Experimental Neurology 261 (2014) 494–500
Contents lists available at ScienceDirect
Experimental Neurology journal homepage: www.elsevier.com/locate/yexnr
Commentary
Neural stem cells in models of spinal cord injury Mark H. Tuszynski a,b,⁎, Yaozhi Wang a, Lori Graham a, Karla McHale a, Mingyong Gao a, Di Wu a, John Brock a, Armin Blesch c, Ephron S. Rosenzweig a, Leif A. Havtond, Binhai Zheng a, James M. Conner a, Martin Marsala e, Paul Lu a,b a
Dept. of Neurosciences, University of California — San Diego, La Jolla, CA, USA Veterans Administration Medical Center, San Diego, CA, USA Spinal Cord Injury Center, University Hospital Heidelberg, 69118 Heidelberg, Germany d Dept. of Neurology, University of California — Irvine, Irvine, CA, USA e Dept. of Anesthesiology, University of California — San Diego, La Jolla, CA, USA b c
a r t i c l e
i n f o
Article history: Received 27 May 2014 Revised 9 July 2014 Accepted 22 July 2014 Available online 28 July 2014
a b s t r a c t Replication of published studies is an important and respected aspect of the conduct of science. Most would argue that the interpretation of “negative” outcomes is still more challenging than the interpretation of “positive” findings, however, due to uncertainty in knowing precisely why a hypothesized outcome was not observed: in particular, are “negative” findings in replication studies a result of invalidity of the original experimental hypothesis, or due to a methodological failure, insensitivity of the applied instruments of analysis, or other factors? These points must be carefully considered. Steward and colleagues report findings of a study in which multipotent neural progenitor cells were grafted to sites of T3 complete transection. Unlike our study, cells failed to fill the lesion site, leaving collagenous rifts between rostral and caudal graft components. This “anatomical” failure precluded formation of neural relays across the lesion site, and was predictably associated with a failure to detect functional improvement. In summarizing outcomes of the study, Steward and colleagues did not clearly link the failure to achieve graft continuity in the lesion cavity with functional outcomes, despite the central role of this observation in cogently interpreting results of the replication study. In addition, the authors stated that they failed to replicate our report of “extensive” host axonal regeneration into grafts, but we did not report “extensive” host anatomical regeneration; moreover, underexposed images may have contributed to Steward's underestimation of host axonal penetration. The authors also stated that our original study excluded some animals from functional analysis, and this is incorrect. While replication studies are important and necessary, this particular report contained several errors, and the failure to form a continuous neural progenitor cell bridge across the lesion site limited the ability to conclude whether continuous grafts can restore function. In subsequent experiments we too have observed rift formation in animals grafted at long delays (N2 weeks) after SCI, and we confirm that animals with rifts do not exhibit functional improvement; we are developing methods to remove or prevent rift formation. The replication study confirmed the cardinal finding of our original report: that early-stage neural precursors extend very large numbers of axons over remarkably long distances through the lesioned adult spinal cord. Published by Elsevier Inc.
The ability to replicate published reports is essential to the practice of science, and we wholeheartedly support such efforts. A confirmatory replication provides the field with greater confidence that a particular avenue of research will prove informative. A failure to replicate is more complex. Failure to replicate leaves the first impression that the original findings were invalid, a potential case of Type I error in which the null hypothesis has been incorrectly rejected. Type I errors can result from chance, from underpowering of a study, from misinterpretation of the data, or more
⁎ Corresponding author at: Dept. Neurosciences, 0626, University of California, San Diego, La Jolla, CA 92093, USA. Fax: +1 858 534 5220. E-mail address: [email protected] (M.H. Tuszynski).
http://dx.doi.org/10.1016/j.expneurol.2014.07.011 0014-4886/Published by Elsevier Inc.
ominously, due to intentional or unintentional experimental bias or outright fraud. These are serious considerations. A failure to replicate can also result from other causes. Most important among these is that the conditions of the original study might not have been reproduced. If the original study methods were not used, or if the attempt to use the original study methods fails to generate the expected change in the dependent variable that is required to go on and test the hypothesis, then one is unable to draw a clear conclusion. One cannot in this case draw the conclusion that the study failed replication; one can only state that the necessary conditions to investigate the hypothesis could not be established. To provide an example, imagine that an expedition claims to have discovered a unique species of great ape in the Rwandan highlands. In
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
an effort to confirm this discovery, a second group sets off, but finds that a key bridge they must traverse to reach the site has been destroyed. They attempt to build a new bridge, but are unsuccessful; they are unable to reach the region to confirm or deny the report. Clearly, they cannot conclude that the species does not exist. They can only say that their effort was unsuccessful because they were unable to build the bridge. Moving from an example to the current case, the premise of our report in Cell (Lu et al., 2012) was that neural progenitor cells grafts can fill a severe spinal cord lesion site, enabling the formation of neural relays across the lesion site that support electrophysiological and functional improvement. Steward and colleagues, in an effort to replicate the study (Sharp et al., 2014), failed to achieve consistent filling of the spinal cord transection site. Not surprisingly, they subsequently failed to observe formation of functional relays across the lesion and functional improvement. But the failure to fill the lesion site eliminated the biological mechanism for testing the hypothesis that functional relays could be formed to support functional recovery. Accordingly, the study should either have been repeated with renewed effort to fill the lesion, or the report should have been limited to stating that the reported axonal outgrowth was replicated, but that the lack of cell fill precluded meaningful analysis of functional outcomes. There were other problems with the study, including: 1) a stated inability to communicate clearly with the first author, 2) statements that “extensive” axonal regeneration from host to graft were not replicated, although we did not report that host axonal regeneration was “extensive”, and 3) the statement that animals in our original study were “selected” for presentation of functional results, a statement that is untrue. Findings of replication studies should be presented in a thoughtful manner that appropriately considers and balances the numerous factors that can determine experimental outcomes. We provide detailed examples below that address some of these points. Specific points Gaps in grafts The authors of the replication study state (Sharp et al., 2014): “In the original study, two different methods were used for grafting. This was not reported in their Methods section, but was communicated to us by the original author.” This is inaccurate. Our original report used a single grafting method that consisted of dural penetration by a pulled glass micropipette and injection of cells into closed, established lesion cavities. This method corresponds to “Method 2” reported by Steward and colleagues. The first author of this response letter has used an open grafting method in non-human primate models, wherein the established spinal cord lesion is re-opened and accumulated cellular debris and “scar” are removed to provide an open and continuous grafting cavity (“Method 1” in the Steward paper). Around the time of the Steward replication we had begun to assess whether the open grafting method could be brought to the rat model in order to eliminate fibrous scar formation that occurred when rats were grafted at prolonged time points after injury (3 months post-lesion). The Steward replication experiment included this method, but the original study (Lu et al, 2012) did not. Steward and colleagues were mistaken in believing that the method had been used in our original study. There were several other errors in their report that are cited below. Our original report offered evidence that grafts that fill a spinal cord lesion site form a new neural relay across the complete transection site (Lu et al., 2012). Host axons regenerated into the graft (but not beyond the graft) and formed synapses; grafted neurons extended very large numbers of axons out of the lesion site and for long distances in the
