3 or 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 02010 . 7554/eLife . 11273 . 021Figure 5—figure supplement 1 . Interaction of K1-K2 chimeras with lipid membranes measured using Trp fluorescence . ( A–E , left panel ) Trp emission spectra of K1 ( D1N ) ( A ) , K1 ( K2L2 ) ( B ) , K1 ( K2L3 ) ( C ) , K1 ( K2L4 ) ( D ) and K1 ( K2CT ) ( E ) in the absence ( black ) and presence ( blue , 2 mM ) of lipid vesicles ( 1:1 mix of POPC:POPG ) . ( A–D , right panel ) Relative fluorescence intensity at 320 nm ( F/F0 ) as a function of available lipid concentrations ( 60% of total lipid concentration ) . Smooth curves correspond to partition functions with Kx = ( 1 . 76 ± 0 . 06 ) × 106 and F/F0max = 3 . 79 ± 0 . 02 for K1 ( D1N ) , Kx = ( 4 . 77 ± 1 . 39 ) × 105 and F/F0max = 1 . 31 ± 0 . 02 for K1 ( K2L2 ) , Kx = ( 1 . 02 ± 0 . 04 ) × 104 and F/F0max = 3 . 69 ± 0 . 02 for K1 ( K2L3 ) , and Kx = ( 0 . 90 ± 0 . 04 ) × 106 and F/F0max = 2 . 59 ± 0 . 02 for K1 ( K2L4 ) . Error bar corresponds to SEM ( n = 3 or 4 ) . Smooth curve of K1-ΔL was shown as a dotted line for comparison throughout the figures . ( F ) Comparisons of mole-fraction partition coefficient ( Kx ) of toxins . Error bars correspond to SEM ( n = 3 or 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 021 We conducted similar experiments for K1 and K2 separately , and observed that both K1 and K2 exhibit blue shifts upon addition of LUVs , although interestingly the blue shift of K2 was considerably smaller compared to K1 ( Figure 5B , C ) . Fitting of a partition function to the data yielded Kx values of ( 3 . 9 ± 1 . 1 ) × 105 for K1 and ( 2 . 0 ± 0 . 2 ) × 104 for K2 , suggesting that the interaction of K1 with lipid membranes is stronger compared to K2 . Because the fluorescence changes for K2 were small and did not saturate even at higher lipid concentrations ( Figure 5C ) , we conducted a similar experiment using the bivalent constructs K1K1 and K2K2 , which we expected would increase the Kx values due to avidity ( Figure 5D , E ) . Indeed , both K1K1 ( [2 . 9 ± 0 . 3] × 106 ) and K2K2 ( [8 . 1 ± 2] × 104 ) showed increased Kx values compared to K1 and K2 , respectively , and K1K1 exhibited a higher Kx value when compared to K2K2 , confirming that partitioning of K1 into membranes is more favorable than K2 . These results are remarkable when considering that K2 has a higher affinity for TRPV1 compared to K1 , leading us to propose a model whereby partitioning of DkTx into membranes would be disproportionately mediated by K1 , leaving K2 available to engage TRPV1 and initiate formation of the toxin-channel complex ( see Discussion ) . We also examined chimeras of K1 and K2 to identify which regions determine their distinct membrane partitioning energies . Interestingly , transfer of the N-terminus and loops 2 and 4 from K2 into K1 produced either no change , or a moderate increase in Kx , even though the region transferred was from the weaker partitioning K2 knot ( Figure 5—figure supplement 1 ) . It is possible that the C-terminus of K2 is responsible for the weaker partitioning of that knot; however , transfer of that region into K1 greatly diminished the blue-shifts observed on addition of membrane vesicles ( Figure 5—figure supplement 1E ) , precluding an accurate determination of Kx . Although bivalency in DkTx clearly plays an important role in channel activation , it also increases the local concentration of the two lobes relative to each other and may promote interactions between the two . Indeed , the measured free energies of membrane partitioning ( ΔG° = −RT lnKx ) for DkTx ( ΔG° = −8 . 5 kcal mol−1 ) , K1K1 ( ΔG° = −8 . 7 kcal mol−1 ) and K2K2 ( ΔG° = −6 . 6 kcal mol−1 ) reveal a systematic energetic penalty for bivalency between 4 . 8 to 6 . 3 kcal mol−1 if we calculate the theoretical free energies for bivalent toxins ( DkTx ΔG° = −13 . 3 kcal mol−1; K1-K1 ΔG° = −15 kcal mol−1; K2-K2 ΔG° = −11 . 6 kcal mol−1 ) from that measured for the monovalent toxins ( K1 ΔG° = −7 . 5 kcal mol−1; K2 ΔG° = −5 . 8 kcal mol−1 ) assuming complete additivity of the free energies for partitioning of each lobe . One possible explanation for this energetic penalty is that the linker between the two knots constrains their orientations and prevents optimal membrane interactions by both lobes concurrently . To test this possibility we constructed a version of DkTx wherein the linker was replaced with a highly flexible linker comprised of 7 Gly residues ( 7G-DkTx ) , and observed a Kx value that is indistinguishable from wild-type DkTx ( [4 . 4 ± 0 . 03] × 106 , Figure 5F ) . This result seems to rule out the possibility that the linker restricts the dynamics of the two lobes significantly . Therefore , we deduce that the suboptimal membrane partitioning of DkTx probably owes to direct interactions between the two lobes , possibly mediated through the same amphipathic surfaces that interact with the membrane . Although we could not observe NOEs between K1 and K2 in the NMR spectra of DkTx in solution , it is conceivable that transient and non-specific hydrophobic interactions could go undetected , or that they are prevalent only when the toxin dynamics becomes restricted by the membrane surface . As noted previously ( Cao et al . , 2013 ) , the most striking features of the cryo-EM structure of TRPV1 bound to DkTx , relative to the unbound structure , are the dilation of the outer pore within the selectivity filter ( SF ) region , and the opening of the intracellular gate that is formed at the crossing between the S6 helices . Based on our improved structures of TRPV1 with and without DkTx , we set out to systematically analyze the key differences between these two states to explore the mechanism by which toxin binding promotes channel opening . This comparative analysis indicates that binding of DkTx promotes changes in the transmembrane architecture of the channel that pertain not only to the internal structure of each of the channel subunits , but also to their relative arrangement ( Figure 6 ) ; as a result , these seemingly cooperative changes open up the constrictions observed in the apo structure , both in the SF region and the intracellular gate . These structural changes appear to be effected through displacements in the pore helix ( P ) and S6 helices relative to the S1-S4 unit and the transmembrane S5 segment; individually , the S1-S5 and P-S6 units are largely unchanged ( RMSD ~ 0 . 7–0 . 8 Å ) ( Figure 6A–C ) . That the P and S6 helices become displaced relative to S1-S5 ( Figure 6D ) is consistent with the fact that their respective N-termini are the primary contacts for the toxin on the channel extracellular surface , as discussed above . The changes in P and S6 within a given subunit correlate very clearly with a pronounced rearrangement of the SF and pore loop in that same subunit ( Figure 6D ) , but also with changes in the intracellular side of the channel . The rationale for this remote effect is that the P and S6 helices are the main interfacial elements between adjacent subunits in the transmembrane domain of the channel ( Figure 6E ) ; therefore , displacements in these elements are propagated to the adjacent subunits , and compounded to the changes in their own internal structure . As a result , the S1-S4 units ( which as mentioned remain largely unchanged ) become noticeably displaced relative to each other , thus affecting the crossing angle of helices S5 and S6 , and hence the degree of opening of the intracellular gate ( Figure 6F ) . 10 . 7554/eLife . 11273 . 022Figure 6 . Conformational mechanism of activation of TRPV1 inferred from the cryo-EM structures of apo and DkTx-bound channel . ( A ) Overlay of the S1-S4 unit in the apo ( orange ) and toxin-bound ( marine ) structure; the root-mean-square difference ( RMSD ) of the C-α trace is only ~0 . 7 Å , indicating the internal structure of this unit is not altered upon toxin binding . ( B ) Overlay of the S1-S4 unit , plus the transmembrane portion of the neighboring S5 , which belongs to the adjacent protomer ( hence referred to as S5’ ) ; the RMSD is again ~0 . 7 Å . ( C ) Overlay of the pore helix ( P' ) and the transmembrane portion of S6'; the RMSD is also ~0 . 7 Å . ( D ) Overlay of the S1-S4 unit , plus S5' , the pore helix P' and S6’ . Only the S1-S4 unit and S5' ( in gray ) are used for fitting , to highlight the relative displacement of S6' and the P' helix , which , as shown in panel ( C ) , move as a largely rigid unit . ( E ) Overlay of the pore helix P’ and the transmembrane portion of S6 across the protomer interface , i . e . S6"; the RSMD is ~ 0 . 6 Å . ( F ) Overlay of two adjacent channel subunits , using the same fit as in panel ( D ) . Owing to the quasi-rigid structure of the P-S6 interface across subunits ( see panel E ) , the displacements induced upon toxin binding on the pore and S6 helices result in a change in the relative orientation of the S1-S4 units , which we propose leads to a change in the crossing-angle of S5 and S6 on the intracellular side , and thus the opening of the lower gate . ( G , H ) Close-up view of the changes induced upon toxin ( yellow ) binding on a hydrophobic cluster formed by side-chains from S5 , the pore helix , the selectivity filter ( SF ) and S6 . Mutation of the side-chains colored in gray has been shown to diminish toxin-induced TRPV1 activation ( see text ) . ( I ) Evaluation of the compactness of the hydrophobic cluster shown in ( G , H ) during MD simulations of apo and toxin-bound TRPV1 , in terms of the number of contacting atom pairs of a given side-chain with all others . Fewer contacts imply a less compact arrangement with greater exposure to the solvent , and potentially , an increased heat-capacity of the channel . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 022 Interestingly , the structural changes in the main-chain of the channel that are observed upon DkTx binding correlate with the disruption of a cluster of hydrophobic interactions behind the SF , at the interface between S5 , the pore helix and S6 , in close proximity to the contact region with the toxin ( Figure 6G , H; Video 3 ) . This network involves , among others , residues I599 , F659 and V595 , F649 , as well as T650 . Only the latter two residues are nearby DkTx in the toxin-bound structure; however , mutation of any of these five residues diminishes toxin-induced opening of the channel , particularly in the case of I599A , F649A and F659A ( Bohlen et al . , 2010 ) . To evaluate whether the distinct arrangement of this hydrophobic cluster is a significant feature of the apo and toxin-bound structures , despite their limited resolution , a MD simulation of apo TRPV1 was also carried out and contrasted with that discussed above for the toxin-bound structure ( Figure 4 ) , using an analogous methodology . When the degree of compactness of this hydrophobic network is quantified in terms of the number and persistence of pairwise side-chain contacts , it is apparent that there is a generalized reduction of these contacts in DkTx-bound TRPV1 , relative to apo TRPV1 ( Figure 6I ) . In DkTx , two of the residues nearest to this hydrophobic cluster are M25 in K1 and L65 in K2 , a relatively conserved position on the two knots that makes frequent contact with channel residues in the MD simulations ( Figure 4D and Figure 6G ) . Accordingly , both M25 and L65 are predicted to play energetically important roles based on our computational alanine-scan of the toxin-channel interface ( Figure 4—figure supplement 2 ) . Although we have not yet tested other toxin residues that make direct contact with TRPV1 ( such as W11/W53 and K14/K56 ) , we mutated L65 in K2 to Ala and measured the concentration-dependence for activation of TRPV1 relative to wild-type K2 . The L65A mutant dramatically reduced activation of K2 and increased the rate of dissociation ( Figure 7A , B ) , confirming this residue mediates important interactions with the outer pore of TRPV1 . 10 . 7554/eLife . 11273 . 023Video 3 . Structural changes in TRPV1 resulting from DkTx binding , and close-up view of a cluster of buried hydrophobic residues that become exposed upon channel opening . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 02310 . 7554/eLife . 11273 . 024Figure 7 . The L65A mutation in K2 perturbs activation of the TRPV1 channel . ( A ) Representative TRPV1 current time-course in HEK293 cells depicting activation of the channel by L65A and WT K2 toxins in the whole-cell configuration at +60 and −60 mV . Currents were measured every 3 s by applying voltage-ramps of 1 s duration from −120 to +140 mV starting at a holding potential of −90 mV . Currents were recorded under constant perfusion with control solution or with the test solutions containing either NMDG+ instead of Na+ as permeant ion ( gray ) , or the K2 toxins together with 130 mM NaCl ( black and blue ) , as indicated by the continuous horizontal lines . The dotted red line denotes the zero-current level . Removing external Na+ leads to constitutive activation of TRPV1 to a similar extent as a saturating concentration of DkTx or K2 ( Jara-Oseguera , 2016 ) , and therefore provides a convenient means of normalizing response to the toxins . ( B ) Normalized concentration-response relation for TRPV1 activation at −60 mV by WT K2 toxin ( blue ) or the L65A mutant ( black ) obtained in the whole-cell configuration using HEK293 cells . Currents at −60 mV were measured from voltage-ramps as in ( A ) , and normalized to the current value at +60 mV measured before application of the toxin , in the presence of an external solution containing 130 mM NMDGCl instead of NaCl ( see ( A ) , gray ) . Initial currents measured in control solution at +60 and −60 mV were subtracted from the data . A single toxin variant ( WT or L65A ) was evaluated per experiment to construct dose-response curves , and one or more toxin concentrations were tested per cell . Data are shown as mean ± sem , with n = 3–14 for each data point . The continuous curves are fits to the Hill equation with parameters: WT , KD = 8 . 5 µM , n = 4 . 0; L65A , KD > 0 . 2 mM , n = 1 . 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 024 The goal of this study was to determine the structure of DkTx and explore its interaction with the TRPV1 channel using the recently reported electron density maps of TRPV1 in complex with the toxin . The solution NMR structure of DkTx indicates that the toxin is composed of two well-ordered ICK lobes , connected by a 7 residue linker ( Figure 1 ) , as previously surmised ( Cao et al . , 2013 ) . Our structures show that K1 and K2 lobes have similar amphipathic surfaces , which are formed by clusters of solvent-accessible hydrophobic residues ( aromatic and aliphatic residues in loop 2 and loop 4 ) and surrounding basic and acidic residues ( Figure 1 ) . In an improved model of the toxin docked onto the TRPV1 channel , these amphipathic surfaces on K1 and K2 can be seen to intimately interact with the TRPV1 channel , as well as with lipids in the surrounding membrane ( Figures 3 , 4 ) . Indeed , conserved aromatic residues in DkTx ( W11 and F27 in K1 , W53 and F67 in K2 ) reach into the void formed by S4 , S6 and pore-helix of TRPV1 , which lipid molecules would fill in the absence of the toxin ( Figures 3 , 4 ) . Our results demonstrate that DkTx interacts favorably with membranes in the absence of the channel such that the conserved Trp residues reside within the membrane environment and exhibit blue-shifted fluorescent emission spectra ( Figure 5 ) . Taken together , our results provide direct structural evidence for a model wherein DkTx interacts with TRPV1 within the lipid membrane . Although the bivalency of DkTx clearly helps to prolong the lifetime of the toxin-channel complex ( Bohlen et al . , 2010 ) , each lobe has relatively small protein-protein interfaces with TRPV1 ( 655 Å2 for K1 and 556 Å2 for K2 ) , suggesting that the interaction of the toxin with the surrounding membrane is important for stabilizing the toxin-channel complex . It is interesting to compare the present structural picture of DkTx binding to TRPV1 with the mode of PcTx1 binding to ASIC , as well as that described for tarantula toxins binding to voltage-sensing domains in Kv channels . DkTx , PcTx1 and voltage-sensor toxins like hanatoxin and GxTx-1E adopt similar folds , and display amphipathic surfaces that enable interactions with lipid membranes ( Figure 8 ) ( Gupta et al . , 2015 ) . In DkTx , residues in loops 2 and 4 engage the outer pore of TRPV1 , while interacting with lipids in the surrounding membrane ( Figures 3 , 4 , 8A , B ) , and contain mostly hydrophobic residues ( compare Figure 8A , B with Figure 1C , D where the toxins are oriented identically ) . In PcTx1 , residues in loops 1 and 4 form a clamp-like structure for the toxin to bind to helix-5 within the extracellular thumb domain of the ASIC channel ( Figure 8D ) ( Baconguis and Gouaux , 2012; Dawson et al . , 2012 ) . Interestingly , the voltage-sensor toxin GxTx-1E also employs residues in loops 1 and 4 to bind to the S3b helix within the voltage-sensing domain of Kv channels , and these form a surface that resembles the one used by PcTx1 to bind to ASIC ( Figure 8E ) ( Gupta et al . , 2015 ) . If we orient the two lobes of DkTx with channel-binding surfaces directed down , as if viewing the toxin-channel complex from a side view in the membrane , and oriented both PcTx1 and GxTx-1E by backbone superposition ( as is done in Figure 8A , B , D , E ) , the channel-binding surfaces for PcTx1 and GxTx-1E would both be displaced laterally by roughly 90° ( compare the location of residues highlighted in red in Figure 8A , B for K1 and K2 with Figure 8D for PcTx1 and Figure 8E for GxTx-1E ) . Interestingly , the amphipathic surface of GxTx-1E is similarly displaced relative to the amphipathic surfaces of both lobes of DkTx , which makes sense because those surfaces for both toxins are important for interacting with membranes and their target channels . Overall , this comparison suggests that the common fold of all these toxins provides at least two distinct surfaces for engaging with ion channel proteins , while maintaining favorable interactions with the lipid membrane . 10 . 7554/eLife . 11273 . 025Figure 8 . Comparison of active surfaces of different classes of tarantula toxins . ( A , B ) Active surfaces of K1 ( A ) and K2 ( B ) . Side-chains in DkTx are colored based on the probability of finding TRPV1 channel residues within 6 Å in a simulation of the complex ( except cysteine residues colored in yellow ) . The value specified in the scale bar reflects the number residues of the channel with one or more atoms within 6 Å of the toxin residue , on average . Orientation of K1 and K2 are the same as in Figure 1A–D , with loops 2 and 4 pointing down . ( C ) Sequences of tarantula toxins . Conserved cysteine residues are shown in yellow and residues are colored according to schemes in A , B , D or E as appropriate . ( D ) Structure of PcTx1 bound to helix-5 of the ASIC1a channel ( pastel yellow ) oriented the same as K1 and K2 in A and B by superimposing backbones ( PDB entry 4FZ0 ) . Backbone of the toxin is colored in pastel green , and side-chains are colored by changes in solvent-accessible surface area between the structure of PcTx1 when the toxin is bound to ASIC1a and the structure of toxin without channel ( Gupta et al . , 2015 ) ( ΔSASA in Å2 ) . ( E ) Solution structure of GxTx1E oriented the same as K1 and K2 in A and B by superimposing backbones ( PDB entry 2WH9 ) . Side-chains are colored according to perturbation in free energy of binding ( ΔΔG° in kcal mol−1 ) ( Gupta et al . , 2015 ) . ( F ) Surface representation of the PcTx1 structure where hydrophobic residues are colored green , basic residues blue and acidic residues red . Orientation is the same as in D . ( G ) Surface representation of the GxTx-1E structure where hydrophobic residues are colored green , basic residues blue and acidic residues red . Orientation is the same as in E . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 025 Our results also provide important insight into the bivalent structure of DkTx . Previous studies have demonstrated that the K2 lobe is a more effective activator of TRPV1 when compared to K1 ( see also Figure 3F ) , and that tethering of the two lobes of DkTx dramatically enhances the lifetime of the toxin-channel complex ( Bae et al . , 2012; Bohlen et al . , 2010 ) , similar to the avidity effect of a bivalent antibody . We find that the two lobes of DkTx also have very different energetics for partitioning into lipid membranes , with the K1 lobe displaying considerably stronger interactions when compared to K2 , a trend that can be recapitulated in the bivalent K1K1 construct when compared to the bivalent K2K2 construct ( Figure 5 ) . Using bivalent K1K1 and K2K2 constructs , we also found that differences in binding affinity can explain why K2 is a much better activator of TRPV1 compared to K1 ( Figure 3D ) . Thus , bivalency not only increases the lifetime of the toxin-channel complex , but also enables the two lobes to be separately tuned for optimal membrane partitioning or affinity for binding to the channel . This feature of DkTx would suggest that the toxin preferentially uses the K1 lobe to interact with membranes and thus raise the local concentration of K2 at the membrane surface , thereby promoting the initial formation of the toxin-channel complex using K2 within the interfacial region of the membrane ( Figure 9 ) . It will be fascinating to further explore the structural and biophysical basis of these mechanisms of recognition , to better understand how protein-protein interactions are modulated by the membrane environment . 10 . 7554/eLife . 11273 . 026Figure 9 . Proposed mechanism of DkTx partitioning into membranes and binding to TRPV1 . Partitioning of DkTx into the lipid membrane ( gray ) before recognition of TRPV1 is preferentially driven by the K1 lobe ( green ) , whereas K2 ( cyan ) has a higher affinity for binding to the channel . After binding to TRPV1 , both K1 and K2 retain a significant number of interactions with the lipid membrane , presumably so as to stabilize the complex , which features a relatively small protein-protein interface . DOI: http://dx . doi . org/10 . 7554/eLife . 11273 . 026 In docking of DkTx into the cryo-EM maps of TRPV1 and optimizing the structures of both the toxin-bound and apo channels , we identified an interesting hydrophobic cluster behind the SF of TRPV1 that undergoes a noticeable conformational change . This cluster appears to be a determining factor in the energetics of channel opening , as several mutations in this cluster diminish toxin activation of the channel even though most do not directly contact the toxin ( Bohlen et al . , 2010 ) . It is tantalizing to speculate that the disruption of this cluster , which implies an increased solvent exposure of more than a dozen hydrophobic side-chains across four channel subunits , leads not only to the structural changes required for ion permeation , but also to the kind of increase in the heat capacity of the protein that has been postulated to explain temperature sensing in TRP channels , including TRPV1 ( Clapham and Miller , 2011 ) . Notably , mutations at positions T633 , F640 and Y653 within the hydrophobic cluster have been shown to alter temperature sensitivity of the TRPV1 channel without affecting activation of the channel by capsaicin ( Grandl et al . , 2010; Myers et al . , 2008; Ryu et al . , 2007 ) . In addition , it is interesting that an external Na+ binding site in the outer pore of TRPV1 has been shown to tightly regulate temperature-sensor activation , and that DkTx renders TRPV1 temperature-insensitive over a wide temperature range ( Jara-Oseguera , 2016 ) . The hypothesis that this rearranging hydrophobic cluster is involved in temperature sensing will require careful experimental examination , and could begin to provide a unifying framework to rationalize how very different stimuli cooperate to regulate TRP channel activation . K1 , K2 and DkTx were prepared as previously described ( Bae et al . , 2012 ) . The sequence of bivalent K1 ( K1K1 ) is: GDCAKEGEVCSWGKKCCDLDNFYCPMEFIPHCKKYKPYVPVTTDCAKEGEVCSWGKKCCDLDNFYCPMEFIPHCKKYK . The toxin was prepared using a protocol similar to DkTx ( Bae et al . , 2012 ) , where K1K1 were produced in E . coli as a fusion protein with ketosteroid isomerase ( KSI ) . K1K1 was cleaved from KSI with hydroxylamine , refolded and purified using reversed-phase HPLC . The sequence of bivalent K2 ( K2K2 ) is: NCAKEGEVCGWGSKCCHGLDCPLAFIPYCEKYRPYVPVTTNCAKEGEVCGWGSKCCHGLDCPLAFIPYCEK . The toxin was prepared by cloning synthetic K2K2 DNA into the pET28a vector with an additional Met added at the N-terminus of the K2K2 sequence . The toxin was produced in E . coli as a 6 Histidine tagged form , refolded and purified using Ni-affinity chromatography . The N-terminal fusion protein was removed by CNBr cleavage and the toxin purified using reversed-phase HPLC . For experiments shown in Figure 5C , K2 was prepared using an identical protocol to that for K2K2 , but an additional Met was inserted before the N-terminus of N41 in the K2K2 construct such that CNBr cleavage would yield K2 . 15N labeled DkTx was prepared by producing the toxin in M9 minimal media where 15NH4Cl was included as a nitrogen source , and otherwise prepared as previously described ( Bae et al . , 2012 ) . K1 and K2 were dissolved at concentrations of 1 mM in 90% H2O/10% D2O ( pH 4 . 0 ) containing trimethylsilyl propionate ( TSP ) as an internal standard , and double-quantum-filtered correlation spectroscopy ( DQF-COSY ) , total correlation spectroscopy ( TOCSY ) and nuclear Overhauser effect spectroscopy ( NOESY ) spectra recorded at 298 K or 310 K in a Bruker 600 MHz spectrometer . Mixing times of TOCSY and NOESY experiments were 80 ms and 250 ms , respectively . 15N labeled DkTx was dissolved at a concentration of 0 . 4 mM in 10 mM sodium phosphate buffer ( pH 4 ) containing 10% D2O and TSP as an internal standard , and 1H-15N NOESY-heteronuclear single quantum coherence ( HSQC ) and 1H-15N TOCSY-HSQC spectra recorded at 298 K in a 900 MHz Bruker DRX900 spectrometer equipped with cryogenic probe . All spectra were processed with NMRPipe ( Delaglio et al . , 1995 ) , and analyzed and assigned with NMRview ( Kirby et al . , 2004 ) . J-coupling constants were estimated from DQF-COSY spectra ( Kim and Prestegard , 1989 ) and imposed as dihedral angle restraints for structure calculation based on the following rules: for 3JHNHα values of <5 . 5 Hz , the phi angle was constrained in the range of −65 ± 25°; For 3JHNHα values of >8 . 0 Hz , the phi angle was constrained in the range of −120 ± 40° . Interproton distance restraints were obtained from unambiguous NOE peaks that were manually assigned using NMRview ( Kirby et al . , 2004 ) . The initial structure of the toxins were generated by a simulated annealing protocol in torsion angle space using Cyana2 . 1 ( Guntert et al . , 1997 ) by imposing interproton distance and dihedral angle restraints . The structure was further refined using Xplor-NIH 2 . 37 ( Bermejo et al . , 2012; Schwieters et al . , 2006 ) , using interproton distances , dihedral angle restraints , and disulfide bond restraints based on sequence homology with other ICK toxins were imposed along with a multidimensional torsion angle potential of mean force ( Bermejo et al . , 2012 ) , a gyration volume term to enforce proper packing ( Schwieters and Clore , 2008 ) and standard covalent and nonbonded energy terms . The quality of each of the 20 ensemble structures were analyzed by Protein Structure Validation Software suite ( http://psvs-1_5-dev . nesg . org ) . Analysis of Ramachandran plots with Procheck gave the following results: for K1 , most favored region ( 76 . 8% ) , additionally allowed regions ( 22 . 2% ) , generously allowed regions ( 0 . 8% ) and disallowed regions ( 0 . 2% ) ; for K2 , favored region ( 58 . 2% ) , additionally allowed regions ( 31 . 4% ) , generously allowed regions ( 6 . 4% ) and disallowed regions ( 4 . 1% ) . Xplor-NIH was used to dock the NMR structures of K1 and K2 into a cryo-EM map for TRPV1 bound to DkTx/RTx with imposed twofold symmetry ( kindly provided Dr . Yifan Cheng and colleagues ) . Energy-minimized structures of K1 and K2 were placed in the EM map in a clockwise ( CW ) or counter-clockwise ( CCW ) configuration , and combined with a truncated model of the TRPV1 structure comprising residues 518–550 and 585–645 from one chain and 644–669 from the adjacent chain ( from PDB entry 3J5Q ) . Electron density map fitting was achieved using the probDistPot energy term ( Gong et al . , 2015 ) with a force constant of 100 kcal/mol throughout . Energy-minimized structures of the truncated complex were determined by imposing electron density map along with interproton distance restraints of K1 and K2 from solution NMR experiments . In addition to these experimental terms , the multidimensional torsion angle potential of mean force , empirical backbone hydrogen bonding potential of mean force ( Grishaev and Bax , 2004 ) and standard covalent and nonbonded terms were employed . 20 structures were calculated and the 10 lowest energy models were then used for further analysis . To prepare the structures of apo TRPV1 and the TRPV1-DxTx complex for refinement , we first modeled the missing loops , side-chains and termini into the existing structures ( PDB entries 3J5P and 3J5Q ) , using MODELLER version 9 . 10 ( Eswar et al . , 2006 ) ; for the TRPV1-DxTx complex , the truncated Xplor-NIH models ( CCW and CW ) were merged with the published channel structure . The fit-to-density protocol ( DiMaio et al . , 2009 ) of ROSETTA version 2014wk05 ( Leaver-Fay et al . , 2011 ) was then used to refine the three models ( apo and DxTx-bound , CW or CCW ) . In a first stage , 100 models were generated for each system following the ‘relax’ application in 5 cycles , using a high-resolution energy function for proteins in the membrane environment ( Barth et al . , 2007; Yarov-Yarovoy et al . , 2006 ) in addition to the experimental cryo-EM maps . The fit-to-density score was determined with a 9-residue sliding-window , and was added to the total ROSETTA score with a weight of 0 . 2 . The symmetry of the cryo-EM maps was explicitly imposed on all the models; a fourfold symmetric map was used for apo TRPV1 , and a twofold symmetric map for the DkTx-TRPV1 complex . Each of these 100 models required approximately 280 computer-hours on an Intel Xeon 2 . 4 GHz core-processor for the apo TRPV1 channel and 800 hr for the complex . The models with the best ROSETTA score for apo TRPV1 and TRPV1 with DxTx in a CCW configuration were then used as the seed for a second refinement stage . In this stage , positional restraints were imposed on the backbone and on well-resolved side-chains , and the configuration of all remaining side-chains was refined through extensive sampling and scoring . In particular , we ranked the degree of confidence in the position of each side-chain atom by evaluating a score equal to −maρ ( xa , ya , za ) , where ma is the mass of the atom and ρ ( xa , ya , za ) the normalized density signal at the hypothetical atom position ( Wu et al . , 2013 ) . Atoms for which the calculated score was in the top 20% were considered to be well resolved and were not modified further . To generate a plausible configuration for the remaining side-chain atoms , we generated 12 , 000 new models , using the ROSETTA energy function only ( i . e . without a fit-to-map restraint ) . On average , each new model of apo TRPV1 and TRPV1-DkTk required approximately 15 and 25 core-minutes , respectively . To identify the most representative model for each of these ensembles , we used the clustering algorithm of Daura and colleagues as implemented in GROMACS version 4 . 6 . 5 ( Daura et al . , 1999; Hess et al . , 2008 ) , using a similarity cut-off of 1 Å for apo TRPV1 and 0 . 08 Å for the toxin-channel complex . The two selected models correspond to the central structure of the most populated cluster obtained in each case ( ~1 , 100 and ~7 , 900 structures , respectively ) . The improved structures of apo and DkTx-bound TRPV1 were embedded in a cubic simulation box of side length of 145 Å containing a bilayer of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) lipids , surrounded by water . The simulation systems were prepared using GRIFFIN ( Staritzbichler et al . , 2011 ) . The set of protonation states in the channel and toxin was selected on the basis of a Monte-Carlo/Poisson-equation simulation , as described previously ( Eicher et al . , 2014 ) , for neutral pH . Energetically favorable positions for buried water molecules within the protein were identified with DOWSER ( Zhang and Hermans , 1996 ) . Na+ and Cl− ions were introduced to achieve electroneutrality and a 100 mM electrolyte concentration . Each of the simulation systems amounts to ~315 , 000 atoms in total , including over 500 lipid molecules and ~69 , 000 water molecules . The simulations were carried out using the CHARMM36 force-field for proteins and lipids ( Best et al . , 2012; Klauda et al . , 2010 ) as implemented in NAMD version 2 . 9 ( Phillips et al . , 2005a ) , at constant pressure and temperature , and with periodic boundary conditions . The equations of motion were integrated with a time-step of 2 fs . The pressure ( 1 atm ) was maintained constant with a Nose-Hoover Langevin-piston barostat ( Feller et al . , 1995 ) , allowing variations in the volume of the simulation cell but keeping a constant ratio in the membrane-plane dimensions . The temperature ( 298 K ) was maintained with a Langevin thermostat . Electrostatic interactions were calculated using Particle-Mesh Ewald ( PME ) ( Darden et al . , 1993 ) with a real space cutoff of 12 Å . The same cut-off was used for truncating van-der-Waals interactions , modeled with a shifted Lennard-Jones potential . To equilibrate the simulation systems , we used conventional MD simulations with gradually weaker positional restraints applied to the protein , the toxin and buried water molecules over 12 ns . For the toxin-channel complex , a subsequent MD simulation was carried out for 400 ns using the twofold symmetric cryo-EM map of the complex as a three-dimensional restraint , via the MDFF module in NAMD ( Trabuco et al . , 2009 ) ( with 0 . 3 kcal/mol as the scaling factor ) . During the last 200 ns of this simulation , we coupled the χ1 and χ2 torsions of all side-chains at the toxin-channel interface to a fictitious temperature of 3 , 000 K , using an extended-Lagrangian approach ( Iannuzzi et al . , 2003 ) , so as to accelerate the configurational sampling of that interface . The set of accelerated side-chains was determined by analyzing the first 200 ns of simulation using a simple definition of toxin-channel contacts based on a distance cut-off of 3 . 5 Å; this set comprises all the residues specified in Figure 4 . Finally , snapshots from the last 200 ns of simulation were analyzed to identify the most persistent contacts between toxin , channel and membrane . A similar scheme was followed for the apo channel , comprising an MDFF simulation of 200 ns ( using the corresponding fourfold symmetric cryo-EM map ) , with enhanced side-chain sampling in the last 100 ns , which were considered for analysis . To quantify the significance of a hypothetical contact between a given side-chain in the channel or toxin ( group of atoms A ) and one or more side-chains in the other protein , or the head-group and/or tail regions of the lipid membrane ( group of atoms B ) , we calculated the non-normalized distribution function of the distance r between each of the atoms in group A and each of the atoms in group B ( only non-hydrogen atoms were considered ) . This collective distribution function g ( r ) was then integrated over the first shell of interaction , which we assumed to be 6 Å: Np=∫064πr2g ( r ) dr The resulting value , denoted by Np , therefore represents the number of contacting atom pairs in groups A and B; note this value scales with the size of the groups , but we reasoned that so does the strength of their interaction . All analyses were performed with VMD version 1 . 9 . 1 ( Humphrey et al . , 1996 ) . The computational alanine-scanning of the residues at the toxin-channel interface was carried out with ROSETTA version 2014wk05 ( Kortemme et al . , 2004; Leaver-Fay et al . , 2011 ) . Specifically , the algorithm was used to independently identify the toxin residues at the interface with the channel , to replace these residues individually with alanine , and to estimate the effect of this mutation on the binding free energy of the complex . Positive values of this estimate imply that the alanine mutation is predicted to destabilize the complex , while negative values imply a stabilizing effect . The scan was carried for 200 , 000 different input configurations , extracted from the last 200 ns of the MD simulation of the complex; the results for each toxin residue were then averaged . After each alanine substitution , interfacial residues within 10 Å were re-optimized using the ROSETTA repacking method . Large unilamellar vesicles ( LUVs ) were prepared by extruding lipid suspensions made of a 1:1 mix of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) and 1-palmitoyl-2-oleoyl-sn-glycero-3-[phospho-rac- ( 1-glycerol ) ] ( POPG ) through 100 nm polycarbonate film . Toxins ( 2 µM or 5 µM in 10 mM HEPES , 1 mM EDTA , pH 7 . 0 buffer ) in cuvette were excited with 280 nm wavelength and emission spectra were measured from 300 to 450 nm in the presence or absence of LUV using a SPEC FluoroMax . Scattering of light from lipid vesicles were corrected ( Gupta et al . , 2015; Ladokhin et al . , 2000; Milescu et al . , 2007 ) . Smooth curves were obtained by fitting the following partition function , F/F0 ( L ) = 1+ ( F/F0max−1 ) Kx[L]/ ( [W]+Kx[L] ) , to the data points , where F/F0 ( L ) is relative fluorescence intensity of 320 nm at a given lipid concentration , F/F0max is fluorescence intensity when partitioning is saturated , [L] is molar concentration of accessible lipid ( 60% of total lipid , corresponding to the outer leaflet ) , [W] is molar concentration of water ( 55 . 3 M ) , and Kx is mole-fraction partition coefficient . Xenopus laevis oocytes were surgically removed and gently shaken for 60 min in a solution of 82 . 5 mM NaCl , 2 . 5 mM KCl , 1 mM MgCl2 , 5 mM HEPES and 2 mg/mL collagenase . A rat TRPV1 construct ( generously provided by D . Julius , UCSF ) was cloned into the pGEM-HE vector , and used to generate cRNA . The cRNA was then injected into oocytes , which were then incubated for 1–3 days at 17°C in ND-96 solution ( 96 mM NaCl , 2 mM KCl , 1 . 8 mM CaCl2 , 5 mM HEPES , 1 mM MgCl2 and 50 μg/mL gentamycin , titrated to pH 7 . 6 with NaOH ) . TRPV1 activity was recorded under voltage clamp using a two-electrode voltage clamp ( OC-725C; Warner Instruments ) in a 150-μL recording chamber . The recorded data were filtered at 1 kHz and digitized at 5 kHz using a digidata analog/digital converter and pClamp software ( Molecular Devices ) . Microelectrode resistances were 0 . 1–1 MΩ when filled with 3 M KCl . The external recording solution contained 115 mM NaCl , 2 . 5 mM KCl , 1 . 5 mM MgCl2 and 10 mM HEPES , titrated to pH 7 . 4 with NaOH . All experiments were performed at room temperature ( ~22°C ) . HEK293 cells were transiently transfected with rTRPV1 and Green Fluoresence Protein ( pGreen-Lantern , Invitrogen ) cDNAs using FuGENE6 ( Roche ) transfection reagent following manufacturer’s instructions , and used for recording 12–24 hr after transfection . Standard whole-cell patch clamp recordings at room temperature ( 22–24°C ) were performed . Data was acquired with an Axopatch 200B amplifier ( Axon Instruments ) , filtered with an 8-pole low-pass Bessel filter ( model 900 , Frequency Devices ) and digitized with a Digidata 1322A interphase and pClamp10 software ( Axon Instruments ) . All data was analyzed using Igor Pro 6 . 34A ( Wavemetrics Inc . ) . Pipettes were pulled from borosilicate glass and heat-polished to final resistances between 2–4 MΩ . 80–95% series resistance ( Rs ) compensation was used . An agar bridge ( 1 M KCl , 3% agarose ) was used to connect the recording chamber with the ground electrode . A holding potential of –90 mV was used in all experiments . Data were acquired at 10 kHz and low-pass filtered at 2 kHz . For the voltage-ramps , voltage was stepped down from the holding to −120mV for 50 ms , then ramped up to +140 mV in 1s and returned to −90 mV for 50 ms . A ramp was applied every 3 s . Recordings were done using isometric solutions consisting of ( in mM ) : 130 NaCl , 10 HEPES , 10 EGTA , pH 7 . 4 . Solutions were applied using a gravity-fed rapid solution exchange system ( RSC-200 , BioLogic ) . Cells were lifted from the coverslip and placed in front of glass capillaries perfused with the different solutions . Fused silica tubing ( 250 µM internal diameter , Polymicro Technologies ) was used to deliver toxin . Fresh toxin solutions were prepared every day . Dose-response curves were fit to the Hill equation: ITxINMDG=Imin+Imax−Imin1+KD[Tx]s , where ITx is the current measured at the various toxin concentrations at −60 mV , INMDG is the current measured in the presence of 130 mM NMDGCl at +60 mV , Imin is the minimal current remaining after subtraction of the currents in control conditions ( i . e . 130 mM NaCl before application of the toxins ) , Imax is the maximal activation achieved by the toxins relative to 130 mM NMDGCl , KD is the apparent dissociation constant , [Tx] is the molar concentration of the toxin and s is the Hill coefficient .	Humans and other mammals sense heat using a protein called the transient receptor potential vanilloid ( TRPV1 ) channel . This protein is found in the membranes of a particular type of nerve cell , and it forms a pore that allows certain ions to pass through the membrane . Along with sensing heat , TRPV1 can also be activated by a toxin called double-knot toxin – which is found in spider venom – and by capsaicin , the active ingredient in chilli peppers . One way to investigate how a protein works is to study its three-dimensional ( 3D ) structure . Here , Bae , Anselmi et al . use a technique called nuclear magnetic resonance ( NMR ) spectroscopy to produce a detailed model of the 3D structure of double-knot toxin . This model is then combined with 3D maps of TRPV1 from previous studies to predict where the toxin binds to TRPV1 . This suggests that the toxin binds to a section of TRPV1 that is buried within the membrane . Moreover , the new models highlight a 'hydrophobic' region of the TRPV1 channel that may work as the heat sensor . Together , Bae , Anselmi et al . ’s findings reveal a new way in which a toxin can bind to a target protein in membranes . The next step is to test the idea that the hydrophobic region identified in this work is the part of TRPV1 that senses heat .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2016	Structural insights into the mechanism of activation of the TRPV1 channel by a membrane-bound tarantula toxin
Adolescents are particularly vulnerable to nicotine , the principal addictive component driving tobacco smoking . In a companion study , we found that reduced activity of the translation initiation factor eIF2α underlies the hypersensitivity of adolescent mice to the effects of cocaine . Here we report that nicotine potentiates excitatory synaptic transmission in ventral tegmental area dopaminergic neurons more readily in adolescent mice compared to adults . Adult mice with genetic or pharmacological reduction in p-eIF2α-mediated translation are more susceptible to nicotine’s synaptic effects , like adolescents . When we investigated the influence of allelic variability of the Eif2s1 gene ( encoding eIF2α ) on reward-related neuronal responses in human smokers , we found that a single nucleotide polymorphism in the Eif2s1 gene modulates mesolimbic neuronal reward responses in human smokers . These findings suggest that p-eIF2α regulates synaptic actions of nicotine in both mice and humans , and that reduced p-eIF2α may enhance susceptibility to nicotine ( and other drugs of abuse ) during adolescence . Tobacco use is a major global health problem with enormous economic and social costs . It remains the leading cause of preventable death worldwide , with tobacco-related illnesses estimated to kill more than 6 million people annually ( World Health Organization , 2011 ) . In the United States , the direct and indirect financial costs of smoking are estimated at more than $300 billion each year ( U . S . Department of Health and Human Services , Centers for Disease Control and Prevention , National Center for Chronic Disease Prevention and Health Promotion , Office on Smoking and Health , 2014; Xu et al . , 2015 ) . Adolescents are particularly at risk for initiating tobacco use , with a vast majority of all smokers beginning at age 18 or younger ( U . S . Department of Health and Human Services , Centers for Disease Control and Prevention , National Center for Chronic Disease Prevention and Health Promotion , Office on Smoking and Health , 2014 ) . A growing body of evidence from both human and animal studies indicates that adolescents are more susceptible than adults to the cellular and behavioral effects of nicotine , the main addictive component of tobacco ( Adriani and Laviola , 2004; Counotte et al . , 2011; Lydon et al . , 2014; Smith et al . , 2015 ) . Nicotine modifies dopamine ( DA ) signaling in key regions of the brain’s reward system ( Jasinska et al . , 2014; De Biasi and Dani , 2011 ) . Human neuroimaging studies of smokers have shown that exposure to nicotine alters reward-related activity in dopaminergic reward regions ( Rose et al . , 2012 ) . Moreover , in rodents , while nicotine is known to potentiate excitatory synaptic connections to DA neurons in the reward-related ventral tegmental area ( VTA ) ( Mansvelder and McGehee , 2000; Saal et al . , 2003 ) , the precise molecular mechanism underlying nicotine-induced long-term potentiation ( LTP ) remains unclear . In a companion article ( Huang et al . , 2016 ) , we found that the translation initiation factor eIF2α regulates adolescent vulnerability to the synaptic and behavioral effects of cocaine . Briefly , adolescent ( 5 weeks old ) mice proved to be more sensitive to a lower dose of cocaine than adult ( 3–5 months old ) mice with regard to reduced phosphorylation of eIF2α , the induction of LTP , and cocaine-induced behavior [conditioned place preference ( CPP ) , a common behavioral task that reflects behavioral reinforcement underlying the development of drug addiction] ( Koo et al . , 2012 ) . Consistent with these findings , genetic and pharmacological reduction in p-eIF2α-mediated translational control increased the susceptibility of adult mice to the synaptic and behavioral effects of cocaine , making adult mice resemble adolescents in this respect . Furthermore , other drugs of abuse ( including nicotine ) , that are known to induce LTP in the VTA ( Saal et al . , 2003 ) , also reduced p-eIF2α in the VTA ( Huang et al . , 2016 ) , thus highlighting the role of p-eIF2α as a common effector implicated in the initiation of addictive behavior . In the present study , we found that , like cocaine , mice with reduced p-eIF2α-mediated translation are more susceptible to nicotine-evoked synaptic potentiation in the VTA . Furthermore , using functional magnetic resonance imaging ( fMRI ) in humans , we identified a functional genetic variation in the promoter of the Eif2s1 gene encoding eIF2α that alters brain responses to rewarding stimuli in human tobacco smokers . In the accompanying study , we found that adolescent mice are more susceptible to cocaine-induced synaptic potentiation in VTA DA neurons . We therefore asked whether this was also true for nicotine . To answer this question , we measured glutamate-mediated excitatory postsynaptic currents ( EPSCs ) in VTA DA neurons from adolescent ( 5 weeks old ) and adult ( 3–5 months old ) mice 24 hr after single intra-peritoneal ( i . p . ) injection of either saline or different doses of nicotine . The peak amplitudes of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor ( AMPAR ) and N-methyl D-aspartate receptor ( NMDAR ) -mediated components of the evoked EPSCs ( recorded at +40 mV ) were isolated and used to calculate the AMPAR/NMDAR ratio as described ( Huang et al . , 2016 ) . An increase in this ratio was taken as an index of LTP . We found that a relatively low dose of nicotine ( 0 . 4 mg/kg i . p . ) was sufficient to induce LTP in VTA DA neurons from adolescent mice , but not in adult mice ( Figure 1a , b and Figure 1—figure supplement 1 ) . By contrast , a higher dose of nicotine ( 1 . 0 mg/kg ) was required to elicit comparable LTP in VTA DA neurons from adult mice ( Figure 1b and Figure 1—figure supplement 1 ) . Thus , like cocaine , nicotine induces LTP in VTA DA neurons at a significantly lower dose in adolescent mice . 10 . 7554/eLife . 12056 . 003Figure 1 . Reduced p-eIF2α-mediated translational control increases the susceptibility to nicotine-induced LTP . ( a-b ) Left , Representative traces of AMPAR and NMDAR EPSCs recorded from VTA DA neurons 24 hr after i . p . injection of saline or the indicated dose of nicotine . A relatively low dose of nicotine ( 0 . 4 mg/kg ) induced LTP , shown by an increase in AMPAR/NMDAR ratio in VTA DA neurons ( a , Right , P<0 . 01 , n=6 , 6 saline/0 . 4 mg/kg nicotine , t10=4 . 026 ) from adolescent mice ( 5 weeks old ) , but not in those from adult mice ( 3–5 months old , b , Right , P=0 . 802 , n=6/7/6 saline/0 . 4 mg/kg nicotine/1 . 0 mg/kg nicotine , F2 , 16=9 . 029 ) . A higher dose of nicotine ( 1 . 0 mg/kg ) was required to increase the AMPAR/NMDAR ratio in VTA DA neurons from adult mice ( b , Right , P<0 . 05 vs . saline or 1 . 0 mg/kg nicotine , n=6/7/6 saline/0 . 4 mg/kg nicotine/1 . 0 mg/kg nicotine , F2 , 16=9 . 029 ) . ( c-d ) A low dose of nicotine ( 0 . 4 mg/kg ) reduced p-eIF2α in the VTA of adolescents ( c , P<0 . 05 , n=9/5 saline/0 . 4 mg/kg nicotine , t12=2 . 479 ) , but not adult mice ( d , P=0 . 5710 , n=7/11 saline/0 . 4 mg/kg nicotine , t16=0 . 5784 ) . ( e ) A higher dose of nicotine ( 1 mg/kg ) was required to reduce p-eIF2α in VTA of adult mice ( P<0 . 01 , n=11/5 saline/1 mg/kg nicotine , t14=3 . 428 ) . ( f ) A low dose of nicotine ( 0 . 4 mg/kg ) failed to induce LTP in VTA DA neurons from adult WT ( Eif2s1S/S ) mice ( Left , P=0 . 964 , n=5 per group , t8=0 . 05 ) , but elicited significant LTP in adult Eif2s1S/A mice ( Right , P=0 . 003 , n=5 per group , t8=6 . 73 ) . ( g ) A low dose of nicotine ( 0 . 4 mg/kg ) induced LTP in ISRIB-injected adult mice compared to vehicle-injected mice ( P<0 . 001 , n=7/7 nicotine+vehicle/nicotine+ISRIB , t12=5 . 222 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 00310 . 7554/eLife . 12056 . 004Figure 1—figure supplement 1 . Adolescent mice are more susceptible than adult mice to nicotine-induced synaptic potentiation . Adolescent ( 5 weeks old , n=6–7 per group ) or adult mice ( 3–5 months old , n=6–7 per group ) were i . p-injected with saline or nicotine at indicated doses and whole-cell recordings were performed in VTA DA neurons . An increase in the AMPAR/NMDAR ratio ( an index of LTP ) was induced with the 0 . 4 mg/kg dose of nicotine ( F2 , 32=4 . 34 , P<0 . 01 vs . saline ) in adolescent mice , whereas 1 . 0 mg/kg was required for a significant increase in adults ( F2 , 32=4 . 34 , P<0 . 05 vs . saline or 0 . 4 mg/kg nicotine ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 004 In the accompanying article ( Huang et al . , 2016 ) we found that a low dose of cocaine reduces phosphorylation of eIF2α only in the VTA of adolescent mice . To test whether the same is true for nicotine , we injected adolescent and adult mice with a low ( 0 . 4 mg/kg ) and a relatively high ( 1 mg/kg ) dose of nicotine , respectively . Consistent with our findings with cocaine , we found that a low dose of nicotine ( 0 . 4 mg/kg ) reduced p-eIF2α only in the VTA of adolescent mice ( Figure 1c , d ) , whereas a higher dose of nicotine ( 1 mg/kg ) was required to reduce p-eIF2α in the VTA of adult mice ( Figure 1e ) . Thus , like cocaine , low doses of nicotine selectively reduce p-eIF2α in the VTA of adolescent mice , highlighting the involvement of p-eIF2α-mediated translational control during this period of heightened vulnerability to the effects of drugs of abuse . Based on these findings , we predicted that adult mice with reduced p-eIF2α-mediated translational control would be more susceptible to the synaptic effects of nicotine . To test this prediction , we injected both adult control and Eif2s1S/A mice ( in which p-eIF2α in VTA is reduced by about 50% because the phosphorylation site is mutated to alanine ( Huang et al . , 2016 ) with a low dose of nicotine ( 0 . 4 mg/kg i . p . ) . This low dose does not typically induce LTP in adult wild-type ( WT ) mice ( Figure 1b ) , and as expected , it failed to induce LTP in control WT Eif2s1S/S mice ( Figure 1f ) . By contrast , the same low dose of nicotine elicited LTP in adult Eif2s1S/A mice ( Figure 1f ) . Thus , like adolescent mice , adult mice with reduced eIF2α phosphorylation are more susceptible to the synaptic effects of nicotine . To further support these findings , we used the recently discovered small molecule ISRIB ( Sidrauski et al . , 2013 ) , which selectively blocks p-eIF2α-mediated translational control ( Sidrauski et al . , 2013 ) . Briefly , adult WT mice were acutely injected with both ISRIB ( 2 . 5 mg/kg ) and a low dose of nicotine ( 0 . 4 mg/kg ) and LTP was recorded in VTA DA neurons 24 hr later . Indeed , a low dose of nicotine ( 0 . 4 mg/kg ) induced LTP only in adult mice in which p-eIF2α-mediated translation was blocked pharmacologically with ISRIB ( Figure 1g ) . Hence , like adolescent mice , adult mice with reduced p-eIF2α-mediated translational control are more susceptible to nicotine-induced LTP . Thus , our findings that reducing p-eIF2α-mediated translational control renders animals more susceptible to the effects of both cocaine ( Huang et al . , 2016 ) and nicotine underscores a key role of p-eIF2α as a common regulator of drug-induced synaptic potentiation and behavior . Several studies in rodents have shown that certain genes or signaling pathways are implicated in the behavioral effects of drugs of abuse ( Robison and Nestler , 2011 ) , but their clinical relevance to humans remains unknown . Because p-eIF2α crucially regulates drug-induced changes in synaptic strength and behavior in mice , we sought to determine whether single nucleotide polymorphisms ( SNPs ) in the eIF2α signaling pathway could be associated with reward-induced changes in neuronal activity in human smokers . Indeed , by studying specific SNPs chosen on the basis of this hypothesis , we bypassed the problems inherent to large exploratory data analyses and multiple comparisons typical of genome-wide association studies . In this way we increased the chances of finding true effects related to biological processes , while reducing the possibility of false positives . According to neuroimaging studies , neuronal activity in key reward areas of the human brain is strongly associated with indices of drug use ( Rose et al . , 2014 ) . We therefore measured reward-mediated activity in the caudate and putamen—brain regions with crucial reward-related functional connections to the VTA ( Koob and Volkow , 2010 ) —of tobacco smokers and non-smokers ( Figure 2—figure supplement 1 ) . To elicit reward responses in the fMRI scanner , participants received small ( 1 mL ) squirts of sweet juice orally while functional MRI images of their brains were collected ( see Material and Methods and Figure 2—figure supplement 2 ) . Consistent with previous findings in cocaine and tobacco users ( Rose et al . , 2012; Rose et al . , 2014 ) , we observed significantly lower reward-induced activity in the caudate and putamen of smokers ( Figure 2a and c ) , indicating that smokers find sweet drinks less rewarding . More importantly , we identified an SNP ( rs10144417 ) in the Eif2s1 ( eIF2α ) gene that revealed an interaction between genotype and smoking status . While smokers carrying the AG/GG genotype showed lower reward-dependent activity compared with that of the AA smokers , such activity did not differ between non-smokers of both genotypes ( Figure 2d ) . These data indicate that rs10144417 in the Eif2s1 gene is associated with both reward signaling and tobacco use . 10 . 7554/eLife . 12056 . 005Figure 2 . The effect of a single nucleotide polymorphism ( SNP ) in the promoter of the Eif2s1 gene on reward-dependent striatal activity in human tobacco smokers . ( a-c ) Reward-related activity in caudate/putamen is lower in smokers than non-smokers ( b , P<0 . 01 , n=33/55 , t86=2 . 678 ) . ( b-c ) Transverse ( b ) and sagittal ( c ) views of significant fMRI BOLD signal in caudate/putamen of non-smokers compared to smokers in response to juice reward . ( d ) Interaction between smoking and rs10144417 genotype ( P<0 . 05 , F1 , 86=5 . 836 ) . ( e ) Partial alignment of Eif2s1 promoter sequences in human and related animals . Note high level of nucleotide conservation: the rs10144417 SNP is indicated in blue and the non-conserved nucleotides in red . ( f ) Schematic of firefly luciferase reporter constructs , in which a 5-kb Eif2s1 promoter fragment containing either the A or G allele was cloned upstream of the firefly luciferase gene in the p15Amp reporter vector . A renilla luciferase reporter was co-transfected with reporters containing the A or G variant and firefly luciferase ( Fluc ) activity was normalized to renilla ( Rluc ) activity . ( g ) Effects of A and G variants of SNP rs10144417 on the transcriptional activity of the Eif2s1 promoter as assessed by a luciferase reporter assay in HEK-293 cells . The data are from three independent experiments ( P<0 . 001 , n=6 per group , t10=5 . 405 ) . ( h ) Western blotting showing that overexpression of Eif2s1 ( pRc/CMV-Eif2s1 ) increased eIF2α levels compared to control ( vector alone pRc/CMV ) . ( i ) Diagram of the 5′ UTR-Ophn1-Fluc reporter , which consists of the 5’UTR of Ophn1 mRNA fused to the coding region of Firefly luciferase ( Fluc ) . A renilla luciferase ( Rluc ) reporter vector was co-transfected into HEK293T as a transfection control . ( j ) Overexpression of Eif2s1 reduced expression of 5′ UTR-Ophn1-Fluc ( P<0 . 01 , n=6 , t10=3 . 9425 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 00510 . 7554/eLife . 12056 . 006Figure 2—figure supplement 1 . Demographic information of human participants involved in fMRI studies . ( a ) Table showing the number of participants by gender , age , and smoking status carrying the A or the G variant in the Eif2s1 gene . ( b ) Table showing the number of participants by their self-reported ethnicities The participants have no history of any other drug dependance . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 00610 . 7554/eLife . 12056 . 007Figure 2—figure supplement 2 . fMRI recording paradigm and reward-stimulus pairing in human smokers . ( a ) The fMRI recording session consisted of four 5–7 min blocks of light-juice pairings . A self-paced break ( “inter-run interval” ) was included between runs to allow participants to ask any questions and to allow the investigators to provide feedback on participant motion within the scanner . ( b ) A total of fifty-five light-juice ( 1 mL ) pairings were presented . The delay between light and juice was 7 s . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 00710 . 7554/eLife . 12056 . 008Figure 2—figure supplement 3 . ATF4-Luciferase construct design and activity with Eif2s1 . ( a ) Diagram of the 5 UTR-ATF4 Fluc reporter , which consists of the 5’UTR of ATF4 mRNA fused to the coding region of Firefly luciferase ( Fluc ) . A renilla luciferase ( Rluc ) reporter vector was co-transfected into HEK293T as a transfection control . ( b ) Overexpression of Eif2s1 reduced expression of 5 UTR-ATF4-Fluc ( P<0 . 0001 , n=4 per group , t6=11 . 33 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12056 . 008 In silico analysis revealed that rs10144417 spans a highly conserved region of the Eif2s1 promoter ( Figure 2e ) . To examine the functional effect of the A/G polymorphism ( rs10144417 ) at this site , we measured firefly luciferase reporter activity ex-vivo using a 5 kilobase ( kb ) segment of the Eif2s1 promoter ( Figure 2f ) . Briefly , human embryonic kidney ( HEK293T ) cells were co-transfected with a firefly luciferase reporter ( Fluc ) containing either the A or G variant and a renilla luciferase reporter ( Rluc ) , which was used as a transfection control . We found that , compared to the A variant , the normalized G variant Fluc/Rluc ratio was significantly increased ( ~40% , Figure 2g ) , reflecting increased expression of Eif2s1 . To determine whether overexpression of Eif2s1 could affect p-eIF2α–mediated translational control , we co-expressed in HEK293T cells Eif2s1 ( Donzé et al . , 1995 ) ( Figure 2h ) , a 5’UTR-Ophn1 firefly luciferase reporter ( 5’UTR-Ophn1-Fluc; Figure 2i ) , whose translation is known to be enhanced by conditions that increase p-eIF2α ( Di Prisco et al . , 2014 ) , and a renilla Luciferase reporter ( Rluc ) , which was used to calculate the relative Fluc/Rluc translation ratio . Strikingly , increased expression of Eif2s1 significantly reduced the normalized Fluc/Rluc ratio , indicating a significant reduction in translation of Ophn1 mRNA ( Figure 2j ) . Additionally , expression of Eif2s1 selectively reduced 5’UTR-Ophn1-Fluc activity , but had no effect on Rluc activity ( data not shown ) . These data were further supported by our findings that overexpression of Eif2s1 also reduced translation of a 5’UTR-Atf4 firefly luciferase reporter ( Figure 2—figure supplement 3 ) , which is typically up-regulated by increased p-eIF2α ( Sidrauski et al . , 2013; Lu et al . , 2004 ) . Hence , overexpression of the alpha subunit of eIF2 ( Eif2s1 ) reduces p-eIF2α–mediated translation . The mechanism by which increased expression of Eif2s1 inhibits p-eIF2α–mediated translation remains to be determined . Several different mechanisms could be at play . Overexpression of Eif2s1 could affect eIF2α-mediated translation by altering: a ) the assembly of eIF2 complex [eIF2 is a heterotrimer consisting of an alpha ( encoded by Eif2s1 ) , a beta ( encoded by Eif2s2 ) , and a gamma subunit ( encoded by Eif2s3 ) , b ) the binding of eIF2 to key regulatory proteins ( e . g . , eIF2B and/or eIF5 ) , or c ) by titrating the phosphorylated alpha subunit away from the eIF2 complex . Moreover , overexpression of Eif2s1 could alter the expression of a given eIF2α kinase or phosphatase . Such compensatory translational homeostatic mechanisms have been observed when the levels of key translation initiation factors ( e . g . , eIF4E , 4E-BPs , PABP , Paips ) are either increased or decreased ( Khaleghpour et al . , 1999; Yanagiya et al . , 2012; Yoshida et al . , 2006 ) . Collectively , our mouse and human data suggest that reduced p-eIF2α-mediated translational control mediates a genetic predisposition to greater risk for drug-induced changes in synaptic strength , which may account for the greater vulnerability of adolescents , even as first time drug users . For instance , mice with reduced eIF2α-mediated translation are more susceptible to nicotine-induced changes in synaptic function . Similarly , smokers carrying the G variant , who also have reduced eIF2α-mediated translation , show reduced reward-induced activity in the caudate and putamen , suggesting that such individuals are likely to consume more drugs to obtain reward activity comparable to that of non-smokers . Finally , these insights may hold promise for new p-eIF2α-based approaches to treating drug abuse . All experiments were conducted on male and female mice from the C57BL/6 background . Eif2s1S/A mice were previously described ( Di Prisco et al . , 2014 ) . Mice were kept on a 12h/12 hr light/dark cycle ( lights on at 7:00 am ) and had access to food and water ad libitum . Animal care and experimental procedures were approved by the institutional animal care and use committee ( IACUC ) at Baylor College of Medicine , according to NIH Guidelines . No statistical methods were used to predetermine sample sizes . All sample sizes met the criteria for corresponding statistical tests—our sample sizes are similar to those reported in previous publications ( Bellone and Lüscher , 2006; Ungless et al . , 2001 ) . Nicotine was dissolved in 0 . 9% saline and injected at a volume of 5 ml/kg . ( - ) -Nicotine hydrogen tartrate was obtained from Sigma-Aldrich ( St . Louis , MO ) . ISRIB ( provided by P . Walter ) was dissolved in DMSO and further diluted in PEG-400 ( 1:1 ratio ) as previously described ( Sidrauski et al . , 2013 ) . For both electrophysiological and behavioral experiments , ISRIB ( 2 . 5 mg/kg ) or vehicle ( DMSO/PEG-400 , 2 ml/kg ) was injected 90 min before nicotine or control saline injections . Electrophysiological recordings were performed as previously described ( Ungless et al . , 2001 ) , investigators blind to genotype . Each electrophysiological experiment was replicated at least three times . Briefly , mice were anesthetized with a mixture of ketamine ( 100 mg/kg ) , xylazine ( 10 mg/kg ) , and acepromazine ( 3 mg/kg ) . Mice were transcardially perfused with an ice-cold , oxygenated solution containing ( in mM ) NaCl , 120; NaHCO3 , 25; KCl , 3 . 3; NaH2PO4 , 1 . 2; MgCl2 , 4; CaCl2 , 1; dextrose , 10; sucrose , 20 . Horizontal slices ( 225–300 μm thick ) containing the VTA were cut from the brains of adolescent ( 5 week old ) or adult ( 3–5 month old ) C57BL/6J mice using a vibrating tissue slicer ( VF-100 Compresstome , Precisionary Instruments , San Jose , CA , or Leica VT 1000S , Leica Microsystems , Buffalo Grove , IL ) . Slices were next incubated at 34°C for 40 min then kept at room temperature for at least 30 min before they were transferred to a recording chamber where they were continuously perfused with artificial cerebrospinal fluid ( ACSF ) at 32°C at a flow rate of 2–3 ml/min . The recording ACSF differed from the cutting solution in regard to the concentration of MgCl2 ( 1 mM ) and CaCl2 ( 2 mM ) . Recording pipettes were made from thin-walled borosilicate glass ( TW150F-4 , WPI , Sarasota , FL ) . After filling with intracellular solution ( in mM ) : 117 CsMeSO3; 0 . 4 EGTA; 20 HEPES; 2 . 8 NaCl , 2 . 5 ATP-Mg 2 . 0; 0 . 25 GTP-Na; 5 TEA-Cl , adjusted to pH 7 . 3 with CsOH and 290 mOsmol/l , they had a resistance of 3–5 MΩ . Data were obtained with a MultiClamp 700B amplifier , digitized at 20 kHz with a Digidata 1440A , recorded by Clampex 10 and analyzed with Clampfit 10 software ( Molecular Devices ) . The signals were filtered online at 4 kHz with a Bessel low-pass filter . A 2 mV hyperpolarizing pulse was applied before each EPSC to evaluate the input and access resistance ( Ra ) . Data were discarded when Ra was either unstable or greater than 25 MΩ , holding current was > 200 pA , input resistance dropped > 20% during the recording , or baseline EPSCs changed by > 10% . Traces illustrated in Figures are averages of 10–15 consecutive traces . After establishing a gigaohm seal ( > 2 GΩ ) and recording stable spontaneous firing in cell-attached voltage clamp mode ( at -70 mV holding potential ) , cell phenotype was determined by measuring the width of the action potential ( Ford et al . , 2006 ) . AMPAR/NMDAR ratios were calculated as previously described ( Ungless et al . , 2001 ) . Briefly , neurons were voltage-clamped at +40 mV until the holding current stabilized ( at < 200 pA ) . Monosynaptic EPSCs were evoked at 0 . 05 Hz with a bipolar stimulating electrode placed 50–150 μm rostral to the lateral VTA . Picrotoxin ( 100 μM ) was added to the recording ACSF to block GABAAR-mediated IPSCs . After recording the dual-component EPSC , DL-AP5 ( 100 μM ) was bath-applied for 10 min to remove the NMDAR component , which was then obtained by offline subtraction of the remaining AMPAR component from the original EPSC . The peak amplitudes of the isolated components were used to calculate the AMPAR/NMDAR ratios . Picrotoxin and DL-AP5 were purchased from Tocris Bioscience and all other reagents and experimental compounds were obtained from Sigma-Aldrich . The Eif2s1 promoter region was cloned out of human BAC RP11-713C11 using primers 5’ sense 5’-ATTCGCGAGGGAAAGATTTCAATTC-3’ , and antisense 5’-TCTGCAATTTAAACAAAAGAATTAAGTAAGT-3’ . The firefly luciferase was amplified from Luciferase-pcDNA3 ( Promega , Madison , WI ) using the sense primer 5’-ACTTACTTAATTCTTTTGTTTAAATTGCAGAATGGAAGACGCCAAAAACATAAAG3’ and antisense primer 5’-ACATTTCCCCGAAAAGTGCCACCTGCCATAGAGCCCACCGCATCCCCAG-3’ . The p15a_Amp was amplified from pWSTK6 using sense primer 5’-CAGGTGGCACTTTTCGGGGAAATGT-3’ and antisense primer 5’-GAATTGAAATCTTTCCCTCGCGAATGCTAGCGGAGTGTATACTGGCTTAC-3’ . All three fragments were Gibson cloned and confirmed by DNA sequencing . The resulting plasmid ( Eif2s1-A-luc ) carries the A in the SNP rs101444417 position up-stream of firefly luciferase . In a mutagenesis reaction using the Eif2s1-A-luc plasmid as template and sense primer 5’-GGTTACTTAGCTAAAGCTAAGGTAGCGAGAAACAGTACTTTTCAG-3’ and antisense primer 5’-CCTGAAAAGTACTGTTTCTCGCTACCTTAGCTTTAGCTAAGTAACC-3’ , we generated a clone ( Eif2s1-G-luc ) containing the G in the SNP rs101444417 position up-stream of firefly luciferase . HEK293T cells ( a widely used human cell line ) were grown in 24-well plates in Gibco DMEM+Glutamax ( Life Technologies ) supplemented with 10% FBS , 100 units of Pen/Strep per ml . Eif2s1-G-luc and Eif2s1-A-luc were co-transfected with a renilla luciferase ( Rluc ) plasmid pRL-TK ( Promega , Madison , WI ) into HEK293T cells at 50–80% confluency using Lipofectomine LTX plus ( Life Technologies ) . pRc/CMV and pRc/CMV-eIF2s1 vectors , which were previously reported ( Donzé et al . , 1995 ) , were co-transfected with 5’UTR-Ophn1-Fluc and RLuc into HEK293T , as described above . Similarly 5’UTR-Atf4-Fluc ( Sidrauski et al . , 2013 ) was co-transfected with RLuc into HEK293T cells . 24 hr after transfection , cell extracts were prepared in passive lysis buffer and samples were collected in pre-chilled microcentrifuge tubes and lysed in homogenizing buffer [200 mM HEPES , 50 mM NaCl , 10% Glycerol , 1% Triton X-100 , 1 mM EDTA , 50 mM NaF , 2 mM Na3VO4 , 25 mM β-glycerophosphate , and EDTA-free complete ULTRA tablets ( Roche , Indianapolis , IN ) ] . All procedures were approved by Baylor College of Medicine Internal Review Board . Smokers and non-smokers were recruited from the Houston metropolitan area via fliers , newspaper , and internet advertisements and received a small monetary compensation . All participants were pre-screened to rule out non-tobacco substance dependence or MRI contraindications ( e . g . head injuries , foreign metal in the body , claustrophobia , or pregnancy ) . Smokers currently seeking cessation treatment were also excluded . Smoking history and dependence were evaluated using the Fagerström Test for Nicotine Dependence ( FTND ) ( Heatherton et al . , 1991 ) , Shiffman-Jarvik Withdrawal Questionnaire ( SJWQ ) ( Shiffman and Jarvik , 1976 ) , and the Positive and Negative Affect Schedule ( PANAS ) ( Watson and Clark , 1994 ) . All participants read and signed an Informed Consent to participate in this research protocol . No statistical methods were used to predetermine sample sizes but sample sizes were similar to those previously reported ( Rose et al . , 2012; Salas et al . , 2010 ) . The subjects were either smokers ( n=35 , average age=43 . 26 years ± 12 . 19 years , 77 . 1% males , 22 . 9% females , smoking at least 10–15 cigarettes per day and had smoked for at least the past year ) or non-smokers ( n=54 , average age=31 . 37 ± 11 . 60 years , 44 . 4% males 55 . 5% females , lifetime incidence of smoking less than 50 cigarettes or no cigarettes for the last year ) . For more information on participant demographic , see Figure 2—figure supplement 1 . Participants were scanned in 3T Siemens Trio scanners in the Center for Advanced MRI at Baylor College of Medicine . Structural MPRAGE images were collected as 160 1×1×1 mm axial slices ( TE=2 . 66 ms , TR=1200 ms , flip angle=12° , 256 × 256 matrix ) while functional images were collected as 2x2x2 mm epi scans ( TR 2 s , TE 40 ms ) over a 44 mm “slab” covering the regions of interest . Light-juice pairings were presented during functional scanning using ePrime software ( Psychology Software Tools , Sharpsburg , PA ) . For each of a maximum of 55 light-juice pairings , a Standard Infuse/Withdraw Harvard 33 Twin Syringe Pump ( Harvard Apparatus , Holliston , MA ) delivered 1 mL of juice 7s after a light cue . Prior to scanning , participants were given a choice of sugar-free sweet drinks ( such as lemonade , iced tea with peach or fruit punch , and the preferred juice was given during scanning ) . fMRI data were analyzed using a standard AFNI processing stream ( Cox , 1996 ) . Briefly , the first four TRs were removed to establish a stable baseline . Data were then corrected for slice-time acquisition ( 3dTshift ) , aligned to the first image and measured for motion ( 3dvolreg ) , registered to the high resolution MPRAGE , and transformed to MNI space using a single spatial transform ( @auto_tlrc , 3dAllineate ) . A 4 . 5 mm smoothing kernel was applied in 3dmerge and submitted to a General Linear Model ( GLM ) regression in 3dDeconvolve . GLM regressors for linear , quadratic , and cubic linear trends; x , y , z , roll , pitch , and yaw motion parameters; and the two stimulus conditions: visual cue and juice reward were included . An analysis of the interaction between smoker/non-smoker group and genotype ( e . g . rs10144417 AA vs . GG/AG ) was performed using a 3D multivariate model ( 3dMVM ) and family-wise error correction . The percent signal change in BOLD activation for individual subjects was extracted from caudate and putamen regions of interest ( ROIs ) and restricted to those voxels identified as having a P<0 . 05 and alpha<0 . 05 in the multivariate model of group by genotype interaction . No effect of age , number of years smoking , average number of cigarettes per day , gender , or ethnicity was observed in the fMRI analysis . DNA was isolated from buccal swabs and genotyping was completed using Illumina ( San Diego , CA ) HumanOmniExpress-12 v1 . 1 BeadChip arrays ( cat# WG-312-1120 ) containing approximately 741K SNPs . Genomic DNA ( 200 ng ) from each sample was processed following Illumina’s Infinium HD Ultra Assay protocol . BeadChip images were captured using Illumina iScan System . The Illumina chip contained the Eif2s1 rs10144417 SNP ( Global minor allele frequency MAF:0 . 2993 ) . fMRI activity was studied according to the genotype of analyzed subjects . Subjects with AG and GG genotypes were pooled for analysis and compared to subjects with the AA genotype using AFNI’s statistical software . All data are presented as mean ± s . e . m . Statistical analyses were performed with SigmaPlot ( Systat Software ) . Data distribution normality and homogeneity of variance were assessed by Shapiro-Wilk and Levene tests , respectively . The statistics were based on the two-sided Student’s t test , or one- or two-way ANOVA with Tukey’s HSD ( or HSD for unequal sample sizes where appropriate ) to correct for multiple post hoc comparisons . Within-groups variation is indicated by standard errors of the mean of each distribution , which are depicted in the graphs as error bars . P<0 . 05 was considered significant ( P<0 . 05 , P<0 . 01 , P<0 . 001 , **P<0 . 0001 ) .	Nicotine addiction is a serious public health problem . People who start using nicotine during adolescence are more likely to become addicted to it during adulthood , but the reasons for this are not well understood . Nicotine causes long-lasting changes in the brain that are responsible for the feelings of pleasure and reward . In particular , nicotine strengthens the connections between neurons at structures called synapses and increases communication between reward-related neurons in key reward areas of the brain . This hijacking of the natural reward system requires new proteins to be made . However , the relationship between protein synthesis and adolescents being particularly vulnerable to nicotine addiction was not known . In a related study , Huang et al . found that the reduced activity of a protein called eIF2α , which controls the production of new proteins , accounts for why adolescents are more likely to become addicted to cocaine than adults . Thus , Placzek et al . wanted to know whether the same was true for nicotine and whether the proteins controlled by eIF2α are involved in the way human nicotine addicts experience reward . Placzek et al . found that adolescent mice are more susceptible than adult mice to the changes in synaptic strength that are caused by nicotine . This increased susceptibility results from reduced activity levels of the protein eIF2α . Reducing the activity of eIF2α in adult mice made their synapses as likely to change strength in response to nicotine as the synapses of adolescent mice . Placzek et al . also used a technique called functional magnetic resonance imaging and found that compared to non-smokers , the brain activity of human smokers was significantly reduced when given a natural reward . Further studies revealed a variation in the gene encoding the eIF2α protein that affects how smokers respond to a reward , suggesting that this variant is linked to the likelihood that a person will be addicted to nicotine . This work raises several important questions . In addition to regulating the initial adaptive changes induced in the brain by nicotine , does eIF2α activity affect compulsive nicotine use ? If so , could targeting parts of the eIF2α pathway help treat nicotine addiction ? Finally , further studies could explore whether the gene variant identified by Placzek et al . affects how users of other drugs ( such as cocaine or alcohol ) respond to natural rewards .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2016	Translational control of nicotine-evoked synaptic potentiation in mice and neuronal responses in human smokers by eIF2α
