Available online at Received 16 December 2002; revised 19 June 2003; accepted 23 June 2003

Transcription

1 Available online at R Experimental Neurology 184 (2003) Delayed grafting of BDNF and NT-3 producing fibroblasts into the injured spinal cord stimulates sprouting, partially rescues axotomized red nucleus neurons from loss and atrophy, and provides limited regeneration C.A. Tobias, a J.S. Shumsky, a M. Shibata, a M.H. Tuszynski, b I. Fischer, a A. Tessler, a,c and M. Murray a, * a Department of Neurobiology and Anatomy, Drexel University College of Medicine, 2900 Queen Lane, Philadelphia, PA 19129, USA b Department of Neuroscience, University of California San Diego, La Jolla, CA 92093, USA c Department of Veterans Affairs Hospital, Philadelphia, PA 19141, USA Received 16 December 2002; revised 19 June 2003; accepted 23 June 2003 Abstract Ex vivo gene therapy, utilizing modified fibroblasts that deliver BDNF or NT-3 to the acutely injured spinal cord, has been shown to elicit regeneration and recovery of function in the adult rat. Delayed grafting into the injured spinal cord is of great clinical interest as a model for treatment of chronic injury but may pose additional obstacles that are not present after acute injury, such as the need to remove an established scar, increased retrograde cell loss and/or atrophy, and diminished capacity for regeneration by neurons which may be doubly injured. The purpose of the present study was to determine if delayed grafting of neurotrophin secreting fibroblasts would have anatomical effects similar to those seen in acute grafting models. We grafted a mixture of BDNF and NT-3 producing fibroblasts or control fibroblasts into a complete unilateral cervical hemisection after a 6-week delay. Fourteen weeks after delayed grafting we found that both the neurotrophin secreting fibroblasts and control fibroblasts survived, but that only the neurotrophin secreting grafts provided a permissive environment for host axon growth, as indicated by immunostaining for RT-97, a marker for axonal neurofilaments, GAP-43, a marker for elongating axons, CGRP, a marker for dorsal root axons, and 5-HT, a marker for raphe spinal axons, within the graft. Anterograde tracing of the uninjured vestibulospinal tract showed growth into neurotrophin producing transplants but not into control grafts, while anterograde tracing of the axotomized rubrospinal tract showed a small number of regenerating axons within the genetically modified grafts, but none in control grafts. The neurotrophin expressing grafts, but not the control grafts, significantly reduced retrograde degeneration and atrophy in the injured red nucleus. Grafts of BDNF NT-3 expressing fibroblasts delayed 6 weeks after injury therefore elicit growth from intact segmental and descending spinal tracts, stimulate modest regenerative growth by rubrospinal axons, and partially rescue axotomized supraspinal neurons and protect them from atrophy. The regeneration of rubrospinal axons into delayed transplants was much less than has been observed when similar transplants were placed acutely into a lateral funiculus or, after a 4-week delay, into a hemisection lesion. This suggests that the regenerative capacity of chronically injured red nucleus neurons was markedly diminished. The increased GAP43 reactivity in the corticospinal tracts ipsilaterally and contralaterally to the combination grafts suggests that these axons remain responsive to the neurotrophins, that the neurotrophins may stimulate both regenerative and sprouting responses, and that the grafted cells continue to secrete the neurotrophins Elsevier Inc. All rights reserved. Keywords: Ex vivo; Gene therapy; Neurotrophins; NT-3; BDNF; Regeneration; Spinal cord injury; Sprouting Introduction Injury to the spinal cord damages ascending and descending pathways and often results in permanent loss of * Corresponding author. Department of Neurobiology and Anatomy, Drexel University College of Medicine, 2900 Queen Lane, Philadelphia, PA Fax: address: murray@drexel.edu (M. Murray). sensory and motor function because of the limited capacity of adult CNS neurons to regenerate (Bracken et al., 1992). The abortive regeneration after spinal cord injury has been attributed to a nonpermissive environment at the injury site, to atrophy and/or retrograde death of axotomized neurons (Mori et al., 1997; Himes et al., 1994, 2001), and to the limited ability of injured CNS neurons to express regeneration associated genes (Kobayashi et al., 1997; reviewed by /$ see front matter 2003 Elsevier Inc. All rights reserved. doi: /s (03)