495
host spinal cord beyond. Graft-derived axons formed synapses with host neurons in the spinal cord caudal (and rostral) to the lesion site. Functional improvement on the BBB motor scale occurred over several weeks. When the host spinal cord was stimulated three spinal segments rostral to the T3 complete transection/graft site, evoked responses were detected in the host spinal cord three spinal segments caudal to the lesion site. To determine whether host axons were specifically mediating either the functional improvement or neural activity recorded below the lesion, animals were re-lesioned in the host spinal cord immediately rostral to the lesion site. The re-lesion entirely abolished evoked responses below the lesion and eliminated the behavioral improvement. We presented these findings as indicative of evidence of formation of novel neural relays formed across the lesion site (Lu et al., 2012). The conceptual basis of this experiment is that novel neural relays are formed. An effort to replicate the experiment must generate experimental conditions in which a graft is placed that fills the lesion site sufficiently well that host axons can enter the graft and connect with a graft neuron, which in turn (perhaps through intermediary neurons) extends an axon outside the graft. The replication effort of Steward and colleagues failed to generate these basic conditions to enable testing of the hypothesis. That is, animals had non-neural tissue rifts that effectively segregated the graft into rostral and caudal components across which axons did not extend. The rifts in grafts of the Steward replication are under-appreciated because grafts are too darkly stained. The first author of this letter requested that the authors re-stain their material to generate normally rather than over-intensely labeled tissue, but this was not done. To illustrate the impact of this tissue processing issue, we used Adobe Photoshop software to modify only contrast and brightness in the published images from the authors' article (Fig. 1). When overexposure is reduced, one can see quite clearly that both of the authors' “two best” grafts (see their Table VI) lack rostral-to-caudal continuity (Fig. 1). Rostral-to-caudal continuity is present over only about 10% (estimated) of the proximal host/graft interface and 7% of the distal interface in their animal #3. Moreover, these interfaces were segregated from one another within the graft by a large central cavity (Fig. 1). Their second best example, animal #18, exhibited 43% integration in the proximal interface and 20% in the distal interface, but once again these regions were poorly aligned with one another and would be unlikely to provide a rostral-to-caudal pathway for bridging of the lesion site by a neural relay (Fig. 1). After these two “best” animals, the authors state that other animals had yet worse graft integration (Sharp et al., 2014). These conditions fail to generate the essential conditions for determining whether the experimental findings of neural relay formation with functional improvement can be replicated. In our view, it would have been appropriate to either repeat grafting to make a continuous graft bridge, or else state that the methods used in this experiment were unable to reconstitute a bridge, and that the ability to interpret functional outcomes was accordingly limited. The authors could then have focused on the more relevant issue of the rift formation itself, which is indeed an important consideration for further development of these methods. Why did rifts form in these grafts, unlike the study of Lu et al. (2012)? Dr. Lu performed the grafting at UC Irvine, using the same methods (in half of the grafted animals) that were used in his original report. We cannot identify with certainty the reason that rifts formed in the attempted replication, but several points may be relevant. First, we too observe rifts in our own laboratory when animals are grafted at delays longer than two weeks after the original transection. These rifts consist of collagen, and clearly divide the graft into rostral and caudal compartments and axons do not cross between these sections (Fig. 2). We do not observe functional recovery when rifts form. Rifts appear to originate from the pia/arachnoid membranes overlying the lesion, and do not form when a spinal cord lesion is more ventrally placed and is closed to the overlying dura. Second, grafts were prepared by a different individual in our laboratory than in the original Lu et al.
496
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Fig. 1. Graft integration. Shown are the two best examples (according to Steward and colleagues) of graft integration from the Steward paper. (A) Original panel from the Steward paper with very dark GFP labeling of graft in animal #3. Although a large cavity in the center of the graft is evident, poor integration of the graft with the host is obscured by the dark labeling. Horizontal section, rostral is toward the left. (B) Lightening the authors' original image in Adobe Photoshop, only one small region of continuity of the graft with the rostral region of the graft/host is visible, indicated by the arrowhead. Two white lines indicate the region of continuity, shown at higher magnification in C. Thus, there is a very small region from which host axons might penetrate the central region of the graft. Toward the caudal end of the graft/host interface, two very small regions of continuity are indicated between parallel white lines. These are shown at higher magnification in D. (C) Toward the rostral host/graft interface, the sole region of continuity from host into the core of the graft is indicated by the arrowhead and between the two parallel lines. Arrows point to areas of discontinuity that were obscured in the darkly labeled section of the original manuscript figure. (D) Similarly, the caudal portion of the graft shows small regions of continuity of the graft core to its caudal aspects between parallel white lines. White arrows indicate regions of discontinuity. Overall, the graft in animal #3 shows very poor graft continuity across the lesion site. (E) The authors' original report also cited animal #18 as one of two animals with the best graft continuity. However, this section, like others, was darkly labeled. (F) Lightening of original panel from the Steward paper shows limited regions of graft integration with the host. Potential graft continuity with the rostral host spinal cord (between parallel white lines) amounts to 43% of the available area, and on the caudal end of graft is 20%. However, the rostral and caudal regions of graft continuity with the host are not aligned: the rostral region of continuity is central, whereas the caudal region is lateral. Thus, there would be little rostral-to-caudal parenchymal connectivity through the graft. Arrows indicate regions of non-continuity. Scale bars cannot be provided because they were omitted in the Steward manuscript (their Fig. 3; (Sharp et al., 2014)).
(2012) report, and there may have been differences in tissue harvesting and preparation that predisposed to rift formation or incomplete graft filling. Dr. Lu is the most experienced individual in preparing these
grafts, but he did not prepare grafts for the replication because he was already at UC Irvine preparing for surgery. Third, the grafted cells remained immobile in solution for a longer time period in the Steward
Fig. 2. Fibrous septation in chronic grafts. When we implant E14 spinal cord-derived multipotent neural progenitor cells to sites of T3 complete transection at a delay of three months after the original injury, we also observe formation of fibrous rifts (arrows) in the lesion site that separate the proximal and distal aspects of the spinal cord. Functional recovery is not observed in our animals with rift formation across the lesion site. Scale bar 0.75 mm (image courtesy of K. Kadoya, UCSD).
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
497
replication due to the need to complete all cell harvest prior to grafting, then transport cells to Irvine from San Diego (adding experimental time of several hours). Cell viability was not tested or reported by Steward and colleagues at the conclusion of animal grafting, an oversight that leads to a lack of data for determining whether cell viability was adversely affected. Fourth, the grafting solution consisting of fibrinogen/thrombin matrix and growth factors was prepared at UC Irvine, unlike our original experiment. Fifth, the operating environment was different from UCSD and there may be introduction of unknown differences that could impact engraftment technique and graft survival/ integration.