Photoactivation ( 'uncaging’ ) is a powerful approach for releasing bioactive small-molecules in living cells . Current uncaging methods are limited by the random distribution of caged molecules within cells . We have developed a mitochondria-specific photoactivation method , which permitted us to release free sphingosine inside mitochondria and thereafter monitor local sphingosine metabolism by lipidomics . Our results indicate that sphingosine was quickly phosphorylated into sphingosine 1-phosphate ( S1P ) driven by sphingosine kinases . In time-course studies , the mitochondria-specific uncaged sphingosine demonstrated distinct metabolic patterns compared to globally-released sphingosine , and did not induce calcium spikes . Our data provide direct evidence that sphingolipid metabolism and signaling are highly dependent on the subcellular location and opens up new possibilities to study the effects of lipid localization on signaling and metabolic fate . Sphingolipids are one of the major lipid species in cellular membranes of all eukaryotic cells . Apart from maintaining structural properties of membranes , at least four sphingolipid metabolites , sphingosine , ceramide , sphingosine 1-phosphate ( S1P ) and ceramide 1-phosphate ( C1P ) , are also signaling messengers that regulate fundamental cellular processes ( Aguilera-Romero et al . , 2014; Hannun and Obeid , 2008; Maceyka and Spiegel , 2014; Atilla-Gokcumen et al . , 2014 ) . Due to their essential roles , aberrant sphingolipid levels are linked to a broad range of diseases , including cancers , diabetes , inflammation and neurodegeneration ( Maceyka and Spiegel , 2014; Platt , 2014; Guri et al . , 2017 ) . Interestingly , sphingolipid messengers could serve distinct functions depending on their subcellular localization , but this is currently difficult to address ( Hannun and Obeid , 2008 ) . For example , extracellular S1P activates receptors on the cell surface ( Lee et al . , 1998; Van Brocklyn et al . , 1998; Liu et al . , 2000 ) , while S1P generated inside the nucleus has been proposed to regulate histone acetylation through direct interactions with the histone deacetylases ( HDAC ) ( Hait et al . , 2009 ) . Furthermore , several protein domains have been suggested to bind to S1P , which might affect the recruitment of many PH domains ( Vonkova et al . , 2015 ) . Although cellular localization is likely to be an important regulator of the metabolism and functions of sphingolipids , there is no direct evidence to demonstrate this . In the last years some new techniques were introduced to explore lipid metabolism ( Wenk , 2005; da Silveira Dos Santos et al . , 2014; Papan et al . , 2014 ) and to map lipid-protein interactions ( Haberkant et al . , 2016; Saliba et al . , 2016 ) . While proven to be valuable tools , they do not permit the acute manipulation of lipid amounts in living cells . Recently developed optogenetic ( Levskaya et al . , 2009 ) or chemical induced dimerization techniques ( Feng et al . , 2014 ) have been applied to rapidly change lipid levels by controlling the localization of lipid-metabolizing enzymes in living cells . Additionally , photoswitchable fatty acids and their derivatives also enables precise optical control of specific signaling lipid-protein interactions ( Frank et al . , 2015; Frank et al . , 2016 ) . However , successful design often requires detailed mechanistic understanding of the protein-lipid interactions . Alternatively , photoactivation ( 'uncaging’ ) has become a common way to quickly increase the lipid supply . This method requires the chemical protection of the lipid head group using a photo-labile moiety , which blocks the lipid bioactivities . Once inside cells , a strong light flash can easily remove the protecting group within milli-seconds to seconds , and thereby release the bioactive lipids with high temporal resolution ( Höglinger et al . , 2014; Klán et al . , 2013 ) . Equipped with confocal fluorescence microscopy , the uncaging approach offers a convenient platform to study lipid signaling at the single cell level with minimal perturbation . To date , a number of caged lipids , including sphingosine ( Höglinger et al . , 2015; Höglinger et al . , 2017 ) , S1P ( Qiao et al . , 1998 ) , fatty acids ( Nadler et al . , 2015 ) , and phosphatidylinositol phosphates ( PIPs ) ( Subramanian et al . , 2010 ) have been successfully applied in lipid signaling studies . To perform the photoactivation experiments , coumarin and nitrophenyl derivatives are widely used as caging groups to mask the bioactive small molecules ( Klán et al . , 2013 ) . Nevertheless , the caged molecules are usually distributed in the cell randomly . Selective uncaging is only achieved through sophisticated fluorescence microscopy that is able to produce a highly focused laser beam in a defined area at the subcellular level . This method has disadvantages , though , because it is very difficult to exclusively activate the compartment of interest due to the diffraction limit of light . More importantly , this 'selective uncaging’ strategy is limited to single-cell analysis , and hence is not suitable for biochemical assays that are needed to study metabolism . Only a few exceptions allow uncaging with subcellular specificity , either from the outer leaflet of plasma membrane because the probes cannot enter into cells ( Nadler et al . , 2015 ) , or in mitochondria to uncouple the membrane potential , but the latter system did not permit to visualize the caged compound once inside the cell making it difficult to determine its precise localization ( Chalmers et al . , 2012 ) . Here , we report a new mitochondria-targeted photoactivation method , which enabled the rapid release of sphingosine inside the mitochondria in living cells . We combined this method with lipid analysis by mass spectrometry ( da Silveira Dos Santos et al . , 2014; Han , 2016 ) to study mitochondrial sphingosine metabolism in intact cells . We observed that after photo-release , free sphingosine was rapidly phosphorylated into S1P , driven by sphingosine kinases . Inhibition of sphingosine kinases largely suppressed the S1P generation . We also applied this technique to monitor the turnover of sphingosine and compared the mito-caged sphingosine ( Mito-So ) with the globally distributed caged sphingosine ( Sph-Cou ) ( Höglinger et al . , 2015; Höglinger et al . , 2017 ) . In both cases , sphingosine was quickly lost indicating that these signaling sphingolipids are rapidly metabolized even when targeted to mitochondria . Using stable isotope-labeled caged sphingosine precursors , we investigated the conversion of sphingosine into ceramides and sphingomyelins after uncaging , and found that sphingosine metabolism was highly dependent on its subcellular localization . Mitochondria released sphingosine was used to continuously produce ceramides , whereas globally released sphingosine was more rapidly metabolized through sphingolipid metabolic network . Finally , we compared the two caged probes in calcium mobilization experiments . In contrast to the globally distributed Sph-Cou , the Mito-So failed to trigger calcium mobilization , demonstrating the importance of lipid localization in signal transduction . Coumarin and nitrophenyl derivatives are the most popular caging molecules for biological studies . Although both can be efficiently cleaved from the caged molecules inside living cells , only coumarin-based molecules generate fluorescent signals and thus can be easily visualized under fluorescence microscopy . In order to determine the subcellular localization of caged molecules in cells , we chose to modify the 7-amino coumarin ( 1 ) by introducing a carboxylic linker and a methyl group . SeO2-mediated hydroxylation at the benzylic position afforded ( 3 ) . TFA deprotection of the t-Bu ester allowed facile conjugation to a triphenylphosphonium ( TPP ) group via an amide bond to afford TPP-Cou-OH ( 4 ) . Introduced by Murphy and co-workers ( Murphy , 1997; Murphy , 2008 ) , the lipophilic TPP cation is the best known mitochondria tag that facilitates accumulation of functional groups into mitochondrial matrix . Using the TPP-labeled coumarin alcohol ( TPP-Cou-OH ) , we synthesized caged sphingosine ( Mito-So ) and sphinganine ( Mito-Sa ) following standard protocols ( Figure 1 ) . To evaluate the photo-cleavage efficiency of the mitochondria-targeted caged lipids , we performed uncaging experiments in aqueous solution using a powerful UV lamp . Previously described diethylaminocoumarin caged sphingosine ( Sph-Cou ) was used as a reference compound ( Höglinger et al . , 2015 ) . We analyzed the caging probes by LC-MS at different time points ( Figure 2B ) . Most of the sphingosine was cleaved from the Mito-So after one minute , judging from the coumarin absorbance ( Figure 2B ) . Compared to Sph-Cou , both Mito-So and Mito-Sa were completely cleaved after two minutes ( Figure 2C ) , indicating the TPP-Cou-OH is an effective photolabile molecule . On the other hand , our measurement also indicated that the caged probes are fluorescent dyes that emit in the blue region ( Figure 1—figure supplement 1 ) . Next , we treated Hela cells with Mito-So for 15 min in the presence of MitoTracker . 5 μM of Mito-So was used to obtain sufficient signals since it is a relatively weak fluorophore . The pattern of Mito-So fluorescence completely overlaps with MitoTracker ( Figure 2D , E , Figure 2—figure supplement 1 ) , demonstrating Mito-So and Mito-Sa are predominantly targeted to mitochondria . Like other TPP-containing mitochondria targeting molecules ( Murphy , 2008 ) , this accumulation relies on the integrity of mitochondrial membrane potential . When the potential is disrupted by CCCP , the mitochondria staining of Mito-So was not seen ( Figure 2F ) and the probe was randomly distributed throughout the cell . In mammalian cells , sphingosine can be phosphorylated into pro-proliferative S1P , or converted into pro-apoptotic ceramide . To explore the metabolism of mitochondrial sphingosine , we performed lipidomics on Hela cells incubated with Mito-So and exposed to UV light ( 350–450 nm ) for 2 min , which should be long enough to uncage most of the Mito-So ( Figure 2B ) . Immediately after uncaging on ice , cells were collected for lipid extraction . The results show that both sphingosine and S1P levels were dramatically elevated compared to the controls ( Figure 3A , B ) . Similarly , sphinganine and sphinganine 1-phosphate ( Sa-1P ) were also increased after uncaging of Mito-Sa ( Figure 3C , D ) . Interestingly , the sphinganine level after Mito-Sa uncaging is higher compared to the sphingosine level after uncaging Mito-So; the opposite effects were observed in the phosphorylated form . Since we treated the cells with equal amount of the probes , the results might reflect the fact that sphingosine is better recognized as a substrate for phosphorylation and/or that that Mito-Sa is more efficiently accumulated inside the cells . As a control , we examined the stability of Mito-So and Mito-Sa without photoactivation in the intracellular environments , and found no significant effects on sphingoid bases , showing that the Mito-caged lipids were sufficiently stable and were not cleaved by enzymatic activities such as esterases during the experiments . In addition , although sphingosine and sphinganine are structurally very similar , the sphinganine level remained stable after Mito-So uncaging ( Figure 3—figure supplement 1A ) , and vice versa ( Figure 3—figure supplement 1B , C ) . As a further control , we assessed the potential effects of UV light on the sphingolipid levels , by exposing the cells to continuous UV illumination without any caged probe and measuring the sphingosine and S1P levels . Although both lipids were slightly reduced after UV exposure ( Figure 3—figure supplement 2 ) , the reduction is rather minimal and is in the opposite direction compared to the effects of uncaging , showing that the effect of UV illumination is insignificant under these conditions . Likewise , the major lipid species and ceramides were not affected after Mito-So and Mito-Sa uncaging ( Figure 3—figure supplement 1D , E ) . Taken together , our results demonstrate that these photochemical probes , upon illumination , effectively released sphingoid bases with high spatiotemporal resolution . As sphingosine is the major sphingoid base in mammalian cells , we focused on Mito-So in the following studies . Next , we sought to confirm that the elevated S1P after photo-releasing sphingosine required enzymatic phosphorylation and occurred after uncaging . Since sphingosine and S1P are interconvertible molecules , we assumed that the increased amount of S1P was due to phosphorylation of sphingosine by sphingosine kinases . There are two isoforms of sphingosine kinases in mammalian cells , sphingosine kinase 1 ( SphK1 ) and sphingosine kinase 2 ( SphK2 ) , both of which are capable of phosphorylating sphingosine into S1P ( Maceyka et al . , 2005 ) . To investigate the role of SphKs , we employed the CRISPR/Cas9-based genome editing strategy to generate a control and the double kinase knock-out cell line based on HeLa MZ cells ( Figure 4—figure supplement 1 ) ( Ran et al . , 2013; Liao et al . , 2015; Harayama and Riezman , 2017 ) . After showing that removing sphingosine kinases did not disrupt mitochondria morphology or membrane potential ( Figure 4—figure supplement 2 ) , we carried out uncaging experiments using Mito-So in the cell lines and measured their sphingolipid levels . Consistent with previous results ( Figure 3 ) , sphingosine levels were dramatically increased after uncaging in the cells ( Figure 4A ) . As expected , SphK double knock-out ( dKO ) cells failed to raise the S1P level after uncaging , in contrast to the control cells in which significantly elevated S1P was detected ( Figure 4B ) . The level of S1P in the control cells was not as high as in Figure 3 , most likely due to the intrinsic differences between the cell lines used initially and the cell line used for generating knockout cells , even though at the beginning they were both selected from HeLa cells . Nevertheless , the Hela MZ is the most appropriate control for the mutant lines we produced . To provide additional evidence supporting the role of SphKs in the S1P production , we performed uncaging experiments using HeLa cells in the presence of the sphingosine kinase inhibitor SKI-II , which inhibits both SphK1 and SphK2 ( French et al . , 2003 ) . Consistent with the results in the knockout cells , SKI-II treated cells generated less S1P , and increased the level of sphingosine ( Figure 4C , D ) . The SKI-II did not fully block S1P synthesis , which might be due to the fact that the SKI-II and sphingosine are competitive substrates for the kinases ( Lim et al . , 2012; French et al . , 2003 ) , and/or the access of SKI-II to the mitochondria might be limited . Taken together , our results clearly show that the rapid S1P accumulation after uncaging was catalyzed by sphingosine kinases . As we detected a significant amount of S1P rapidly after uncaging , it is essential to know whether the S1P accumulation was driven by very fast enzymatic reactions , or if phosphorylation occurred on the caged sphingosine molecules ( Mito-So ) prior to UV illumination . Therefore , we extracted lipids from cells that were incubated with Mito-So , uncaged them after extraction and measured sphingosine and S1P levels . There was no significant increase in S1P following this protocol , while , as expected , the amount of sphingosine was greatly increased ( Figure 4—figure supplement 3 ) . These results show that sphingosine kinases were not able to use the caged lipids as substrates in vivo even though the hydroxyl group was available , which further supports our conclusion that S1P accumulation is due to sphingosine kinase activity on the uncaged sphingoid bases . Although the sphingoid bases were initially photo-released in mitochondria , we cannot formally rule out that free bases were rapidly transported out of the mitochondria and phosphorylated in the cytoplasm . To investigate this issue , we isolated mitochondria from mouse liver and performed uncaging experiments using Mito-So ( Figure 5 ) . Consistent with live-cell experiments , after UV irradiation , we detected an elevated sphingosine level ( Figure 5A ) . Importantly , we also observed a significant increase in S1P , which was suppressed by the addition of the SKI-II inhibitor ( Figure 5B ) . Compared with Figure 3B , the phosphorylation of sphingosine into S1P was less efficient in purified mitochondria , which might have been due to the different experimental settings and the intrinsic activity of purified mitochondria compared to mitochondria in vivo . For instance , it is possible that the SPHKs are more efficient in their native environments than in isolated mitochondria . In addition , if some of the S1P generated inside mitochondria was exported during the experiments it would have been lost during our isolation and extraction procedure ( see following ) . To make sure the newly generated S1P was not caused by the contamination of mitochondria during purification and therefore produced outside of the mitochondria , we added a large amount of S1P ( 10 pmol , approx . 10 times more than produced in vitro ) in a mock experiment without Mito-So to the purified mitochondria , and then proceeded with pelleting and lipid extraction , but did not detect any significant change of either sphingosine or S1P ( Figure 5C , D ) . We therefore conclude that the S1P found in our assay was indeed inside mitochondria as extra-mitochondrial S1P was not detected in our assay , probably due to its relatively high solubility in aqueous buffers . Collectively , our data demonstrate that , upon illumination , free sphingosine was quickly released from Mito-So in the mitochondria and was partially and rapidly converted into S1P by SphKs . The rapid phosphorylation , together with our results using purified mitochondria , provide convincing evidence that at least part of the released sphingosine was phosphorylated inside mitochondria . This is consistent with reports finding that sphingosine kinases are partially localized to mitochondria ( Lim et al . , 2012; Strub et al . , 2011 ) . Sphingolipids are key components of cellular membranes as well as signaling molecules involved in essential physiological and pathological processes ( Maceyka and Spiegel , 2014; Aguilera-Romero et al . , 2014; Platt , 2014 ) . Their accumulation at specific intracellular locations could possibly be important for their physiological effects . Indeed , it has been proposed that sphingolipid signaling molecules could exhibit different cellular functions depending on their subcellular localization ( Hannun and Obeid , 2008 ) . Therefore , understanding their local metabolism is crucial to reveal the complexities of sphingolipid homeostasis and signaling . To address the effect of subcellular localization on metabolism , we applied the photochemical probe to release sphingosine in mitochondria , and thereafter monitored the decay of sphingosine over time . We also performed parallel experiments using a recent reported caged probe ( Sph-Cou , Figure 6A [Höglinger et al . , 2015] ) that releases sphingosine without subcellular selectivity . During the experiments , after 2 min uncaging on ice , the cells were incubated at 37°C and then were collected for lipid extraction at different time points ( Figure 6B ) . Not surprisingly , in both cases , the released sphingosine was quickly metabolized without affecting the amounts of other major lipid species ( Figure 6—figure supplement 1 ) . By simply fitting the time-course data with one-phase decay functions , we were able to estimate the half-lives of sphingosine . Notably , the decay of sphingosine ( t1/2 = 2 . 63 min , Figure 6C ) generated from Mito-So is much faster than the one from Sph-Cou ( t1/2 = 4 . 21 min , Figure 6E ) . Also , unlike the Mito-So , sphingosine released from Sph-Cou did not drop to the basal level even after 20 min , suggesting that part of the globally released sphingosine was stored inside cells , which prevented further metabolism during the time course . According to the sphingolipid metabolic network , sphingosine can either converted into ceramide by ceramide synthases , or be phosphorylated into S1P , which can be further degraded by sphingosine-1-phosphate lyase ( SGPL ) , hydrolyzed to phosphate and sphingosine by lipid phosphatases , or secreted extracellularly ( Aguilera-Romero et al . , 2014 ) . To explore the rate of conversion to ceramide , we uncaged Mito-So in SphK dKO cells , where the latter pathways are blocked due to a lack of formation of S1P and quantified the sphingosine level over time . Even though major lipid species were not affected ( Figure 6—figure supplement 1 ) , sphingosine levels still declined over time , but the half-life was significantly longer than in wild type cells ( t1/2 = 4 . 2 min , Figure 6D ) . Since S1P formation and sphingoid base turnover were blocked , the most logical explanation is that sphingosine turnover was the result of its conversion into ceramide by ceramide synthases , but this did not affect the total ceramide pool since steady state levels of ceramide in HeLa cells are much higher than sphingosine or S1P levels . In order to further investigate the metabolic fate of sphingosine into more abundant lipid species , it is important to distinguish the newly formed ceramide from the endogenous ones . For this purpose , we synthesized the d7-Mito-So and d7-Sph-Cou using heavy isotope labeled d7-sphingosine as a precursor . Following the same time frame and experimental settings , we measured the d7-sphingolipids over time after uncaging in live Hela cells ( Figure 7 ) . To better understand the sphingolipid metabolism under various subcellular contexts , we performed experiments using direct addition of the d7-sphingosine to cells , and compared this with the mito-released and globally-released sphingosine probes . Among the downstream sphingolipids , we focused on the formation of isotope-labeled C16 and C24 ceramides , the most abundant ceramides in HeLa cells , which also showed the most pronounced increase with negligible background signals in our study . As shown in Figure 7 ( left ) , we detected both d7-labeled ceramides immediately after uncaging of d7-Mito-So , demonstrating a rapid conversion to ceramide . The amount of newly formed ceramides steadily increased even after 20 min . Interestingly , globally localized d7-Sph-Cou generated a distinct metabolic pattern ( Figure 7 , middle ) . Fewer ceramides were formed at 0 min and both ceramides reached their maximum after 5 to 10 min and dropped significantly after 20 min , suggesting that the ceramides were quickly consumed to synthesize other sphingolipids . As for directly added d7-sphingosine , we observed continuous accumulation and sharp increase of ceramides over time , indicating that sphingosine entered from outside was efficiently metabolized by cells . As the most abundant sphingolipids on the plasma membrane , sphingomyelins are produced by sphingomyelin synthases using ceramides as precursors in the biosynthetic pathway . We also analyzed the signals of d7 C16 sphingomyelin and observed that , while the signals were increased over time in all cases , the signals generated after global release of d7 Sph-Cou uncaging was close to reaching a plateau , unlike the externally added and mitochondrial released d7 Sph that were still in an early phase . The lower sphingomyelin signals , compared to ceramides , suggests that only part of the ceramide population was readily converted to sphingomyelin . Sphingolipid metabolism is a complex process that involves a number of proteins and enzymes , and we are still in an early stage of our understanding . It has been suggested that sphingolipids serve distinct functions depending on their subcellular localizations , but it is technically difficult to address this issue . Our data provide direct evidence that subcellular localization is important for sphingolipid metabolism . The rapid turnover also suggests that lipid metabolism studies require techniques with high temporal resolution . Recently , Höglinger et al . reported that uncaging sphingosine from Sph-Cou causes an acute release of calcium from acidic stores via the two-pore channel 1 ( Höglinger et al . , 2015 ) . To investigate whether subcellular localization of sphingosine is relevant for this signaling event , we tested the Sph-Cou and Mito-So probes in live Hela cells using a ratiometric calcium dye ( Fluo-4 ) as readout . As shown here , global uncaging of sphingosine quickly induced calcium release as previously reported , whereas mitochondria specific uncaging failed to trigger any calcium mobilization in the saame time frame ( Figure 8 , Figure 8—figure supplement 1 ) . To verify that this is not due to a difference in total cellular amounts of sphingosine , we quantified the sphingosine levels generated from the two probes extracted from cells . Since the calcium curves were obtained from single-cell analysis which does not provide quantitative information of photo-released sphingosine , we incubated the two probes in culture dishes , extracted lipids , performed uncaging in the lipid suspension , and measured sphingosine levels by mass spectrometry ( same protocol as in Figure 4—figure supplement 3 ) . The amount of sphingosine found after uncaging was approximately two times higher for Mito-So than for Sph-Cou ( Figure 8—figure supplement 2 ) , showing that differences in the amount of probe taken up by the cells is not the explanation for the different physiological consequences . Our data thus provide direct evidence that the intracellular sphingoid base compartmentalization can be a deciding factor in the regulation of intracellular signal transduction . It is clear that glycerophospholipids and sphingolipids are unequally distributed among the subcellular structures of the cell , however , the transport of lipids is very rapid , with phosphatidylethanolamine being transported from its site of the synthesis to the plasma membrane in less than a couple of minutes ( Sleight and Pagano , 1983 ) . This rapid movement raises the question of the influence of localization on metabolism and signaling . To address this , we have developed a highly efficient photo-cleavable probe ( TPP-Cou-OH ) that was easily installed on the amino group of free sphingoid bases , and which can be potentially applied to mask other functional groups , allowing the caging of a wide variety of lipids and other bioactive molecules . Carrying a hydrophobic and positively charged triphenylphosphonium ( TPP ) cation , the caged probes accumulated in mitochondria , into which the entry is tightly regulated . As these probes are also fluorescent dyes , their subcellular localization could be conveniently visualized under fluorescence microscopy . With few exceptions ( Nadler et al . , 2015; Chalmers et al . , 2012 ) , most available caged molecules are randomly distributed , and thus spatiotemporal control has to rely on sophisticated microscopy at a single-cell level . In contrast , our caged probes can be used for single-cell microscopic analysis as well as biochemical experiments to follow metabolism . In the recent years , mass-spectrometry ( MS ) -based lipidomic approaches have emerged as powerful tools that provide comprehensive and quantitative analysis of cellular lipid profiles ( da Silveira Dos Santos et al . , 2014; Han , 2016 ) . Combining this approach with our caged probes produced a robust system that enabled us to monitor the local sphingosine metabolism in living cells . We observed that , upon uncaging , some liberated sphingosine was rapidly converted into S1P . Suppressing sphingosine kinase activity , either by genetic ablation or chemical inhibitors , largely abolished S1P synthesis , confirming that phosphorylation was driven by sphingosine kinases . Importantly , uncaging sphingosine in purified mitochondria also generated a significant amount of S1P , providing strong evidence that mito-released sphingosine was phosphorylated inside mitochondria , consistent with previous studies that claimed SPHKs are detected in the mitochondria ( Strub et al . , 2011 ) . It has been known that ceramide is pro-apoptotic , whereas S1P promotes cell growth and proliferation . But the functions of sphingosine , which can be metabolized into ceramide and S1P , are not well understood . Since mitochondria is major a hub of energy supply and has been highly associated with apoptosis , it is interesting to investigate the metabolism of mitochondrial sphingosine , and to compare this with the metabolism globally distributed sphingosine . Using the mitochondria-specific caged probe ( Mito-So ) and globally caged probe ( Sph-Cou ) , we monitored the sphingosine level over time , from which we derived the half-lives of decay by fitting a one phase function . Our results indicate that mitochondria released sphingosine was quickly lost over time ( t1/2 = 2 . 63 min , Figure 6C ) , whereas globally released sphingosine was slowly lost and did not come back to the basal level ( t1/2 = 4 . 21 min , Figure 6E ) . Since ceramides are much more abundant lipids than sphingoid bases in cells ( ( Simanshu et al . , 2013; Shaner et al . , 2009; Canals et al . , 2010 ) , Figure 3—figure supplement 1 ) , we prepared stable isotope labelled caged sphingosine so that the newly produced ceramides can be distinguished from the endogenous pool of ceramides by mass spectrometry . After releasing sphingosine at different subcellular localizations , we have found that sphingosine can be metabolized through distinct metabolic patterns , depending on where free sphingosine is supplied . Focusing on two most abundant ceramides in mammalian cells , our data show that both C16 and C24 ceramides were continuously generated after Mito-So uncaging . Nevertheless , after Sph-Cou uncaging , both lipids were produced and also quickly consumed after 5 to 10 min . Since ceramide synthases ( CerS ) , which catalyze the conversion of sphingosine into ceramide , are mainly located in endoplasmic reticulum ( ER ) , the lack of substrate accessibility to CerS may prevent immediate ceramide formation from liberated sphingosine in mitochondria . It is likely that the slow sphingosine exit from mitochondria resulted in the continuous increase of ceramides . On the other hand , globally uncaged sphingosine is easily accessible to CerS , which could explain the rapid ceramide production and degradation . We did not detect a preferential synthesis of ceramides with specific acyl chain length dependent upon the localized source of sphingosine . Interestingly , we detected less ceramides from globally released sphingosine . One possibility is that mitochondrial sphingosine is more efficiently , but less rapidly transferred to the ER , ( Figure 8—figure supplement 2 ) the site of ceramide production . It is clearly shown from our data that sphingosine produced in the mitochondria can enter the sphingolipid biosynthetic pathway , demonstrating that there is a pathway to transfer sphingosine from the mitochondrial matrix to the ER . Overall , the data provided here provide direct evidence that sphingosine metabolism is different depending upon its subcellular localization . To provide further evidence that sphingosine localization is important for signaling processes , we investigated the impact of mitochondria-specific uncaging in lipid signaling studies . We demonstrated a role of compartmentalization in restricting sphingosine signaling for calcium release . Again , sphingosine produced in the mitochondria was clearly not equivalent to sphingosine produced elsewhere in the cell . This clearly shows that the mitochondrial sphingosine does not freely and rapidly diffuse in the cells because it would have produced a release of calcium . Why does it not produce a calcium release ? Two factors might influence this . One , it could be that the exit from mitochondria is too slow to be seen in our time course . Second , it could be that the mitochondrial sphingosine is transported preferentially to the ER . This would be consistent with the increased efficiency of ceramide synthesis from mitochondrial sphingosine compared to global sphingosine . It is possible that externally added sphingosine is also targeted preferentially to the ER as it behaves similarly to mitochondrial sphingosine . One interesting question for the future is to determine where the sphingosine must be located to induce a calcium release from acidic stores . A similar approach as taken here that localizes caged-sphingosine to different intracellular locations could be used to address this question and determine if the sphingosine must be inside or outside of the lysosome . Furthermore , extending our concept of subcellular release of caged lipids to other compartments and other lipids should allow experimentation to understand the role of localization in lipid metabolism and signaling in general . Lipid messengers play active roles in essential cellular functions and their metabolism is tightly regulated through a complex lipid metabolic network . Although it is believed that sphingolipids could have different functions depending on their subcellular locations , it is difficult to acquire direct evidence because of technical challenges to spatiotemporally control bioactive lipids . Our approach offers a novel technique that permits the elucidation of the relevance of lipid localization in bioactive lipid metabolism and signaling . Combining this technique with mass spectrometry ( MS ) -based lipidomic approach we were able to monitor real-time metabolism of a key lipid regulator in mitochondria . More importantly , our combined results provide direct evidence that sphingosine has distinct metabolism and signaling patterns depending on subcellular localization . We envision that site-specific photochemical probes will become valuable tools in lipid signaling and metabolism studies . Unless otherwise stated , chemicals and reagents were purchased from commercial sources ( Sigma-Aldrich , Acros , Alfa-Aesar ) and were used without purification . Sphingosine was purchased from Echelon Bioscience . Sphingosine-d7 was purchased from Avanti Polar Lipids . MitoTracker Orange CMTMRos was a gift from Jean Gruenberg lab ( Department of Biochemistry , University of Geneva ) originally purchased from Thermo Fisher Scientific . Fluo-4 AM was purchased from Thermo Fisher Scientific . Sph-Cou was synthesized according to previously described protocols ( Höglinger et al . , 2015 ) . Deuterated solvents were obtained from Cambridge Isotope Laboratories , Inc . The plasmid pX330 was deposited to Addgene ( plasmid #42230 ) by Feng Zhang ( Broad Institute ) . All cells were cultured at 37°C and 5% CO2 in Dulbecco’s Modified Eagle Medium ( DMEM , Invitrogen ) with 4 . 5 g/L glucose , supplemented with 10% fetal calf serum ( FCS , Hyclone ) and 1% Pen/Strep ( Gibco ) . Cell numbers were quantified by Countess II Automated Cell Counter ( Invitrogen ) following manufacturerâ€™s protocol . For fluorescence microscopy experiments , cells were cultured in 35 mm glass bottom MatTek dishes to reach 80% confluency . In the subcellular localization experiments , cells were incubated with 5 μM Mito-So or Mito-Sa , together with 50 MitoTracker Orange in 1 mL imaging buffer ( 20 mM HEPES , 115 mM NaCl , 1 . 2 mM MgCl2 , 1 . 2 mM K2HPO4 , 1 . 8 mM CaCl2 and 0 . 2% glucose , pH 7 . 40 ) for 15 min at 37°C before replacing with new imaging buffer . In the calcium mobilization experiments , cells were incubated with blank , 5 μM Mito-So , or Sph-Cou , respectively , in the presence of 5 μM Fluo-4 AM in 1 mL imaging buffer for 15 min at 37 before replacing with new imaging buffer . Mouse livers were dissected , rinsed with ice-cold MB buffer ( Mannitol 210 mM , Sucrose 70 mM , HEPES 10 mM , EDTA 1 mM , pH 7 . 5 ) and homogenized in 10 ml of MB buffer by twenty passages in a glass homogenizer . The resultant homogenate ( liver homogenate ) was centrifuged at 1500 g for 5 min at 4°C to remove nuclei and unbroken cells . The supernatant was further centrifuged twice at 6000 g for 10 min to provide a mitochondria-enriched pellet . Mice were euthanized by CO2 inhalation . All experimental procedures were performed according to guidelines provided by the Animal Welfare Act and Animal welfare ordinance , the Rectors’ Conference of the Swiss Universities ( CRUS ) policy and the Swiss Academy of Medical Sciences/Swiss Academy of Sciences’ Ethical Principles and Guidelines for Experiments on Animals , and were approved by the Geneva Cantonal Veterinary Authority ( authorization number: 28038/GE86/16 ) . Experiments were performed using a 1000 Watt Arc Lamp Source ( #66924 , NewPort ) equipped with a dichromic mirror ( 350–450 nm , #66226 ) . For live-cell uncaging , cells were seeded and cultured in 60 mm dishes until full confluency . The dishes were loaded with caged probes in 2 mL imaging buffer for 15 min at 37°C . After replacing with new imaging buffer , cells were placed on ice under the lamp at a distance of 20 cm , and irradiated 120 s at 1000 Watt . Cells were either placed back in a 37°C incubator , and/or immediately processed for lipid extraction . For the uncaging experiments in purified mitochondria , mitochondria in MB buffer was re-suspended in KCl buffer ( 125 mM KCl , 0 . 5 mM EGTA , 4 mM MgCl2 , 5 mM K2HPO4 , 10 mM HEPES , pH 7 . 4 ) containing 5 mM ATP . Mito-So was added and the suspension was incubated at 37°C for 10 min , irradiated on ice for 2 min , re-incubated at 37°C for 15 min , centrifuged at 14 , 000 rpm for 5 min prior to lipid extraction . Lipids were extracted following previously described protocols with minor modifications ( Guan et al . , 2009; Loizides-Mangold et al . , 2012; da Silveira Dos Santos et al . , 2014 ) . Briefly , cells were washed with cold PBS and scraped off in 500 μl cold PBS on ice . The suspension was transferred to a 1 . 5 ml Eppendorf tube in which it was spin down at 2500 rpm for 5 min at 4 . After taking off the PBS , samples were stored at −20°C or directly used for further extraction . For sphingoid base analysis , samples were re-suspended in 150 uL extraction buffer ( ethanol , water , diethyl ether , pyridine , and 4 . 2 N ammonium hydroxide ( 15:15:5:1:0 . 018 , v/v ) ) . A mixture of internal standards ( 0 . 04 nmol of C17 sphingosine , 0 . 04 nmol of C17 sphinganine , 0 . 4 nmol of C17 sphingosine-1-phosphate , 0 . 4 nmol of C17 sphinganine-1-phosphate ) was added . The samples were vigorously vortexed using a Cell Disruptor Homogenizer ( Disruptor Genie , Scientific Industries ) for 10 min at 4°C and incubated on ice for 20 min . Cell debris were pelleted by centrifugation at 14 , 000 rpm for 2 min at 4°C , and the supernatant was collected . The extraction was repeated once more without ice incubation . The supernatants were combined and dried under vacuum in a CentriVap ( Labconco , Kansas City , MO ) . The samples were re-suspended in a mixture of solvents composed of 70 μl of borate buffer ( 200 mM boric acid pH 8 . 8 , 10 mM tris ( 2-carboxyethyl ) -phosphine , 10 mM ascorbic acid and 33 . 7 μM 15N13C-valine ) , and 10 μl of formic acid solution ( 0 . 1% aqueous solution ) , derivatized by reacting with 20 μl 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate ( AQC ) solution ( 2 . 85 mg/ml in acetonitrile ) for 15 min at 55°C . After overnight incubation at 24°C , samples were analyzed by LC-MS/MS in an Accela HPLC system ( ThermoFisher Scientific , Waltham , USA ) coupled to a TSQ Vantage ( ThermoFisher Scientific , Waltham , USA ) . MRM-MS was used to identify and quantify sphingoid bases . The amounts of sphingolipids were normalized with respect to the amount of C17 internal standards and cell numbers . For ceramide and phospholipid analysis , samples were prepared following the MTBE protocol ( Matyash et al . , 2008 ) . Briefly , cells were re-suspended in 100 μL of water and transferred into a 2 ml Eppendorf tube . 360 μl of MeOH and a mixture of internal standards ( 0 . 4 nmol of DLPC , 1 nmol of PE31:1 , 1 nmol of PI31:1 , 3 . 3 nmol of PS31:1 , 2 . 500 nmol of C12 sphingomyelin , 0 . 5 nmol of C17 ceramide and 0 . 1 nmol of C8 glucosylceramide ) was added . Samples were vortexed , following the addition of 1 . 2 ml of MTBE . The samples were vigorously vortexed at maximum speed for 10 min at 4°C and incubated for 1 hr at room temperature on a shaker . Phase separation was induced by addition of 200 μL MS-grade water and incubation for 10 min . Samples were centrifuged at 1000 g for 10 min . The upper phase was transferred into a 13 mm glass tube and the lower phase was re-extracted with 400 μl of a MTBE/MeOH/H2O mixture ( 10:3:1 . 5 , v/v ) . The extraction was repeated one more time . The combined upper phase was separated into three equal aliquots before drying under nitrogen flow . One aliquot was treated by alkaline hydrolysis to enrich for sphingolipids , one was used for glycerophospholipid analysis and the third was kept as a backup . To deacylate glycerophospholipids , the sample was re-suspended in 1 ml freshly prepared monomethylamine reagent ( methylamine/H2O/n-butanol/methanol at 5:3:1:4 ( v/v ) ) and incubated at 53 for 1 hr in a water bath before dried by nitrogen flow . The excess salts were removed by extracting samples in 300 μL water-saturated n-butanol solution and 150 μL MS-grade water ( Clarke and Dawson , 1981 ) . The organic phase was collected , and the extraction was repeated with 300 μL water-saturated n-butanol . The combined organic phase was dried by nitrogen flow . Identification and quantification of phospholipid and sphingolipid molecular species were performed using multiple reaction monitoring with a TSQ Vantage Triple Stage Quadrupole Mass Spectrometer ( Thermo Scientific ) equipped with a robotic nanoflow ion source , Nanomate HD ( Advion Biosciences ) . Each individual ion dissociation pathway was optimized with regard to collision energy . Lipid concentrations were calculated with respect to the corresponding internal standards and were presented as percentage of all lipid signals ( Guan et al . , 2013 ) . Subcellular localization experiments were performed on a Leica SP8 confocal microscope using a 63 x oil immersion objective . A 405 nm and 532 nm laser were used with appropriate filter settings during image acquisition . Photoactivation experiments were performed using a Nikon A1r microscope with 63 x oil immersion objective at 37 with 5% CO2 . A 100 mW 402 nm laser was used for photo-releasing sphingosine , and a 488 nm laser was used for recording the time-lapse images . Specifically , uncaging was performed within a circle in 3 . 0 μm diameter . Cells were irradiated for 4 s ( 2 . 2 μs/pixel ) at 100% laser power in the region of interest ( mitochondria in the case of Mito-So uncaging ) . Raw imaging data were analyzed by measuring the mean intensity of laser-illuminated cells . The fluorescence images were analyzed by Fiji software ( Schindelin et al . , 2012 ) . The fluorescence staining images were presented in their original form . Time-lapse images were extracted by measuring the mean intensities of photo-stimulated individual cells during the acquisition . The single-cell traces were normalized to baseline of each cell before exporting to GraphPad Prism . Data represent at least the average of three independent experiments . Error bars represent standard error of the mean ( SEM ) as indicated . Statistical significance was calculated based on two-tailed unpaired student's t-test . Mutant cells were generated by the CRISPR/Cas9 system from Streptococcus pyogenes ( Ran et al . , 2013 ) , using the HPRT co-targeting strategy ( Liao et al . , 2015 ) as previously described ( Harayama and Riezman , 2017 ) . Target sequences ( listed below ) were selected based on high specificity and efficacy scores predicted by the CRISPOR algorithm ( Haeussler et al . , 2016 ) , and the corresponding pairs of oligo DNA were synthesized ( Microsynth AG , Balgach , St . Gallen , Switzerland ) . Plasmids were constructed by assembling annealed oligo DNA pairs and plasmids in single tube reactions using FastDigest Bpi I ( Thermo Fisher Scientific ) and quick ligase ( New England Biolabs , Ipswich , MA , USA ) in quick ligase buffer . Three cycles of restriction at 37 and ligation at 25 were repeated ( 5 min for each step ) , followed by Bpi I restriction for one hour to remove empty vectors . pX330 plasmids ( Ran et al . , 2013 ) were used for assembly , except for HPRT guide RNA , for which pUC-U6-sg plasmid ( Harayama and Riezman , 2017 ) was used . Plasmids were transformed into chemical competent STBL3 bacterial cells ( Thermo Fisher Scientific , Waltham , MA , USA ) , sequence-verified by Sanger sequencing ( Fasteris SA , Plan-les-Ouates , Geneva , Switzerland ) , and purified using GenElute Plasmid Miniprep kit ( Sigma-Aldrich , St . Louis , MO , USA ) followed by endotoxin removal by isothermal Triton X-114 extraction ( Ma et al . , 2012 ) . Plasmids ( 98 ng total of plasmids for target ( s ) and 2 ng plasmid for HPRT ) were reverse transfected into HeLa MZ cells using Lipofectamine 3000 ( Thermo Fisher Scientific ) in a 96 well plate , cell culture areas were scaled up before reaching over confluency , and selected with 6 μg/mL 6-thioguanine ( Sigma-Aldrich ) 5 days post-transfection . After one week of selection , the resistance against 6-thioguanine caused by the mutations in the co-targeted HPRT gene led to enrichment for mutations in the target genes ( Liao et al . , 2015 ) . To evaluate mutation rates , individual loci were analyzed by PCR direct sequencing ( using primers listed below ) followed by TIDE ( tracking indels by deconvolution ) analysis ( Brinkman et al . , 2014 ) . PCR reactions were peformed using ExTaq polymerase ( TAKARA Clonthech , Otsu , Shiga , Japan ) . 1H and 13C-NMR spectra were recorded on a Bruker AMX-400 MHz or 500 MHz spectrometer . Chemical shifts are given in ppm ( δ ) using the NMR solvent as internal references and J values are reported in Hz . Splitting patterns are designated as follows: s , singlet; d , doublet; t , triplet; q , quartet; m , multiplet; b , broad . 13C-NMR spectra were broadband hydrogen decoupled . LC-MS spectra were recorded using a Thermo Electron Corporation HPLC with a Thermo Scientific Finnigan Surveyor MSQ Spectrometer System . High-resolution mass spectra were recorded on a QSTAR Pulsar ( QqTOF ) mass spectrometer . Flash chromatography purification was carried out using silica gel 40–63 μm ( 200–400 mesh ) from SiliCycle in solvent systems as described . Thin layer chromatography ( TLC ) was performed on aluminum-backed , pre-coated silica gel plates ( Merck TLC silica gel 60 F254 ) . Spots were detected by a UV lamp under 254 nm or 365 nm wavelength . Reverse phase HPLC purification was performed using an Agilent Technologies 1260 infinity HPLC equipped with a ZORBAX 300 SB-C18 column ( 9 . 4 × 250 mm ) . UV-Vis spectra were recorded using a JASCO V-650 spectrophotometer equipped with a stirrer and a temperature . Fluorescence measurements were performed with a FluoroMax-4 spectrofluorometer ( Horiba Scientific ) equipped with a stirrer and a temperature controller . ( 3-Aminopropyl ) triphenylphosphonium bromide , compound 1 and subsequently methylated product were synthesized according to previously published procedures ( Kaur et al . , 2015; Hagen et al . , 2005 ) , respectively .	Fatty or oily molecules called lipids are essential components of the membranes of cells and important signaling molecules too . They are made in specific compartments of the cell , but most are found in all membranes , albeit in varying amounts . Their widespread distribution suggests that there are extensive networks for transporting lipids within cells . Yet scientists know little about lipid transport inside living cells because it is difficult to detect their movements . Mitochondria are cellular compartments that are often referred to as the “powerhouses of the cell” . Many lipids are found in mitochondria including one called sphingosine , which is a common component of many other cell membranes too . Sphingosine can increase the concentration of calcium ions inside the cells , and when converted to a molecule called sphingosine 1 phosphate it forms a signaling molecule that regulates fundamental processes like cell survival and migration . However , it was not known if sphingosine localized in the mitochondria was processed differently to the same molecule elsewhere in the cell , or if its signaling activity was affected by its location . In the laboratory , Feng et al . synthesized an inactive sphingosine-like molecule that would only localize to mitochondria and which could be activated with a flash of light . By adding this molecule to human cells , they showed that sphingosine could be converted to sphingosine 1 phosphate within the mitochondria , before being exported rapidly to another compartment in the cell . The experiments allowed Feng et al . to observe the process in enough detail to to conclude that , despite its rapid transport , when localized only inside mitochondria , sphingosine could not trigger its normal signaling response . This new light-activated lipid molecule will be a useful tool for many researchers studying both metabolism and signaling . In principle , a similar tool could be developed for many compounds and it should also be possible to localize the compound to different locations within the cell . This new generation of compounds would give scientists a better understanding of mitochondria biology . They could be applied to the study of diseases where the mitochondria do not function as they should , for example Barth syndrome , where a mitochondria specific lipid called cardiolipin is not properly synthesized .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology", "tools", "and", "resources" ]	2018	Mitochondria-specific photoactivation to monitor local sphingosine metabolism and function
Serotonin is implicated in mood and affective disorders . However , growing evidence suggests that a core endogenous role is to promote flexible adaptation to changes in the causal structure of the environment , through behavioral inhibition and enhanced plasticity . We used long-term photometric recordings in mice to study a population of dorsal raphe serotonin neurons , whose activity we could link to normal reversal learning using pharmacogenetics . We found that these neurons are activated by both positive and negative prediction errors , and thus report signals similar to those proposed to promote learning in conditions of uncertainty . Furthermore , by comparing the cue responses of serotonin and dopamine neurons , we found differences in learning rates that could explain the importance of serotonin in inhibiting perseverative responding . Our findings show how the activity patterns of serotonin neurons support a role in cognitive flexibility , and suggest a revised model of dopamine–serotonin opponency with potential clinical implications . Serotonin ( 5-HT ) is classically known to be implicated in mood and affective disorders ( Dayan and Huys , 2009; Cools et al . , 2011; Li et al . , 2012 ) , but it also plays a fundamental role when organisms need to adapt to sudden changes in the causal structure of an environment , such as during extinction and reversal learning paradigms ( Clarke et al . , 2004 , 2007; Boulougouris and Robbins , 2010; Bari et al . , 2010; Brigman et al . , 2010; Berg et al . , 2014 ) . These studies have shown that 5-HT depletion , particularly in the orbitofrontal cortex ( OFC ) of primates , causes perseverative errors , that is , difficulties in stopping responses to previously rewarded stimuli which are no longer reinforced , without affecting learning of new associations or retention of learned associations ( Clarke et al . , 2007 ) . Such results seem to stem from two functions of endogenous 5-HT activation: inhibiting learned responses that are not currently adaptive ( Soubrié , 1986; Bari and Robbins , 2013 ) and driving plasticity to reconfigure them ( Maya Vetencourt et al . , 2008; Jitsuki et al . , 2011; He et al . , 2015 ) . These mirror dual functions of dopamine ( DA ) in invigorating reward-related responses ( Niv et al . , 2007; Panigrahi et al . , 2015 ) and promoting plasticity that reinforces new ones ( Tsai et al . , 2009; Kim et al . , 2012; Steinberg et al . , 2013 ) . However , while DA neurons are known to be activated by reward prediction errors ( Schultz et al . , 1997; Cohen et al . , 2012; Eshel et al . , 2015 ) , consistent with theories of reinforcement learning ( Sutton and Barto , 1998; Schultz et al . , 1997 ) , the reported firing patterns of 5-HT neurons ( Liu et al . , 2014; Cohen et al . , 2015; Li et al . , 2016 ) do not accord with any existing theories ( Daw et al . , 2002; Boureau and Dayan , 2011; Cools et al . , 2011; Nakamura , 2013 ) . Indeed , 5-HT neurons have been proposed to signal worse-than-expected outcomes by being activated by negative reward prediction errors in the reinforcement learning framework ( Daw et al . , 2002; Boureau and Dayan , 2011 ) , but there is little experimental evidence for such a signal in 5-HT neurons ( Cohen et al . , 2015; Hayashi et al . , 2015; Li et al . , 2016 ) and 5-HT activation does not appear to drive aversive learning processes ( Dugué et al . , 2014; Liu et al . , 2014; McDevitt et al . , 2014; Qi et al . , 2014; Miyazaki et al . , 2014; Fonseca et al . , 2015 ) the way DA drives appetitive learning ( Tsai et al . , 2009; Kim et al . , 2012; Steinberg et al . , 2013 ) . To investigate how 5-HT neurons could be involved in cognitive and behavioral flexibility in changing environments , we recorded their activity over several days in mice engaged in a reversal learning task in which the associations between neutral odor cues and different positive and negative outcomes are first well-learned and then suddenly changed . We reasoned that the scarcity of prediction error–like responses in previous recordings of identified 5-HT neurons ( Liu et al . , 2014; Cohen et al . , 2015; Li et al . , 2016 ) or unidentified raphe neurons ( Ranade and Mainen , 2009; Hayashi et al . , 2015 ) might be due to inadequately strong prediction errors . In these studies , the omission of rewards in a small fraction of trials was used to generate prediction errors . While increasing the variability of the outcome , this results in expected uncertainty . In contrast , in a reversal task , there is an abrupt violation of previously stable predictions and a step increase in the frequency of the prediction errors , termed unexpected uncertainty . Expected and unexpected uncertainty may differentially activate neuromodulatory systems ( Yu and Dayan , 2005 ) . We first sought causal evidence that 5-HT neurons were linked to reversal learning in mice engaged in such a task by using a pharmacogenetic approach to silence 5-HT neurons ( Ray et al . , 2011; Teissier et al . , 2015; Armbruster et al . , 2007 ) . Transgenic mice expressing CRE recombinase under the 5-HT transporter promoter ( Gong et al . , 2007 ) ( SERT-Cre , n = 8 ) were transduced with a Cre-dependent adeno-associated ( AAV . Flex ) virus expressing the synthetic receptor Di ( DREADD , hM4D ) ( Armbruster et al . , 2007 ) injected in the dorsal raphe nucleus ( DRN ) , the major source of 5-HT to the forebrain ( Figure 1A ) . These mice and their wild-type littermates ( WT , n = 4 ) were trained in a head-fixed classical conditioning paradigm in which one of four odor cues ( conditioned stimuli , CSs ) was randomly presented in each trial . After a fixed 2 s trace period , each odor was followed by a tone and a specific outcome , or unconditioned stimulus ( US ) ( Figure 1B top ) . For two odors the US was a water reward , and for the other two it was nothing ( that is , only the tone was played ) . After training , mice showed learning of the odor–outcome contingencies , as indicated by differences in the anticipatory lick rate ( Figure 1B bottom ) . 10 . 7554/eLife . 20552 . 003Figure 1 . Inhibition of DRN 5-HT neurons causes perseverative responding . ( A ) Injections of Cre-dependent hM4Di-mCherry ( right ) in the dorsal raphe nucleus ( DRN ) of SERT-Cre mice ( left ) . ( B ) Trial structure of the task ( top ) and mean lick rate of an example session along the four trial types ( bottom ) . ( C ) Reversal procedure ( top ) and example of adaptation in mean anticipatory licking ( baseline lick rate subtracted ) across trials around reversals ( bottom , gray ) , with exponential fits to the reversed odors ( red and black traces ) . Gray shade represents the trials of sessions after CNO injection . ( D ) Mean exponential fits of anticipatory licking for each group of mice after reversal . ( E ) Mean time constants for the groups in ( D ) ( one-way ANOVA , F2 , 19 = 6 . 28 , p=0 . 008 for negative reversal , F2 , 16 = 0 . 34 , p=0 . 715 for positive reversal; multiple comparisons indicated in the figure ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 00310 . 7554/eLife . 20552 . 004Figure 1—figure supplement 1 . Anticipatory licking is more perseverative when DRN 5-HT neurons are inhibited . ( A ) Detailed schematics of the manipulation experiment , including the viral transduction and i . p . injection received by experimental and control animals . ( B ) Exponential fits of anticipatory licking for individual mouse–odor pairs after the reversal ( grouped as a negative or positive reversal ) . ( C ) In negative reversals ( but not in positive reversals or in WT controls receiving CNO in both reversals ) , adaptation in anticipatory licking takes longer when experimental animals receive CNO injection , compared to when they receive vehicle ( one-sample Student’s t-test ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 004 To test the impact of inhibiting DRN SERT-Cre expressing neurons ( hereafter simply ‘5-HT neurons’ ) we used a within-animal cross-over design in which each mouse experienced two reversals ( Figure 1C top ) , receiving the DREADD ligand clozapine-N-oxide ( CNO ) during one and vehicle during the other; WT mice , which always received CNO , served as additional controls ( Figure 1—figure supplement 1A ) . As expected , mice adjusted their anticipatory licking according to the new associations in both reversals ( Figure 1C bottom , gray traces ) . For worse-than-expected outcomes ( negative reversals ) , the kinetics of adaptation to the new contingencies were significantly slower in hM4D mice receiving CNO , compared to hM4D no-CNO controls and WT controls ( Figure 1C , D , E; Figure 1—figure supplement 1B , C ) . In contrast , for better-than-expected outcomes ( positive reversals ) , there was no significant difference between treatment and control groups ( Figure 1D , E ) . This experiment shows that a population of 5-HT neurons in the DRN contributes to inhibiting perseverative responding , suggesting an anatomical and genetic substrate for previous results obtained with pharmacological and lesion experiments ( Clarke et al . , 2004 , 2007; Boulougouris and Robbins , 2010; Bari et al . , 2010; Brigman et al . , 2010 ) . These findings also defined an access point to assess how the net activity of a specific population of 5-HT neurons could account for its effects on reversal learning . To obtain a broad view of DRN 5-HT activity and compare our results to other DRN recording studies ( Hayashi et al . , 2015; Cohen et al . , 2015 ) , for the next series of recording experiments we used a second reversal task in which mice learned to associate four odors with four different outcomes: a large water reward , a small water reward , nothing ( neutral ) and a mild air puff to the eye ( Figure 2A ) . After approximately two weeks of training , mice showed robust CS-triggered anticipatory licking correlated to the reward value of the associated USs ( large water > small water > neutral ≈ air puff ) and eye-blink responses to the delivery of air puffs ( Figure 2B ) . We then reversed the CS–US associations in pairs , such that the CSs associated with the large and small rewards now predicted the air puff and neutral outcomes , respectively , and vice versa ( Figure 2C ) . Upon this reversal , mice experienced strong violations of CS-based expectations ( unexpected uncertainty ) , both positive and negative in value , when the unexpected USs were delivered . Anticipatory licking measurements showed that mice adapted to reversal of contingencies over 1–3 additional sessions ( Figure 2D ) . 10 . 7554/eLife . 20552 . 005Figure 2 . Behavior of head-fixed mice trained in a reversal task . ( A ) Schematics of the trial structure in the classical conditioning task ( before reversal ) with four different outcomes . In each trial , one of four odors was randomly selected and presented for 1 s after a variable foreperiod ( Forep ) . The associated outcome was delivered after a 2 s trace period , together with a tone ( same tone for all trial types ) . Mice were presented with 140 to 346 interleaved trials ( mean ± SD: 223 ± 30 ) per session ( day ) . ( B ) Top: Mean lick rate of SERT-Cre mice in this task ( n = 10 ) along the duration of each trial type . For each mouse , three sessions of the classical conditioning task where initial associations had already been learned were averaged . Bottom: Mean eye movement of SERT-Cre mice ( n = 6 ) along the duration of each trial type . Shaded areas represent s . e . m . ( C ) Reversal of CS–US contingencies ( negative reversal: CS 1 and 2; positive reversal: CS 3 and 4 ) . ( D ) Anticipatory licking ( mean of 500–2800 ms after odor onset , after subtracting the baseline ) across mice for sessions around reversal , showing that the lick rate triggered by the presentation of each odor is adjusted after reversal ( n = 8 , two-way ANOVA with factors day ( days −2 and −1 are considered together ) and mouse , main effect of day: F4 , 2597 = 722 . 14 , p<0 . 001 for odor 1 , F4 , 2554 = 355 . 53 , p<0 . 001 for odor 2 , F4 , 2513 = 104 . 93 , p<0 . 001 for odor 3 , F4 , 2559 = 381 . 55 , p<0 . 001 for odor 4 ) . Colors follow odor identity as in ( A ) . **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 005 To record the population activity of 5-HT neurons across days around the time of the reversal , we used photometry to monitor the activity of these DRN 5-HT neurons through an implanted optical fiber ( Tecuapetla et al . , 2014 ) ( Figure 3A ) . SERT-Cre mice were infected in the DRN using two AAV . Flex viruses containing the genetically-encoded calcium indicator GCaMP6s ( Chen et al . , 2013 ) and the activity-insensitive fluorophore , tdTomato ( Figure 3B , C ) . We verified the specificity of GCaMP6s expression to DRN 5-HT neurons using histological methods ( Figure 3—figure supplement 1 ) . We used a regression-based method to decompose the dual fluorescence signals into a GCaMP6s-specific component , reflecting activity-dependent changes , and a shared component , reflecting general fluorescence changes ( for example , movement artifacts; see Methods and Figure 3—figure supplement 2 ) . We validated the effectiveness of this approach in control mice ( n = 4 ) infected in the DRN with yellow fluorescent protein ( YFP; replacing GCaMP ) and tdTomato ( Figure 3—figure supplements 1 and 2 ) . 10 . 7554/eLife . 20552 . 006Figure 3 . Responses of 5-HT and DA neurons before reversal . ( A ) Fiber photometry with movement artifact correction in head-fixed mice . L: laser; PMT: photomultiplier tube; D . M: dichroic mirror; Ex: excitation; Em: emission; F: filter . ( B ) Cre-dependent fluorophores used . ( C ) Coronal section showing expression of GCaMP6s and tdTomato in the DRN of a SERT-Cre mouse ( scale bar: 200 µm ) . PAG: periaqueductal gray; Aq: Aqueduct . ( D ) Mean responses of 5-HT neurons to the four CSs and USs during an example session of a mouse before reversal . Shaded areas represent 95% confidence interval ( CI ) . ( E ) Coronal section showing expression of GCaMP6s and tdTomato in the ventral tegmental area ( VTA ) of a TH-Cre mouse ( scale bar: 200 µm ) . RLi: rostral linear nucleus of the raphe; RPC: red nucleus , parvicellular part; IPR: interpeduncular nucleus . ( F ) Mean responses of DA neurons to the four CSs and USs during an example session of a mouse before reversal . Shaded areas represent 95% CI . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 00610 . 7554/eLife . 20552 . 007Figure 3—figure supplement 1 . Expression of GCaMP6s and of tdTomato in DRN 5-HT neurons . ( A ) Confocal picture of a coronal slice showing GCaMP6s expression in DRN 5-HT neurons of a SERT-Cre mouse ( scale bar: 100 μm ) . ( B ) Confocal pictures showing DAPI staining , GCaMP6s expression , 5-HT immunoreactivity and overlay ( scale bar: 20 μm ) . ( C ) Schematics of a coronal slice showing the DRN and the area in ( A ) , indicated by a violet rectangle ( top , black scale bar: 1 mm ) . Quantification of specific expression of GCaMP6s in 5-HT neurons ( bottom , specificity: 93 . 2% ± 0 . 3% , infection success: 73 . 3% ± 3 . 5% , mean ± s . e . m , n = 4 mice ) . ( D ) Expression of GCaMP6s and tdTomato in DRN 5-HT neurons ( scale bars: 100 μm , aq: aqueduct ) . ( E ) Expression of YFP and tdTomato in DRN 5-HT neurons ( scale bars: 100 μm ) . ( F ) Total number of cells expressing green fluorophore ( GCaMP6s or YFP ) , tdTomato , or both in SERT-Cre experimental and control mice ( counted from three sections at the center of infection for each mouse , n = 6 GCaMP6s mice , n = 3 YFP mice ) . ( G ) Percentage of green and red cells that express both fluorophores in GCaMP6s and YFP infected mice ( n = 6 GCaMP6s mice , n = 3 YFP mice , no statistical difference was obtained between the percentage of green cells that co-express tdTomato in GCaMP vs YFP mice , nor between the percentage of red cells that co-express a green fluorophore in GCaMP vs YFP mice , Mann-Whitney U test , n . s . for p<0 . 05 ) . ( H ) Anteroposterior location of the center of infection ( circles ) and of the fiber tip location ( crosses ) in the SERT-Cre mice used in behavioral experiments with good histology ( n = 9 for GCaMP6s infection center , n = 8 for GCaMP6s fiber placement , n = 3 for YFP control mice ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 00710 . 7554/eLife . 20552 . 008Figure 3—figure supplement 2 . Linear regression approach to eliminate movement artifacts from neuronal photometric data . ( A ) , ( B ) Surface plots showing raw GCaMP6s ( A ) and tdTomato ( B ) fluorescence signals aligned on the onsets of lick bouts during an example session of a SERT-Cre mouse infected with the corresponding fluorophore . Gray dots represent single licks . Inset in ( A , right ) shows the distribution of inter-lick intervals for this session: lick bouts were defined as sequences of licks separated by no more than 315 ms . The surface plots are aligned from the longer lick bouts at the top to the shorter ones at the bottom ( the last ones are single licks that do not belong to any bout ) . Fluorescence data are shown from 2 s before to 2 s after the bouts . ( C ) Bar plot showing R-squared values comparing GCaMP6s , tdTomato and licking signals , calculated during three imaging sessions and averaged across all SERT-Cre mice ( n = 9 ) . ( D ) , ( H ) Mean licking ( top ) , mean filtered green fluorescence of GCaMP6s ( D ) or YFP ( H ) , and mean filtered tdTomato fluorescence in large water reward trials in an example session of a SERT-Cre mouse transduced with GCaMP6s and tdTomato , and of a SERT-Cre example mouse transduced with YFP and tdTomato . Data in ( D ) belong to the same session represented in ( A , B ) . ( E ) , ( I ) Scatter plot and linear regression between tdTomato and GCaMP6s ( E ) or YFP ( I ) signals for the same sessions as before . tdTomato signals were often bimodal , and well fit as a sum of two Gaussians ( top; orange dotted curves , individual Gaussians; black curve , their sum ) . In order to avoid fitting noise , only data from the Gaussian not centered at zero were included in the linear regression calculation ( black dots to the left of the vertical line ) . Orange lines indicate regression curves . ( F ) , ( J ) Mean filtered raw GCaMP6s ( F ) or YFP ( J ) signals and corresponding artifact predictions ( calculated using linear regression as shown in E and I ) . ( G ) , ( K ) Corrected fluorescence signal obtained after the subtraction of the filtered raw signal by the artifact signal presented in ( F ) and ( J ) . While a calcium signal is still visible in the mouse infected with GCaMP6s , no signal is observed in the YFP control . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 00810 . 7554/eLife . 20552 . 009Figure 3—figure supplement 3 . Responses of DRN 5-HT neurons to odor cues and to predicted outcomes . ( A ) Left: Mean ( ± s . e . m ) CS responses ( corrected fluorescence ) of all SERT-Cre mice expressing GCaMP6s ( n = 9 ) to the four odors ( CSs ) , after having learned the CS-US associations; Middle: quantification of the mean ( ± s . e . m ) CS response amplitude ( in a 1 . 5 s period from odor onset ) across mice ( three-way ANOVA with factors day , mouse and trial type , main effect of trial type F3 , 94 = 16 . 22 , p<0 . 001 , only statistically significant effects of trial type after post hoc correction for multiple comparisons are shown: vertical ticks signal events being compared , horizontal bars without vertical ticks below them are used for grouping non-statistically different events ) ; Right: z-scores of mean response amplitude for individual mice ( gray dots , n = 9 mice ) and their mean ( ± s . e . m . , black dots ) . ( B ) Same as ( A ) but for US responses of SERT-Cre mice ( F3 , 94 = 12 . 41 , p<0 . 001 ) . The predicted large reward US responses of 5-HT neurons have a slightly negative value because of the response that the corresponding CS triggered before , and which takes longer to go back to baseline than the 2 s trace period . p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 00910 . 7554/eLife . 20552 . 010Figure 3—figure supplement 4 . Fluorescence changes to odor cues and to predicted outcomes in YFP control mice . ( A ) Left: Mean ( ± s . e . m . ) CS responses ( corrected fluorescence ) of all SERT-Cre mice infected with YFP instead of GCaMP6s ( n = 4 ) to the four odors after having learned the CS–US associations; Right: quantification of the mean ( ± s . e . m . ) CS response amplitude ( in a 1 . 5 s period from odor onset ) across mice ( three-way ANOVA with factors day , mouse and trial type , main effect of trial type F3 , 39 = 0 . 42 , p=0 . 7426 ) . ( B ) Same as ( A ) , but for US responses of YFP-expressing SERT-Cre mice ( F3 , 39 = 4 . 85 , p=0 . 0058 ) . p<0 . 05 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 01010 . 7554/eLife . 20552 . 011Figure 3—figure supplement 5 . Responses of midbrain DA neurons before reversal . ( A ) Confocal picture of a coronal slice showing GCaMP6s expression in VTA DA neurons of a TH-Cre mouse ( left , scale bar: 100 μm ) ; close up showing DAPI staining , GCaMP6s expression , TH immunoreactivity and overlay ( middle , scale bar: 20 μm ) ; schematics of a coronal slice showing the VTA and the area on the left signaled by a violet rectangle ( scale bar: 1 mm ) with quantification of specific expression of GCaMP6s in DA neurons in the infection center ( right , specificity: 89 . 9% ± 2 . 5% , infection success: 86 . 7% ± 2 . 5% , mean ± s . e . m , n = 4 mice ) . ( B ) Left: Mean ( ± s . e . m ) CS responses ( corrected fluorescence ) of all TH-Cre mice expressing GCaMP6s ( n = 3 ) to the four odors ( CSs ) , after having learned the CS-US associations; Middle: quantification of the mean ( ± s . e . m ) CS response amplitude ( in a 1 . 5 s period from odor onset ) across mice ( three-way ANOVA with factors day , mouse and trial type , main effect of trial type F3 , 28 = 52 . 11 p<0 . 001 , only statistically significant effects of trial type after post hoc correction for multiple comparisons are shown: vertical ticks signal events being compared , horizontal bars without vertical ticks below them are used for grouping non-statistically different events ) ; Right: z-scores of mean response amplitude for individual mice ( gray dots , n = 3 mice ) and their mean ( ± s . e . m . , black dots ) . ( C ) Same as ( B ) , but for US responses of TH-Cre mice ( F3 , 28 = 1 . 05 , p=0 . 386 ) . p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 011 Before reversal , photometric 5-HT responses were similar to previous electrical ( Liu et al . , 2014; Cohen et al . , 2015 ) and photometric ( Li et al . , 2016 ) recordings of identified 5-HT neurons: 5-HT neurons were activated by reward-predicting CSs and air puffs ( Figure 3D , Figure 3—figure supplement 3 ) . YFP control mice implanted and recorded in the same manner showed no photometric responses to these events ( Figure 3—figure supplement 4 ) . To compare directly how DA neurons respond in the same paradigm , we infected TH-Cre mice and targeted neurons in either the posterior lateral ventral tegmental nucleus ( VTA ) or the substantia nigra pars compacta ( SNc ) ( Figure 3E , Figure 3—figure supplement 5 ) . DA photometry responses in these two areas were similar and were therefore combined . As expected , DA neurons were activated by reward-predicting cues , and showed small responses to predicted rewards ( Figure 3F ) . To understand the pattern of 5-HT neural activity that could underlie adaptation to reversal of contingencies , we first analyzed US responses , which could contribute to or modulate reinforcement learning . In general , we found that the abrupt reversal of cue–outcome associations caused immediate changes in 5-HT and DA US responses , much more so than in reward omission tests ( Ranade and Mainen , 2009; Cohen et al . , 2015; Hayashi et al . , 2015; Li et al . , 2016 ) , consistent with sensitivity to the sudden increase in uncertainty that occurred upon reversal after extensive training . We first examined the case of positive reversals . 5-HT neurons showed little or no response to large water rewards before reversal when they were predicted by the preceding CS , but responded robustly to the same events when they were unpredicted , after reversal ( Figure 4A , B ) . Thus , 5-HT neurons showed an excitatory response to a better-than-expected outcome , or positive reward prediction error ( RPE ) . The response to the small reward was also modulated by reward expectation ( Figure 4—figure supplement 1 ) , although to a lesser degree , perhaps due to the presence of a small response even after extensive training ( Figure 3—figure supplement 3B ) . Like 5-HT neurons , DA neurons also showed stronger excitatory responses to water rewards immediately after reversal when they violated cue-based predictions , as opposed to before reversal when they occurred as predicted ( Figure 4C , Figure 4—figure supplement 1C ) . Therefore , both 5-HT and DA neurons showed an increase in activity in response to positive RPEs , and both showed a larger response for the larger magnitude RPE . 10 . 7554/eLife . 20552 . 012Figure 4 . US responses of 5-HT and DA neurons to the large reward during reversal . ( A ) Schematic of the reversal procedure following the large reward US . ( B ) Top: Mean large reward US responses of an example mouse ( SERT1 ) across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean large reward response amplitude ( z-scored across days ) : gray dots represent individual mice ( n = 8 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day is F4 , 2592 = 31 . 47 p<0 . 001; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( C ) Same as ( B ) for DA neurons ( n = 3 mice ) : F4 , 853 = 32 . 46 , p<0 . 001 . p<0 . 05 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 01210 . 7554/eLife . 20552 . 013Figure 4—figure supplement 1 . US responses of 5-HT and DA neurons to small reward during reversal . ( A ) Schematic of the reversal procedure following the small reward US . ( B ) Top: Mean small reward US responses of an example mouse ( SERT3 ) across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean small reward response amplitude ( z-scored across days ) : gray dots represent individual mice ( n = 8 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day F4 , 2532 = 5 . 98 , p<0 . 001; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( C ) Same as ( B ) for DA neurons ( n = 3 mice ) : F4 , 911 = 8 . 52 , p<0 . 001 . p<0 . 05 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 013 We next examined the response to the neutral USs . Before reversal , this US elicited little response from either 5-HT or DA neurons . After reversal , the neutral US was presented when a small water reward was predicted . Therefore , it represented a reward omission or negative RPE . Interestingly , 5-HT neurons showed a robust excitatory response to the neutral US after reversal ( Figure 5B ) . In contrast , DA neurons showed an inhibitory response to the same event ( Figure 5C ) . 10 . 7554/eLife . 20552 . 014Figure 5 . US responses of 5-HT and DA neurons to neutral outcome during reversal . ( A ) Schematic of the reversal procedure following neutral US . ( B ) Top: Mean neutral US responses of an example mouse ( SERT1 ) across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean neutral response amplitude ( z-scored across days ) : gray dots represent individual mice ( n = 8 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day F4 , 2535 = 10 . 71 , p<0 . 001; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( C ) Same as ( B ) for DA neurons ( n = 3 mice ) : F4 , 843 = 4 . 54 , p=0 . 001 . p<0 . 05 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 01410 . 7554/eLife . 20552 . 015Figure 5—figure supplement 1 . US responses of 5-HT and DA neurons to air puff during reversal . ( A ) Schematic of the reversal procedure following the air puff US . ( B ) Top: Mean air puff US responses of an example mouse ( SERT1 ) across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean air puff response amplitude ( z-scored across days ) : gray dots represent individual mice ( n = 8 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day F4 , 2564 = 2 . 55 , p=0 . 037; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( C ) Same as ( B ) for DA neurons ( n = 3 mice ) : F4 , 881 = 5 . 21 , p<0 . 001 . p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 015 Taking the neutral and rewarding USs together , the results show that midbrain DA neurons respond to positive and negative RPEs with modulation of the opposite sign , as reported previously in reward omission paradigms ( Cohen et al . , 2012; Schultz et al . , 1997 ) ; but see Matsumoto and Hikosaka ( 2009 ) ; Lammel et al . ( 2011 ) ; Kim et al . ( 2016 ) ; Matsumoto et al . , 2016 ) . On the other hand , SERT-positive DRN 5-HT neurons show excitatory responses to both positive and negative RPEs . Thus , DRN 5-HT responses to rewards and reward omissions resemble an ‘unsigned RPE’ or ‘surprise’ signal ( see Discussion ) . Finally , we examined the response of 5-HT and DA neurons to predicted and unpredicted air puffs . In contrast to other USs , DRN 5-HT neurons were mildly activated by air puff USs , even after extensive training ( Figure 3; Figure 3—figure supplement 3B ) . Upon reversal , despite the fact that the air puff US now represented a large negative RPE ( since the large water reward was predicted ) , 5-HT neurons showed no significant response ( Figure 5—figure supplement 1B ) . Midbrain DA neurons , on the other hand , showed no response to the air puff US after training , but showed a small but significant inhibitory response after reversal ( Figure 5—figure supplement 1C ) . The results for all USs are summarized in Figure 6 . Overall , midbrain DA responses adhered closely to the model of a ‘signed RPE’ , including for the air puff , whereas the DRN 5-HT neurons resembled an ‘unsigned RPE’ with respect to rewards and reward omissions , but they diverged from this model for air puff responses ( see Discussion for further interpretation ) . Thus , 5-HT and DA neurons are both sensitive to violations of expectation that occur during an abrupt reversal , with the two systems responding in the same way to better-than-expected outcomes but in opposite ways to worse-than-expected outcomes . 10 . 7554/eLife . 20552 . 016Figure 6 . Responses of 5-HT and DA neurons to outcomes are differentially modulated by expectations . ( A ) Mean ( ± s . e . m . ) response of 5-HT neurons , across mice , to the four USs before ( day −1 , filled bars ) and right after ( day 0 , open bars ) reversal ( n = 8 mice , two-way ANOVA with factors mouse and day , the main effect of day F1 , 764 = 84 . 36 , p<0 . 001 for large reward , F1 , 748 = 3 . 49 , p=0 . 062 for small reward , F1 , 756 = 38 . 17 , p<0 . 001 for neutral , F1 , 766 = 2 . 79 , p=0 . 095 for air puff ) . ( B ) Same as ( A ) for midbrain DA neurons ( n = 3 mice , F1 , 249 = 67 . 9 , p<0 . 001 for large reward , F1 , 277 = 8 . 49 , p=0 . 004 for small reward , F1 , 278 = 10 . 95 , p=0 . 001 for neutral , F1 , 250 = 12 . 74 , p<0 . 001 for air puff . p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 016 To further investigate the idea that 5-HT neurons might report prediction errors , we examined responses to USs delivered outside of the normal context . For this , five days after reversal , on a small fraction ( 20% ) of trials , a randomly-selected US was delivered at the time that a CS was normally presented ( Figure 7A ) . We found that water rewards produced larger 5-HT responses when they were presented in this way , compared to when preceded by a well-learned cue ( Figure 7B ) . Of particular interest was that even neutral tones produced an excitatory response when an odor was expected ( Figure 7B; Figure 7—figure supplement 1 ) . Therefore , 5-HT neurons were activated by the substitution of one neutral stimulus with another . DA neurons also responded strongly to uncued rewards , as previously reported ( Schultz et al . , 1997; Cohen et al . , 2012 ) , but little to other uncued USs ( Figure 7C ) ( Matsumoto et al . , 2016 ) . Thus , consistent with the responses following CS–US reversal , this experiment also showed that , with respect to water rewards and reward omissions , 5-HT neurons respond in the same manner to unexpected events , whether negative , neutral or positive , whereas DA neurons are primarily sensitive to unexpected events that have some reward value . 10 . 7554/eLife . 20552 . 017Figure 7 . DRN 5-HT neurons respond more to uncued outcomes . ( A ) Behavioral task diagram . ( B ) Mean ( ± s . e . m . ) response of DRN 5-HT neurons across mice to the four USs when they are predicted ( filled bars ) and when they are unpredicted ( open bars ) ( n = 4 mice , two-way ANOVA with factors type ( predicted or unpredicted ) and mouse , the main effect of type: large reward F1 , 923 = 45 . 17 , p<0 . 001 , small reward F1 , 944 = 8 . 42 , p=0 . 0038 , neutral F1 , 924 = 5 . 36 , p=0 . 0208 , air puff F1 , 924 = 0 . 61 , p=0 . 4331 ) . ( C ) Same as ( B ) but for midbrain DA neurons ( n = 3 mice , large reward F1 , 642 = 175 . 05 , p<0 . 001 , small reward F1 , 589 = 17 . 53 , p<0 . 001 , neutral F1 , 673 = 0 . 52 , p=0 . 4707 , air puff F1 , 601 = 0 . 34 , p=0 . 5598 ) . p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 01710 . 7554/eLife . 20552 . 018Figure 7—figure supplement 1 . Responses of 5-HT and DA neurons to predicted and unpredicted outcomes . ( A ) Mean US responses of SERT-Cre mice ( n = 4 ) to predicted ( black ) and unpredicted ( red ) outcomes ( shaded areas represent s . e . m . ) . ( B ) Same as ( A ) for TH-Cre mice . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 018 US responses are appropriate to drive learning across trials , but occur too late within a given trial to inhibit CS-driven behavioral responses directly . If it is to intervene in time to prevent a response , behavioral inhibition should be triggered by predictive CS cues . We therefore examined the CS responses of 5-HT and DA neurons carefully , to test how they might contribute to reversal learning . Before reversal , both 5-HT and DA neurons showed CS responses that correlated with the relative value of the US predicted by the CS ( large reward > small reward > neutral ≈ air puff ) ( Figure 3 , Figure 3—figure supplements 3 and 5 ) . After the reversal , both adjusted to the new contingencies such that , by three days post-reversal , the CS responses reflected their new US associations ( Figure 8 ) . Thus , despite small differences in their relative magnitudes , and in contrast to their distinct US responses , DA and 5-HT neurons showed CS responses that were remarkably similar , both before and after reversal learning . If DA and 5-HT have opposing direct effects on behavior ( for example , Cools et al . , 2011 ) , these results suggest that they would simply cancel one another out . 10 . 7554/eLife . 20552 . 019Figure 8 . 5-HT and DA CS responses are relearned after the reversal . ( A ) Mean ( ± s . e . m . ) response of 5-HT neurons across mice to the four CSs before reversal ( filled bars ) and after adaptation to the reversed contingencies ( open bars ) ( n = 8 mice , two-way ANOVA with factors day and mouse , the main effect of day: large reward F1 , 906 = 17 . 35 , p<0 . 001 , small reward F1 , 902 = 14 . 87 , p<0 . 001 , neutral F1 , 882 = 0 . 13 , p=0 . 72 , air puff F1 , 914 = 17 . 12 , p<0 . 001 ) . ( B ) Same as ( A ) for midbrain DA neurons ( n = 3 mice , large reward F1 , 294 = 15 . 35 , p<0 . 001 , small reward F1 , 336 = 71 . 72 , p<0 . 001 , neutral F1 , 282 = 3 . 45 , p=0 . 06 , air puff F1 , 312 = 6 . 56 , p=0 . 01 ) . p<0 . 05 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 019 However , when we analyzed the time course of the adaptation of the CS responses , we found that 5-HT CS responses had a markedly slower rate of adaptation to the new contingencies than did DA CS responses ( Figure 9A , B , Figure 9—figure supplements 1 and 2 ) . The difference in the time constant of CS adaptation was significant for both negative and positive reversals , and was not due to differences in learning rates between groups of mice ( Figure 9C ) . We also tested whether US responses , which presumably reflect , in part , CS-related learning , also show a difference in the time course of adaptation . However , because the US signals showed a smaller signal-to-noise ratio than the CS signals , reliable time courses could not be extracted . 10 . 7554/eLife . 20552 . 020Figure 9 . Distinct speed of CS reversal learning in DRN 5-HT and midbrain DA neurons . ( A ) Normalized exponential fits ( black traces ) to the mean amplitude of the CS responses ( gray traces ) across trials for CS 2 and CS 3 of an example SERT-Cre mouse . Insets on top show mean CS response ( and 95% CI ) on days −1 ( left ) and 3 ( right ) . ( B ) Same as ( A ) for an example TH-Cre mouse . ( C ) Mean time constants ( ± s . e . m . , green and purple dots ) of the exponential fits of CS responses obtained for TH-Cre and SERT-Cre mice during reversal learning ( neural activity: unpaired t-tests , p<0 . 001 for negative reversal , p=0 . 0023 for positive reversal; no significance obtained for anticipatory licking ) . Gray dots represent individual mouse–odor pairs for each category of reversal type; gray dots with darker edges represent odors 2 or 4 , while the remaining dots represent odors 1 or 3 . ( D ) Difference in the mean fitted amplitude of CS response between DA and 5-HT during negative reversal ( left ) and during positive reversal ( right ) . p<0 . 01 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 02010 . 7554/eLife . 20552 . 021Figure 9—figure supplement 1 . CS responses of DRN 5-HT neurons during reversal . ( A ) Top: Mean responses of an example mouse ( SERT7 ) to odor one across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean response amplitude to odor 1 ( z-scored across days ) : gray dots represent individual mice ( n = 8 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day: F4 , 2597 = 72 . 1 , p<0 . 001; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( B ) , ( C ) , ( D ) Same as ( A ) , but for odor 2 ( F4 , 2554 = 33 . 37 , p<0 . 001 ) , odor 3 ( F4 , 2513 = 10 . 88 , p<0 . 001 ) , and odor 4 ( F4 , 2559 = 73 . 36 , p<0 . 001 ) , respectively . p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 02110 . 7554/eLife . 20552 . 022Figure 9—figure supplement 2 . CS responses of midbrain DA neurons during reversal . ( A ) Top: Mean responses of an example mouse ( TH2 ) to odor one across days around reversal ( shaded areas represent 95% CI ) ; Bottom: change in mean response amplitude to odor 1 ( z-scored across days ) : gray dots represent individual mice ( n = 3 ) , black dots average ( ± s . e . m . ) of mice ( two-way ANOVA with factors day and mouse , the main effect of day: F4 , 882 = 223 . 44 , p<0 . 001; multiple comparisons with the two days before reversal , corrected using Scheffé’s method , are indicated in the figure ) . ( B ) , ( C ) , ( D ) Same as ( A ) , but for odor 2 ( F4 , 856 = 84 . 97 , p<0 . 001 ) , odor 3 ( F4 , 898 = 29 . 86 , p<0 . 001 ) , and odor 4 ( F4 , 852 = 151 . 95 , p<0 . 001 ) , respectively . *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 022 A potentially important consequence of the difference in CS learning time constants is that it implies an asymmetry between DA and 5-HT systems in positive and negative reversals ( Solomon and Corbit , 1974 ) . During a positive reversal , because the adaptation of 5-HT cue responses is much slower than that of DA cue responses , the net signal will be transiently biased towards the effects of DA ( Figure 9D , right ) . Conversely , during a negative reversal , because 5-HT cue responses persist longer than those of DA , the difference will be biased towards the effects of 5-HT ( Figure 9D , left ) . This suggests a novel mechanism by which 5-HT can contribute to preventing perseverative responding during negative reversals ( Clarke et al . , 2007 ) , by directly inhibiting behavioral responses to CSs that have undergone decreases in associated outcome values . The DREADD inactivation experiment ( Figure 1 ) supported the contribution of 5-HT to negative reversal learning , but did not distinguish whether the relevant activity occurs during the CS or the US . To test for a contribution of the CS-related activity , we asked whether there was a correlation in the animal-to-animal variability in the time constant of behavioral adaptation ( anticipatory licking ) and neural adaptation ( CS magnitude ) . Remarkably , we observed a significant correlation between the time constant of DRN 5-HT CS response and the time constant of CS-related licking for the negative reversals but not the positive one ( Figure 10 ) , suggesting that these responses could be involved in adapting to negative reversals . Moreover , during such negative reversals the time constant of DRN 5-HT responses was slower than that of anticipatory licking for all animals ( Figure 10; see Figure 10—figure supplement 1 for DA ) . We note that , while we expected that the adaptation of 5-HT CS responses to the reversal should be at least as slow as that of anticipatory licking for the two to be causally related , the fact that it was much slower ( around eight times as slow ) requires an explanation . One possibility is that our behavioral readout ( that is , tongue protrusions long enough to be detected by our sensor ) is just a ‘tip-of-the-iceberg’ of motor responses to appetitive cues , and that other , covert , movements also need to be suppressed by 5-HT during relearning , and thus 5-HT neurons need to be active until all motor responses to appetitive cues have disappeared . Alternatively , it may be the case that 5-HT CS responses could serve more than a mere motor suppression function during reversal learning , and contribute to the longer-lasting learning processes required for reversal learning ( He et al . , 2015 ) , such as those that prevent spontaneous recovery following extinction training ( Karpova et al . , 2011 ) . 10 . 7554/eLife . 20552 . 023Figure 10 . The correlation between the speed of DRN 5-HT cue learning and anticipatory licking . ( A ) Correlation between time constants of 5-HT CS responses and anticipatory licking for the negative reversal . A significant linear relationship was found: y = 8 . 4x + 34; r2: 0 . 288; F = 5 . 67 , p=0 . 032 . ( B ) Same as ( A ) for positive reversals ( no relationship was found ) . Diagonal dashed lines represent y = x . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 02310 . 7554/eLife . 20552 . 024Figure 10—figure supplement 1 . Time constant of DA CS response versus time constant of corresponding anticipatory licking . ( A ) For negative reversals . ( B ) For positive reversals . Diagonal dashed lines represent y = x . DOI: http://dx . doi . org/10 . 7554/eLife . 20552 . 024 First , we found that 5-HT US responses were strongly sensitive to changes in cue–outcome contingency after the reversal . Remarkably , 5-HT neurons responded with a similar transient excitation to violations of expectation that were either better-than-expected or worse-than-expected reward outcomes . Midbrain DA neurons , on the other hand , responded oppositely to better-than-expected and worse-than-expected outcomes . Thus , whereas DA neurons could be described as reporting a signed RPE , 5-HT neurons appeared to report , in part , an unsigned RPE ( but see below for discussion of responses to aversive events ) . That is , 5-HT neurons were sensitive not to the direction of error but to its magnitude . These responses could also be described as a type of ‘surprise’ signal ( for example , Courville et al . , 2006 ) . Supporting this idea , we found that 5-HT neurons were also sensitive to substitution of one neutral cue for a cue of another modality ( sound for odor ) . It remains to be determined whether these responses were dictated entirely by small differences in reward value , or whether they reflect sensory as well as value prediction errors . Unsigned prediction error signals have been proposed on theoretical grounds to be ideal for regulating learning and attention based on uncertainty ( Pearce and Hall , 1980; Courville et al . , 2006 ) . By reporting such signals , 5-HT US responses would be suitable to drive plasticity and re-learning during reversal of contingencies . The strong excitatory response of the 5-HT system to negative RPEs , caused by reward omissions , provides a possible explanation for why inhibiting this system impairs negative reversal learning ( Clarke et al . , 2007; Bari et al . , 2010 ) ( Figure 1 ) . That is , during negative reversals or extinction learning , the 5-HT system , either directly or through an interaction with the DA system ( Boureau and Dayan , 2011 ) , could facilitate trial-by-trial undoing of DA-dependent learning . Since 5-HT neurons also respond during positive prediction errors , such as during positive reversal or initial learning , such activation might compete with co-occurring DA signals , slowing positive learning , as has been described ( Fletcher et al . , 1999 ) . The preferential involvement of 5-HT in ‘unlearning’ responses could be explained by the relative effects of 5-HT release on downstream targets , where 5-HT may favor long-term depression ( LTD ) and DA long term potentiation ( LTP ) ( He et al . , 2015 ) . 5-HT US responses contained one notable divergence from an idealized prediction error: air puff USs continued to evoke responses , even after extensive training , and showed only minor sensitivity to the presence of a predictive cue — observations consistent with a previous report ( Cohen et al . , 2015 ) . One possible explanation is that mice failed to learn the predictive relationship between the CS and the air puff . Indeed , mice showed air puff–triggered blink responses , but failed to learn anticipatory blinking responses despite extensive training . This result likely depends on the relatively long duration of the CS–US trace period , here 2 s ( Reynolds , 1945; Boneau , 1958; Cohen et al . , 2012 , 2015; Caro-Martín et al . , 2015; cf . Matsumoto et al . , 2016 ) . This is consistent with the idea that mice did not learn the CS–air puff association . If mice formed no CS-dependent predictions about the air puff , then they might have experienced each air puff as ‘unpredicted’ , whether before or after reversal . In this case , the presence of robust air puff US responses would be consistent with an unsigned value prediction error . However , since we have no explanation for how mice could succeed in learning a CS–reward association while failing to learn the CS–air puff association , other explanations should also be considered . A second possible line of explanation for the observation that the air puff did not elicit an increased response after the reversal is that 5-HT neurons report at least two qualitatively distinct signals: one relating to the processing of rewards and the other to the processing of aversive stimuli . In principle , following a reversal from large reward to air puff , one would have expected a contribution of the reward omission response to the US response , as seen in the small reward to neutral reversal . The lack of such a response could indicate either simple saturation or a suppressive influence of the air puff on the reward omission signal . A distinction between the encoding of rewarding vs . aversive events by the DA system has been proposed ( Fiorillo , 2013 ) . The presence of dual signals might reflect the inclusion of multiple 5-HT neuronal populations within our photometric recordings . In future experiments , these could be distinguished using a pathway-specific labeling , as has been done in the DA system ( Lerner et al . , 2015; see further discussion below ) . On the other hand , VTA DA neurons have been reported to integrate reward and aversive outcome values , but with aversive responses being strongly modulated by the rate of reward available in the current context ( Matsumoto et al . , 2016 ) . In future experiments , it will be important to understand how individual 5-HT neurons integrate information from combinations of outcomes , and in different reward contexts . An alternative possibility is that the pattern of 5-HT US responses could be understood together as a variation on a prediction error signal . Whereas mice can control the consumption of available water , they cannot control the delivery of air puffs; they are afforded no means to escape in the head-fixed configuration . It is therefore interesting to consider the possibility that 5-HT neurons might report errors of control rather than errors of prediction . Under this hypothesis , an aversive outcome such as the air puff continues to generate a response in 5-HT neurons because the organism has not managed to control this aspect of its environment . If the mouse were offered a means to escape , we would expect to see the air puff response diminish . Conversely , because the 5-HT US response is also sensitive to errors of a positive nature , we would also expect to see continued responses to a non-controllable reward , for example , direct oral infusion of sucrose ( Li et al . , 2016 ) . Such ‘unsigned control errors’ could provide the organism with a signal of the magnitude of cognitive or behavioral effort required to adapt to a given situation , a signal that could be read out for the purpose of energizing or deenergizing actions . Consistent with the control error hypothesis , predictable but uncontrollable shocks robustly activate the immediate early gene Fos in DRN 5-HT neurons ( Takase et al . , 2004 ) , and this activation is lowered by controllability signals from the ventral medial prefrontal cortex ( Bland et al . , 2003; Amat et al . , 2005 ) . This proposal also finds support in a recent study showing that DRN 5-HT activity mediates short-term sensorimotor adaptation in zebrafish , by reporting the difference between the expected and actual sensory consequences of motor commands ( Kawashima et al . , 2016 ) . However , further experiments will clearly be necessary to test these ideas as explanations of the present data . 5-HT could thus contribute to cognitive flexibility not only through learning and plasticity , but also by directly suppressing activity in systems responsible for violated predictions . Indeed , 5-HT has been strongly associated with suppressing both impulsive and perseverative responses through ‘behavioral inhibition’ ( Clarke et al . , 2007; Boureau and Dayan , 2011; Cools et al . , 2011 ) . In addition to US signals that could explain the contribution of 5-HT to uncertainty-driven learning , we also found CS or cue responses that could explain a direct and immediate contribution to behavioral control during environmental change . We found that 5-HT CS responses , like DA CS responses , were strongly positively correlated with CS value , consistent with previous reports ( Liu et al . , 2014; Cohen et al . , 2015; Hayashi et al . , 2015 ) . Indeed , 5-HT and DA CS signals were qualitatively extremely similar , both after initial training and after relearning . Given that 5-HT and DA are thought to drive opposing processes of behavioral inhibition and invigoration , respectively ( Boureau and Dayan , 2011; Cools et al . , 2011 ) , this would suggest that the two systems effectively cancel one another out . However , surprisingly , we found that the CS responses of 5-HT neurons were not only much slower than DA neurons to adapt to new associations after the reversal , but were also maintained throughout the extinction of the maladaptive perseverative response , as would be needed to prevent interference of the old appetitive response . Furthermore , there was a significant correlation across animals in the post-reversal learning rates of trial-by-trial 5-HT activity and that of anticipatory licking ( Figure 10 ) . This difference in rates of adaptation between the two systems , which to our knowledge was not previously reported in any neuromodulatory system , implies that the net balance between DA and 5-HT will undergo specific dynamics during learning that resemble the classical proposal concerning opponent processes by Solomon & Corbit ( 1974 ) . Specifically , because DA cue responses are quicker to establish , cues undergoing positive changes in outcome value will temporarily favor DA signals . Conversely , because DA cue responses are also quicker to withdraw , cues undergoing negative changes in outcome value will temporarily favor 5-HT signals . Thus , positive changes will favor DA and behavioral invigoration , and negative changes will favor 5-HT and behavioral suppression . This may explain why 5-HT is specifically critical in preventing responses to cues that were previously rewarding , which is observed experimentally ( Figure 1; Clarke et al . , 2007 ) . The origins of the differences in 5-HT and DA learning dynamics will be important to uncover , and might arise from differences in the systems feeding into the two neuromodulators . Interestingly , neural responses in the caudate nucleus , a major recipient of DA projections ( Clarke et al . , 2011 ) , adapt faster during reversal learning , while the PFC , a major target of 5-HT projections ( Muzerelle et al . , 2016 ) , adapts more slowly ( Pasupathy and Miller , 2005 ) . The technique of fiber photometry of genetically-encoded calcium indicators provides excellent genetic specificity and stable long-term recordings , but does not allow the resolution of single-neuron responses . It is therefore possible that differential activity patterns within specific subpopulations of DRN 5-HT neurons exist that could not be resolved by this recording method . In fact , several studies point to a heterogeneity among DRN neurons , both in terms of physiological responses ( Ranade and Mainen , 2009 ) and in terms of projection targets of DRN cell groups ( Muzerelle et al . , 2016 ) and single neurons ( Gagnon and Parent , 2014 ) . This would suggest that the different signals we observed—for example , CS vs . US or rewarding vs . aversive USs—could have different origins and functions . Even if this is the case to some extent , and given the consistency of our optical fiber targeting ( Figure 3—figure supplement 1 ) , we believe that such heterogeneity probably will not substantially impact our conclusions for several reasons . First , importantly , we established that the population from which we are recording contributes to reversal learning , and it is therefore a relevant population . Second , activity patterns were consistent across mice ( Figure 3—figure supplements 3 and 5 , Figure 4 , Figure 4—figure supplement 1 , Figure 5 , Figure 5—figure supplement 1 , Figure 9—figure supplements 1 and 2 ) , despite inevitable small variations in infections and fiber placements ( Figure 3—figure supplement 1 ) , indicating that the findings are robust to the precise population monitored . Third , single-unit recordings ( Cohen et al . , 2015; Hayashi et al . , 2015 ) show that rewarding and aversive events activate the same individual DRN neurons ( including identified 5-HT neurons ) , and are therefore not generated by distinct populations . Finally , because individual 5-HT neurons have broad projection fields ( Muzerelle et al . , 2016 ) and transmit primarily by volume conduction ( Dankoski and Wightman , 2013 ) , heterogeneity will tend to be averaged out through pooling by downstream targets . Another limitation of our study relates to the pharmacogenetic approach to inhibiting 5-HT neurons . While it has good genetic specificity , its spatial resolution is limited by the spread of the viral particles containing the hM4Di receptor in the DRN , and its temporal resolution is on the order of dozens of minutes . Additionally , although we know this approach should inhibit 5-HT neurons in vivo ( Teissier et al . , 2015 ) , we did not test the efficacy of this inhibition in our animals . The limited temporal resolution of this approach makes it impossible to distinguish the contribution of CS and US 5-HT signals to behavioral flexibility . Still , we have an indication that CS responses might play a role in behavioral inhibition of perseverative responding . This could potentially be resolved in future experiments using optogenetic inhibition . Our results support , in a general sense , the long-standing notion of DA–5-HT opponency ( Boureau and Dayan , 2011 ) , but call for a refinement of this view . Rather than carrying the positive and negative sides of a single-signed prediction error ( Daw et al . , 2002; Boureau and Dayan , 2011; Cools et al . , 2011 ) , DA–5-HT opponency seems to be more complex and subtle . As has been classically described , the activity of DA neurons which we recorded closely resembled a so-called signed RPE ( Schultz et al . , 1997 ) . The US-related 5-HT signals , on the other hand , resemble in important respects , but don’t perfectly match , the concept of an ‘unsigned RPE’ signal . Thus , 5-HT neurons responded not to an opposing class of events , but to an overlapping and broader range of events compared to DA . In this respect , they might be acting as a kind of inhibitory ‘surround’ to DA’s excitatory ‘center’ , helping to sharpen the focus of behavioral attention . Nevertheless , just as DA signaling is increasingly acknowledged to be more complex than classically described ( Cohen et al . , 2012; Eshel et al . , 2015; 2016; Matsumoto et al . , 2016; Wise , 2004 ) , attributing a single function to 5-HT neurons is also clearly an oversimplification . With respect to CS responses , 5-HT neurons showed a remarkably similar pattern of activity to that of DA neurons , scaling closely with the value of the stimuli . A possible explanation for this observation is that 5-HT CS responses could be learned by the same DA-dependent process that generates DA CS responses . If this were the entire story , then 5-HT and DA CS responses might simply balance and nullify one another . However , the fact that 5-HT CS responses evolved much more slowly than did DA CS responses means that such a balance will not hold in dynamic environments . This dynamic balance between positive and negative forces resembles the balance of excitation and inhibition in the cortex ( for example , Wehr and Zador , 2003 ) , albeit on a much slower time scale . Such a temporal asymmetry between opponent processes endows the joint system with novel and potentially important dynamics , which may be an important substrate in the dynamics of learning , as previously proposed ( Solomon and Corbit , 1974 ) . CS and US responses of a similar nature to those observed in 5-HT and DA neurons also appear to be observed in other neuromodulatory systems as well ( Yu and Dayan , 2005; Dayan and Yu , 2006; Sara and Bouret , 2012; Hangya et al . , 2015 ) . This suggests that , contrary to the notion that each neuromodulator reports a completely distinct signal ( Daw et al . , 2002; Doya , 2008; Dayan , 2012 ) , they have highly overlapping signals , presumably derived from partly overlapping inputs , but with more subtle differences through which their joint actions are orchestrated . This description of the dynamics of 5-HT neurons during reversal learning provides novel insights into how this system can contribute to cognitive flexibility . Moreover , the results also suggest the need for a refinement in conventional conceptions of 5-HT’s function in the regulation of mood , with implications for understanding its role in depression and other psychiatric disorders . More than reporting the affective value of the environment ( Boureau and Dayan , 2011; Luo et al . , 2016 ) , we suggest that 5-HT facilitates the ability of an organism to adapt flexibly to dynamic environments through plasticity and behavioral control . The clinical benefits of an enhancement of 5-HT function would therefore stem not from directly biasing affective states toward the positive , but by preventing the negative consequences of maladaptive world views and facilitating adaptive change ( Branchi , 2011 ) . All procedures were reviewed and performed in accordance with the European Union Directive 2010/63/EU and the Champalimaud Centre for the Unknown Ethics Committee guidelines , and approved by the Portuguese Veterinary General Board ( Direcção Geral de Veterinária , approvals 0420/000/000/2011 and 0421/000/000/2016 ) . Thirty-four C57BL/6 male mice between two and nine months of age were used in this study . Mice resulted from the backcrossing of BAC transgenic mice into Black C57BL for at least six generations , and expressed the Cre recombinase under the control of specific promoters . Twenty-six mice expressed Cre under the serotonin transporter gene ( Tg ( Slc6a4-cre ) ET33Gsat/Mmucd ) from GENSAT ( Gong et al . , 2007 ) ; RRID:MMRRC_017260-UCD ) , four mice under the tyrosine hydroxylase gene , two mice ( Tg ( Th-cre ) FI12Gsat/Mmucd ) from GENSAT ( Gong et al . , 2007 ) ; RRID:MMRRC_017262-UCD ) , and two mice ( B6 . Cg-Tg ( Th-Cre ) 1Tmd/J ) from the Jackson Laboratory ( Savitt et al . , 2005 ) ; RRID:IMSR_JAX:008601 ) . Animals ( 25–45 g ) were group-housed prior to surgery and individually housed post-surgery and kept under a normal 12 hr light/dark cycle . All experiments were performed in the light phase . Mice had free access to food . After training initiation , mice used in behavioral experiments had water availability restricted to the behavioral sessions . Mice were deeply anaesthetized with isoflurane mixed with O2 ( 4% for induction and 0 . 5–1% for maintenance ) and placed in a stereotaxic apparatus ( David Kopf Instruments ) . Butorphanol ( 0 . 4 mg/kg ) was injected subcutaneously for analgesia and Lidocaine ( 2% ) was injected subcutaneously before incising the scalp and exposing the skull . For SERT-Cre mice a craniotomy was drilled over lobule 4/5 of the cerebellum , and a pipette filled with a viral solution was lowered to the DRN ( bregma −4 . 55 anteroposterior ( AP ) , −2 . 85 dorsoventral ( DV ) ) with a 32° angle toward the back of the animal . For the two TH-Cre mice from The Jackson Laboratory , the pipette was targeted to the VTA ( bregma −3 . 3 AP , 0 . 35 mediolateral ( ML ) , −4 . 2 DV ) with a 10° lateral angle , and for the two TH-Cre mice from GENSAT we targeted the SNc ( bregma −3 . 15 AP , 1 . 4 ML , −4 . 2 DV ) . Although the TH-Cre lines have been characterized as less specific than other DA-specific lines ( Lammel et al . , 2015 ) , we targeted our fibers to areas where this specificity problem is reduced ( Lammel et al . , 2015 ) and that are known to contain the classical DA neurons that show RPE activity and are involved in reward processing ( Lammel et al . , 2011 , 2012; Matsumoto and Hikosaka , 2009; Lerner et al . , 2015; Kim et al . , 2016 ) . Viral solution was injected using a Picospritzer II ( Parker Hannifin ) at a rate of approximately 38 nl per minute . The expression of hM4D and of all fluorophores was Cre-dependent , and all viruses were obtained from the University of Pennsylvania ( with 1012 or 1013 GC/mL titers ) . For hM4D experiments 1 µl AAV2/5 - Syn . DIO . hM4D . mCherry was injected in the DRN of 8 SERT-Cre mice . No virus was injected in WT controls ( n = 4 ) . For analysis of GCaMP6s specific expression in 5-HT neurons , four SERT-Cre mice were transduced in the DRN with 1 µl of viral stock solution of AAV2/1 - Syn . Flex . GCaMP6s . WPRE . SV40 . For behavioral experiments in control mice ( four SERT-Cre mice ) , 1 . 5 µl of a mixture of equal volumes of AAV2/1 . EF1a . DIO . eYFP . WPRE . hGH and of AAV2/1 . CAG . FLEX . tdTomato . WPRE . bGH was used . For the remaining mice , a mixture of equal volumes of AAV2/ ( 1 or 9 ) . Syn . Flex . GCaMP6s . WPRE . SV40 and of AAV2/1 . CAG . FLEX . tdTomato . WPRE . bGH was injected: 1 . 5 µl in ten SERT-Cre mice ( distributed around six points around the target coordinates ) and 0 . 75 µl of 10 times diluted mixture in four TH-Cre mice ( distributed around four points around the target ) . For photometry experiments , optical fiber implantations were done after infection and a head plate for head fixation was placed above Bregma; the skull was cleaned and covered with a layer of Super Bond C and B ( Morita ) . An optical fiber ( 300 µm , 0 . 22 NA ) housed inside a connectorized implant ( M3 , Doric Lenses ) was inserted in the brain , with the fiber tip positioned at the target for SERT-Cre mice and 200 µm above the infection target for TH-Cre mice . The implants were secured with dental acrylic ( Pi-Ku-Plast HP 36 , Bredent ) . Mice were water-deprived in their home cage on the day of surgery , or up to five days before it . During water deprivation each mouse’s weight was maintained above 80% of its original value . Following infection and implantation surgery , mice were habituated to the head-fixed setup by receiving water every 4 s ( 6 µl drops ) for three days , after which training in the odor-guided task started . A mouse nose poke ( 007120 . 0002 , Island Motion Corporation ) using an IR photoemitter-photodetector was adapted to measure licking as IR beam breaks . To deliver air puffs , a pulse of air was delivered through a tube to the right eye of the mouse . Sounds signaling the beginning of the trial and the outcomes were amplified ( PCA1 , PYLE Audio Inc . ) and presented through speakers ( Neo3 PDRW W/BC , Bohlender-Graebener ) . Water valves ( LHDA1233115H , The Lee Company ) were calibrated and a custom made olfactometer designed by Z . F . M . ( Island Motion ) was used for odor delivery . The behavioral control system ( Bcontrol ) was developed by Carlos Brody ( Princeton University ) in collaboration with Calin Culianu , Tony Zador ( Cold Spring Harbor Laboratory ) and Z . F . M . Odors were diluted in mineral oil ( Sigma-Aldrich ) at 1:10 and 25 µl of each diluted odor was placed inside a syringe filter ( 2 . 7 µm pore size , 6823–1327 , GE Healthcare ) to be used in two sessions ( ~100 trials for each odor ) . Odorized air was delivered at 1000 ml/min . Odors used were carvone ( R ) - ( - ) , 2-octanol ( S ) - ( + ) , amyl acetate and cuminaldehyde . For the behavioral task used in the hM4D experiment , these odors were associated with reward , reward , nothing and nothing , respectively . For the behavioral task used in the GCaMP6s experiment , they were associated with a large reward ( 4 µl water drop ) , small reward ( 2 µl water drop ) , neutral ( no outcome ) and punishment ( air puff to the eye ) before reversal , and with punishment , neutral , small reward and large reward after the reversal of the cue–outcome associations , respectively . In each trial , white noise was played to signal the beginning of the trial and to mask odor valve sounds . A randomly selected odor was presented for 1 s . Following a 2 s trace period , the corresponding outcome was available . Mice completed one session per day . For hM4D experiments , odors were introduced in pairs . For photometry experiments , training started by presenting only the large and small reward trials to the mice , followed by the introduction of the neutral type of trial in the next session , and finally the punishment trial in the following one . Punishment trials were presented gradually until all four types of trials had the same probability of occurrence and each session consisted of 140–346 trials ( minimum to maximum , 223 ± 30 , mean ± SD ) . Time to odor ( foreperiod ) , trace period and inter-trial interval ( ITI ) were also gradually increased during training until mice could do the task with their final values: foreperiod was 3 to 4 s , taken from a uniform distribution , trace was fixed at 2 s , ITI was 4 to 8 s taken from a uniform distribution . hM4D experiments were run in two batches: the experiment was run first on the WT animals and then on the SERT-Cre animals ( with some overlapping days ) . Photometry experiments were run in five batches in the following sequence: 3 ( SERT-Cre , experimental ) +3 ( SERT-Cre , experimental and YFP controls ) +2 ( SERT-Cre YFP controls ) +6 ( SERT-Cre , experimental ) +4 ( TH-Cre , experimental ) . For the hM4D experiments , mice received a daily injection of vehicle ( saline 0 . 9% and DMSO 0 . 25% ) 40 min . before session start . The volume of these daily injections of vehicle was determined according to each mouse’s weight , and it required an adjustment of the water drop size for each mouse to keep them motivated to do 150 trials per session . On the reversal day and the two following days , for experimental mice , CNO was diluted in the vehicle solution and delivered at a concentration of 3 mg/kg . In both reversal learning tasks used , we ensured that mice could correctly perform the task on at least three consecutive days before reversing the odor–outcome contingency for the first time . On the reversal day , mice started the session as before and the contingencies were reversed at trial 50 in the hM4D experiment , and between the 32th and the 100th trial ( 73 ± 12 , mean ± SD ) in the GCaMP6s experiments . One SERT-Cre mouse was excluded from the hM4D analysis for not showing a differential lick rate within 1 . 5 s of US delivery , between odors 1 and 2 ( rewards ) and odors 3 and 4 ( nothing ) . Two mice were excluded from the GCaMP6s data analysis for bad fiber placement assessed after histology analysis ( more than 400 µm away from the infection area ) : one SERT-Cre and one TH-Cre mouse . Additionally , another SERT-Cre mouse was discarded from the reversal data analysis because of experimental problems with the fiber during the reversal session . In four SERT-Cre mice and in all TH-Cre mice , at five to six days after the reversal , we introduced uncued US trials during the task . These trials represented approximately 20% of the total number of trials in a session during which no odor cue was presented; the typical white noise of the foreperiod was immediately followed by one of the four possible outcomes , randomly selected ( 11 ± 4 uncued vs 44 ± 8 cued trials per session , mean ± SD ) . To analyze these data , four sessions with cued and uncued outcomes were pooled together for each mouse . All GCaMP6s experiments were performed within the limit of one month from the viral injection date , to avoid cell death due to over-expression of GCaMP6s in neurons . The dual-color fiber photometry acquisition setup consists of a three-stage tabletop black case containing optical components ( filters , dichroic mirrors , collimator ) , two light sources for excitation and two photomultiplier tubes ( PMTs ) for acquisition of fluorescence of a green ( GCaMP6s ) and of a red ( tdTomato ) fluorophore . We used a 473 nm ( maximum power: 30 mW ) and a 561 nm ( maximum power: 100 mW ) diode-pumped solid-state laser ( both from Crystalaser ) for excitation of GCaMP6s and of tdTomato , respectively . Beamsplitters ( BS007 , Thorlabs ) and photodiodes ( SM1PD1A , Thorlabs ) were used to monitor the output of each laser . The laser beams were attenuated with absorptive neutral density filters ( Thorlabs ) , and each was aligned to one of the two entrances of the three-stage tabletop black case ( Doric Lenses ) . At the corresponding entrances the excitation filters used were 473 nm ( LD01-473/10-25 Semrock ) and 561 nm ( LL02-561-25 Semrock ) . Inside the black case three interchangeable/stackable cubes ( Doric Lenses ) with dichroic mirrors were used: one to separate the 473 nm excitation light from longer wavelengths ( Chroma T495LP ) , one to collect the emission light of GCaMP6s ( FF552-Di02−25 × 36 Semrock ) , and one to separate the 561 nm excitation light from tdTomato’s fluorescence ( Di01-R561−25 × 36 ) . A collimator ( F = 12 mm , NA = 0 . 50 , Doric Lenses ) focused the laser beams in a single multimode silica optical fiber with 300 µm core and 0 . 22 NA ( MFP_300/330/900–0 . 22_2 . 5m-FC_CM3 , Doric Lenses ) , which was used for transmission of all excitation and emission wavelengths . The three-stage tabletop black case had two exits , one for each fluorophore emission , at which we placed the corresponding emission filters ( Chroma ET525/50m for GCaMP6s and Semrock LP02-568RS-25 for tdTomato ) , and convergent lenses ( F = 40 mm and F = 50 mm , Thorlabs ) before the photodetectors ( photomultiplier tube module H7422-02 , Hamamatsu Photonics ) . The output signals of the PMTs were amplified by a preamplifier ( C7319 , Hamamatsu ) , acquired in a Micro1401-3 unit at 5000 Hz and visualized in Spike2 software ( Cambridge Electronic Design ) . Light power at the tip of the patchcord fiber was 200 µW for each wavelength ( 473 nm and 561 nm ) for all experiments ( measured before each experiment with a powermeter PM130D , Thorlabs ) . This patchcord fiber was attached to the fiber cannula each animal had implanted ( MFC_300/330–0 . 22_5 mm_RM3_FLT Fibre Polymicro , polymide removed ) through a titanium M3 thread receptacle . All data were analyzed in MATLAB ( RRID:SCR:001622 ) . For the behavioral experiments , lick rate was acquired at 1 KHz and smoothed using convolution with a Gaussian filter of 50 ms standard deviation . Mean anticipatory licking was calculated for each trial as the mean lick rate in the period of 500–2800 ms after odor onset , after subtracting the mean lick rate over a baseline period of −500 to 500 ms around odor onset . To evaluate the aversiveness of the air puff delivered to the mice in the photometry experiment , we used a CCD camera ( Point Grey ) to record the right eye of six mice during several sessions at 60 Hz . To quantify blinking in video data , we manually selected the eye area in each session and calculated the mean pixel value for that area; then , for each frame , we subtracted this value from the previous frame to obtain a measure of movement . The start and end of blinking created a sudden increase and decrease , respectively , in the difference between the mean pixel value of consecutive video frames . In the time course analysis of the licking behavior in the hM4D experiment , trials of sessions around reversal were concatenated and smoothing over three trials was performed along the trials . For each reversed odor and each mouse , the last 50 trials before reversal were fit by a constant function of the form ( A+B ) ; the first 200 trials after the reversal were fit by an exponential function of the form ( A+Bexp ( -t/τ ) ) using fminsearch in MATLAB . The conditions for this fitting to be done were: the last 100 trials before reversal had to be statistically different from trials 100–200 after the reversal ( t-test ) , the change in licking pattern had to follow the correct trend of the reversal ( increase in licking for positive reversals and decrease in licking for negative ones ) , and the time constant obtained had to be larger than 1 . Mouse–odor pairs that did not fulfill this condition were excluded ( that is , odor 4 of mice M#4 and M#5 ) . Time constants were grouped according to the type of reversal and genotype with drug treatment , and compared using one-way ANOVA . Then , for each SERT-Cre mouse , the time constant of the reversal with the vehicle was subtracted from the reversal with CNO . The same was done for WT mice , but subtracting the time constant of reversal two from that of reversal 1 ( since CNO was delivered in both ) . t-tests were used to determine whether these differences had means significantly different from zero . Fluorescence data were downsampled to 1 kHz and smoothed using convolution with a Gaussian filter of 100 ms standard deviation . For each trial , the relative change in fluorescence , ΔF/F0 = ( F-F0 ) /F0 , was calculated by taking F0 to be the mean fluorescence during a 1 s period before the odor presentation for both the red and the green channels ( [ΔF/F0] GREEN and [ΔF/F0] RED ) . For each session and each mouse , the distribution of green and red values of ΔF/F0 was fitted by the sum of two Gaussians along the red channel , and the crossing point between these two Gaussians was used as a boundary ( excluding the first and last 1000 ms of each trial because of filtering artifacts ) . All values of [ΔF/F0]RED below this boundary were used , together with the corresponding [ΔF/F0]GREEN , to fit a linear regression line . Then , for each trial we corrected the green ΔF/F0 values using the parameters ( a - slope; b - offset ) obtained with the regression model of that mouse in that session: [ΔF/F0]GREEN_corr = [ΔF/F0] GREEN - a[ΔF/F0] RED - b . Behavioral data were organized as a function of US type and divided into CS and US responses . [ΔF/F0]GREEN_corr US responses were normalized by subtracting the mean [ΔF/F0]GREEN_corr over the 1 s interval before US onset . The CS or US response was considered the mean of the signal during the 1 . 5 s period after CS or US onset , respectively . For each mouse , all CS and US responses were z-scored in the expert phase , so that the amplitudes of responses to the different events could be compared . Analysis of US responses across days was performed by z-scoring all US responses of each mouse across days for each US type . Statistical analysis was done by comparing each day with pre-reversal days −1 and −2 . For each mouse , mean amplitude of response to each US on the reversal day was also compared to the day before the reversal . For analysis of uncued US trials , four days of each mouse were pooled together due to the small number of uncued trials of each US type in each session . For the analysis of CS response time courses during a reversal , each mean amplitude change across trials was fitted by an exponential function with maximum time constant of 225 trials ( minimum number of trials after the reversal for any US type of any mouse ) . The same criteria and parameters used for the hM4D experiments were used here . Time constants for mouse–odor pairs were pooled together in pairs ( odors 1 and 2 , and odors 3 and 4 ) which correspond to the negative and positive reversals , respectively . The data are available from the Dryad Digital Repository: http://dx . doi . org/10 . 5061/dryad . 649nk ( Matias , 2016 ) . Mice were deeply anesthetized with pentobarbital ( Eutasil , CEVA Sante Animale ) , exsanguinated transcardially with cold saline and perfused with 4% paraformaldehyde ( P6148 , Sigma-Aldrich ) . Coronal sections ( 40 µm ) were cut with a vibratome and used for immunohistochemistry . For SERT-Cre mice used in expression specificity analysis , anti-5-HT ( 36 hr incubation with rabbit anti-5-HT antibody 1:2000 , Immunostar , RRID:AB_572263 , followed by 2 hr incubation with Alexa Fluor 594 goat anti-rabbit 1:1000 , Life Technologies ) and anti-GFP immunostaining ( 15 hr incubation with mouse anti-GFP antibody 1:1000 , Life Technologies , followed by 2 hr incubation with Alexa Fluor 488 goat anti-mouse 1:1000 , Life Technologies ) were performed sequentially . For SERT-Cre mice used in behavioral experiments , anti-GFP immunostaining was performed ( 15 hr incubation with rabbit polyclonal anti-GFP antibody 1:1000 , Life Technologies , followed by 2 hr incubation with Alexa Fluor 488 goat anti-rabbit 1:1000 , Life Technologies ) . For TH-Cre mice , anti-GFP ( 15 hr incubation with rabbit polyclonal anti-GFP antibody 1:1000 , Life Technologies , followed by 2 hr incubation with Alexa Fluor 488 goat anti-rabbit 1:1000 , Life Technologies ) and anti-TH immunostaining ( 15 hr incubation with mouse monoclonal anti-TH antibody 1:5000 , Immunostar , RRID:AB_572268 , followed by 2 hr incubation with Alexa Fluor 647 goat anti-mouse , 1:1000 , Life Technologies ) were performed sequentially . To quantify the specificity of GCaMP6s expression in 5-HT neurons of SERT-Cre mice , we used a confocal microscope ( Zeiss LSM 710 , Zeiss ) with a 20X objective ( optical slice thickness of 1 . 8 µm ) to acquire z-stacks of three slices around the center of infection . Images for DAPI , GFP and Alexa Fluor 592 were acquired , and cells expressing GCaMP6s and cells stained with 5-HT antibody were quantified in a 200 × 200 µm window in the center of the DRN . The same was done for quantification of specificity in DA neurons of TH-Cre mice , but acquiring Alexa Fluor 647 instead of 592 , and taking the 200 × 200 µm window on the infection side . To evaluate fiber location in relation to infection , images for DAPI , YFP or GFP and tdTomato were acquired with an upright fluorescence scanning microscope ( Axio Imager M2 , Zeiss ) equipped with a digital CCD camera ( AxioCam MRm , Zeiss ) with a 10X objective . The location of the fiber tip was determined by the most anterior brain damage made by the optical fiber subtracted by its radius . The center of infection was estimated through visual inspection of slices as the location where there were most infected cells . The distance between the fiber tip location and center of infection was calculated as an anterior–posterior distance , which was estimated by comparing each corresponding location in the mouse brain atlas ( Paxinos and Franklin , 2001 ) . To determine the overlap between cells expressing YFP or GCaMP6s and tdTomato in SERT-Cre mice , we used a confocal microscope ( Zeiss LSM 710 , Zeiss ) with a 20X objective ( optical slice thickness of 1 . 8 µm ) to image three slices around the center of infection ( slices −1 , 0 and 1 , relative to it ) . All cell counts were done using the Cell Counter plugin of Fiji ( RRID:SCR_002285 ) .	Serotonin is a molecule that plays various roles in the human body . In the brain , it is involved in regulating mood and emotions . Growing evidence suggests that serotonin also helps animals – including humans – adapt their behavior to changes in their environment . To allow for such behavioral flexibility , serotonin might promote changes in the underlying brain structures and activity . In a type of learning known as ‘reversal learning’ , for instance , it is necessary to adapt to a sudden change in a previously familiar environment . For example , if there were a road closure on a person’s way to work , they might want to learn to stop following their usual route and learn a new and better one . Previous research has shown that when serotonin signaling is reduced , people persevere . That is , they will keep following the old route even if it is no longer the best choice . How this process works is still largely unknown . To start unraveling these mechanisms , Matias et al . trained mice in a reversal learning task while manipulating and recording the activity of the neurons that produce serotonin . The results showed that when the activity in serotonin neurons was experimentally blocked , the mice tended to keep looking for a reward that was no longer available . Then , by recording the activity of serotonin neurons , Matias et al . found that it was the surprise of discovering a change in a previously familiar environment that activates serotonin neurons . It did not matter whether the change was better or worse than expected . The findings suggest that together with dopamine , another molecule involved in learning from rewards , serotonin could play an important role during reversal learning . One next step will be to determine if serotonin mainly stops perseverance in its tracks , or whether it works by helping to unlearn the old behavior , or a combination of both . In the future , this could further our understanding of depression , which can be viewed as a disorder characterized by patients being unable to adapt to adverse situations , leaving them trapped to repeat behaviors and thoughts that are not beneficial . Future studies could also build on these findings to guide the development of new treatments for depression that involve serotonin .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Activity patterns of serotonin neurons underlying cognitive flexibility
Entosis is a form of epithelial cell cannibalism that is prevalent in human cancer , typically triggered by loss of matrix adhesion . Here , we report an alternative mechanism for entosis in human epithelial cells , driven by mitosis . Mitotic entosis is regulated by Cdc42 , which controls mitotic morphology . Cdc42 depletion enhances mitotic deadhesion and rounding , and these biophysical changes , which depend on RhoA activation and are phenocopied by Rap1 inhibition , permit subsequent entosis . Mitotic entosis occurs constitutively in some human cancer cell lines and mitotic index correlates with cell cannibalism in primary human breast tumours . Adherent , wild-type cells can act efficiently as entotic hosts , suggesting that normal epithelia may engulf and kill aberrantly dividing neighbours . Finally , we report that Paclitaxel/taxol promotes mitotic rounding and subsequent entosis , revealing an unconventional activity of this drug . Together , our data uncover an intriguing link between cell division and cannibalism , of significance to both cancer and chemotherapy . Cellular cannibalism is an ancient form of feeding used by bacteria ( González-Pastor et al . , 2003 ) and predatory amoebae ( Waddell and Duffy , 1986 ) in response to starvation . A similar phenomenon is observed among epithelial cells in human cancer ( Brouwer et al . , 1984; Overholtzer et al . , 2007; Overholtzer and Brugge , 2008; Sharma and Dey , 2011; Yang and Li , 2012; Cano et al . , 2012 ) , suggesting this primeval process may promote survival within the tumour microenvironment ( Fais , 2007; Matarrese et al . , 2008; He et al . , 2013; Lozupone and Fais , 2015 ) . Homotypic epithelial cell cannibalism can occur by entosis , an intriguing process through which one live and viable cell is completely engulfed by another , yielding a ‘cell-in-cell’ structure ( Overholtzer et al . , 2007 ) . The vast majority of internalised , entotic cells are ultimately killed and digested by their hosts , through a mechanism involving non-canonical autophagy and lysosomal degradation ( Yuan and Kroemer , 2010; Florey et al . , 2011 ) . Entosis is observed in a wide range of human cancers ( Overholtzer et al . , 2007; Overholtzer and Brugge , 2008 ) and is believed to mediate pleiotropic effects on cancer biology . On one hand , entotic cell killing can limit outgrowth through the elimination of internalised cells ( Florey et al . , 2011 ) , representing a possible means of tumour suppression . Conversely , entosis simultaneously promotes host cell survival and transformation , by providing valuable nutrients ( Fais , 2007; Krajcovic et al . , 2013 ) and driving genomic instability ( Krajcovic et al . , 2011 ) . Consistent with these pro-tumorigenic effects , the frequency of entosis is found to increase with tumour grade ( Krajcovic et al . , 2011 ) , and cell-in-cell formation correlates with poor patient outcome ( Schwegler et al . , 2015; Schenker et al . , 2017 ) , suggesting this process may be associated with tumour progression . Finally , entosis can mediate cancer cell competition ( Sun et al . , 2014a ) , allowing one population to preferentially engulf and kill another , and may therefore contribute to shaping tumour evolution . Together , these findings indicate that entosis can mediate both tumour suppressive and promoting effects ( Krishna and Overholtzer , 2016 ) , but the overall impact of entotic cell cannibalism on tumour biology and progression remains to be fully understood ( Durgan and Florey , 2015 ) . Mechanistically , entosis involves the formation of adherens junctions and the generation of actomyosin-based contractility , which enables one cell to actively push or ‘invade’ into a more deformable neighbour , in an unconventional mode of engulfment ( Overholtzer et al . , 2007 ) . This process is known to be triggered by matrix deadhesion , which renders the cells unanchored and this contractile force unopposed . ROCK-mediated myosin phosphorylation is indispensable for entosis in cultured cells ( Overholtzer et al . , 2007; Sun et al . , 2014a; Wan et al . , 2012; Sun et al . , 2014b ) and during embryonic implantation ( Li et al . , 2015 ) . Accordingly , regulated changes in actomyosin contractility , as induced by oncogenic K-Ras , can modulate entosis in suspension ( Sun et al . , 2014a ) ; a similar mechanism operates during the early stages of matrix adhesion ( Wan et al . , 2012 ) . Rho-family small GTPases regulate many fundamental cellular processes ( Jaffe and Hall , 2005; Heasman and Ridley , 2008 ) , including entosis . RhoA controls actomyosin contractility through ROCK , and blebbing through mDia , and is therefore indispensable for cell-in-cell formation ( Overholtzer et al . , 2007; Yamada and Nelson , 2007; Purvanov et al . , 2014 ) . Similarly , Rac1 can regulate myosin phosphorylation to modulate entotic cell competition ( Sun et al . , 2014a ) . Other Rho-family members seem likely to influence cell cannibalism , through effects on the actomyosin cytoskeleton , cell-cell contacts or cell-matrix adhesion , but their contributions have yet to be investigated . The present study was initiated to explore a possible role for Cdc42 , a master regulator of epithelial cell biology ( Heasman and Ridley , 2008; Joberty et al . , 2000; Etienne-Manneville , 2004; Martin-Belmonte et al . , 2007; Jaffe et al . , 2008; Wallace et al . , 2010; Roignot et al . , 2013 ) . Unexpectedly , this work has uncovered a novel mechanism of entotic cell-in-cell formation , driven by mitosis . We present our findings on the relationship between epithelial cell division and cannibalism and demonstrate its relevance to both human cancer and chemotherapy . To assess a possible role in entosis , Cdc42 was depleted in 16HBE human bronchial epithelial cells , using multiple distinct and non-overlapping RNAi reagents . All Cdc42-specific duplexes and hairpins yield a robust knockdown ( Figure 1a ) , and induce expected functional changes , such as disruption of adherens ( AJ ) and tight junction ( TJ ) maturation ( Figure 1b ) ( Wallace et al . , 2010 ) . To initiate entosis , cells were cultured in suspension for 8 hr ( Figure 1c–d ) , inducing cell-in-cell formation among ~10% of control cells , consistent with previous reports ( Krajcovic et al . , 2011 ) . Surprisingly , despite clear effects on AJ maturation , Cdc42 depletion has no major impact on cell-in-cell formation under these conditions , suggesting that primordial junctions are sufficient to support entosis . Strikingly , however , we found instead that Cdc42 depletion promotes robust cell-in-cell formation in adherent culture ( Figure 1e–f ) , in which entosis would not be expected to occur . This surprising phenotype is reproducible and statistically significant across all RNAi reagents tested , corresponding closely with knockdown efficiency ( compare siCdc42 . 1/2 ) , consistent with a specific , on-target effect of Cdc42 . Furthermore , adherent cell-in-cell formation is also observed in Cdc42-depleted MCF7 breast epithelia ( Figure 1g–i ) , indicating that this process is consistent across multiple cell lines , derived from different tissues of origin . Together , these data reveal that Cdc42 does not play a significant role in conventional entosis , but unexpectedly controls a novel form of cell-in-cell formation among adherent cells . 10 . 7554/eLife . 27134 . 003Figure 1 . Cdc42 controls entosis in adherent epithelial cells . ( a ) Control and siRNA or shRNA Cdc42-depleted 16HBE cell lysates were probed for Cdc42 and GAPDH expression by western blotting . ( b ) Representative confocal images of control and Cdc42-depleted 16HBE monolayers stained for β-catenin ( adherens junctions ) , ZO-1 ( tight junctions ) and DNA . Scale bar = 20 μm . ( c ) Representative images of cell-in-cell structures formed in matrix detached control and Cdc42-depleted 16HBE cells . Cell were stained for DNA ( blue ) and imaged by IF/confocal and DIC . Scale bar = 5 μm . ( d ) Quantification of suspension cell-in-cell formation in control and Cdc42-depleted cells . >200 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . ns = no significant difference , t-test . ( e ) Representative images of a cell-in-cell structure formed under adherent conditions in Cdc42-depleted 16HBE cells . Cells were stained for cell body ( green ) and DNA ( blue ) and imaged by IF/confocal and DIC . Scale bar = 10 μm . ( f ) Quantification of adherent cell-in-cell formation . >200 16HBE cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 002; p<0 . 0002; **p<0 . 0001 , t-test . ( g ) Control and shRNA Cdc42-depleted MCF7 cell lysates were probed for Cdc42 and GAPDH expression by western blotting . ( h ) Representative images of a cell-in-cell structure formed under adherent conditions in Cdc42-depleted MCF7 cells . Cells were stained for cell body ( green ) and DNA ( blue ) and imaged by IF/confocal and DIC . Scale bar = 10 μm . ( i ) Quantification of adherent cell-in-cell formation . >200 MCF7 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 02 , t-test . ( j ) Lysates from 16HBE cells co-depleted of Cdc42 and α-catenin ( aCat ) were probed for α-catenin , Cdc42 and GAPDH by western blotting . ( k ) Representative confocal images of 16HBE cells co-depleted of Cdc42 and siControl or α-catenin and stained for β-catenin ( green ) and DNA ( blue ) . Scale bar = 20 μm . ( l ) Quantification of cell-in-cell structures in adherent 16HBE cells treated with siCdc42 and siControl or siα-catenin , treated −/+10 μM Y-27632 ( ROCKi ) , for 3 days . >200 cells were scored per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 002; p<0 . 0002 , t-test . ( m ) Confocal images of a forming cell-in-cell structure in adherent , Cdc42-depleted 16HBE cells fixed and costained for pMLC2 ( S19; green ) , β-catenin ( red ) and DNA ( Hoechst , blue ) . The arrowhead indicates the tail of the internalising cell . Scale bar = 10 μm . ( n ) Cell-in-cell structures in Cdc42-depleted 16HBE cells were fixed and costained for LC3 ( green ) , LAMP1 ( red ) and DNA ( blue ) , and imaged by IF/confocal and DIC . The arrowhead indicates a dying internalised cell . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00310 . 7554/eLife . 27134 . 004Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 004 The adherent cell-in-cell structures observed upon Cdc42 depletion morphologically resemble those formed through entosis . However , entosis is typically triggered by matrix detachment , occuring in suspension ( Overholtzer et al . , 2007 ) , or during the early stages of adhesion ( 6–8 hr after plating ) ( Wan et al . , 2012 ) . To determine whether adherent cell-in-cell formation otherwise bears the hallmarks of entosis , additional characteristics were analysed . Firstly , Cdc42 was co-depleted with α-catenin , a core junctional component that is indispensible for entosis ( Wang et al . , 2015 ) ( Figure 1j–l ) . α-catenin depletion yields profound AJ defects , and a corresponding reduction in cell-in-cell formation , in line with an entotic mechanism . Next , the involvement of actomyosin contractility was assessed , which drives suspension entosis downstream of RhoA/ROCK ( Overholtzer et al . , 2007 ) . Inhibition of ROCK dramatically suppresses cell-in-cell formation among Cdc42-depleted cells ( Figure 1l ) , like entosis under detached ( Overholtzer et al . , 2007; Sun et al . , 2014a , 2014b ) , semi-adherent ( Wan et al . , 2012 ) or in vivo conditions ( Li et al . , 2015 ) . Consistent with this , active , phospho-myosin ( pS10-MLC2 ) is clearly enriched in the internalising cell tail ( Figure 1m ) . Finally , entotic cell cannibalism characteristically involves non-canonical autophagy ( Florey et al . , 2011 ) and lysosomal degradation ( Overholtzer et al . , 2007; Krajcovic et al . , 2013 ) . Consistent with this mechanism , both LC3 ( an autophagy protein ) and LAMP1 ( a lysosomal marker ) are transiently recruited to the vacuoles of Cdc42-depleted cell-in-cell structures ( Figure 1n ) . Taken together , these data reveal that entosis can indeed occur among adherent epithelial cells and suggest that a distinct mechanism must trigger cell-in-cell formation under these conditions . To investigate the mechanism underlying adherent entosis , long-term timelapse imaging was used to track live cell-in-cell formation events among Cdc42-depleted cells . Representative movies and stills are shown ( Figure 2a–c , Videos 1–3 ) . During every cell-in-cell event analysed , the inner cell penetrates its host either during , or shortly after , mitosis . Several permutations of this process were observed , with one or both daughters penetrating the same or different hosts , or one daughter invading the other and both entering an adherent neighbour ( cell-in-cell-in-cell ) . To support these studies , fixed , adherent , entotic events were visualised at a mid-way point by IF/3D-correlative light-electron microscopy ( CLEM; Figure 2d , Videos 4–5 ) . In each case , mitotic cells were observed internalising into adherent neighbours . These comprehensive imaging studies establish a clear relationship between cell division and cell-in-cell formation , suggesting that mitosis may drive entosis in adherent cell populations . To test this model more directly , Cdc42-depleted cells were arrested at the G2/M boundary using a Cdk1 inhibitor ( RO-3306 ) . Strikingly , inhibition of mitosis leads to a profound decrease in cell-in-cell formation in matrix-attached , but not detached conditions ( Figure 2e ) . These data indicate that cell division drives adherent , but not suspension , entosis , highlighting two mechanistically distinct routes to epithelial cell cannibalism . 10 . 7554/eLife . 27134 . 005Figure 2 . Mitosis drives entotic cell cannibalism in adherent cells . ( a–c ) Cdc42-depleted 16HBE cells were analysed by timelapse microscopy . Three different configurations of cell-in-cell formation are shown with timestamps ( hr:min ) . In each case , a mitotic cell ( Cell 1 , outlined white ) enters an adherent entotic host ( Cell 2 , outlined yellow ) . Scale bars = 10 μm . ( d ) Cdc42-depleted 16HBE cells were analysed by 3D-CLEM . ( i ) Live confocal sections from basal and mid-planes of a forming cell-in-cell structure , stained for plasma membrane ( red ) , cell body ( green ) and DNA ( blue ) . A mitotic daughter ( Cell 1 ) is shown internalising into an adherent neighbour ( Cell 2 ) . Scale bar = 10 μm . ( ii ) Corresponding serial blockface scanning electron microscopy ( SBF-SEM ) images of the same forming structure . The arrowhead marks the midbody between daughter cells . ( iii ) Cartoon outline of cell-in-cell structure from ( ii ) . ( e ) Quantification of cell-in-cell formation among control and Cdc42-depleted 16HBE cells , under adherent or suspension conditions , in the presence or absence of RO-3306 ( 5 μM ) , a Cdk1 inhibitor that induces G2/M arrest . >200 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 002; ns = no significant difference , t-test . ( f ) Representative confocal and DIC images of adherent cell-in-cell structures in wild-type-16HBE ( red ) and Cdc42-depleted GFP-16HBE ( green ) co-cultures . Scale bar = 15 μm . ( g ) Quantification of cell-in-cell formation between WT and Cdc42-depleted 16HBE co-cultures as described in ( f ) . >50 cell-in-cell structures were imaged per condition/experiment , across three separate experiments . Error bars denote mean±SEM . ( h ) Cdc42-depleted 16HBE cells were mixed with wild-type cells expressing RFP-zyxin , incubated for 3 days then stained for plasma membrane ( green ) and DNA ( blue ) and imaged by live confocal and DIC microscopy . A representative adherent cell-in-cell structure is shown , with basal and mid-plane sections presented; asterix = host cell nucleus , arrowhead = internalised cell . Scale bar = 15 μm . ( i ) Inner cell fate was analysed in Cdc42-depleted adherent 16HBE entotic structures , stained for DNA ( blue ) . Representative timelapse series show non-apoptotic and apoptotic inner cell death; timestamps = ( hr:min ) . Scale bar = 10 μm . ( j ) Quantification of inner cell fate over 20 hr in adherent Cdc42-depleted 16HBE structures . 154 cell-in-cell structures were analysed over three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00510 . 7554/eLife . 27134 . 006Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00610 . 7554/eLife . 27134 . 007Video 1 . Mitosis-driven entosis in adherent Cdc42-depleted 16HBE cells . DIC images from Widefield timelapse . Cell 1 engulfed by cell 2 post mitosis . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00710 . 7554/eLife . 27134 . 008Video 2 . Mitosis-driven entosis in adherent Cdc42-depleted 16HBE cells . DIC images from Widefield timelapse . Cell 1 is engulfed by cell 2 during mitosis . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00810 . 7554/eLife . 27134 . 009Video 3 . Mitosis-driven entosis in adherent Cdc42-depleted 16HBE cells . DIC images from Widefield timelapse . Cell 1 daughters engulf each other and are then engulfed by cell 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 00910 . 7554/eLife . 27134 . 010Video 4 . Live cell confocal z-stack of a forming cell-in-cell structure in Cdc42-depleted adherent 16HBE cells . Cells are stained with CellTracker green ( cell body ) , CellMASK , red ( membrane ) and Hoechst ( DNA , blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01010 . 7554/eLife . 27134 . 011Video 5 . Corresponding Serial Block Face SEM z-stack of forming cell-in-cell structure in Video 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 011 To investigate the process of mitosis-induced entosis further , we examined the role of Cdc42 more closely , testing whether its depletion affects the internalising mitotic cell , or its host . Control and Cdc42-depleted cells were labelled red or green , respectively , and then reseeded to yield an equally mixed monolayer . Inner/outer cell identities were scored among the adherent , cell-in-cell structures formed ( Figure 2f–g ) . In almost all pairs , the inner cell is Cdc42-depleted . These cells penetrate control or knockdown neighbours with equal frequency , and the results are unchanged by dye reversal ( data not shown ) . These findings establish that loss of Cdc42 specifically promotes internalisation , enabling a dividing cell to ‘invade’ into an adherent neighbour , while having no detectable effect on the host . Given that wild-type cells act efficiently as hosts , we can exclude the possibility that Cdc42-depletion promotes entosis by impairing cell-matrix contacts to ‘mimic’ deadhesion . To verify this further , Cdc42-depleted cells were mixed with wild-type cells expressing RFP-zyxin , an adhesome component ( Horton et al . , 2015 ) . Basal RFP-zyxin-positive foci are observed within entotic hosts ( Figure 2h ) , reinforcing the conclusion that the host cell can be fully matrix-adhered . Finally , we analysed the consequences of mitosis-induced entosis under adherent conditions . Cdc42-depleted 16HBE cell-in-cell structures were followed by timelapse microscopy and the fate of the internalised cell recorded ( Figure 2i–j ) . As in suspension conditions ( Overholtzer et al . , 2007; Florey et al . , 2011 ) , the majority of internalised cells die , by either non-apoptotic or apoptotic means , within the entotic vacuole , establishing mitotic entosis in adherent cells as a means of epithelial cell cannibalism . Together , our data demonstrate that loss of Cdc42 permits a dividing cell to ‘invade’ into a neighbour by entosis , and clarify that the host cell can be wild-type and fully adherent . These findings raise the intriguing concept that normal , adherent epithelia may engulf , kill and digest dividing neighbours under certain conditions . Our findings support a model in which epithelial Cdc42 inhibits entosis among mitotic cells . To investigate the possible mitotic functions of Cdc42 , dividing cells were analysed by flow cytometry and microscopy . No gross defects in cell cycle progression are detected upon Cdc42-depletion by flow cytometric analysis of DNA content ( Figure 3a ) . However , imaging studies reveal significant morphological changes during cell division . Control and Cdc42-depleted cells were visualised during the different phases of mitosis , in live cells stained for plasma membrane and DNA ( Figure 3b ) . Consistent with previous reports , we observed that Cdc42-depletion can misorient the plane of division ( Jaffe et al . , 2008; Roignot et al . , 2013 ) , while control cells divide parallel to the substrate , the plane of division in Cdc42-depleted cells is randomised . However , this phenotype seems unlikely to drive cell-in-cell formation , as aPKC-depletion promotes similar spindle misorientation ( Durgan et al . , 2011 ) , with no detectable effect on entosis under these conditions ( Figure 3—figure supplement 1 ) . Interestingly , we also identified additional , unanticipated changes in mitotic morphology upon Cdc42-depletion ( Figure 3b–d ) . Like other polarised epithelia ( Reinsch and Karsenti , 1994 ) , WT-16HBE cells bulge as they enter prometaphase , but retain both cell-cell and cell-matrix contacts throughout mitosis ( Figure 3b , upper panel ) . In contrast , from prometaphase onwards , Cdc42-depleted cells exhibit reduced spreading , associated with a diminished adhesive surface area , and dramatic rounding , a closely related phenomenon ( Marchesi et al . , 2014 ) ( Figure 3b , lower panel ) . These morphological changes can be quantified by calculating height/length , as a measure of cell spreading ( Figure 3c ) , and circularity , as a score of roundness ( Figure 3d ) . Across multiple reagents and experiments , Cdc42-depletion consistently and significantly augments mitotic deadhesion and rounding . These data uncover a novel role for Cdc42 in controlling mitotic morphology in polarised epithelial cells . 10 . 7554/eLife . 27134 . 012Figure 3 . Cdc42 controls mitotic deadhesion and rounding in polarised epithelial cells . ( a ) Control and Cdc42-depleted 16HBE cells were fixed and stained with propidium iodide . DNA content was analysed by FACS . ( b ) Control and Cdc42-depleted 16HBE cells were stained for plasma membrane ( red ) and DNA ( blue ) , and analysed by live confocal microscopy . Representative sections and z-stacks of different phases of mitosis are shown . Scale bar = 20 μm . Quantification of ( c ) mitotic spreading ( cell height/length ) and ( d ) mitotic rounding ( where 1 = a perfect circle ) in control and Cdc42-depleted cells . >10 metaphase cells were imaged per sample/experiment , across three independent experiments . Error bars denote mean±SD . *p<0 . 0001 , Mann-Whitney U test . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01210 . 7554/eLife . 27134 . 013Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01310 . 7554/eLife . 27134 . 014Figure 3—figure supplement 1 . Cdc42 controls adherent cell-in-cell formation , but aPKC does not . ( a ) Lysates from control and siCdc42 and siaPKC 16HBE cells were probed for Cdc42 , aPKC and GAPDH by western blot . ( b ) Representative confocal images of siRNA-treated cells stained for ZO-1 ( green ) , to visualise tight junctions , and DNA ( blue ) . Arrows indicate cell-in-cell structures . Scale bar = 50 mm . Quantification of ( c ) junctions and ( d ) cell-in-cell formation in siRNA treated cells . p<0 . 002; *p<0 . 0008 , t-test . Depletion of aPKC disrupts tight junction formation , like Cdc42 , but does not induce entosis . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 014 We next considered the molecular mechanisms through which Cdc42 may regulate mitotic morphology and entosis , with RhoA emerging as a compelling candidate player . RhoA , another of the major small GTPases , has been implicated in mitosis ( Chircop , 2014 ) , mitotic rounding ( Maddox and Burridge , 2003; Matthews et al . , 2012 ) and cell-in-cell formation ( Overholtzer et al . , 2007 ) through previous studies . To investigate a possible role here , the spatiotemporal activation of RhoA was assessed using a FRET-based biosensor , RhoA-FLARE ( Sun et al . , 2014b; Pertz et al . , 2006 ) . Control or Cdc42-depleted 16HBE cells expressing the RhoA biosensor were subjected to live confocal imaging for CFP ( FRET donor ) and YFP ( FRET acceptor ) , as well as DNA ( Hoechst ) and DIC ( Figure 4a ) ; RhoA activation is proportional to the FRET/CFP emission ratio . While control 16HBE cells show a relatively low level of RhoA activity during metaphase , a clear enrichment of active RhoA is frequently observed at the metaphase cortex among Cdc42-depleted cells ( 37%; Figure 4b ) . Related to this , an increase in cortical actin can also be observed among Cdc42-depleted , metaphase cells ( Figure 4c ) . Together , these data indicate that Cdc42 constrains mitotic RhoA activation in polarised epithelial cells , and accordingly , that loss of Cdc42 permits overactivation of cortical RhoA during metaphase . 10 . 7554/eLife . 27134 . 015Figure 4 . RhoA activity is spatiotemporally regulated by Cdc42 and controls mitotic spreading . ( a ) 16HBE cells expressing a RhoA FRET biosensor were treated with siControl or siCdc42 . Three days later , cells were subjected to live confocal imaging for CFP ( FRET donor ) , YFP ( FRET acceptor ) , DNA and DIC . RhoA activity is represented by the FRET/CFP emission ratio . Scale bars = 10 μm . ( b ) RhoA activity was measured in >30 metaphase cells per condition , across four independent experiments , and cortical enrichment of active RhoA was scored . p<0 . 003 , t-test . ( c ) Control or Cdc42-depleted 16HBE cells were fixed and stained for actin ( green ) or DNA ( blue ) , and metaphase cells were imaged by IF/confocal . Representative sections and z-stacks are shown . Scale bar = 10 μm . ( d ) 16HBE cells were treated with siControl or siCdc42 and incubated for 3 days . Cdc42-depleted cells were then incubated with C3 ( Rho inhibitor; 1 μg/ml ) , Y-27632 ( ROCK inhibitor; 10 μM ) or Blebbistatin ( myosin inhibitor; 100 μM ) for a further 4 hr . Live cells were stained for plasma membrane ( white ) and DNA ( blue ) and imaged by IF/confocal to assess metaphase morphology; scale bar = 15 μm . Representative z-stacks and basal sections are shown . ( e ) The spread basal area of each metaphase cell was measured . >15 cells were scored/condition , across three independent experiments . **p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01510 . 7554/eLife . 27134 . 016Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 016 To test whether deregulated RhoA activation influences mitotic morphology , live Cdc42-depleted cells were treated with a cell permeable form of C3 Transferase , a toxin that selectively ribosylates and inactivates RhoA/B/C ( Barbieri et al . , 2002 ) , to block downstream signalling ( Figure 4d ) . Although mitotic cells do remain somewhat rounded in the presence of this toxin , inhibition of the Rho proteins has a clear impact on mitotic spreading , yielding a significant increase in basal area , and thus partially reversing the effect of Cdc42-depletion ( Figure 4e ) . These data are consistent with previous studies that have implicated RhoA activity in mitotic cell retraction , rigidity and rounding ( Maddox and Burridge , 2003; Matthews et al . , 2012 ) , along with its targets ROCK ( Maddox and Burridge , 2003; Meyer et al . , 2011 ) and myosin ( Matthews et al . , 2012 ) . To explore this observation further , live Cdc42-depleted cells were treated with additional pathway inhibitors , and suppression of ROCK ( Y-27632 ) or myosin ( blebbistatin ) activity similarly rescued mitotic cell spreading in Cdc42-depleted cells ( Figure 4d–e ) . Together , these data indicate that aberrant activation of a Rho/ROCK/myosin cascade during mitosis can drive enhanced retraction and rounding . Based on these findings , we would predict that RhoA inhibition may also suppress mitotic entosis upon Cdc42-depletion , by reverting these induced changes in mitotic morphology . Unfortunately , it is not possible to demonstrate this link unambiguously , because RhoA inhibition will block entotic cell-in-cell formation regardless of its trigger , due to downstream effects on myosin contractility ( Overholtzer et al . , 2007 ) and actin dynamics ( Purvanov et al . , 2014 ) . As such , while we can clearly implicate RhoA in both mitotic rounding and cell-in-cell formation , we cannot definitively dissect the process of mitotic entosis further by targeting RhoA alone . As an alternative approach to determine whether mitotic deadhesion and rounding is sufficient to drive entosis , mitotic morphology was manipulated through other means . Rap1 is yet another small GTPase , which , importantly here , is known to control post-mitotic spreading ( Dao et al . , 2009; Lancaster et al . , 2013 ) . Consistent with this , we confirm that expression of DN-Rap1 reduces spreading and increases rounding during 16HBE division ( Figure 5a–d ) . Strikingly , inhibition of Rap1 , like Cdc42 , also induces adherent entosis , albeit at a lower level . Cell-in-cell formation among Rap1-inhibited cells occurs during or shortly after mitosis ( Figure 5e–g , Video 6 ) and is inhibited by G2 arrest ( Cdk1i ) . These data are consistent with a model in which enhanced mitotic deadhesion and rounding drive subsequent cell-in-cell formation . In light of these data , we next asked whether mitosis might promote entosis by simply providing an alternative route to matrix deadhesion in one cell of the pair ( the dividing , internalised cell ) . To address this , we tested whether detached , interphase cells can similarly penetrate matrix-attached hosts . WT-16HBE cells were labelled green , detached and then either overlaid onto wild-type monolayers ( adherent hosts ) or maintained in suspension for 8 hr; the resulting cell-in-cell structures were scored ( Figure 5h–i ) . Strikingly , while detached interphase cells undergo efficient entosis in suspension ( ~8% ) , penetration of adherent hosts is barely detectable . These data indicate that simple loss of matrix-attachment is insufficient to drive penetration of an adherent host . Consistent with this , entosis has not been observed among 16HBE-monolayers upon depletion of matrix-adhesion proteins ( e . g . β1-integrin; Figure 5—figure supplement 1 ) . Together , these data indicate that mitotic deadhesion and rounding , which can be augmented by inhibition of Cdc42 or Rap1 , can drive the entotic penetration of an adherent host cell , while matrix-detachment during interphase cannot . These findings infer important mechanistic differences between cell cannibalism under detached and adherent conditions . We conclude that the distinctive biophysical changes associated with mitosis are an essential requirement when entosis occurs in an adherent host . 10 . 7554/eLife . 27134 . 017Figure 5 . Enhanced mitotic deadhesion and rounding can drive entosis . ( a ) Control ( pQC ) and DN-Rap1-HA expressing 16HBE cells were probed for HA , Rap1 and tubulin by western blot . ( b ) Control and DN-Rap1 expressing 16HBE cells were stained for cell body ( green ) and DNA ( blue ) and analysed by live confocal microscopy . Representative midplane x/y , and z sections through the dashed line , are presented . Arrowheads indicate metaphase cells , as identified by nuclear morphology . Quantification of ( c ) mitotic spreading ( cell height/length ) and ( d ) mitotic rounding ( where 1 = a perfect circle ) in control and DN-Rap1 16HBE cells . >10 metaphase cells were imaged per sample/experiment , across three independent experiments . Error bars denote mean±SD . *p<0 . 0001 , Mann-Whitney U test . ( e ) DN-Rap1 expressing 16HBE cells were analysed by timelapse microscopy . A mitotic cell ( Cell 1 ) is outlined in white , the adherent entotic host ( Cell 2 ) in yellow . Timestamps are indicated ( hr:min ) and scale bar = 10 μm . ( f ) Representative confocal image of an adherent cell-in-cell structure in DN-Rap1 expressing 16HBE cells , fixed and stained for the cell body ( green ) and DNA ( blue ) . Scale bar = 10 μm . ( g ) Quantification of cell-in-cell formation in adherent control and DN-Rap1 cells , treated −/+ a Cdk1 inhibitor that induces G2/M arrest ( 5 μM RO-3306; Cdk1i ) . >250 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 03 , t-test . ( h ) Representative images from a co-culture of suspension wild type 16HBE cells , labelled with CellTracker ( green ) , overlaid on an existing monolayer of wild type 16HBE cells stained for DNA ( blue ) . Co-cultures were monitored for 8 hr by timelapse microscopy . DIC/IF images are shown at time 0 and 8 hr; the detached population ( green ) are highlighted with red arrows; these cells can persist throughout the timecourse and very rarely penetrate adherent hosts . ( i ) Quantification of cell-in-cell formation under adherent conditions as described in ( h ) , and under suspension conditions . >300 cells were counted per sample/experiment , across three independent experiments . Error bars denote mean±SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01710 . 7554/eLife . 27134 . 018Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01810 . 7554/eLife . 27134 . 019Figure 5—figure supplement 1 . Loss of β1 integrin ( ITGB1 ) does not induce adherent cell-in-cell formation . ( a ) Lysates from control and shITGB1 and shCdc42 16HBE cells were probed for ITGB1 , Cdc42 and GAPDH by western blot . ( b ) Quantification of cell-in-cell formation in shRNA-treated cells . >300 cells were counted per sample/experiment , across three independent experiments . Error bars denote mean±SEM . p<0 . 002; t-test . ( c ) Representative confocal images of shITGB1-treated cells stained for ZO-1 ( green ) , to visualise tight junctions , and DNA ( blue ) . Note , no cell-in-cell structures are observed and mitotic cells remain spread in ITGB1-depleted cells . Scale bar = 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 01910 . 7554/eLife . 27134 . 020Video 6 . Mitosis-driven entosis in adherent 16HBE cells expressing DN-Rap1 . DIC images from Widefield timelapse . A daughter of cell 1 is engulfed by cell 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 020 Given that tumour cells are prone to undergo deregulated division and that mitotic rounding is proposed to be of relevance in cancer ( Cadart et al . , 2014 ) , we hypothesised that transformed cells may undergo mitotic entosis constitutively . To test this , adherent cancer cell lines were examined by fixed and live microscopy . Interestingly , MCF7 ( breast ) and HCT116 ( colorectal ) cells were found to undergo adherent entosis under basal conditions . Importantly , cell-in-cell formation is coincident with cell division , as shown by timelapse microscopy ( Figure 6a–b , Videos 7–8 ) , and is largely dependent on progression through mitosis , in both 2D and 3D culture ( Figure 6c–d ) . These data indicate that mitosis drives constitutive adherent entosis in certain transformed tumour cell lines and raise the interesting possibility that cell division may promote cell cannibalism in the proliferative environment of a tumour . To investigate this more directly , a tissue microarray of 75 human breast invasive ductal carcinomas was analysed ( Figure 6e–g , Supplementary file 1 ) . Each tumour core was stained for DNA , β-catenin and p-Histone-H3 ( pS10; a mitotic marker ) , and then imaged in full and scored for mitotic activity ( mitotic cells/core ) and entosis ( cell-in-cell structures/core ) . 31/75 ductal carcinomas were positive for entosis , and among these , mitotic activity positively correlates with cell-in-cell formation , with statistical significance ( Figure 6g ) . These data are consistent with the notion that mitosis may drive entotic cell-in-cell formation in vivo . During conventional entosis , the inner cell is typically killed ( Overholtzer et al . , 2007; Florey et al . , 2011; Wan et al . , 2012 ) , while the outer cell is rendered prone to division failure ( Krajcovic et al . , 2011; Wan et al . , 2012 ) . To determine whether similar anti- and pro-tumorigenic consequences accompany mitotic entosis among adherent cancer cells , the outcomes were followed by timelapse and IF/confocal microscopy . Similar to suspension conditions , cell death is the predominant fate for internalised cells following mitotic entosis , which can occur by either non-apoptotic and apoptotic means ( Figure 6h–i ) . Host cell division failure was also analysed by scoring for multinucleation . Again , similar to suspension entosis , host cells are more frequently multi-nucleated than surrounding single cells , suggesting that abscission can be disrupted by entosis in adherent , as well as matrix-detached , conditions ( Figure 6j–k ) . Together , these data establish that mitosis-induced entosis occurs basally in adherent cancer cell lines and human breast carcinomas . Mitosis-induced entosis can drive inner cell killing , a potentially tumour suppressive effect , but also promotes host cell multi-nucleation , a route to tumour-promoting genomic instability . These findings uncover an intriguing relationship between cell division and cannibalism , of potential functional significance during tumour development and evolution . 10 . 7554/eLife . 27134 . 021Figure 6 . Mitosis-induced entosis occurs in cancer cell lines and human tumours , with pleiotropic effects . ( a–b ) Adherent cancer cell lines MCF7 ( breast ) and HCT116 ( colorectal ) were analysed by timelapse microscopy; representative cell-in-cell formation events are shown . In each case , a mitotic cell ( Cell 1 , outlined in white ) is internalised by an adherent neighbour ( Cell 2 , outlined in yellow ) . Timestamps are shown ( hr:min ) and scale bar = 20 μm . ( c ) Quantification of cell-in-cell formation in adherent MCF7 cells cultured in 2D or 3D , treated +/− a Cdk1 inhibitor that induces G2/M arrest ( 5 μM RO-3306; Cdk1i ) . >300 cells for 2D and >50 cells for 3D were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 002; p<0 . 