2 98 C.A. Tobias et al. / Experimental Neurology 184 (2003) Fernandes and Tetzlaff, 2001; Murray and Fischer, 2001). Attempts to modify these conditions utilizing transplants and/or neurotrophins and agents that block inhibitory myelin proteins or degrade or inhibit the synthesis of proteoglycans (reviewed in Murray and Tobias, 2003) have enhanced regeneration of damaged neurons, but most of these successful results have been obtained when the therapy was provided simultaneously or within 1 to 2 weeks of injury. Attention is now being turned to developing strategies to repair more chronic injuries (Houle and Reier, 1988; Grill et al., 1997a; Coumans et al., 2001; Jin et al. 2002; Kwon et al., 2002a; Lu et al., 2002). Delayed transplantation into the injured spinal cord must overcome additional obstacles. For example, transplantation procedures designed to treat chronic spinal cord injury after surgical lesions have been reported to increase retrograde neuron death or atrophy because they subject neurons to a second injury (Houle and Ye, 1999; Kwon et al., 2002b). Nevertheless, specific trophic factors delivered to the injury site can increase survival after a second injury (Houle and Ye, 1999). Therefore, a therapy for chronic injury may need to provide factors that will enable neurons to withstand a second axotomy, as well as factors that will elicit regeneration from the surviving neurons. Other obstacles include an established glial scar at the injury site (Houle and Reier, 1988; reviewed by Reier and Houle, 1988; reviewed by Fawcett and Asher, 1999), lack of induction of regeneration associated genes in axotomized cell bodies (Tetzlaff et al., 1994; Kobayashi et al., 1997, reviewed by Fernandes and Tetzlaff, 2001), and an increase in expression of inhibitory molecules at and around the injury (reviewed by Huber and Schwab, 2000; McKeon et al., 1991). Potential therapies for chronic spinal cord injury must overcome these impediments in order to promote recovery. One of the better characterized therapies for spinal cord injury (SCI) has been the intraspinal transplantation of fibroblasts modified to secrete neurotrophins. Primary fibroblasts modified to express NT-3 support corticospinal tract regeneration and partial recovery of hindlimb function after a dorsal thoracic hemisection (Grill et al., 1997b; Tuszynski et al., 2003). BDNF producing fibroblasts support axonal growth, including rubrospinal tract regeneration, ameliorate atrophy (and/or death) of axotomized red nucleus neurons, and allow partial recovery of forelimb function after a unilateral cervical subtotal hemisection (Liu et al., 1999, 2002; Kim et al., 2001; Schwartz et al., 2003; reviewed in Murray et al., 2002). To test the effectiveness of these transplants in a chronic preparation, we grafted fibroblasts genetically modified to secrete neurotrophins into a complete unilateral cervical hemisection after a 6-week delay. Since many of the ascending and descending spinal tracts that have been lesioned by this injury model express TrkB and TrkC receptors (Bradbury et al., 1998; King et al., 1999), we used a combination of BDNF and NT-3 secreting fibroblasts. In this report we evaluated the presence of axons within the Fig. 1. Structure of the recombinant retroviral vectors. The retroviral LTR promoter drives expression of the (A) human BDNF cdna, (B) human NT-3 cdna, and (C) GFP cdna. The SV40 promoter initiates transcription of the neomycin phosphotransferase gene used for antibiotic