Host axonal regeneration into grafts Steward and colleagues state (Sharp et al., 2014): “Although BDA-labeled [reticulospinal] axons were numerous rostral to the lesion, few extended past the host-graft boundary.” “The fact that a few reticulospinal axons enter the graft and form terminal arbors confirms the findings of Lu et al., although on balance, it is the lack of extension into the graft that is more striking.” “We did not confirm extensive ingrowth of host axons into the graft…” We did not report extensive host axonal ingrowth; the Lu et al. (2012) paper simply stated that host axons penetrated neural stem cell grafts, and sample images were provided. We have subsequently replicated our findings and demonstrated host axons forming synapses onto grafted neurons using ultrastructural methods. Steward and colleagues presented underexposed images to support their claim that few host axons penetrated grafts. Although the exposure times they employed are useful for preventing pixel saturation and examining subtle differences in levels of fluorescence, such underexposed images are distinctly suboptimal for assessing the presence of fine structures. Taking the authors' published images and adjusting only brightness and contrast settings in Adobe Photoshop to compensate for low exposure, more axons become apparent (Fig. 3). Had the authors examined the material at higher magnification and acquired images under improved fluorescent intensity, the presence of more axons might have been appreciated. The number of structures meeting morphological criteria for axons is by our measure 5 in panel A of Fig. 3 (dim image) and 14 in panel B of Fig. 3 (brighter image); the number of structures in Fig. 3C vs. Fig. 3D is 0 vs. 8. These differences are not trivial. Another caveat to the authors' interpretation of host axonal penetration of grafts is that animals with poor graft survival or integration may have fewer penetrating axons. As noted above, some animals rated by the authors as having the best grafts in fact have major fissures, including fissures at the rostral lesion site (animal #3 above, Fig. 1) that would preclude axonal entry to the graft. As with their assessment of reticulospinal axonal penetration of grafts, Steward and colleagues reported that serotonergic penetration of axons into grafts was limited. The authors state:
Fig. 3. Host reticulospinal axonal penetration of grafts. (A) This image from the Steward study (their Fig. 8E) shows limited numbers of host BDA anterogradely-labeled reticulospinal axons present within a region that the authors refer to as a graft. Note that no clear axons are visible in region of dashed lines. (B) The same original figure from the authors' paper is brightened in Adobe Photoshop. Axons present in the original figure are clearer, and several axons that were not visible in the original image are seen, indicated by arrowheads. While background artifact is also accentuated, particularly in the upper right hand corner of the image, linear structures are identifiable as axons. (C) The boxed region in panel A is shown at higher magnification at the exposure of the authors' original figure. No clearly identifiable axons are visible. (D) When the same field is brightened, at least 8 axons are identifiable in the field (arrowheads). We are not certain that this region, stated to be within the graft in the authors' article, is actually within the graft: the axons that are now more readily visible extend in parallel and linear arrays, raising the possibility that the structures are actually still within the host, close to the graft/host interface. In our original study, we would not have considered axons with this morphology to be within a graft. Scale bar A–B 100 μm, C–D 60 μm.
“…our conclusion is that growth of host axons into the graft was meager.”
From the data presented it is not clear how one can conclude that serotonergic axons were “much more likely” graft-derived. Steward and colleagues point out that they found occasional 5HT-labeled cell bodies in the graft, and conclude that serotonergic axons within the graft may have originated from these grafted neurons. However, axons originating from grafted cells should double label for GFP and 5HT, yet such structures were not detected (Figs. 5A–B). While the authors stated that the detection of GFP may have been below detection limits in the serotonergic axons, Fig. 5C–D seems to clearly indicate
However, most images illustrating serotonergic penetration of the graft were of too low a magnification to be clearly interpretable. Fig. 4 illustrates the authors' original images, and the effects of brightening the image. With brightening, more axons become evident. Steward and colleagues also raise doubts that serotonergic axons within and caudal to grafts are host-derived. However, their statements
on this point are incongruous. The Results section accurately states limitations of their dataset: “This means either that the 5HT positive axons actually did arise from the host or that the 5HT axons did not express GFP at sufficiently high levels to be detectable. Thus, the question of the origin of the 5HT axons caudal to the graft remains open.” However, the Discussion section avers: “We cannot exclude the possibility that the 5HT positive axons caudal to the graft were host axons that regenerated through and beyond the graft, but it is much more likely that they are from the 5HT positive neurons within the graft.”
498
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Fig. 4. Host raphespinal axonal penetration of grafts. (A) This unaltered image from the Steward study (their Fig. 9C) shows limited numbers of host 5HT-labeled axons penetrating grafts. Areas in boxes indicate regions shown at higher magnification in panels C–F. The arrow is from the Steward paper and is said to point to axons. This was the highest magnification image provided in the study of Steward and colleagues, and is too low a magnification to clearly determine whether axons are present. (B) The same figure from the authors' paper is brightened in Adobe Photoshop. Objects that may represent axons are brighter. The resolution and focus of the image are not optimal. (C) We have enlarged the boxed region in panel A, retaining the low exposure values of the authors' image. The white arrow points to the structure the authors identified as a serotonergic axon penetrating the graft. (D) The image in panel C is brightened in Adobe Photoshop, and reveals three additional structures indicated with arrowheads that appear to be serotonergic axons, with linear configuration and punctate labeling that are typical of the labeling pattern of this class of axons in the spinal cord. These objects were not visible in the authors' original panel A. Additional linear objects in the field may represent axons but definitive conclusions regarding their identity as axons cannot be made. (E) The boxed region from panel A is shown at higher magnification; this same region is brightened in panel (F) using the same brightening settings employed to generate panel D. No new structures are evident in panel F, suggesting that the brightening procedure that reveals linear structures in panel D is specific rather than non-specific. Scale bars cannot be provided because they were omitted in the Steward manuscript.
detection and separation of 5HT- and GFP-labeled axons. Brighter images and higher magnification make this more clearly appreciable (Figs. 5B–D). Oddly, the authors used only native GFP fluorescence to address the question of 5HT and GFP co-localization, rather than using anti-GFP antibodies that substantially enhance detection sensitivity for fine axons. In addition, for reasons that are unclear, in some cases the authors labeled 5HT axons with a green fluorescent antibody, which overlapped with the native green fluorescence of the graft and subsequently limited the yield of interpretable data. It should also be noted that serotonergic penetration of grafts was assessed in some animals with extensive graft cavitation, a confounding factor in interpreting host axonal penetration (Fig. 6). A potential explanation for the occasional presence of serotonergic somata in grafts in the Steward study is that cells for grafting were prepared in our laboratory by a different individual than the grafts used in the original study by Lu et al. (2012). When grafts from E14 donors are prepared that include portions of the brainstem, it is possible
Fig. 5. Origin of raphespinal axons in grafts. (A) Steward and colleagues state that they cannot be certain whether 5HT labeled axonal structures in and caudal to grafts originate from the host or from the graft. Their original panel C (this panel A) is underexposed, but it indicates a serotonergic axon caudal to the graft with a red arrow, and the axon does not appear to co-localize with GFP; co-localization would indicate origin of the serotonergic axon from a grafted neuron. A neighboring, GFP-labeled axon derived from a grafted neuron is evident, and it is not labeled for serotonin. (B) Brightening the authors' original figure, and (C–D) magnification of these images, seems to indicate no overlap of the red and green channels, suggesting that these serotonergic axons caudal to the graft are likely host-derived. It would appear from these images that the GFP labeling is sufficiently distinct to exclude co-localization of GFP and 5HT labels. Scale bar A–B 250 μm, C 80 μm, D 40 μm.
to have brainstem (raphe) neuronal primordial (Hou et al., 2013). The grafts in our original study were prepared by Dr. Lu and did not contain serotonergic somata. Functional assessments Steward and colleagues state on page 34: “Although not explicitly reported, Dr. Lu communicated to us that rats were excluded post-hoc from the behavioral assessments if histological assessment revealed that grafts did not meet defined criteria.” This statement is incorrect. “Lu et al. only included rats in the BBB analyses in which grafts met particular criteria. His explanation to us was that rats were only included ‘if grafts were adherent to the host over more than 1/3 of their circumference bilaterally’. It was not clear to us how this criterion was applied in practice even after extensive discussions with Dr. Lu.” This statement is incorrect. The authors are indicating, however, that there were communication problems. What Dr. Lu was trying to do was advise the authors how to deal with rifts in their grafts. Clearly, there was a misunderstanding. “The lack of motor recovery here vs. in Lu et al. may be due to the fact that most of our grafts had partitions. It should be noted, however, that the functional data presented in Lu et al. came from selected cases in which the grafts were judged to meet specific criteria (see Results).” This statement is incorrect; our original paper did not select cases. How did the authors misunderstand so many separate points, and report these points inaccurately in their publication? The authors themselves stated that communication was unclear (“It was not clear to us
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
499
Fig. 6. Graft used to assess 5HT labeling. Panels A and B show animal #5 from Steward's study that was one of four animals used to assess graft penetration by host serotonergic axons. The graft has a large central cavity and for this reason may not be an optimal case for examination of host serotonergic penetration. Panel B is a lightened version of Stewards' original image to more clearly indicate graft morphology and integration. Scale bars cannot be provided because they were omitted in the Steward manuscript.