02 , t-test . ( d ) MCF7 cells seeded in 3D matrigel were treated −/+5 μM RO-3306 overnight , then fixed and stained for β-catenin ( green ) and DNA ( blue ) . Two to four cell cysts were imaged to assess mitotic status and cell-in-cell formation . Representative sections are shown; scale bar = 10 μm . ( e ) Human breast invasive ductal carcinoma . A tumour microarray was stained for β-catenin ( green ) , pS10-Histone H3 ( red , mitotic marker ) and DNA ( blue ) and imaged in full by DIC and IF/confocal . A tiled confocal image is presented for one core . Scale bar = 200 μm . ( f ) Entosis in a human invasive breast ductal carcinoma . A representative cell-in-cell structure is shown by DIC and IF/confocal; notably the inner cell is mitotic as judged by pHH3 . Scale bar = 10 μm . ( g ) Quantification of mitotic index and cell-in-cell formation among human breast invasive ductal carcinomas . The median number of mitotic cells/core is 24 . p<0 . 05 , Mann-Whitney test . ( h ) Representative timelapse series showing inner cell death in an MCF7 cell-in-cell structure , stained for DNA ( blue ) . The internalised cell is outlined in yellow; its corpse shrinks over time . Timestamps are indicated ( hr:min ) and scale bar = 10 μm . ( i ) Quantification of inner cell fate in MCF7 entotic structures over 20 hr . Fifty-four cell-in-cell structures were analysed over three independent experiments . ( j ) Representative image of a multinucleated , entotic host cell in adherent MCF7 stained for cell body ( green ) and DNA ( blue ) . Scale bar = 10 μm . ( k ) Quantification of MCF7 multinucleation in single cells versus entotic hosts cells . Error bars denote mean±SEM across three independent experiments . p<0 . 008 , t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 02110 . 7554/eLife . 27134 . 022Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 02210 . 7554/eLife . 27134 . 023Video 7 . Mitosis-driven entosis in adherent MCF7 cells . DIC and DNA ( Hoechst ) images from Widefield timelapse . A daughter of cell 1 is engulfed by cell 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 02310 . 7554/eLife . 27134 . 024Video 8 . Mitosis-driven entosis in adherent HCT116 cells . DIC images from Widefield timelapse . Cell 1 is engulfed by cell 2 during mitosis . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 024 Many cancers are treated chemotherapeutically with taxanes ( e . g . Paclitaxel ) , which can induce prometaphase arrest , multipolar division and cell death ( Zasadil et al . , 2014 ) . As such , we hypothesised that taxane treatment might influence cell cannibalism by modulating mitotic entosis . To test this , 16HBE cells were incubated in the presence or absence of taxol , then imaged by timelapse or IF/confocal microscopy ( Figure 7a–c , Video 9 ) . As expected , taxol treatment increases mitotic index under these conditions , arresting the cells in prometaphase ( in contrast to Cdk1i which arrests at G2/M , inhibiting entry into mitosis ) . Importantly , taxol treatment also consistently increases cell-in-cell formation under these conditions , in line with an induction of mitotic entosis . This effect can be phenocopied with additional drugs that inhibit mitotic progression through different mechanisms , including nocodazole ( microtubule assembly inhibitor ) and STLC ( mitotic kinesin inhibitor ) , suggesting that entosis occurs as a consequence of prometaphase arrest rather than microtubule stabilisation . Strikingly , we also find that taxol , nocodazole and STLC significantly enhance mitotic deadhesion and rounding ( Figure 7d–f ) , thus independently supporting the model that changes in mitotic morphology are closely associated with adherent cell-in-cell formation . To investigate whether taxane-induced cell cannibalism may be of chemotherapeutic significance , MCF7 breast cancer cells were examined . In both cultured cells ( Figure 7g–h ) and mouse xenograft models ( Figure 7i–l ) , 24 hr taxol treatment promotes a significant increase in mitotic index , and a corresponding induction of cell-in-cell formation , consistent with mitosis-induced entosis . Notably , as taxol is reported to inhibit MCF7 entosis in suspension ( Xia et al . , 2014 ) , these data further suggest that the mechanisms of detached versus adherent cell-in-cell formation are quite distinct . Together , our findings establish that taxane treatment enhances mitotic deadhesion and rounding and promotes cell cannibalism through entosis , and reinforce the conclusion that changes in mitotic morphology can drive adherent entosis . Moreover , these findings uncover some intriguing and unconventional new effects of taxane treatment , of potential chemotherapeutic interest . 10 . 7554/eLife . 27134 . 025Figure 7 . Paclitaxel/taxol treatment promotes mitotic deadhesion , rounding and entosis . ( a ) Representative timelapse images of adherent 16HBE cells treated with 1 μM taxol . Cell 1 ( outlined white ) rounds up in prometaphase and subsequently penetrates an adherent interphase neighbour ( Cell 2 , outlined yellow ) . Timestamps are shown ( hr:min ) and scale bar = 10 μm . ( b ) Representative confocal/DIC images of adherent cell-in-cell structures in 16HBE treated with taxol ( 1 μM ) , nocodazole ( 100 ng/ml ) or STLC ( 20 μM ) for 24 hr . Cells were stained for DNA ( blue ) , scale bar = 10 μm . ( c ) Quantification of drug-induced cell-in-cell formation . >150 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 002; p<0 . 02 , t-test . ( d ) Mitotic morphology of 16HBE cells treated with taxol ( 1 μM ) , nocodazole ( 100 ng/ml ) or STLC ( 20 μM ) for 24 hr . Live cells were stained for cell body ( green ) and DNA ( blue ) . Midplane x/y , and z sections through the dashed line , are presented . Scale bar = 10 μm . Quantification of ( e ) mitotic spreading ( cell height/length ) and ( f ) mitotic rounding ( where 1 = a perfect circle ) in control and drug-treated cells . >15 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SD . **p<0 . 0001 , Mann-Whitney U test . ( g ) Representative confocal images of partially and completely formed cell-in-cell structures in MCF7 cells treated with taxol ( 1 μM ) , and stained for β-catenin ( green ) and DNA ( blue ) . The arrowheads point to prometaphase arrested cells internalised by adherent , interphase neighbours . Scale bar = 10 μm . ( h ) Quantification of taxol-induced entosis in MCF7 . >150 cells were counted per sample/experiment , across three separate experiments . Error bars denote mean±SEM . p<0 . 04 , t- test . ( i ) Confocal images of MCF7 xenografts treated −/+ taxol for 24 hr and stained for phospho-Histone H3 ( pHH3 , red ) and DNA ( blue ) . Scale bar = 50 μm . ( j ) Quantification of pHH3-positive , mitotic cells in MCF7 mouse xenografts treated −/+ taxol for 24 hr . ***p<0 . 0001 , t-test . ( k ) Representative confocal and DIC images of an entotic cell-in-cell structure in a taxol-treated MCF7 xenograft , stained for β-catenin ( green ) and DNA ( blue ) . Scale bar = 10 μm . ( l ) Quantification of cell-in-cell formation in MCF7 mouse xenografts treated −/+ taxol for 24 hr . p<0 . 01 , t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 02510 . 7554/eLife . 27134 . 026Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 02610 . 7554/eLife . 27134 . 027Video 9 . Mitosis-driven entosis in adherent 16HBE cells treated with taxol ( 1 μM ) . DIC images from Widefield timelapse . Cell 2 enters mitosis and is engulfed by cell 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 027 Cell cannibalism has been observed within tumours for over a century ( Overholtzer and Brugge , 2008; He et al . , 2013; Lozupone and Fais , 2015 ) , but its underlying mechanisms and functional consequences remain to be fully understood . Entosis is a form of homotypic epithelial cell cannibalism that is typically triggered by matrix-deadhesion , and which proceeds through junction formation and the generation of actomyosin contractility , culminating in cell engulfment and killing ( Overholtzer et al . , 2007 ) . This process is regulated at various stages by Rho-family small GTPases , which govern the cytoskeletal and junctional components upon which entosis depends ( Overholtzer et al . , 2007; Sun et al . , 2014a; Purvanov et al . , 2014 ) . This study set out to address whether Cdc42 , a master regulator of epithelial biology ( Jaffe and Hall , 2005; Heasman and Ridley , 2008; Joberty et al . , 2000; Martin-Belmonte et al . , 2007; Jaffe et al . , 2008; Wallace et al . , 2010; Roignot et al . , 2013 ) , controls entosis . Surprisingly , despite its well-documented control of the cytoskeleton , polarity and epithelial junctions ( Etienne-Manneville , 2004 ) , Cdc42 has little effect on entosis among cells cultured in suspension . Unexpectedly , however , we found that loss of Cdc42 can in fact promote robust cell-in-cell formation among adherent cells . This result was very surprising , because entosis is not expected to occur under these conditions ( Overholtzer et al . , 2007 ) , but , this process otherwise bears its hallmarks , involving cell-cell contacts ( Sun et al . , 2014b; Wang et al . , 2015 ) , polarised actomyosin ( Sun et al . , 2014a ) , autophagy proteins ( Florey et al . , 2011 ) and lysosomes ( Krajcovic et al . , 2013 ) . Notably , some similar , adherent engulfment events have been noted previously ( Lai et al . , 2010; Abreu and Sealy , 2012 ) , and in the present study , we clarify that entosis can indeed occur among matrix-attached cells , and define the associated mechanism . Cdc42 is known to regulate cell-matrix contacts through β1-integrin in certain cancer cells ( Reymond et al . , 2012 ) , so it is plausible to hypothesise that its depletion may weaken focal adhesions , or impair integrin signalling , to mimic detachment and so induce entosis . However , our data are not consistent with this model . Firstly , control monolayers can efficiently internalise Cdc42-depleted neighbours , demonstrating that entotic hosts can be wild-type and therefore fully adherent . Secondly , neither enzymatic deadhesion , nor depletion of β1-integrin , is sufficient to drive one population of cells to penetrate surrounding adherent neighbours . Together , these data indicate that general changes in matrix-binding are unlikely to fully account for adherent entosis . In this study , we identify mitosis as a novel trigger for entosis . We show that mitosis is indispensable for adherent cell-in-cell formation , with entosis occurring during , or shortly after , cell division , and requiring transit through G2/M . In contrast , mitosis is not necessary for entosis in suspension , indicating that quite distinct mechanisms operate under different growth conditions . Our findings identify an important new route to entosis and uncover an intriguing relationship between epithelial cell division and cannibalism , of particular interest in the context of a proliferative and nutrient-deprived tumour . Cdc42 activity is known to be cell-cycle regulated ( Yoshizaki et al . , 2003 ) and to control certain mitotic processes , including kinetochore attachment and chromosome segregation ( Chircop , 2014; Yasuda et al . , 2004 ) ; however , these roles appear unrelated to cell cannibalism . We have also found little evidence to support a role for aPKC as a Cdc42 effector during adherent entosis , likely excluding mechanisms involving the polarity complex ( Joberty et al . , 2000 ) , such as junctional remodelling ( Wallace et al . , 2010 ) , spindle orientation ( Durgan et al . , 2011 ) , or control of the metaphase cortex ( Rosa et al . , 2015 ) . Instead , we have discovered an additional role for Cdc42 in regulating mitotic morphology , consistent with previous observations in NRK ( Zhu et al . , 2011 ) and HeLa ( Mitsushima et al . , 2009 ) cells . Upon Cdc42 depletion , mitotic deadhesion and rounding are significantly enhanced . These phenotypes are associated with a prominent increase in cortical RhoA activity , and can be reverted by inhibition of RhoA , or its downstream effectors ROCK and myosin , consistent with previous work ( Maddox and Burridge , 2003; Matthews et al . , 2012 ) . We conclude that Cdc42 plays a novel role in the regulation of mitotic morphology in polarised epithelial cells , in a RhoA/ROCK/myosin-dependent manner . It is interesting to note that this Rho-dependent signalling axis is also required for entosis among suspension cells ( Overholtzer et al . , 2007; Sun et al . , 2014a; Wan et al . , 2012; Li et al . , 2015 ) , suggesting that some interesting mechanistic parallels exist between these alternative routes to cell cannibalism . The augmentation of mitotic deadhesion and rounding induced by Cdc42 suppression can be phenocopied by inhibition of Rap1 , or through drug-induced prometaphase arrest , and , importantly , in each case is followed by subsequent cell-in-cell formation ( Figure 8 ) . Putting together the data from these diverse conditions , we conclude that the distinctive biophysical changes associated with mitotic deadhesion and rounding may uniquely enable a dividing cell to penetrate an adherent epithelium , by which it is ultimately cannibalised . These findings add to the emerging concept that mitotic shape changes bear important functional consequences ( Lancaster et al . , 2013; Cadart et al . , 2014; Théry and Bornens , 2006; Gibson et al . , 2011; Luxenburg et al . , 2011; Kondo and Hayashi , 2013; Sorce et al . , 2015 ) , of particular significance within tumours ( Sorce et al . , 2015 ) . 10 . 7554/eLife . 27134 . 028Figure 8 . The triggers and consequences of entosis in cancer . Entosis can be triggered among epithelial cells through either matrix deadhesion or aberrant mitosis . Mitotic entosis is associated with enhanced deadhesion and rounding during cell division , which can be induced by inhibition of Cdc42 or Rap1 , or through prometaphase arrest . RhoA activity is important in both suspension and mitosis-induced entosis , driving ROCK-dependent myosin activation . Regardless of the triggering mechanism , entosis promotes both inner cell death and outer cell nutrient gain and multi-nucleation , with the potential to confer both anti- and pro-tumorigenic effects . DOI: http://dx . doi . org/10 . 7554/eLife . 27134 . 028 We report that the process of mitosis-induced entosis is observed basally among certain cancer cell lines , establishing a broader biological incidence of this process . Moreover , we find that mitotic index positively and significantly correlates with cell-in-cell formation in human breast invasive ductal carcinomas , consistent with a pathophysiological occurrence in vivo . Importantly , as mitotic index is one of the key criteria used to stage breast cancers , our findings can in part explain the increased frequency of cell-in-cell structures among higher grade , more proliferative tumours ( Krajcovic et al . , 2011; Gupta and Dey , 2003; Abodief et al . , 2006 ) . Our study thus contributes new insights into the field of cell cannibalism in cancer , building on the emerging notion that cell-in-cell formation correlates with more aggressive disease ( Krajcovic et al . , 2011 ) and may provide prognostic indications ( Schwegler et al . , 2015; Schenker et al . , 2017 ) . In relation to cancer , we also report the chemotherapeutic induction of mitotic entosis as a novel and unanticipated effect of Paclitaxel/taxol treatment . Interestingly , we find that Paclitaxel enhances mitotic deadhesion and rounding , thereby driving subsequent cell-in-cell formation through entosis . There is a significant debate regarding the mechanism of action of taxane-family drugs ( Weaver , 2014 ) , which can cause prometaphase arrest , multipolar divisions and cell death ( Zasadil et al . , 2014 ) . As such , these additional and unconventional activities in modifying mitotic morphology and driving cell cannibalism through entosis may be of potential clinical interest . Finally , we show that the outcome of entosis is broadly conserved , whether it is triggered by deadhesion or mitosis , within an adherent or detached host cell ( Figure 8 ) . On one hand , entosis drives internalised cell killing , a potential means of limiting growth ( Florey et al . , 2011 ) . We find here that the host cell can be wild-type and fully adherent , developing an intriguing model in which aberrantly dividing cells may penetrate , and be eliminated by , the surrounding normal epithelium , in a tumour-suppressive act of ‘assisted suicide’ . On the other hand , we also show that mitosis-induced entosis can disrupt host cell division , with the potential to drive genomic instability ( Krajcovic et al . , 2011 ) . In this case , an alternative model emerges in which one aberrant division promotes more , thereby contributing to tumour progression . These models are not mutually exclusive and the overall impact of entosis on tumour biology remains the focus of ongoing work ( Durgan and Florey , 2015 ) . In conclusion , we propose that there are at least two mechanistic routes to entosis: loss of matrix-attachment and cell division . It is striking that anchorage independence and unrestrained proliferation , two classic hallmarks of cancer ( Freedman and Shin , 1974; Hanahan and Weinberg , 2011 ) , can both drive this form of cell cannibalism , so commonly observed in tumours . It will be worthwhile through future work to assess whether additional features of cancer cells ( e . g . stemness ) , or their environments ( e . g . hypoxia , nutrient deprivation ) , may similarly trigger entosis , and to more comprehensively investigate the effects of cell cannibalism on tumour development , progression and evolution . Cells were obtained from the following sources: 16HBE ( Durgan et al . , 2015 ) ( from lab of Dieter Gruenert , UCSF ) , MCF7 ( Sun et al . , 2014a ) ( Lombardi Cancer Center , Georgetown University ) , HCT116 ( Sun et al . , 2014a ) ( from lab of David Boone , University of Notre Dame ) , 293FT ( Durgan et al . , 2011 ) ( Invitrogen ) ; all tested negative for mycoplasma ( MSKCC core facility ) and were cultured as described previously . The following inhibitors were used: Blebbistatin ( 100 μM; Sigma , UK ) , C3 Transferase ( 1 μg/ml; CT04 Cytoskeleton , Denver , CO ) , Nocodazole ( 100 ng/ml; Sigma ) , RO-3306 ( 5 μM; Sigma ) , STLC ( 20 μM , Santa Cruz , Dalla , TX ) , Taxol ( 1 μM; EMD ) , Y-27632 ( 10 μM; R&D , Minneapolis , MN ) ; high-grade DMSO was used as a carrier control ( 1:1000; Sigma ) . For stable expression of shRNA ( pSUPER , shCdc42 . 1 and 2 , pSiren , shITGB1 . 1 , shITGB1 . 2 ) or protein ( GFP , RFP-zyxin , HA-DN-Rap1 , RhoA biosensor ) , cells lines were generated by retroviral infection and selection as described previously ( Durgan et al . , 2011 ) . Briefly , cells were seeded at 10∧5 cells/6-well , infected by centrifugation and stable cells were selected with Puromycin ( 1 . 5 μg/ml for 16HBE , 2 . 5 μg/ml for MCF7 ) for 2–5 days . Short-term stable pools of cells were prepared for each experiment to avoid clonal effects . shRNA Cdc42 depletions were performed using hairpins cloned into pSUPER Puro vector: shCdc42 . 1 ( cctgatatcctacacaacaaa ) , shCdc42 . 2 ( cagatgtatttctagtctgtt ) . β1 integrin ( ITGB1 ) depletions were performed using hairpins cloned into pSiren Puro vector: shITGB1 . 1 ( gccttgcattactgctgat ) , shITGB1 . 2 ( gccttgcattactgctgatat ) . Stable pools were seeded at 2 . 5 × 10∧4 cells/24-well and incubated for 3 days before analysis . siRNA depletions were performed using the following duplexes ( Dharmacon , Lafayette , CO ) : siControl ( siLamin A/C; D-001620–02 ) , siCdc42 . 1 ( gauuacgaccgcugaguua ) , siCdc42 . 2 ( cggaauauguaccgacugu ) , siα-catenin SMARTpool ( M-010505-01-0005 ) . 16HBE cells were transfected using Lipofectamine LTX as described previously ( Durgan et al . , 2015 ) . Briefly , cells were seeded at 2 . 5 × 10∧4 cells/24-well and transfected with 1 . 25 μl Lipofectamine LTX + 1 . 25 μl 20 μM siRNA ( or 10 μM + 10 μM for co-depletions ) in antibiotic-free media , overnight . Media was changed the following day , with or without inhibitors ( see figure legends ) , and cells incubated for at least 3 days to optimise knockdown levels . Western blotting was performed as described previously ( Durgan et al . , 2008 ) . The following antibodies were used in this study: α-catenin ( Sigma C2081; RRID:AB_476830; 1:1000; blocked with 5% milk ) , Cdc42 ( BD 610929; RRID:AB_398244; 1:250 ) , GAPDH ( Santa Cruz 25778; RRID:AB_10167668; 1:2000 ) , HA ( Covance MMS-101R; RRID:AB_291262; 1:2000; blocked with 5% milk ) , ITGB1 ( cytoSM158 , kindly provided by Dr Filippo Giancotti; 1:2500 blocked with 5% milk ) , Rap1 ( Millipore 07–916; RRID:AB_2177126; 1:1000; blocked with 5% milk ) , α-tubulin ( Serotec MCA77S; 1:2000; blocked with 5% milk ) . Representative images of blots are shown . 2 × 10∧5 shControl or shCdc42-16HBE cells were seeded on 60 mm dishes and incubated for 2 days . These cycling populations were fixed with EtOH and stained with propidium iodide as described previously ( Durgan et al . , 2011 ) . DNA content was analysed using a FACSCaliber to capture 10 , 000–30 , 000 events/sample . Analysis was performed using FlowJo software . Unless otherwise indicated , immunofluorescence ( IF ) was performed as described previously ( Durgan et al . , 2015 ) . Briefly , cells were fixed using 3 . 7% formaldehyde in PBS ( 10 min , RT ) , permeablised in 0 . 5% triton ( 5 min , RT ) and then incubated with primary antibody in PBS ( 4C , overnight ) : β-catenin ( BD 610153; RRID:AB_397554; 1:100 ) , p-MLC2 ( Cell Signalling 3671L; RRID:AB_330249; 1:100 ) , LC3 ( Cell Signalling 4108; RRID:AB_2137703; 1:100 ) , LAMP1 ( BD 555798; 1:100 ) , p-Histone H3 ( Millipore 06–570; RRID:AB_310177; 1:100 ) , ZO-1 ( Invitrogen 61–7300; RRID:AB_2533938; 1:100 ) . Cells were washed in PBS and incubated with Alexa Fluor 488/568 goat anti-mouse/rabbit ( H+L ) secondary ( 1:500 ) and Hoechst 3342 ( 1μg/ml ) for 45 min at RT; where indicated , HCS CellMASK Deep Red ( Thermofisher , H32721 ) or Alexa Fluor 488-phalloidin ( Cell Signalling , 8878S ) were included , to stain the cell body or actin respectively , according to the manufacturers’ guidelines . Cells were washed with PBS , then water , and mounted using Prolong Gold Antifade Mountant ( Thermofisher ) . Image acquisition was performed with a Confocal Zeiss LSM 780 microscope ( Carl Zeiss Ltd ) equipped with a 40X oil immersion 1 . 4 NA objective , using Zen software ( Carl Zeiss Ltd ) . Cells were seeded on glass-bottomed dishes ( Mattek , Ashland , MA ) and incubated/treated as shown in figure legends . Where indicated , cells were stained with CellTracker Green CMFDA , CellTracker Red CMTPX or CellMASK deep red plasma membrane stain ( Invitrogen , C10046 ) for 30 min , as recommended by the manufacturer ( Thermofisher ) and with Hoechst 33342 ( 1 μg/ml , Sigma ) , then washed and returned to normal growth media for imaging . All live microscopy were performed in an incubation chamber at 37°C , with 5% CO2; for overnight imaging , media was overlaid with mineral oil to prevent evaporation . For widefield timelapse microscopy , fluorescent and DIC images were acquired every 8 min using a Flash 4 . 0 v2 sCMOS camera ( Hamamatsu , Japan ) , coupled to a Nikon Ti-E inverted microscope , using a 20 × 0 . 45 NA objective . Image acquisition and analysis was performed with Elements software ( Nikon , Japan ) . For live confocal imaging , the same microscope , camera and software were used as described in the section above . For emission ratio imaging , 16HBE cells stably expressing the RhoA-FLARE biosensor ( a gift from Dr Klaus Hahn , Addgene #12602 ) were seeded on 35 mm glass bottom dishes ( Mattek ) and treated with siRNA as indicated . Hoechst 33342 was added prior to image acquisition . FRET images were acquired with a Zeiss 780 confocal system equipped with a 40 × 1 . 4 NA objective lens . 458 nm excitation light was used to excite the donor ( CFP ) , with donor emission collected , 464–490 nm , and acceptor ( YFP ) emission collected , 535–588 nm . Images were typically taken with a pixel size of 80 nm , a pixel dwell time of 0 . 97 ms and the pinhole set to 2 Airy units . Images were processed using ImageJ . Briefly , all images were background subtracted and the FRET image was used to make a binary mask and selection region . The FRET image was then divided by the CFP image yielding a ratio image reflecting RhoA activity . A linear pseudocolour lookup table was applied . Only cells that had a high enough signal-to-noise ratio in both the CFP and FRET signals were used . Cdc42-depleted cells were cultured on 35 mm gridded glass-bottomed dishes ( MatTek ) . Cells were stained using CellTracker Green , CellMASK Plasma Membrane dye and Hoechst , then analysed by live confocal microscopy . A mitotic cell-in-cell formation event was identified , imaged and its position recorded . The cells were quickly fixed , processed , imaged and analysed by correlative serial block face scanning electron microscopy as described previously ( Russell et al . , 2017 ) . Cells were trypsinised on day 3 post-seeding or transfection , resuspended in growth media , −/+ inhibitors , and seeded at 10∧5 cells/six-well on ultra-low adhesion dishes ( Costar ) for 8 hr . Following suspension culture , 5 × 10∧4 cells were transferred to a glass slide by cytospin at 300 rpm for 3 min , fixed in 10% TCA and analysed by IF/confocal . Cells were seeded on glass coverslips or glass-bottomed dishes , treated −/+ siRNA as indicated , and incubated for 3 days . To analyse the effects of taxol , nocodazole and STLC , the drugs were added to WT cells for a further 24 hr . To analyse the effect of Cdk1 inhibition , a Y-27632 wash-out experiment was performed . Following siRNA transfection ( 16HBE ) or seeding ( 16HBE-DN-Rap1 , MCF7 ) , media was replaced with 10 μM Y-27632 , to inhibit entosis . This treatment allowed a monolayer to form in the absence of cell-in-cell formation , thereby yielding a clean background . 3 days later , Y-27632 was washed out , to permit entosis , and replaced with either control media or 5 μM RO-3306 ( a Cdk1 inhibitor ) . Cells were fixed and analysed 24 hr later . 10∧5 WT or GFP-expressing 16HBE cells were seeded per six-well and transfected with siControl or siCdc42 . 2 , respectively . WT cells were treated with Cell Tracker Red for 30 min , and then both cell lines were trypsinised , mixed in equal proportion , then reseeded at 10∧5 cells/well on 35 mm glass bottomed dishes to yield mixed monolayers . 2 days later , the resulting entotic structures were analysed by live IF/confocal microscopy ( d3 post-siRNA ) . The colours could be reversed with no change in experimental outcome ( ie . siControl/GFP cells , siCdc42/WT-red cells ) . 10∧5 16HBE cells were seeded/35 mm glass-bottomed dish and transfected with siCdc42 . Three days post-transfection , the cells were stained with CellTracker Green and Hoechst for 30 min , then placed in fresh media for imaging . Metaphase cells were imaged by live DIC and confocal microscopy , with full z-stacks acquired . Images were analysed using ImageJ software . To measure mitotic cell spreading , cell length was measured in a basal x/y section , cell height in the z , and height/length ratio was plotted . For mitotic rounding , cell shape was analysed in the midplane x/y section of each metaphase cell , scoring for roundness ( where 1 = a perfect circle ) . 10∧5 16HBE cells were seeded/35 mm glass-bottomed dish and transfected with siControl or siCdc42 . On day 3 post-transfection , Cdc42-depleted cells were stained with Hoechst and treated with C3 ( 1 μg/ml ) , Y-27632 ( 10 μM ) or Blebbistatin ( 100 μM ) for a further 4 hr . 10 min prior to imaging , each dish of cells was treated with CellMASK deep red plasma membrane stain ( Invitrogen , C10046 ) , and cells subjected to live IF/confocal imaging . Metaphase cells were identified by DNA morphology and the basal section imaged . Basal spread area was measured using ImageJ . MCF7 cells were seeded in 3D matrigel using established protocols ( Durgan et al . , 2011 ) , and incubated to form early stage cysts ( 2–4 cells ) in which matrix contacts are retained , to minimise the induction of detachment-induced entosis . Briefly , four-well , glass-bottomed chamber slides ( Lab-Tek II; 155382 ) were coated with a thin layer of 80% Matrigel Growth Factor Reduced Membrane Matrix ( Corning; 356230 ) /20% Rat Collagen I ( Cultrex; 3440-100-01 ) , then overlaid with 5 × 10∧4 cells in 2% Matrigel/media; 10 μM Y-27632 was included to suppress basal detachment-induced entosis during seeding . Cells were incubated for 24 hr to initiate cyst formation , Y-27632 was then washed out and replaced with fresh media −/+ RO-3306 ( 5 μM ) , to allow cell-in-cell formation to proceed in 3D , in the presence or absence of mitosis . Twenty-four hour later , cysts were formalin fixed , stained for actin ( 488-phalloidin ) and DNA ( Hoechst ) and imaged by confocal microscopy . A human breast cancer microarray was obtained from US Biomax ( BR1505b ) , bearing 75 cases of invasive ductal carcinoma ( see Supplementary Figure 2 ) . Each case was represented by duplicate formalin-fixed , paraffin-embedded cores , each with a diameter of 1 mm and a thickness of 5 μm . The array was baked at 55°C for 30 min , deparaffinised in xylene ( 2 × 10 min washes , RT ) and rehydrated through sequential washes ( 2 × 100% EtOH , 1 × 70% EtOH , 1 × 30% EtOH , 3x H2O; 5 min each at RT ) . Antigens were retrieved by boiling in 1x citrate buffer ( Vector Labs ) for 20 min , then cooled back to RT . Blocking ( 45 min , RT ) and antibody incubations were performed in TBS-T/5% BSA/0 . 1M Glycine; TBS-T was used for washes . Antibodies and mounting conditions are stated above ( see IF ) , with primaries incubated 4°C/overnight , secondaries at RT/45 min and DAPI at RT/10 min . The array was stained for: ( 1 ) β-catenin , to visualise adherens junctions ( where present ) , ( 2 ) p-Histone H3 ( S10 ) , a marker of mitosis , and ( 3 ) DAPI to visualise nuclear morphology . Each 1 mm core was imaged in whole by DIC and confocal microscopy , using a 40 × 1 . 4 NA objective and tiling a 6 × 6 grid . For each core , the number of p-Histone H3 ( S10 ) -positive nuclei was counted as an indicator of mitotic activity , and the number of cell-in-cell structures scored as a measure of entosis . Six- to 8-week-old female athymic mice were implanted with β−17 estradiol pellets 3 days before tumor implantation . 1 × 10∧7 MCF7 cells were injected subcutaneously per tumour , in duplicate . When tumours reached a size of ~150 mm∧3 , mice were treated with either vehicle , or 15 mg/kg taxol . Twenty-four hour post-treatment , tumours were excised , formalin-fixed and 20 micron sections were prepared ( these relatively thick sections permit complete visualisation of whole engulfed cell-in-cell structures ) . Sections were processed and stained as described for the TMA , and analysed by DIC and IF/confocal ( β-catenin , p-HH3/DNA ) . To account for tumour heterogeneity , five well-separated fields of view ( x/y ) were captured from each of three distinct sections per tumour ( z; from the top , middle , bottom ) using a 40 × 1 . 4 NA objective . The average number of pHH3-positive cells was scored per FOV , per section . The total number of cell-in-cell structures in each section was scored . Data were analysed by Mann-Whitney U test and Student t-test as indicated , using Prism 6 software . All relevant data are available from the authors .	For over a century , scientists looking down microscopes at samples from human cancers have noticed cells eating other cells – in other words , cell cannibalism . The causes and mechanisms involved in this unusual process , which is also known as entosis , are not well understood and its relationship to cancer is complex . On one hand , cell cannibalism may promote cancer by providing nutrients for growing tumours and making it more likely that genetic errors will occur . On the other hand , this process may resist cancer by eliminating damaged cells . In the laboratory , cell cannibalism has only been seen in cells that are detached from their surroundings . Cells in the body are typically surrounded and supported by a mesh of proteins called the extracellular matrix . However , within a tumour , cancer cells can often begin to grow without being attached to the matrix , which means that cell cannibalism can occur . A protein called Cdc42 plays a part in how cells attach to each other and to the extracellular matrix , but the role of Cdc42 in controlling entosis had not been previously explored . Durgan et al . initially set out to ask whether Cdc42 was involved in the established process of cell cannibalism , as seen in detached cells . However , the experiments showed that removing Cdc42 from human cells grown in the laboratory had little effect on this method of entosis . Unexpectedly , though , the loss of Cdc42 did enable a different form of cell cannibalism in cells that remained attached to the extracellular matrix , which had not been seen before . This new cannibalism process is linked to cell division , with cells that are dividing or that have recently divided being consumed by neighbours . This form of cell cannibalism is more commonly seen in cancers where the cells divide a lot , and some chemotherapy drugs that interfere with cell division also increase the rate of cell cannibalism . During cell division a group of proteins – including RhoA and myosin – cause cells to become rounder and stiffer . Durgan et al . suggest this allows the dividing cells to force their way inside other cells , the key first stage of entosis . Since cancer cells divide often , this form of cell cannibalism may lead to the cancer cells being destroyed by their healthy neighbours , in a form of “assisted suicide” . This reveals an unexpected link between cell division and cell cannibalism , which is relevant to both cancer and chemotherapy . Future work will explore whether entosis can be used to predict how a cancer will progress in a patient , or how they will respond to a given treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2017	Mitosis can drive cell cannibalism through entosis
The etiological underpinnings of amyotrophic lateral sclerosis ( ALS ) are complex and incompletely understood , although contributions to pathogenesis by regulators of proteolytic pathways have become increasingly apparent . Here , we present a novel variant in UBQLN4 that is associated with ALS and show that its expression compromises motor axon morphogenesis in mouse motor neurons and in zebrafish . We further demonstrate that the ALS-associated UBQLN4 variant impairs proteasomal function , and identify the Wnt signaling pathway effector beta-catenin as a UBQLN4 substrate . Inhibition of beta-catenin function rescues the UBQLN4 variant-induced motor axon phenotypes . These findings provide a strong link between the regulation of axonal morphogenesis and a new ALS-associated gene variant mediated by protein degradation pathways . Ubiquilins belong to the ubiquitin-like family of proteins and act broadly as key regulators of degradative processes . The best understood function of ubiquilins is in the ubiquitin proteasome system ( UPS ) , where they recognize and bind polyubiquitinated substrate proteins through a C-terminal ubiquitin-associated ( UBA ) domain and deliver them to the proteasome through an N-terminal ubiquitin-like ( UBL ) domain ( Marín , 2014 ) . Additional roles for ubiquilins in autophagy and ER-associated degradation have also been described ( Hjerpe et al . , 2016; Lee and Brown , 2012; Lim et al . , 2009 ) . The UPS and autophagy are indispensible for multiple aspects of neuronal development ( Cecconi and Levine , 2008; Hamilton and Zito , 2013 ) and are crucial in maintaining homeostasis in the aging nervous system ( Groen and Gillingwater , 2015; Rezania and Roos , 2013 ) . Dysfunction of these pathways contributes to neurodegenerative diseases characterized by protein aggregation ( Huang et al . , 2010; Kim et al . , 2009; Wang et al . , 2011; Wu et al . , 2015 ) . Accordingly , variants in UBQLN1 and UBQLN2 have been linked to neurological disorders . The ubiquilin1 protein , encoded by UBQLN1 , has been associated with neurofibrillary tangles in Alzheimer’s disease ( AD ) and is proposed to contribute to protein aggregates characteristic of AD pathology ( El Ayadi et al . , 2012; Mah et al . , 2000 ) . Single-nucleotide polymorphisms in UBQLN1 were suggested to confer susceptibility to Alzheimer’s disease ( Bertram et al . , 2005; Kamboh et al . , 2006 ) . Mutations in UBQLN2 have been related to familial X-linked ALS/FTD ( Deng et al . , 2011 ) and heterogeneous X-linked dominant neurodegeneration ( Fahed et al . , 2014 ) . Here , we report a novel variant in UBQLN4 that is associated with ALS and demonstrate a mechanism by which wild-type UBQLN4 may regulate motor axon morphogenesis . We also reveal how dysregulation of this mechanism by the ALS-associated variant leads to abnormal motor neuron structure and function through impaired degradation and substrate retention . Mutations in UBQLN2 have been identified in ALS patients with or without dementia ( Deng et al . , 2011 ) . A variant , E54D , in UBQLN1 has been reported in a single patient with atypical motor neuron disease consistent with Brown-Vialetto-Van Laere syndrome ( González-Pérez et al . , 2012 ) . These data suggest a role for ubiquilins in motor neuron diseases . UBQLN4 , like UBQLN1 and UBQLN2 , is widely expressed and shows substantial homology to human UBQLN2 ( Marín , 2014 ) . To test if genetic variants of UBQLN4 are involved in the etiology of ALS , we screened its 11 exons with primers covering coding regions and the exon-intron boundaries . We examined 267 familial ALS index cases and 411 sporadic ALS cases , and identified a variant in a familial ALS case ( Figure 1A ) . This variant , c . 269A>C , was located in exon 3 , leading to the change of aspartate to alanine , p . D90A at the protein level ( Figure 1C ) . The female patient ( III3 ) with this UBQLN4D90A variant had an age of disease onset at 55 years , with disease duration of two years . Her mother ( II3 ) , maternal aunt ( II2 ) and a maternal cousin ( III1 ) also developed ALS and died of respiratory failure two to five years following disease onset . The variant was neither present in our 332 in-house controls , nor in any SNP databases with a total of >15 , 000 sequenced alleles nor in the Exome Aggregation Consortium ( ExAC ) database in a total of 60 , 706 unrelated individuals . The amino acid D90 is adjacent to the UBL domain of UBQLN4 ( Figure 1B ) and is highly conserved during evolution , suggesting its importance in structural and functional properties of the protein . 10 . 7554/eLife . 25453 . 003Figure 1 . The UBQLN4 c . 269A>C ( p . D90A ) variant identified in a familial ALS case . ( A ) Pedigree of a family with ALS . The proband ( III3 , arrow ) had disease onset at 55 years of age , with disease duration of 22 months . Her mother ( II3 ) died of ALS at 62 years of age without clear information regarding disease onset . Her maternal grandfather ( I2 ) died in a traffic accident without any known neurological problems . Her maternal aunt ( II2 ) developed ALS with disease onset at 51 years of age , and disease duration of 36 months . Her cousin ( III1 ) developed ALS at 56 years of age and died five years later . ( B ) Predicted structural and functional domains of UBQLN4 with an arrow indicating the position of the mutation site . Domains include a UBL: ubiquitin-like domain , aa 13–83; four STI1 heat-shock-chaperonin-binding motifs , aa 192–229 , 230–261 , 393–440 and 444–476; and a UBA: ubiquitin-associated domain , aa 558–597 . ( C ) Sequencing chromatograms of UBQLN4 wild-type allele in control and mutant allele in the patient with ALS . An adenine to cytosine substitution is present in the ALS patient , resulting in the change from aspartate to alanine at the ninetieth amino acid , D90A . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 003 To assess the effects of the ALS-associated variant , we expressed wild-type or disease-associated UBQLN4 in cultured mouse spinal motor neurons ( Figure 2A , Figure 2—figure supplement 1 ) . UBQLN4D90A-expressing cells showed a significant increase in the total number of neurites as compared to cells expressing wild-type UBQLN4 , or non-transfected cells ( Figure 2A , C ) . Importantly , these results were validated in vivo in zebrafish . When mRNAs encoding UBQLN4-WT or UBQLN4D90A were injected into zebrafish embryos , we observed abnormal motor axon branching in UBQLN4D90A but not UBQLN4-WT or uninjected embryos ( Figure 2B , D ) . Both UBQLN4-WT and UBQLN4D90A-injected fish embryos otherwise developed normally and showed neither gross morphological abnormalities nor significant changes in motor axon length , suggesting specificity of the motor axon branching phenotype . In all experiments , expression levels of UBQLN4-WT and UBQLN4D90A were comparable ( Figure 2—figure supplement 2A–C ) . These results indicate that the ALS-associated UBQLN4 variant interferes with normal motor axon morphogenesis in culture and in vivo . 10 . 7554/eLife . 25453 . 004Figure 2 . Expression of UBQLN4D90A results in motor axon branching abnormalities in vitro and in vivo . ( A ) Representative images of primary mouse spinal motor neurons transfected with pCAG-GFP alone , or co-transfected with UBQLN4-WT or UBQLN4D90A . Scale bar: 20 μm . ( B ) Representative images of lateral whole-mount zebrafish spinal cords from uninjected , UBQLN4-WT mRNA , or UBQLN4D90A mRNA injected embryos . Scale bar: 50 μm . ( C ) Quantification of total neurites in ( A ) revealed an increase in neurite number in UBQLN4D90A transfected neurons compared to pCAG-GFP-only and UBQLN4-WT transfected neurons ( n = 30 cells per group , p<0 . 0001 ) . Data are quantified from three independent experiments and are mean ± SEM . **p<0 . 0001 , one-way ANOVA with Bonferroni post-hoc test . ( D ) Quantification of percentage of motor axons with aberrant branching in ( B ) revealed an increase in the percentage of affected motor axons in UBLQN4D90A injected zebrafish compared to both uninjected and UBQLN4-WT injected controls ( n = 36 embryos per group , p<0 . 0001 ) . The difference between uninjected and UBQLN4-WT injected fish was not significant ( p=0 . 155 ) . The average motor axon length was not significantly different among three groups ( p=0 . 2034 ) . Data are from three independent experiments and are mean ± SEM . p<0 . 0001 , one-way ANOVA with Bonferroni post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 00410 . 7554/eLife . 25453 . 005Figure 2—figure supplement 1 . Cultured primary motor neurons express the motor neuron marker Islet1 . ( A ) Representative images of primary mouse spinal cord neurons transfected with pCAG-GFP and UBQLN4-WT or UBQLN4D90A , stained with Islet1 ( red ) and DAPI ( blue ) . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 00510 . 7554/eLife . 25453 . 006Figure 2—figure supplement 2 . UBQLN4-WT and UBQLN4D90A are expressed at similar levels in primary mouse motor neurons and in zebrafish embryos . ( A ) Representative images of primary mouse spinal cord neurons transfected with pCAG-GFP and Flag-tagged UBQLN4-WT or UBQLN4D90A , stained with anti-Flag antibody ( red ) and DAPI ( blue ) . Scale bar: 20 μm . ( B ) Western blot of UBQLN4/Flag levels from zebrafish embryos injected with UBQLN4-WT or UBQLN4D90A mRNA , and uninjected controls . Actin Western blot indicates equal protein loading . ( C ) Quantification of Flag signal in ( B ) confirmed that UBQLN4-WT and UBQLN4D90A mRNAs are expressed at similar levels ( p=0 . 77 ) . Data are from three independent experiments and are mean ± SEM . ns: p>0 . 05 , two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 006 Given the role ubiquilins play in the UPS , we sought to determine if UBQLN4D90A affects proteasome-mediated degradation . We used the UbG76V-GFP fusion protein as a reporter for UPS function ( Dantuma et al . , 2000 ) . The G76V substitution produces an uncleavable ubiquitin moiety that acts as a proteasome degradation signal , thereby targeting GFP for proteasomal degradation . Rates of UPS-dependent protein turnover can therefore be monitored by GFP protein level . We transfected NSC-34 cells , a motor neuron-derived cell line , with UbG76V-GFP , UbG76V-GFP + UBQLN4-WT , or UbG76V-GFP + UBQLN4D90A and compared GFP levels . UBQLN4D90A expression resulted in reduced protein turnover , as indicated by significantly greater GFP signal compared to UBQLN4-WT or UbG76V-GFP alone ( Figure 3A , B ) . We further confirmed this finding with cycloheximide protein stability assay ( Figure 3C , D ) . Cells transfected with UbG76V-GFP alone or UbG76V-GFP + UBQLN4-WT showed a gradual reduction in GFP levels due to proteasomal degradation in the presence of protein synthesis inhibitor cycloheximide . Cells transfected with UbG76V-GFP + UBQLN4D90A showed much less GFP degradation , indicating impairment of proteasomal function by the ALS-associated UBQLN4D90A variant ( Figure 3C , D ) . Taken together , these results suggest that UBQLN4D90A impairs proteasomal function . 10 . 7554/eLife . 25453 . 007Figure 3 . Expression of UBQLN4D90A impairs proteasomal degradation and results in beta-catenin accumulation . ( A ) Representative images of NSC-34 cells transfected with UbG67V-GFP alone , or co-transfected with UBQLN4-WT or UBQLN4D90A . DAPI staining is shown in blue . Scale bar: 100 μm . ( B ) Quantification of GFP levels in ( A ) revealed reduced proteasomal turnover following UBQLN4D90A expression . GFP signal , normalized to DAPI , was greater in UBQLN4D90A transfected cells than in UbG67V-GFP-only transfected cells ( p=0 . 0008 ) , or UBQLN4-WT transfected cells ( p=0 . 0088 ) . The difference in GFP level between UbG67V-GFP -only and UBQLN4-WT transfected cells was not significant ( p=0 . 1114 ) . Results are from three independent experiments and are mean ± SEM . p<0 . 001 , p<0 . 01 , one-way ANOVA with Bonferroni post-hoc test . ( C ) Cycloheximide protein stability assay . GFP protein stability is compared between UbG67V-GFP-only , UbG67V-GFP + UBQLN4-WT , and UbG67V-GFP + UBQLN4D90A transfected NSC-34 cells treated with cycloheximide for 0 , 2 , 4 , and 6 hr . Quantification of GFP level revealed impeded protein turnover in UBQLN4D90A transfected cells . GFP level of UbG67V-GFP + UBQLN4D90A transfected cells was significantly greater than that of UbG67V-GFP + UBQLN4-WT transfected cells at 2 , 4 , and 6 hr with cycloheximide treatment ( p=0 . 024 ( 2 hr ) , p=0 . 0038 ( 4 hr ) , and p=0 . 036 ( 6 hr ) ) . GFP level of UbG67V-GFP + UBQLN4D90A transfected cells was significantly greater than that of UbG67V-GFP-only transfected cells ( p=0 . 0068 ( 2 hr ) , p=0 . 0093 ( 4 hours ) , and p=0 . 032 ( 6 hr ) ) . Results are from three independent experiments and are mean ± SEM . p<0 . 01 , p<0 . 05 , one-way ANOVA with Bonferroni post-hoc test . ( D ) Representative Western blot of the cycloheximide protein stability assay . Actin serves as a loading control . ( E ) Western blot of beta-catenin levels from UBQLN4-WT , UBQLN4D90A and non-transfected NSC-34 cells . GAPDH Western blot indicates equal protein loading . ( F ) Quantification of beta-catenin signal in ( D ) indicated greater beta-catenin levels in UBQLN4D90A transfected and non-transfected cells as compared to UBQLN4-WT transfected cells ( p=0 . 0174 and p=0 . 0326 , respectively ) . Results are from three independent experiments and are mean ± SEM . p<0 . 05 , one-way ANOVA with Bonferroni post-hoc test . ( G ) Representative images of primary mouse neurons transfected with pCAG-GFP and UBQLN4-WT or UBQLN4D90A , stained for beta-catenin . Scale bar: 20 μm . ( G ) Quantification of beta-catenin localization in ( F ) revealed increased nuclear localization of beta-catenin in UBQLN4D90A transfected cells as compared to UBQLN4-WT transfected cells ( n = 22 or more cells per group , p=0 . 0038 ) . Data are from three independent experiments and are mean ± SEM . p<0 . 01 , two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 007 Recently , loss of the ubiquitin-like modifier-activating enzyme 1 ( UBA1 ) was shown to cause motor axon abnormalities similar to what we observed . In the UBA1 study , the phenotypes were shown to result from heightened beta-catenin signaling , caused by loss of Uba1 and dysfunction of ubiquitination pathways ( Wishart et al . , 2014 ) . Given our finding that UBQLN4D90A results in reduced proteasomal efficiency , we sought to determine if it also affects beta-catenin levels . We first compared beta-catenin levels in UBQLN4-WT , UBQLN4D90A and non-transfected NSC-34 cells ( Figure 3E ) . There was a significant decrease in beta-catenin in UBQLN4-WT-expressing cells compared to non-transfected cells , suggesting a role for UBQLN4 in facilitating beta-catenin degradation . Moreover , we detected a significant increase of beta-catenin in UBQLN4D90A-expressing cells compared to those expressing UBQLN4-WT , suggesting that the UBQLN4D90A mutation disrupts beta-catenin degradation ( Figure 3F ) . Immunostaining in primary mouse spinal motor neurons validated these observations , and further confirmed nuclear beta-catenin accumulation in UBQLN4D90A-expressing cells ( Figure 3G , H ) . Given our findings that UBQLN4D90A expression results in reduced proteasomal efficiency and nuclear beta-catenin accumulation , we asked if inhibition of beta-catenin could mitigate motor axon morphogenesis defects characteristic of UBQLN4D90A expression . To address this possibility we utilized quercetin , which acts by restricting beta-catenin nuclear localization to reduce beta-catenin-dependent signaling ( Park et al . , 2005 ) . Immunostaining for beta-catenin in mouse spinal motor neurons transfected with UBQLN4-WT or UBQLN4D90A revealed significantly greater nuclear accumulation of beta-catenin in UBQLN4D90A-expressing cells as compared to UBQLN4-WT-expressing cells ( Figure 4A , B ) . This accumulation was reduced by treatment with 0 . 1 μM quercetin , confirming the inhibitor’s effectiveness . We next looked at quercetin’s effects on axon morphogenesis in primary neurons and in zebrafish in the context of UBQLN4-WT or UBQLN4D90A expression . Indeed , the total number of neurites in 0 . 1 μM quercetin-treated UBQLN4D90A-expressing neurons was rescued to wild-type levels ( Figure 4C , D ) . Moreover , the percentage of motor axons with aberrant branching morphology in 50 μM quercetin-treated UBQLN4D90A-expressing zebrafish embryos was significantly reduced as compared to vehicle-treated embryos ( Figure 4E and F ) . Taken together , these findings suggest that motor axon morphogenesis phenotypes characteristic of UBQLN4D90A expression may result from beta-catenin accumulation , and that inhibition of beta-catenin is sufficient to mitigate UBQLN4D90A variant-induced phenotypes . 10 . 7554/eLife . 25453 . 008Figure 4 . UBQLN4D90A-induced phenotypes are rescued by beta-catenin inhibition . ( A ) Representative images of primary mouse neurons transfected with pCAG-GFP and UBQLN4-WT or UBQLN4D90A , treated with 0 . 1 μM quercetin or DMSO . Cells are stained for beta-catenin . Scale bar: 20 μm . ( B ) Quantification of beta-catenin localization in ( A ) revealed a rescue effect of quercetin on increased nuclear localization of beta-catenin caused by UBQLN4D90A . UBQLN4D90A transfected cells showed a dramatic increase of nuclear beta-catenin localization compared to UBQLN4-WT transfected cells ( n = 25 or more cells per group , p<0 . 05 ) . The increase was rescued by the application of quercetin ( n = 25 or more cells per group , p=0 . 48 ) . Data are from three independent experiments and are mean ± SEM . p<0 . 05 , one-way ANOVA with Bonferroni post-hoc test . ( C ) Representative images of primary mouse spinal cord neurons transfected with pCAG-GFP and UBQLN4-WT or UBQLN4D90A , treated with 0 . 1 μM quercetin or DMSO . Scale bar: 20 μm . ( D ) Quantification of total neurite numbers in ( C ) revealed a rescue effect of quercetin on increased neurite number in UBQLN4D90A transfected cells . The number of neurites present in UBQLN4D90A transfected cells was significantly greater than that in UBQLN4-WT transfected cells ( n = 30 cells per group , p<0 . 0001 ) . Quercetin treatment rescued the increased number of neurites induced by UBQLN4D90A transfection ( n = 30 cells per group , p<0 . 0001 ) . Data are from three independent experiments and are mean ± SEM . **p<0 . 0001 , one-way ANOVA with Bonferroni post-hoc test . ( E ) Representative images of lateral whole-mount zebrafish spinal cord from uninjected controls or UBQLN4D90A mRNA injected embryos , treated with DMSO or 50 μM quercetin . Scale bar: 50 μm . ( F ) Quantification of the percentage of motor axons with aberrant branching in ( E ) revealed a rescue effect of quercetin on UBQLN4D90A injected embryos . UBQLN4D90A injected embryos showed a significantly greater percentage of affected motor axons compared to uninjected controls ( n = 60 embryos per group , p<0 . 0001 ) . Quercetin treatment rescued aberrant motor axon branching in UBQLN4D90A injected embryos ( n = 60 embryos per group , p<0 . 0001 ) . Data are from three independent experiments and are mean ± SEM . **p<0 . 0001 , one-way ANOVA with Bonferroni post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 008 Here we provide the first evidence of UBQLN4 involvement in ALS , and reveal the association of beta-catenin-dependent signaling with the disease variant-induced phenotypes . Taken with reports of contributions by UBQLN1 and UBQLN2 to neurodegeneration , our findings suggest prominent neuropathological involvement of the UBQLN gene family . In the present study , we identified abnormalities in spinal motor neuron morphogenesis in primary mouse neurons as well as a zebrafish model in vivo . Furthermore , we demonstrated that the novel ALS-associated UBQLN4 variant impaired UPS function , leading to increased beta-catenin in UBQLN4D90A-expressing cells . We also showed that inhibition of beta-catenin was sufficient to mitigate these phenotypes , suggesting a role for beta-catenin signaling in regulating these physiological and pathological functions . Ubiquilins are characterized by a ubiquitin-like UBL domain and a ubiquitin-associated UBA domain that interact with proteasomes and polyubiquitinated substrates , respectively , to facilitate proteasomal degradation ( Ko et al . , 2004 ) . Because the D90A mutation lies near the UBL domain , the UBQLN4D90A variant may compromise interaction with proteasomes , consistent with our finding of proteasomal impairment in UBQLN4D90A-expressing neurons . Beta-catenin-dependent signaling has been shown to play a critical role in regulating multiple aspects of neuronal development , including neurite outgrowth ( Votin et al . , 2005 ) , axon guidance ( Avilés and Stoeckli , 2016; Maro et al . , 2009 ) , and target innervation ( Salinas and Zou , 2008; Wu et al . , 2012 ) . Therefore UBQLN4D90A expression , which compromises UPS function and causes beta-catenin accumulation , may affect motor neuron development through dysregulated beta-catenin-dependent signaling . Early neurodevelopmental processes affect mature neuron functions; motor axon morphogenesis is essential for action potential transmission and target innervation . Defects in these processes may therefore confer functional vulnerability at later stages , rendering motor neurons susceptible to degeneration in ALS . This is consistent with reports of aberrant motor axon morphology associated with ALS-related genes including TDP-43 ( Kabashi et al . , 2010 ) , SOD1 ( Clark et al . , 2016; Lemmens et al . , 2007; Ramesh et al . , 2010 ) and C9orf72 ( Burguete et al . , 2015; Ciura et al . , 2013 ) . Our observation that UBQLN4D90A expression impairs proteasome function ( Figure 3A–D ) suggests that the variant acts in a dominant-negative manner . However , our finding that beta-catenin level was still reduced in UBQLN4D90A expressing cells compared to nontransfected cells ( Figure 3E , F ) suggests that the variant results in a partial loss-of-function . This may be because in the short term UBQLN4D90A functions with reduced efficiency , but still facilitates degradation of substrates . Overtime , reduced UBQLN4 functional efficiency may overload the proteasome pathway; therefore , an initial partial loss of function can eventually lead to a dominant negative gain of function outcome . This notion is consistent with the progressive nature of ALS . It is notable that our findings link UBQLN4 with beta-catenin in the context of ALS , as recent work suggested a similar role for beta-catenin signaling in the pediatric motor neuron disease Spinal Muscular Atrophy ( SMA ) ( Wishart et al . , 2014 ) . Wishart et al . reported similar morphological phenotypes in UBA1-deficient zebrafish motor neurons in which beta-catenin expression was heightened , and showed a dose-dependent rescue effect by quercetin . The degradation of beta-catenin is known to be carried out through the ubiquitin proteasome pathway ( Aberle et al . , 1997 ) . Disruption of UBQLN4 or UBA1 function , which are both involved in the UPS pathway , could therefore lead to beta-catenin accumulation and aberrant signaling . Given the wide-ranging and critical roles in neuronal development and function played by beta-catenin , its upregulation in ALS and SMA may understandably have dire consequences for the development , function , and survival of affected motor neurons . Collectively these studies suggest heightened beta-catenin activity as a common mechanism between the adult-onset motor neuron disease ALS and the pediatric motor neuron disease SMA . Taken together , we revealed a novel role for UBQLN4 in regulating motor axon morphogenesis through the UPS . Dysregulation of this function by the ALS-associated UBQLN4D90A variant leads to compromised proteasome function and beta-catenin accumulation , conferring abnormalities in motor axon morphogenesis ( Figure 5 ) , and contributing to motor neuron degeneration in ALS . Further exploration of the underlying mechanism may provide new insights for understanding ALS pathogenesis and for therapeutic development . 10 . 7554/eLife . 25453 . 009Figure 5 . Schematic model illustrating proposed roles for wild-type ( A ) and ALS-associated UBQLN4D90A ( B ) in motor axon morphogenesis . ( A ) Wild-type UBQLN4 associates with beta-catenin through its UBA domain , and with the proteasome through its UBL domain . These interactions allow for the degradation of beta-catenin , which in turn modulates gene expression to control motor axon morphogenesis . ( B ) The ALS-associated UBQLN4D90A variant is deficient in mediating proteasomal degradation of beta-catenin , leading to its accumulation and excessive induction of gene expression . Hyperactivation of beta-catenin-controlled genes dysregulates axon morphogenesis , causing aberrant axon branching in motor neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 25453 . 009 This study has been approved by the Northwestern University Institutional Review Board . Blood samples were collected after obtaining written informed consent . Eleven sets of primers ( Supplementary file 1 ) were synthesized for PCR amplification of human UBQLN4 exons and Sanger sequencing using the EQ 8000 Genetic Analysis System ( Deng et al . , 2011 ) . A total of 267 familial ALS index cases and 411 sporadic ALS cases were sequenced . The familial case in which the UBQLN4 variant was identified was compared to numerous control large-scale reference datasets to validate the variant: these included 332 in-house controls , SNP databases with a total of >15 , 000 sequenced alleles , and the Exome Aggregation Consortium ( ExAC ) database with a total of 60 , 706 unrelated individuals ( Lek et al . , 2016 ) . This study has been approved by the Institutional Animal Care and Use Committee of the Lurie Children’s Hospital of Chicago . All studies were conducted in accordance with the US Public Health Service’s Policy on Humane Care and Use of Laboratory Animals . Wild-type AB zebrafish ( RRID:ZIRC_ZL1 ) were obtained from the Zebrafish International Resource Center ( Eugene , OR ) and maintained in standard conditions . Timed-pregnant wild-type CD1 mice ( RRID:IMSR_CRL:086 ) were obtained from Charles River Laboratories ( Chicago , IL ) . A human UBQLN4 cDNA clone was obtained from GE Dharmacon ( Clone ID: 6183942 , Accession #BU149502 ) . The coding region was excised and 5’Fse1 and 3’Asc1 sites were inserted via PCR ( UBQLN4 FseFwd 5’ GATC GGC CGG CCT ACC ATG GCG GAG CCG AGC GGG GCC GAG 3’; UBQLN4 AscRev 5’ GAT CGG CGC GCC TTA GGA GAG CTG GGA GCC CAG CAG 3’ ) . The product was ligated into a pCS2-Flag vector , and site-directed mutagenesis ( Q5 kit , NEB ) was performed to convert the 2747 adenine to cytosine ( UBQLN4 Q5 Fwd 5’ AAG GCT CAA GcT CCA GCT GCT G 3’; UBQLN4 Q5 Rev 5’ CTG AGG GGT CTT GAT GAC 3’ ) . Both wild-type and mutant constructs were verified by sequencing . UbG76V-GFP was a gift from Nico Dantuma ( Addgene plasmid #11941 ) . Capped mRNAs were generated from linearized UBQLN4 wild-type and mutant constructs via in vitro transcription using the mMessage mMachine SP6 Transcription Kit ( Ambion ) . Zebrafish embryos were injected with capped mRNA ( 200–300 pg target ) at the single-cell stage and grown until 32 hr post-fertilization . Embryos were treated with quercetin ( 50 μM in 0 . 25% DMSO in embryo medium ) or vehicle control from six hours post-fertilization until fixation . Primary mouse spinal cord motor neurons were isolated , dissociated , and cultured as described previously ( Miller et al . , 2015 ) . The NSC-34 motor neuron cell line ( RRID:CVCL_D356 ) was provided by Dr . Neil Cashman ( University of British Columbia , Vancouver , Canada; [Cashman et al . , 1992] ) . The NSC-34 line is not included in the International Cell Line Authentication Committee’s Database of Cross-Contaminated or Misidentified Cell Lines , and is negative for mycoplasma . Primary and NSC-34 cells were transfected using Lipofectamine 2000 ( Life Technologies ) according to manufacturer’s instructions . Cultures were fixed for imaging or lysed for Western blot analysis 48 hr post-transfection . For cycloheximide protein stability assay , cells were treated for 0 , 2 , 4 , or 6 hr with 100 μg/ml cycloheximide following UbG76V-GFP transfection . For quercetin treatment , a working concentration of 0 . 1 μM was used . Western blotting was performed as described ( Miller et al . , 2015 ) with the following antibodies and dilutions: mouse anti-beta-catenin ( BD-Biosciences #610153 , RRID:AB_397554; 1:500 ) , goat anti-beta-actin ( Santa Cruz #1616 , RRID:AB_630836; 1:1000 ) , rabbit anti-GAPDH ( Santa Cruz #25778 , RRID:AB_10167668; 1:1000 ) , mouse anti-GFP ( Affymetrix #14-6674-80 , RRID:AB_2572899; 1:300 ) , and mouse anti-Flag M2 ( Sigma #F3165 , RRID:AB_259529 , 1:1000 ) . Whole-mount immunostaining of zebrafish embryos was performed as follows: embryos were fixed overnight in 4% PFA at 4°C . The tissue was permeabilized with proteinase K ( 10 μg/ml ) for 40 min and blocked with 5% BSA and 1% goat serum in PBS for one hour . Embryos were incubated in primary antibody , mouse anti-znp1 ( DSHB #znp - 1 , RRID:AB_2315626; 1:500 ) , in blocking solution overnight at 4°C . Goat-anti-mouse Cy3 secondary antibody ( Jackson ImmunoResearch , 1:250 ) diluted in blocking solution was applied for one hour at room temperature . Immunostaining of cultured mouse neurons was performed as described ( Miller et al . , 2015 ) with the following antibodies and dilutions: mouse anti-DDDDK tag M2 ( Abcam #ab45766 , RRID:AB_731867; 1:1500 ) , mouse anti-Isl1 ( DSHB #40 . 2D6 , RRID:AB_528315; 1:60 ) , and mouse anti-beta-catenin ( BD-Biosciences #610153 , RRID:AB_397554; 1:300 ) . All images were acquired with the Zeiss 510 Meta Laser Scanning Microscope .	Amyotrophic lateral sclerosis , or ALS for short , is a disease in which parts of the brain and spinal cord progressively degenerate . Specifically , the condition causes the nerve cells that control movement – called motor neurons – to die . As a result , people with ALS lose control of their muscles . The cause of ALS is not known , but evidence suggests that a person’s genetics plays a role in the development of the disease . Learning which genes are involved and what they do within cells may help scientists figure out what goes wrong in patients with ALS and how to treat the condition . People with ALS often experience an abnormal build up of proteins in their brain and spinal cord . Cells normally rely on molecules working together in the so-called ubiquitin proteasome system to eliminate unwanted proteins . A few mutations linked with ALS , and some other neurodegenerative conditions , have been traced back to genes encoding parts of this protein disposal system , which may help to explain the build up of proteins . However , our understanding of the genetic causes of the disease is far from complete . Now , Edens et al . report a new mutation in a gene that encodes a protein involved in the ubiquitin proteasome system . The gene , called UBQLN4 , had not previously been linked to ALS , but looking at this gene in nearly 700 patients with ALS revealed a mutation in one patient with an inherited form of the disease . This mutation was not found in public databases that contain genetic information from tens of thousands of people without ALS . To better understand the effect of this newly identified mutation , Edens et al . recreated it in zebrafish embryos and motor neurons from mice . The mutation in UBQLN4 changed the shape of the cells in the spinal cords of the zebrafish and the mouse motor neurons . There was also a build up of excess proteins because the breakdown of proteins by the ubiquitin proteasome system was slowed . Specifically , there was an excess amount of a protein called beta-catenin , which is important for development and activity of the nervous system . Treating the mutant motor neuron cells with a drug called quercetin , which suppresses beta-catenin , reversed the defects seen in the cells . Larger studies are now needed to see how often this mutation occurs in patients with ALS and to determine if other forms of the disease might have a similar cause .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology", "neuroscience" ]	2017	A novel ALS-associated variant in UBQLN4 regulates motor axon morphogenesis
Spontaneous DNA breaks instigate genomic changes that fuel cancer and evolution , yet direct quantification of double-strand breaks ( DSBs ) has been limited . Predominant sources of spontaneous DSBs remain elusive . We report synthetic technology for quantifying DSBs using fluorescent-protein fusions of double-strand DNA end-binding protein , Gam of bacteriophage Mu . In Escherichia coli GamGFP forms foci at chromosomal DSBs and pinpoints their subgenomic locations . Spontaneous DSBs occur mostly one per cell , and correspond with generations , supporting replicative models for spontaneous breakage , and providing the first true breakage rates . In mammalian cells GamGFP—labels laser-induced DSBs antagonized by end-binding protein Ku; co-localizes incompletely with DSB marker 53BP1 suggesting superior DSB-specificity; blocks resection; and demonstrates DNA breakage via APOBEC3A cytosine deaminase . We demonstrate directly that some spontaneous DSBs occur outside of S phase . The data illuminate spontaneous DNA breakage in E . coli and human cells and illustrate the versatility of fluorescent-Gam for interrogation of DSBs in living cells . DNA double-strand breaks ( DSBs ) are the most genome-destabilizing DNA damage ( Jackson and Bartek , 2009 ) . ‘DSBs’ is used here as a collective term that includes two-ended structures ( DSBs , e . g . , as caused by double-strand endonucleases or ionizing radiation ) and single double-stranded ends of DNA ( DSEs , or one-ended DSBs ) , such as are caused by replication-fork collapses ( Kuzminov , 2001 ) . We use ‘DSE’ to refer to each single DSE in a two-ended DSB and to the sole DSE in a one-ended DSB . DSBs ( one- and two-ended ) promote deletions , genome rearrangements ( Hastings et al . , 2009 ) , chromosome loss ( Paques and Haber , 1999 ) , and point mutations ( Harris et al . , 1994; Rosenberg et al . , 1994; Strathern et al . , 1995 ) . DSB-induced genomic instability promotes cancer ( Negrini et al . , 2010 ) and genetic diseases ( O’Driscoll and Jeggo , 2006 ) , evolution of antibiotic resistance ( Cirz et al . , 2005 ) and of pathogenic bacteria ( Prieto et al . , 2006 ) including in biofilms ( Boles and Singh , 2008 ) . The latter reflect the role of DSBs in inducing mutagenesis and genome rearrangement under stress , which may accelerate evolution of bacteria ( Al Mamun et al . , 2012; Rosenberg et al . , 2012 ) , and human cancer cells ( Bindra et al . , 2007 ) . DSBs are implicated in mutation hotspots in cancer genomes ( Nik-Zainal et al . , 2012; Roberts et al . , 2012 ) . Breaks induced by ionizing radiation and alkylating drugs are used as anti-cancer therapy , and conversely DSBs are likely to foretell genomic instability that drives malignancy ( Negrini et al . , 2010 ) . Despite the importance of DSBs to many biological processes , quantification of DSBs has been limited . Moreover , although some mechanisms of DSB formation are being explicated ( Merrikh et al . , 2012 ) , the main mechanisms underlying spontaneous DNA breakage in bacterial ( Pennington and Rosenberg , 2007 ) and human cells ( Vilenchik and Knudson , 2003; Kongruttanachok et al . , 2010 ) remain elusive . DSBs have been quantified via neutral sucrose gradients ( e . g . , Bonura and Smith , 1977 ) , or pulse-field gels ( PFGE ) ( Michel et al . , 1997 ) , neither of which routinely detects DSBs present in fewer than ∼10% of a population of molecules , far above DSB levels that occur in cells spontaneously ( Pennington and Rosenberg , 2007 ) . The standard single-cell gel electrophoresis ( ‘comet’ ) assay ( Olive et al . , 1990 ) detects single-strand ( ss ) DNA nicks and DSBs , and thus is not specific to DSBs , whereas the neutral comet assay ( Wojewodzka et al . , 2002 ) is DSB-specific , but lacks sensitivity . The terminal transferase dUTP nick end-labeling ( TUNEL ) assay detects free ends of DNA , and so ( nonspecifically ) labels both ssDNA nicks and DSBs ( Gavrieli et al . , 1992 ) . Cytological assays for foci of DSB-repair proteins identify locations of DSBs in situ via surrogate markers γ-H2AX ( Rogakou et al . , 1999 ) , Mre11 , Rad50 ( Maser et al . , 1997 ) , Rad51 ( Haaf et al . , 1995 ) , Rad52 ( Liu et al . , 1999 ) , BRAC1 ( Scully et al . , 1997 ) , Ku80/70 ( Koike et al . , 2011 ) , and 53BP1 ( Rappold et al . , 2001 ) in eukaryotes , and RecA ( Renzette et al . , 2005 ) , RecFON ( Kidane et al . , 2004 ) , and bacterial Ku ( Kobayashi et al . , 2008 ) in bacteria . Only some of these may be DSB-specific . γ-H2AX and 53BP1 , the most commonly used DSB markers in mammalian cells , are indirect markers . Antibodies to γ-H2AX and 53BP1 detect a modified histone and a DNA repair protein respectively , rather than DNA ends , and are likely to label sites not currently possessing a DSB . γ-H2AX is a histone variant phosphorylated at sites of DNA damage during DNA damage signaling . γ-H2AX spreads over up to ∼2-Mbp regions that comprise 500 to 8000 γ-H2AX molecules , so does not pinpoint DSB sites ( Rogakou et al . , 1999 ) . Numbers of γ-H2AX foci induced by DNA damage may not represent true numbers of DSBs ( Bouquet et al . , 2006 ) , and γ-H2AX focus formation is variable and can occur at non-DSB sites ( Han et al . , 2006 ) . Thus , γ-H2AX may not always signify a physical break . 53BP1 is a non-homologous end-joining ( NHEJ ) protein and forms nuclear foci upon ionizing radiation ( IR ) treatment , dependently on several histone modifications including γ-H2AX , H2A/X ubiquitylation and H4K20 methylation ( Lukas et al . , 2011b; Fradet-Turcotte et al . , 2013 ) . Because these histone modifications do not exist equally throughout the genome , the efficiency and DSB-specificity of 53BP1 are not known . Moreover , γ-H2AX , and all histological markers provide ‘snapshots’ of fixed cells and do not allow the possibility of measuring accrual of DSBs over time . Although Ku is likely to be most specific for DSBs via its function in NHEJ ( Taccioli et al . , 1994 ) , Ku also functions at telomeres ( Gravel et al . , 1998 ) and appears to interact with RNA polymerase II ( Dynan and Yoo , 1998 ) , raising questions about its specificity . Ku also binds ssDNA nicks , gaps , and regions of transition between single- and double-stranded structures , supporting concerns about its specificity ( Paillard and Strauss , 1991; Blier et al . , 1993 ) . In bacteria , RecA binds ssDNA assisted by RecF . Neither is specific for DSEs of DNA . Bacterial Ku has also been used to visualize DNA damage ( Kobayashi et al . , 2008 ) . Quantification of DSBs in living E . coli was achieved by measuring the RecB-dependent ( DSB-specific ) induction of the SOS response in individual cells by flow cytometry ( Pennington and Rosenberg , 2007 ) . Although this assay allowed quantification of rates of formation of living cells with DSBs ( Pennington and Rosenberg , 2007 ) , it could not determine the numbers of DSBs per cell . Thus , to date , neither precise rates of spontaneous DNA breakage in living cells nor the mechanisms that underlie most spontaneous DNA breakage are known , even in E . coli . DSBs can arise by several different mechanisms , many involving DNA replication . Replication-fork collapse at ssDNA nicks creates one-ended DSBs ( Kuzminov , 2001 ) ( illustrated below ) . Collisions of the replisome with transcription complexes ( Merrikh et al . , 2012 ) and other proteins ( Gupta et al . , 2013 ) cause DNA breakage . In stationary-phase ( non-replicating ) cells , RNA/DNA hybrids left by transcription ( R-loops ) promote DNA breakage apparently by priming replication forks that then collapse at ssDNA nicks ( Wimberly et al . , 2013 ) . Though replication can generate DSBs , whether it is the principle generator of DSBs spontaneously , in cells/sites not specifically engineered to maximize collisions , has not been addressed . Similarly , spontaneous 53BP1 foci can be detected in G1 ( non-replicating ) human cells ( Lukas et al . , 2011a ) . Given the uncertainly of the DSB-specificity of 53BP1 , whether these reflect genuine DSBs formed outside of S phase , when most replication occurs , is unclear . Additional processes may generate DSBs in human cells . Most humans express up to nine primate-specific ssDNA deaminases , AID , APOBEC1 , and seven distinct APOBEC3s , all of which convert DNA cytosines to uracils ( Conticello et al . , 2007 ) . In human , uracils in DNA can be processed by UNG2 into abasic sites and subsequently into ssDNA nicks by APEX . Such nicks could potentially produce DSBs via opposing strand nicks or replication-fork collapses , as occurs with uracil excision from DNA in E . coli ( Kuzminova and Kuzminov , 2008 ) . Although DSBs have been inferred and associated with AID for Ig translocations in some B-cell cancers ( Robbiani and Nussenzweig , 2013 ) , DSBs have yet to be identified directly . DSBs have also been inferred as intermediates in the generation of strand-biased cytosine mutation clusters in many cancers ( Nik-Zainal et al . , 2012; Roberts et al . , 2012; Burns et al . , 2013b ) . This inference is further supported by the demonstration of similar-sized mutation hotspots targeted to DSBs created by I-SceI endonucleolytic cleavage in E . coli ( Shee et al . , 2012 ) , and by similar mutation clustering in yeast cells exposed to alkylating agents or engineered to express various human DNA deaminases ( Roberts et al . , 2012; Taylor et al . , 2013 ) . In this study , we develop engineered proteins for specific detection of DSBs in bacterial and mammalian cells , and use them to illuminate spontaneous DNA breakage in both . We created fluorescent-protein fusions of the highly DSE-specific ( Williams and Radding , 1981; Akroyd and Symonds , 1986; Abraham and Symonds , 1990 ) Gam protein of phage Mu for detection of DSBs as foci upon its expression from the E . coli chromosome , or from vectors in mammalian cells . Gam is the ortholog of eukaryotic and bacterial Ku ( d’Adda di Fagagna et al . , 2003 ) , but , unlike Ku proteins , does not perform DNA repair reactions nor bind any other known protein . During phage infection , Gam binds and protects ends of linear phage DNA , preventing degradation by host exonucleases ( Akroyd and Symonds , 1986 ) . Biochemically , Mu Gam is a highly specific DSE-binding protein ( Williams and Radding , 1981; Abraham and Symonds , 1990 ) . We demonstrate the utility of regulatable GamGFP fusion proteins for detecting DSBs in E . coli and in mammalian cells and use them to determine the rates of spontaneous DNA breakage in E . coli . Moreover , we track the origins of spontaneous DSBs in live , proliferating E . coli . We find precise correlation of DSBs with the numbers of divisions , implying that replication-dependent mechanism ( s ) underlie most spontaneous DNA breakage . In human and mouse cells , we show that GamGFP labels DSBs and we provide evidence that— ( i ) GamGFP competes with Ku for DSBs; ( ii ) 53BP1 appears less specific for DSBs than GamGFP; ( iii ) GamGFP inhibits end resection at DSBs; ( iv ) DNA cytosine deamination produces DSBs in human cells , identifying a potentially primate-specific mechanism of DNA breakage; and ( iv ) G1 cells show multiple clustered foci , implying that some spontaneous DNA breakage occurs outside of S phase when most replication takes place . We constructed a regulatable chromosomal expression cassette of Mu gam and a Mu gam-gfp fusion gene in the E . coli chromosome , controlled by the doxycyline/tetracycline-inducible PN25tetO promoter ( ‘Materials and methods’ , Figure 1A ) . Promoter-only and GFP-only controls were also constructed . Production of GFP , Gam , and GamGFP were verified by SDS-PAGE and western analyses ( Figure 1—figure supplement 1 ) . gam-gfp and a derivative gam-EmGFP fusion gene were sub-cloned into a mammalian expression system using the E . coli chromosomal construct as a template . 10 . 7554/eLife . 01222 . 003Figure 1 . GamGFP production mimics recB double-strand-exonuclease defect . ( A ) Doxycycline-inducible gam-gfp fusion construct in the E . coli chromosome . Constitutively produced TetR protein represses the PN25tetO promoter , which produces GamGFP upon doxycycline induction . oriC , origin of replication; ter , replication terminus; arrows , directions of transcription . ( B ) Phage λ assay for end-blocking activity by Mu Gam and GamGFP . Rolling-circle replication of phage λred gam is inhibited by E . coli RecBCD , which causes small plaques of λred gam on wild-type E . coli ( Smith , 1983 ) . Mu Gam protein binds and protects DNA ends from RecBCD exonuclease activity ( Akroyd and Symonds , 1986 ) and so is expected to allow rolling-circle replication of λred gam and therefore allow formation of large plaques . ( C ) λred gam plaques are small on recB+ ( WT ) and large on recB-deficient cells ( recB- ) . Plaques produced on WT cells carrying gam and gam-gfp are small when Gam and GamGFP proteins are not produced ( Uninduced ) . ( D ) λred gam produce large plaques on WT cells if Gam or GamGFP are produced ( Induced ) . ( E ) UV sensitivity of E . coli recB-null mutant compared with recB+ ( WT ) , and uninduced gam and gam-gfp carrying cells . WT ( ) , recB− ( ) , WT GamGFP , ( ) ; WT Gam , ( ) . ( F ) Induction of Gam or GamGFP with 200 ng/ml doxycycline causes UV sensitivity similar to that of recB-null mutant cells , indicating that Gam or GamGFP block RecBCD action on double-stranded DNA ends . WT , SMR14327; recB , SMR8350; WT GamGFP , SMR14334; WT Gam , SMR14333 . Representative experiment performed three times with comparable results . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 00310 . 7554/eLife . 01222 . 004Figure 1—figure supplement 1 . Production of Gam and GamGFP fusion proteins in E . coli . Doxycycline induction and detection of plasmid-borne Gam and chromosomally encoded GamGFP and GFP are performed by Coomassie blue staining following electrophoresis ( Gam ) or by western blot immunodetection ( GamGFP , GFP ) , in upper and lower panels , respectively . Cultures grown , as described in ‘Materials and methods’ , were incubated in the presence or absence ( + or − ) of 100 ng/ml doxycycline . For the western blot , protein was visualized using antibodies against GFP . Arrows indicate positions of Gam , GFP , and GamGFP within the gel . Molecular weights of protein standards are indicated to left . Strains are: promoter only , SMR14311; gam in plasmid , SMR13908; chromosomal gam-gfp , SMR14334; chromosomal gam , SMR14333; and chromosomal gfp , SMR14332 . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 00410 . 7554/eLife . 01222 . 005Figure 1—figure supplement 2 . Long-term GamGFP production reduces E . coli viability . Greater viability loss with GamGFP than Gam implies that GamGFP is a superior DSE trap . ( A ) We quantified the effect of long-term Gam and GamGFP production on cell viability by inducing Gam or GamGFP production briefly in split log-phase cultures , then plating them for viable colony-forming units ( cfu ) on inducing or non-inducing solid medium with long-term overnight incubation . Saturated LBH cultures of SMR14327 ( WT ) , SMR14333 ( Gam ) and SMR14334 ( GamGFP ) were diluted 1:100 in fresh LBH medium and grown shaking at 37°C for 90 min , then either induced with 200 ng/ml of doxycycline or not , and grown for an additional 2 hours shaking at 37°C prior to plating for cfu on LBH solid medium with or without 200 ng/ml doxycycline . The colonies were scored after overnight incubation at 37°C . We observe that induced cultures of the Gam- and GamGFP-producing strains show , respectively , 32 ± 9% and 0 . 47 ± 0 . 06% the number of viable cfu as either WT cells with no gam gene or uninduced gam- or gam-gfp-containing cells ( mean ± SD , three experiments ) . These data imply that DSBs bound by GamGFP are not repaired or are repaired inefficiently . Whereas the viability of Gam producers is similar to that of liquid cultures of E . coli recBC DSB-repair-defective cells , which typically contain ∼30% viable cells ( e . g . , Miranda and Kuzminov , 2003 ) , GamGFP producers have lower viability . These data suggest that GamGFP is a more permanent DSE-trap and blocker of repair than Gam is , and that there is residual RecBC-independent DSB repair in recBC-defective cells . We speculate that the GFP moiety might confer more permanence to the GamGFP binding of DSBs either because the GamGFP protein inherently possesses a reduced dissociation constant or perhaps because the GFP moiety instigates multimerization with other GFP moieties in other GamGFP molecules present in the cell . This could both confer its outstanding focus-forming ability and might additionally retard end dissociation . ( B ) We find no cfu-reducing effect of production of GFP alone . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 005 We show that chromosomally encoded Gam and GamGFP are functional in E . coli by demonstrating that their production blocks the action of RecBCD , a highly DSE-specific dsDNA exonuclease , in two assays . First , phage lambda ( λ ) lacking its own Gam protein ( which , unlike Mu Gam , is a RecBCD-binding protein ) and Red recombination proteins forms small plaques because RecBCD DSE-dependent exonuclease prevents λ rolling-circle replication ( Smith , 1983 ) ( Figure 1C ) . By contrast , λred− gam− forms large plaques on recB- ( or recC or recD ) -defective E . coli because rolling-circle replication occurs ( Smith , 1983 ) ( Figure 1C , D ) . We show that wild-type E . coli producing either Mu Gam or GamGFP allow large plaque formation by λred− gam− , equivalent to those seen on recB-null-mutant E . coli ( Figure 1D ) . We conclude that chromosomally encoded Mu Gam and GamGFP block RecBCD exonuclease activity , implying that they are functional for the Mu Gam DSE-binding activity ( Williams and Radding , 1981; Akroyd and Symonds , 1986; Abraham and Symonds , 1990 ) . Second , recB-null cells are highly sensitive to ultraviolet ( UV ) light ( Willetts and Clark , 1969 ) because they are DSB-repair deficient , and UV-induced damage can lead to DSBs ( Bonura and Smith , 1975 ) , which are lethal if not repaired ( Figure 1E ) . We find that induction of Gam or GamGFP in wild-type E . coli creates a phenocopy of the recB UV sensitivity that is almost identical to that of recB− cells ( Figure 1F ) . These data , and those from the λ assay , show that Gam or GamGFP blocks RecBCD DSE-dependent exonuclease/DSB-repair activity , implying that chromosomally produced GamGFP binds DSEs of DNA in living E . coli . Additionally , we also show that long-term Gam or GamGFP production confers poor viability , expected for DSB-repair-deficient cells , supporting a complete block to DSB repair ( Figure 1—figure supplement 2 ) . recBC null-mutant cells show severe loss of viability , with only about 30% of cells present in liquid cultures being viable and able to produce colonies ( e . g . , Miranda and Kuzminov , 2003 ) . This results , presumably , from reduced repair of spontaneous DSBs . In Figure 1—figure supplement 2 , we show that long-term Gam production causes a similar low viability of 32 ± 9% viable cells relative to uninduced or wild type ( WT ) cells . GamGFP shows even further reduced viability ( 0 . 5% ± 0 . 06% ) , even though GFP production alone causes no reduction in viability . As discussed ( Figure 1—figure supplement 2A legend ) , the data imply that GamGFP is a better blocker of DSB repair than Gam , and that it blocks RecBC-dependent and also one or more RecBC-independent , residual DSB-repair pathways . We used the chromosomal regulatable I-SceI double-strand endonuclease ( Gumbiner-Russo et al . , 2001; Ponder et al . , 2005 ) and chromosomal I-SceI cutsites ( I-sites ) ( Shee et al . , 2012 ) to make site-specific DSBs in the E . coli chromosome and show that GamGFP forms foci at DSBs in living cells ( Figure 2A–C ) . First , when GamGFP is produced for 3 hr in cells without I-SceI endonuclease , spontaneous foci ( presumably reflecting spontaneous DSBs , validated below ) are visible in ∼7 . 5% of cells , a number higher than the 2 . 1% of cells with DSBs reported in a previous assay ( Pennington and Rosenberg , 2007 ) . This discrepancy , we show below , reflects different growth medium ( which affects growth rate and the number of chromosomes per cell ) from that of the previous assay . Second , GamGFP forms foci in almost all cells when I-SceI is induced in cells carrying an I-SceI cutsite ( Figure 2B , C ) , and not in cells expressing only the enzyme ( no cutsite , not shown ) , or carrying the cutsite but no enzyme ( Figure 2C , “spontaneous” ) . These data imply that DSBs underlie foci . Third , we varied the number of DSBs per cell by using rapidly growing cells with the I-site either near the replication origin ( ori ) ( more DNA copies , so more DSBs upon I-SceI induction ) or near the replication terminus ( fewer DNA/I-site copies , so fewer DSBs , Figure 2A ) . qPCR showed ∼2–3 times more ori- than ter-proximal DNA copies under these conditions ( Figure 2—figure supplement 1 ) . We find that although the number of cells with foci is the same with the I-site near ori as ter , the number of foci per cell is significantly greater when cleavage is near ori than when it is near ter ( Figure 2B , C ) . Whereas the cells with an ori-proximal I-site had 57 ± 2% of cells with >1 focus , those with the ter-proximal I-site had a significantly lower 14 ± 3% of cells with >1 focus ( p=0 . 0001 , Student’s t test ) . The data show that foci form proportionately to the number of DSBs per cell and imply that foci form at the DSB sites ( supported independently below ) . 10 . 7554/eLife . 01222 . 006Figure 2 . GamGFP foci at DSBs in living E . coli . ( A ) Strategy . In log-phase replicating E . coli , cells have more copies of origin ( oriC ) -proximal than terminus ( ter ) -proximal DNA and so will have more DSBs per cell when cleaved by chromosomally encoded I-SceI ( Ponder et al . , 2005 ) at a cutsite ( red arrow/green flash ) near ori than near ter . ( B ) Representative data ( arrows indicate foci ) . ( C ) Quantification from multiple experiments shows correlation of GamGFP foci with numbers of DSBs per cell . Cells have >1 focus when cleaved by I-SceI near ori , usually 1 focus per cell when cleaved by I-SceI near ter , far fewer cells with foci when only spontaneous DSBs are present ( no I-SceI cleavage ) , and <0 . 03% of cells with foci when GFP alone is produced . E . coli strains: GFP only , SMR14332; GamGFP , SMR14350; oriC DSB , SMR14354; ter DSB , SMR14362 . Error bars , ± SEM . ( D ) Strategy: a site-specific ssDNA nick made by TraI ssDNA endonuclease at oriT in the F plasmid becomes a one-ended DSB upon replication by fork collapse ( Kuzminov , 2001 ) . ( E ) TraI-dependent GamGFP foci imply that GamGFP detects one-ended DSBs . Cells with no F plasmid ( F− ) , an F′ plasmid encoding TraI ( F′ ) , or an isogenic traI-deleted F′ ( F′ΔtraI ) : strains SMR14015 , SMR16387 , and SMR16475 . ( F ) GamGFP foci are correlated with dose of DSB-producing γ-radiation . Figure 2—figure supplement 2A shows linear correlation of foci with dose . Strain , SMR14350 . Cells with 1 focus , green; >1 focus , red . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 00610 . 7554/eLife . 01222 . 007Figure 2—Figure supplement 1 . Quantitative real-time PCR shows ∼three-fold more DNA copies near ori than ter in log-phase , regardless of I-SceI cleavage , implying that some cells have two and some have four ori:ter regions . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 00710 . 7554/eLife . 01222 . 008Figure 2—figure supplement 2 . Linear gamma-ray dose-response and bleomycin induction of GamGFP foci in E . coli . ( A ) Numbers of GamGFP foci are linearly correlated with dose of DSB-producing γ-irradiation . Numbers of foci at different doses of γ-irradiation were calculated from the data displayed in Figure 2F . ( B ) GamGFP foci form in response to bleomycin induced DSBs in living E . coli . Twenty µg/ml bleomycin ( BLC ) promotes GamGFP foci . Green , 1 focus per cell; red , >1 focus per cell . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 008 These data also imply that a single focus occurs at each I-SceI-induced DSB: one focus for each of the two DSEs present in the DSB . Because two foci per cell can be detected when more than one chromosome has a DSB ( Figure 2A–C , E , F ) , we infer that there is not an inherent limit by which only one focus can form in a cell . Rather the two DSEs present in a single DSB appear to be kept close enough to each other , either by GamGFP or perhaps by an E . coli DNA-repair or other protein ( s ) , that only a single focus is visible . DSBs formed by I-SceI cleavage contain two double-stranded DNA ends ( DSEs ) . Our data suggest that GamGFP binds at both DSEs next to each other and forms a single focus . By contrast a one-ended DSB is expected to result when replication encounters a ssDNA nick ( Figure 2D ) via fork collapse ( Kuzminov , 2001 ) , a postulated frequent occurrence . We mimicked such fork collapses using the constitutive ssDNA nicking that occurs in the E . coli F conjugative plasmid . F plasmids are constitutively nicked by TraI ssDNA endonuclease , at a specific site , oriT , to initiate conjugal DNA transfer ( Traxler and Minkley , 1988 ) . ssDNA nicks become single DSEs ( one-ended DSBs ) upon replication-fork collapse ( Kuzminov , 2001 ) ( Figure 2D ) . We observe a TraI-dependent , four-fold increase in GamGFP foci in F′-carrying cells compared with isogenic F− cells ( Figure 2D , E ) , implying that GamGFP also labels one-ended DSBs , in addition to two-ended DSBs such as those created by double-strand endonuclease I-SceI . We used two known DSB-inducing treatments to quantify the efficiency of GamGFP focus formation relative to known efficiencies of DNA breakage by one of these agents , and to generalize our conclusions with the other . Gamma rays are widely used to induce DSBs , and the DSB load per gray ( Gy ) of ionizing radiation ( IR ) in E . coli is documented ( Bonura and Smith , 1977 ) . We find that focus formation is linearly related to gamma-ray dose over the range of 0–140 Gy ( r2 = 0 . 991 ) ( Figure 2F , Figure 2—figure supplement 2A ) . We observed 3 . 07 foci per cell given 140 Gy ( 3 . 12 foci/cell at 140 Gy less 0 . 043 foci/cell at 0 Gy ) , or 0 . 022 foci/cell/Gy . Comparing this with the figure obtained by sucrose sedimentation of 0 . 031 DSBs/cell/Gy for E . coli ( Bonura and Smith , 1977 ) , we infer an efficiency of detection of DSBs as GamGFP foci of 71% ( 0 . 022/0 . 031 = 0 . 71 ) . The 71% efficiency of detection should be considered a rough estimate because although we used identical growth medium and conditions to those used previously ( Bonura and Smith , 1977 ) , and the number of DSBs per E . coli cell per Gy is expected to be constant , we did not perform independent measurement of DSBs after IR by neutral sucrose gradients as per Bonura and Smith ( 1977 ) . Therefore , some variation is possible . However , using an independent method below , we obtained a roughly similar estimate of efficiency ( see GamGFP pinpoints subgenomic locations of DSBs in E . coli ) . The smaller number of foci per cell with the zero dose ( 0 . 043 ) than observed in Figure 2C ( spontaneous , 0 . 075 ) , reflects the poorer growth medium used in these IR experiments: M9 0 . 4% glucose exactly as per Bonura and Smith ( 1977 ) vs rich LBH medium in Figure 2C . E . coli produces more chromosome copies per cell in LBH than M9 medium ( Pennington , 2006 ) , and so is expected to have more spontaneous DSBs per cell in rich medium than poor ( also shown below ) . Bleomycin also induces DSBs ( Hecht , 2000 ) . We see that log-phase cells treated with bleomycin show significantly increased foci . About 60% have one focus and ∼34% have ≥2 foci ( 94% of cells with foci total ) , about 14-fold higher than spontaneous focus levels ( Figure 2—figure supplement 2B ) . These results generalize the DSB-labeling activity of GamGFP , and the results with gamma rays indicate that a high proportion , ∼71% , of DSBs are detected by GamGFP in E . coli . We show that GamGFP foci indicate the subcellular/subgenomic locations of DSBs in E . coli using site-specific I-SceI cleavage combined with a fixed chromosomal tetracycline operator ( tetO ) array bound by a Tet repressor ( TetR ) -mCherry fusion protein , which forms a focus at a site near oriC ( Figure 3A ) . In cells carrying this chromosomal-site label , we introduced an I-SceI cutsite ( I-site ) either 10 kb , 55 kb , 80 kb or 2 . 4 Mb away in different strains ( Figure 3B , D , F , H ) and quantified co-localization of GamGFP and TetR-mCherry ( representative data , Figure 3C , E , G , I ) . We find that GamGFP foci co-localize with the TetR-mCherry focus , producing a yellow focus , about 80% of the time with the I-site 10 kb from the array ( Figure 3C , J ) . The remaining 20% of cells had average interfocal distances of ∼0 . 3 μm ( Figure 3K ) . With 55 kb , 80 kb , and 2 Mb separating the I-site and tetO array , the mean distances between green and red foci increased to 0 . 45 μm , 0 . 52 μm , and to 0 . 57 μm respectively ( Figure 3K ) and the number of cells with overlapping ( yellow ) foci decreased ( Figure 3J ) . With the most distant I-site , most cells had one green ( ter-proximal ) focus and two red ( ori-proximal ) foci with the average distance ∼0 . 84 μm ( 2 . 4 Mb far , Figure 3K ) and 0 . 57 μm ( 2 . 4 Mb near , Figure 3K ) between the single green focus and each of the two red foci . The percentages of co-localization were significantly different for 10 , 55 and 80 kb ( p=0 . 00003 , 0 . 00001 , and 0 . 00006 , Student’s t test ) and not between 80 kb and 2 . 4 Mb ( p=0 . 053 ) . These data imply that sites farther than 80 kb apart are not necessarily further apart in space within the bacterial nucleoid ( 3-D chromosome structure ) , at least not after the nucleoid has suffered double-strand cleavage . Interfocal distances were more variable and less significantly different than the proportion of co-localization , probably reflecting the dynamic nature of the chromosome within the nucleoids of these living cells . Our data show that co-localization of GamGFP and TetR-mCherry foci occurs when the cutsite is near to and not when it is far from the array . These results indicate first , that GamGFP can diagnose subgenomic locations of DSBs . Second , the data imply that beyond 80 kb , genomic locations are roughly the same physical distance apart in cleaved bacterial nucleoids ( shown previously for sites >200 kb apart in uncleaved nucleoids [Wang and Sherratt , 2010] ) . Third , the data support the conclusion ( above ) that GamGFP foci occur at DSB sites . 10 . 7554/eLife . 01222 . 009Figure 3 . Subcellular/subgenomic localization of DSBs in living E . coli . ( A ) Strategy: we varied the location of I-SceI cleavage sites ( I-sites ) in different strains relative to a fixed-position chromosomal TetR-mCherry-bound tetO array , with GamGFP temperature inducibly produced from chromosomal λPR ( cIts857 PRgam-gfp ) . Red circle , plasmid that produces TetRmCherry . ( B–H ) Diagrams of E . coli chromosomes with inducible I-SceI endonuclease and I-sites engineered ( B ) 10 kb , ( D ) 55 kb , ( F ) 80 kb , and ( H ) 2 . 4 Mb from the tetO array . Co-localization of TetR-mCherry ( ) and GamGFP foci ( ) results in a yellow focus ( ) . ( C , E , G , I ) Representative fluorescence microscopy results show co-localization of mCherry and GamGFP ( yellow foci ) at 10 kb ( C ) , and non-overlapping foci at 55 kb ( E ) , 80 kb ( G ) , 2 . 4 Mb ( I ) in strains SMR16600 , SMR16711 , SMR16713 , and SMR16606 . ( J ) Percentage of cells with yellow overlapped foci at each distance . ( K ) Mean interfocal distances . At 2 . 4 Mb , there were frequently two red foci per one green focus , reflecting more copies of ori- than ter-proximal DNA during replication . The greater interfocal distance ( far ) is plotted separately from the shorter ( near ) , and cells with 1:1 ratios were counted separately . Data represent three independent experiments , error bars indicate SEM , with the number of cells counted in all three totaling: 298 , 10 kb; 204 , 55 kb; 333 , 80 kb; and 1347 , 2 . 4 Mb . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 009 These data also allow an independent ( though equivocal ) estimation of the efficiency of GamGFP focus formation at DSBs . Because red tetO-array foci label chromosomes , we can use the fraction of green ( GamGFP ) per red ( chromosome ) focus to approximate the efficiency of GamGFP focus formation at DSBs per chromosome . This method has the caveat that we do not know the efficiency of I-SceI cleavage of the chromosomes under our induction conditions . Nevertheless , using the construct in which the tetO array and I-site are nearby , at 55 kb away , we observe 0 . 82 ± 0 . 03 ( mean ± SD , three experiments ) green per red focus , indicating a rough efficiency of 82% of DSBs with a focus . This is similar to our estimate of ∼71% of DSBs with a focus , above . Previously , we estimated the steady-state frequency of proliferating E . coli with one or more spontaneous DSBs to be between 0 . 5% and ∼2 . 1% , and from this derived rates of formation of break-carrying cells between 0 . 25% to ∼1% per generation ( Pennington and Rosenberg , 2007 ) . However , two problems cloud interpretation of the previous data . First , the previous method could not distinguish whether most spontaneous DSBs occur singly in cells or via multi-break catastrophes . Second , whether most spontaneous DSBs occur replication- and thus generation-dependently was unknown ( reviewed in ‘Introduction’ ) . GamGFP allowed solution of both problems . First , time-lapse microfluidic imaging shows that most spontaneous DSBs form with precise correlation to numbers of cell divisions , and as above ( e . g . , Figure 2A–C ) they form mostly 1 DSB per cell , not in multi-break catastrophes . In microfluidic chambers , we captured images of growing microcolonies from the 1-cell to ∼100-cell stage measuring divisions and appearance of spontaneous GamGFP foci while varying cell-division rates by withdrawal of glucose from the flowing medium ( Figure 4 , Figure 4—figure supplement 1 ) . If spontaneous DSBs form independently of replication/generations , then the focus appearance might correlate with time not generations , whereas replication-dependent mechanisms of DSB formation predict correspondence with generations . In Figure 4A , cells that were kept dividing in log-phase for 9 hr , then shifted to no-glucose for an additional 18 hr , show severely slowed divisions after the shift , and a highly precise correspondence of the numbers of spontaneous DSB foci with numbers of cell divisions at all division rates ( Figure 4A , Figure 4—figure supplement 1 ) . To verify that the cells experiencing slow/no growth were still capable of forming GamGFP foci had breaks been present , we gave 20 μg/ml of DSB-producing agent phleomycin after 27 hr and found that 45 ± 5% ( mean ± SEM ) of cells then formed GamGFP foci ( Figure 4—figure supplement 1 , 32 hr ) . Thus , new DSBs could have been detected if they had formed . These results provide the first demonstration that most spontaneous DSBs in E . coli form generation-dependently and support replicative models for the origins of most spontaneous DSBs . 10 . 7554/eLife . 01222 . 010Figure 4 . Generation-dependence of spontaneous GamGFP focus formation in proliferating E . coli . Log-phase GamGFP-pre-induced cells were loaded into a microfluidic chamber in which single cells anchor then divide to form single-cell-layer microcolonies . The numbers of cell divisions and appearance of spontaneous foci were captured with time-lapse photography . Rapid growth in glucose during the first 9 hr was followed by washing cells in the same medium lacking glucose for 18 hr to slow and halt cell divisions . ( A ) Spontaneous DSB foci are correlated with numbers of cell divisions . Summary of data for six cells that became microcolonies . Blue ( ) , number of cell divisions; green ( ) , cumulative number of spontaneous foci that appear in each microfluidics micro-colony ( mean ± SEM , six microcolonies ) . ( B ) Representative 2-hr micro-colony with a GamGFP focus ( arrow ) . ( C ) Representative 15-hr micro-colony with GamGFP foci ( arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 01010 . 7554/eLife . 01222 . 011Figure 4—Figure supplement 1 . Representative data on the origins of spontaneous DSBs over time during growth , or growth retardation , visualized and quantified per ‘Materials and methods’ , Microfluidics and time-lapse fluorescence microscopy of E . coli . Numbers in the lower right of each frame are hours after loading into the microfluidic chamber . GamGFP foci are indicated with arrows . Note that cells expressing gam-gfp show variation in the amount of GFP per cell documented for E . coli ( Elowitz et al . , 2002 ) and other cells expressing any fluorescent-protein gene . This variation represents stochastic variation in transcription and mRNA accumulation , but the data scored are foci ( arrows ) . Cells were bathed in medium with glucose for 9 hr to allow log-phase growth , then cell divisions slowed and ultimately halted ( Figure 4A ) by switching to the same medium without glucose . Then at 27 hr , 20 µg/ml phleomycin was added to induce DSBs , visible as foci in 45± 5% of cells at 32 hr , to verify that had DSBs formed in the starving cells , they would have been visible as foci . These images were taken under very low-dose ( 30 ms ) exposure to fluorescent light to minimize fluorescence-induced DNA damage ( Ge et al . , 2013 ) and possible induction of GamGFP foci . Control experiments summarized in ‘Materials and methods’ show that these brief pulses did not induce GamGFP foci ( Microfluidics and time-lapse microscopy of E . coli , ‘Evidence that fluorescence exposure did not contribute to the spontaneous GamGFP foci scored ) ’ . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 011 Second , the microfluidic data provide the rate of DNA breakage per cell division directly , as follows . Figure 4A shows a constant frequency of cells with a single focus ( mean 0 . 0145 ± 0 . 006 SEM per cell division; none had >1 focus ) independently of cell-division rate . In all six experiments summarized in Figure 4A , one focus appeared per cell , and cells with a focus did not divide further ( probably because GamGFP is a DSE ‘trap’ that prevents repair of the break ( Figure 1F ) and shown for native Mu Gam protein in phage λ repair assays , Thaler et al . [1987] ) . Therefore , 0 . 0145 ± 0 . 006 foci per cell division represents the rate of formation of foci per division . Correcting this rate for ∼71% efficiency of detection of DSBs as foci ( above ) , provides a rate of 0 . 021 ± 0 . 008 DSBs per cell division . Third , the data obtained here additionally allow accurate interpretation of previous data from experiments that used flow-cytometric detection of cells with one or more DSBs ( Pennington and Rosenberg , 2007 ) for a separate rate measurement as follows: ( i ) we now know that most of these cells have one DSB , not multi-break catastrophes and so can estimate real DSB formation rates; ( ii ) whereas previously , we assumed generation-dependence , a point that was not known , here we showed that spontaneous breaks do form generation-dependently ( Figure 4 ) . Thus we can translate the previously estimated rate to real rates of DSBs per cell division . Previously , the rate of formation of cells with ≥1 DSB per generation was estimated to be 0 . 01 ( Pennington and Rosenberg , 2007 ) . We can correct this for the number of DSBs per cell by noting that under the same growth conditions ( exponential growth in minimal glucose ) we observed 108 foci in 98 cells with foci , or ∼1 . 1 foci per focus-carrying cell . Applying this function to an estimate of 0 . 01 of cells producing ≥1 DSBs per generation ( Pennington and Rosenberg , 2007 ) , we have 0 . 01 × 108/98 = 0 . 011 DSBs per cell division . This is similar to the 0 . 021 ± 0 . 008 DSBs per cell division obtained from the microfluidic data above , and both are far lower than initial estimates ( Cox et al . , 2000 ) , discussed below . These rates hold for log-phase cells grown in minimal 0 . 1% glucose medium , in which most cells possess two chromosomes ( Pennington and Rosenberg , 2007 ) , the growth medium and condition used by Pennington and Rosenberg ( 2007 ) . When growing in rich LBH medium , as we did in Figure 2A–C , replication is faster , the number of chromosomes per cell is increased , and the frequency of cells with DSB ( s ) is higher ( Pennington , 2006 ) . Similarly , we observed the higher 0 . 075 foci per cell in rich medium ( Figure 2C ) . In the moderately richer M9 0 . 4% glucose medium used in the zero-dose of the IR experiments ( Figure 2F ) , we observed the moderately higher frequency of 0 . 043 foci per cell . These data support the correlation between growth rate/the number of chromosomes per cell and spontaneous focus/DSB frequency or rate . We find that HeLa cells expressing GamGFP , in which DNA is sheared by a laser beam across the nucleus , display recruitment of fluorescence signal to the laser line ( Figure 5A , Figure 5—figure supplement 1A ) . GamGFP co-localized with 53BP1 visualized by immunofluorescence staining in the same laser-treated and fixed samples ( Figure 5A ) . We observed less robust GamGFP localization at laser damage in living ( Figure 5—figure supplement 1A ) than fixed ( Figure 5A ) HeLa cells , perhaps because pre-extraction of soluble GamGFP during fixation increases the apparent signal from the bound GamGFP ( Figure 5—figure supplement 1B ) . We tested whether competition with Ku for DNA end-binding might be a factor affecting Gam localization to DSBs , as seen in yeast ( d’Adda di Fagagna et al . , 2003 ) . We found ∼three-fold better labeling of laser-induced DSBs in Ku-deficient cells ( lacking Ku80 , also known as Xrcc5−/−; in which Ku70 is also downregulated , Taccioli et al . , 1994 ) compared with heterozygous Ku80-competent cells ( Xrcc5+/− ) or Lig4−/− ( end-joining-defective but Ku-competent ) mouse embryonic fibroblast cells ( MEFs ) ( Figure 5B–D , Figure 5—figure supplement 2 ) . Because GamGFP is inhibited by Ku even in end-joining-defective cells lacking LigIV ( Figure 5—figure supplement 2 ) , we infer that competition with Ku reduces GamGFP recruitment to DSBs independently of NHEJ , not that end-joining reduces the numbers of DSBs present for GamGFP to bind . These data indicate that GamGFP labels DSBs in mammalian cells . 10 . 7554/eLife . 01222 . 012Figure 5 . GamGFP marks DSBs in mammalian cells and is inhibited by Ku . ( A ) GamGFP co-localizes with 53BP1 on laser-induced DNA breaks . ( B ) Ku inhibits recruitment of GamGFP to laser-induced damage , live cells . ( C and D ) Ku inhibits recruitment of GamGFP , fixed cells . Mean ± SEM of three experiments , n >25 cells each . ( E ) GamGFP forms IR-induced foci in Ku80-defective MEFs . ( F ) Zoomed image from E . ( G ) IR-induced foci containing Gam only , 53BP1 only or both Gam and 53BP1 ( >2600 total foci counted in three independent experiments ) . Error bars , SEM . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 01210 . 7554/eLife . 01222 . 013Figure 5—figure supplement 1 . GamGFP marks DSBs in mammalian cells . ( A ) Live analysis of GamGFP localization to laser-induced DNA damage . Hela cells producing GamGFP were laser damaged along the cell track indicated by the red line at 0 min ( m ) and images were taken at the indicated times as shown . ( B ) GamGFP co-localizes with γH2AX in fixed Hela cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 01310 . 7554/eLife . 01222 . 014Figure 5—Figure supplement 2 . Ku inhibits GamGFP recruitment at DSBs independently of non-homologous end joining . Cells lacking either Ku80 or LigIV are defective in non-homologous end joining ( NHEJ ) , yet the presence of Ku still inhibits recruitment of GamGFP to laser-induced DSBs even in NHEJ-defective cells , and thus independently of the cell's ability to complete NHEJ . Whereas it could have been possible that reduced GamGFP recruitment in the presence of Ku was caused by reduced persistence of DSBs due to their repair by NHEJ , our data show instead that Ku inhibits recruitment independently of successful NHEJ and imply that Ku binding to DSEs itself is inhibitory . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 014 We examined co-localization of GamGFP with γ-H2AX and 53BP1 foci in Ku80-deficient MEFs , in which GamGFP binds DSBs efficiently , without inhibition by Ku ( Figure 5B–D ) , as discussed above . We find that γ-H2AX labeling of laser damage co-localizes with GamGFP ( Figure 5C ) . Interestingly , though both GamGFP and 53BP1 form foci on Gamma-irradiated Ku80-deficient MEFs , only ∼31% of foci per cell were coincident 53BP1 and GamGFP ( Figure 5G ) . About 46% showed only 53BP1 and ∼23% showed only GamGFP ( Figure 5G ) . The coincident foci of GamGFP with γ-H2AX ( Figure 5C ) and 53BP1 ( Figure 5A , E–G ) validate both of these markers , both indirect inferred DSB markers , as genuine DSB markers . However , the data also imply that , as expected , some of the sites that 53BP1 binds might not , at the moment of binding , possess a frank DSE because 46% are not bound simultaneously by GamGFP ( Figure 5G ) . This result is expected because , first , Gam is specific for flush DSEs with up to a four-base single-strand DNA overhang ( Akroyd and Symonds , 1986 ) , not the long single-strand DNA overhangs created by resection of DSBs by repair exonucleases ( Symington and Gautier , 2011 ) . Therefore , there are expected to be 53BP1 foci unoccupied by GamGFP , as we observe in Figure 5G , because GamGFP is a more specific reagent . Second , 53BP1 is expected to have a post-DSB-repair presence because it binds modified nucleosomes ( Lukas et al . , 2011b; Fradet-Turcotte et al . , 2013 ) rather than DNA , which is also compatible with our data . The mechanism that predominates remains to be determined . Similarly , the ∼23% of foci per cell with GamGFP alone indicates that some DSBs are present but not bound by sufficient 53BP1 to form a visible focus . These DSBs could lack DNA-damage-response signaling due to occlusion of the ends by GamGFP or exist in areas without the chromatin response that recruits 53BP1 . At least in E . coli , GamGFP focus formation is very rapid , with 91 ± 0 . 5% ( mean ± SEM , three independent experiments ) of the foci resulting from a 10-min ( pulse ) exposure to 20 µg/ml phleomycin appearing at 10 min . Whether because of timing , location or signaling , the data suggest that a more accurate picture of the locations of current DSBs containing frank DSEs might be obtained with GamGFP than 53BP1 . We find that GamGFP appears to block resection . We quantified RAD51 foci ( single-stranded DNA ) ( Raderschall et al . , 1999 ) induced by IR , thus including DSBs , in S-G2 ( CyclinA-positive ) cells that either were or were not simultaneously transfected with the GamGFP vector . We observed an inverse correlation between RAD51 foci and GamGFP-positive cells . Most of the cells that produced GamGFP had few ( 0–5 ) RAD51 foci , in optically sectioned nuclei in which all or most foci are expected to have been detected ( Figure 6 ) . Those nuclei without GamGFP had increased RAD51 focus formation . These data indicate exclusivity of the presence of GamGFP and resection , implying that as in E . coli ( Figure 1C–F ) , GamGFP blocks exonuclease activity at DSEs in mammalian cells . 10 . 7554/eLife . 01222 . 015Figure 6 . GamGFP inhibits IR-induced RAD51 foci , apparently blocking end resection . We quantified RAD51 foci ( single-stranded DNA ) ( Raderschall et al . , 1999 ) induced by IR , so presumably at DSBs , in S-G2 ( CyclinA-positive ) cells that either did or did not produce GamGFP , from the same transfections . ( A ) The GamGFP-positive Ku80-defective MEFs display reduced RAD51 foci upon IR treatment in S/G2 cells . Cells were analyzed by immunofluorescence with the indicated antibodies . S/G2 cells were identified by positive staining of CyclinA . Dotted white lines mark cell nuclei . ( B ) Quantification of RAD51 foci in CyclinA-positive cells with or without GamGFP production ( cumulative values from three experiments with >75 cells total ) . Each cell is ‘Z-stacked’ ( optically sectioned ) so that all RAD51 foci were examined . The data indicate that most GamGFP-positive cells have few ( 0–5 ) RAD51 foci per cell , and that those cells with more RAD51 foci ( classes 6–10 , 11–15 , 16–20 and >20 ) are enriched among the GamGFP-negative cells . These data indicate a partial mutual exclusivity of GamGFP presence and RAD51 foci , as would be expected if Gam binding to DSEs blocks the resection that creates the ssDNA onto which RAD51 binds . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 015 APOBEC3A is one of the most potent members of a family of DNA cytosine deaminase enzymes ( Carpenter et al . , 2012 ) . Induction of APOBEC3A , or the related enzyme APOBEC3B , results in both γ-H2AX and 53BP1 foci and frequent DNA breakage as determined by the comet assay ( Landry et al . , 2011; Shinohara et al . , 2012; Taylor et al . , 2013; Burns et al . , 2013a ) . We expressed GamEmGFP ( emerald GFP ) , one of the two forms we constructed in mammalian expression vectors . We find that GamEmGFP forms foci in ∼35% of cells when GamEmGFP and APOBEC3A are co-induced , and many of the GamEmGFP foci co-localize with 53BP1 ( Figure 7 ) . The appearance of foci requires the catalytic glutamate of ABOBEC3A , strongly implying that DNA cytosine deamination leads to DSBs in human cells . As in MEFs ( Figure 5 ) , incomplete localization of 53BP1 with GamEmGFP ( Figure 7 ) implies that some 53BP1-bound sites may not have had Gam-recognizable DSBs , and some DSBs detectable by GamEmGFP did not possess sufficient 53BP1 to form a visible focus . 10 . 7554/eLife . 01222 . 016Figure 7 . APOBEC3A induces DSBs in human cells . ( A ) HeLa cells co-transfected with GamEmGFP and APOBEC3A-mCherry or catalytic mutant , APOBEC3A-E72A-mCherry . ( B ) Summary of foci observed in cells producing both GamEmGFP and A3A-mCherry or A3A-E72A-mCherry ( two independent experiments; n = 100 per experiment ) . Data are percent of co-transfected cells . ( C ) Mean number of foci per focus-positive cell co-transfected with and expressing both GamEmGFP and A3A-mCherry . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 016 Most spontaneous DSBs are thought to occur during S-phase , presumably by any of several possible DNA replication-dependent mechanisms ( reviewed in ‘Introduction’ ) . However , 53BP1 forms unusually large nuclear foci exclusively in a subset of undamaged G1 cells ( Harrigan et al . , 2011; Lukas et al . , 2011a ) . The large foci in otherwise undamaged cells are only in G1 , as shown by their CyclinA-negative state and tracking fluorescent 53BP1 throughout the cell cycle ( Harrigan et al . , 2011; Lukas et al . , 2011a ) . Although this might suggest that some spontaneous DSBs might form outside of S phase , it has been argued that not all 53BP1 foci indicate DSBs because different types of DNA damage nucleate 53BP1 foci ( Lukas et al . , 2011a ) . In this study , we examined the large nuclear 53BP1 foci in undamaged cells , previously shown to form exclusively in G1 ( Harrigan et al . , 2011; Lukas et al . , 2011a ) , and find that ∼60% of the 53BP1 foci in Ku80-deficient cells correspond to genuine GamGFP-detectable DSBs ( Figure 8 ) . Moreover , the GamGFP that coincides with 53BP1 foci contain multiple individual foci , implying that these DSBs may be in large multi-break clusters . Because GamGFP does not spread along DNA , it can resolve multiple nearby DSBs , which was not possible with 53BP1 ( Harrigan et al . , 2011; Lukas et al . , 2011a ) . These data indicate that some spontaneous DSBs form outside of S phase , when most replication occurs , and do so in clusters . 10 . 7554/eLife . 01222 . 017Figure 8 . Spontaneous DNA breakage in G1-phase cells: GamGFP shows large spontaneous G1 53BP1 foci to contain multi-break clusters . The large spontaneous 53BP1 foci in undamaged cells , which occur solely in G1 ( Harrigan et al . , 2011; Lukas et al . , 2011a ) , contain multiple DSBs that are marked by GamGFP . The GamGFP-53BP1 co-localization is more apparent in the absence of Ku . Data are from three ( Ku80-proficient ) or four ( Ku80-defective ) independent experiments of >25 cells each with Z-stacked ( optically sectioned ) nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 01222 . 017 We showed that fluorescent-protein-fusion derivatives of the highly DSE-specific Gam protein of phage Mu allow direct identification of DSBs in bacterial and mammalian cells , and we used GamGFP to illuminate origins of spontaneous DSBs in both . In living E . coli , chromosomally encoded GamGFP forms foci at I-SceI endonuclease-produced two-ended DSBs ( Figures 2A–C and 3 ) , at one-ended DSBs made by replication-fork collapse ( Figure 2D , E ) , and DSBs generated by gamma-irradiation and bleomycin ( Figure 2F , Figure 2—figure supplement 2 ) . GamGFP detects about 71–82% of DSBs , a robust efficiency . A single GamGFP focus occurs per two-ended DSB ( Figures 2A–C and 3 ) or per one-ended DSB ( Figure 2E ) . Either E . coli DNA-repair or other proteins , or GamGFP itself , might keep the two ends in close enough proximity that only a single focus forms ( discussed in ‘Results’: GamGFP pinpoints subgenomic locations of DSBs in E . coli ) . GamGFP also allowed subcellular localization of site-specific DSBs relative to a chromosomal mCherry marker in E . coli , which are distinct even as close as 55 kb ( Figure 3 ) . Similarly , GamGFP and GamEmGFP protein fusions produced in mammalian cells showed localization at laser-induced DSBs ( Figure 5 ) , were inhibited by Ku in those cells ( Figure 5B–D ) , and inhibited resection at DSBs ( Figure 6 ) , indicating that GamGFPs label DSBs also in mammalian cells . GamGFP and its derivatives also allowed important novel conclusions about the origins of spontaneous DNA breaks in E . coli and mammalian cells . In mammalian cells , GamGFPs may have considerable utility as a marker for DSBs , particularly when levels of DNA insults/damage are high . However , we note that a failure to detect DSBs using Gam may be due to competition with Ku ( Figure 5B , C ) and/or to relatively low levels of DNA damage . Additionally , in mammalian cells GamGFP foci were more dramatic in fixed than living cells ( Figure 5A , Figure 5—figure supplement 1 ) , perhaps because depletion of background unbound GamGFP in the fixed cells improved contrast . Nevertheless , this tool can be extremely valuable for a number of studies . For instance , GamGFPs allowed us to validate indirect markers γ-H2AX and 53BP1 as genuine DSB markers by showing co-localization at laser and IR-induced DSBs ( Figures 5A , C , E–G , 7 , 8 ) , and GamGFP inhibits DNA end exonucleolytic resection ( Figure 6 ) . Consistent with these results , we also demonstrate that 53BP1 foci in G1 contain DSBs that previously had been inferred only ( Figure 8 ) . We are working to develop Ku-resistant DSE-binding fusions currently . In work published while this paper was in review , Britton et al . ( 2013 ) use Ku foci for detection of DSBs similarly to our use of GamGFP . They show that Ku interacts with chromatin via RNA and that foci can be seen in RNase-treated samples . An advantage of Ku is that competition with Ku is not a problem , whereas an advantage of GamGFP is its greater specificity for double-strand ends ( reviewed in ‘Introduction’ ) . Second , one aspect of the incomplete co-localization of 53BP1 with GamGFPs is that there are DSBs identifiable with GamGFPs that are not seen with 53BP1 ( Figures 5G and 7C ) . One possible explanation is that , particularly in the Ku-defective MEFs ( Figures 5G and 8 ) , GamGFPs may be more sensitive than 53BP1 to DSBs , and might make visible DSB foci before DNA-damage signaling has allowed sufficient 53BP1 accumulation to produce a visible focus . At least in E . coli GamGFP labels DSBs very rapidly , showing most foci by 10 min after phleomycin exposure . Thus , in some circumstances , GamGFPs may be more sensitive than 53BP1 . Conversely , the presence of 53BP1 foci at sites that do not show GamGFP foci ( Figures 5E–G , 7B , C ) , even in Ku80-deficient MEFs ( Figure 5D ) , may indicate that some of the sites bound by 53BP1 may not possess frank DSEs , the Gam DNA substrate ( Williams and Radding , 1981 ) . This could be because either the DSB has been repaired , but the 53BP1 at the site has not yet dissipated , or because exonucleolytic ssDNA resection has created a long ssDNA overhang , which cannot be bound by Gam ( Williams and Radding , 1981; Akroyd and Symonds , 1986; Abraham and Symonds , 1990 ) . With either possibility , the data suggest that GamGFPs may be more selective and specific markers for DSBs . Given the widespread use of 53BP1 as a DSB marker , understanding its potential limitations of possibly lower specificity relative to GamGFPs is important . Third , we demonstrate directly that primate-specific deaminase APOBEC3A caused GamGFP foci , and thus DSBs in human cells , indicating that DSBs result from DNA cytosine deamination . This supports previous , less specific evidence from the appearance of γ-H2AX ( Landry et al . , 2011; Burns et al . , 2013a ) and 53BP1 ( Taylor et al . , 2013 ) foci induced by APOBEC3A expression . A possible mechanism could be that removal of uracil from DNA after cytosine deamination followed by cleavage at the abasic site creates a ssDNA nick , which becomes a one-ended DSB either by replication-fork collapse ( Figure 2D ) , or similar ssDNA nick creation on the opposing strand . DNA breaks occur upon uracil excision from DNA in E . coli ( Kuzminova and Kuzminov , 2008 ) . Regardless of the specific mechanisms , the data indicate that cytosine-deamination may be a general mechanism of DSB generation in mammalian cells . Cytosine deamination clearly contributes a significant part of the cancer genome mutational landscape ( Nik-Zainal et al . , 2012; Roberts et al . , 2012; Taylor et al . , 2013; Burns et al . , 2013a , 2013b ) and it is likely that some of this could be by DSB induction . In E . coli , GamGFP allowed demonstration that , first and importantly , spontaneous DNA breakage is precisely correlated with the number of cell divisions ( Figure 4 ) , providing the first evidence that most spontaneous breakage results from DNA replication-based mechanisms . Previously , DNA replication has been implicated in several mechanisms of DSB formation . These include fork collapse in engineered lambda phage ( Kuzminov , 2001 ) , fork-regression or ‘chicken-foot’ formation , to produce one-ended DSBs , in cells with defective replication proteins ( Michel et al . , 1997 ) , and DSBs created by collisions of replication forks with transcription complexes ( Merrikh et al . , 2012 ) , and other proteins ( Gupta et al . , 2013 ) . However , nearly all of the work required engineered constructs or situations ( to maximize collisions , fork collapses , etc ) or mutant proteins , and so did not address how spontaneous DSBs occur normally . Our results demonstrate the correspondence of spontaneous DSBs with the number of cell divisions and imply that some or all of these mechanisms , and replication generally , are in fact predominant causes of spontaneous DNA breakage normally in proliferating E . coli . In separate work , we found that one particular replicative mechanism , in which RNA ( R-loop ) -primed replication-fork collapse produces DSBs , occurs spontaneously in stationary-phase ( non-proliferating ) E . coli ( Wimberly et al . , 2013 ) . Second , the demonstration that spontaneous DSBs are generation dependent ( Figure 4A ) , and our finding that most spontaneous DSBs result as single events per cell , not multi-break catastrophes ( Figures 2C and 4 , Figure 4—figure supplement 1 ) , allowed calculation of the first true rates of spontaneous DNA breakage . Our microfluidic data directly show rates of 0 . 021±0 . 008 spontaneous DSBs per cell division , and the use of our data to re-interpret and correct previous indirect , low-resolution data ( Pennington and Rosenberg , 2007 ) indicates a roughly similar rate of 0 . 011 spontaneous DSBs per cell division ( ‘Results’ ) . These true rates are 10–20-times lower than initial postulates ( Cox et al . , 2000 ) , but in line with each other . Because genomic rearrangement frequencies remain the same regardless of estimations of DSB numbers , these rarer and more accurate break rates imply that each DSB is 10–20 times more genome destabilizing than had initially been postulated . Spontaneous DSBs are infrequent but dangerous , and now that their generation-dependence is unequivocal , their origins in replication are supported . Whereas mechanisms underlying spontaneous DNA breakage have been elusive , our data support major roles for replication in E . coli , and for some DNA breakage outside of S phase , in G1 in human cells . Our results show a novel DNA breakage mechanism in human cells , and indicate the utility of fluorescent-Gam for interrogating DSB formation , location , and dynamics in living cells . Understanding the origins of spontaneous DNA breakage obtainable with improved synthetic reagents such as GamGFP will illuminate the underlying causes , and possible means of prevention , of important biological consequences of DNA breakage in living cells . E . coli strains used are given in Supplementary file 1A . Bacteria were grown in LBH ( Torkelson et al . , 1997 ) or M9 minimal medium ( Miller , 1992 ) supplemented with 10 μg/ml thiamine ( vitamin B1 ) and 0 . 1% glucose as carbon source . Other additives were used at the following concentrations ( μg/ml ) : ampicillin , 100; chloramphenicol , 25; kanamycin , 50; tetracycline , 10; gentamycin , 15; sodium citrate 20 mM . Human HeLa cells were grown in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , 100 µg/ml penicillin , 100 mg/ml streptomycin and 2 mM L-glutamine . SV40 immortalized MEF cells lines ( Ku80-defective Xrcc5+/− , Ku80-proficient Xrcc5−/− , Lig4−/−; kindly provided by the Lazzerini Denchi lab , Scripps Research Institute ) were grown in the same medium containing 15% FBS . gam-gfp was amplified from E . coli SMR14354 genomic DNA with 5′-NNN NAA GCT TGC CAC CAT GGC TAA ACC AGC AAA ACG-3′ and 5′-GCA GCC GGA TCC CTT ATT TGT ATA GTT C-3′ , digested with HindIII and BamHI , and ligated into similarly cut pcDNA3 . 1 ( + ) ( Invitrogen , Carlsbad , CA , USA ) . gam-EmGFP was constructed by first amplifying gam from pcDNA3 . 1 ( + ) -Gam-EGFP using 5′-NNN NAA GCT TGC CAC CAT GGC TAA ACC AGC AAA ACG-3′ and 5′-CCT CGC CCT TGC TCA CCA TAT GTA TAT CGG CGG CGG CGG C-3′ . Second , EmGFP was amplified from plasmid pcDNA6 . 2-C-EmGFP-DEST ( Invitrogen , Carlsbad , CA , USA ) with 5′-GCC GCC GCC GCC GAT ATA CAT ATG GTG AGC AAG GGC GAG G-3′ and 5′-NNN NGG ATC CTT ACT TGT ACA GCT CGT CCA TGC CG-3′ . Finally , these two fragments were used in an overlapping PCR with the 5′-Gam and 3′-EmGFP primers above to make a gam-EmGFP fusion gene , which was digested with BamHI and HindIII , and ligated into similarly cut pcDNA3 . 1 ( + ) ( Invitrogen , Carlsbad , CA , USA ) . The constructs were verified by restriction digestion and DNA sequencing prior to functional assays . The APOBEC3A-mCherry construct and the catalytic mutant construct , APOBEC3A-E72A-mCherry , were previously described ( Lackey et al . , 2012 ) . HeLa cells at 75% confluence were co-transfected transiently with 150 ng of gam-EmGFP and APOBEC3A-mCherry or APOBEC3A-E72A-mCherry constructs using TransIt-LT1 ( Mirus , Madison , WI , USA ) and fixed 20 hr later in 4% paraformaldehyde . 53BP1 foci were detected by rabbit anti-53BP1 ( 1:1000; Novus Biologicals , Littleton , CO , USA ) followed by goat anti-rabbit Cy5 ( 1:500; Abcam , UK ) . Nuclei were stained with 0 . 1% Hoechst dye for imaging at 600X magnification using a DeltaVision deconvolution microscope ( Applied Precision , Issaquah , WA , USA ) . Production of Gam and GamGFP in E . coli was determined by SDS-PAGE and western blot analysis . Saturated LBH cultures grown at 37°C were diluted 1:100 in a fresh medium and grown 4 hr with or without addition of 200 ng/ml doxycycline . Total proteins were extracted from 1-ml culture with BugBuster® master Mix extraction buffer ( Novagen® , Madison , WI , USA ) . Samples were denatured with 6X SDS loading dye , heated to 95°C for 5 min and 10 μl of total proteins were loaded onto a 12% polyacrylamide gel . Sodium dodecyl sulfate-polyacrylamide gel ( 12% ) electrophoresis ( SDS-PAGE ) under reducing conditions was performed as per Laemmli ( 1970 ) . Relative molecular weights were determined using Precision Plus Protein Kaleidoscope Standards . Proteins were detected by staining the gel with 0 . 1% Coomassie brilliant blue R-250 . For western blot analysis of GFP and GamGFP , gels were transferred to PVDF membrane according to manufacturer’s specifications ( Amersham , GE Healthcare , Piscataway , NJ , USA ) . The membranes were blocked with 5% non-fat dry milk in PBS-Tween and probed with primary rabbit anti-GFP polyclonal IgG antibodies from Santa Cruz Biotechnology ( Santa Cruz , CA , USA ) . Bound antibodies were detected using secondary antibody ECL Plex goat-a-rabbit IgG conjugated with fluorescent dye Cy5 after 1000-fold dilution and visualized by scanning in multicolor imager Typhoon detection system . Functional validation of Mu Gam and GamGFP produced in E . coli was determined by a lambda ( λ ) plaque assay . A λred gam Chi0 strain was used to test the activity of Mu Gam and GamGFP in E . coli . Saturated tryptone broth ( TB ) cultures of E . coli were diluted 1:10 in fresh TB with 0 . 2% maltose , 5 mM MgSO4 , 10 μg/ml thymine , 10 μg/ml vitamin B1 and grown for half an hour prior to the addition of 200 ng/ml of doxycycline to induce production of Gam and GamGFP . After 2 hr of induction by doxycycline shaking at 37°C , an equal volume of 10 mM Tris 10 mM Mg ( TM ) buffer , pH 7 . 5 was added and cells were mixed with an appropriate volume of λred gam Chi0 suspension , adsorbed without shaking for 10 min at room temperature , then plated with the addition of 2 . 5 ml molten 50°C soft BBL agar ( 1% BBL trypticase peptone , 0 . 1M NaCl , 0 . 7% agar , pH 7 . 5 ) onto BBL plates ( same medium solidified with 1% agar ) with or without 200 ng/ml doxycycline . The plates were incubated overnight at 37°C before scoring size of plaques . Saturated LBH cultures of E . coli were diluted 1:100 in fresh LBH medium and grown shaking at 37°C for 90 min , at which time 200 ng/ml of doxycycline was added to induce Gam and GamGFP production . After 2 hr of induction by doxycycline at 37°C shaking , cells were plated on LBH solid medium containing 200 ng/ml doxycycline and the plated cells irradiated with different UV doses , then incubated in the dark over night at 37°C for colony quantification . Control cultures without doxycycline induction were treated otherwise identically . Saturated overnight cultures of E . coli carrying the chromosomal gam-gfp expression cassette were diluted 1:100 in fresh medium , 100 ng/ml doxycycline added to induce GamGFP , incubated for 1 hr shaking at 37°C , then 0 . 1% arabinose added to induce I-SceI production and incubated 3 hr more , then subjected to microscopic examination . Live cells were visualized using a Zeiss inverted fluorescence microscope with Axion vision software . Quantification of foci was done by Nick software . Genomic DNA was prepared from E . coli incubated as described above for the I-SceI induced DSB-associated GamGFP focus experiments using a CTAB preparation protocol ( Porebski et al . , 1997 ) . Reactions containing 10 ng genomic DNA and 350 µM primers P21 ( oriC specific ) or P22 ( terC specific ) in 1X KAPA SYBR Fast ABI prism qPCR mix ( 20 µl total volume ) were run in an Applied Biosystems ( Foster City , CA , USA ) 7900HT RT thermocycler in 96-well plates . Relative copy numbers of oriC and terC sequences were determined by the ΔCt method: the difference in amplification rates of oriC and terC ( ΔCt ) for each sample ( average of four replicate reactions ) were normalized to the ΔCt for a wild-type strain grown to saturation ( stationary phase ) shaking in LBH at 37°C for 28 hr , to contain equal numbers of oriC and terC sequences per cell . The saturated cultures of E . coli grown in LBH ( for bleomycin experiments ) or M9 0 . 4% glucose ( for gamma irradiation ) medium were diluted 1:100 into fresh LBH or M9 0 . 4% glucose , then grown 1 hr shaking at 37°C , then given 100 ng/ml doxycycline to induce GamGFP , grown an additional 1 hr , and treated with 20 µg/ml of bleomycin , or different doses of gamma radiation . A Cs137 or a Faxitron X-ray machine ( Faxitron X-ray Corporation , Tucson , AZ , USA ) was used for gamma irradiation ( experiments in Figures 2 and 5 , respectively ) . After treatment with DNA-damaging agents or irradiation , cells were incubated for an additional 3 hr before microscopic examination unless stated otherwise . Gamma irradiation was done in exponential cultures grown in M9 0 . 4% glucose with shaking at 37°C to mimic the identical growth medium and growth conditions used previously for quantification of DSBs/Gy IR in E . coli by Bonura and Smith ( 1977 ) . Elsewhere in the paper , cultures are grown in M9 vitamin B1 0 . 1% glucose , or LBH , and gave different numbers of foci per cell ( more in LBH , less in M9 B1 0 . 1% glucose ) in these richer and poorer growth media . TetR mCherry was expressed from the arabinose inducible PBAD promoter in plasmid pDB340 in cells induced for GamGFP production at 37°C , at which temperature the chromosomal λcIts857-controlled PRgam-gfp cassette is transcribed . Estimation of average inter-focal distances and co-localization of GamGFP and TetR-mCherry was determined from cells that contained both green and red foci at a 1:1 ratio . In the case of cells with DSBs near ter , many cells had two red foci and one green focus . In this case , interfocal distances are plotted separately for cells that contained either a 1:1 or 1:2 ratio of green and red foci ( Figure 3K , 2 . 4 Mb ) . In the case of two red foci , the longer and shorter interfocal distances between the red and green foci are plotted separately as ‘1:2 far’ and ‘1:2 near’ , respectively ( Figure 3K ) . Multicolor fluorescent beads on slides were used to align independent color channels . All images were acquired with a Zeiss Axio Imager Z1 microscope plus Hamamatsu Electron Multiplier charge-coupled device ( CCD ) camera . We followed the growth of single cells into microcolonies using the CellASIC ONIX Microfluidic Platform ( Millipore , Billerica , MA , USA ) including microfluidic perfusion system , microfluidic flow chamber for bacteria ( BO4A plates ) and FG software . All experiments were performed at 2psi ( flow rate of 3 µl/hr ) . Time-lapse microscopy was performed using a Zeiss HAL100 inverted fluorescence microscope . Fields were acquired at 100× magnification with an EM-CCD camera ( Hammamatsu , Japan ) . Bright field and fluorescence images ( GFP cube = Chroma , #41017; X-Cite120 fluorescence illuminator , EXFO Photonic Solutions ) were acquired and image analysis performed using AxioVision Rel . 4 . 6 ( Zeiss , Germany ) . The microscope was housed in an incubation system consisting of Incubator XL-S1 ( PeCon , Germany ) controlled by TempModule S and Heating Unit XL S ( Zeiss , Germany ) to maintain a constant 37°C environment throughout the experiments . For laser-induced damage , cells were grown on glass-bottomed dishes ( Willco Wells , Netherlands ) and pre-sensitized by adding 1 . 5 µM BrdU ( 5-bromo-2′-deoxyuridine ) for >20 hr followed by laser micro-irradiation . DNA damage was created using a 405 nm solid-state laser focused through a 63X objective lens in the epifluorescence path of the microscope system on a Fluoview 1000 confocal microscope ( Olympus , Japan ) . Laser settings ( 60% laser power , 150 scans at 20 ms/pixel ) were used that generated DNA damage specifically along the laser path in a BrdU-dependent manner . The cells were analyzed in TimeScan mode using Zero-drift compensation for auto focusing at an interval of 1–2 min for up to 30 min post-laser damage for live analysis . Images were exported as TIFF files and analyzed in ImageJ . After DNA damage induction , the cells were washed three times with cold 1× PBS , then pre-extracted by incubating the dishes in CSK buffer ( 10 mM PIPES pH 6 . 8 , 100 mM NaCl , 300 mM sucrose , 3 mM MgCl2 , 1 mM EGTA , 0 . 5% TritonX-100 ) for 1 to 5 min on ice . Cells were washed three times in room-temperature-PBS and fixed with 2% paraformaldehyde for 15 min at room temperature followed by three washes with 1× PBS , then blocked with 3%BSA in PBS for 15 min , and incubated with either 53BP1 ( Novus , Littleton , CO , USA , NB100-304 ) or γ-H2AX ( Millipore , Billerica , MA , USA , #05-636 ) for 1 hr at room temperature , or overnight at 4°C . The cells were washed three times in PBS followed by a 45 min incubation in PBS containing Alexa Fluor 594 or 647 goat-anti rabbit IgG to detect 53BP1 or Alexa Fluor 594 goat-anti mouse IgG to detect γ-H2AX . Hoechst dye ( 0 . 1% ) was added to this incubation step to stain nuclear DNA , followed by washing three times in PBS and mounting on coverslips containing Vectashield . The cells were imaged using an inverted Fluoview 1000 confocal microscope ( Olympus , Japan ) . To quantify laser recruitment of GamGFP , the cells were laser damaged and analyzed by immunofluorescence as described . Images from individual cells were taken and equal areas both in undamaged and damaged areas ( as indicated by 53BP1 positive staining ) were quantified for Gam signal using Fluoview software . The cells that showed a >30% increase in intensity of GamGFP at the damage site compared with the undamaged site were labeled positive . This analysis was performed in both Ku80-proficient and Ku80-defective MEFs for >50 cells and results are means ± SEM of three independent experiments . For quantification of GamGFP IR foci , cells were damaged and analyzed by immunofluorescence as indicated . Individual images were taken and foci counted that either contained GamGFP only , 53BP1 only , or both . The average total number of foci per cell 30 min post-5Gy IR was 34 . For >2600 foci counted , the mean percentage of foci for each of the three categories is graphed for three independent experiments ± SEM . For quantification of RAD51 foci in CyclinA-positive cells , Ku80-defective MEFs were transfected with GamGFP , treated with IR and analyzed 4 hr post-5 Gy IR . The cells were analyzed by immunofluorescence as described above except without pre-extraction with CSK buffer . RAD51 ( Abcam , ab88572 ) and CyclinA ( Santa Cruz , Santa Cruz , CA , USA , sc-751 ) antibodies were applied for 1 hr room temperature . AlexaFluor 594 goat anti-mouse and AlexaFluor 647 goat anti-rabbit IgG were used to fluorescently label RAD51 and CyclinA respectively and the cells were fixed as described . Z-stacked images were collected using a Fluoview FV1000 confocal microscope and RAD51 foci were counted in CyclinA-positive cells with or without GamGFP signal . Data from three independent experiments from >75 cells are graphed ( Figure 6 ) . For detection of spontaneous/endogenous G1 foci , we followed the demonstration of Harrigan et al . ( 2011 ) ; Lukas et al . ( 2011a ) that large 53BP1 foci in undamaged cells occur almost exclusively in G1 phase , as shown by their CyclinA-negative state and tracking of fluorescent 53BP1 throughout the cell cycle . We , therefore , used the criterion of large spontaneous 53BP1 foci to identify G1 cells .	Cells have developed a variety of mechanisms for repairing DNA molecules when breaks occur in one or both of the DNA strands . However , we know relatively little about the causes of these breaks , which often occur naturally , or even about how common they are . Learning more about the most common forms of DNA breakage is important because the genomic changes caused by these breaks are driving forces behind both cancer and evolution , including the evolution of drug resistance in bacteria . Shee et al . have developed a new method for detecting double-strand breaks in both bacterial and mammalian cells . The method involved combining a natural virus protein called Gam with a fluorescent protein called GFP ( short for green fluorescent protein ) to make a fusion protein called GamGFP . Gam was chosen because it binds only to double-strand breaks , traps double-strand breaks , and does not bind to any proteins . Genetic engineering techniques were used to introduce GamGFP into cells , with DNA breaks in these cells showing up as fluorescent spots when viewed under a microscope . Shee et al . used this approach to detect double-strand breaks in both Escherichia coli cells and mammalian cells , and to measure the rate of spontaneous DNA breakage in E . coli . The number of double-strand breaks in E . coli was proportional to the number of times the cells had divided , which provides support for DNA replication-dependent models of spontaneous DNA breakage . The GamGFP method also provided various insights into DNA breaks in mouse and human cells . In particular , Shee et al . found evidence for a mechanism of DNA breakage that appears to be specific to primates . This mechanism involves an enzyme that is only found in the innate immune system of primates removing an amine group from a cytosine . In future , this approach might allow the trapping , mapping and quantification of DNA breaks in all kinds of cells , and the highly specific way GamGFP binds to breaks could make it the preferred tool for studying DNA breakage in mammalian cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2013	Engineered proteins detect spontaneous DNA breakage in human and bacterial cells
In response to Ca2+ influx , a synapse needs to release neurotransmitters quickly while immediately preparing for repeat firing . How this harmonization is achieved is not known . In this study , we found that the Ca2+ sensor synaptotagmin 1 orchestrates the membrane association/disassociation cycle of Rab3 , which functions in activity-dependent recruitment of synaptic vesicles . In the absence of Ca2+ , synaptotagmin 1 binds to Rab3 GTPase activating protein ( GAP ) and inhibits the GTP hydrolysis of Rab3 protein . Rab3 GAP resides on synaptic vesicles , and synaptotagmin 1 is essential for the synaptic localization of Rab3 GAP . In the presence of Ca2+ , synaptotagmin 1 releases Rab3 GAP and promotes membrane disassociation of Rab3 . Without synaptotagmin 1 , the tight coupling between vesicle exocytosis and Rab3 membrane disassociation is disrupted . We uncovered the long-sought molecular apparatus linking vesicle exocytosis to Rab3 cycling and we also revealed the important function of synaptotagmin 1 in repetitive synaptic vesicle release . In nerve terminals , neurotransmitters are packaged into synaptic vesicles ( SVs ) and released by Ca2+-induced exocytosis ( Sudhof , 2004 ) . Fast and precise neuronal reaction requires that SVs are clustered in front of the release site , the presynaptic active zone . SVs then dock at the active zone , where they are primed to adopt a competent “ready-for-fusion” state . An action potential induces the opening of Ca2+ channels , and the rising Ca2+ concentration stimulates SV-plasma membrane fusion . The basic membrane fusion reaction is mediated by evolutionarily conserved soluble NSF attachment protein receptors ( SNAREs ) and related proteins like Munc13 and Munc18 ( Weber et al . , 1998; Sudhof , 2004; Brunger , 2005; Jahn and Scheller , 2006; Lang and Jahn , 2008; Jahn and Fasshauer , 2012 ) . However , the Ca2+-sensing process that starts the SNARE engine is primarily carried out by the synaptotagmin family ( Chapman , 2002; Jahn and Fasshauer , 2012 ) . Through their C2 domains , synaptotagmins bind to Ca2+ , thus triggering membrane fusion ( Sudhof , 2004 ) . After exocytosis , SVs undergo endocytosis and recycling and are refilled with neurotransmitters for repeated rounds of release . Rab3 protein is highly enriched in the nervous system and is specifically localized on SVs ( Fischer von Mollard et al . , 1991; Fischer von Mollard et al . , 1994; Geppert et al . , 1994; Stahl et al . , 1996 ) . Like other Rabs , Rab3 cycles on and off its target membranes according to its GTP- or GDP-bound state . On the vesicles , the active GTP-bound form of Rab3 is complexed with effector proteins like rabphilin and RIM ( Rab3-interacting molecule ) ( Shirataki et al . , 1993; Li et al . , 1994; Wang et al . , 1997 , 2000 ) , thus facilitating the recruitment/docking of SVs ( Nonet et al . , 1997; Leenders et al . , 2001; Tsuboi and Fukuda , 2006 ) . Ca2+-induced exocytosis can trigger disassociation of Rab3 from SV membranes through the GTP hydrolysis process ( Fischer von Mollard et al . , 1991; Fischer von Mollard et al . , 1994; Stahl et al . , 1996 ) , but the underlying mechanisms are not clear . The GTP-to-GDP conversion not only removes Rab3 from SVs , but also simultaneously dissociates Rab3 from its binding effectors , which disassembles the docking complex so that both Rab3 and Rab3 effectors can be recycled for the next round of release ( Wang et al . , 1997 , 2000 ) . Given their unique features , Rab3 and synaptotagmin have been considered as the Yin and Yang of membrane fusion , respectively ( Geppert and Südhof , 1998 ) . However , the functional regulatory interaction between synaptotagmin 1 and Rab3 cycling has not been identified nor has the mechanism by which this interaction is coupled to fast and repetitive neurotransmitter release . Here , we found that synaptotagmin 1/SNT-1 in C . elegans is crucial for the SV association of RAB-3 protein . SNT-1 promotes the GTP-bound state of RAB-3 by inhibiting RAB-3 GAP . The catalytic subunit of RAB-3 GAP ( RBG-1 ) localizes on SVs and directly binds to SNT-1 . Ca2+ treatment disrupts the direct association between SNT-1 and RBG-1 . In addition , Ca2+-binding activity of SNT-1 is essential for the dissociation of RAB-3 from SVs . Thus , our study reveals the pivotal dual role of synaptotagmin 1 in coupling SV exocytosis with the Rab3 membrane association and dissociation cycle . In C . elegans motor neurons , RAB-3 fused with Green Fluorescent Protein ( GFP ) adopts a punctate pattern of localization along the length of the ventral and dorsal cords ( Mahoney et al . , 2006 ) . This punctate RAB-3 pattern is similar to that of other SV proteins , including synaptobrevin and synaptotagmin ( Nonet et al . , 1993; Nonet , 1999; Zhen and Jin , 1999 ) . A previous report also showed that most Rab3 protein is associated with SV membranes ( Fischer von Mollard et al . , 1990 ) . In the absence of the RAB-3 GEF , AEX-3 ( Iwasaki et al . , 1997 ) , GFP::RAB-3 no longer shows a punctate pattern and becomes diffusely distributed in neuron cell bodies and axons ( Figure 1A ) . AEX-3 is responsible for converting RAB-3 protein from the membrane-dissociated GDP-bound form to the membrane-associated GTP-bound form ( Figure 1C ) . Therefore , the punctate localization of RAB-3 in wild type likely represents the GTP-bound , SV membrane-associated form of RAB-3 , while the diffuse GFP::RAB-3 signal may represent the dissociated GDP-RAB-3 . 10 . 7554/eLife . 05118 . 003Figure 1 . RAB-3 synaptic vesicle association requires SNT-1 . ( A ) Punctate distribution of GFP::RAB-3 in C . elegans motor neurons in wild-type animals ( top ) . The GFP::RAB-3 puncta become diffuse in aex-3 , rep-1 , hmgs-1 , and snt-1 mutants ( lower panels ) . Yellow arrows indicate the cell bodies along the ventral cord . A representative line-scanning image for each genotype is shown in the right panel . ( B ) Quantification of the synaptic enrichment in wild-type , aex-3 , rep-1 , hmgs-1 , and snt-1 animals . Data are presented as mean ± SD; p < 0 . 01 . ( C ) Schematic representation of the RAB-3/SV association and dissociation cycle . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 00310 . 7554/eLife . 05118 . 004Figure 1—figure supplement 1 . snt-1 is require for RAB-3 synaptic vesicle localization . ( A ) Multiple snt-1 alleles show the diffuse GFP::RAB-3 phenotype . White arrows indicate the cell bodies . ( B ) GFP::RAB-3 expressed pan-neuronally under the control of the Prab-3 promoter displays punctate distribution . ( C ) GFP::RAB-3 is diffuse in all Prab-3-expressing cells . Red boxed areas are enlarged in panels underneath the image taken from a whole animal ( B′ , B″ , B‴ , C′ , C″ and C‴ ) . ( D ) The diffuse GFP::RAB-3 phenotype in snt-1 mutants is rescued by expressing wild-type snt-1 gene pan-neuronally using the Psnt-1 promoter or in DD , VD , and AS neurons using the Phmr-1 promoter . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 004 We speculated that mutations in components required for Rab3/SV association may lead to a diffuse RAB-3::GFP phenotype similar to that in aex-3 animals . Hence , we conducted a genetic screen and isolated multiple mutants in which GFP::RAB-3 lost its punctate localization pattern . Through SNP mapping , complementation testing , and fosmid rescue , we cloned all of these mutations . Six of them ( xd58 , xd137 , xd142 , xd143 , xd148 , and xd149 ) turned out to be new alleles of aex-3 . In addition , we obtained four rep-1 alleles ( xd56 , xd138 , xd139 , and xd140 ) and three hmgs-1 alleles ( xd128 , xd129 , and xd145 ) . rep-1 encodes the sole Rab escort protein ( Rep ) ( Tanaka et al . , 2008 ) . Rep proteins bind newly synthesized Rab proteins and facilitate the addition of geranylgeranyl groups to Rabs ( Seabra et al . , 1992a , 1992b; Andres et al . , 1993 ) ( Figure 1C ) . HMGS-1 is orthologous to the human hydroxymethylglutaryl-CoA synthase ( HMGS ) which is required for synthesis of the geranylgeranyl moiety ( Mehrabian et al . , 1986; Shi and Ruvkun , 2012 ) ( Figure 1C ) . Thus , both rep-1 and hmgs-1 are critical for RAB-3 membrane targeting . The diffuse GFP::RAB-3 phenotype in rep-1 and hmgs-1 mutants further suggested that the diffuse signal indeed comes from membrane-dissociated RAB-3 protein . In the meantime , we hypothesized that the molecules controlling the RAB-3 cycle may be associated with SV cycling . Thus , we systematically examined SV cycle-related mutants . Interestingly , we found that in snt-1 ( md290 ) animals , which lack snt-1 function , the GFP::RAB-3 puncta disappeared and the GFP signal was diffusely distributed throughout the neuronal processes , similar to aex-3 , rep-1 , and hmgs-1 mutants ( Figure 1A , B ) . The strong phenotypic similarity between snt-1 and other RAB-3/SV-association defective mutants suggests that SNT-1 plays an important role in RAB-3/SV localization . snt-1 encodes the synaptotagmin 1 homologue in C . elegans ( Nonet et al . , 1993 ) . Neuronal synaptotagmins function as Ca2+ sensors for synaptic exocytosis , but their role in Rab3 localization has not been revealed . To determine whether loss of snt-1 function indeed leads to the diffuse RAB-3 phenotype , we examined other snt-1 alleles . We found that the n2665 , md220 , md125 , and md172 alleles of snt-1 all display a diffuse GFP::RAB-3 phenotype similar to md290 ( Figure 1—figure supplement 1A ) . Both the snt-1 and rab-3 genes are broadly expressed in the nervous system ( Nonet et al . , 1993 , 1997 ) . To determine whether snt-1 influences RAB-3 in all neurons , we examined RAB-3 localization with a pan-neuronal marker Prab-3::GFP::RAB-3 . In wild-type animals , GFP::RAB-3 displays a punctate pattern , while in snt-1 mutants , GFP::RAB-3 is completely diffuse in neuronal processes , including the nerve ring , ventral cord , and dorsal cord regions ( Figure 1—figure supplement 1B , C ) . These data indicate that the effect of snt-1 on RAB-3 localization is widely preserved in the nervous system . In addition , the diffuse GFP::RAB-3 phenotype was fully rescued when wild-type snt-1 was introduced into mutant animals ( Figure 1—figure supplement 1D ) , suggesting that SNT-1 is indeed essential for localization of RAB-3 on SVs . The diffuse RAB-3 phenotype in snt-1 may be caused by failure of SV clustering at the synaptic terminal . Therefore , we examined the localization of another synaptic vesicle protein SNB-1 . SNB-1 is the C . elegans synaptobrevin homologue ( Nonet , 1999 ) . In worm DD and VD motor neurons , SNB-1 is distributed evenly in punctate structures along neuronal processes , similar to RAB-3 ( Figure 2A ) ( Zhen and Jin , 1999 ) . In snt-1 animals , some of the SNB-1 puncta are enlarged , but the punctate distribution of SNB-1 is not altered ( Figure 2A , B ) . This observation is consistent with recent findings ( Yu et al . , 2013 ) , suggesting that SV clustering is probably not affected by snt-1 . 10 . 7554/eLife . 05118 . 005Figure 2 . Synaptic vesicle clustering is unaffected by loss of snt-1 function . ( A ) SNB-1::GFP puncta distribution in wild type and snt-1 mutants . ( B ) The synaptic enrichment of SNB-1::GFP puncta is indistinguishable in wild type and snt-1 . Data are presented as mean ± SD; NS , not significant . In both wild type ( C , C′ and C″ ) and snt-1 ( D , D′ and D″ ) , the SNB-1::GFP puncta are present in the synaptic area on the ventral cord , which is outside of the cell body ( C″ and D″ ) . In unc-104 ( E , E′ and E″ ) or snt-1 unc-104 double mutants ( F , F′ and F″ ) , SNB-1::GFP accumulates in the cell bodies on the ventral cord ( E″ and F″ ) . ( G ) In wild type , GFP::RAB-3 is distributed in a punctate pattern in the pre-synaptic regions on the ventral cord ( G“ ) . ( H ) GFP::RAB-3 is diffuse throughout the whole axon including both dorsal ( H' ) and ventral ( H″ ) processes . ( I ) GFP::RAB-3 accumulates in ventral cell bodies ( I″ ) . ( J ) In snt-1 unc-104 double mutants , GFP::RAB-3 is diffuse throughout the whole axon in both dorsal ( J' ) and ventral ( J″ ) regions . Yellow boxes indicate part of the dorsal cord , which is enlarged in the lower left panels . Red boxes indicate part of the ventral cord , which is enlarged in the lower right panels ( white arrows indicate DD cell bodies in the ventral cord ) . A schematic drawing of a DD neuron during the L1 stage is presented underneath the fluorescence images of each genotype , with the SNB-1::GFP or GFP::RAB-3 signal shown in green . Small green dots represent the pre-synaptic areas . Individual DD cell bodies are indicated as large ovals at the bottom right of each diagram . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 005 unc-104 encodes the cytosolic kinesin responsible for synaptic vesicle trafficking from cell bodies to nerve terminals ( Hall and Hedgecock , 1991 ) . In the absence of UNC-104 kinesin , few SVs are transported to synaptic termini , while neuron cell bodies have a surfeit of SVs . To further address whether the diffuse RAB-3 phenotype is indeed caused by the dissociation of RAB-3 from SVs in snt-1 mutants , we performed a serial mutant analysis utilizing both GFP::RAB-3 and SNB-1::GFP markers . Because the snt-1;unc-104 double mutant animals are arrested during larval development , the synaptic phenotypes were examined in newly hatched L1 animals ( larval stage 1 ) . In L1 animals , among DD , VD , and AS motor neurons , only DDs are born ( Sulston , 1976; Sulston and Horvitz , 1977 ) , and they form pre-synapses along the ventral cord . Thus , in wild-type L1 animals labeled by Punc-25::SNB-1::GFP , the GFP signal could only be detected along the ventral cord ( Figure 2C , C″ ) but not the dorsal cord ( Figure 2C' ) . In snt-1 mutants , the SNB-1::GFP distribution is indistinguishable from wild type ( Figure 2D , D' , and 2D″ ) . In unc-104 single mutants , we found that SNB-1::GFP accumulated in cell bodies ( Figure 2E , E″ , white arrow ) and little GFP signal could be detected outside of cell bodies or on the dorsal cord ( Figure 2E' ) , which is consistent with the role of UNC-104 in SV transport . In snt-1;unc-104 double mutants , the SNB-1::GFP signal accumulated in cell bodies ( Figure 2F , F″ , white arrow ) like in unc-104 single mutants , suggesting that further removal of SNT-1 in unc-104 mutant animals does not alter the dependence of SVs on UNC-104 for intracellular trafficking . Interestingly , the effect of snt-1 or snt-1 unc-104 mutations on RAB-3 and SNB-1 is quite different . As shown in Figure 2H , J , GFP::RAB-3 is still diffuse in cell bodies and axons in both snt-1 and snt-1 unc-104 animals , and GFP signal could be detected even in the non-synaptic dorsal cord region ( Figure 2H' and 2J' ) . In contrast , in unc-104 animals , the RAB-3 puncta are retained within cell bodies , just like SNB-1 is in unc-104 mutants ( Figure 2I , I″ ) . These data strongly support the notion that the diffuse phenotype of RAB-3 is not caused by the dispersion of SV clusters , but rather by the specific dissociation of RAB-3 from SV membranes . How does mutation of snt-1 affect the SV membrane association of RAB-3 ? Previous studies showed that the localization of RAB-3 on SV membranes is tightly associated with its GTP-bound state ( Zerial and McBride , 2001 ) . Therefore , we tested whether the loss of RAB-3 from SVs in snt-1 mutants is caused by reduction of GTP-bound RAB-3 . The active GTP-Rab3 binds to the RBD domain of its effector RIM , while the inactive GDP-Rab3 does not . Previous reports demonstrated that the RBD domain of mammalian RIM2 could bind to the worm GTP-RAB-3 ( Wang et al . , 1997; Mahoney et al . , 2006 ) . Thus , we performed pull-down assays to examine the GTP-RAB-3 level in vivo . In wild-type worm lysates , the active GTP-bound form of RAB-3 protein was efficiently pulled down by GST-RBD ( Figure 3A ) . In contrast , the amount of GTP-RAB-3 pulled down by RIM2 RBD was significantly reduced in snt-1 lysates ( Figure 3A , B ) . In the absence of RAB-3 GEF , the GDP-bound RAB-3 cannot be converted to the GTP-bound RAB-3 . Indeed , in aex-3 animals , the amount of RAB-3 that can be pulled down by GST-RBD is also greatly decreased ( Figure 3A , B ) . Thus snt-1 , similar to aex-3 , affects the level of GTP-bound RAB-3 in vivo . 10 . 7554/eLife . 05118 . 006Figure 3 . The GTP-bound form of RAB-3 is decreased in snt-1 mutants . ( A ) A GST-fused RBD domain of RIM2 binds active GTP-RAB-3 . The amount of GTP-RAB-3 pulled down by RBD is decreased in both aex-3 and snt-1 animals . ( B ) Quantification of the GTP-RAB-3 level in wild type , aex-3 , and snt-1 . Data are presented as mean ± SD; p < 0 . 01 . ( C ) The amount of GFP-RAB-3 in the cytosolic fraction is increased in snt-1 mutants . ( D ) Localization of GFP::RAB-27 puncta is affected by mutation of aex-3 , but not by mutation of snt-1 . The white arrows indicate the cell bodies . ( E ) Over-expression of aex-3 does not rescue the snt-1 mutant phenotype . Yellow arrows indicate the cell bodies . ( F ) The AEX-3::GFP level is unchanged in snt-1 mutants compared to wild type . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 006 GTP-RAB-3 is associated with SV membranes , while GDP-RAB-3 is diffused in the cytosol . Thus , we performed cell fractionation experiments to further examine the GTP or GDP status of RAB-3 . In wild type , RAB-3 is highly enriched in membrane fractions ( Figure 3C ) , which is consistent with the SV localization of GTP-RAB-3 . In contrast , the RAB-3 protein distribution is shifted to the soluble fraction when snt-1 is removed , suggesting a cytosolic localization of RAB-3 in snt-1 mutants ( Figure 3C ) . These data suggest that snt-1 indeed promotes the GTP-bound form of RAB-3 . How does loss of function of snt-1 lead to the reduction of GTP RAB-3 ? One possibility is that SNT-1 may regulate the RAB-3 GTP-GDP cycle by promoting GEF activity . We performed the following experiments to test this possibility . Firstly , AEX-3 is the GEF molecule for both RAB-3 and RAB-27 ( Mahoney et al . , 2006 ) . If snt-1 indeed affects AEX-3 activity , we would expect that the localization of RAB-27 on SVs will be affected by the absence of SNT-1 . We made a GFP::RAB-27 reporter and expressed it in motor neurons in worms . The GFP::RAB-27 protein is enriched in synaptic regions and displays a punctate expression pattern similar to RAB-3 ( Figure 3D ) . In aex-3 mutants , GFP::RAB-27 becomes diffuse , consistent with the role of AEX-3 as a GEF for RAB-27 ( Figure 3D ) . In contrast , GFP::RAB-27 still displays a punctate distribution indistinguishable from wild type in snt-1 mutants ( Figure 3D ) , suggesting that the GEF activity of AEX-3 , at least for RAB-27 , is not altered by mutation of snt-1 . Second , if SNT-1 promotes AEX-3 GEF activity , we would expect that increasing the aex-3 expression level may rescue the diffuse RAB-3 phenotype in snt-1 mutants . However , no such rescue was observed ( Figure 3E ) . Lastly , we examined the expression level of AEX-3 and found that it was indistinguishable in wild-type and snt-1 animals ( Figure 3F ) . Together , these results suggest that it is unlikely that snt-1 regulates the GTP-RAB-3 level by promoting RAB-3 GEF activity . Alternatively , the decreased GTP-RAB-3 level may be caused by increased RAB-3 GTPase activity in snt-1 mutants . Rab3 GTPase activity is greatly facilitated by Rab3-specific GTPase-activating protein ( GAP ) . Rab3 GAP is composed of the catalytic subunit Rab3GAP1 and the noncatalytic subunit Rab3GAP2 . rbg-1 and rbg-2 encode Rab3GAP1 and Rab3GAP2 , respectively in worms ( Figure 4—figure supplement 1A , B ) ( Fukui et al . , 1997; Nagano et al . , 1998 ) . In the absence of Rab3 GAP , the RAB-3 synaptic enrichment is enhanced , which is consistent with the role of Rab3 GAP in assisting GTP hydrolysis ( Figure 4—figure supplement 1C , D ) . If RAB-3 GTP hydrolysis activity is indeed increased in snt-1 mutants , we would expect that loss of GAP function will suppress the snt-1 mutant phenotype . Indeed , in rbg-1;snt-1 double mutants , we found that the diffuse GFP::RAB-3 phenotype of snt-1 single mutants is significantly suppressed ( Figure 4A , B ) . Furthermore , mutation of the rbg-2 gene also suppressed the diffuse RAB-3 phenotype in snt-1 mutants ( Figure 4—figure supplement 1E ) . In contrast , the diffuse GFP::RAB-3 signal caused by aex-3 mutation could not be suppressed by rbg-1 ( Figure 4A , B ) . We next performed RIM2-RBD pull-down assays to test whether the GTP-RAB-3 level was restored in rbg-1;snt-1 mutants . In contrast to the greatly reduced GTP-RAB-3 level in snt-1 lysates , the amount of GTP-bound RAB-3 is significantly increased in rbg-1;snt-1 samples ( Figure 4C ) . Taken together , these results suggest that snt-1 indeed regulate the RAB-3/SV association specifically by inhibiting RAB-3 GTP hydrolysis . 10 . 7554/eLife . 05118 . 007Figure 4 . RAB-3 GAP mutations suppress the snt-1 mutant phenotype . ( A ) The punctate distribution of GFP::RAB-3 is restored in rbg-1;snt-1 animals , while the aex-3 phenotype could not be suppressed by mutation of rbg-1 . Yellow arrows indicate the cell bodies along the ventral cord . Scale bar , 5 µm . A representative line-scanning image for each genotype is shown in the right panel . ( B ) Quantification of the synaptic enrichment of GFP::RAB-3 signal in the genotypes shown in ( A ) . Data are represented as mean ± SD . p < 0 . 01; NS , not significant . ( C ) The amount of GTP-RAB-3 pulled down by RBD is increased in rbg-1;snt-1 animals compared to snt-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 00710 . 7554/eLife . 05118 . 