selection and cloning of independent cell lines. graft, regeneration of the rubrospinal tract and sprouting of the vestibulospinal tract, and the extent of survival and atrophy of the axotomized red nucleus neurons. In a separate report we evaluate recovery of motor and sensory function (see Shumsky et al., 2003). Materials and methods Engineering of primary fibroblasts expressing BDNF, NT-3, or GFP BDNF (Fb/BDNF), NT-3 (Fb/NT-3), and green fluorescent protein (Fb/GFP) producing fibroblasts were used in these experiments. Detailed procedures for the generation of retrovirally engineered fibroblasts have been previously described (Grill et al., 1997a, b; Nakahara et al., 1996). Briefly, the BDNF, NT-3, or GFP recombinant retroviral DNA constructs (Figs. 1A C, respectively) were transfected into the PA317 amphotropic producer cell line (Miller, 1990) to create three different producer cell lines making infectious but replication-incompetent viral particles. A neomycin resistance gene was inserted downstream of a constitutively active SV40 promoter to allow for in vitro selection of the cells in the presence of the antibiotic G418 (Life Technologies, Grand Island, NY). Next, three rounds of infection of primary Long-Evans fibroblasts were performed with media from the three different producer cell lines. G418 was added to the media of the infected cells, and clones were selected after 2 weeks in culture. The level of BDNF and NT-3 expression was determined by ELISA. All groups of engineered fibroblasts were then assayed for recombinant neurotrophin bioactivity by an E8 chick DRG bioassay.

3 C.A. Tobias et al. / Experimental Neurology 184 (2003) To establish control fibroblasts (Fb/GFP), primary Long- Evans fibroblasts were engineered with a retroviral construct that expressed green fluorescent protein, GFP (Fig. 1C). The GFP expressed by these cells had no neurotrophic effects and elicited no neurite outgrowth in an E8 chick embryo DRG assay. We used these cells to determine whether any effects on axonal outgrowth were due to the primary fibroblasts alone. Bioassay of recombinant BDNF and NT-3 activity To determine if the recombinant BDNF and NT-3 produced by these separate cells were functional, we analyzed supernatant with an E8 chick dorsal root ganglion (DRG) explant assay. This assay has been previously described (Horie et al., 1991) and used to study Fb/BDNF growing in culture (Liu et al., 1999; Tobias et al., 2001). Briefly, standard growth media were removed from Fb/BDNF, Fb/NT-3, or Fb/GFP growing at 50% confluency in culture, and then cells were washed three times with sterile saline and grown in fresh media containing 0.1% fetal calf serum. Twentyfour hours later the conditioned media were removed from each cell line and studied. At that time 400 l of conditioned media from Fb/BDNF, Fb/NT-3, and Fb/GFP cell lines or control media (DMEM 0.1% FCS) were added to separate wells of a 12-well plate containing the DRGs. Conditioned media were also assayed at dilutions of 1:10 and 1:100. DRG neurite outgrowth was observed after h and compared to that elicited by media containing 15, 45, or 450 ng of recombinant human BDNF or NT-3 (Upstate Biotechnology, Lake Placid, NY). Surgical procedures Adult female Sprague Dawley rats (N 24) weighing g (Taconic, Germantown, NY) received a complete right-sided hemisection of the C3/C4 segment of the spinal cord followed after a 6-week delay by a transplant of neurotrophin secreting (Fb/BDNF Fb/NT-3) or non-neurotrophin secreting control fibroblasts (Fb/GFP). The rats were anesthetized by an intraperitoneal (ip) injection of acepromazine maleate (0.5 mg/kg, Fermenta Animal Health Co., Kansas City, MO), ketamine (63 mg/kg, Fort Dodge Animal Health, Fort Dodge, IA), and xylazine (6.3 mg/kg, Bayer Co., Shawnee Mission, KS), and the C4 segment of the spinal cord was exposed by a laminectomy. The dura was cut in the midline and a 30-gauge needle was used to make an incision between the C3 and C4 segments of the right side of the spinal cord. The spinal cord was