how this criterion was applied in practice even after extensive discussions with Dr. Lu”). If the authors were unclear on a point, they could simply have contacted Dr. Tuszynski, a communicating author of the original study, or the other study authors directly for further clarification. This is important to ensure accuracy of the replication effort, particularly when rendering statements regarding the conduct of our original study. Dr. Steward may have been reluctant to communicate with more authors of the original study, lest the independence of the study be called into question. Yet the alternative — publishing statements that one did not truly understand — ran the substantial risk of reporting inaccurate information. Indeed, several inaccurate statements regarding our study were made and, unfortunately, published. To address the doubt that the authors raise regarding our functional data and selection of animals, Table 1 presents all of our original functional scoring data from all animals. Eight animals in our grafted group were enrolled in the behavioral study; one animal died after two weeks, prior to grafting, and another animal died five weeks after grafting. All remaining six animals were included in the behavioral analysis; selective data were not presented in our original study. Finally, with regard to functional outcomes of the Steward study, we believe that there is reason to be cautious. First, the extent of functional deficit in controls as measured by Steward and colleagues was roughly half that of our lesioned controls, suggesting either that lesions were incomplete, or that there were fundamental differences in how BBB scoring was performed. Indeed, more than half of Steward's lesioned controls had BBB scores above 3, with scores ranging as high as 7. We do not encounter rats with BBB scores in this higher range after T3 complete transections. Higher scores in the control lesioned group would render the detection of functional improvements in grafted animals more difficult (if adequate grafts were present). BBB scores can be overestimated if reflex movements associated with a passage of a fecal pellet are counted; we exclude these reflex movements. Second, two
grafted rats with higher BBB scores (mean = 6) were re-transected in the Steward study, and their scores dropped to 2. This suggests either that the initial lesions were incomplete, or that these animals had functional bridges. The authors did not perform histology on these animals, unfortunately. Third, the authors state that they have only two animals with a “Category 1” graft which has “at least a partial tissue bridge between rostral and caudal segments” of the graft: all of the other grafts were more poorly continuous than this. It is not feasible to draw conclusions regarding functionality with so few animals We also note that Dr. Tuszynski reviewed slides with the first author of the Steward paper several months prior to its publication, and communicated that there were extensive problems with their graft survival, integration or continuity across the lesion site. Yet these statements were not relayed in the author's manuscript, and only Dr. Lu's comments were relayed (erroneously, in many instances). The reasons for this are unknown. Outgrowth of neural stem cell-derived axons The cardinal finding of our original study was that grafts of multipotent neural progenitor cells extend very high numbers of axons into the lesioned adult spinal cord, through white matter and gray matter, and extend axons over very long distances (Lu et al., 2012). This growth is not dependent on rostral-to-caudal graft continuity, as grafted neurons surviving on either side of the lesion site can act as sources of axons that grow into the host. Indeed, Steward and colleagues reported a greater number of graft axons growing into the host than our original study: 48,420 axons compared to our report of 29,000 axons. As stated in our original paper (Lu et al., 2012), we erred on the conservative side in generating axon counts because axons are so dense and fasciculated at times that individual axons cannot be resolved, and in these cases we counted bundles of axons as one axon.
Table 1 BBB scores from all study subjects (Lu et al., 2012). Group
Week 1
Week 2
Week 3 (p1)
Week 4 (p2)
Week 5 (p3)
Week 6 (p4)
Week 7 (p5)
Week 8 (p6)
Week 9 (re-trans)
E14 E14 E14 E14 Lesion Lesion E14 E14 E14 E14 Lesion Lesion Lesion Lesion
0 0 0.5 0 0 0.5 0 0 0 0 0 0 0 0
0.5 0.5 0.5 (dead) 0.5 0 0 0.5 0 0 0 0.5 0 0
1 0.5 1.5
2 1.5 3.5
4.5 2.5 5
6.5 6 8
6 5.5 8
7 7 7.5
2.5
0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0 0
2 1 1.5 4 1 1.5 0.5 1 1 1
1 1 4 2.5 3 1 1 1.5 1.5 1
1.5 2 7 6.5 5.5 7 3 1.5 1.5 2.5
2 2.5 8.5 7 3 1 (sick) 1 1 1.5 1
2 2.5 8 7 3.5 (dead) 0.5 1 1 2
2.5 1 3.5 0.5 1 2 1 2
500
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Discussion section of Steward's report The Discussion section of Steward and colleagues paper begins with this statement: “The original report of Lu et al. was remarkable and exciting because of the reported long-distance growth of axons from the graft, ingrowth of host axons, which could form a relay circuit, and fivepoint improvement on the BBB scale reflecting recovery of the ability to move all joints of the hindlimbs. Of these 3 key findings, our results confirmed only the long-distance outgrowth of axons from the graft. We did not confirm extensive ingrowth of host axons into the graft or enhanced locomotor recovery in rats that received grafts.” By dissociating the failure of their replication effort to form a continuous tissue bridge across the lesion from the functional outcomes, the authors ignore a fundamental limitation in their study that precludes their ability to draw reasonable conclusions: without a bridge, there is no basis for relay formation and functional recovery. When the authors learned that their grafts did not form bridges, they either should have repeated the study, or simply reported that rifts were present that precluded the drawing of conclusions regarding functional outcomes. What the authors do show, in fact, is that isolated grafts fail to support functional recovery, a finding consistent with our original report that continuous grafts are required to form neural relays across the complete T3 transection site. Conclusions The fact that multipotent neural progenitor cells failed to reconstitute a continuous tissue bridge across the lesion site limited the ability
of Steward and colleagues to test the hypothesis that such grafts can influence functional outcomes. We agree that rift formation in grafts is a challenge after delayed grafting, and this methodological issue requires refinement as the initial effort to transfer this technology to other laboratories occurs. The authors' faint imaging methods yielded poor visualization of penetrating host axons, and it is prudent to limit one's conclusions when the methods are limited. The authors' statements regarding our original study, including grafting methods that were used in our study and the exclusion of animals from functional scoring, are simply incorrect. The cardinal finding of our report, the remarkable extension of axons over very long distances from grafts of multipotent neural progenitor cells, was confirmed. This very extensive axonal growth, and the concurrent growth of host axons into grafts, are the subject of several investigations now in progress that we hope will lead to basic insights into the nature of axonal growth after adult CNS injury, and contribute to the eventual development of novel therapies for neural trauma. References Hou, S., Tom, V.J., Graham, L., Lu, P., Blesch, A., 2013. Partial restoration of cardiovascular function by embryonic neural stem cell grafts after complete spinal cord transection. J. Neurosci. 33, 17138–17149. Lu, P.,Wang, Y.,Graham, L.,McHale, K.,Gao, M.,Wu, D.,Brock, J.,Blesch, A.,Rosenzweig, E.S., Havton, L.A., Zheng, B., Conner, J.M., Marsala, M., Tuszynski, M.H., 2012. Long-distance growth and connectivity of neural stem cells after severe spinal cord injury. Cell 150, 1264–1273. Sharp, K.G., Yee, K.M., Steward, O., 2014. A re-assessment of long distance growth and connectivity of neural stem cells after severe spinal cord injury. Exp. Neurol. 257, 186–204.