008Figure 4—figure supplement 1 . rbg-2 suppresses the snt-1 mutant phenotype . ( A ) Schematic representation of the rbg-1 ( ok1660 ) deletion mutation . ( B ) Schematic representation of the rbg-2 ( ok3195 ) deletion mutation . Solid boxes indicate exons and thin lines indicate introns . The bar below the gene indicates the deleted region . ( C ) The enlarged GFP::RAB-3 puncta phenotype in rbg-1 mutants is rescued by expressing a wild-type copy of the rbg-1 gene . ( D ) Quantification of the synaptic enrichment of GFP::RAB-3 signal in the genotypes shown in ( C ) . Data are represented as mean ± SD; p < 0 . 01 . ( E ) Loss of rbg-2 function suppresses the snt-1 mutant phenotype . White arrows indicate the cell bodies . ( F ) RAB-3 is co-precipitated with RBG-1 . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 008 As the catalytic subunit of Rab3 GAP , RBG-1 can interact with RAB-3 ( Figure 4—figure supplement 1F ) . To understand how SNT-1 inhibits RAB-3 GTP hydrolysis , we examined the sub-cellular localization of RBG-1 . We created a functional mCherry::RBG-1 construct and injected it into rbg-1 mutant animals . In wild-type animals , the mCherry::RBG-1 signal displayed a punctate expression pattern along the ventral and dorsal cords . Double staining further showed that the mCherry::RBG-1 puncta co-localized with the SV marker SNB-1::GFP ( Figure 5C ) . Next , we tested whether the punctate localization of RBG-1 relies on the UNC-104-based intracellular transport system like other SV-associated proteins . We found that in unc-104 mutants , the mCherry::RBG-1 puncta no longer appeared in the putative synaptic region; instead they were retained in the cell bodies ( Figure 5B ) . Together , the data above suggest that RBG-1 is localized on SVs . 10 . 7554/eLife . 05118 . 009Figure 5 . Localization of RBG-1 on synaptic vesicles requires SNT-1 . ( A and B ) mCherry::RBG-1 has a punctate distribution in wild type ( A ) but accumulates in cell bodies ( yellow arrows ) in unc-104 mutants ( B ) . ( C ) mCherry::RBG-1 ( red ) is co-localized with SNB-1::GFP puncta ( green ) . ( D ) In snt-1 mutants , mCherry::RBG-1 loses its punctate localization and becomes diffuse in axons , while SNB-1::GFP retains its punctate pattern . ( E ) mCherry::RBG-1 retains its punctate distribution and is co-localized with SNB-1::GFP in rab-3 mutants . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 009 SNT-1 resides on SVs and its function in RAB-3 localization is executed by RAB-3 GAP . Could the SV localization of RBG-1 be regulated by snt-1 ? We examined RBG-1 distribution in snt-1 animals and found that the RBG-1/SV co-localization is lost and mCherry::RBG-1 fluorescence becomes diffuse ( Figure 5D ) . In contrast , SNB-1::GFP still retains its punctate expression pattern in snt-1 animals , similar to wild type ( Figure 5D ) . Therefore , SNT-1 is required for the SV localization of RBG-1 . RBG-1 binds to its substrate RAB-3 ( Figure 4—figure supplement 1E ) . Thus , we also tested whether the SV localization of RBG-1 is controlled by RAB-3 . In rab-3 ( js49 ) mutants , we found that mCherry::RBG-1 retains its punctate expression and still co-localizes with SNB-1::GFP ( Figure 5E ) , suggesting that the SV localization of RBG-1 does not rely on RAB-3 protein . The SNT-1-dependent SV association of RBG-1 suggests a direct association between SNT-1 and RBG-1 . We co-expressed full-length RBG-1 and the cytosolic domain of SNT-1 ( C2AB ) in HEK293FT cells . After affinity purification , the RBG-1 protein was incubated with the SNT-1 C2AB fragment . In contrast to the mock-transfected sample , RBG-1 was effectively co-precipitated with the cytosolic region of SNT-1 ( Figure 6A ) . The SNT-1 cytosolic portion was also co-precipitated by RBG-1 ( Figure 6B ) . We next asked which domain of the SNT-1 cytosolic region is required for this binding . The cytosolic region of SNT-1 contains a C2A and a C2B motif . When the C2A domain was deleted , the remaining C2B motif retained the RBG-1 binding activity ( Figure 6C ) . In contrast , when C2B was removed , the C2A domain alone could not bind to RBG-1 ( Figure 6C ) . Therefore , the C2B domain is required for SNT-1 binding to RBG-1 . 10 . 7554/eLife . 05118 . 010Figure 6 . RBG-1 associates with the C2B domain of SNT-1 . ( A ) RBG-1 is precipitated by the intracellular domain ( C2AB ) of SNT-1 . ( B ) The SNT-1 intracellular domain is precipitated by RBG-1 . ( C ) The C2B domain of SNT-1 binds to RBG-1 . ( D ) SNT-1 without the C2B domain fails to rescue the snt-1 mutant phenotype . Yellow arrows indicate cell bodies . The schematic diagram shows the transmembrane ( TM ) and intracellular calcium-binding domains ( C2A and C2B ) of SNT-1 . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 010 We further investigated whether the function of SNT-1 in regulating the RAB-3/SV association is mediated through the C2B domain . In comparison with full-length SNT-1 , which fully rescues the GFP::RAB-3 mis-localization defect in snt-1 mutants , we found that SNT-1 without the C2B domain ( deltaC2B ) had no rescue effect ( Figure 6D ) . In contrast , when the C2A domain is removed ( deltaC2A ) from SNT-1 , the remaining protein still possesses the rescue activity ( Figure 6D ) . Together , these data suggest that the C2B domain is critical for SNT-1 function . SNT-1 lacking the trans-membrane domain ( deltaTM ) also failed to rescue the diffuse RAB-3 phenotype ( Figure 6D ) , suggesting that the SV localization function of SNT-1 is required in addition to the C2B domain for regulating RAB-3/SV association in vivo . Ca2+-mediated exocytosis activates the dissociation of Rab3 from the SV membrane ( Fischer von Mollard et al . , 1991; Fischer von Mollard et al . , 1994 ) . Could synaptotagmin 1 , as the Ca2+ sensor for SV exocytosis , be the trigger to initiate the Rab3 SV dissociation process ? We showed above that SNT-1 directly associates with Rab3GAP1/RBG-1 . Therefore , we wondered whether the Ca2+ level could affect the binding of SNT-1 to RabGAP1/RBG-1 , and whether the inhibition of RAB-3 GAP by SNT-1 is relieved by Ca2+ binding , thus allowing dissociation of RAB-3 from the SV during exocytosis . To test the ideas above , we asked whether the presence of Ca2+ disrupts the binding between SNT-1 and RBG-1 . We purified RBG-1 and SNT-1 proteins and performed co-IP experiments with increasing concentrations of Ca2+ . A Ca2+ concentration of 0 . 5 mM or 1 mM significantly compromised the RBG-1/SNT-1interaction ( Figure 7A , B ) . Thus , upon Ca2+ binding , SNT-1 releases RBG-1 . 10 . 7554/eLife . 05118 . 011Figure 7 . Dissociation of RAB-3 from synaptic vesicles requires the Ca2+-binding activity of SNT-1 . ( A ) Ca2+ treatment diminishes the binding between RBG-1 and the intracellular domain ( C2AB ) of SNT-1 . ( B ) Quantification of the relative binding between RBG-1 and the C2AB domain upon Ca2+ treatment . ( C ) SNT-1 without Ca2+-binding sites ( C2AB ) still binds to RBG-1 in the presence of Ca2+ . ( D ) Quantification of the relative binding between RBG-1 and the C2AB domain upon Ca2+ treatment . ( E ) Mutant SNT-1 proteins without Ca2+-binding activity stabilize RAB-3 on SVs . Ex snt-1 , over-expression of SNT-1; Ex snt-1 C2A* and Ex snt-1 C2B* , over-expression of SNT-1 with mutant C2A domain or C2B domain , respectively; Ex snt-1 C2AB , over-expression of SNT-1 with mutant C2A and C2B domains . Scale bar , 5 µm . ( F ) Quantification of the synaptic enrichment in the genotypes shown in ( E ) . ( G ) SNT-1 functions as a molecular switch controlling RAB-3/SV association and disassociation during SV exocytosis . Data are represented as mean ± SD . p < 0 . 05; p < 0 . 01; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 01110 . 7554/eLife . 05118 . 012Figure 7—figure supplement 1 . Exocytosis is uncoupled from RAB-3 synaptic vesicle dissociation in snt-1 mutants . ( A ) The failure of RAB-3/SV dissociation caused by exocytosis mutants , including unc-2 and unc-13 , is bypassed by mutation of snt-1 . Yellow arrows indicate the cell bodies along the ventral cord . Scale bar , 5 µm . ( B ) Quantification of the synaptic enrichment in the genotypes shown in ( A ) . Data are presented as mean ± SD . p < 0 . 05; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05118 . 012 Next , we asked whether SNT-1-RBG-1 binding is still affected by Ca2+ treatment if the Ca2+-binding sites are removed from SNT-1 . The amino acids critical for Ca2+ binding in the C2A domain ( D248 and D250 ) and the C2B domain ( D383 and D385 ) were mutated , and the resulting SNT-1 mutant protein ( C2AB ) was purified and tested for its ability to bind RBG-1 . We found that the RBG-1-binding capability of C2AB was high in the absence or presence of Ca2+ ( Figure 7C , D ) , suggesting that the Ca2+-binding activity of SNT-1 is essential for attenuation of the SNT-1/RBG-1 interaction when Ca2+ is present . Ca2+ treatment decreases the binding between SNT-1 and RBG-1 . Is the inhibition of RAB-3 GAP by SNT-1 alleviated when the Ca2+concentration rises ? If so , SNT-1 that lacks Ca2+-binding capability will fail to release RAB-3 GAP during Ca2+ influx , and thus the dissociation of RAB-3 from SVs will be inhibited . To test this hypothesis , we analyzed the GFP::RAB-3 pattern in transgenic animals that express different mutant forms of SNT-1 . The two amino acids necessary for Ca2+ binding in the C2A domain of SNT-1 are D248 and D250 . We created the SNT-1 mutant C2A* by replacing these two aspartic acids with alanines and found that C2A* could rescue the GFP::RAB-3 mis-localization defect ( Figure 7E ) . In addition to this rescue phenomenon , we noticed that the GFP::RAB-3 puncta were enlarged and the GFP signal was more enriched in the punctate regions in comparison to wild-type animals over-expressing snt-1 ( Figure 7E , F ) . A similar GFP::RAB-3 enrichment effect was observed in worms expressing the C2B* mutant form of SNT-1 , which contains a C2B domain that cannot bind Ca2+ ( D383 and D385 were replaced with alanines ) ( Figure 7E , F ) . We further replaced the Ca2+-binding sites in both C2A and C2B ( C2AB ) . When this construct was expressed , the GFP::RAB-3 signal was also increased in the punctate regions ( Figure 7E , F ) . Could the enhanced GFP:RAB-3 signal in the puncta indeed reflect the failure of exocytosis ? Previous studies suggest that RAB3 dissociation from SV membranes is inhibited when SV exocytosis is disrupted by blocking either Ca2+ influx or membrane fusion ( Fischer von Mollard et al . , 1991; Fischer von Mollard et al . , 1994; Stahl et al . , 1996 ) . We examined two exocytosis mutants , unc-2 and unc-13 . unc-2 encodes the alpha subunit of the voltage-gated Ca2+ channel ( Schafer and Kenyon , 1995 ) . In unc-2 mutants , we found that the GFP signal is significantly enriched within the punctate regions and the GFP::RAB-3 puncta are larger and brighter compared to wild type ( Figure 7—figure supplement 1A , B ) . unc-13 encodes the Munc13 homolog in worms and loss of function of unc-13 results in blockage of membrane fusion during SV exocytosis ( Aravamudan et al . , 1999; Richmond et al . , 1999; Varoqueaux et al . , 2002 ) . We found that the GFP signal was enhanced in the punctate regions in unc-13 mutants , as in unc-2 mutants ( Figure 7—figure supplement 1A , B ) . Blocking SV exocytosis resulted in failure of RAB-3 to dissociate from SVs; thus the enlarged GFP::RAB-3 puncta indeed indicate the decreased dissociation of RAB-3 protein from SVs . Taken together , the data above indicate that the Ca2+-binding capability is essential for SNT-1 function in mediating the RAB-3 SV dissociation induced by Ca2+-triggered exocytosis . Interestingly , in unc-13;snt-1 or unc-2;snt-1 double mutants , the RAB-3::GFP signal is diffuse throughout whole neuronal cells ( Figure 7—figure supplement 1A , B ) , as seen in snt-1 single mutants . Because the diffuse RAB-3::GFP signal suggests a dissociation of RAB-3 from SV membranes , the above observation implies that in the absence of SNT-1 , the dissociation of RAB-3 from SVs can occur even when Ca2+-induced exocytosis is blocked . As a consequence , the coupling between SV exocytosis and RAB-3 dissociation is disrupted when SNT-1 is missing . These data are consistent with the inhibition-of-inhibition role of SNT-1 on RAB-3/SV association . Together , these results suggest that SNT-1 plays a dual role in the RAB-3/SV cycle , inhibiting dissociation of RAB-3 from SVs during the resting state ( no Ca2+ influx ) and triggering dissociation of RAB-3 from SVs upon Ca2+ binding ( Figure 7G ) . Acute and precise neuronal activity requires precise coordination between the SV cycle and the Rab3 cycle . As the trigger of regulated vesicle secretion , synaptotagmin 1 is known to bind the membrane and the SNARE complex to give the final push for complete assembly of the SNARE complex for membrane fusion ( Davletov and Südhof , 1993; Chapman and Davis , 1998; Dai et al . , 2007; Choi et al . , 2010; Vrljic et al . , 2010 ) . Here , we revealed that SNT-1/synaptotagmin 1 functions as an on-and-off switch to regulate Rab3 membrane association , thus facilitating repeated release . Our study has many important implications . In the resting state , Rab3 protein is associated with SVs in the GTP-bound form , and Rab3 GTP , together with its corresponding effectors , docks the vesicles at the active zone region ( Sudhof , 2004 ) . However , Rab3 GAP , the negative regulator of Rab3 , is also enriched in the synaptic fraction and is localized on SVs ( this study ) ( Oishi et al . , 1998 ) . How therefore is the active form of Rab3 , and hence the proper docking complex , maintained when GAP is close by ? Based upon our study , it is entirely possible that synaptotagmin 1 prevents the hydrolysis of Rab3 GTP by directly sequestering or inhibiting Rab3 GAP before Ca2+ influx . In fact , the SV localization of Rab3GAP1/RBG-1 is particularly interesting , given the fact that the Rab3 cycle must be spatially regulated so that Rab3 is kept in close proximity to SVs for repeated neurotransmitter release . Indeed , when synaptotagmin 1/SNT-1 binds Ca2+ , the inhibition on RAB-3 GAP is alleviated , so the locally enriched Rab3 GAP can freely and quickly access GTP-Rab3 and hydrolyze Rab3 GTP to GDP . Together with the functional Ca2+switch , synaptotagmin 1 can therefore efficiently coordinate the Rab3 cycle with the SV cycle . The concentration of Ca2+ needed for exocytosis ranges from ∼10 to 200 µM in the nerve terminals , so the Ca2+ concentration that is required to release RBG-1 from SNT-1 is relatively high ( 500 µM ) . However , it is known that the rather low affinity of synaptotagmin 1 for Ca2+ ( Ubach et al . , 1998; Fernandez et al . , 2001 ) can be strongly affected by the presence of synaptotagmin-binding partners , especially membrane lipids ( Chapman , 2002 ) . Therefore , it will be interesting to test whether plasma membrane-enriched lipids such as PIP2 ( Lee et al . , 2010 ) can act synergistically with Ca2+ to regulate the binding of SNT-1 to RBG-1 . At the molecular level , synaptotagmin 1 is known to bind the plasma membrane and the SNARE complex in the presence of Ca2+ ( Davletov and Südhof , 1993; Chapman and Davis , 1998; Fernandez et al . , 2001; Dai et al . , 2007; Choi et al . , 2010; Vrljic et al . , 2010 ) . The interaction of synaptotagmin with membranes and SNARE proteins has well-documented consequences , including creating local positive membrane curvature and displacing the clamping factor complexin from the SNARE complex ( Martens et al . , 2007; Hui et al . , 2009 ) . Combining these previous reports with data presented in this study , we think that synaptotagmin-Ca2+ may not only give the final push for complete assembly of the SNARE complex for membrane fusion , but also play a role in terminating vesicle docking by indirectly deactivating Rab3 , thus facilitating repetitive transmitter release . Among more than 60 Rabs in humans and mice , Rab3 and Rab27 seem to be specifically involved in stimulated secretion in a variety of secretory cells ( Fukuda , 2008 , 2013 ) . Synaptotagmins have also evolved specifically to regulate secretion . Interestingly , the regulatory role of SNT-1 on RAB-3 does not extend to RAB-27 , which is functionally closely related to RAB-3 ( Mahoney et al . , 2006 ) . Simultaneous knockdown of Rab3 and Rab27 causes secretion defects more severe than single knockdown in worms and PC12 cells ( Mahoney et al . , 2006; Tsuboi and Fukuda , 2006 ) . Both Rab27 and Rab3 are localized on SVs and can bind to RIM , thereby docking the vesicles ( Wang et al . , 1997; Fukuda , 2003 ) . Rab3 and Rab27 share the same GEF , which is AEX-3 in worms and DENN/MADD in mammals ( Levivier et al . , 2001; Coppola et al . , 2002; Mahoney et al . , 2006 ) . In contrast , Rab3 GAP serves as a specific GAP for Rab3 ( Fukui et al . , 1997; Nagano et al . , 1998; Itoh and Fukuda , 2006 ) . Here , SNT-1 apparently only affects the RAB-3 cycle and this action is mediated by inhibition of the Rab3-specific GAP . Therefore , although Rab3 and Rab27 play redundant roles in SV exocytosis , they can be differentially controlled through their specific regulators . We would like to point out that humans and worms have multiple synaptotagmins , and it remains to be determined whether any of the other synaptotagmins play a similar regulatory role on Rab27 . Rab3 GAP consists of the catalytic subunit Rab3GAP1 and the noncatalytic subunit Rab3GAP2 ( Fukui et al . , 1997; Nagano et al . , 1998 ) . Rab3GAP1 and Rab3GAP2 form a complex in vitro and co-immunoprecipitate in vivo ( Nagano et al . , 1998 ) . Although Rab3GAP2 does not affect the in vitro GAP activity of Rab GAP1 , it may act to stabilize , regulate , or localize Rab3GAP1 correctly in cells . The functional characteristics are consistent with their close biochemical interactions . Loss-of-function mutations in Rab3GAP1 and Rab3GAP2 produce clinically almost indistinguishable conditions , Warburg Micro syndrome and Martsolf syndrome , characterized by brain , eye , and endocrine abnormalities ( Aligianis et al . , 2005 , 2006 ) . We have now revealed that both Rab3GAP1/rbg-1 and Rab3GAP1/rbg-2 mutations can suppress the RAB-3 mis-localization phenotype in snt-1 mutants , implying that Rab3GAP does indeed function as a complex to participate in the SNT-1-mediated regulation of the RAB-3 cycle . However , we noticed that loss of function of rbg-2 alone leads to additional synaptic or axonal defects ( data not shown ) compared to loss of rab-3 or rbg-1 . This suggests that RBG-2 may play roles in nervous system development other than forming the Rab3GAP complex with RBG-1 during SV exocytosis . The GTP-bound active form of Rab promotes membrane trafficking by interacting with specific effectors . In contrast with Rab effectors that function in secretory vesicle trafficking , relatively little is known about the specific Rab GEFs and GAPs and how they are regulated during vesicle secretion . Evidence that the sub-cellular localization and activity of RAB-3 GAP can be regulated by synaptotagmin/SNT-1 strongly hints that regulators of Rabs could be subjects for active manipulation during various types of intracellular membrane trafficking . A recent intriguing study is in agreement with our notion . In the amoeba Dictyostelium discoideum , vacuolar Ca2+ release activates the Rab GAP CnrF , thus subsequently down-regulating Rab11a ( Donato et al . , 2013 ) . Taken together , current data suggest a vital role of Ca2+ as the functional switch for regulated secretion processes . Because the key principles and regulatory components of different intracellular vesicle trafficking events are broadly conserved , the mechanism that we have uncovered is likely to represent a conserved mode of action . Understanding how the sequential activation of Rab GTPases is achieved during vesicle trafficking is a central theme of cell biology . This in turn raises the question of how regulatory Rab GEFs or Rab GAPs are activated in the right place and at the right time . Synaptotagmins act as the primary Ca2+-sensors in most forms of Ca2+-induced exocytosis , as exemplified by synaptic transmission . Our studies have proposed the elegant molecular machinery by which Rab GAP can be temporally and spatially regulated in response to acute cellular signals , so that correctly activated Rabs can perform their appropriate functions and allow vesicle fusion to occur in an orderly fashion . Strain maintenance and genetic manipulations were performed as described ( Brenner , 1974 ) . Strains used in this study are: LG I: EG2710 [unc-57 ( ok310 ) ] , CB450 [unc-13 ( e450 ) ] , VC2481 [rbg-2 ( ok3195 ) ] , CB47 [unc-11 ( e47 ) ] . LGII: NM204 [snt-1 ( md290 ) ] , CB1265 [unc-104 ( e1265 ) ] , MT6977 [snt-1 ( n2665 ) ] , RM1606 [snt-1 ( md172 ) ] , RM1603 [snt-1 ( md125 ) ] , RM1620 [snt-1 ( md220 ) ] , NM791 [rab-3 ( js49 ) ] . LGIII: XD1366 [rep-1 ( xd56 ) ] , NM1278 [rbf-1 ( js232 ) ] . LGIV: EG3027 [unc-26 ( s1710 ) ] , CB169 [unc-31 ( e169 ) ] . LGV: XD1925 [hmgs-1 ( xd145 ) ] , CB268 [unc-41 ( e268 ) ] , NM467 [snb-1 ( md247 ) ] , RM956 [ric-4 ( md1088 ) ] . LGX: XD1199 [aex-3 ( xd58 ) ] , RB1453 [rbg-1 ( ok1660 ) ] , CB81 [unc-18 ( e81 ) ] , CB55 [unc-2 ( e55 ) ] , CB102 [unc-10 ( e102 ) ] , CX51 [dyn-1 ( ky51 ) ] . Additional strains are: XD2188 [xdEx1380; Paex-3::AEX-3::GFP] , XD2702 [xdEx1214; Punc-25::mCHERRY::RBG-1] , XD3132 [xdEx1461; Phmr-1b::SNT-1FL] , XD3017 [xdEx1419; Phmr-1b::SNT-1 deltaTM] , XD3134 [xdEx1463; Phmr-1b::SNT-1 deltaC2A] , XD3034 [xdEx1398; Phmr-1b::SNT-1 deltaC2B] , XD3032 [xdEx1396; Phmr-1b::SNT-1 C2A] , XD3033 [xdEx1397; Phmr-1b::SNT-1C2B] , XD3133 [xdEx1462; Phmr-1b::SNT-1C2AB] . Mutagenesis was carried out in the xdIs7 ( Phmr-1b::GFP::RAB-3 ) strain treated with ethylmethane sulfonate . From 5000 genomes , 13 mutations were isolated . Subsequent genetic and molecular analysis revealed that we had isolated four alleles of rep-1 ( xd56 , xd138 , xd139 , and xd142 ) , six alleles of aex-3 ( xd58 , xd60 , xd137 , xd143 , xd148 , and xd149 ) , and three alleles of hmgs-1 ( xd128 , xd129 , and xd145 ) . rep-1 ( xd138 ) , rep-1 ( xd139 ) , hmgs-1 ( xd128 ) and hmgs-1 ( xd129 ) animals are larval lethal . The rest of the identified mutants are fertile . Promoters , GFP , mCherry , and various cDNA or genomic DNA fragments were cloned into the deltapSM vector through standard procedures . Site-directed mutagenesis was performed using standard PCR-based methods . Transgenic animals were produced as previously described ( Song et al . , 2010 ) . Integrated strains were obtained by UV irradiation . All integrated transgenic animals were out-crossed at least 3 times . DNA fragments containing the RBD domains of rat RIM2 were inserted into the pGEX-4T-3 vector . Expression and purification of GST fusion proteins in E . coli were carried out according to standard procedures . Worms with different genotypes were collected and washed in M9 buffer . 800 µl of homogenizing buffer ( 50 mM Tris-Cl pH8 . 0 , NaCl 150 mM , 0 . 5% sodium deoxycholate , 1% Triton-X 100 ) was added and samples were disrupted with a Dounce homogenizer ( Cheng-He Company , Zhuhai , China ) ( Chen et al . , 2010 ) on ice for 5 min . Debris was removed by centrifuging at 12 , 000 rpm for 10 min at 4°C . The amount of GFP-RAB-3 input in each experiment was equalized before the pull-down assay . The worm lysates were incubated for 4 hr at 4°C with GST-tagged RBD RIM2 coupled to glutathione-Sepharose 4B ( GE Healthcare , USA ) . After washing three times , the GFP-RAB-3 level was analyzed by 10% SDS-PAGE followed by standard western blotting with an anti-GFP antibody ( 1:5000 dilution ) ( Santa Cruz Biotechnology ) . To express proteins in HEK293FT cells , cDNA fragments were amplified and cloned into modified pcDNATM3 . 1/myc-HIS ( − ) or pFLAG-CMV-2 vectors through standard procedures . HET293FT cells were cultured in DMEM medium supplemented with 12% FBS . Plasmid transfections were carried out using Lipofectamine 2000 ( Invitrogen , USA ) . 24 hours after transfection , cells were harvested and lysed for 10 min at 4°C . After centrifugation , the supernatants were incubated with anti-FLAG or anti-myc beads at 4°C for 4 hr . Samples were resolved by standard immunoblotting techniques . For co-immunoprecipitation experiments with purified proteins , the immunoprecipitated samples were eluted with elution buffer ( Thermal Scientific , USA ) and neutralized with Tris buffer ( pH 9 . 0 ) . For GFP-tagged proteins , anti-GFP antibody ( Abcam , USA ) was incubated with the protein supernatant . Worms with different genotypes were collected and washed in M9 buffer . 500 µl of lysis buffer ( 250 mM sucrose , 50 mM Tris–HCl with pH6 . 8 , 1 mM EDTA ) were added and worm samples were homogenized with a Dounce homogenizer ( Cheng-He Company , Zhuhai , China ) ( Chen et al . , 2010 ) on ice for 15 min . The nuclear pellet was removed by centrifuging at 3000 rpm for 10 min at 4°C . The supernatant was further centrifuged at 40 , 000 rpm for 1 hr . The new supernatant was collected as the cytosolic fraction . The pellet was further washed and centrifuged at 40 , 000 rpm for 45 min . All samples were mixed with 2xSDS loading buffer before 10% SDS-PAGE gel analysis . The GFP-RAB-3 and tubulin levels in each fraction were analyzed by standard western blotting procedures . Images were captured using a Plan-Apochromat 40X/1 . 4 objective on an Olympus confocal microscope . All images were taken at the young adult stage unless specifically indicated . Images were analyzed with custom Image J software . Two main parameters were determined: puncta number ( PN ) and synaptic enrichment ( SE ) ( Ch'ng et al . , 2008 ) . These were calculated from the punctal fluorescence ( PF ) , which measures the signal intensity at the pre-synaptic specialization , and the inter-punctal fluorescence ( IPF ) , which measures the signal intensity in axons between synapses . An individual punctum is defined when the peak PF/average peak IPF is ≥2 . Synaptic enrichment is defined as total PF/total IPF within a 100-µm length of middle dorsal cord region . All data are shown as mean ± SD . Statistical analyses were performed with Student's t-test . For each genotype , more than 20 animals were imaged and analyzed .	Neurons communicate with one another at junctions called synapses . The arrival of an electrical signal called an action potential causes calcium ions to enter the first cell , which in turn triggers the release of molecules called neurotransmitters into the gap between the neurons . The binding of these molecules to receptors on the second cell then enables the action potential to be regenerated . For cells to respond rapidly and reliably to incoming electrical signals , they must maintain a supply of vesicles—the packages that contain neurotransmitters—close to the site where they are released from the first cell . The vesicles are held in contact with the cell membrane by a structure called the docking complex . A number of the proteins in this docking complex have been identified , including two that have been referred to as the ‘yin and yang’ of vesicle fusion: synaptotagmin , which promotes fusion , and Rab3 , which limits it . However , little is known about how these and other proteins interact to keep vesicles docked at the membrane . Cheng , Wang et al . have now clarified the docking process with the aid of experiments in nematode worms . In resting neurons that are not releasing neurotransmitters , synaptotagmin ( ‘yin’ ) binds to an enzyme called GAP and prevents it from converting GTP—an energy-storage molecule—into GDP . Given that Rab3 ( ‘yang’ ) requires a molecule of GTP to power its own activity , the actions of synaptotagmin ensure that Rab3 has enough energy to remain bound to other proteins within the docking complex . However , when an action potential arrives , calcium ions enter the neuron , and some of them bind to synaptotagmin . This disrupts its interaction with the GAP enzyme , which thus becomes free to convert the GTP molecule bound to Rab3 into GDP . The loss of its energy source causes Rab3 to separate from its binding partners , and docking complex collapses . As a result , vesicles fuse with the membrane and release neurotransmitter molecules into the synapse . Given that Rab3 and synaptotagmin have changed little over the course of evolution , it is highly likely that the same indirect interaction between these two proteins also regulates the release of transmitter at synapses in the mammalian brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2015	Synaptotagmin 1 directs repetitive release by coupling vesicle exocytosis to the Rab3 cycle
Modulation of synaptic vesicle retrieval is considered to be potentially important in steady-state synaptic performance . Here we show that at physiological temperature endocytosis kinetics at hippocampal and cortical nerve terminals show a bi-phasic dependence on electrical activity . Endocytosis accelerates for the first 15–25 APs during bursts of action potential firing , after which it slows with increasing burst length creating an optimum stimulus for this kinetic parameter . We show that activity-dependent acceleration is only prominent at physiological temperature and that the mechanism of this modulation is based on the dephosphorylation of dynamin 1 . Nerve terminals in which dynamin 1 and 3 have been replaced with dynamin 1 harboring dephospho- or phospho-mimetic mutations in the proline-rich domain eliminate the acceleration phase by either setting endocytosis at an accelerated state or a decelerated state , respectively . Synaptic transmission relies on a steady supply of release competent neurotransmitter-filled synaptic vesicles ( SVs ) to maintain information transmission in neural circuits . At small CNS nerve terminals the size of the recycling SV pool is often limited ( Harata et al . , 2001; de Lange et al . , 2003; Fernandez-Alfonso and Ryan , 2004; Kim and Ryan , 2010; Denker et al . , 2011 ) and therefore control of steps in SV recycling will likely prove important in setting overall synaptic performance . Given the importance of Ca2+ in regulating exocytosis , the role of this ion in regulating SV recycling steps has been of significant interest for over 30 years ( Ceccarelli et al . , 1979; Ceccarelli and Hurlbut , 1980 ) . At hippocampal and cortical nerve terminals synaptic vesicles are retrieved largely in a dynamin ( Ferguson et al . , 2007; Raimondi et al . , 2011 ) , clathrin ( Granseth et al . , 2006 ) and AP-2 dependent fashion ( Kim and Ryan , 2009 ) . Single vesicle retrieval studies ( Balaji and Ryan , 2007 ) demonstrated that endocytosis occurs in a probabilistic fashion , and the mean endocytosis time constant for large bursts ( 100 AP 10 Hz ) is a cell-wide property that can vary significantly from neuron to neuron ( Armbruster and Ryan , 2011 ) . Numerous studies have revealed a modulatory role for Ca2+ in endocytosis at nerve terminals in different synaptic preparations ( Balaji et al . , 2008; Sankaranarayanan and Ryan , 2001; von Gersdorff and Matthews , 1994; Wu et al . , 2009; Yamashita et al . , 2010; Yao et al . , 2009 ) although the precise steps at which Ca2+ acts in this process have not been identified . The direction of the Ca2+ modulation is also in dispute , some studies suggest an accelerating role of Ca2+ , ( Sankaranarayanan and Ryan , 2001; Wu et al . , 2009 ) while others suggest that Ca2+ slows endocytosis ( Balaji et al . , 2008; von Gersdorff and Matthews , 1994; Sun et al . , 2002; Sankaranarayanan and Ryan , 2000 ) . Although a number of proteins with Ca2+ sensing domains have been implicated in SV endocytosis ( synaptotagmin , calmodulin , calcineurin ) direct links connecting endocytic behavior to consequences of Ca2+ sensing have not been established . We took advantage of the ability to map endocytosis kinetics with high fidelity using a pHluorin-tagged synaptic vesicle protein to show that endocytosis has a pronounced acceleration phase with increasing number of action potentials used to elicit exocytosis . This initial acceleration phase is followed by a gradual slowing with increasing stimulus number . Our previous studies had missed this acceleration phase , in part due to the variability of endocytosis time constants across different neurons and in part owing to the fact that it is only readily apparent at physiological temperatures ( 37°C ) . We show that both phases of stimulus-dependent endocytosis are modulated by Ca2+ . The existence of two phases with opposite sign implies endocytosis has an optimal minimal value that is tuned by calcium-dependent processes . Dynamin , a mechano-chemical enzyme that plays a key role in membrane fission ( Ferguson and De Camilli , 2012 ) , was identified in a search for proteins whose phosphorylation decreases upon Ca2+ entry at nerve terminals ( Robinson and Dunkley , 1983 ) . Subsequent studies demonstrated two specific serines in dynamin’s proline rich domain as substrates for the calcium-dependent phosphatase calcineurin and the proline-directed serine/threonine kinase CDK5 ( Graham et al . , 2007 ) . In order to investigate the role of dynamin’s phosphorylation in controlling endocytosis we made use of the fact that eliminating the two major brain isoforms ( dynamin 1 and 3 ) results in a dramatic ( >10-fold ) slowing in endocytosis kinetics ( Raimondi et al . , 2011 ) . This strong phenotype allowed us to sensitively probe the ability of different dynamin isoforms to restore endocytic function in response to varying stimulus conditions . We show that mutating the two key phosphorylated serines in dynamin 1 to either alanine or aspartate both rescue the major endocytic defect of the dynamin 1/3 KO however they both eliminate the activity-dependent acceleration of endocytosis kinetics . These studies thus pinpoint dynamin 1 as a critical substrate in activity-dependent modulation of synaptic vesicle endocytosis and that it forms a basis for fine-tuning the retrieval process . We made use of pHluorin-tagged vesicular glutamate transporter ( vG-pH ) transfected into primary neurons to provide high sensitivity optical assays of synaptic vesicle endocytosis . vG-pH fluorescence is quenched by the acidic lumen of the synaptic vesicle ( pH 5 . 6 ) . Upon exocytosis the vesicular fluorescence increases ∼20-fold and is requenched by reacidification after endocytosis ( Sankaranarayanan et al . , 2000 ) . The endocytic time constant can be deconvolved from reacidification and measured from the fluorescence decay after stimulation ( Granseth et al . , 2006; Balaji et al . , 2008 ) where it has been shown in numerous studies to follow simple single exponential decay kinetics over a broad range of stimulus conditions ( Balaji and Ryan , 2007; Armbruster and Ryan , 2011; Yao et al . , 2011; Kwon and Chapman , 2012; Willox and Royle , 2012 ) . Changes in reacidification rates are only expected to minimally impact endocytosis time constant estimates ( see ‘Materials and methods’ ) and for the relevant stimulus conditions asynchronous release would unlikely affect the measures of endocytosis ( Atluri and Ryan , 2006; Granseth et al . , 2006 ) . We recently extended this technology to characterize endocytosis at individual boutons ( Armbruster and Ryan , 2011 ) where the probe allows for many rounds of stimulation and recovery within a given field of view , thus allowing the same synaptic boutons to be probed many times with different stimulus conditions . Our earlier examination of stimulus dependence relied more heavily on analysis of ensemble behavior across many cells ( Balaji et al . , 2008 ) where we saw little evidence for modulation of endocytosis kinetics for stimuli <100 AP . Comparison of fluorescence recovery profiles for exocytosis triggered with a 100 AP and 10 AP at 10 Hz ( at 30°C ) at the same boutons however revealed that the exponential endocytic decay was slower for the larger stimulation ( Figure 1A ) . Experiments carried out at an individual set of boutons for a range of stimuli ( 10 , 25 , 50 , and 100 AP ) showed that this trend was continuous , with gradual slowing of the endocytic time constant ( τendo ) with increasing stimulus number ( Figure 1B ) . The degree of slowing was most easily parameterized as a linear dependence on the number of action potentials used to drive exocytosis ( Figure 1B ) . We examined this dependence on stimulus number across collections of boutons from many individual neurons ( N = 44 ) and found both the degree of slowing ( i . e . , the slope in s/AP ) and the extrapolated fastest time constant expected at 1 AP ( the intercept ) to vary significantly across neurons ( Figure 1C ) but were uncorrelated with each other . The mean degree of slowing for the population of neurons was 0 . 058 ± 0 . 004 s/AP . 10 . 7554/eLife . 00845 . 003Figure 1 . Calcium slows endocytosis at 30°C . ( A ) Endocytosis decays from a 100 AP 10 Hz run ( black ) and 10 AP 10 Hz runs ( 2–3 runs averaged ) ( gray ) , inset shows endocytosis phase fit with a single exponential decay ( red ) to measure endocytosis time constant = 13 . 9 ± 0 . 10 s , 6 . 8 ± 0 . 3 s 47 boutons . adj . R-square of fits 0 . 95 , 0 . 997 respectively . ( B ) An example cell probed multiple times at 10 , 25 , 50 , 100 AP at 10 Hz at 30°C , fit with a linear dependence with a slope of 0 . 058 ± 0 . 004 s , a predicted 1 AP time constant of 6 . 09 ± 0 . 11 s . ( C ) Across 44 cells , the slope ( s/AP ) is plotted against the predicted 1 AP time constant ( s ) . Average slope 0 . 053 ± 0 . 008 s , average predicted 1 AP time constant 8 . 31 ± 0 . 64 s . There is a CV of 100% in the slope and 52% in the predicted 1 AP time constant . ( D ) Endocytosis for 25 , 50 and 100 AP delivered in 2 mM and 4 mM external Ca2+ for the same set of boutons from one cell . Each condition was probed 1–4 times and averaged over 46 ROIs . Slope of 2 mM is 0 . 01 ± 0 . 01 s/AP , 4 mM is 0 . 13 ± 0 . 03 s/AP . The intercept of 2 mM is 9 . 3 ± 0 . 7 s , 4 mM is 5 . 7 ± 2 . 0 s . ( E ) Across 10 cells the percentage change in slope and predicted 1 AP time constant when changing from 2 mM to 4 mM external Ca2+ slope and predicted 1 AP time constant . The change in slope is significantly different from 0 , ( log corrected one sample t-test p<0 . 004 ) . ( F ) 25 , 50 , and 100 AP at 10 Hz endocytosis time constant compared before and after 90 s load of 100 μM EGTA-AM reveals an acceleration phase of endocytosis for low stimulus number ( 25 AP significantly different before and after EGTA treatment , paired sample t-test p<0 . 03 , N = 6 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 003 Comparisons of the slowing behavior in the same synapses for stimuli delivered in 4 mM compared to 2 mM external Ca2+ demonstrated that this inhibition of endocytosis appears to be enhanced under conditions that would lead to more elevated intracellular calcium ( Figure 1D ) . These experiments revealed that the activity-dependent slowing was always steeper when stimuli were delivered in the higher Ca2+ concentration . On average ( n = 9 ) the slope increased by ∼700% ( Figure 1E ) while the intercept was unchanged . The difference in slopes is not dependent simply on examining how τendo changes with stimulus number and can be readily observed if one alternatively examines the relationship of τendo vs exocytosis ( Figure 2—figure supplement 1 ) . We further examined a possible influence of Ca2+ on the activity-dependence of endocytosis by measuring the activity dependence before and after loading nerve terminals with the Ca2+ chelator EGTA-AM which at 30°C ( the temperature used for these experiments ) still allows for significant exocytosis . These experiments revealed that buffering intracellular Ca2+ slowed endocytosis compared to control across stimuli but eliminated slowing as a function of stimulation . Additionally incubation with EGTA-AM unmasked a modest acceleration phase for the lowest stimuli ( Figure 1F ) . Similar to the impact of varying external Ca2+ , the impact of EGTA on stimulus-dependent slowing was independent of any changes in exocytosis ( data not shown ) . The continuous slowing of τendo with increasing stimulus number in the 10–100 AP range predicts that the value of τendo for single AP stimulation ( i . e . , the predicted intercept in Figure 1B ) would be the fastest of any stimulus . However we and others have previously shown that the endocytic recovery for single AP stimuli were similar to that obtained for much larger stimuli ( Granseth et al . , 2006; Balaji and Ryan , 2007 ) . Additionally our experiments with EGTA buffering suggests that endocytosis would have the opposite behavior in the low stimulus regime , given that smaller stimuli lead to less total Ca2+ entry . To directly examine this we systematically examined endocytosis following single AP stimulation as well as following 25 , 50 and 100 AP in the same boutons . ( Figure 2A , B ) . The observed single AP τendo was slower than for that obtained at 25 AP ( Figure 2A ) , as well as than that predicted from the linear regression at higher stimulus levels ( Figure 2B , C ) . Taken together these data thus demonstrate two distinct phases of activity-dependence: an acceleration of endocytosis for stimuli >1 AP and a slowing for stimuli > approximately 10 AP . This predicted acceleration phase agrees with the results of our EGTA experiments ( Figure 1F ) where buffering intracellular Ca2+ during bursts of AP revealed an acceleration phase for lower stimulus numbers . 10 . 7554/eLife . 00845 . 004Figure 2 . Acceleration of endocytosis for small stimuli . ( A ) Individual example traces of a cell probed with 1 AP ( black ) and 25 AP at 10 Hz ( gray ) , inset shows endocytosis phase with their fits ( red ) at 30°C . 1 AP time constant τ = 12 . 36 ± 1 . 1 s , 25 AP 10 Hz time constant τ = 8 . 57 ± 0 . 67 s , adj . R-square of fits 0 . 93 , 0 . 99 respectively . ( B ) The same example cell probed at 1 , 25 , 50 , 100 AP at 10 Hz at 30°C , with a linear fit to the 25 , 50 , 100 AP data . Each point is an average of 2–3 runs based on 76 ROIs . Predicted 1 AP time constant = 6 . 55 ± 0 . 54 s . ( C ) Across 10 cells , the predicted 1 AP time constant based upon linear fit to 25 , 50 , 100 AP 10 Hz data compared to observed 1 AP time constant . The difference is significant paired sample t-test p<0 . 01 . ( D ) At physiological temperature , ( 37°C black ) probing 5 , 10 , 15 , 25 , 50 , 100 AP at 10 Hz , normalized for 100 AP 10 Hz tau for each cell , showing the acceleration of endocytosis . N = 8 cells . 5 AP ( maximum ) vs 25 AP ( minimum ) is significant p<0 . 01 , 25 AP vs 100 AP is significant p<0 . 03 paired sample t-tests , 5 AP vs 10 AP is significant p<0 . 009 paired sample t-tests . At 30°C ( red ) there is no significant acceleration over a similar range of stimuli 1 AP vs 10 AP ( minimum ) p>0 . 05 paired t-test . N = 8 , 7 , 12 , 14 , 12 , 14 cells for 1 , 5 , 10 , 25 , 50 , and 100 AP 10 Hz respectively . ( E ) Paired comparisons of 1 AP stimulation at 2 mM and 4 mM extracellular Ca2+ showing a Ca2+ dependent acceleration ( N = 6 cells , significant difference in remaining fluorescence at 20 s paired t-test p<0 . 05 ) . ( F ) Acceleration of endocytosis measured for 30 Hz AP bursts normalized to the value obtained at 15 AP , but including the relative value measured for 1 AP ( 2 mM ) in D . N = 3 , 4 , 5 , 4 , 4 cells for stimuli 100 , 50 , 25 , 15 10 AP at 30 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 00410 . 7554/eLife . 00845 . 005Figure 2—figure supplement 1 . Comparing the effects of 2 mM and 4 mM external Ca2+ on the slopes of endocytosis corrected for changes in exocytosis . ( A ) Example from a single cell plotting endocytosis time constant for a variety of stimuli at 2 mM and 4 mM compared to the endocytic load ( Fluorescence at the end of the stimulus ) . Linear fits to the slopes at 2 mM and 4 mM Ca2+; Slopes 0 . 01 ± 0 . 002 s/ΔF , 0 . 039 ± 0 . 006 s/ΔF respectively; Intercepts at 0 ΔF were 13 . 0 ± 1 . 2 s , 0 . 7 ± 3 . 3 s respectively . ( B ) Statistics on the ratios of the slopes ( s/ΔF ) , 0 . 42 ± 0 . 16 , and the intercepts ( s ) , 1 . 4 ± 0 . 5 . The slope is significantly different from 1 , one-sample t-test p<0 . 01 . This indicates that the changes in slope associated with changes in Ca2+ are not due to differences in the endocytic load . N = 10 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 00510 . 7554/eLife . 00845 . 006Figure 2—figure supplement 2 . Mapping the acceleration notch curve at 30 Hz , at 100 , 50 , 25 , 15 , 10 AP . Each cell is normalized to its 100 AP 10 Hz time constant . The acceleration minimum is difficult to resolve with the increased Ca2+ influx; N=3 , 4 , 5 , 4 , 4 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 006 The existence of these two opposite phases of Ca2+ modulation on endocytosis thus appears critical in determining the endocytosis time constant and would likely be influenced by any factors that impact the degree of calcium entry and accumulation during AP bursts . One such critical physiological parameter is temperature . The experiments above were all carried out at 30°C . Although this temperature was chosen to provide a practical signal to noise ratio for experiments , given the known temperature-dependence on Ca2+ handling ( Sabatini and Regehr , 1998 ) and action potential waveforms ( Hlavova et al . , 1970 ) we reasoned that endocytic optima might be hard to pinpoint unless one worked at physiological temperature . For this reason we explored the activity-dependent behavior of endocytosis for a range of stimuli ( between 1 AP and 100 AP at 10 Hz ) at 37°C and 30°C ( Figure 2D ) . Given the large variability from cell to cell in endocytosis time constant , in order to compare stimulus-dependent variation across cells in each experiment the data were internally normalized to the value for τendo obtained for 100 AP . Data normalized in this fashion revealed the expected slowing behavior between 10 AP and 100 AP . Although the acceleration between 1 AP and 10 AP was readily apparent for individual cells ( Figure 2B ) at 30°C , on average this acceleration of endocytosis was not easily resolvable at this cooler temperature ( Figure 2D ) . However these experiments revealed a clear and robust acceleration phase for endocytosis at physiological temperature with stimuli ranging from 5 AP to 25 AP and milder slowing phase for stimuli above this than that seen at 30°C , with a minimum endocytosis time constant in the vicinity of 25 AP at 10 Hz . At physiological temperature , on average , endocytosis accelerated by 42 ± 7% ( N = 9 ) going from 5 AP to 25 AP and then slowed by 18 ± 4% ( N = 8 ) at the 100 AP stimulus level . The prominent slowing phase of endocytosis present at 30°C however was still observed at 37°C but it required larger stimuli ( 300 AP ) to show obvious ( 55 ± 20% [N = 7] ) slowing when compared to 100 AP 10 Hz . The more pronounced acceleration of endocytosis in the low stimulus regime implicates calcium as a likely modulator for mediating the acceleration as intracellular calcium would likely accumulate during brief bursts . We tested this notion explicitly by comparing the endocytic time constant for single AP stimulation at 2 mM and 4 mM Ca2+ at 37°C . These experiments showed that increasing Ca2+ accelerated the post-stimulus endocytosis kinetics significantly ( Figure 2E ) . The acceleration is more clearly present at physiological temperature , we expect through reduced Ca2+ influx on a per action potential basis . Conversely at a higher stimulation ( 30 Hz ) frequency the acceleration occurs over a very small stimulus range ( Figure 2—figure supplement 2 ) . This makes the optimum stimulus for endocytosis hard to discern , but shows a larger relative impact on endocytosis relative to single AP responses ( Figure 2F ) . Increasing the stimulation from 5 AP to 10 AP at 10 Hz accelerates the endocytosis time constant ( Figure 2D ) , which cannot be explained by changes in the reacidifcation kinetics ( see ‘Materials and methods’ ) . However it is presumably a Ca2+ dependent process ( Figure 1F ) . In order to narrow down possible mechanisms of Ca2+ action in this process we sought to determine how long the effect of a burst of stimulation would last in accelerating endocytosis . To examine this we designed a protocol where we examined endocytosis following a 5 AP burst delivered at different inter-burst intervals for five total bursts . For an inter-burst interval of 0 s it is the equivalent to looking at a single prolonged burst of 5 , 10 , 15 , or 25 AP at 10 Hz . A representative example using a 30 s inter-burst interval is shown in Figure 3A . These experiments revealed that the acceleration of endocytosis caused by 5 AP persists for at least 15 s , but is lost for intervals >30 s . To compare results across many cells data for each cell were normalized to the value of τendo for a single 5 AP burst for each experiment and inter-burst intervals of 0 s ( Figure 3C ) , 15 s ( Figure 3D ) , 20 s ( Figure 3E ) and 30 s ( Figure 3F ) were examined for many cells . For the 15 s inter-burst interval ( Figure 3D ) we relied on characterizing the endocytic time scale as ( 1/rate ) as the time frame was too compressed to allow accurate exponential fitting . These experiments showed that endocytosis was accelerated for a continuous 10 AP burst compared to a 5 AP ( Figure 3C , similar to Figure 2D ) . The impact of a single 5 AP burst on accelerating endocytosis however persisted for at least 20 s , as τendo for a second 5 AP burst was accelerated to the same extent as providing a continuous 10 AP burst even if the second burst was delivered 20 s later . Measurements for a 30 s inter-burst interval however showed no significant acceleration . Example traces of endocytosis for the first 2 bursts for 20 s and 30 s intervals from two different cells ( Figure 3B ) illustrate this point . These experiments show that the acceleration persists for ∼20 s between stimuli but is dissipated after ∼30 s . Given that elevations in intracellular Ca2+ following a 5 AP burst decays on much faster time scales ( <1 s ) these data imply that Ca2+ is likely acting through a second messenger system to control endocytosis in the acceleration phase . 10 . 7554/eLife . 00845 . 007Figure 3 . Persistence time of endocytic acceleration . ( A ) A sequence of 5 bursts of 5 AP 10 Hz with inter-burst interval of 30 s apart from one cell averaged over 4 runs , each decay is fit with an exponential decay . Endocytosis time constants = 9 . 7 ± 0 . 8 s , 7 . 9 ± 0 . 5 s , 7 . 2 ± 0 . 4 s , 9 . 2 ± 0 . 5 s , 6 . 7 ± 0 . 3 s respectively for pulses 1–5 . ( B ) Example endocytic decays from pulses 1 and 2 with 20 s and 30 s spacing illustrating the acceleration of endocytosis for 20 s inter-burst interval spacing: 20 s time constants 8 . 6 ± 0 . 5 s and 4 . 8 ± 0 . 2 s for first and second burst respectively , adj . R-square of fits 0 . 88 , 0 . 93 respectively; 30 s time constants 8 . 3 ± 0 . 2 s and 7 . 7 ± 0 . 2 s first and second burst respectively , adj . R-square of fits 0 . 96 and 0 . 97 respectively . Data were averaged over 5–12 runs across 30–50 boutons . ( C ) Inter-burst interval of 0 s , based upon data from Figure 2D , normalized to 5 AP 10 Hz shows significant acceleration of endocytosis , p<0 . 02 for 10 AP , 15 AP , and 25 AP 10 Hz pulses n = 8 cells . ( D ) 15 s spacing with linear fits for the decays , plotting 1/Rate of endocytosis , showing significant acceleration of endocytosis compared to the first burst for all subsequent bursts , p<0 . 03 for bursts 2–5 . n = 11 cells . ( E ) 20 s spacing , fit with exponential time constants , bursts 2 and 3 are significantly accelerated compared to the first burst p<0 . 05 , n = 9 cells . ( F ) 30 s inter-burst interval with exponential fits , only the fifth burst is significantly accelerated compared to the first burst p<0 . 04 , n = 9 cells . All tests one sample t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 007 A potential candidate for an endocytic mechanism that is mediated by a second messenger downstream of Ca2+ is the control of the phosphorylation state of dephosphin proteins . Dynamin 1 , the founding member of the dephosphin family and by far the most abundant neuronal dynamin , is constitutively phosphorylated at serines 774 and 778 by Cdk5 and dephosphorylated in a stimulus- induced manner at the same sites by the Ca2+ dependent phosphatase calcineurin ( Liu et al . , 1994 ) . A major effect of this phospho-regulation is to control the interaction of dynamin 1 with syndapin , whose binding to dephospho-dynamin is abolished by phosphorylation ( Anggono et al . , 2006; Anggono and Robinson , 2007 ) . Interestingly , dephosphorylation of dynamin after a stimulatory burst has been shown to persist for ∼40 s under certain stimulus conditions ( Robinson et al . , 1994 ) . We previously showed that dynamin 1/3 DKO mice have severely impaired endocytosis; however , nerve terminals lacking these major dynamin isoforms are still able to undergo multiple rounds of vesicle recycling ( Raimondi et al . , 2011 ) . We re-examined endocytosis in dynamin 1/3 DKO neurons at 37°C which showed a very similar phenotype to our previous studies ( Figure 4A ) : the absence of both dynamin 1 and 3 results in a severe impairment of endocytosis , which can be fully rescued by re-introduction of dynamin 1 . These studies in mouse neurons necessitated using cortical neurons rather than hippocampal neurons due to the small size of the hippocampus in newborn mice , and the smaller total neuronal population . We therefore reexamined the stimulus dependence of endocytosis in mouse cortical neurons . These experiments showed very similar behavior as rat hippocampal neurons with the exception that the slowing phase was much less pronounced and was only evident at much higher stimulus levels ( see below ) . The acceleration phase however was readily apparent in both the mouse cortical neuron controls and dynamin 1/3 DKO mouse cortical neurons in which dynamin 1 has been reintroduced ( dynamin 1 rescue , Figure 4B ) showing a 31 ± 12% ( N = 9 ) and 21 ± 8% ( N = 7 ) acceleration between 10 AP and 100 AP in the dynamin 1 rescue and control respectively ( compared to 42 ± 7% [N = 9] ) for rat hippocampal neurons . Consistent with the need to use greater stimulation to see the slowing phase , mouse cortical neurons appear to have their acceleration phase shifted slightly to higher stimulus numbers compared to rat hippocampal neurons as well . In order to examine the possible role of dynamin 1 dephosphorylation in mediating acceleration of endocytosis we mutated serines 774 and 778 to either alanine or aspartate , used these isoforms of dynamin 1 to rescue the endocytic defect in dynamin 1/3 DKO neurons and examined the stimulus-dependence of endocytosis . Previous studies have shown that mutations of these serines to alanine and aspartate mimic the dephosphorylated and phosphorylated states respectively with respect to dynamin 1’s ability to bind syndapin ( Anggono et al . , 2006 ) . Although both phosphomimetic ( S774/778D ) and phospho-deficient ( S774/778A ) mutants could efficiently rescue the severe dynamin 1/3 DKO endocytosis defect , neither showed any activity dependent acceleration ( Figure 4C , D ) . The phosphomimetic form was effectively locked in a slower endocytic state , while the phospho-deficient form was faster across all stimuli ( Figure 4C , E ) . Thus , neurons expressing mutants of dynamin 1 that lock the phosphorylation sites in a specific state do not undergo stimulus-dependent regulation and constitutively proceed at overall faster ( phospho-deficient mutant ) or slower ( phospho-mimetic mutant ) endocytic speeds . Dynamin 2 , which has different activity dependent phosphorylation sites ( Chircop et al . , 2011 ) , was only able to partially rescue the 100 AP 10 Hz time constant and was not tested further . Increasing the stimulus number from 100 AP to 300 AP showed a slowing of endocytosis for all conditions similar to the relationship described at 30°C ( Figure 4F ) . Using the bafilomycin method ( Sankaranarayanan and Ryan , 2001 ) we detected no difference in the size of the recycling pool of synaptic vesicles or in the rate of exocytosis between control , dynamin 1 , phosphomimetic , or phospho-deficient rescue ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 00845 . 008Figure 4 . Dephosphin control of endocytic acceleration . ( A ) 100 AP 10 Hz stimulation of dynamin 1/3 DKO , dynamin 1 rescue expressed in DKO , or the dynamin 1 Het dynamin 3 KO control genotype at 37°C . n = 5 , 7 , 10 cells respectively . ( B ) Comparison of endocytosis acceleration in rat neurons ( top ) cortical mouse neurons ( middle ) and cortical mouse dynamin 1/3 DKO neurons rescued with dynamin 1 . For each cell data is normalized to the value obtained for 100 AP 10 Hz , N = 8 rat hippocampus , 7 mouse cortical , 9 dynamin 1 rescue . Mouse cortical 10 AP compared to 50 AP ( minimum ) is significant p<0 . 03; dynamin 1 rescue 10 AP compared to 100 AP ( minimum ) is significant p<0 . 05 paired sample t-tests . ( C ) Endocytosis vs stimulation for dynamin 1/3 DKO rescued with the full length dynamin 1 ( black , replotted from B , bottom ) , phospho-deficient , S774/8A ( pink ) , or the phosphomimetic S774/8D mutants of dynamin 1 ( gold ) shows that mutations at these serines block acceleration and lock endocytosis in a fast or slow state . Individual traces are normalized to the 100 AP 10 Hz value for the dynamin 1 rescue . N = 9 cells dynamin 1 rescue , N = 6 cells phospho-deficient , N = 8 cells phosphomimetic . ( D ) Example traces of 10 AP compared to 100 AP for dynamin 1 rescue ( adj . R-square of fits 0 . 97 , 0 . 97 100 AP , 10 AP respectively ) , phospho-deficient rescue ( adj . R-square of fits 0 . 98 , 0 . 89 100 AP , 10 AP respectively ) , and phosphomimetic rescue ( adj . R-square of fits 0 . 999 , 0 . 97 100 AP 10 AP respectively ) of dynamin 1/3DKO showing the lack of acceleration for the phosphorylation mutants . Single cells based upon 30–50 boutons and 1–3 runs . Time constants of decays: dynamin 1 rescue 25 . 1 ± 0 . 4 s and 9 . 7 ± 0 . 4 s , phospho-deficient rescue 6 . 8 ± 1 . 2 s and 9 . 4 ± 0 . 5 , phosphomimetic rescue 12 . 4 ± 1 . 1 s and 12 . 0 ± 0 . 1 s for 10 AP and 100 AP respectively . ( E ) The 100 AP 10 Hz time constants of the dynamin 1/3 DKO rescued with dynamin 1 , phosphorylation mutants and dynamin 2 . Phosphomimetic is significantly different from control KS-test p<0 . 003 . Dynamin 2 is significantly different from control KS-test p<0 . 0004 . N = 7 , 10 , 9 , 12 , and 11 cells respectively . ( F ) Paired 100 AP and 300 AP 10 Hz stimulus , endocytosis time constants for the phosphomutants and control . Data normalized to 100 AP 10 Hz paired t-test p<0 . 03 for all conditions . 300 AP τendo phospho-deficient 1 . 39 ± 0 . 13 s , phosphomimetic 1 . 33 ± 0 . 13 s , control 1 . 66 ± 0 . 20 s . N = 9 , 15 , 9 cells respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 00810 . 7554/eLife . 00845 . 009Figure 4—figure supplement 1 . Exocytosis and pool size controls for dynamin rescues . Exocytosis and recycling pool size are measure by applying the vesicular proton pump inhibitor bafilomycin and stimulating for 1000 AP at 10 Hz . The rate of fluorescent increase gives a measure of exocytosis and the absolute fluorescence represents the size of the recycling pool . ( A ) No changes are observed in the size of the recycling pool , for the full length dynamin 1 rescue , control , dynamin 1 phospho-deficient mutant , or dynamin 1 phosphomimetic mutant as measured with the bafilomycin assay . ( B ) No significant change is observed in the bafliomycin time constant , a measure of exocytosis for the full length dynamin 1 rescue , control , dynamin 1 phospho-deficient mutant , or dynamin 1 phosphomimetic mutant . N = 9 , 5 , 5 , 7 for dynamin 1 , control , phospho-deficient , phosphomimetic . DOI: http://dx . doi . org/10 . 7554/eLife . 00845 . 009 Using high-sensitivity pHluorin assays of synaptic vesicle endocytosis for individual neurons we revealed the presence of two phases of stimulus and Ca2+ dependence of synaptic vesicle endocytosis: an acceleration phase prominent for small stimuli and a slowing phase prominent for larger stimuli . Our experiments made use of pHluorin-tagged vGlut expressed in dissociated neurons . Although it is likely that these data sets arise from a mixture of gabaergic and glutamatergic neurons , we previously showed that these two neuron classes differ little in endocytic behavior , even when vGlut is expressed in a gabaergic neuron ( Armbruster and Ryan , 2011 ) , perhaps owing to the tracer-level expression of this probe ( Balaji and Ryan , 2007 ) . While the vGlut-pHluorin reporter only tracks the internalization of vGlut , previous synaptic vesicle endocytosis studies have shown SynaptopHluorin ( Vamp2 ) , Synaptophysin-pHluorin , and Synaptotagmin-pHluorin all showing the same kinetics as vGlut-pHluorin as well as the same dependence on the clathrin adaptor AP-2 ( Kim and Ryan , 2009 ) . Our data helps reconcile numerous reports on the Ca2+ sensitivity of synaptic vesicle endocytosis . Ca2+ has long been implicated in controlling synaptic vesicle endocytosis and numerous calcium-sensing proteins have been implicated in synaptic vesicle retrieval including calcineurin , synaptotagmin , and calmodulin ( Poskanzer et al . , 2003; Nicholson-Tomishima and Ryan , 2004; Poskanzer et al . , 2006; Yao et al . , 2011; Yao and Sakaba , 2012 ) . While inhibiting or mutating these Ca2+ sensors impact synaptic vesicle endocytosis , to our knowledge these putative modulators of endocytosis had not been directly linked to demonstrated modulation of endocytosis kinetics . Our experiments show that dynamin 1 phosphorylation sites that were previously demonstrated to be calcineurin substrates are critical specifically for the acceleration phase during brief action potential bursts . The dephosphorylation accelerates the kinetics of endocytosis without fundamentally changing the endocytic properties suggesting that this represents a tuning of the mechanism rather than a distinct pathway . The dynamin 1/3 DKO provides the ideal background to test the significance of the phosphorylation sites , as it is necessary to also remove dynamin 3 since it has a similar phospho-box motif as dynamin 1 ( Larsen et al . , 2004 ) . Although recent studies implicate dynamin’s phosphorylation in controlling bulk endocytosis , analysis in dynamin 1/3 DKO nerve terminals indicates that this form of endocytosis proceeds unimpaired in the absence of dynamin ( Yumei Wu , Shawn Ferguson and Pietro De Camilli , in preparation ) . Analysis of single synaptic vesicle retrieval demonstrated that endocytosis is stochastic and the mean time for endocytosis is determined by a single rate-limiting step ( Balaji and Ryan , 2007 ) . As dynamin is considered to be a critical enzyme in endocytic membrane fission ( Ferguson and De Camilli , 2012 ) , the fact that endocytosis kinetics can be accelerated by dephosphorylation of dynamin 1 suggests that under these conditions membrane fission , or another dynamin-dependent event , is the rate limiting step in endocytosis . Phosphorylation of dynamin at serines 774 and 778 abolishes the interaction with the F-BAR protein syndapin and dephosphorylation promotes it ( Anggono et al . , 2006; Anggono and Robinson , 2007 ) . Our data therefore suggest that the cooperative participation of syndapin and dephosphorylated dynamin improves the efficiency of the endocytic process: when dynamin cannot bind syndapin ( phosphomimetic mutation ) no acceleration occurs , while when dynamin can interact with syndapin even at rest ( phospho-deficient mutation ) , endocytosis is already fast and no stimulus-dependent acceleration occurs . Only once further atomic-detail of the basis of syndapin’s interaction with dynamin have been revealed will it be possible to test specifically the potential role of syndapin in this acceleration . Consistent with this scenario however , genetic ablation of syndapin 1 , the major syndapin isoform expressed in brain , revealed a number of pleotropic phenotypes consistent with a failure to properly recruit dynamin to membranes ( Koch et al . , 2011 ) and syndapin is one of the proteins whose levels are more strongly reduced in mice that lack dynamin 1 and 3 ( Raimondi et al . , 2011 ) . The Ca2+ dependent slowing phase we observed was more prominent at lower temperatures and agrees well with previous findings that show Ca2+ inhibition of endocytosis ( Balaji et al . , 2008; von Gersdorff and Matthews , 1994; Sun et al . , 2002 ) . The Ca2+ sensor for this mechanism is not known , although we showed that it is not affected by setting dynamin’s phosphorylation state at the calcineurin sites . Previously a number of studies examined the role of the specific enzymes , calcineurin and CDK5 , that control dynamin phosphorylation ( Liu et al . , 1994; Tan et al . , 2003; Tomizawa et al . , 2003 ) , in controlling vesicle recycling . Although manipulation of these enzymatic activities impacted synaptic vesicle endocytosis , it is seems likely that these manipulations do not solely alter dynamin activity given that they target numerous substrates . Recent studies have recently shown for example that both these enzymes profoundly modulate Ca2+ influx ( Kim and Ryan , 2013 ) precluding simple interpretations of such pharmacological manipulations on endocytosis . Our data indicate that at 37°C synaptic vesicle retrieval appears to be optimized for brief action potential bursts and that the basis for this tuning is based on a balance of calcineurin activation vs an additional calcium-dependent inhibitory effect that dominates during prolonged stimulation . Furthermore dynamin appears to be centrally important in this short term optimization as preventing changes in the phosphorylation at two key serines in dynamin 1 eliminates this tuning of endocytosis . This data compels one to speculate that cells might tune their endocytic profile to be optimized for individual firing patterns . The tuning could be achieved by altering the balance of CDK5 and calcineurin activity as has been demonstrated for certain forms of homeostatic plasticity ( Kim and Ryan , 2010 ) , or by altering routes of Ca2+ entry or clearance . These phosphorylation sites in dynamin are well conserved across species with a nervous system suggesting that that endocytic optimization is a fundamental property of nervous system function . Hippocampal CA3–CA1 regions were dissected from 1- to 3-day-old Sprague Dawley rats , dissociated , and plated onto poly-ornithine-coated glass and grown for 14–26 days as described previously ( Ryan , 1999 ) . For experiments utilizing the dynamin knockout mice and littermate controls ( dynamin 1 het , dynamin 3 KO ) cortexes were dissected from postnatal 0 to 1-day-old mice were dissociated and plated onto poly-ornithine coated glass as previously described ( Ferguson et al . , 2007; Raimondi et al . , 2011 ) . Cultures were transfected with calcium-phosphate 7–8 days after plating and imaging was performed 13–26 days after plating ( 5–18 days after transfection ) . The reporter used was a chimera of the pH sensitive GFP , pHluorin and the vesicular glutamate transporter made by the Voglmaier lab ( UCSF ) . Constructs for human dynamin 1 ( aa spliceform ) , rat dynamin 2-mRFP ( AAB spliceform ) , human dynamin 1 S774/778A ( aa spliceform ) , and human dynamin 1 S774/778D ( aa spliceform ) were used . Between species ( mouse/rat ) and ( mouse/human ) there is >99% amino acid identity . Coverslips were mounted in a rapid-switching , laminar-flow perfusion and stimulation chamber ( volume ∼75 μl ) on the stage of a custom-built laser-illuminated epifluorescence microscope . Cells were perfused with a solution containing in mM: 119 NaCl , 2 . 5 KCl , 2CaCl2 , 2MgCl2 , 25 HEPES ( buffered to pH 7 . 4 ) , 30 glucose supplemented with 10 μM 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX ) , and 50 µM D , L-2-amino-5-phosphonovaleric acid ( AP5 ) . For experiments involving 4 mM Ca2+ tyrodes solution , CaCl2 was swapped for MgCl2 . All chemicals were obtained from Sigma-Aldrich ( St Louis , MO ) . Due to the low surface fraction of vG-pH ( Balaji and Ryan , 2007 ) , we gave brief bursts with 6 APs at 30 Hz every 4 s to find transfected cells in a dish . Identity of genotype and transfected plasmids were known to the investigator when performing imaging . Perfusion was kept between 75–250 μl per minute to ensure prolonged cell survival . Cells were imaged either at 30°C , or 36 . 8°C by heating the microscope objective with a flexible resistive heater ( Omega , Stamford , CT ) utilizing an on–off controller ( Minco , Minneapolis , MN ) , which maintained the temperature at the objective within ±0 . 1°C as readout by a 100 Ω platinum thermistor ( Minco ) . Cells were illuminated utilizing a 488 nm diode pumped solid state laser ( Coherent , Santa Clara , CA ) , shuttered using an acousto-optic modulation during all periods without data acquisition . Fluorescence excitation and collection was through a 40X 1 . 3 NA Fluar Zeiss objective using 515–560 nm emission and 510 nm dichroic filters ( Chroma , Bellows Falls , VT ) and a 1 . 6X Optivar tube lens . Laser power at the back aperture was ∼1 mW , imaging onto a Andor iXon+ ( model number DU-897E-BV ) back-illuminated electron-multiplying charge coupled device camera . Action potentials were evoked by passing 1 ms current pulses , yielding fields of ∼10 V/cm via platinum-iridium electrodes from an Isolated current stimulator ( World Precision Instruments , Sarasota , FL ) . Images were analyzed in ImageJ ( http://rsb . info . nih . gov/ij/ ) using a custom-written plugin ( http://rsb . info . nih . gov/ij/plugins/time-series . html ) . 2 µm diameter circular ROIs were placed on all varicosities based upon the ΔF image of a 100 AP 10 Hz run , between 25–120 ROIs were used per cell . Only boutons that did not split or merge , remained in focus and responded throughout all trials were chosen . All small stimuli were additionally averaged over several rounds ( up to 10 ) of stimulation to increase the signal to noise before fitting . All fitting was done using OriginPro ( OriginLab , Northampton , MA ) with the Levenberg-Marquardt algorithm . Fits of the endocytosis time constant were single exponential decays with a temporal offset for reacidification of ∼2–3 s at 36 . 8°C and ∼5 s at 30°C as described previously ( Balaji and Ryan , 2007 ) . To assess the contribution of changing reacdification times to our endocytosis fits , we ran simulations based upon the biexponential model of synaptic vesicle endocytosis ( Granseth et al . , 2006 ) and our fitting protocols . A change in reacdification from 1 . 6 s to 2 . 6 s introduces a +8% error in the measurement of endocytosis for a 10 s time constant , suggesting that it is unlikely to explain our observations . In general , fits were conducted on the ensemble average of each run , or multiple runs for very small stimuli . All fits were visually inspected and we did not observe deviation from the single exponential characteristic . For 1 AP data at physiological temperature the remaining fluorescence 15 s after the end of stimulation is quantified as a more robust measure given the low signal to noise and very slow decays of some traces . For 30 s and 20 s spacing between bursts ( Figure 3 ) decays were fit with single exponential decays with constant time windows for each burst . For 15 s spacing , there is insufficient time to carry out a robust fit to a single exponential . For these experiments endocytic performance was estimated by a linear fit to the decay ( Rate ) and plotted as 1/Rate . Time constants or 1/Rate measures were normalized for the measure of the first burst in the 5 burst sequence . For physiological Ca2+ dependence studies , each cell was normalized to its 100 AP 10 Hz behavior . All statistical tests were done using OriginPro ( OriginLab ) , all statistical tests were two-sided . One sample t-test , and paired t-test were used for internal comparisons for changes within an individual cell depending upon normalization; unless otherwise stated data met the test criteria and was not transformed . Kolmogorov-smirnov tests were used to compare time constants between cells as has been previously established ( Armbruster and Ryan , 2011 ) . Dynamin 1 rescue expressed in dynamin 1/3 DKO had 2 outliers which were >3 standard deviations away from the mean at 28 s and 36 s . These cells were treated as outliers and are not included .	Neurons communicate with each other at specialized junctions called synapses . When signals travelling along a neuron reach the presynaptic cell , this triggers small packages ( vesicles ) containing neurotransmitter molecules to release their contents into the synapse , and these molecules then cross the gap and bind to receptors on the postsynaptic neuron . To release their cargo , individual vesicles fuse with the plasma membrane of the presynaptic neuron and form a ‘pore’ through which neurotransmitter molecules can leave the cell . However , to avoid running out of vesicles , the neuron must recycle and rebuild them through a process known as endocytosis . This involves recapturing the proteins that make up the synaptic vesicle and internalizing them back into the presynaptic terminal . Exactly how endocytosis is regulated has been the subject of much debate in recent years . Now , Armbruster et al . have used fluorescent markers to study the timing of endocytosis in unprecedented detail . Observations of individual synapses reveal that when a series of action potentials ( spikes of electrical activity ) occurs in a neuron , endocytosis accelerates during the first few action potentials , and then slows . However , this acceleration was only detectable at a physiological temperature of 37°C—markedly higher than the 30°C at which synaptic endocytosis is typically studied . The new study showed that acceleration of endocytosis depends on the phosphorylation status of dynamin , a mechano-chemical enzyme long known to be crucial for endocytosis , which helps to sever the connection between the endocytosing membrane and the surface of the cell . Phosphorylation is a common mechanism for controlling enzyme activity , and involves the addition of phosphate groups to specific amino acids by enzymes called kinases . Phosphatase enzymes reverse the process by removing the phosphate groups . Dynamin is usually phosphorylated at two specific amino acids , but when levels of calcium in the cell increase ( as occurs during action potentials ) , a phosphatase called calcineurin dephosphorylates these sites . Using versions of dynamin that were either permanently phosphorylated or never phosphorylated , Armbruster et al . showed that a decrease in dynamin phosphorylation was required for the initial acceleration of endocytosis . This type of regulation seems to optimize the recycling of vesicles to enable neurons to respond effectively to brief bursts of stimulation . Given that dynamin phosphorylation is conserved in evolution , it is likely that regulation of synaptic endocytosis is a key mechanism for ensuring the efficient functioning of the nervous system . Future research will investigate how calcium influx mediates the later slowing of endocytosis , and help to further unravel this previously unknown regulatory process .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2013	Dynamin phosphorylation controls optimization of endocytosis for brief action potential bursts