removed between these segments by gentle aspiration. The rostrocaudal length of the lesion was approximately 2 mm. Care was taken to remove completely all spinal cord tissue on the right side laterally, dorsally, and ventrally along the midline while leaving the left side of the spinal cord intact. When hemostasis was achieved, gelfoam (UpJohn Pharmaceuticals, Kalamazoo, MI) was placed into the injury site. The dura was closed with interrupted 10-O sutures (Ethicon), the muscle closed in layers with 4-O sutures (Ethicon), and the skin closed with surgical staples. At the time of injury all rats received one bolus intravenous injection of methylprednisolone (30 mg/kg of body wt, iv) (Pharmacia & Upjohn Company, Kalamazoo, MI) through the tail vein within 5 min of the conclusion of surgery as in our previous studies (Liu et al., 1999) and because methylprednisolone treatment is commonly used clinically. After surgery, rats were kept on heating pads, observed until fully awake, and then returned to their home cages. At the time of the second surgery 6 weeks later, the injury site was reexposed, gelfoam removed, and the cord was debrided of scar tissue, thus extending the wound approximately 1 mm rostrally and caudally (the total length of lesion after this debriding was approximately 4 mm). Extending the wound reinjures spinal cord axons but also exposes them to the transplant without the barriers imposed by the astrocytic scar. Once hemostasis was achieved, gelfoam soaked in a solution of neurotrophin-secreting fibroblasts (Fb/BDNF Fb/NT-3, N 15) or control fibroblasts (Fb/GFP, N 6) resuspended at 100,000 cells/ l was placed into the lesion cavity until the space was filled completely. Before the dura was closed, an additional 10 l of engineered cells ( cells) resuspended in culture media was injected into the lesion site with a Hamilton syringe. Since approximately cells adhere to the gelfoam used as a transplant and cells are injected into the lesion site, approximately engineered cells were transplanted into each lesion cavity. Three rats received gelfoam again at the time of the second surgery to serve as an additional group of controls. The injury was closed and the rats received two additional intravenous injections of methylprednisolone in the same manner as after the first surgery (see above). All procedures were carried out in accordance with a protocol approved by Drexel University College of Medicine s Institutional Animal Care and Use Committee and followed the NIH guidelines for the care and use of laboratory animals. Immune suppression with cyclosporin A (CSA) All rats were immune suppressed with cyclosporin A (CSA, Novartis Pharmaceutical Corp., East Hanover, NJ) by daily subcutaneous injections of 1 mg/100g of body weight beginning 3 days before grafting and continuing for 2 weeks after injury. After that all rats received oral CSA resuspended in their drinking water at a final concentration of 50 g/ml for the duration of the study. Anterograde tracing of descending spinal tracts with biotinylated dextran amine (BDA) BDA injections were made 14 days before sacrifice. Rats were anesthetized (see Surgical procedures above) and placed into a stereotaxic apparatus. Labeling of the rubro-

4 100 C.A. Tobias et al. / Experimental Neurology 184 (2003) spinal tract and vestibulospinal tract was achieved by injecting BDA (Molecular Probes, Eugene, OR) into the left red nucleus (RN, N 8) and left lateral vestibular nucleus (LVN, N 8). The following coordinates were used to drill a burr hole to inject 0.5 l of 10% BDA, over 5 min, stereotactically with a 1- l Hamilton syringe, using Bregma as the zero point: RN, anterior posterior (AP) 5.8 mm; medial lateral (ML) 0.7 mm; and dorsal ventral (DV, from dural surface) 7.0 mm; LVN, AP 11.0 mm, ML 2.2 mm, DV 6.2 mm. After injection the Hamilton syringe was left in place for an additional 5 min to allow for diffusion of the BDA before retraction from the brain. Because the rubrospinal pathway is largely crossed, injections into the RN were made