Contents lists available at ScienceDirect
Experimental Neurology journal homepage: www.elsevier.com/locate/yexnr
Commentary
Neural stem cells in models of spinal cord injury Mark H. Tuszynski a,b,⁎, Yaozhi Wang a, Lori Graham a, Karla McHale a, Mingyong Gao a, Di Wu a, John Brock a, Armin Blesch c, Ephron S. Rosenzweig a, Leif A. Havtond, Binhai Zheng a, James M. Conner a, Martin Marsala e, Paul Lu a,b a
Dept. of Neurosciences, University of California — San Diego, La Jolla, CA, USA Veterans Administration Medical Center, San Diego, CA, USA Spinal Cord Injury Center, University Hospital Heidelberg, 69118 Heidelberg, Germany d Dept. of Neurology, University of California — Irvine, Irvine, CA, USA e Dept. of Anesthesiology, University of California — San Diego, La Jolla, CA, USA b c
a r t i c l e
i n f o
Article history: Received 27 May 2014 Revised 9 July 2014 Accepted 22 July 2014 Available online 28 July 2014
a b s t r a c t Replication of published studies is an important and respected aspect of the conduct of science. Most would argue that the interpretation of “negative” outcomes is still more challenging than the interpretation of “positive” findings, however, due to uncertainty in knowing precisely why a hypothesized outcome was not observed: in particular, are “negative” findings in replication studies a result of invalidity of the original experimental hypothesis, or due to a methodological failure, insensitivity of the applied instruments of analysis, or other factors? These points must be carefully considered. Steward and colleagues report findings of a study in which multipotent neural progenitor cells were grafted to sites of T3 complete transection. Unlike our study, cells failed to fill the lesion site, leaving collagenous rifts between rostral and caudal graft components. This “anatomical” failure precluded formation of neural relays across the lesion site, and was predictably associated with a failure to detect functional improvement. In summarizing outcomes of the study, Steward and colleagues did not clearly link the failure to achieve graft continuity in the lesion cavity with functional outcomes, despite the central role of this observation in cogently interpreting results of the replication study. In addition, the authors stated that they failed to replicate our report of “extensive” host axonal regeneration into grafts, but we did not report “extensive” host anatomical regeneration; moreover, underexposed images may have contributed to Steward's underestimation of host axonal penetration. The authors also stated that our original study excluded some animals from functional analysis, and this is incorrect. While replication studies are important and necessary, this particular report contained several errors, and the failure to form a continuous neural progenitor cell bridge across the lesion site limited the ability to conclude whether continuous grafts can restore function. In subsequent experiments we too have observed rift formation in animals grafted at long delays (N2 weeks) after SCI, and we confirm that animals with rifts do not exhibit functional improvement; we are developing methods to remove or prevent rift formation. The replication study confirmed the cardinal finding of our original report: that early-stage neural precursors extend very large numbers of axons over remarkably long distances through the lesioned adult spinal cord. Published by Elsevier Inc.
The ability to replicate published reports is essential to the practice of science, and we wholeheartedly support such efforts. A confirmatory replication provides the field with greater confidence that a particular avenue of research will prove informative. A failure to replicate is more complex. Failure to replicate leaves the first impression that the original findings were invalid, a potential case of Type I error in which the null hypothesis has been incorrectly rejected. Type I errors can result from chance, from underpowering of a study, from misinterpretation of the data, or more
⁎ Corresponding author at: Dept. Neurosciences, 0626, University of California, San Diego, La Jolla, CA 92093, USA. Fax: +1 858 534 5220. E-mail address: [email protected] (M.H. Tuszynski).
http://dx.doi.org/10.1016/j.expneurol.2014.07.011 0014-4886/Published by Elsevier Inc.
ominously, due to intentional or unintentional experimental bias or outright fraud. These are serious considerations. A failure to replicate can also result from other causes. Most important among these is that the conditions of the original study might not have been reproduced. If the original study methods were not used, or if the attempt to use the original study methods fails to generate the expected change in the dependent variable that is required to go on and test the hypothesis, then one is unable to draw a clear conclusion. One cannot in this case draw the conclusion that the study failed replication; one can only state that the necessary conditions to investigate the hypothesis could not be established. To provide an example, imagine that an expedition claims to have discovered a unique species of great ape in the Rwandan highlands. In
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
an effort to confirm this discovery, a second group sets off, but finds that a key bridge they must traverse to reach the site has been destroyed. They attempt to build a new bridge, but are unsuccessful; they are unable to reach the region to confirm or deny the report. Clearly, they cannot conclude that the species does not exist. They can only say that their effort was unsuccessful because they were unable to build the bridge. Moving from an example to the current case, the premise of our report in Cell (Lu et al., 2012) was that neural progenitor cells grafts can fill a severe spinal cord lesion site, enabling the formation of neural relays across the lesion site that support electrophysiological and functional improvement. Steward and colleagues, in an effort to replicate the study (Sharp et al., 2014), failed to achieve consistent filling of the spinal cord transection site. Not surprisingly, they subsequently failed to observe formation of functional relays across the lesion and functional improvement. But the failure to fill the lesion site eliminated the biological mechanism for testing the hypothesis that functional relays could be formed to support functional recovery. Accordingly, the study should either have been repeated with renewed effort to fill the lesion, or the report should have been limited to stating that the reported axonal outgrowth was replicated, but that the lack of cell fill precluded meaningful analysis of functional outcomes. There were other problems with the study, including: 1) a stated inability to communicate clearly with the first author, 2) statements that “extensive” axonal regeneration from host to graft were not replicated, although we did not report that host axonal regeneration was “extensive”, and 3) the statement that animals in our original study were “selected” for presentation of functional results, a statement that is untrue. Findings of replication studies should be presented in a thoughtful manner that appropriately considers and balances the numerous factors that can determine experimental outcomes. We provide detailed examples below that address some of these points. Specific points Gaps in grafts The authors of the replication study state (Sharp et al., 2014): “In the original study, two different methods were used for grafting. This was not reported in their Methods section, but was communicated to us by the original author.” This is inaccurate. Our original report used a single grafting method that consisted of dural penetration by a pulled glass micropipette and injection of cells into closed, established lesion cavities. This method corresponds to “Method 2” reported by Steward and colleagues. The first author of this response letter has used an open grafting method in non-human primate models, wherein the established spinal cord lesion is re-opened and accumulated cellular debris and “scar” are removed to provide an open and continuous grafting cavity (“Method 1” in the Steward paper). Around the time of the Steward replication we had begun to assess whether the open grafting method could be brought to the rat model in order to eliminate fibrous scar formation that occurred when rats were grafted at prolonged time points after injury (3 months post-lesion). The Steward replication experiment included this method, but the original study (Lu et al, 2012) did not. Steward and colleagues were mistaken in believing that the method had been used in our original study. There were several other errors in their report that are cited below. Our original report offered evidence that grafts that fill a spinal cord lesion site form a new neural relay across the complete transection site (Lu et al., 2012). Host axons regenerated into the graft (but not beyond the graft) and formed synapses; grafted neurons extended very large numbers of axons out of the lesion site and for long distances in the