Large herbivores influence ecosystem functioning via their effects on vegetation at different spatial scales . It is often overlooked that the spatial distribution of large herbivores results from their responses to interacting top-down and bottom-up ecological gradients that create landscape-scale variation in the structure of the entire community . We studied the complexity of these cascading interactions using high-resolution camera trapping and remote sensing data in the best-preserved European lowland forest , Białowieża Forest , Poland . We showed that the variation in spatial distribution of an entire community of large herbivores is explained by species-specific responses to both environmental bottom-up and biotic top-down factors in combination with human-induced ( cascading ) effects . We decomposed the spatial variation in herbivore community structure and identified functionally distinct landscape-scale herbivory regimes ( ‘herbiscapes’ ) , which are predicted to occur in a variety of ecosystems and could be an important mechanism creating spatial variation in herbivory maintaining vegetation heterogeneity . Spatial patterns in species distribution , abundance and community composition are manifestations of underlying ecological mechanisms operating at a range of spatial scales ( Levin , 1992 ) . These patterns emerge from dynamic interactions between environmental bottom-up , biotic top-down and biotic parallel factors in combination with stochastic effects . Although the role of bottom-up factors in shaping species distributions has been intensively studied in recent decades ( Elith and Leathwick , 2009; Guisan and Thuiller , 2005 ) , we still know little about the importance of the biotic factors underlying most spatial patterns . Especially , the role of species interactions , within and across trophic levels , including those involving humans , remain largely unexplored ( Darimont et al . , 2015; Schmitz et al . , 2017; Wiens , 2011; Worm and Paine , 2016 ) . It has recently been proposed that community ecology should be ‘rediscovered’ as an integrative study of species interactions and spatial distributions ( Schmitz et al . , 2017 ) , while accounting for direct and indirect anthropogenic effects on species distributions and behavior ( Berger , 2007; Worm and Paine , 2016 ) . Large mammalian herbivores influence terrestrial ecosystem structure and functioning ( Gordon et al . , 2004; Hobbs , 1996; Schmitz , 2008 ) via their direct effects on vegetation structure ( Charles-Dominique et al . , 2016; Churski et al . , 2017; Didion et al . , 2009; Hempson et al . , 2015; Kuijper et al . , 2010a ) and indirect effects on nutrient cycling ( Murray et al . , 2013 ) . In this way , herbivory influences vegetation at large spatial scales , from the local landscape up to the biome level ( Moncrieff et al . , 2016; Woodward et al . , 2004 ) , and can lead to herbivory-mediated cascading effects on other trophic levels ( Gordon et al . , 2004; Palmer et al . , 2015; Schmitz , 2008 ) . There is often strong spatial variation in herbivory impact , resulting from the space use of different functional groups of herbivores . This spatial variation is driven primarily by the interactive effects of abiotic factors , disturbances , forage quality and quantity in combination with life-history traits , such as herbivore body mass ( Anderson et al . , 2016; Cromsigt et al . , 2009; Hempson et al . , 2015; Hopcraft et al . , 2010; Ogutu et al . , 2010 ) . However , the actual distribution of many herbivores often differs from the expected distribution derived purely from interactions with bottom-up factors . This discrepancy results from herbivores also responding to landscape gradients induced by biotic top-down interactions ( Anderson et al . , 2010; Hopcraft et al . , 2010; Kauffman et al . , 2007 ) . In effect , this landscape of interacting ecological gradients , both bottom-up and top-down , creates spatial heterogeneity in the availability and suitability of habitats for different large herbivore species within a community ( Cromsigt et al . , 2009; Fryxell , 1991; Hopcraft et al . , 2010 ) . Thus , to assess the ecosystem-level impact of large herbivore communities requires full understanding of the factors driving spatial heterogeneity in their community structure across a landscape ( Gordon et al . , 2004; Weisberg and Bugmann , 2003 ) . Recently , there has been much attention given to the role of large carnivores in structuring ecosystems via their effects on herbivore communities ( Estes et al . , 2011; Ripple et al . , 2014; Terborgh et al . , 2006 ) . In addition to their density-mediated effects ( i . e . impact on prey population size ) , behaviorally mediated effects ( i . e . impact on prey behavior ) on prey species are a crucial mechanism explaining the trophic cascades driven by large carnivores ( sensu Ripple et al . , 2016 ) . Prey species react to the presence of large carnivores by adjusting their spatio-temporal patterns of landscape use ( Creel et al . , 2005; Kohl et al . , 2018; Laundré et al . , 2001; Valeix et al . , 2009 ) . These spatial interactions between trophic levels are usually context-dependent and are shaped by the biophysical characteristics of a landscape ( Kauffman et al . , 2007; Schmitz et al . , 2017; Valeix et al . , 2009 ) . Many studies have addressed how carnivores affect their prey species , but these have generally used a single carnivore - single prey species approach , whereas many ecosystems host multiple carnivore and multiple prey species . In such systems , different carnivore species can create contrasting risk effects ( Creel et al . , 2017; Preisser et al . , 2007; Thaker et al . , 2011 ) . Moreover , in multi-species communities , some prey species perceive more risk than others from a carnivore species ( Anderson et al . , 2016; Laundré et al . , 2001; Valeix et al . , 2009 ) . This suggests that spatial distributions of predation-sensitive prey species may be mainly driven by carnivore top-down effects , whereas distributions of predation-insensitive prey or non-target species by gradients in resources availability ( Hopcraft et al . , 2010 ) . When a community of prey species consists of ecologically or functionally similar species , changes in the abundance and distribution of one species may be buffered by another species ( Ford et al . , 2015; Rosenfeld , 2002 ) . These so-called redundancy effects , can prevent apex predators from creating trophic cascading effects when taking the response of the entire herbivore community into account , despite them significantly impacting one or more prey species ( Ford et al . , 2015; Liu et al . , 2016 ) . There is a growing awareness that including humans in community studies is critical for improving our understanding of ecosystem functioning ( Darimont et al . , 2015; Worm and Paine , 2016 ) and for predicting species distributions in increasingly anthropogenic environments . Due to the recovery of some large carnivore populations and expansion of human populations , carnivores are increasingly sharing landscapes with humans world-wide ( Carter and Linnell , 2016; Chapron et al . , 2014 ) . Humans are also increasingly being considered a coherent part of complex trophic interaction chains ( Darimont et al . , 2015; Strong and Frank , 2010; Kuijper et al . , 2016 ) . The resulting complex , cascading interactions urgently need to be considered when studying the spatial distributions of herbivores , their effects within the landscape and the functional role large carnivores can play in landscapes that are becoming increasingly anthropogenic ( Kuijper et al . , 2016 ) . In this study , we investigated how the interactive effects of bottom-up and natural top-down factors ( two large carnivore species and humans ) , determine the landscape distribution and community composition of five native ungulate species in Białowieża Forest ( BF , Poland; Figure 1 ) . BF is regarded to be one of the best preserved temperate European lowland forest systems and is inhabited by a natural community of large mammals ( Jędrzejewska and Jędrzejewski , 1998 ) . In addition , BF is also embedded within an anthropogenic landscape typical for many terrestrial systems . We hypothesized that spatial variation in the composition of the large herbivore community is explained by the interactive effects of species-specific responses to major environmental and risk gradients operating at the landscape level . We aimed to answer the following questions: 1 ) Do large carnivores have species-specific effects on the distributions of ungulates in our multiple predator-prey system ? , 2 ) How does human activity mediate predator-prey interactions at the landscape scale ? , 3 ) Does this lead to ecologically distinct herbivory regimes ( sensu Hempson et al . , 2015 ) with differential vegetation impact at the landscape scale ? Using detailed data on species distributions ( 894 camera trap locations ) , landscape structure ( high-resolution GIS and remote sensing data ) and detailed woody vegetation surveys ( 385 study plots ) along with a novel spatially-explicit hierarchical modelling approach we decomposed the spatial variation in herbivore community structure into ecologically distinct landscape-scale herbivory regimes . With data from a complex , multi-species and human-influenced system , we aimed to exemplify the functional role that large carnivores and their herbivorous prey can play in increasingly human-affected ecosystems . The study was carried out in Białowieża Forest ( BF ) in Poland ( c . 580 km2; Figure 1 ) . This harbors a natural assemblage of central European ungulate species , with red deer ( Cervus elaphus ) being the most abundant ( 6 . 0 individuals/km2 ) , followed by wild boar ( Sus scrofa; 5 . 4/km2 ) and roe deer ( Capreolus capreolus; 2 . 0/km2 ) ; and European bison ( Bison bonasus; 0 . 5/km2 ) and moose as the rarest ungulates ( Alces alces; 0 . 08/km2 ) ( Borowik et al . , 2016 ) . Two large carnivores occur in BF: the Eurasian lynx ( Lynx lynx; c . 15 individuals ) and wolf ( Canis lupus; 4 packs of 7–12 individuals ) ( Jędrzejewski et al . , 2002; Schmidt and Kuijper , 2015 ) . Part of BF , the core of Białowieża National Park ( BNP; c . 47 km2 ) , has been strictly protected since 1921 . Since this time , human activities such as hunting and forestry have been banned . In 1996 BNP was enlarged to c . 100 km2 . Outside the national park , the forest is managed by the State Forest Holding , hence timber production takes place and ungulate hunting is allowed , but here also exists a network of nature reserves ( c . 130 km2 , see Figure 1 ) . For a more detailed description of this area see Faliński and Falińska ( 1986 ) and Jędrzejewska and Jędrzejewski ( 1998 ) . From a landscape perspective , it is important that the boundaries of BF ecosystem are well defined . From the west , BF is surrounded by agricultural fields , and from the north and south by a mosaic of agricultural and fragmented forest landscapes . In the east , a tall wire-fence along the border with Belarus ( built in 1981 ) prevents movements by ungulates . These conditions create an opportunity to study the spatial distribution of a whole community of ungulates in a spatially restricted , complex ecosystem with natural predators present and varying levels of human management and impact . In contrast to telemetry , which allows the study of both predator and prey movement patterns by placing sensors on individual animals , we used point-based camera trapping to record and reconstruct the distributional patterns of large carnivores and ungulates . We sampled the continuous BF landscape with camera traps placed randomly with respect to species movements and hypothetical mechanisms driving their distributions ( habitat structure , humans , predator-prey interactions ) . We used an intensive , large-scale and high-resolution camera trap network covering the entire study area to collect detailed , spatially-explicit information on species distributions . We argue that camera trapping is the most objective and efficient method for collecting this type of spatially-explicit community data . This kind of study would be inherently impossible to do with telemetry: to record the spatial patterns we were looking for , the entire populations of each species had to be observed simultaneously . Moreover , because of logistical and financial limitations and ethical issues ( related to live-trapping of protected species in Europe ) it would be practically impossible to obtain sufficient data for all the studied species of large herbivores and carnivores using telemetry . We used digital trail cameras ( Ecotone SGN-5210A ) triggered by passive infrared sensors with a detection angle of c . 35° and range of c . 20 m . After detection , with a time lag of c . 1 s , a photograph was taken and the camera recorded a 60 s video . When an animal stayed , this procedure was repeated without trigger delay . During low-light conditions , cameras switched to a stealth infrared mode . Cameras were attached to a tree at a height of c . 1 m at locations with a clear view of at least 20 m . Whenever possible , we randomly chose an acceptable place to mount a camera as close as possible to the locations given by coordinates pre-computed prior to the field work . We quantified the spatial distribution of the ungulate community by using a spatially extensive , high-resolution network of camera traps . The data were collected over 2 years ( May 2012 - May 2014 ) of intensive camera trapping in all seasons , except for days with the strongest winter conditions ( snowing heavily or temperatures below −20°C ) . We conducted 34 trapping sessions , each lasting 14 days on average . During each session between 30 to 40 camera traps were pseudo-randomly deployed in different parts of the forest and within a minimum distance of 100 m from the nearest roads ( both paved and unpaved ) , large clearings and settlements . Additionally , we kept a minimum distance of 100 m between all camera locations . The coordinates for all locations were pre-computed prior to the field work in QGIS software ( QGIS Development Team , 2017 ) . In total , we collected data at >1 k sites . However , because of logistical errors , camera failures , stealing of equipment and heavy snowing in winter ( leading to blocked view and uninterpretable gaps in data ) we had to exclude >100 sites ( ~10% ) from further analysis . Finally , we used data collected at 894 sites , covering the whole BF landscape ( Figure 1 ) . Large carnivores generally tend to have a higher detection probability at forest roads and trails ( Cusack et al . , 2015 ) because they often prefer to move along linear landscape structures ( Zimmermann et al . , 2014 ) and/or to increase the probability of encountering prey ( Whittington et al . , 2011 ) . This specific space use resulted in a very low trapping rates of both carnivores ( wolf and lynx ) during the ungulate survey ( Appendix 1—table 1 ) , when camera traps were placed randomly in the forest . Hence , to better quantify the space use of the two large carnivores we ran an additional camera trapping survey in September-October 2015 . We deployed 73 camera traps on the sides of forest roads across the whole landscape for one month ( Figure 1 ) . The core areas of wolf pack territories are related to the locations of breeding dens ( Jędrzejewski et al . , 2001 ) . During the reproductive season ( spring-summer ) the spatial distribution of a wolf pack is restricted to their core area , whereas outside this period they regularly return to it ( Jędrzejewski et al . , 2001 ) . In August-September , pups begin to travel with other pack members and move more widely through their territory , returning to the core of their territory on a regular basis ( Jędrzejewski et al . , 2001 ) . Lynx have a similar , typical pattern of movements whereby females restrict their movements in May-July , while tending their kittens ( Schmidt , 1998 ) . Therefore , late summer-autumn is the best period to quantify the space use of both carnivores at the landscape scale . The ungulate and carnivore surveys were conducted during different time periods , with different sampling intensities ( May 2012 to May 2014 versus September to October 2015 , respectively ) and using different camera placement strategies . To check if our results were not spatio-temporally confounded because of these differences , we ran the same model for all ungulate species using only a subset of the camera trap data covering a 3-month period ( August - October ) matching the period of the carnivore survey as closely as possible ( the models did not converge for data from only one or two months ) . The obtained results were similar to those based on the full dataset ( Appendix 1: Figs . S36-S37 ) ; we thus chose to use the full dataset , which provided a larger sample size and better spatial coverage of the study area , both of which are needed for making robust inferences using complex hierarchical spatial models such as ours ( see Statistical model section ) . Moreover , between-season variation in the trapping rate of ungulates was directly accounted for in the model . Lastly , all the studied ungulate species are non-migratory; in other words , there is no large ( landscape ) -scale seasonal movement of ungulates in BF ( Jędrzejewska and Jędrzejewski , 1998; Kamler et al . , 2008; Podgórski et al . , 2013 ) . The winter distribution of European bison is to a large extent driven by the location of supplementary feeding sites . This results in concentration of bison at these locations during the winter season . After downloading camera trap data , both ungulates and wolf datasets were organized and classified using TRAPPER software ( Bubnicki et al . , 2016 ) . Species , sex , age and group size were determined for every recorded image or video containing an observation of focal species . We defined the independence interval between successive captures ( i . e . event; see Meek et al . , 2014 ) as five minutes . Let us consider a landscape divided into j=1 , 2 , . . . , G grid cells of equal sizes and i=1 , 2 , . . . , S sampling locations monitored with camera traps for di days each , where each grid cell can contain zero , one or multiple camera trap locations . This results in a multi-scale design where smaller subunits ( camera trap sites ) are nested within larger units ( grid cells ) ( Kery and Royle , 2016 ) . Counts , yi , are Poisson random variables given by ( 1 ) yi\|Nj[i]∼Poisson ( Nj[i]γidi ) where the intensity parameter is a product of Nji , the number of individuals using a landscape grid cell j during a study period , γi , the expected detection ( trapping ) rate per sampling occasion ( here 1 day ) and di , the number of sampling occasions ( days ) during a survey at a camera trap location i . In the state-space formulation of our hierarchical model Nj is the Negative Binomial distributed latent variable ( 2 ) Nj∼NegBinλj , ϕwith the parameter λj being the expected number of individuals using a landscape grid cell j and ϕ the dispersion parameter . Both parameters , λj and γi can depend on covariates describing for example environmental gradients ( at different scales ) , biotic interactions and seasonal differences in species activity level . This variation can be modelled using standard generalized linear regression techniques using log-link functions ( 3 ) logγi=xγi'βγ+ϵi ( 4 ) logλj=xλj'βλwhere xγ⋅' and xλ⋅' are transposed rows of design matrices Xγ and Xλ , respectively , β⋅ are vectors of linear predictor coefficients and ϵi are identically and independently distributed iid camera trap measurement errors . It is necessary to note that , while the linear predictor of λj explains the variation in data arising from the ecological processes , the linear predictor of γi ( detection/trapping rate ) deals with both the ecological ( e . g . habitat selection , movement , seasonal activity levels ) and observational processes ( e . g . detection issues , camera failures ) . In the latter case , informative and ecologically meaningful covariates are needed to disentangle these otherwise confounded sources of variation . In order to directly account for the potential spatial dependence between landscape grid cells ( i . e . to capture all the spatial variation not explained by the included covariates ) we introduced spatial random effects ( “spatial residuals” ) into the linear predictor of λj: ( 5 ) log ( λj ) =xλj′βλ+ωjwhere ω=ω1 , . . . , ωG is the realization of a Gaussian spatial process on a discrete ( gridded ) spatial domain . Specifically , we implemented a Restricted Spatial Regression model ( RSR ) ( Hughes and Haran , 2013; Johnson et al . , 2013 ) , which is the restricted version of the intrinsic Conditional Autoregressive ( iCAR ) model . The RSR model was developed to solve the issue of confounding between the spatial process and the fixed-effects covariates in spatial regression models . The byproduct of this solution is its computational efficiency , as the RSR model is a reduced dimension model . The model as described above was fitted to our camera trapping data of the wolf and lynx and each of the five ungulate species . To model the intensity of landscape use ( λj in Equation 5 ) , we overlaid a grid of 2303 cells 500 m per side ( 25 ha each ) over the study area . Based on this grid we compiled a set of spatial ( raster ) covariates describing the ecologically relevant ( sensu Elith and Leathwick , 2009 ) environmental and human-induced gradients ( Figure 2 ) . For GIS data processing , we used QGIS ( QGIS Development QGIS Development Team , 2017 ) and GRASS GIS ( Neteler et al . , 2012 ) open source software . We standardized ( scaled ) all covariates , by subtracting the mean and dividing the result by the SD of the original variables . Finally , we ensured that the Pearson correlation coefficient for all pairs of included covariates was lower than 0 . 7 ( Appendix 1—figure 1 ) . For more details about all spatial ( raster ) covariates and the processing of GIS and remote sensing data see Appendix 1 . Based on existing knowledge ( Kuijper et al . , 2015; Schmidt et al . , 2009; Theuerkauf et al . , 2003 ) both carnivore species in the Białowieża forest utilize the entire landscape , although with clear spatial patterns in the intensity of use . Previous work in this study area has shown that this is primarily determined by human related factors ( Theuerkauf et al . , 2003 ) . In BF tourist traffic concentrates mainly within the central parts of the forest where roads ( open for the public and cars ) connect the three major settlements in the area , that is Hajnówka , Białowieża and Narewka ( see Figure 1 and Appendix 1—figure 3 ) . For the reasons above , the following raster layers were chosen as landscape covariates likely to influence carnivore space-use: distance to major settlements , distance to touristic trails , density of touristic infrastructure ( POIs ) , density of protected areas ( BNP and nature reserves ) and elevation ( Figure 2 ) . Elevation was included as in this flat landscape the lowest , often swampy areas are the least accessible for humans and could therefore be preferred by the wolf and lynx . We interpreted the parameter λ as large carnivore space use intensity and included rasters with predicted values of λj for both species as covariates in all models for ungulates . We assumed that from a prey perspective , λj is proportional to the predator encounter rate , hence it quantifies potential risk as perceived by ungulates . To explain the variation in the landscape-scale distribution of ungulates , we considered three environmental gradients primarily related to major biophysical properties of the forest environment that are known to affect space use of ungulates at multiple scales: percentage of landscape openness , tree stand canopy height and percentage share of coniferous species ( Churski et al . , 2017; Jedrzejewska et al . , 1994; Kuijper et al . , 2009; Kuijper et al . , 2010a; Figure 2 ) . For landscape openness and percentage share of coniferous species , we additionally included their quadratic effects , allowing for the existence of an optimum value for each variable ( e . g . species preference for a mixed forest or for intermediate levels of canopy closure ) . The other covariates included were potential predation risk variables ( space use of large carnivores ) and two human-related landscape gradients , namely distance to all settlements and density of protected areas ( BNP and nature reserves; Figure 2 ) . Additionally , we computed the same set of environmental covariates but with a higher resolution ( 100 m ) and used them to model variation in detection rate ( γi in Equation 4 ) at camera trap sites . Here , we assumed that a part of this variation comes from a resource selection process operating at a scale smaller than the landscape unit we defined ( see e . g . Johnson , 1980 ) , influencing at-site detection rates and observed counts in the end . Another source of variation is species movement behavior and activity level , which can both change between seasons . To control for these temporal effects , we considered a quadratic function of temperature and snow cover as covariates for detection rate . The models were implemented within a Bayesian framework in Python using PyMC 2 . 3 . 6 software ( Patil et al . , 2010 ) . We used Markov chain Monte Carlo ( MCMC ) for inference and sampled from the posterior distributions with Metropolis-Hastings and Adaptive Metropolis step methods , both available in PyMC . To speed up the model and improve the MCMC convergence , we marginalized out the latent variable N and implemented the integrated likelihood function ( Guillera-Arroita et al . , 2012; Royle , 2004 ) . To make the integration over N values finite , we assumed 100 as the maximal possible number of individuals using a single grid cell j ( 25 ha ) . We defined the priors for the linear predictor coefficients β⋅ as diffuse normal priors N0 , 10-3 . The priors for the measurement errors ϵi were given by N0 , 1/σ2 with the hyper-parameter σ∼U0 , 100 ( Gelman and Hill , 2006 ) . We followed Royle et al . ( 2007 ) and we chose the gamma distributed prior G0 . 1 , 0 . 1 for the precision parameter of the RSR model . The reason for not choosing a vague prior for this parameter ( as e . g . in Johnson et al . , 2013 ) was that we expected the spatial covariates included in the models would not account for all the spatial dependence alone . Part of this ( unexplained ) variation is likely related to species movement behavior occurring at multiple spatial scales . It is also commonly known that the variance components are poorly identified in these types of models ( Royle et al . , 2007 ) . To obtain posterior distributions of parameters , we ran a MCMC sampler with three chains for 500 , 000 iterations each ( removing the first 400 . 000 as a burn-in phase of the sampling process ) and with the thinning parameter set to 20 to avoid autocorrelation between samples . The convergence was assessed through visual inspection of MCMC trace plots and Gelman–Rubin diagnostics provided by the PyMC software . We evaluated the fit of the models through visual inspection of standard model diagnostics plots ( see Appendix 1—figures 28–35 ) . We used posterior predictive distribution and Bayesian ‘p-value’ to assess the goodness of fit of each model ( Kery and Royle , 2016 ) . The source code of our models is available at https://github . com/mripasteam/herbiscapes/ ( Bubnicki et al . , 2019; copy archived at https://github . com/elifesciences-publications/herbiscapes ) . We explored the distribution of ungulates across the landscape , as predicted by species-specific models , in the context of the community . First , the spatial overlap between species was evaluated by means of pairwise Pearson correlations of their relative density surfaces . Next , we converted ungulate relative densities to herbivore biomass using the following average weights per species: red deer female 90 kg , red deer male 150 kg , roe deer 20 kg , wild boar 80 kg , moose 200 kg and European bison 400 kg ( Borowik et al . , 2016; Jedrzejewska et al . , 1994 ) . Total biomass was estimated as the sum of each species’ biomass for each landscape cell ( 25 ha pixel ) in our prediction grid . We further calculated and mapped the index of the functional diversity of the entire ungulate community ( FDis; Laliberté and Legendre , 2010 ) using the R statistical software ( Core Team , 2019 ) and the R package FD v . 1 . 0–12 . The FDis was calculated based on the predicted relative densities of all ungulates and the following species-specific traits: body mass , diet type and gut type . The FDis is the mean distance in multidimensional trait space of individual species to the centroid of all species . It can account for species abundances by shifting the position of the centroid toward the more abundant species and weighting distances of individual species by their relative abundances ( Laliberté and Legendre , 2010 ) . The input data for a FDis calculation are 1 ) a matrix with species traits and 2 ) a matrix with relative densities that describe how much weight to assign to each individual observation . We compiled a six row ( species ) by three column ( traits ) matrix with one quantitative and two qualitative traits , namely body mass , gut type and diet type ( Appendix 1—table 2 ) . By means of hierarchical cluster analyses ( Lê et al . , 2008 ) , we grouped all landscape grid cells ( 25 ha each ) with similar values of FDis , and total and species-specific biomass into clusters representing ecologically distinct landscape-scale herbivory regimes , or ‘herbiscapes’ . Specifically , following the approach of Hempson et al . ( 2015 ) , we used the R package FactoMineR v . 1 . 39 ( Lê et al . , 2008 ) and its HCPC ( hierarchical clustering on principle components ) function . The HCPC requires that PCA ( principal component analysis ) is performed on variables prior to clustering , which limits the impact of covariance amongst variables on the subsequent clustering algorithm . To build the cluster tree , we used HCPC default values for the metric ( Euclidean distance ) and method ( Ward’s ) parameters . Similarly to Hempson et al . ( 2015 ) , the number of clusters was determined by assessing the inertia ( i . e . change in within cluster homogeneity ) gained by cutting the tree at different levels and the ecological interpretability of the resulting clusters . Eventually , we split the cluster tree into five independent clusters ( Appendix 1—figure 2 ) . The hierarchical clustering analysis allowed us to learn how the combined effect of predation risk and resource quality translates into the composition and abundance of the ungulate community and , in consequence , into the diversification of herbivory pressure on the ecosystem . Using an independent vegetation dataset from a large-scale inventory of tree regeneration ( part of the LIFE+ ForBioSensing project , contract number LIFE13 ENV/PL/000048 ) , we tested if the predicted variation in the landscape-scale distribution of large herbivores , synthesized into ecologically distinct herbivory regimes that is herbiscapes , affects tree browsing intensity and regenerating tree species composition . We used data collected in 2017 at 385 plots spread randomly across the entire BF . Each plot contained two concentric sub-plots: 1 ) with a radius of 1 . 3 m ( area of 5 m2 ) at which all trees with height <30 cm excluding seedlings were recorded , and 2 ) with a radius of 2 . 52 m ( area of 20 m2 ) at which all trees with height ≥30 cm and diameter at breast height <2 cm were recorded . Additionally , each individual tree was checked for any sign of fresh or 1-year-old browsing of its main shoot . The <30 cm tree sapling community is structured mainly by bottom-up factors ( and/or forest management practices ) and only minimally influenced by ungulate herbivory ( Kuijper et al . , 2010a ) , whereas the ≥30 cm tree sapling community is within the foraging height class preferred by ungulate herbivores ( Kuijper et al . , 2013 ) and is therefore largely structured by ungulate top-down factors ( Kuijper et al . , 2010a; Kuijper et al . , 2010b ) . Based on this data , for each plot we calculated the cumulative browsing intensity index , expressed as the proportion of browsed individual trees out of all tree saplings ≥ 30 cm , and the difference between the two tree height classes in the proportional shares of Carpinus betulus ( Carpinus ) and Acer platanoides ( Acer ) . The latter two parameters represent a measure of recruitment from the sapling-bank ( <30 cm ) to the taller size-class ( ≥30 cm ) . This process is to a great extent driven by large herbivores in the studied system ( see Kuijper et al . , 2010a and Churski et al . , 2017 ) . We specifically focused on the response of two contrasting species , Carpinus and Acer . While both species are palatable and strongly selected by the ungulate community ( see Churski et al . , 2017 ) , Carpinus is highly browsing-tolerant ( a typical ‘brown-world’ species sensu Churski et al . , 2017 ) and Acer is highly-sensitive to ungulate browsing ( a typical ‘green-world’ species ) . Long-term exclosure studies have also shown that Carpinus typically increases while Acer decreases in dominance in response to ungulate herbivory ( Kuijper et al . , 2010a ) . As both species are very common throughout the forest , we see them as suitable indicator species for the impact of ungulate herbivory on tree species composition . We explained the variation in the calculated parameters by fitting simple linear models with two interacting factors: herbiscape and reserves . The latter was added to account for potential differences in forest structure ( see Jedrzejewska et al . , 1994 ) and ungulate behaviour ( see Kamler et al . , 2008 ) between protected and unprotected areas in BF . The main factor associated with the space use of both large carnivore species was human activity , as indicated by the positive effect of distance to major settlements on their spatial distributions ( Figure 3 ) . For the wolf , the density of protected areas was another important variable related to its landscape use . Wolves more often used large nature reserves ( including BNP ) than parts of the landscape dominated by managed forest . There was no statistically important effect of elevation , distance to touristic trails and density of touristic infrastructure on landscape distribution of neither the wolf nor lynx . However , wolves tended to use lower areas more intensively . For the lynx there was a clear tendency to use parts of the landscape further away from major human settlements; however , this effect was less evident than for wolf ( the credible intervals overlapped at zero , Figure 3D ) . The density of protected areas had no effect on lynx distribution . To test the quality of our predictions , we compared them with existing radio-tracking data from collared wolves collected over 20 years ago ( 1994–1999 ) in BF . This showed that our model fitted to the camera trapping dataset ( 471 wolf detections ) not only conforms to the general pattern of the wolves’ space use determined by telemetry 20 years ago , but also reflects the wolves’ response to ongoing environmental changes in the study area . See the Appendix 1 and Figure 4 for more details . In contrast to the other ungulates , the red deer , the main prey of the wolf ( Jędrzejewski et al . , 2002 ) , was the only species whose landscape use was associated with that of wolves ( Figures 4 and 5 ) . The predicted relative densities of both red deer females and males were lowest in parts of the landscape intensively used by wolves , and this effect was more pronounced for females . Lynx distribution was not related to the landscape use of any ungulate species . Red deer females were also positively associated with protected areas ( BNP and nature reserves ) and mixed deciduous forest with an intermediate level of landscape openness ( Figure 4 ) . Red deer males showed similar tendencies , but these factors were not statistically important predictors of their landscape distribution . Instead , red deer males showed a negative association with distance to human settlements , which could have resulted from their more intensive use of open meadows surrounding settlements , especially during the rutting period . Forage habitat availability was the main factor associated with wild boar distribution , as indicated by its clear association with closed-canopy and deciduous species dominated forest stands ( as indicated by the strong negative effect of coniferous species dominated forest stands; Figure 4 ) . The presence of deciduous forest was also the main factor positively associated with the distribution of bison , followed by proximity to human settlements ( as indicated by the negative association with distance to human settlements ) . The latter is likely related to the presence of meadows surrounding settlements , which provide optimal foraging habitats for bison ( Bocherens et al . , 2015; Cromsigt et al . , 2012 ) . The moose was the only species whose landscape-scale distribution was positively associated with low elevation areas ( river valleys and wetlands ) . Roe deer were positively associated with intermediate levels of landscape openness . However , no other predictors that we used were related to the distribution of roe deer , which may be due to the low density of this species in our study system . All ungulate species except red deer males showed some level of remaining spatial auto-correlation not explained by the raster covariates included in the models ( Figure 6 ) . The spatial random effects were particularly large for the bison , whose spatial distribution in BF is strongly influenced by supplementary winter feeding and use of open areas outside the forest throughout the year ( Kowalczyk et al . , 2011 ) . An interesting hot-spot of unexplained variation in distribution of bison and red deer females was the south of the national park , in an area without hunting and rich old-growth deciduous stands and with known higher densities of deer ( Jędrzejewska et al . , 1997 ) . At the scale of single camera trap sites ( 1 ha ) , increased canopy openness led to higher detection rates of red deer and moose ( Figure 7 ) . This could be explained by cameras having better detection in open areas ( Marcus Rowcliffe et al . , 2011 ) and/or a preference of these species to forage in canopy gaps ( Churski et al . , 2017; Kuijper et al . , 2009 ) . Red deer females were associated with even larger canopy gaps , whereas for red deer males and moose there was an optimum value of canopy openness , above which the detection rate started to decrease . Roe deer followed a similar pattern as red deer but there was no statistical support for this result . At-site detection rate of wild boar was highest in deciduous forest patches with relatively closed canopies ( in line with Kuijper et al . , 2009 ) . However , wild boar , as well as red deer , roe deer and bison were also relatively frequently detected at forest patches dominated by coniferous tree species , indicating a context-dependence in the selection of small habitat patches . For example , as forage availability changes between seasons , deciduous patches may be preferred in the green season while coniferous patches in winter . Temperature affected detection rate non-linearly by influencing the activity level of red deer males , with the optimum at the yearly average temperature . In the case of the wild boar the highest detection rates were found at high temperatures , coinciding with their reproductive period in summer . And in case of the moose , the highest activity level was found at moderately high temperatures . The bison was the only species whose detection rate was strongly affected by snow cover . Winter supplementary feeding causes bison to aggregate near feeding stations or outside BF ( Kowalczyk et al . , 2011 ) , hence dramatically decreasing their detection rate in other parts of the forest . All ungulate species except moose showed some level of pairwise positive spatial associations as indicated by the Pearson correlation of their relative density surfaces predicted by the models ( Figure 8 ) . However , the strength of these associations was relatively low , indicating substantial spatial variation in the structure of the whole community . Unsurprisingly , the strongest overlap in space was between red deer females and males . However , the estimated correlation ( 0 . 71 ) was far from a ‘perfect’ overlap , indicating red deer are sexually segregated in space in BF ( see Kamler et al . , 2008 ) . The relatively high spatial overlap between red deer males and roe deer ( 0 . 58 ) likely resulted from their utilization of similar parts of the landscape , often close to the forest edge and large clearings with human settlements . The moose was a clear exception showing a negative spatial association with all other ungulate species . This indicates that the moose has a specific ( spatial ) niche in our system , strongly associated with low lying areas like river valleys and wetlands ( see Figures 2 , 4 and 5 ) . Interestingly , when comparing the surfaces of spatial random effects , there was a relatively strong positive spatial association ( 0 . 59 ) between bison and wild boar ( Figure 6 ) . A possible explanation for this pattern may be that wild boar are attracted to supplementary food at bison feeding stations , where next to hay and silage , beetroots are provided for bison ( Kowalczyk et al . , 2011 ) . When mapped , both the estimated total biomass of the ungulate community and the functional diversity index ( FDis ) showed distinct patterns across the studied landscape ( Figure 9 ) . Interestingly , the lowest values of both parameters were associated with coniferous-dominated tree stands at higher elevations ( Figures 2 and 5 ) belonging to the least productive parts of this landscape ( Faliński and Falińska , 1986; Kwiatkowski , 1994 ) . By means of hierarchical cluster analyses , we further grouped all landscape grid cells ( 25 ha each ) with similar values of FDis , and the total and species specific biomasses . We identified five clusters , characterized by different sets of risk-related and environmental factors and composed of different sets of ungulate species ( Figure 10 ) . The ‘red’ cluster ( id = 1 ) was characterized by high wolf and lynx use and low quality foraging habitat ( low elevation , mixed tree stands with large shares of coniferous tree species , Figure 10B ) . In terms of ungulate biomass , this cluster was dominated by moose , red deer and wild boar ( Figure 10C ) . However , red deer relative density was the lowest out of all five clusters . The total ungulate biomass in this cluster was low but the FDis value was relatively high ( Figure 10D ) . A large part of the ‘red’ cluster was spatially associated with the lowest elevated areas , that is marshlands and river valleys ( compare Figure 10A with Figure 2 ) . The ‘green’ cluster ( id = 3 ) was characterized by high predator presence and high-quality foraging habitat ( moderate elevation , mixed tree stands with large shares of deciduous tree species ) , which increased the numbers of seemingly risk-insensitive species like wild boar and bison . Red deer males , which had a less pronounced negative spatial association with wolves than females ( Figure 4 ) , also had a higher biomass in this cluster than in the ‘red’ one . Moreover , the ‘green’ cluster was characterized by low values of total biomass and high values of FDis . This cluster in the landscape , mainly covered remote areas of low human activity and high use by wolf , and occurred in protected areas ( BNP and nature reserves ) . The ‘blue’ cluster ( id = 2 ) was characterized by moderate use by predators and low-quality foraging habitat ( high elevation , coniferous dominated tree stands ) , and had the lowest values of both total biomass and FDis . However , the wolf used these areas less intensively than the ‘red’ and ‘green’ clusters , and the biomass of red deer was higher than in those high wolf use clusters . The ‘purple’ cluster ( id = 4 ) was characterized by low use by predators and high-quality foraging habitat ( high elevation , mixed tree stands with large share of deciduous tree species ) , and was dominated by red deer . Both sexes of red deer were at their most abundant in this cluster , which covered areas of high human activity and the lowest use by wolf . These seemingly ‘safe’ parts of the landscape were also characterised by high-quality habitat for red deer – a mosaic of deciduous and mixed tree stands at higher elevations and relatively close to forest edges and large clearings surrounding human settlements . The roe deer was also spatially associated with this cluster . The ‘orange’ cluster ( id = 5 ) was characterized by moderate use by predators and high-quality foraging habitat ( moderate elevation , deciduous dominated tree stands ) and had the highest values of total biomass and FDis . This cluster was dominated by the bison - the main grazer and largest species in our system . Also , the wild boar was at its highest relative density in this cluster . It is worth mentioning that a large part of this cluster was in the southern part of Białowieża National Park , comprising some of the best preserved parts of this forest . The highest proportions of browsed trees were found in herbiscapes 3 , 4 and 5 ( Figure 11c ) , which are characterized by high values of total ungulate biomass and/or high functional diversity index ( FDis ) . These three herbiscapes covered the more fertile parts of the landscape dominated by productive deciduous and mixed forests ( Figure 10B ) with abundant tree regeneration dominated by Acer platanoides and Carpinus betulus ( Acer and Carpinus hereinafter , Figure 11d ) . The variation in browsing intensity between the herbiscapes was reflected in the recruitment patterns towards taller tree sapling size classes ( >30 cm ) of Acer and Carpinus . In accordance with previous experimental exclosure studies in BF ( Churski et al . , 2017; Kuijper et al . , 2010a; Hedwall et al . , 2018 ) , the proportion of the palatable but browsing intolerant Acer in the community of regenerating trees decreased as herbivore pressure increased ( Figure 11a , d ) . In contrast , the palatable but highly browsing-tolerant Carpinus , showed the opposite pattern ( Figure 11b , d ) . These supposedly herbivore-driven shifts in the species composition of regenerating trees were most pronounced in nature reserves , whereas areas outside the reserves broadly showed the same patterns but less closely followed the observed patterns in cumulative browsing intensity . All ungulates in our study showed specific , non-uniform distributional patterns that were associated with species-specific combinations of bottom-up and/or top-down forces , including predation , human presence , availability of resources and the ( bio ) physical properties of the landscape ( Figures 4 and 5 ) . The substantial spatial variation in each of these landscape components , when combined , resulted in an aggregated , non-uniform pattern of landscape distribution of all ungulates , as predicted by Fryxell ( 1991 ) and Hopcraft et al . ( 2010 ) . The spatial distribution of each species ( Figure 5 ) followed a characteristic shape of a hollow or sigmoidal curve ( Figure 13 ) when plotted as a graph of ranked relative densities predicted for each 25 ha landscape pixel . This pattern of spatial variation in abundance of different populations has been shown to be universal over large spatial domains and for different taxa ( Brown et al . , 1995 ) . Our study contributes to this finding by showing that similar spatial patterns can be observed within a population and within a local landscape . The shape of these ranked-abundance curves is informative for the properties of a given continuous ( landscape ) surface ( Rocchini and Neteler , 2012 ) , for example the sigmoidal curve of wild boar indicates a highly heterogeneous distribution with many hot and cold spots ( high and low use ) , whereas the hollow curve of moose indicates a rather homogeneous distribution at low density with only some hot-spots . Interestingly , the degree of spatial aggregation of different ungulate species in our study system was not related to their body size . This contrasts with studies from African savannas where larger herbivore species were more evenly distributed over the landscape than the smaller species ( Cromsigt et al . , 2009; du and Owen-Smith , 1989 ) . The two largest herbivores in our study system , the European bison and moose , showed highly aggregated landscape distributions . The moose was strongly associated with lower elevation habitats ( i . e . river valleys and wetlands ) , whereas landscape use by bison was strongly related to locations of supplementary winter feeding and open areas inside the forest , which are distributed sparsely within the study area and often associated with anthropogenic activity . The distribution pattern of bison could also be the result of partial migrations outside the forest in winter ( Kowalczyk et al . , 2013; Kowalczyk et al . , 2011 ) . This mainly human-driven distribution of bison was also manifested in the largest values for unexplained spatial variation in the model output ( fitted spatial random effects , Figure 6 ) . The distributions of all species , when combined , revealed substantial spatial variation in the composition of the entire ungulate community ( Figure 10 ) . To our knowledge this is the first empirical study presenting a synthesized , high-resolution and spatially explicit approach ( sensu Royle et al . , 2007 ) combining bottom-up and top-down factors to explain the landscape-scale variation of an entire large herbivore community in a temperate forest ecosystem . Including both resource- and predator-related factors was critical for achieving this goal as they can operate simultaneously and interactively ( Anderson et al . , 2010; Fryxell , 1991; Hopcraft et al . , 2010 ) . Similar studies in African ecosystems have shown that the distribution and diversity of a community of large herbivores can be driven by bottom-up ( e . g . habitat heterogeneity; Cromsigt et al . , 2009 ) , top-down ( e . g . anthropogenic fire; Klop and Prins , 2008 ) or interactive effects of both factors ( e . g . distance to water and settlements; Ogutu et al . , 2010 ) . More recent studies , also from African ecosystems , have extended this approach by including predation-related factors , revealing the trade-offs native ungulates make to cope with changes in forage availability , human disturbance and predation risk ( Schuette et al . , 2016 ) and showing that a top predator can have species-specific spatial associations with herbivores ( Anderson et al . , 2010 ) . For example , the latter study showed that lions were positively associated with large-bodied migratory ungulates but negatively associated with smaller non-migrants . Most comparably to our study , Anderson et al . ( 2016 ) used an extensive network of camera traps and a spatially-explicit occupancy modeling framework to quantify the spatial distribution of African savanna herbivores . Interestingly , they quantified pairwise interactions between all modeled species demonstrating the emergence of strong positive spatial associations among a diverse group of savannah herbivores . This is in line with our results where all ungulates except moose showed positive spatial associations . However , in our study , these spatial associations were rather weak , indicating substantial spatial variation in the structure of the ungulate community . Our contribution to all the above studies is a high-resolution picture of the spatial structure of an entire community of large herbivores that incorporates both bottom-up and predation- and human-related top-down factors . We believe one of our major points of novelty is in providing information on these spatial interactions for a temperate ecosystem . The red deer was the only species whose landscape use was associated with that of large carnivores , with lower red deer presence in areas with higher wolf use . This is in line with other studies that have shown that predator top-down effects can work selectively on some members of large herbivore communities ( Sinclair et al . , 2003; Valeix et al . , 2009 ) and have a complex and context-specific nature especially in multiple-prey and multiple-predator systems ( Davies et al . , 2016; Moll et al . , 2016 ) . However , wolves , by seemingly affecting the spatial distribution of red deer ( see below ) , one of the dominant species in BF , re-structured the composition of the entire community of large herbivores ( Figure 14 ) and increased the degree of its spatial heterogeneity ( Figure 10 ) . The red deer is the main prey species for wolves in our study area and experiences the highest predation pressure by wolves , in contrast to the European bison , moose and roe deer , which comprise only a small proportion of their diet , and wild boar , which is a secondary prey species ( Jędrzejewska et al . , 1997; Jędrzejewski et al . , 2002 ) . Hence , this might explain why the the red deer was also the species that seemed to react most strongly to the space use of its main predator . The most intensively used part of a wolf territory ( on an annual basis ) is related to the location of their dens during the reproductive period , and is generally far from human settlements ( Jędrzejewski et al . , 2001; Kuijper et al . , 2015 ) . These high wolf use areas likely only have higher predation rates during the reproductive period ( Kuijper et al . , 2013 ) . Over the rest of the year , wolves move more widely across their territories ( Jędrzejewski et al . , 2001 ) and kills may be distributed more widely across annual wolf territories . Hence , the areas that seemed to be avoided by red deer are not necessarily areas with the highest predation risk on an annual basis . This finding adds to the growing recognition that prey species perceive risk based on various factors such as the space use of large carnivores and physical landscape and not necessarily by kill site distribution ( Kohl et al . , 2018; Gaynor et al . , 2019 ) . The distributional pattern of red deer was similar for both sexes ( one independent model was fitted for each ) , but females had a stronger negative association with wolf space use than males . This indicates that females are more sensitive to wolf presence than males and is consistent with the selective killing of this sex and juveniles by wolves in our study system ( Jędrzejewski et al . , 2000; Jędrzejewski et al . , 2002 ) . These apparent effects of wolf space use on red deer distribution could have resulted from both non-lethal ( behaviorally mediated ) as well as lethal ( density-mediated ) effects . With our data we were not able to distinguish between these two mechanisms , although previous studies have shown that non-lethal risk effects play an important role in affecting the responses of red deer at both fine- and large-scales in this system ( Kuijper et al . , 2015; Kuijper et al . , 2013; Kuijper et al . , 2014 ) . In contrast to the red deer , environmental bottom-up factors , particularly landscape topography and resource availability ( natural or supplemented by humans ) , had the strongest associations with the spatial distribution of the other , less predation-sensitive ungulate species ( Figures 4 and 5 ) . Although the wild boar is a secondary prey species for wolves ( Jędrzejewski et al . , 2002 ) , its high abundance during t study period means that predation by wolves contributed little to its annual mortality ( Jędrzejewska et al . , 1997 ) . This may explain why we did not observe a negative association between wild boar and wolf space use . Moreover , a previous study of ours found that the wild boar displayed behaviour suggesting it perceived no predation risk in response to the presence of fresh wolf or lynx scats ( Kuijper et al . , 2014; Wikenros et al . , 2015 ) . Our results seem contrary to those of Theuerkauf and Rouys ( 2008 ) , who carried out a similarly focused study in the same study area with use of pellet counts . They concluded that habitat alteration by forest exploitation and hunting by humans influenced the density distribution of ungulates , including red deer , more than predation risk by wolves . Despite our results indicating a spatial mismatch between the red deer and wolves’ landscape use , this is likely to be caused by a cascading effect of humans . Moreover , the patterns of ungulate space use shown by our analyses revealed an additional source of variation , which involves a species-specific response to the inter-relationships between human and predator space use . We were surprised to find that the Eurasian lynx – the other large carnivore in our study area – had no apparent effect . As ungulates constitute the bulk of the lynx’s diet , with the roe deer being the major prey ( 60% of the diet ) and the red deer being the alternative prey ( 22% ) ( Okarma et al . , 1997 ) , it is an important apex predator in this system . Lynx predation is a major mortality factor for both cervids , taking 21–36% of roe deer and 6–13% of red deer population numbers ( Okarma et al . , 1997; Jędrzejewski et al . , 1993; Jędrzejewska et al . , 1997; Jędrzejewska and Jędrzejewski , 2005 ) . Moreover , it was recently revealed that red deer clearly react with anti-predatory behavior to olfactory cues of the lynx in BF ( Wikenros et al . , 2015 ) . It is thus striking that these highly sensitive behavioral responses do not lead to changes in the spaces use of ungulates at the landscape scale . This may be a result of the combined effects of the low densities of both the roe deer and the lynx and the particular hunting mode of lynx , which , typically for felids , relies on fine-scale habitat characteristics that allow the predator to exploit prey independently of their spatial distribution ( Podgórski et al . , 2008; Schmidt , 2008 ) . Although there are many studies on the impact of large carnivores on the space use of their prey species in temperate systems , the majority of these have focused on single carnivore - single prey relationships ( see e . g . Creel et al . , 2005; Kauffman et al . , 2007; Lima and Dill , 1990; Mao et al . , 2005 ) . Our study is the first we are aware of to show how these carnivore top-down effects structure an entire ungulate community in a temperate landscape . This knowledge is relevant as most terrestrial ecosystems are characterized by a diversity of herbivorous prey species and the differential responses of functionally similar prey species to apex predators can to a large extent determine the potential for trophic cascading effects of large carnivores ( Ford and Goheen , 2015; Ford et al . , 2015; Rosenfeld , 2002 ) . Humans can drive complex interactions between species , particularly by affecting keystone species like large carnivores ( Worm and Paine , 2016 ) . We found that humans were the main factor associated with the spatial distributions of both the wolf and lynx , which had lower activities in parts of the landscape heavily used by humans ( in line with Theuerkauf et al . , 2003 ) . In this way , human presence can be beneficial for ungulate prey species as large carnivores generally avoid human presence and activity more strongly than their ungulate prey species ( Rogala et al . , 2011 ) , leading to so-called ‘human shields’ ( Berger , 2007 ) . The observed spatial ( re ) distribution of red deer in our system seems to result from human-induced shifts in space use of the wolf . These kinds of three-way trophic cascades ( following the definition of Ripple et al . , 2016 ) involving humans , large carnivores and ungulates have been found in different systems , mainly in landscapes with moderate human activity ( Berger , 2007; Hebblewhite et al . , 2005 ) and are likely much more pronounced in highly human-dominated landscapes ( Kuijper et al . , 2016 ) . We carried out our study in the forest considered to be the best preserved lowland forest ecosystem in Europe , which is replete with natural wildlife communities and ecological processes ( including trophic interactions ) still operating at the landscape scale . Thus , we believe it provides new knowledge on the fundamental structuring and functioning of European forest ecosystems that will help to predict the ecological effects of the ongoing large carnivore recolonisation of more human-altered habitats across Europe ( Chapron et al . , 2014 ) . How the consumptive off-take by large mammal herbivores varies in space at a continental-scale has recently been shown by Hempson et al . ( 2015 ) , who classified African herbivore communities into ecologically distinct herbivory regimes , or herbivomes . In our study , we down-scaled this approach to the landscape level , exploring the spatial variation and functional diversity of the herbivore community within a single herbivome , decomposing it into ecologically distinct landscape-scale herbivory regimes . We refer to these as herbiscapes . Additionally , we extended the herbivome concept by including the direct and indirect top-down effects of higher trophic levels: large carnivores ( keystone species ) and humans ( hyper-keystone species ) . Thus , similarly to a spatial ‘profile’ of each single species , the major dimensions of a herbiscape are oriented along both environmental ( landscape feature , resource availability ) and trophic interaction induced ( direct and perceived predation risk , human activity ) gradients operating over a landscape ( Figure 10 ) . As large herbivores influence many important processes of ecosystem functioning ( Hobbs , 1996 ) , the observed landscape-scale variation in the structure of the large herbivore community ( i . e . herbiscapes ) , when stable enough , could result in a differential impact of herbivores on ( woody ) vegetation . Our tree regeneration analysis revealed that the patterns of ungulate browsing intensity followed the variation in the modeled community-level distribution of ungulates in BF . The highest browsing intensities occurred in herbiscapes characterized by high ungulate biomass and/or functional diversity . Compositional changes in regenerating trees across herbiscapes were in accordance with our earlier experimental studies ( Churski et al . , 2017; Kuijper et al . , 2010a ) : with higher levels of browsing intensity the proportion of a common browsing-tolerant tree increased and the proportion of a browsing-intolerant tree decreased . These patterns strongly suggest that the impact of ungulate communities on tree regeneration varies greatly at the landscape-scale , even in relatively homogeneous forest landscapes like BF . Earlier experimental studies showed that the species composition of the small tree sapling community in BF is shaped by environmental bottom-up factors , whereas herbivores are the main factor limiting recruitment towards taller size classes ( above 50 cm; Kuijper et al . , 2010a ) . We therefore interpret the observed shift in dominance towards browsing-tolerant species in the taller height class ( >30 cm ) as the result of herbivore top-down effects , as indicated by the clear differences in browsing intensity ( Kuijper et al . , 2010a; Churski et al . , 2017 ) . Humans are an inherent part of these multi-trophic interactions as they affect both carnivore and species-specific herbivore space use , cascading down in a complex but predictable way to the lower trophic levels . Our study suggests , that the existence of more stable herbiscapes at larger spatial scales could create a mosaic of differential impact on woody plant communities and consequently spatial variation in expressed tolerance traits . Herbiscapes could therefore strongly contribute to the creation of a heterogeneous patchwork of green ( less browsing tolerant ) and brown ( more browsing tolerant ) worlds ( sensu Churski et al . , 2017 ) driven by variation in herbivory pressure at multiple spatial scales in temperate forest systems . Recent views on the biome concept have emphasized the role of biotic interactions ( e . g . herbivory ) in creating multiple biome stable states under similar climatic conditions ( Moncrieff et al . , 2016; Woodward et al . , 2004 ) . This approach has proven to be very useful in explaining the spatial distributions of different vegetation communities maintained by functionally distinct guilds of herbivores at continental-scales ( Charles-Dominique et al . , 2016; Hempson et al . , 2015 ) . On the basis of the high-resolution data we collected on ungulate species distribution , our study revealed the presence of a large continuous variation in herbivore community structure at a much finer , within-landscape scale . Although European assemblages of wild herbivores are not as species-rich as those found in African savannas , the species we studied are clearly functionally diversified ( see for example Hofmann , 1989 ) . These differences in resource use and foraging behavior among species within the community could differentially impact the vegetation if certain parts of the landscape are dominated by different functional types of herbivores for a long enough period . Studies from African savannas have indicated that the stability of these relationships is poorly understood , but is likely maintained by strong large-scale positive feedbacks between vegetation , abiotic resources and consumers ( Charles-Dominique et al . , 2016 ) . This long-term stability likely provides the basis for co-evolutionary dynamics between functional types of herbivores and plants in these systems . Whether such feedbacks could also be operating at smaller spatial scales ( i . e . in herbiscapes within a herbivome ) and play a role in creating small-scale variation in vegetation structure is an intriguing question . While we do not believe they could be stable on an evolutionary time-scale , there could be sufficient stability over several decades to significantly impact woody plant communities ( Kuijper et al . , 2010a ) . In our system , these feedbacks have been controlled by landscape scale anthropogenic factors that have been present in a similar spatial arrangement for decades or even centuries ( i . e . villages and roads have their origins in historical times; see for example Samojlik et al . , 2016 ) . As a result , the distribution of herbiscapes , driven by human-induced carnivore space use is arguably stable enough to create differential impact on vegetation in different herbiscapes . As humans are a crucial factor determining and restricting space use of large carnivores worldwide ( Tucker et al . , 2018 ) , they likely contribute toward stabilizing herbiscapes in many human influenced landscapes across the globe . Similarly , the results of our study have implications for patterns in vegetation structure that have been observed at large spatial scales . Herbiscapes will likewise also occur within the herbivomes described for the African continent . Our study indicates that predators can be a main factor structuring herbivore communities by redistributing predation-sensitive species over the landscape . As African systems harbor one of the most carnivore-rich communities in the world ( Ripple et al . , 2014 ) , large carnivores are expected to greatly influence variation in herbivore community structure within African landscapes ( Thaker et al . , 2011; Valeix et al . , 2009 ) . A major difference between our , and African study systems , is that humans supposedly play a less pronounced role in determining the space use of large carnivores and herbivores ( and hence the spatial arrangement of herbiscapes ) in the latter ( but see Tucker et al . , 2018 ) . The importance of this landscape-scale ( or within-herbivome ) variation in herbivore community structure in creating clear spatial patterns in vegetation , have already been illustrated for African systems . For example , Ford et al . ( 2014 ) showed that both predation risk and plant defenses enabled plants to thrive in different parts of a landscape; consequently , the thorniness of tree communities decreased across the landscape , contributing to intra-biome variation driven by predator-prey interactions ( Ford et al . , 2014 ) . The knowledge obtained in this study on spatial variation in the densities of local wildlife populations dependent on species-specific responses to habitat and disturbance factors gives us a better understanding of wildlife communities and may be relevant for effective wildlife management . In contrast to the common assumption that for management purposes wildlife populations are uniformly distributed , our study emphasizes the existence of large spatial variation in the landscape scale densities of ungulates . Such an approach fits particularly well in the recently developed concept of ‘hunting for fear’ , which promotes the spatio-temporal diversification of management techniques based on perception of risk in wildlife ( see Cromsigt et al . , 2012 ) . Moreover , maps visualizing cold- and hot-spots of ungulates across landscapes , together with information about community compositions , could also be useful tools for wildlife managers . Our results also introduces valuable knowledge relevant to the conservation of natural habitats . We argue that a unique , and unfortunately still undervalued character of Białowieża Forest is that ecosystem processes are still operating here at a landscape scale , as shown by our study . The ‘herbiscapes’ proposed in this paper are another unique aspect of the ecological processes that should be preserved within this area and will contribute to the ongoing debate on future conservation strategies for Białowieża Forest as UNESCO World Heritage ( Mikusiński et al . , 2018 ) . In conclusion , our study has illustrated that spatial variation in the structure of an entire large herbivore community results from interactive effects of species-specific responses to major ecological gradients operating at the landscape scale . Humans were a crucial factor associated with the landscape use of wolves and lynx . While lynx were not associated with the space use patterns of any ungulate , wolves were strongly ( negatively ) associated with the spatial distribution of their main prey species ( red deer ) , affecting the ungulate co-occurrence patterns at the landscape scale . The space use of European bison , moose , roe deer and wild boar was related to food resources . These processes led to the distribution of different functional types of herbivores over the landscape and created a clear spatial structure in the herbivore community , which we referred to as herbiscapes . Vegetation analyses suggested that herbivore impact measured by browsing intensity and regeneration of browsing-tolerant tree species consistently differed between herbiscapes . When these herbiscapes are stable enough they could be an important mechanism driving variation in herbivore impact on woody vegetation and thus maintain heterogeneity in a wide range of ecosystems .	In almost every ecosystem on Earth , communities of herbivores are kept in check by both predators and the availability of the plants they eat . As herbivores move in response to these pressures , they shape local plant communities and impact vegetation across entire landscapes . Yet the role of large plant-eating mammals in structuring ecosystems is often overlooked . Indeed , most research on this topic has looked at African ecosystems , like open savannahs , and fewer researchers have studied temperate forests like those found across Europe , Asia and North America . Bubnicki et al . have now examined factors influencing the distribution of five large herbivore species and resulting plant communities in Białowieża Forest in eastern Poland , the best-preserved European lowland forest . Their method involved measuring the cascading interactions of plants and animals in the forest using cameras set at nearly 900 locations , satellite images and other remote sensing technologies , and on-the-ground surveys . Added to this were patterns of human activity inferred from the available data for the study area . This approach allowed Bubnicki et al . to explore how humans are influencing the forest ecosystem , too . The analysis revealed that humans are the main factor influencing the movements of carnivorous predators in Białowieża Forest , but not the herbivores directly . Wolves and lynxes avoided areas heavily used by humans whereas large herbivores responded primarily to different environmental factors . Wild boar and bison are influenced by the availability of plant food and preferred habitat for foraging; moose and roe deer by the features of the landscape , like elevation or openness . The red deer was the only large herbivore species whose distribution was strongly linked to that of its main predator , the wolf . From this , Bubnicki et al . identified distinct areas in the forest which have emerged from the interactions at play , describing these areas as ‘herbiscapes’ for the herbivores that shaped them . These findings provide new understanding of the complex ecological processes shaping the Białowieża Forest and serve as a model to help understand other ecosystems around the world . The knowledge will also contribute to the ongoing management and conservation of this UNESCO World Heritage Area .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "ecology" ]	2019	Linking spatial patterns of terrestrial herbivore community structure to trophic interactions
Enzymes can increase the rate of biomolecular reactions by several orders of magnitude . Although the steps of substrate capture and product release are essential in the enzymatic process , complete atomic-level descriptions of these steps are difficult to obtain because of the transient nature of the intermediate conformations , which makes them largely inaccessible to standard structure determination methods . We describe here the determination of the structure of a low-population intermediate in the product release process by human lysozyme through a combination of NMR spectroscopy and molecular dynamics simulations . We validate this structure by rationally designing two mutations , the first engineered to destabilise the intermediate and the second to stabilise it , thus slowing down or speeding up , respectively , product release . These results illustrate how product release by an enzyme can be facilitated by the presence of a metastable intermediate with transient weak interactions between the enzyme and product . As it is becoming increasingly clear that proteins populate a variety of ‘intermediate’ states during their function ( Dobson , 2003; Sekhar and Kay , 2013 ) , it is essential to determine the structures of such states in addition to defining the native conformations . Protein intermediates are involved in folding , misfolding , and aggregation processes , as well as in events associated with molecular recognition , catalysis , and allostery ( Dobson , 2003; Sekhar and Kay , 2013; Tzeng and Kalodimos , 2013 ) . These species are transient in nature and as such they have been difficult to characterise . Nuclear magnetic resonance ( NMR ) spectroscopy has emerged in this context as a powerful technique to define such states as exemplified by the characterisation of the structures of species involved in folding ( Korzhnev et al . , 2010 ) , molecular recognition ( Tang et al . , 2006 ) , and aggregation ( Neudecker et al . , 2012 ) . In the present paper , we describe a study of the mechanism involved in the process by which an enzyme releases its products . This is one of the three major steps in an enzymatic catalysis process ( Fersht , 1999 ) . In the first step , the enzyme forms a complex with the substrate . In the second step , the transition state of the reaction is reached within the favourable environment provided by the catalytic site enabling the conversion of the substrate into product . In the third step , which is often rate-limiting , the product is released and the enzyme returns to its original state . Each of these steps is usually rather complex and involves reaction intermediates , which are transient in nature and difficult to characterise . In order to investigate the third step , we have studied here lysozyme , the first enzyme to be crystallised ( Blake et al . , 1965 ) , and whose structural properties have been characterised in great detail ( Blake et al . , 1965; Phillips , 1967; Artymiuk and Blake , 1981; Radford et al . , 1992 ) . The native structure of this enzyme is divided into a α domain ( residues 1 to 38 , and 86 to 130 ) and β domain ( residues 39 to 85 ) , containing primarily α-helical and β-sheet secondary structures , respectively ( Blake et al . , 1965; Phillips , 1967; Artymiuk and Blake , 1981; Radford et al . , 1992 ) . This enzyme degrades bacterial cell walls by catalysing the hydrolysis of the 1 , 4-β-linkages of the cell wall peptidoglycans , with a reaction that has been the object of intense scrutiny ( Chipman and Sharon , 1969; Warshel and Levitt , 1976; Post et al . , 1986; Vocadlo et al . , 2001 ) . According to the mechanism originally proposed by Phillips on the basis of his structure ( Phillips , 1967 ) , lysozyme binds to a peptidoglycan molecule in the binding site within the cleft between its two domains thus causing the substrate to adopt a strained conformation similar to that of the transition state of the hydrolysis . Here , we study the product release process . To this end , we used a well-characterised oligosaccharide product having an inhibitory effect on the enzyme , N , N' , N“-triacetylchitotriose ( triNAG ) ( Turner and Howell , 1995 ) , which has been frequently used for studying lysozyme–product interactions ( Post et al . , 1986 ) . We have presented the atomic resolution structure of an intermediate associated with the product release in an enzymatic reaction . We have validated this structure by identifying a distinctive structural characteristic of this state , a transient hydrogen bond between the side-chains of residues N44 and E35 . As this interaction stabilises the intermediate state but not the ground state , we introduced a mutational variant ( N44A ) that , by removing specifically the hydrogen bond , reduces the stability of the intermediate state but not that of the ground state and thus inhibits the release process . Our results provide an illustration of the manner in which conformational fluctuations can play a central role in enzymatic reactions by creating low-population intermediate states that facilitate the challenging step of release of the products of the catalytic reaction . Human lysozyme was expressed in Pichia pastoris and purified on an ion exchange column , as previously described ( Johnson and et al . , 2005 ) . 15N ammonium sulfate and 13C methanol were used to 15N and 13C label the protein , respectively . NMR experiments were carried out using a 700 MHz spectrometer at 37°C in a buffer at pH 5 . 0 containing 20 mM potassium phosphate and 10% D2O; the pH was re-adjusted after the addition of the protein . Protein concentrations were in the range of 200–350 μM . For the measurements of the bound state , N , N′ , N”-triacetylchitotriose ( triNAG ) sugar was purchased from Sigma and dissolved in water to constitute a concentrated stock solution . For the assignment of the free state at pH 5 . 0 and 37°C , we used a previously published assignment ( Ohkubo et al . , 1991; Hagan and et al . , 2010 ) , which was confirmed using HNCA measurements , which was run with a spectral width of 1561 Hz and 68 points in the 15N dimension , and a spectral width of 5456 Hz and 64 points in the 13C dimension ( Grzesiek and Bax , 1992 ) . In total , 126 backbone amides were assigned in the 1H-15N spectrum . For the full assignment of human lysozyme bound to triNAG , we performed titrations of 1H-15N HSQC spectra of a 200 μM sample of 15N human lysozyme , which were recorded using progressive concentration of the ligand ( 0 , 0 . 3 , 0 . 5 , 0 . 8 , 1 . 1 , 1 . 6 , 2 . 4 , 3 . 1 , 5 . 2 , and 10 equivalents ) , allowing us to sample different points along the binding curve . HSQC spectra were recorded with a spectral width of 1621 Hz and 128 points in the 15N dimension ( Figure 1—figure supplement 2 ) . Additional information was obtained using HNCA and HNCACB experiments of a triNAG-saturated human lysozyme sample ( Grzesiek and Bax , 1992; Muhandiram and Kay , 1994 ) . The HNCA experiment was carried out with the same settings as for the free state ( see above ) . The HNCACB experiment was carried out with a spectral width of 1561 Hz and 68 points in the 15N dimension and with a spectral width of 13 , 210 Hz and 72 points in the 13C dimension . These complementary data allowed us to obtain the full assignment of the 1H-15N spectra ( Figure 1—figure supplement 1 ) . Residual dipolar couplings ( RDCs ) were measured by orienting the free and triNAG-bound states in two different bicelle solutions , neutral and charged ( Ottiger and Bax , 1998; Schwalbe and et al . , 2001 ) . The neutral bicelle solution contained 5% wt/vol of a mixture of DMPC and DHPC ( q = 2 . 9 ) , whereas CTAB was used to create a positively charged solution of 10% wt/vol of the ( DMPC:DHPC:CTAB ) = ( 2 . 9:1:0 . 2 ) composition . Splitting of the 2H signal was recorded before and after the IPAP experiments , to ensure that alignment had remained constant during the course of the NMR experiment . IPAP experiments were recorded on the isotropic sample as well as on the two anisotropic samples ( neutral and charged ) ( Ottiger et al . , 1998 ) . These experiments were performed using a spectral width of 2447 Hz with 320 points in the indirect 15N dimension for the in-phase ( IP ) or anti-phase ( AP ) spectra . J-couplings were extracted in each medium and RDCs were derived , discarding overlapping and poorly defined peaks . For the free state , we extracted 109 RDCs in the steric medium and 110 in the charged medium; 109 RDCs were extracted for the bound state , both for steric and charged media . 3J HNHα couplings were obtained using HNHA experiments ( Vuister and Bax , 1993 ) , which were performed on the free and bound states using a 700 MHz spectrometer and a spectral width of 1454 Hz with 68 points in 15N and 9800 Hz with 72–80 points in the indirect 1H dimension . The 3J HNHα couplings were extracted using the ratio of intensities of cross- ( IX ) and diagonal ( ID ) peaks ( Kuboniwa et al . , 1994 ) ( 1 ) IX/ID=−tan2 ( 2Πξ3J ) with ξ = 13 . 05 ms . Errors in the 3J HNHα coupling values were based either on a 5% uncertainty or on the noise level for cross-peaks with intensities below the RMS noise of the HNHA spectrum , estimated using Sparky ( Goddard , T . D . , and D . G . Kneller . SPARKY 3 . University of California , San Francisco , 2004 ) . Errors on intensities were propagated according to Equation ( 1 ) to yield the error on 3J HNHα couplings . Residues with overlapping diagonal peaks were discarded , as well as glycine residues . As a starting structure for the ligand-free state , we used the crystal structure of human lysozyme at 1 . 9 Å resolution ( PDB code 2ZIJ ) . For the bound state , we used the crystal structure of the human lysozyme A96L variant bound to triNAG at 1 . 8 Å resolution ( PDB code 1BB5 ) . This structure was modelled by mutating back residue 96 from L to A , as in the wild-type sequence . Molecular dynamics simulations were performed by using AMBER99SB with corrections on backbone ( Best and Hummer , 2009 ) and side chains ( Lindorff-Larsen et al . , 2010 ) dihedral angles as the force field ( EFF ) for the protein . triNAG was modelled using the GLYCAM06 force field ( Kirschner et al . , 2008 ) . The protein and protein/triNAG systems were solvated using the TIP3P water model ( Jorgensen et al . , 1983 ) . A time step of 2 fs was used together with LINCS constraints ( Hess , 2008 ) . Systems were energy minimised and equilibrated with positional restrained simulations of 20 ns , in which the heavy atoms of the protein and triNAG molecules were restrained to their Cartesian coordinates . For the free state , the resulting system box after equilibration was 5 . 55 × 6 . 16 × 5 . 56 nm3 , with 5698 water molecules for a total of 19 , 123 atoms . For the bound state , the resulting system box after equilibration was 6 . 15 × 5 . 62 × 5 . 99 nm3 , with 6131 water molecules for a total of 20 , 509 atoms . The simulations were performed in the NPT ensemble by weak coupling the pressure and temperature with external baths . Temperature coupling was performed with the v-rescale method ( Bussi et al . , 2007 ) with a coupling constant of 0 . 1 ps . The pressure was kept constant using the Berendsen method ( Berendsen et al . , 1984 ) , with a coupling constant of 1 . 0 ps and at a reference pressure of 1 bar . The isotropic compressibility value was set to 4 . 5 × 10−5 bar−1 . Electrostatic interactions were treated by using the particle mesh Ewald method ( Essmann and et al . , 1995 ) . We used replica-averaged RDC restraints in molecular dynamics simulations ( De Simone et al . , 2011; Montalvao et al . , 2011; De Simone et al . , 2013a; De Simone et al . , 2013b ) . This method has been tested for its ability to sample interdomain motions in proteins ( De Simone et al . , 2011; De Simone et al . , 2013b ) , as well as in multiple conformational states in fast exchange in the NMR measurements ( De Simone et al . , 2013a ) . A recent study was carried to generate accurate ensembles of hen egg white lysozyme using RDC measured under the same conditions of the present work ( De Simone et al . , 2013b ) . This investigation has defined the sampling method that we have used here to characterise the conformational properties of lysozyme using RDC restraints . The accuracy of the resulting ensemble was benchmarked using a large variety of NMR observables , including eight sets of RDCs . Briefly , in this approach ( De Simone et al . , 2011; De Simone et al . , 2013b ) , the structural information provided by RDC measurements is imposed to restrain the molecular dynamics simulations by adding a term , ERDC , to a standard molecular mechanics force field , EPot: ( 2 ) ETot=EPot+ERDC . The resulting force field , ETot , is employed in the integration of the equations of motion . In this work , the restraint term , ERDC , is given by ( De Simone et al . , 2011; De Simone et al . , 2013b ) : ( 3 ) ERDC=α∑i ( Dexp−Dcalc ) 2 , where α is the weight of the restraint term , and Dexp and Dcalc are the experimental and calculated RDCs , respectively . The RDC of a given bond vector is calculated as ( De Simone et al . , 2011; De Simone et al . , 2013b ) : ( 4 ) Dcalc=1M∑mDm , where m runs over the M replicas and Dm is the RDC of replica m , which is given by: ( 5 ) D=Dmax∑ij〈Aij〉cosφicosφj , where φi and φj are the angles between the internuclear vector and the molecular reference frame , the indices i and j run over the three Cartesian coordinates , x , y , and z , and 〈Aij〉 is the ( i , j ) component of the alignment tensor . The use of replica-averaged molecular dynamics simulations enables one to generate an ensemble of conformations compatible with the experimental data according to the maximum entropy principle ( Pitera and Chodera , 2012; Cavalli et al . , 2013; Roux and Weare , 2013 ) , at least in the limit of large M and α . We have previously shown ( Cavalli et al . , 2013 ) , however , that it is possible to effectively achieve this limit even if the values of M and α remain relatively small and thus obtain conformational ensembles that provide a good agreement between experimental and calculated observables . Following these procedures , we used here M = 16 and for the weight , α , we first carried out an initial equilibration simulation at 310 K , during which the agreement between the calculated and experimental data was allowed to converge by gradually raising α to the largest possible value that did not generate numerical instabilities . Subsequently , we performed a series of 50 cycles of simulated annealing between 310 and 500 K to sample the conformational space . Each cycle was carried out for a total of 250 ps ( 125 , 000 molecular dynamics steps ) . For each cycle , we collected 24 , 000 structures ( 1 per ps in the final 50 ps of the final 30 cycles of each of the 16 replicas ) . These structures were employed for the analyses reported in this study . The alignment tensor is calculated from the shape and charge of the protein molecule using a procedure recently described ( Montalvao et al . , 2011 ) . We adopted such an approach here rather than the more commonly used singular value decomposition ( SVD ) method ( Clore and Schwieters , 2004b; Clore and Schwieters , 2004a ) because in the presence of conformational fluctuations of relatively large amplitude , such as those exhibited by hen lysozyme , the SVD method , when used in combination with the replica-averaging procedure of Equations 2–5 , is less effective in capturing the motions of a protein ( De Simone et al . , 2013b ) . The reason is that the SVD method does not necessarily provide the actual alignment tensor of a given structure but rather the alignment tensor that generates the RDC values in the closest agreement with the experimental ones and hence is less well suited in describing the specific differences between the structures considered in the averaging procedure in Equation ( 3 ) ( Montalvao et al . , 2011; De Simone et al . , 2013b ) . This structure-based method was used here to calculate the orientations of lysozyme in two alignment media , one steric ( DMPC/DHPC ) and one electrostatic ( DMPC/DHPC/CTAB ) . The Q factors for the refined ensembles of the free and bound states of human lysozyme were 0 . 10 in both cases . In addition to the previous extensive benchmarks performed on the structural ensembles of the hen egg white lysozyme ( De Simone et al . , 2013a ) , which were obtained using the same protocol employed in this work , we performed here a set of additional validations based on NMR measurements not used as restraints in the simulations and by comparing the resulting experimental values with those back-calculated from our ensemble of human lysozyme ( Figure 1—figure supplement 5 ) . N44A mutation and E35D or D53N ( control mutations ) were introduced into the pPIC9/HuLys wt by using QuikChange XL II mutagenesis kit ( Qiagen , Venlo , The Netherlands ) . The pPIC9 plasmid containing the point mutations of HuLys cDNA was linearised by digestion with StuI . Transformation into Pichia pastoris GS115 was performed by using Pichia EasyComp Transformation Kit ( Life Technologies ) , according to manufacturer's instructions . Cell colonies were screened for lysozyme expression level by quantifying by NuPAGE analysis the amount of lysozyme produced in 10-ml mini-cultures . Protein expression and purification were performed as previously reported ( Johnson and et al . , 2005 ) . Protein purity exceeded 95% as estimated by NuPAGE analysis . Protein concentrations were determined by absorbance measurements at 280 nm using theoretical extinction coefficients calculated with Expasy ProtParam . Surface plasmon resonance ( SPR ) experiments were performed using a Biacore 3000 system ( GE Healthcare ) . CM5 sensor chip surfaces were activated by using an amine coupling kit ( GE Healthcare ) . WT and N44A lysozyme variants were immobilised to the activated surfaces by amine coupling at a density of 2500–3000 resonance units ( RU ) . Single chain kinetic experiments were performed at 25°C using a flow rate of 20 µl/min in 50 mM phosphate pH 6 . 2 , 100 mM NaCl . Serial dilutions ( 200 µM , 100 µM , 50 µM , 25 µM , and 12 . 5 µM ) of N , N′ , N′′-Triacetylchitotriose ( Tri-NAG , Sigma Aldrich ) were sequentially injected every 700 s using a contact time of 250 s for each injection . Data fitting was performed with the single chain kinetic module provided with the Biaevaluation software ( Biacore GE lifesciences ) . Hydrolase activity assay was performed using Micrococcus lysodeikticus cells ( Sigma Aldrich ) as the substrate . Cells of Micrococcus were suspended at 0 . 3 mg/ml in 100 mM potassium phosphate , pH 6 . 2 , shortly before the assay . The decrease of Absorbance at 450 nm was monitored at 25⁰C in the presence of 20 nM lysozyme variants .	Enzymes are proteins that catalyse biochemical reactions . They bind to their target molecules—known as substrates—and help to change them to make ‘products’ . Afterwards , the products are released and the enzymes are free to bind to the next molecules . To perform this activity , an enzyme can change its structure several times , but it has been challenging to characterise the intermediate shapes because of their transient nature . De Simone et al . took advantage of a technique called nuclear magnetic resonance spectroscopy to get a better look at the structures adopted by the human enzyme lysozyme . This enzyme helps to protect us from bacterial infections because it breaks the links between peptidoglycan molecules , which make up the wall that surrounds bacterial cells . The experiments show that two ‘arms’ in the lysozyme structure move to create an intermediate shape during the final step—the release of the product—in the chemical reaction . This type of flexibility gives the enzyme the ability to tightly bind the peptidoglycan at the start and to let go of the product when the chemical reaction is complete . Next , to confirm their findings , De Simone et al . examined what happened when they introduced particular mutations in the gene that makes lysozyme . The first mutation was meant to destabilise the intermediate shape of the enzyme , which resulted in the enzyme binding more tightly to the peptidoglycan in the final step and releasing it more slowly . A second mutation was made to stabilize the structure of the intermediate shape , which , as expected , allowed lysozyme to release the peptidoglycan more quickly . De Simone et al . 's findings explain how intermediate shapes can be involved in the release of the product from lysozyme and other enzymes . The next challenges will be to characterise the structure of the intermediate shape that binds to the substrate and , more generally , to extend this type of approach to other enzymes .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Structure of a low-population intermediate state in the release of an enzyme product