contralateral to the injury side to label axons that had regenerated into the grafted region. Because the lateral vestibulospinal tract is largely uncrossed, injections of the LVN were made contralateral to the injury side to evaluate uninjured axons that had sprouted into the transplant. After injections were completed, the skin was closed with 4-O sutures (Ethicon) and the rats were kept on heating pads, observed until fully awake, and then returned to their home cages. Tissue preparation, histology, and immunohistochemistry Rats were anesthetized with Nembutal at the end of the behavioral studies and transcardially perfused with 150 ml of physiological saline and then 500 ml of ice-cold 4% paraformaldehyde in 0.1 M phosphate buffer (PB) solution, ph 7.4. Spinal cords and brains were dissected and the transplant region was washed with PB for 2 h and then placed in 0.1 M PB containing 30% sucrose for 72 h. Specimens were then frozen in OCT compound (Tissue Tek, Sakura Inc., Japan) and serially sectioned on a freezing microtome at 20 m for spinal cord tissue and 40 m for brain tissue. Sectioned tissue was processed for Nissl or Nissl/myelin staining and fluorescence immunohistochemistry. The procedures for immunohistochemistry have been reported in detail elsewhere (Liu et al., 1999; Chow et al., 2000). The sectioned tissue was washed three times for 5 min in PBS, incubated for 5 min in PBS containing 0.2% Triton X-100, washed with PBS, incubated for 1 h in PBS containing 10% normal goat serum (NGS), and then transferred to a solution of 2% NGS the appropriate dilution of primary antibody in PBS overnight at room temperature in a humidified chamber. The following day, sections were washed with PBS, incubated with one of the fluorescently conjugated secondary antibodies in PBS for 2 h, washed with PBS, and then coverslipped with Vectashield (Vector, San Diego, CA) containing the nuclear counterstain DAPI. Primary antibodies were polyclonal CGRP, to identify dorsal root axons (1:5000) (Peninsula Laboratories, Inc., Belmont, CA), polyclonal 5-HT to identify raphe-spinal axons (1:10,000) (Eugene Tech., Ridgefield, NJ), monoclonal RT-97 to identify neurofilaments (1:100) (Boehringer Mannheim, GmbH, Germany), monoclonal GFAP to identify astrocytes (1:100) (Boehringer Mannheim, GmbH), and polyclonal GAP-43 to identify growing axons (1:2500; kindly provided by Dr. Larry Benowitz). Fluorescent secondary antibodies (diluted 1:200) were either FITC- or rhodamine-conjugated goat antimouse (Jackson Immuno- Research, West Grove, PA) for monoclonal antibodies or FITC- or rhodamine-conjugated goat antirabbit (Jackson ImmunoResearch) for polyclonal antibodies. The immunohistochemical procedure for the polyclonal GAP-43 antibody was performed as previously described (Benowitz et al., 1988, 1999). BDA-labeled fibers were visualized using an ABC elite kit (Vector Labs, Burlingame, CA) with DAB as the chromagen. This procedure has been described previously (Liu et al., 1999). Slides containing spinal cord tissue were rinsed three times for 30 min each with TBST (50 mm Tris-buffered saline containing 0.5% Triton X-100, ph 10.0) and then incubated with the avidin-biotinylated complex (ABC kit) at room temperature for 2 h. Slides were rinsed three times for 30 min in TBST and then 30 s in 50 mm Tris buffer and finally reacted with Sigma Fast-DAB compound (Sigma Chemical Co., St Louis, MO) in the dark for 30 min. The reaction was stopped by placing the slides in dh 2 O, and the slides were dehydrated and coverslipped with DPX (Fluka Chemie AG, Buchs, Switzerland). Quantitation of GAP-43 immunolabeling within transplants The procedure for quantifying the total fiber length of immunostained axons in the central portion of transplant tissue has been reported previously (Tobias et al., 2001). An outlined square field containing an area of m 2 in the central portion of grafted spinal cord from either the combination-treated group (Fb/BDNF Fb/NT-3, N 6) or control group (Fb/GFP or gelfoam alone, N 6) containing GAP-43 immunostaining