495
host spinal cord beyond. Graft-derived axons formed synapses with host neurons in the spinal cord caudal (and rostral) to the lesion site. Functional improvement on the BBB motor scale occurred over several weeks. When the host spinal cord was stimulated three spinal segments rostral to the T3 complete transection/graft site, evoked responses were detected in the host spinal cord three spinal segments caudal to the lesion site. To determine whether host axons were specifically mediating either the functional improvement or neural activity recorded below the lesion, animals were re-lesioned in the host spinal cord immediately rostral to the lesion site. The re-lesion entirely abolished evoked responses below the lesion and eliminated the behavioral improvement. We presented these findings as indicative of evidence of formation of novel neural relays formed across the lesion site (Lu et al., 2012). The conceptual basis of this experiment is that novel neural relays are formed. An effort to replicate the experiment must generate experimental conditions in which a graft is placed that fills the lesion site sufficiently well that host axons can enter the graft and connect with a graft neuron, which in turn (perhaps through intermediary neurons) extends an axon outside the graft. The replication effort of Steward and colleagues failed to generate these basic conditions to enable testing of the hypothesis. That is, animals had non-neural tissue rifts that effectively segregated the graft into rostral and caudal components across which axons did not extend. The rifts in grafts of the Steward replication are under-appreciated because grafts are too darkly stained. The first author of this letter requested that the authors re-stain their material to generate normally rather than over-intensely labeled tissue, but this was not done. To illustrate the impact of this tissue processing issue, we used Adobe Photoshop software to modify only contrast and brightness in the published images from the authors' article (Fig. 1). When overexposure is reduced, one can see quite clearly that both of the authors' “two best” grafts (see their Table VI) lack rostral-to-caudal continuity (Fig. 1). Rostral-to-caudal continuity is present over only about 10% (estimated) of the proximal host/graft interface and 7% of the distal interface in their animal #3. Moreover, these interfaces were segregated from one another within the graft by a large central cavity (Fig. 1). Their second best example, animal #18, exhibited 43% integration in the proximal interface and 20% in the distal interface, but once again these regions were poorly aligned with one another and would be unlikely to provide a rostral-to-caudal pathway for bridging of the lesion site by a neural relay (Fig. 1). After these two “best” animals, the authors state that other animals had yet worse graft integration (Sharp et al., 2014). These conditions fail to generate the essential conditions for determining whether the experimental findings of neural relay formation with functional improvement can be replicated. In our view, it would have been appropriate to either repeat grafting to make a continuous graft bridge, or else state that the methods used in this experiment were unable to reconstitute a bridge, and that the ability to interpret functional outcomes was accordingly limited. The authors could then have focused on the more relevant issue of the rift formation itself, which is indeed an important consideration for further development of these methods. Why did rifts form in these grafts, unlike the study of Lu et al. (2012)? Dr. Lu performed the grafting at UC Irvine, using the same methods (in half of the grafted animals) that were used in his original report. We cannot identify with certainty the reason that rifts formed in the attempted replication, but several points may be relevant. First, we too observe rifts in our own laboratory when animals are grafted at delays longer than two weeks after the original transection. These rifts consist of collagen, and clearly divide the graft into rostral and caudal compartments and axons do not cross between these sections (Fig. 2). We do not observe functional recovery when rifts form. Rifts appear to originate from the pia/arachnoid membranes overlying the lesion, and do not form when a spinal cord lesion is more ventrally placed and is closed to the overlying dura. Second, grafts were prepared by a different individual in our laboratory than in the original Lu et al.
496
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Fig. 1. Graft integration. Shown are the two best examples (according to Steward and colleagues) of graft integration from the Steward paper. (A) Original panel from the Steward paper with very dark GFP labeling of graft in animal #3. Although a large cavity in the center of the graft is evident, poor integration of the graft with the host is obscured by the dark labeling. Horizontal section, rostral is toward the left. (B) Lightening the authors' original image in Adobe Photoshop, only one small region of continuity of the graft with the rostral region of the graft/host is visible, indicated by the arrowhead. Two white lines indicate the region of continuity, shown at higher magnification in C. Thus, there is a very small region from which host axons might penetrate the central region of the graft. Toward the caudal end of the graft/host interface, two very small regions of continuity are indicated between parallel white lines. These are shown at higher magnification in D. (C) Toward the rostral host/graft interface, the sole region of continuity from host into the core of the graft is indicated by the arrowhead and between the two parallel lines. Arrows point to areas of discontinuity that were obscured in the darkly labeled section of the original manuscript figure. (D) Similarly, the caudal portion of the graft shows small regions of continuity of the graft core to its caudal aspects between parallel white lines. White arrows indicate regions of discontinuity. Overall, the graft in animal #3 shows very poor graft continuity across the lesion site. (E) The authors' original report also cited animal #18 as one of two animals with the best graft continuity. However, this section, like others, was darkly labeled. (F) Lightening of original panel from the Steward paper shows limited regions of graft integration with the host. Potential graft continuity with the rostral host spinal cord (between parallel white lines) amounts to 43% of the available area, and on the caudal end of graft is 20%. However, the rostral and caudal regions of graft continuity with the host are not aligned: the rostral region of continuity is central, whereas the caudal region is lateral. Thus, there would be little rostral-to-caudal parenchymal connectivity through the graft. Arrows indicate regions of non-continuity. Scale bars cannot be provided because they were omitted in the Steward manuscript (their Fig. 3; (Sharp et al., 2014)).
(2012) report, and there may have been differences in tissue harvesting and preparation that predisposed to rift formation or incomplete graft filling. Dr. Lu is the most experienced individual in preparing these
grafts, but he did not prepare grafts for the replication because he was already at UC Irvine preparing for surgery. Third, the grafted cells remained immobile in solution for a longer time period in the Steward
Fig. 2. Fibrous septation in chronic grafts. When we implant E14 spinal cord-derived multipotent neural progenitor cells to sites of T3 complete transection at a delay of three months after the original injury, we also observe formation of fibrous rifts (arrows) in the lesion site that separate the proximal and distal aspects of the spinal cord. Functional recovery is not observed in our animals with rift formation across the lesion site. Scale bar 0.75 mm (image courtesy of K. Kadoya, UCSD).
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
497
replication due to the need to complete all cell harvest prior to grafting, then transport cells to Irvine from San Diego (adding experimental time of several hours). Cell viability was not tested or reported by Steward and colleagues at the conclusion of animal grafting, an oversight that leads to a lack of data for determining whether cell viability was adversely affected. Fourth, the grafting solution consisting of fibrinogen/thrombin matrix and growth factors was prepared at UC Irvine, unlike our original experiment. Fifth, the operating environment was different from UCSD and there may be introduction of unknown differences that could impact engraftment technique and graft survival/ integration.