A founding paradigm in virology is that the spatial unit of the viral replication cycle is an individual cell . Multipartite viruses have a segmented genome where each segment is encapsidated separately . In this situation the viral genome is not recapitulated in a single virus particle but in the viral population . How multipartite viruses manage to efficiently infect individual cells with all segments , thus with the whole genome information , is a long-standing but perhaps deceptive mystery . By localizing and quantifying the genome segments of a nanovirus in host plant tissues we show that they rarely co-occur within individual cells . We further demonstrate that distinct segments accumulate independently in different cells and that the viral system is functional through complementation across cells . Our observation deviates from the classical conceptual framework in virology and opens an alternative possibility ( at least for nanoviruses ) where the infection can operate at a level above the individual cell level , defining a viral multicellular way of life . Viruses introduce their genome into a susceptible cell and highjack diverse cell functions to complete their replication cycle . Once this ‘cell-autonomous’ cycle is completed , virus particles exit the infected cell and enter a new healthy one where a similar cycle is reiterated . This universal view of the viral way of life has been applied to multipartite viruses since their discovery , over half a century ago ( Brakke et al . , 1951; Lister , 1966; van Kammen and van Griensven , 1970; Gokhale and Bald , 1987 ) , despite the fact that such a conceptual framework fails to explain the evolution and even the functioning of these viral systems ( Iranzo and Manrubia , 2012; Sicard et al . , 2016; Lucía-Sanz and Manrubia , 2017 ) . Depending on the viral species , multipartite viruses have their genome composed of two to eight segments of DNA or RNA ( single or double stranded ) , each encapsidated individually in a separate virus particle . They have been reported to infect frequently plants and fungi where they represent at least 35–40% of the viral genera and families described ( Hull , 2014 ) , rarely insects ( Hu et al . , 2016; Ladner et al . , 2016 ) , and only hypothetically vertebrates ( Ladner et al . , 2016 ) . The benefits of the individual encapsidation of distinct segments are unclear and debated ( Sicard et al . , 2016; Lucía-Sanz and Manrubia , 2017 ) . In contrast , the reduced probability to infect individual host cells with at least one copy of each segment is unanimously acknowledged as a major cost that increases with the number of segments composing the viral genome ( Iranzo and Manrubia , 2012 ) . This cost appears insurmountable , at least for viruses with more than three genome segments , and sets the existence of such highly multipartite viruses as an enigma in general biology ( Iranzo and Manrubia , 2012; Sicard et al . , 2016; Lucía-Sanz and Manrubia , 2017 ) . Nevertheless , its empirical basis has never been questioned . Using a highly multipartite virus with eight genome segments , we here propose that the conceptual framework in Virology should be amended to account for such viral systems . Indeed , in our specific experimental model species , we demonstrate that the distinct segments do not need be together in individual cells for the system to be functional . They accumulate independently in and complement across neighboring cells , defining a multicellular way of life . Our first aim was to experimentally verify whether the different segments of a multipartite virus have to be together in individual cells for the system to be functional . We used the highly multipartite faba bean necrotic stunt virus ( FBNSV , genus Nanovirus , Family Nanoviridae ) , whose genome is composed of eight circular ssDNA segments ( Figure 1A ) ( Grigoras et al . , 2009 ) . We first compared the localization of distinct segments in individual cells of faba bean host plants with specific fluorescent probes and confocal microscopy . In plants submitted to productive viral infection ( see Materials and methods ) , the visualization of pairs of segments respectively labeled with green and red fluorochromes immediately revealed that distinct segments do not necessarily co-occur but are most often found in different cells ( Figure 1B–I and Figure 1—figure supplement 1 ) . We tested 7 out of the 28 possible pairs of segments . In all cases we observed a similar situation with one segment most often highly accumulated in the absence of the other in individual cells . Remarkably , this applied even to the pairs from the three segments encoding for the three basic functions of plant viruses: replication ( segment R encoding M-Rep ) , encapsidation ( segment S encoding CP ) and intra-host movement ( segment M encoding MP ) , suggesting that cells containing all eight segments are extremely rare . Though challenging the current view of the viral cell-autonomous replication cycle , a simple explanation of our observation is that the FBNSV can function while its genome segments occur in distinct neighboring cells . This possibility calls i ) for further evidence that the accumulation of a given segment is independent of the accumulation of the others in individual cells and ii ) for the proof of concept that the function encoded by a given viral segment can complement the others at a distance , in cells where this segment is absent . One may argue , sticking to the paradigm that the viral genetic information is replicated as an integral genome within individual cells , that all FBNSV segments are present in infected cells but that the apparent absence of some in Figure 1 is due to the detection limit of our technique . Because every technique has its limit , whatever the technology implemented , it would not be reasonable to certify that the absence of detection is a proof of the absence of the corresponding segment . We thus imagined an approach where the detection limit becomes irrelevant . For a given pair of segments , we quantified and compared both green and red fluorescence in all individual cells where at least one of the two was observed above background ( for detailed quantification procedure see the Materials and methods section ) . By doing so , we alleviate the problem of the limit of detection and rather question whether the accumulation of one segment of the pair is dependent on that of the other . As a positive control of this approach , we first produced two fluorescent probes , each specifically labeling a different region of the same segment R ( probes R1 and R2 ) . Figure 2 illustrates that all cells labeled by R1 are also labeled by R2 ( Figure 2A ) . Moreover , plotting the average intensity of green over red fluorescence for each of these cells resulted in a highly significant linear relationship ( correlation coefficient r = 0 . 90 , p=1 . 98 10−23 in the example of petiole N° 42 in Figure 2B ) . Four independent repeats of this control similarly showed strong correlations ( Table 1 ) . This result validates our approach by demonstrating that when we monitor two viral DNA sequences whose accumulation should be highly correlated , such as two regions of the same segment , we indeed find that they co-localize and accumulate at highly correlated levels . This approach was then extensively used for the pair of segments R/S because they encode the two key viral functions , replication and encapsidation . In total , we tested 10 petioles , each from a different plant infected independently . No significant correlation could be found ( Figure 2C , Table 1 ) , whether at early or late stages of infection ( Table 1 and also see Figure 1—figure supplement 2 ) , showing that the accumulation of the segment S in a cell is independent of that of the replication-encoding segment R . We extended this analysis to five additional petioles from five independent plants , two for the pair R/M and three for the pair S/M , and again found no significant positive correlation; a result indicative of independent accumulation ( Table 1 ) . To further support this conclusion , we used the earlier reported variation in the relative frequency of the segments across infected plants ( Sicard et al . , 2013 ) . Within an infected petiole , if two segments independently enter and accumulate in individual cells , then their relative frequencies in infected tissues should be proportional to the ratio of the number of cells they respectively infect . We quantified by qPCR the relative frequency of the two segments of the pair in a fragment of eight of the analyzed petioles ( Supplementary file 1: Table S1 ) . Figure 2D shows a significant correlation between the relative frequency of the segments in the tissues and the relative number of cells they respectively infect ( n = 7 , F = 12 , R = 0 . 677 , p-value=0 . 0122 ) . Hence , two different technologies ( FISH-confocal microscopy and quantitative real-time PCR ) concur to demonstrate that the accumulation of distinct FBNSV segments in individual cells is independent . To make sense out of the above observation we assumed that a viral function can be present and act in a cell where its encoding segment is not , and thus that the viral system can be functional despite the independent dispersal of its distinct genes in distinct cells . The validation of this assumption was set up with a focus on the replication function . We first confirmed that the segments detected in individual cells are the product of replication . Then we searched for the presence of the protein M-Rep in these cells , to support replication even when its encoding segment R is absent . Two possible alternatives could lead to the detection of a segment as a fluorescent signal spread out in a large volume of the nucleus . Either one ( or very few ) copy entered , replicated and invaded the nucleus to detectable level , or a number of copies large enough to be detectable entered and diffused throughout the nucleus with no replication . When looking at images shown in Figure 1 , the latter appears unrealistic because it would imply the specific sorting of different segments at their entry in distinct nearby cells . Nevertheless , to experimentally dismiss this possibility , we similarly investigated the distribution of two alleles of the same segment ( differing solely by a small inserted marker sequence ) for which no specific sorting can be expected . In the absence of sorting , if the segment massively enters cells to detectable level without replication , the two alleles should be co-detected in most cells . In contrast , if one ( to very few ) copy of this segment enters each individual cell and then accumulates through replication , allele-specific FISH labeling should reveal solely one of the two alleles in most cells . Three independent experiments were carried out for each of the segments S and N where two genetic markers have earlier been inserted ( Gallet et al . , 2017 ) . They consistently revealed that most individual cells ( 54% to 100% ) contained only one detectable allele ( Figure 3—figure supplement 1 ) , confirming that one to very few copies of the segment initially enter individual cells and thus that detection all over the nucleus can only result from replication . We then confronted the cells where the segment S is detectable ( thus where it has replicated ) to the detection of either segment R or its protein product M-Rep , by a combination of FISH and immunofluorescence-labeling ( Figure 3A and B ) . While segment R was detectable in only a minority of these cells ( approx . 40% ) , its protein product M-Rep was positively revealed in nearly 85% ( Figure 3C ) , indicating that the expression product of R segment can move to neighboring cells . That the protein M-Rep is undetectable in a few cells replicating segment S ( approx . 15% ) can be explained by distinct turnovers for the genome segments and their expression products . Indeed , both mRNAs and proteins have a relatively rapid turnover within a cell , whereas the DNA viral sequences can be stored indefinitely into the cell nucleus either encapsidated or as stable minichromosomes ( Gronenborn , 2004; Ramesh et al . , 2017; Deuschle et al . , 2016; Rodríguez-Negrete et al . , 2014 ) . Thus , we assume that for all cells where at least one FBNSV segment is detectable the protein M-Rep is or has been present , but that this protein may have disappeared in some cases at the moment of the observation . Consistent with this assumption is the observation of a similar proportion of cells containing the segment R where its own expression product M-Rep is no longer detectable ( Figure 3C ) . It is important to note the detection of the M-Rep protein , often with intense signal ( Figure 3A & B ) , in a large number of cells where segment R is absent ( Supplementary file 1: Table S1 , petioles 36–39 ) . Moreover , Figure 3C ( orange middle and bleu-grey right boxplots ) indicates that the protein M-Rep is not more associated to cells containing its own encoding segment R than to cells containing other segments . Although they represent indirect evidence , these observations together further support our conclusion that either the mRNA or the protein M-Rep itself can travel from the producing cells ( those where segment R accumulates ) to other cells of the host , as further discussed below . Altogether , our results demonstrate that key genome segments of the FBNSV accumulate in individual cells independently of the presence and accumulation of the others . We thus propose that the different parts of the viral genome can complement across distinct neighboring cells and can only sustain the productive infection at a multicellular tissue level . As numerous other plant virus species ( Hipper et al . , 2013; Folimonova and Tilsner , 2018 ) , nanoviruses are restricted to vascular tissues and replicate in phloem companion and parenchyma cells ( Shirasawa-Seo et al . , 2005 ) . A model compatible with our observations is that each genome segment entering and expressing within a cell can export its gene products as ‘common goods’ in neighboring cells and/or sieve elements , so that these common goods are redistributed among interconnected cells and complement the system . The demonstration that the protein M-Rep accumulates and functions in cells devoid of its encoding DNA-R is a proof of concept fully supporting this model . Numerous studies have shown that plant mRNAs ( Kehr and Kragler , 2018 ) and proteins can move from cell to cell or long distance ( Turnbull and Lopez-Cobollo , 2013; Lopez-Cobollo et al . , 2016; Paultre et al . , 2016 ) , even from root to shoot . Unfortunately , the mechanisms and specific control of these mRNAs and proteins mobility is vastly elusive . For FBNSV , each segment may export and redistribute its product in a distinct way , some may produce mobile mRNA , others may produce mobile protein , and others may be complemented for movement by the product of a specific segment ( or even by a host factor ) that may act as a carrier . Specifying these diverse possibilities is beyond the scope of this report . Likewise , whether this non-cell-autonomous model can be extended to viral species invading non-phloem tissues where cell communication is more restricted is unknown . Non phloem-restricted viruses could induce the formation of symplastic domains , either by manipulating the endogenous capacity of the host plant to do so ( Faulkner , 2018 ) or by opening plasmodesmata through the action of their non-cell-autonomous movement protein ( Lucas , 2006 ) , but this possibility awaits further investigation . Here we introduce an additional concept in virology , which is compatible with empirical observations and which partially alleviates the insurmountable cost in highly multipartite viral systems such as FBNSV ( Iranzo and Manrubia , 2012 ) : because concomitant infection of individual cells by all genomic segments is not necessary , the associated putative cost should be much smaller if not nil at the within-host level . We earlier discussed the fact that the analogous cost upon between-host transmission and the mechanisms of its compensation remain to be understood ( Gallet et al . , 2018 ) . Such a multicellular way of life could be adopted in other multicomponent genetic entities , such as other multipartite viruses , segmented viruses which often fail to encapsidate all genome segments together ( Luque et al . , 2009; Wichgers Schreur and Kortekaas , 2016; Brooke , 2017 ) , satellites , and defective interfering particles , if functional complementation could occur at a supra-cellular level . Faba bean ( Vicia faba , cv . ‘Sevilla’ ) was used as the host plant in all experiments , and seeded , maintained and inoculated as described ( Sicard et al . , 2013; Gallet et al . , 2017; Gallet et al . , 2018; Sicard et al . , 2015 ) . The viral infectious clone is from the species Faba bean necrotic stunt virus , genus Nanovirus , family Nanoviridae . Its construction in agrobacterium plasmids has been earlier described in details ( Grigoras et al . , 2009 ) , as well as all conditions and procedures for agro-inoculation ( Grigoras et al . , 2009; Sicard et al . , 2013; Gallet et al . , 2017; Gallet et al . , 2018; Sicard et al . , 2015 ) . Plants were sowed as one seed per pot . The plantlets were agro-inoculated nine days later and symptoms appeared from 9 days post inoculation ( dpi ) till 19 dpi , depending on individual plants . As shown in all the publications cited in this paragraph , the inoculation of plants with this infectious clone routinely yields a productive viral infection , meaning that the FBNSV particles can be purified from these infected plants and very efficiently transmitted to new healthy ones by aphids . Infected plants were sampled at different days post inoculation: at 11 , 12 , 13 , 17 , 20 , 25 and 27 dpi ( Table 1 and Supplementary file 1: Table S1 ) . Nanoviruses , and thus FBNSV , replicate and accumulate primarily in phloem vascular bundles of the upper part of the plant , in companion cells and phloem parenchyma cells of the stem , petioles and main leaf veins . A 1 to 3 cm-long section of the petiole of the upper leaf level was cut off from each infected plant and immediately fixed and processed for fluorescent in-situ hybridization ( FISH ) and/or immunofluorescence labeling ( IF ) . In some cases , the petiole was divided into two equal portions , one for FISH and/or IF and the other for DNA extraction and qPCR . The FBNSV genome is composed of eight circular ssDNA segments of around 1 kb each ( Figure 1A ) . Total DNA extraction from infected petioles and qPCR reactions specific to each of the eight segments were performed as described previously ( Gallet et al . , 2017 ) , including primer pairs , PCR conditions , and post-PCR data analysis . The relative frequency of two segments of a pair within a given petiole was obtained by dividing the estimated copy number of a segment by that of the other . Random priming and incorporation of Alexa Fluor-labeled dUTP ( either Alexa Fluor 488 or Alexa Fluor 568 for green and red labeling , respectively ) were performed to prepare probes specific to each of the eight segments , with the BioPrime DNA labeling system kit ( Invitrogen ) and according to the manufacturer instructions except that dUTP-Alexa Fluor were used in place of dCTP-biotine of the kit . For each probe , the template DNA corresponded to a PCR-amplified fragment restricted within the viral gene coding sequence , which is the sequence unique to each segment . The primer pairs used to amplify the coding sequence of the targeted segments are listed in Supplementary file 1: Table S2 . In some cases , short probes were synthesized by the company Eurogentec and covalently linked to either ATTO-488 ( green ) or ATTO-565 ( red ) fluorochrome ( sequences also available in Table S2 ) . The specificity of each probe was tested on membranes , on dotted plasmids each containing one of the eight viral segments , as well as on healthy plants . Each petiole harvested from an infected plant as described above was fixed by gentle stirring overnight at 4°C in PBS buffer containing 4% paraformaldehyde and 0 . 2% Tween-20 . After one rinse with 2 mL of PBS buffer containing 0 . 1M glycine , the petiole was transferred into ethanol 70% and stored at 4°C until use ( max . storage time: 1 month ) . Fixed petioles were then placed individually into Eppendorf tubes containing 8% low-melting agarose in water at 40°C . Petiole trunks were maintained into an upright position until the agarose cooled down and polymerized , and placed at 4°C for 4 hr . The jellified agarose blocks were extorted from the tubes and cross-sections of 80–100 microns were produced with a Vibratome HM650V ( Microm ) , set up in mode CPC , program 50 , speed 23 , frequency 100 , and amplitude 0 . 6 . Cross-sections were treated in 1 mL of Carnoy solution ( six volumes chloroform , three volumes ethanol , and 1 vol acetic acid ) during 1 hr under gentle stirring and then rinsed once in PBS buffer . An RNAse treatment ( 100 μg/mL of RNAse1 in PBS ) was applied to all samples for 45 min . at 37°C , in order to eliminate viral mRNAs , followed by one rinse in PBS and three additional rinses in hybridization buffer ( 20 mM Tris-HCl pH8 , 0 . 9M NaCl , 0 . 01% SDS , 30% Formamide ) . Fluorescent probes were diluted thirty times in hybridization buffer ( 10 μL in 300 μL total ) , denatured 10 min . at 100°C and then rapidly cooled on ice for 15 min . Petiole cross sections were then incubated overnight at 37°C in the diluted and heat-denatured probes solutions into embryo dishes sealed with parafilm membranes . After three rinses of 5 min with hybridization buffer and one with PBS , petiole sections were mounted on microscopy slides in Vectashield antifade mounting medium containing 1 . 5μg/mL DAPI for staining nuclei . Some samples were further treated for immuno-fluorescent labeling of M-Rep protein . In these cases the FISH-treated petiole cross sections were collected after the last PBS rinse and blocked for 1 hr and 30 min . in PBS + 5% BSA . The incubation with the primary antibody ( M-Rep specific antibody ‘FBNYV-M-Rep 8th bleed’ ( Vega-Arreguín et al . , 2005 ) , diluted 1/300 ) was in PBS + 5% BSA overnight at 4°C , and that with the secondary antibody ( goat anti-rabbit Alexa Fluor 594 IgG conjugate , diluted 1/250 , Life Technologies ) was in PBS + 5% BSA for 1 hr at 37°C . Samples were submitted to three rinses of 10 min each in PBS + 0 . 05%Tween-20 at room temperature after incubations with primary and secondary antibodies . Samples were finally transferred into PBS and then mounted on microscopy slides as above . Observations were all performed in sequential mode using a Zeiss LSM700 confocal microscope . Alexa Fluor 488 and ATTO 488 were excited with a 488 nm laser and the variable secondary dichroic ( VSD ) beam splitter was set to recover fluorescence up to 535 nm . Alexa Fluor 568/594 and ATTO 565 were excited with a 555 nm laser and the VSD was set to recover fluorescence up to 615 nm . DAPI was excited with a 405 nm laser and the VSD was set to recover fluorescence up to 626 nm , with the additional use of a short pass SP490 eliminating wave lengths > 490 nm . Images were acquired with 40x or 63x objectives at variable resolutions , depending on their intended use , with a pinhole aperture of 1 airy unit . Acquisitions were either in plane or stack mode , and the images used from stacks correspond to maximum intensity projections . The settings for the acquisition of images shown in Figures 1 , 2 and 3 , Figure 1—figure supplement 1 and Figure 3—figure supplement 1 are detailed in Supplementary file 1: Tables S3 . All images used for fluorescence quantification were acquired with the 40x objective , and the resolution chosen ( 512 × 512 ) corresponded to a compromise between image quality and acquisition time . All were acquired in stack mode and quantification performed on maximum intensity projections . The absence of potential biases induced by image acquisition at distinct resolution for fluorescence quantification has been verified as described in the next section . For constructing Figures 1 , 2 and 3 , Figure 1—figure supplement 1 and Figure 3—figure supplement 1 , as well as for selecting all cells where the fluorescence has been quantified , raw images were processed using ImageJ software version 1 . 50c4 . The overall intensity of green and red signals was adjusted for each individual image up to the limit where the green and red auto-fluorescence of the cell wall , cytoplasm and nuclei of the xylem and mesophyll cells visually disappeared ( the FBNSV being phloem-restricted , it does not invade xylem and mesophyll cells ) . Individual cells in phloem bundles were considered containing a segment when either green or red fluorescence ( or both ) could be visualized above background . Above-background fluorescence was never observed in non-phloem cells , consistent with the phloem restriction of this virus . Infection by FBNSV distorts the anatomy of the phloem tissues and so distinguishing between phloem parenchyma cells , companion cells and sieve elements is difficult and sometimes impossible in our petiole cross-sections . For this reason , and because we were primarily interested by cells where the viral DNA replicates ( ssDNA viruses replicate and accumulate in the nucleus of their host cells ) , we quantified the fluorescence only in those cells where a nucleus was clearly revealed by the DAPI staining , so in cells other than the sieve tubes . The DAPI staining allowed to precisely encircle the nucleus , and thus to define an ovoid corresponding area in each selected individual cell . Both green and red fluorescence were then quantified for each pixel within this area and the average pixel fluorescence intensity was estimated . This gave one average intensity value ( in arbitrary fluorescence units ) for green and one for red fluorescence in each selected individual cell within an infected petiole and all quantitative results are provided in Supplementary file 2: Table S4 . As indicated above , the setting of the confocal microscope was adjusted depending on the planned ulterior use of images . For example , to prepare Figure 1 , the plane mode , higher resolution ( >512×512 ) and higher magnification ( 63x objective ) were generally preferred . In contrast , for fluorescence quantification , the stack mode and maximum intensity projection images appeared best appropriate in order to capture the whole fluorescence of the nuclei . For obvious practical reasons , the numerous images needed for quantification were acquired more rapidly at the feasible resolution of 512 × 512 , and with a lower magnification ( 40x objective ) in order to screen microscopy fields potentially allowing the scoring of more segment-containing cells ( examples of such images are shown in Figure 1B , and Figure 3A ) . We wished to confirm that different resolution settings do not differentially affect green and red fluorescence . Another potential concern was the time spanning between microscopy slide preparation and the end of the observation/quantification . Due to the experimental design , it was important to confirm that the stability of red and green probes was similar and that a possible differential degradation could not bias our localization and correlation/regression analysis . We first quantified the fluorescence of R1-Green and R2-Red probes in petiole N°43 with fast acquired images ( resolution 512 × 512 ) . We then repeated the observation 2 weeks later on the same microscopy slides with a resolution 1024 × 1024 . The cells quantified at the two dates were not necessarily the same as we could not be sure to retrieve all of them . The regression analysis from these two sets of images gave remarkably similar results , demonstrating that the resolution setting and the aging of the preparation could only have negligible effect ( if at all ) on our conclusions . Results from these controls are shown as Figure 2—figure supplement 1 . Linear regression analyses were performed with the JMP 13 . 2 . 0 software . Each petiole was analyzed separately . We did not analyze the fluorescence quantification data set as a whole ( pooling cells from distinct petioles ) because the virus-associated fluorescence intensity is not directly comparable . The total viral load varies in between distinct petioles because of a ‘natural’ variance across infected plants and because some were harvested at different dpi . The auto-fluorescence of the infected tissues also varied across petioles because of the time of infection , the total viral load and presumably other unknown factors . The proportion of cells containing both S and R segments , both S segment and M-Rep protein , or both R segment and M-Rep protein ( Figure 3C , Supplementary file 1: Table S1 ) was analyzed with Generalized linear models ( GLM ) with ‘treatment’ ( S segment – R segment/S segment – M-Rep/R segment – M-Rep ) as a categorical explanatory factor , and a quasi-binomial error type . The GLM model was computed using R software 3 . 1 . 3 . The name of each statistical test and its outcome is indicated in the text and figure legends . All data are available in the manuscript and in Supplementary files . Raw data of all quantified green and red fluorescence within individual cells of infected plants are provided as a separate Excel file , Supplementary file 2: Table S4 . To allow repeat/reproduce all correlation tests , the 508 raw/unprocessed images ( . lsm format ) used for preparing all figures and for fluorescence quantification in individual cells have been deposited in the public repository figshare . They can be accessed at the DOI: 10 . 6084/m9 . figshare . 5981968 .	Many viruses are small particles consisting of genetic material surrounded by a coat made of proteins . They are unable to multiply on their own and so they must enter a host cell and trick it into reading their genetic information to produce new virus particles . It is generally thought that the process of making new virus particles happens independently in each infected cell . This idea assumes that a given particle contains the entire set of genetic material ( known as the genome ) of that virus , but this is not always the case . Many so-called ‘multipartite’ viruses have genomes that are split into several segments carried in separate particles: in this case , a single particle only contains a portion of the entire viral genome . Faba bean necrotic stunt virus ( or FBNSV for short ) is a multipartite virus that infects and causes disease in members of the pea and bean family . There are eight types of FBNSV particle that each carries a distinct genome segment , a small section of the entire viral genome . There is a low probability that a single cell could be infected with all eight different types of particle at the same time and receive the complete FBNSV genome . So how is this virus able to successfully multiply within a plant ? To address this question , Sicard et al . used microscopy to study FBNSV genome segments as they infected the cells of faba bean plants . The experiments confirmed that the eight different segments of the FBNSV genome were not necessarily found together within the same cell , but instead accumulated independently in different cells . This means that a cell infected with FBNSV may be unable to make all of the proteins needed to assemble new virus particles . However , additional experiments demonstrated that infected cells may be exchanging virus proteins , which could enable them to create complete virus particles . The findings of Sicard et al . demonstrate that FBNSV hijacks groups of host cells to manufacture new virus particles , rather than relying on individual cells as previously thought . It is possible that other multipartite and non-multipartite viruses work a similar manner . Ultimately , this knowledge may reshape what we know about how viruses infect their hosts .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "short", "report", "microbiology", "and", "infectious", "disease" ]	2019	A multicellular way of life for a multipartite virus
The chromatin landscape and promoter architecture are dominated by the interplay of nucleosome and transcription factor ( TF ) binding to crucial DNA sequence elements . However , it remains unclear whether nucleosomes mobilized by chromatin remodelers can influence TFs that are already present on the DNA template . In this study , we investigated the interplay between nucleosome remodeling , by either yeast ISW1a or SWI/SNF , and a bound TF . We found that a TF serves as a major barrier to ISW1a remodeling , and acts as a boundary for nucleosome repositioning . In contrast , SWI/SNF was able to slide a nucleosome past a TF , with concurrent eviction of the TF from the DNA , and the TF did not significantly impact the nucleosome positioning . Our results provide direct evidence for a novel mechanism for both nucleosome positioning regulation by bound TFs and TF regulation via dynamic repositioning of nucleosomes . Dynamic access to specific genetic information is critical for cellular development and response to the environment . Thus , processes such as transcription must be mediated by mechanisms that regulate gene function rapidly and reliably ( Barrera and Ren , 2006; Kornberg , 2007 ) . In eukaryotic cells , proper transcriptional regulation depends upon transcription factors ( TFs ) that bind to specific DNA-binding sites ( Kadonaga , 2004 ) . Additionally , the repression of transcription has often been correlated with the presence of nucleosomes , the basic units of chromatin structure , in which histone–DNA interactions act as a barrier for RNA polymerase elongation along DNA ( Li et al . , 2007; Petesch and Lis , 2012; Teves et al . , 2014 ) . Therefore , understanding the relationship between TF binding and nucleosomes is essential in understanding gene expression and regulation ( Voss and Hager , 2014 ) . Chromatin landscape and promoter architecture are dominated by the interplay of nucleosome and TF binding . Nucleosomes and TFs have been shown to compete for binding to DNA ( Mirny , 2010; Moyle-Heyrman et al . , 2011; Lickwar et al . , 2012 ) . This competition is based on the respective affinities of the TF and nucleosome for DNA , and depends upon DNA sequence , histone variants , and histone modifications . However , a nucleosome may also be repositioned through the action of chromatin remodelers , suggesting additional levels of transcription regulation . Some TFs are known to recruit nucleosome remodelers . Previous earlier studies focused on how these recruiting TFs affect the outcomes of nucleosome remodeling ( Nagaich et al . , 2004; Boeger et al . , 2008; Dechassa et al . , 2010; Voss et al . , 2011 ) . However , it is unclear how TFs that do not recruit remodelers influence the chromatin landscape . We hypothesize that nucleosome remodeling , without remodeler recruitment , may regulate the state of a bound TF . Specifically , a remodeler may attempt to move a nucleosome to or through a site pre-occupied by a TF . During such an encounter , the TF may be displaced , or it may act as a roadblock for nucleosome remodeling . Thus , chromatin remodeling may serve as an alternative mechanism to regulate transcription through its influence on a bound TF , and a bound TF may dictate the location of a remodeled nucleosome . Here , we studied the influence of nucleosome remodeling on a bound TF in a single molecule assay . We used a DNA unzipping technique ( Jiang et al . , 2005; Shundrovsky et al . , 2006; Hall et al . , 2009; Jin et al . , 2010; Dechassa et al . , 2011; Li and Wang , 2012; Inman et al . , 2014 ) to characterize the locations of a bound TF and a nucleosome simultaneously , on long DNA templates to near base pair accuracy . By examining the remodeling behavior upon encountering a bound TF , we determined that the relationship between TFs and nucleosome remodeling not only plays a critical role in nucleosome positioning , but also reveals a novel mechanism for how a TF can be dynamically recycled by nucleosome remodeling . In this work , we needed to precisely locate the positions of a nucleosome and a TF before and after nucleosome remodeling . We thus employed the DNA unzipping technique ( Figure 1—figure supplement 1 ) , which has been demonstrated to be a powerful single molecule technique for accurate and precise determination of positions and strengths of DNA–protein interactions ( Jiang et al . , 2005; Shundrovsky et al . , 2006; Hall et al . , 2009; Jin et al . , 2010; Dechassa et al . , 2011; Li and Wang , 2012; Inman et al . , 2014 ) . To evaluate the precision of this approach , we constructed a DNA template containing a single Gal4 sequence for binding to the Gal4 DNA-binding domain ( Gal4DBD ) and a single 601 nucleosome positioning sequence ( 601NPE ) for uniquely positioning a nucleosome . Gal4DBD contains only the 147 amino acids of the N terminal domain of the Gal4 protein and does not have any known remodeler recruitment function . Figure 1 shows representative traces from unzipping DNA molecules without nucleosome remodeling . The top trace of Figure 1 shows the result when the DNA template was unzipped starting from the Gal4 side . Both Gal4DBD ( a single smaller peak ) and a nucleosome ( two clusters of larger peaks ) were readily detected above the baseline of the corresponding naked DNA . The bottom trace shows the result when the DNA template was unzipped starting from the nucleosome side . Although the nucleosome unzipping signature was readily detectable , the unzipping signature of Gal4DBD was sometimes masked by that of the nucleosome . Therefore , it was often necessary to carry out unzipping experiments from both directions . Analysis of these unzipping signatures confirmed that unzipping mapped the position of the TF and the nucleosome to near base pair precision ( ‘Materials and methods’; Figure 1—figure supplement 2; Figure 1—figure supplement 3a; Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 06249 . 003Figure 1 . Single molecule unzipping technique detects Gal4DBD and nucleosome to near base-pair accuracy . DNA molecules , each containing a nucleosome and a bound Gal4DBD , were unzipped . All unzipped DNA molecules used in this work were in the region of 600 bp to 1 . 2 kbp . For clarity , much smaller regions are shown in all figures , with the origin of a template sequence defined as center position ( the dyad ) of the 601NPE . Shaded regions indicate locations of the Gal4 binding sequence and the 601NPE . ( top panel ) Cartoon illustrating the unzipping template design using for this experiment . A Gal4 sequence was separated from a 601NPE by 10 bp . The orientation of the 601NPE sequence is indicated by a white arrow . ( middle panel ) Unzipping in the direction in which the bound Gal4DBD was encountered first . ( bottom panel ) Unzipping in the direction in which the nucleosome was encountered first . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00310 . 7554/eLife . 06249 . 004Figure 1—figure supplement 1 . Unzipping experimental configuration . The DNA template was attached , at one end , to the surface of a glass coverslip via a digoxigenin–antidigoxigenin linkage , and at its other end to a microsphere via a biotin–streptavidin linkage . As the coverslip was moved away from the trapped microsphere , using a loading-rate clamp , the dsDNA was sequentially converted into ssDNA upon base pair separation . The presence of force peaks above the naked DNA baseline revealed the detected locations of protein–DNA interactions . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00410 . 7554/eLife . 06249 . 005Figure 1—figure supplement 2 . Characterization of the precision and accuracy of detection of the locations of Gal4DBD and nucleosome . Single molecule unzipping detected Gal4DBD and a nucleosome simultaneously . The histograms for detected locations of Gal4DBD ( green ) and nucleosome ( red ) were obtained by pooling data from multiple single molecule traces , with the expected bound locations represented by their respective dashed lines . For each histogram , the precision was determined by the standard deviation of each histogram , and the accuracy by the difference between the mean of the histogram and the expected value ( the vertical dashed line ) . These data demonstrate both the precision and accuracy to be near base-pair . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00510 . 7554/eLife . 06249 . 006Figure 1—figure supplement 3 . Characterization of Gal4DBD binding . To characterize Gal4DBD binding to its binding sequence , DNA unzipping was carried out in the presence of known Gal4DBD concentrations . ( A ) A representative unzipping trace of a bound Gal4DBD . The location of the binding sequence is shown as a shaded region . Naked DNA unzipping baseline is shown in gray . ( B ) Fraction of bound Gal4DBD vs the concentration of Gal4DBD . For a given concentration of Gal4DBD , measurements were on multiple DNA molecules to obtain the fraction of Gal4DBD . Data points are represented as ( mean ± s . e . m . ) . The relation for the fraction bound vs [Gal4DBD] was fit to:[Gal4DBD][Gal4DBD]+Kd ( red smooth curve ) , which yielded the dissociation equilibrium constant Kd= 3 . 4 nM . ( C ) Fraction of bound Gal4DBD vs time . This relation shows no significant Gal4DBD dissociation from its binding sequence over a course of one hour . Data were fit to a straight line to guide the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00610 . 7554/eLife . 06249 . 007Figure 1—figure supplement 4 . Detection of Gal4DBD binding . The presence of a bound Gal4DBD was determined by the magnitude of the force peak at the Gal4 binding sequence . In the presence of a bound Gal4DBD , the peak force increased substantially and was readily differentiable from the baseline DNA force . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 007 These unzipping experiments also revealed tight binding of Gal4DBD to its recognition sequence and slow dissociation . Under our experimental conditions , the equilibrium dissociation constant of Gal4DBD was determined to be 3 . 4 nM ( Figure 1—figure supplement 3b ) . Our experiments were carried out with 95% of Gal4 sites bound to Gal4DBD . In addition , the bound Gal4DBD's lifetime was much longer than 1 hr ( the typical duration of a single molecule experiment ) ( Figure 1—figure supplement 3c ) . For all experiments involving Gal4DBD , including those in Figure 1 , Gal4DBD was allowed to equilibrate with the DNA , and remaining free Gal4DBD was then flushed from the sample chamber . Thus , subsequent remodeling reactions were carried out without free Gal4DBD in solution . Chromatin remodelers utilize ATP hydrolysis to move nucleosomes by altering histone–DNA interactions , with the two major families of chromatin remodelers , ISWI and SWI/SNF , differing in their outcomes of the remodeling reaction ( Clapier and Cairns , 2009 ) . SWI/SNF family remodelers are known to be associated with nucleosome disruption ( Imbalzano et al . , 1996; Logie and Peterson , 1997; Aoyagi et al . , 2002 ) and transcriptional activation ( Kwon et al . , 1994; Hassan et al . , 2001; Gkikopoulos et al . , 2011 ) ; while ISWI family remodelers have been shown to contribute to the formation of evenly spaced nucleosome arrays ( Tsukiyama et al . , 1999; Fyodorov and Kadonaga , 2002; Lusser et al . , 2005; Torigoe et al . , 2011 ) . Despite these opposing characteristics , both remodeler families have been implicated in transcriptional activation and repression ( Kwon et al . , 1994; Moreira and Holmberg , 1999; Hassan et al . , 2001; Martens and Winston , 2002; Whitehouse et al . , 2007; Yadon et al . , 2010 ) . On mononucleosome substrates , many ISWI remodelers have been shown to be sensitive to naked DNA segments flanking the nucleosome , preferentially sliding the nucleosome towards the longer segment of DNA ( Yang et al . , 2006; Blosser et al . , 2009; Deindl et al . , 2013 ) . This sensitivity to linker DNA is believed to underlie their ability to generate evenly spaced nucleosomal arrays ( Gelbart et al . , 2001; Stockdale et al . , 2006 ) . SWI/SNF remodelers , on the other hand , can shift a histone octamer up to 50 bp off the end of a short DNA fragment ( Kassabov et al . , 2003 ) . On dinucleosomal templates , SWI/SNF remodelers have been found to shift one nucleosome onto another , indicating nucleosome disruption and eviction characteristics of these remodelers ( Engeholm et al . , 2009; Dechassa et al . , 2010 ) . Here , we employed ySWI/SNF and yISW1a as model systems to study how a bound Gal4DBD may affect the remodeling of an adjacent nucleosome . First , we investigated the initial direction of nucleosome remodeling in the presence of a bound Gal4DBD in close proximity . This was achieved by limiting the remodeling reaction to the first remodeling event which we define as a single round of remodeler binding , nucleosome remodeling , and remodeler detachment from the nucleosomal DNA ( Shundrovsky et al . , 2006 ) . We engineered a DNA template in which a Gal4 binding sequence and a 601NPE were separated by 10 bp . The DNA template containing a positioned nucleosome was then remodeled , by either SWI/SNF or ISW1a , for a short period of time ( ∼1 min ) , with or without the addition of Gal4DBD ( Figure 2; Figure 2—figure supplement 1; ‘Materials and methods’ ) . During such a short remodeling time , ∼56% of nucleosomes pooled from measurements of multiple single molecules were found to remain at the original location , suggesting a lack of remodeling ( Figure 2—figure supplement 2 ) . Of the remaining ∼45% of the nucleosomes that were remodeled , we estimate that ∼73% were remodeled only once and ∼27% were remodeled more than once , using a method we previously established ( Shundrovsky et al . , 2006 ) . 10 . 7554/eLife . 06249 . 008Figure 2 . A bound Gal4DBD affects the directionality of SWI/SNF remodeling and ISW1a remodeling differently . Nucleosomes were remodeled by either 1 nM ISW1a or 1 nM SWI/SNF with 1 mM ATP for 1 min , a time sufficiently short that the majority of nucleosomes were not remodeled ( Figure 2—figure supplement 2 ) . Each DNA template was subsequently unzipped . For templates used in ( D ) – ( G ) , the 601NPE was separated from the Gal4 binding sequence by 10 bp . ( A ) Distribution of the location of a nucleosome before remodeling . Data were pooled from measurements on multiple nucleosomal DNA molecules . ( B ) Distribution of the location of a nucleosome remodeled by ISW1a in the absence of Gal4DBD . ( C ) Distribution of the location of a nucleosome remodeled by SWI/SNF in the absence of Gal4DBD . ( D ) Distribution of the location of a nucleosome remodeled by ISW1a with a bound Gal4DBD initially located upstream of the 601NPE . ( E ) Distribution of the location of a nucleosome remodeled by SWI/SNF with a bound Gal4DBD initially located upstream of the 601NPE . ( F ) Distribution of the location of the nucleosome remodeled by ISW1a with a bound Gal4DBD initially located downstream of the 601NPE . ( G ) Distribution of the location of a nucleosome remodeled by SWI/SNF with a bound Gal4DBD initially located downstream of the 601NPE . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00810 . 7554/eLife . 06249 . 009Figure 2—source data 1 . Comparison of unzipping force signatures of a nucleosome before and after remodeling . We used unzipping to characterize the structure of a nucleosome before or after remodeling by either ISW1a or SWI/SNF , in the presence or absence of Gal4DBD . The structural features include the maximum force in the first force cluster , the maximum force in the second force cluster , the width of each cluster , and the distance between the two clusters . Errors show s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 00910 . 7554/eLife . 06249 . 010Figure 2—figure supplement 1 . Directionality of ISW1a and SWI/SNF remodeling of a nucleosome in close proximity to a bound Gal4DBD . The figure panels A to G show the raw traces for the corresponding panels in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01010 . 7554/eLife . 06249 . 011Figure 2—figure supplement 2 . Determination of fractions of remodeled nucleosome . ( A ) For each nucleosome distribution , remodeled by either ISW1a or SWI/SNF as shown in Figure 2B , C , was fit to a double Gaussian function . One Gaussian ( narrow green curve ) corresponds to the distribution of unremodeled nucleosomes and the other ( broader green curve ) to the distribution of remodeled nucleosomes . The sum of the two Gaussians is shown as the red curve . ( B ) A table summarizing fractions of unremodeled percentage and remodeled percentage for both ISW1a and SWI/SNF . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 011 In the absence of Gal4DBD , although the nucleosome unzipping signature did not appear to be altered after remodeling by either ISW1a or SWI/SNF ( Figure 2—figure supplement 1; Shundrovsky et al . , 2006 ) , the positions of the nucleosomes were spread out from the original location . Both ISW1a and SWI/SNF were able to move a nucleosome bi-directionally ( Figure 2B , C ) without inducing significant changes in the nucleosome structure ( Figure 2—source data 1 ) . The slight asymmetric distribution of the remodeled nucleosome was likely due to the non-palindromic feature of the 601 sequence ( Lowary and Widom , 1998 ) which leads to some asymmetry in the protein–DNA interactions at the two halves of a nucleosome ( Hall et al . , 2009 ) . The results from SWI/SNF remodeling were also consistent with those from an earlier study ( Shundrovsky et al . , 2006 ) . Interestingly , in the presence of Gal4DBD , ISW1a moved the nucleosome away from Gal4DBD ( Figure 2D ) , whereas SWI/SNF moved the nucleosome towards Gal4DBD ( Figure 2E ) . To determine whether such a differential behavior was a result of the DNA sequence used , we engineered another DNA template that was identical to this one , except that the Gal4 binding site was located on the other side of the 601NPE . After adding Gal4DBD , ISW1a again moved the nucleosome away from the Gal4DBD ( Figure 2F ) , while SWI/SNF again moved the nucleosome towards the Gal4DBD ( Figure 2G ) . These data rule out the possibility of a DNA sequence effect on the directionality of nucleosome movement by the two remodelers . Therefore , we conclude that the bound Gal4DBD affects the directionality of nucleosome movement by the two types of remodelers differently: away from the TF for ISW1a and toward the TF for SWI/SNF . Our findings on the TF directed SWI/SNF nucleosome remodeling are entirely novel; while our findings on the TF directed ISW1a nucleosome remodeling are in agreement with a previous study that used NURF ( a homolog of ISWI complexes in Drosophila ) in the presence of Gal4DBD ( Kang et al . , 2002 ) . Since ISW1a moved a nucleosome away from an adjacent Gal4DBD , the bound Gal4DBD may provide a barrier to ISW1a remodeling . To test this , we designed an unzipping template with a 601NPE at the end of the template and a Gal4 binding at a greater spacing ( 75 bp ) from the 601NPE ( Figure 3—figure supplement 1 ) . The use of an end-positioned nucleosome should dictate that the nucleosome movement could only take place towards a bound Gal4DBD . After ISW1a remodeling for 10 min , which was sufficiently long to allow for multiple rounds of remodeling , the distributions of the nucleosome location showed a significant difference between the absence and presence of Gal4DBD ( Figure 3A , B ) . In the absence of Gal4DBD , the nucleosome was moved away from the template end by several hundred base pairs towards the center region of the template , generating a rather broad distribution . In contrast , in the presence of Gal4DBD , although nucleosomes were still moved away from the end of the template , they were not able to pass the location of the Gal4DBD ( Figure 3B ) . Instead , the distribution peaked at the midpoint between the Gal4 binding sequence and the 601NPE . 10 . 7554/eLife . 06249 . 012Figure 3 . ISW1a remodeling is blocked by a bound Gal4DBD . Nucleosomes were remodeled by 1 nM ISW1a with 1 mM ATP for 10 min with or without Gal4DBD . Shaded regions indicate locations of Gal4 binding sequence and 601NPE . ( A ) Distributions of the locations of the nucleosome and bound Gal4DBD , either before remodeling or without Gal4DBD , as controls . ( B ) Distributions of the locations of the nucleosome after ISW1a remodeling in the presence of Gal4DBD on three different templates of increasing separation between the Gal4 binding sequence and the 601NPE . For each template , the nucleosome position distribution is dominated by a narrow population , but has a few outliers which were moved a much greater distance and some of which even passed the Gal4 binding sequence . These outliers ( ∼5% ) were likely a result of templates that did not have a bound Gal4DBD initially ( ∼5%; see main text ) . This is further supported by the observation that none of these outlier traces revealed a bound Gal4DBD . Nonetheless , in order to avoid possible bias , these nucleosome positions were still used for further analysis in ( C ) and thus contributed to the error bars in ( C ) . ( C ) Relationship between the distance the remodeled nucleosome moved and the separation between the Gal4 binding sequence and the 601NPE . Error bars are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01210 . 7554/eLife . 06249 . 013Figure 3—figure supplement 1 . Single molecule unzipping simultaneously detects Gal4DBD and an end-positioned nucleosome . Shown is a representative unzipping trace of a DNA molecule containing an end-positioned nucleosome and a bound Gal4DBD . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01310 . 7554/eLife . 06249 . 014Figure 3—figure supplement 2 . ISW1a centers a nucleosome between a bound Gal4DBD and the template end . This relation shows that ISW1a remodeler tends to center a nucleosome between the Gal4 binding site and the DNA end . Error bars are SEM . The dashed line indicates the center position on the DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 014 To further examine the relationship between the location of Gal4DBD and the ISW1a-remodeled nucleosome , we used two additional templates with shorter ( 24 bp and 50 bp ) distances between the Gal4 binding site and the 601NPE ( Figure 3B ) . After ISW1a remodeling in the presence of Gal4DBD , we found that the dyad locations of the remodeled nucleosomes always nearly centered between the bound Gal4DBD and the template end ( Figure 3C; Figure 3—figure supplement 2 ) . These results demonstrate that Gal4DBD is a physical barrier for ISW1a mediated nucleosome remodeling and that ISW1a is able to use Gal4DBD as a reference point to reposition a nucleosome . This novel finding is of particular relevance to in vivo nucleosome spacing , especially near transcription start and termination sites . Although previous studies have suggested a possible role for ISWI remodelers to space nucleosomes using bound TFs near these sites ( Pazin et al . , 1997; Gkikopoulos et al . , 2011; Yen et al . , 2012 ) , our results provide direct evidence that ISWI remodelers can indeed sense and respond to the presence of a DNA-bound protein such as a TF , which acts as barrier to dictate the placement of nucleosomes . In order to test whether a bound Gal4DBD is a physical barrier for SWI/SNF remodeling and what the fate of a bound Gal4DBD is upon nucleosome remodeling , we used the same template as the one that we used in single round experiments and performed 10 min remodeling on template both in the absence and in the presence of Gal4DBD . After a single round of nucleosome remodeling , the nucleosome will likely overlap the Gal4 binding site ( Figure 2 ) . However , the force signature of a bound Gal4DBD is subtle compared to that of a nucleosome ( Figure 1 ) , and thus the presence or the absence of a bound Gal4DBD cannot be definitively differentiated from a nucleosome by the unzipping force . Therefore , in order to determine whether a Gal4DBD was present after nucleosome remodeling , we allowed SWI/SNF to carry out multiple rounds of remodeling reactions to potentially reposition the nucleosome sufficiently far from the Gal4 binding site to allow for a definitive assay of the state of binding at the Gal4 binding site . For this experiment , we designed a long template ( ∼1200 bp ) with a 601NPE separated from a Gal4 binding sequence by 11 bp . The 601NPE was located near the center of a long DNA template to allow ample distance for possible bidirectional sliding of the nucleosome via multiple rounds of remodeling , such that the remodeled nucleosome and possible presence of Gal4DBD could be independently detected . After a 10 min remodeling by SWI/SNF , nucleosomes were repositioned from the center of the template to random locations along the entire sequence . Both in the absence and presence of Gal4DBD , remodeled nucleosomes were detected on both sides of the original Gal4 binding position ( Figure 4A , Figure 4—figure supplement 1 ) . This indicates that Gal4DBD is not a physical barrier for SWI/SNF mediated nucleosome remodeling . 10 . 7554/eLife . 06249 . 015Figure 4 . SWI/SNF remodeling evicts a bound Gal4DBD from its DNA template . Nucleosomes were remodeled by 1 . 5 nM SWI/SNF with 1 mM ATP for 10 min with or without Gal4DBD . Shaded regions indicate locations of Gal4 binding sequence and 601NPE . ( A ) Distributions of the locations of the nucleosome and bound Gal4DBD before remodeling ( upper plot ) , after remodeling without Gal4DBD ( middle plot ) , and after remodeling with Gal4DBD ( lower plot ) . ( B ) Representative traces in the case of before remodeling ( top plot; N = 55 ) and after remodeling ( middle and bottom plots; N = 50 ) . The middle plot shows an example trace where a nucleosome was remodeled to the opposite side of Gal4DBD relative to its original position; while the bottom plot shows an example trace where a nucleosome was remodeled to the same side of Gal4DBD relative to its original position . Gray traces were taken from the corresponding naked DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01510 . 7554/eLife . 06249 . 016Figure 4—source data 1 . SWI/SNF is unable to evict a bound Gal4DBD in the absence of a nucleosome or in the absence of ATP . To determine whether SWI/SNF with ATP alone is able to displace a bound Gal4DBD in the absence of a nucleosome , we carried out experiments on a DNA template preloaded with Gal4DBD but without a nucleosome in 1 . 5 nM SWI/SNF with 1 mM ATP for 10 min . DNA molecules were subsequently unzipped to determine the presence of Gal4DBD . The fraction of templates containing a bound Gal4DBD remained the same before and after the remodeling reaction , indicating that SWI/SNF with ATP alone is not able to displace a bound Gal4DBD . To rule out the possibility that Gal4DBD disruption was due to binding of SWI/SNF to DNA or the nucleosome and not due to nucleosome remodeling , we carried out a control experiment on a DNA template containing a bound Gal4DBD and a nucleosome by incubating the sample with 1 . 5 nM SWI/SNF for 10 min in the absence of ATP . We subsequently unzipped the DNA template to determine if Gal4DBD was still bound . The fraction of templates containing a bound Gal4DBD was comparable to that of a template without a nucleosome and without SWI/SNF and ATP added , indicating that in the absence of ATP , SWI/SNF is unable to evict a bound Gal4DBD even in the presence of a nucleosome adjacent to a bound Gal4DBD . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01610 . 7554/eLife . 06249 . 017Figure 4—figure supplement 1 . Distributions of the locations of SWI/SNF remodeled nucleosomes as determined by unzipping from both directions . To determine whether the distributions of the locations of remodeled nucleosomes were similar for measurements made by unzipping the DNA in one direction vs in the other direction , we unzipped multiple DNA molecules , each containing either an unremodeled or a remodeled nucleosome , from both directions . Nucleosome remodeling was carried out in 1 nM SWI/SNF with 1 mM ATP for 5 min . Our data show similar distributions for data obtained in both directions . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 01710 . 7554/eLife . 06249 . 018Figure 4—figure supplement 2 . Nucleosome remodeling by SWI/SNF on a template with the Gal4 binding site separated from the 601NPE by 24 bp . To test whether SWI/SNF is able to evict Gal4DBD via nucleosome remodeling when a bound Gal4DBD is located farther away from a nucleosome , we used a template where the Gal4 binding site was separated from the 601NPS by 24 bp and carried out unzipping experiments under identical conditions as those shown in Figure 4 . Out of all traces where the nucleosome was repositioned to the opposite side of the Gal4 binding site by SWI/SNF ( N = 13 ) , we did not detect any Gal4DBD binding signature on the template , indicating eviction of Gal4DBD . Shown are example traces , with arrows indicating the unzipping directions . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 018 What is the fate of the Gal4DBD after a nucleosome has been remodeled ? To answer this question , we allowed a nucleosome to be remodeled by SWI/SNF in the presence of Gal4DBD . We then analyzed each trace to determine whether a nucleosome was remodeled to the opposite side of the Gal4 sequence or to the same side of the Gal4 sequence , relative to the 601NPE . For the traces where nucleosomes were remodeled to the opposite side of the Gal4 sequence , we did not detect any Gal4DBD unzipping signature ( Figure 4B; Figure 1—figure supplement 4 ) . This indicates that SWI/SNF was able to move the nucleosome in such a way that the Gal4DBD was evicted from its binding sequence . For traces where a nucleosome was remodeled to the same side of the Gal4 sequence relative to the 601NPE ( Figure 4B ) , we also did not detect any Gal4DBD unzipping signature . This implies that these nucleosomes were likely first remodeled towards the bound Gal4DBD , as indicated by Figure 2 , resulting in the eviction of Gal4DBD . This was followed by subsequent remodeling events that moved nucleosomes to other positions . Control experiments SWI/SNF is unable to evict a bound Gal4DBD in the absence of a nucleosome or in the absence of ATP ( Figure 4—source data 1 ) . An increase in the separation between the Gal4 binding site and 601NPS to 24 bp still permits SWI/SNF remodeling to evict Gal4DBD ( Figure 4—figure supplement 2 ) . Therefore , we conclude that SWI/SNF remodeling is able to evict a bound TF and this eviction requires the presence of a nucleosome . Our findings provide the first direct evidence that SWI/SNF nucleosome remodeling is capable of evicting a bound TF . Previously , SWI/SNF was shown to move one nucleosome to invade and eventually disrupt an adjacent nucleosome ( Dechassa et al . , 2010 ) . Taken together with our findings , SWI/SNF nucleosome remodeling appears to be powerful machinery capable of actively overcoming and removing a variety of obstacles in its vicinity . To rule out the possibility that the interaction between a TF and a nucleosome demonstrated above is specific to Gal4DBD , we replaced Gal4 binding sequence with a Lac repressor binding sequence and repeated the above experiments using the Lac repressor . Because the Lac repressor is only found in prokaryotic cells and has no known relationship with any chromatin remodeler in eukaryotic cells , it can act as a biologically neutral bound protein . When ISW1a remodeled an end-positioned nucleosome on a template also containing a bound Lac repressor , the Lac repressor effectively dictated the position of the remodeled nucleosome , with ISW1a centering the nucleosome on the DNA with the Lac repressor acting as a barrier ( Figure 5A ) . In contrast , SWI/SNF was able to slide a nucleosome in either direction , displacing the bound Lac repressor ( Figure 5B ) . Therefore , we conclude that the mechanism of TF regulation by nucleosome remodeling is likely general without any specificity to a particular TF . 10 . 7554/eLife . 06249 . 019Figure 5 . ISW1a remodeling is blocked by Lac repressor , while SWI/SNF remodeling evicts Lac repressor from the template . Shaded regions indicate locations of Lac repressor binding sequence and 601NPE . ( A ) Distributions of locations of nucleosomes before remodeling ( upper plot ) , after remodeling by ISW1a without Lac repressor ( middle plot ) , and after remodeling by ISW1a with Lac repressor ( lower plot ) . Lac repressor binding sequence was separated from the 601NPE by 50 bp . Nucleosomes remodeling was carried out in 1 nM ISW1a with 1 mM ATP for 10 min with or without Lac repressor . ( B ) Representative traces in the case of before SWI/SNF remodeling ( top plot; N = 25 ) and after remodeling ( middle and bottom plots; N = 27 ) . Lac repressor binding site was separated from the 601NPE by 10 bp . Nucleosomes were remodeled by 1 . 5 nM SWI/SNF with 1 mM ATP for 10 min . The middle plot shows an example trace where a nucleosome was remodeled to the other side of the Lac repressor and the bottom plot shows an example trace where a nucleosome was remodeled to the same side of Lac repressor . DOI: http://dx . doi . org/10 . 7554/eLife . 06249 . 019 Our results showed opposite directionality for nucleosome positioning when a nucleosome in close proximity to a TF was remodeled by ISW1a and SWI/SNF ( Figure 2 ) . Biochemical studies and the crystal structure of ISW1a indicate that the DNA-binding domain of ISW1a binds to ∼29 bp of the extranucleosomal DNA , which has been proposed to act as an anchor to pull the nucleosome towards ISW1a ( Gangaraju and Bartholomew , 2007; Hauk and Bowman , 2011; Yamada et al . , 2011; Hota et al . , 2013 ) . Our finding that ISW1a moves nucleosome away from a TF adjacent to the nucleosome is consistent with an important role of the DNA-binding domain in engaging fully accessible DNA immediately flanking the nucleosome . In contrast , while it has been widely acknowledged that SWI/SNF does not require extra-nucleosomal DNA binding to remodel a nucleosome , cryo-EM and DNA-crosslinking experiments have shown that the Snf6 subunit of SWI/SNF binds to ∼15 bp of the extra-nucleosomal DNA and the rest of the SWI/SNF slides the nucleosome away from where the Snf6 subunit binds ( Dechassa et al . , 2008 ) . Although the Snf6 subunit has not been shown to be essential for remodeling , it has DNA-binding affinity ( Sengupta et al . , 2001; Dechassa et al . , 2008 ) . We speculate that it may help to orient SWI/SNF binding on the nucleosome . In the presence of a barrier adjacent to a nucleosome , Snf6 may prefer to bind to the side of the nucleosome with more linker DNA and thus orient SWI/SNF to slide a nucleosome towards the TF . Our study of ISW1a remodeling demonstrates that Gal4DBD is an effective barrier for ISW1a-mediated nucleosome remodeling and the ISW1a is able to use Gal4DBD as a reference point to reposition nucleosomes ( Figure 3 ) . These results have significant implications for the mechanism of nucleosome spacing in vivo . Genome-wide nucleosome mapping in budding yeast revealed that deletion of ISWI in yeast disrupts nucleosome spacing ( Gkikopoulos et al . , 2011 ) , suggesting that ISW1 remodelers are key players in generating evenly distributed nucleosomal arrays . In addition , several recent studies have shown that certain DNA-binding factors located at the promoter region are also responsible for nucleosome positioning ( Whitehouse et al . , 2007; Yadon et al . , 2010; Zhang et al . , 2010; Bai et al . , 2011; Hughes et al . , 2012; Yen et al . , 2012; Parikh and Kim , 2013; Struhl and Segal , 2013; Lieleg et al . , 2014 ) . Our finding that a bound Gal4DBD is a barrier to ISW1a now provides direct evidence to illustrate that ISW1a can potentially use a TF around the promoter region as a reference point to evenly position nucleosomes into the gene body . Our study shows that TF eviction is an intrinsic property of SWI/SNF remodeling and is independent of SWI/SNF recruitment ( Figures 4 and 5 ) . It has been previously shown that SWI/SNF recruitment by the glucocorticoid receptor ( GR ) induced histone loss in nucleosomes and this was immediately followed by GR and SWI/SNF eviction from the template ( Nagaich et al . , 2004 ) . Our current work demonstrates that , in the absence of remodeler recruitment , TF eviction via nucleosome remodeling can take place without substantial nucleosome loss . Although SWI/SNF can translocate along naked DNA ( Lia et al . , 2006; Zhang et al . , 2006; Sirinakis et al . , 2011 ) , raising the possibility for TF eviction solely by SWI/SNF , we found that in the absence of a nucleosome , SWI/SNF did not displace Gal4DBD from its binding site ( Figure 4—source data 1 ) . Thus , SWI/SNF translocation alone is insufficient to displace a bound Gal4DBD and TF eviction requires nucleosome remodeling . Previous work by Owen-Hughes and coworkers ( Lia et al . , 2006 ) found that translocation by RSC was highly sensitive to a force in the DNA . Therefore , although SWI/SNF is known to translocate along naked DNA , it may have limited ability in dealing with a road block , such as a bound protein . It is also possible that SWI/SNF is unable to efficiently locate a bound protein in the absence of a nucleosome . We speculate that TF removal may be accelerated once a nucleosome is repositioned over the bound TF . Indeed , a recent single molecule fluorescence study of Gal4 binding kinetics on nucleosomal DNA showed that a nucleosome regulates Gal4 binding not only by preventing Gal4 binding , but also by dramatically increasing the Gal 4 dissociation rate from the DNA ( Luo et al . , 2014 ) . SWI/SNF family remodelers are known to be involved in transcriptional activation . Genome-wide mapping of yeast indicates that , apart from localizing to nucleosomes around transcription start sites , SWI/SNF family remodelers are also enriched upstream of the promoter regions ( Yen et al . , 2012 ) . Genome-wide analysis of the locations of human chromatin remodelers also found that Brg1 , Chd4 , and Snf2h are highly enriched at the promoter and distal upstream regions ( Morris et al . , 2014 ) . Since many relevant transcriptional modulators , such as enhancers ( Ren , 2010 ) and insulators ( Bell et al . , 2001 ) , are located further upstream of promoters , SWI/SNF family remodelers could move promoter nucleosomes to dynamically regulate these factors . Thus , although SWI/SNF alone does not possess any ability to remove TFs on its own , our work shows that SWI/SNF can slide nucleosomes to displace neighboring TFs around the promoter region , providing a mechanistic basis for dynamically clearing both nucleosomes and other bound factors upon SWI/SNF recruitment ( Nagaich et al . , 2004 ) . The plasmids containing the Gal4 binding site and the 601NPE with varied distances were prepared using standard PCR and cloning methods . The cloning segments were generated by standard PCR from the 601 plasmid ( Lowary and Widom , 1998 ) using special primers , one of which contains one Gal4 binding site . The distance between the primer containing the Gal4 binding site and the 601NPE determines the distance between the Gal4 binding site and 601NPE . Then , the PCR product was cloned into the pDrive vector ( Qiagen , Valencia , CA ) . The finished constructs were confirmed by DNA sequencing . Nucleosomal DNA templates were prepared using methods similar to those previously described ( Koch et al . , 2002; Li and Wang , 2012 ) . Briefly , each DNA construct consisted of two separate segments . A ∼1 . 1 kbp anchoring segment was amplified , by PCR , from plasmid pRL574 using a digoxigenin-labeled primer and then subsequently digested with BstXI ( NEB , Ipswich , MA ) to produce an overhang . The unzipping templates were amplified , by PCR , from the plasmids described above and amplified with a biotin-labeled primer , digested with BstXI , and dephosphorylated using CIP ( NEB , Ipswich , MA ) to introduce a nick into the final DNA template . Nucleosomes were assembled from purified HeLa histones onto the unzipping fragment by a well-established salt dialysis method ( Lee and Narlikar , 2001 ) . The two segments were joined by ligation immediately prior to use . This produced a complete template labeled with a single dig tag on one end and a biotin tag located 7 bp after the nick in one DNA strand . yISW1a and ySWI/SNF were purified as previously described ( Gangaraju and Bartholomew , 2007; Dechassa et al . , 2008 ) . yGal4DBD was purchased from Santa Cruz Biotechnology , Inc . , Dallas , TX . After the ligation of the anchoring segment and unzipping segment containing a nucleosome , we incubated 20 nM of the nucleosomal DNA with 200 nM Gal4DBD at 16°C for 30 min . Single molecule sample preparation was performed according to protocols previously described ( Li and Wang , 2012 ) . The remodeling experiments were conducted in a sample chamber after the DNA tethers are formed . SWI/SNF remodeling reactions contained 1 . 5 nM purified ySWI/SNF , and 1 mM ATP in the SWI/SNF remodeling buffer ( 10 mM Tris⋅Cl , pH 8 . 0 , 100 mM NaCl , 7 mM MgCl2 , 2 mM DTT , 0 . 1 mg/ml acBSA ) . ISW1a remodeling reactions contained 1 . 5 nM purified yISW1a , and 1 mM ATP in the ISW1a remodeling buffer ( 30 mM HEPES , pH 7 . 6 , 3 mM MgCl2 , 5 mM NaCl , 0 . 1 mM EGTA , 0 . 02 mM EDTA , 5% glycerol , 0 . 2 mg/ml acBSA ) . Both types of remodeling reactions were incubated at 25°C with duration specified . The reactions were stopped by the addition of 10 mM EDTA and 0 . 25 mg/ml Salmon Sperm DNA and incubation for 5 min at 25°C . Finally , the sample chamber was rinsed with 100 μl sample buffer ( 10 mM Tris⋅Cl pH 7 . 5 , 1 mM EDTA , 100 mM NaCl , 1 . 5 mM MgCl2 , 1 mM DTT , 3% ( vol/vol ) glycerol , 0 . 02% ( vol/vol ) Tween 20 , and 2 mg/ml BSA ) . Single molecule unzipping measurements were subsequently performed in this sample buffer . An optical trapping setup as previously described ( Brower-Toland and Wang , 2004 ) was used to unzip a single DNA molecule by moving the microscope coverslip horizontally away from an optical trap . The unzipping methods have been previously described ( Li and Wang , 2012 ) and briefly summarized here . Whenever the unzipping fork encountered an interaction that prevented the fork progression , the unzipping force was ramped up linearly with time ( 15 pN/s ) until the interaction was disrupted . When two interactions occurred in close vicinity , upon the disruption of the first interaction the force was unable to relax back to the baseline before being ramped up again for the second interaction , subjecting this subsequent interaction to a higher initial force . Therefore , for each region of interactions , the dwell time histogram highlighted the edge of the region first encountered . Another feature of this method was the display of the distinctive force signature for a nucleosome , allowing for robust identification of the nucleosome structure . Data were low pass filtered to 5 kHz , digitized at ∼12 kHz , and later low pass filtered to 60 Hz . The precision and accuracy of the experimental curves were improved to near base pair level by cross-correlation of regions immediately before the Gal4DBD disruption and after the nucleosome disruption , using methods as previously described ( Hall et al . , 2009; Li and Wang , 2012 ) . For the experimental curves where the nucleosomes are located at the end of the template , the cross-correlation was carried out for a region immediately before the Gal4DBD disruption or nucleosome disruption . To account for minor instrumental drift , trapping bead size variations , and DNA linker variations , the alignment allowed for a small additive shift ( ∼10 bp ) and multiplicative linear stretch ( <2% ) using algorithms similar to those previously described ( Hall et al . , 2009 ) . Gal4DBD showed a distinct unzipping signature with a single force peak at 8 bp from the center of the consensus sequence ( Figure 1—figure supplement 2 and Figure 1—figure supplement 3a ) , indicating the front end of the Gal4DBD footprint on the DNA . The disruption force peak was 18–20 pN , significantly larger than the baseline force of ∼15 pN . Therefore , we determined the center position of a bound Gal4DBD by first detecting the peak force location and then shifting this location by 8 bp in the direction of unzipping . The positioned nucleosome displayed a much more complex force signature with multiple force peaks and a significantly greater overall force , reflecting the multiple finer and stronger histone–DNA interactions within a nucleosome ( Shundrovsky et al . , 2006; Hall et al . , 2009 ) . We determine the dyad position of a nucleosome by first measuring mean force location within the first force cluster and then shifting this position by 43 bp in the direction of the unzipping ( Figure 1—figure supplement 2 ) .	Cells contain thousands of genes that are encoded by molecules of DNA . In yeast and other eukaryotic organisms , this DNA is wrapped around proteins called histones to make structures called nucleosomes . This compacts the DNA and allows it to fit inside the tiny nucleus within the cell . The positioning of the nucleosomes influences how tightly packed the DNA is , which in turn influences the activity of genes . Less active genes tend to be found within regions of DNA that are tightly packed , while more active genes are found in less tightly packed regions . To activate a gene , proteins called transcription factors bind to a section of DNA within the gene called the promoter . Enzymes known as ‘chromatin remodelers’ can alter the locations of nucleosomes on DNA to allow the transcription factors access to the promoters of particular genes . In yeast , the SWI/SNF family of chromatin remodelers can disassemble nucleosomes to promote gene activity , while the ISW1 family organises nucleosomes into closely spaced groups to repress gene activity . However , it is not clear if , or how , chromatin remodelers can influence transcription factors that are already bound to DNA . Here , Li et al . studied the interactions between a transcription factor and the chromatin remodelers in yeast . The experiment used a piece of DNA that contained a bound transcription factor and a single nucleosome . Li et al . used a technique called ‘single molecule DNA unzipping’ , which enabled them to precisely locate the position of the nucleosome and transcription factor before and after the nucleosome was remodeled . The experiments found that a chromatin remodeler called ISW1a moved the nucleosome away from the transcription factor , while a SWI/SNF chromatin remodeler moved the nucleosome towards it . Significantly , Li et al . also found that a transcription factor is a major barrier to ISW1a's remodeling activity , suggesting that ISW1a may use transcription factors as reference points to position nucleosomes . In contrast , SWI/SNF was able to slide a nucleosome past the transcription factor , which led to the transcription factor falling off the DNA . Therefore , SWI/SNF is able to move transcription factors out of the way to deactivate genes . Li et al . propose a new model for how chromatin remodelers can move nucleosomes and regulate transcription factors to alter gene activity . A future challenge will be to observe these types of activities in living cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	Dynamic regulation of transcription factors by nucleosome remodeling

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.