was measured. Using NIH image software, a micrometer at the same magnification (200 ) was used to convert pixels to micrometers, and all GAP-43-immunostained fibers were traced within the outlined field in the central grafted region of one slide/animal. This measurement provided a total length of central host axons containing GAP-43 immunostaining for each section counted. The total mean length of axon staining in the combination treatment (Fb/BDNF Fb/NT-3) and control treatment (Fb/GFP or gelfoam alone) was compared using an unpaired Student s t test. Counting and size analysis of axotomized red nucleus neurons The procedure for analyzing red nucleus neuron survival and size after axotomy in Nissl-stained sections has been previously reported (Mori et al., 1997; Shibata et al., 2000; Liu et al., 2002). Neurons in the magnocellular portion of the red nucleus at the level of the interpeduncular nucleus

5 C.A. Tobias et al. / Experimental Neurology 184 (2003) were counted. This region extends approximately 480 m rostral from the caudal pole (Reid et al., 1975; Paxinos and Watson 1986; Mori et al., 1997). Red nucleus neurons are evenly distributed within 300 m from the caudal pole, but 300 to 480 m from the caudal pole ventrolateral and dorsomedial subgroups can be distinguished (Reid et al., 1975). The number of red nucleus neurons was counted using stereological counting methods, which include an optical fractionator technique (Williams and Rakic, 1988; West, 1993; Howard and Reed, 1998; West et al., 1991, 1996). To obtain a random sampling, twelve 40- m serial sections were cut from the caudalmost 480 m of the red nucleus starting from the caudal pole. These sections were labeled 1 12, and neurons in every other section were counted. Shrinkage after tissue processing and Nissl staining resulted in a uniform decrease to a mean thickness of m, which was similar to that reported previously (Liu et al., 2002). The first section to be counted was chosen randomly so that the possibility of counting sections 1, 3, 5, 7, 9, and 11 equaled that of counting sections 2, 4, 6, 8, 10, and 12. The extreme caudalmost 300 m of the red nucleus was considered a single subnucleus, whereas the dorsomedial and ventromedial divisions of the red nucleus located between 300 and 480 m from the caudal pole were considered two individual subnuclei. Each subnucleus was further subdivided into four quadrants and a random number (1 4) was generated to determine the quadrant to be counted. For cell counting using the optical fractionator technique, an IP-Lab program was written to drive a Photometric Sensys KAF-1400 CCD camera attached to a Leica DMRBE microscope to acquire a series of images through tissue sections with a counting box 15 m thick in the z axis at a magnification of 400 (counting frame equal to 39,235 m 2 in which red nucleus neurons are counted). A cell was counted within the 400 counting box if it displayed a clear neuronal morphology with a nucleus and did not contact the left or bottom edge of the counting box. The area of the red nucleus was then measured under 100 magnification using an NIH image macro program. An estimate of the total number of neurons in each red nucleus (N) was determined by obtaining the number of neurons identified by the optical disector (n) and the reciprocal of the fraction of section sampled (1/f), the fraction of the sectional area sampled (1/f a ), and the fraction of section thickness sampled (1/s t ). In brief, total red nucleus neuron counts N n 1/f 1/f a 1/s t. The cross-sectional area of Nissl-stained red nucleus neurons was measured using an NIH image macro under 400 magnification. The counting box used was the same as above and the maximal cross-sectional area for each neuron was measured. A total of red nucleus neurons was measured for each red nucleus. Neurons were then subdivided into three groups according to cross-section area as being m 2 (small), m 2 (medium), or m 2 (large). Statistical analysis of red nucleus neuron number and size Statistical analyses