Host axonal regeneration into grafts Steward and colleagues state (Sharp et al., 2014): “Although BDA-labeled [reticulospinal] axons were numerous rostral to the lesion, few extended past the host-graft boundary.” “The fact that a few reticulospinal axons enter the graft and form terminal arbors confirms the findings of Lu et al., although on balance, it is the lack of extension into the graft that is more striking.” “We did not confirm extensive ingrowth of host axons into the graft…” We did not report extensive host axonal ingrowth; the Lu et al. (2012) paper simply stated that host axons penetrated neural stem cell grafts, and sample images were provided. We have subsequently replicated our findings and demonstrated host axons forming synapses onto grafted neurons using ultrastructural methods. Steward and colleagues presented underexposed images to support their claim that few host axons penetrated grafts. Although the exposure times they employed are useful for preventing pixel saturation and examining subtle differences in levels of fluorescence, such underexposed images are distinctly suboptimal for assessing the presence of fine structures. Taking the authors' published images and adjusting only brightness and contrast settings in Adobe Photoshop to compensate for low exposure, more axons become apparent (Fig. 3). Had the authors examined the material at higher magnification and acquired images under improved fluorescent intensity, the presence of more axons might have been appreciated. The number of structures meeting morphological criteria for axons is by our measure 5 in panel A of Fig. 3 (dim image) and 14 in panel B of Fig. 3 (brighter image); the number of structures in Fig. 3C vs. Fig. 3D is 0 vs. 8. These differences are not trivial. Another caveat to the authors' interpretation of host axonal penetration of grafts is that animals with poor graft survival or integration may have fewer penetrating axons. As noted above, some animals rated by the authors as having the best grafts in fact have major fissures, including fissures at the rostral lesion site (animal #3 above, Fig. 1) that would preclude axonal entry to the graft. As with their assessment of reticulospinal axonal penetration of grafts, Steward and colleagues reported that serotonergic penetration of axons into grafts was limited. The authors state:
Fig. 3. Host reticulospinal axonal penetration of grafts. (A) This image from the Steward study (their Fig. 8E) shows limited numbers of host BDA anterogradely-labeled reticulospinal axons present within a region that the authors refer to as a graft. Note that no clear axons are visible in region of dashed lines. (B) The same original figure from the authors' paper is brightened in Adobe Photoshop. Axons present in the original figure are clearer, and several axons that were not visible in the original image are seen, indicated by arrowheads. While background artifact is also accentuated, particularly in the upper right hand corner of the image, linear structures are identifiable as axons. (C) The boxed region in panel A is shown at higher magnification at the exposure of the authors' original figure. No clearly identifiable axons are visible. (D) When the same field is brightened, at least 8 axons are identifiable in the field (arrowheads). We are not certain that this region, stated to be within the graft in the authors' article, is actually within the graft: the axons that are now more readily visible extend in parallel and linear arrays, raising the possibility that the structures are actually still within the host, close to the graft/host interface. In our original study, we would not have considered axons with this morphology to be within a graft. Scale bar A–B 100 μm, C–D 60 μm.
“…our conclusion is that growth of host axons into the graft was meager.”
From the data presented it is not clear how one can conclude that serotonergic axons were “much more likely” graft-derived. Steward and colleagues point out that they found occasional 5HT-labeled cell bodies in the graft, and conclude that serotonergic axons within the graft may have originated from these grafted neurons. However, axons originating from grafted cells should double label for GFP and 5HT, yet such structures were not detected (Figs. 5A–B). While the authors stated that the detection of GFP may have been below detection limits in the serotonergic axons, Fig. 5C–D seems to clearly indicate
However, most images illustrating serotonergic penetration of the graft were of too low a magnification to be clearly interpretable. Fig. 4 illustrates the authors' original images, and the effects of brightening the image. With brightening, more axons become evident. Steward and colleagues also raise doubts that serotonergic axons within and caudal to grafts are host-derived. However, their statements
on this point are incongruous. The Results section accurately states limitations of their dataset: “This means either that the 5HT positive axons actually did arise from the host or that the 5HT axons did not express GFP at sufficiently high levels to be detectable. Thus, the question of the origin of the 5HT axons caudal to the graft remains open.” However, the Discussion section avers: “We cannot exclude the possibility that the 5HT positive axons caudal to the graft were host axons that regenerated through and beyond the graft, but it is much more likely that they are from the 5HT positive neurons within the graft.”
498
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Fig. 4. Host raphespinal axonal penetration of grafts. (A) This unaltered image from the Steward study (their Fig. 9C) shows limited numbers of host 5HT-labeled axons penetrating grafts. Areas in boxes indicate regions shown at higher magnification in panels C–F. The arrow is from the Steward paper and is said to point to axons. This was the highest magnification image provided in the study of Steward and colleagues, and is too low a magnification to clearly determine whether axons are present. (B) The same figure from the authors' paper is brightened in Adobe Photoshop. Objects that may represent axons are brighter. The resolution and focus of the image are not optimal. (C) We have enlarged the boxed region in panel A, retaining the low exposure values of the authors' image. The white arrow points to the structure the authors identified as a serotonergic axon penetrating the graft. (D) The image in panel C is brightened in Adobe Photoshop, and reveals three additional structures indicated with arrowheads that appear to be serotonergic axons, with linear configuration and punctate labeling that are typical of the labeling pattern of this class of axons in the spinal cord. These objects were not visible in the authors' original panel A. Additional linear objects in the field may represent axons but definitive conclusions regarding their identity as axons cannot be made. (E) The boxed region from panel A is shown at higher magnification; this same region is brightened in panel (F) using the same brightening settings employed to generate panel D. No new structures are evident in panel F, suggesting that the brightening procedure that reveals linear structures in panel D is specific rather than non-specific. Scale bars cannot be provided because they were omitted in the Steward manuscript.
detection and separation of 5HT- and GFP-labeled axons. Brighter images and higher magnification make this more clearly appreciable (Figs. 5B–D). Oddly, the authors used only native GFP fluorescence to address the question of 5HT and GFP co-localization, rather than using anti-GFP antibodies that substantially enhance detection sensitivity for fine axons. In addition, for reasons that are unclear, in some cases the authors labeled 5HT axons with a green fluorescent antibody, which overlapped with the native green fluorescence of the graft and subsequently limited the yield of interpretable data. It should also be noted that serotonergic penetration of grafts was assessed in some animals with extensive graft cavitation, a confounding factor in interpreting host axonal penetration (Fig. 6). A potential explanation for the occasional presence of serotonergic somata in grafts in the Steward study is that cells for grafting were prepared in our laboratory by a different individual than the grafts used in the original study by Lu et al. (2012). When grafts from E14 donors are prepared that include portions of the brainstem, it is possible
Fig. 5. Origin of raphespinal axons in grafts. (A) Steward and colleagues state that they cannot be certain whether 5HT labeled axonal structures in and caudal to grafts originate from the host or from the graft. Their original panel C (this panel A) is underexposed, but it indicates a serotonergic axon caudal to the graft with a red arrow, and the axon does not appear to co-localize with GFP; co-localization would indicate origin of the serotonergic axon from a grafted neuron. A neighboring, GFP-labeled axon derived from a grafted neuron is evident, and it is not labeled for serotonin. (B) Brightening the authors' original figure, and (C–D) magnification of these images, seems to indicate no overlap of the red and green channels, suggesting that these serotonergic axons caudal to the graft are likely host-derived. It would appear from these images that the GFP labeling is sufficiently distinct to exclude co-localization of GFP and 5HT labels. Scale bar A–B 250 μm, C 80 μm, D 40 μm.
to have brainstem (raphe) neuronal primordial (Hou et al., 2013). The grafts in our original study were prepared by Dr. Lu and did not contain serotonergic somata. Functional assessments Steward and colleagues state on page 34: “Although not explicitly reported, Dr. Lu communicated to us that rats were excluded post-hoc from the behavioral assessments if histological assessment revealed that grafts did not meet defined criteria.” This statement is incorrect. “Lu et al. only included rats in the BBB analyses in which grafts met particular criteria. His explanation to us was that rats were only included ‘if grafts were adherent to the host over more than 1/3 of their circumference bilaterally’. It was not clear to us how this criterion was applied in practice even after extensive discussions with Dr. Lu.” This statement is incorrect. The authors are indicating, however, that there were communication problems. What Dr. Lu was trying to do was advise the authors how to deal with rifts in their grafts. Clearly, there was a misunderstanding. “The lack of motor recovery here vs. in Lu et al. may be due to the fact that most of our grafts had partitions. It should be noted, however, that the functional data presented in Lu et al. came from selected cases in which the grafts were judged to meet specific criteria (see Results).” This statement is incorrect; our original paper did not select cases. How did the authors misunderstand so many separate points, and report these points inaccurately in their publication? The authors themselves stated that communication was unclear (“It was not clear to us
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
499
Fig. 6. Graft used to assess 5HT labeling. Panels A and B show animal #5 from Steward's study that was one of four animals used to assess graft penetration by host serotonergic axons. The graft has a large central cavity and for this reason may not be an optimal case for examination of host serotonergic penetration. Panel B is a lightened version of Stewards' original image to more clearly indicate graft morphology and integration. Scale bars cannot be provided because they were omitted in the Steward manuscript.
how this criterion was applied in practice even after extensive discussions with Dr. Lu”). If the authors were unclear on a point, they could simply have contacted Dr. Tuszynski, a communicating author of the original study, or the other study authors directly for further clarification. This is important to ensure accuracy of the replication effort, particularly when rendering statements regarding the conduct of our original study. Dr. Steward may have been reluctant to communicate with more authors of the original study, lest the independence of the study be called into question. Yet the alternative — publishing statements that one did not truly understand — ran the substantial risk of reporting inaccurate information. Indeed, several inaccurate statements regarding our study were made and, unfortunately, published. To address the doubt that the authors raise regarding our functional data and selection of animals, Table 1 presents all of our original functional scoring data from all animals. Eight animals in our grafted group were enrolled in the behavioral study; one animal died after two weeks, prior to grafting, and another animal died five weeks after grafting. All remaining six animals were included in the behavioral analysis; selective data were not presented in our original study. Finally, with regard to functional outcomes of the Steward study, we believe that there is reason to be cautious. First, the extent of functional deficit in controls as measured by Steward and colleagues was roughly half that of our lesioned controls, suggesting either that lesions were incomplete, or that there were fundamental differences in how BBB scoring was performed. Indeed, more than half of Steward's lesioned controls had BBB scores above 3, with scores ranging as high as 7. We do not encounter rats with BBB scores in this higher range after T3 complete transections. Higher scores in the control lesioned group would render the detection of functional improvements in grafted animals more difficult (if adequate grafts were present). BBB scores can be overestimated if reflex movements associated with a passage of a fecal pellet are counted; we exclude these reflex movements. Second, two
grafted rats with higher BBB scores (mean = 6) were re-transected in the Steward study, and their scores dropped to 2. This suggests either that the initial lesions were incomplete, or that these animals had functional bridges. The authors did not perform histology on these animals, unfortunately. Third, the authors state that they have only two animals with a “Category 1” graft which has “at least a partial tissue bridge between rostral and caudal segments” of the graft: all of the other grafts were more poorly continuous than this. It is not feasible to draw conclusions regarding functionality with so few animals We also note that Dr. Tuszynski reviewed slides with the first author of the Steward paper several months prior to its publication, and communicated that there were extensive problems with their graft survival, integration or continuity across the lesion site. Yet these statements were not relayed in the author's manuscript, and only Dr. Lu's comments were relayed (erroneously, in many instances). The reasons for this are unknown. Outgrowth of neural stem cell-derived axons The cardinal finding of our original study was that grafts of multipotent neural progenitor cells extend very high numbers of axons into the lesioned adult spinal cord, through white matter and gray matter, and extend axons over very long distances (Lu et al., 2012). This growth is not dependent on rostral-to-caudal graft continuity, as grafted neurons surviving on either side of the lesion site can act as sources of axons that grow into the host. Indeed, Steward and colleagues reported a greater number of graft axons growing into the host than our original study: 48,420 axons compared to our report of 29,000 axons. As stated in our original paper (Lu et al., 2012), we erred on the conservative side in generating axon counts because axons are so dense and fasciculated at times that individual axons cannot be resolved, and in these cases we counted bundles of axons as one axon.
Table 1 BBB scores from all study subjects (Lu et al., 2012). Group
Week 1
Week 2
Week 3 (p1)
Week 4 (p2)
Week 5 (p3)
Week 6 (p4)
Week 7 (p5)
Week 8 (p6)
Week 9 (re-trans)
E14 E14 E14 E14 Lesion Lesion E14 E14 E14 E14 Lesion Lesion Lesion Lesion
0 0 0.5 0 0 0.5 0 0 0 0 0 0 0 0
0.5 0.5 0.5 (dead) 0.5 0 0 0.5 0 0 0 0.5 0 0
1 0.5 1.5
2 1.5 3.5
4.5 2.5 5
6.5 6 8
6 5.5 8
7 7 7.5
2.5
0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0 0
2 1 1.5 4 1 1.5 0.5 1 1 1
1 1 4 2.5 3 1 1 1.5 1.5 1
1.5 2 7 6.5 5.5 7 3 1.5 1.5 2.5
2 2.5 8.5 7 3 1 (sick) 1 1 1.5 1
2 2.5 8 7 3.5 (dead) 0.5 1 1 2
2.5 1 3.5 0.5 1 2 1 2
500
M.H. Tuszynski et al. / Experimental Neurology 261 (2014) 494–500
Discussion section of Steward's report The Discussion section of Steward and colleagues paper begins with this statement: “The original report of Lu et al. was remarkable and exciting because of the reported long-distance growth of axons from the graft, ingrowth of host axons, which could form a relay circuit, and fivepoint improvement on the BBB scale reflecting recovery of the ability to move all joints of the hindlimbs. Of these 3 key findings, our results confirmed only the long-distance outgrowth of axons from the graft. We did not confirm extensive ingrowth of host axons into the graft or enhanced locomotor recovery in rats that received grafts.” By dissociating the failure of their replication effort to form a continuous tissue bridge across the lesion from the functional outcomes, the authors ignore a fundamental limitation in their study that precludes their ability to draw reasonable conclusions: without a bridge, there is no basis for relay formation and functional recovery. When the authors learned that their grafts did not form bridges, they either should have repeated the study, or simply reported that rifts were present that precluded the drawing of conclusions regarding functional outcomes. What the authors do show, in fact, is that isolated grafts fail to support functional recovery, a finding consistent with our original report that continuous grafts are required to form neural relays across the complete T3 transection site. Conclusions The fact that multipotent neural progenitor cells failed to reconstitute a continuous tissue bridge across the lesion site limited the ability
of Steward and colleagues to test the hypothesis that such grafts can influence functional outcomes. We agree that rift formation in grafts is a challenge after delayed grafting, and this methodological issue requires refinement as the initial effort to transfer this technology to other laboratories occurs. The authors' faint imaging methods yielded poor visualization of penetrating host axons, and it is prudent to limit one's conclusions when the methods are limited. The authors' statements regarding our original study, including grafting methods that were used in our study and the exclusion of animals from functional scoring, are simply incorrect. The cardinal finding of our report, the remarkable extension of axons over very long distances from grafts of multipotent neural progenitor cells, was confirmed. This very extensive axonal growth, and the concurrent growth of host axons into grafts, are the subject of several investigations now in progress that we hope will lead to basic insights into the nature of axonal growth after adult CNS injury, and contribute to the eventual development of novel therapies for neural trauma. References Hou, S., Tom, V.J., Graham, L., Lu, P., Blesch, A., 2013. Partial restoration of cardiovascular function by embryonic neural stem cell grafts after complete spinal cord transection. J. Neurosci. 33, 17138–17149. Lu, P.,Wang, Y.,Graham, L.,McHale, K.,Gao, M.,Wu, D.,Brock, J.,Blesch, A.,Rosenzweig, E.S., Havton, L.A., Zheng, B., Conner, J.M., Marsala, M., Tuszynski, M.H., 2012. Long-distance growth and connectivity of neural stem cells after severe spinal cord injury. Cell 150, 1264–1273. Sharp, K.G., Yee, K.M., Steward, O., 2014. A re-assessment of long distance growth and connectivity of neural stem cells after severe spinal cord injury. Exp. Neurol. 257, 186–204.