of the numerical data (red nucleus number and cross-sectional area) were performed using Stat View software (SAS Institute Inc., Cary, NC). The number and size of neurons in the uninjured red nucleus of combination-treated (Fb/BDNF Fb/NT-3) rats were compared to those in the uninjured red nucleus of control treated (Fb/GFP or gelfoam alone) rats. We found no statistical difference when these data were analyzed with an unpaired Student s t test. Therefore, the injured/uninjured ratio of each animal s red nucleus was used as the basis of comparison. The overall significance of the ratio data was determined using an unpaired Student s t test. Results Engineered fibroblasts secrete bioactive neurotrophins ELISA indicated that Fb/BDNF secreted 94 ng of BDNF/million cells/24 h and Fb/NT-3 secreted 47.7 ng of NT-3/million cells/24 h. An E8 chick DRG assay (Horie et al., 1991; Liu et al., 1999; Tobias et al., 2001) was performed to demonstrate biological efficacy, using conditioned supernatants from fibroblasts engineered to produce BDNF, NT-3, and GFP and on unconditioned media (Fig. 2). Conditioned media from Fb/BDNF (Fig. 2F), Fb/NT-3 (Fig. 2E), and a mixture of media from Fb/BDNF Fb/ NT-3 (Fig. 2G) elicited neurite outgrowth similar to that of the positive controls of recombinant human BDNF (Fig. 2D) and NT-3 (Fig. 2C). In contrast, unconditioned media (Fig. 2A) and media from Fb/GFP (Fig. 2B) did not elicit any neurite outgrowth. These results indicate that the engineered fibroblasts that we used in our studies released bioactive neurotrophins. Histology and immunohistochemistry of grafted region At sacrifice all grafted regions appeared translucent and were closely adherent to the host cord. The region of the spinal cord containing the graft was sectioned parasagittally to determine the length of the lesion/transplant and to study axon penetration into the graft. The mid-transplant region was measured in Nissl-stained sections in all animals to determine the rostral caudal extent of the lesions. The mean rostral to caudal length of the lesions was mm for rats that received Fb/GFP or gelfoam alone and mm for rats that received Fb/BDNF Fb/NT-3. These values are not significantly different (P 0.05). The smaller rostral caudal distance of the lesions compared to gross measurements during surgery was due to tissue shrinkage of approximately 50% after fixation and process-

6 102 C.A. Tobias et al. / Experimental Neurology 184 (2003) Fig. 2. Photomicrographs of the DRG explant assay demonstrating the bioactivity of recombinant neurotrophins. (A) Unconditioned media and (B) culture media from Fb/GFP did not elicit neurite outgrowth, whereas (C) recombinant human NT-3 and (D) recombinant human BDNF, used as the positive controls, and conditioned media from (E) Fb/NT-3, (F) Fb/BDNF, and (G) a combination of Fb/BDNF and Fb/NT-3 secreted biologically active neurotrophins that elicited neurite growth (arrows). Original magnification, 100.

7 C.A. Tobias et al. / Experimental Neurology 184 (2003) Fig. 3. Photomicrographs of Nissl-stained montages of parasagittal sections through cervical spinal cord 14 weeks after implantation of a delayed transplant (20 weeks after initial injury). (A) Fb/BDNF Fb/NT-3, (B) Fb/GFP, and (C) gelfoam alone. Note the survival of cells and the close apposition to the host spinal cord in (A) and (B). Few cells infiltrate into the gelfoam alone (C). H is host; G is graft. Dotted lines indicate the graft/host interface. Scale bars: A C, 1 mm. ing (see Materials and methods). Nissl/myelin staining of transverse sections rostral and caudal to the lesion was used to evaluate the amount of demyelination that occurred after injury. Examination revealed demyelination unilaterally and dorsally on the same side of the lesion in the entire fasciculus gracilis rostral to the lesion, occasionally extending slightly contralateral to the lesion, but the contralateral corticospinal tract was intact in almost all cases. Demyelination was also evident caudal to the lesion unilaterally in the dorsolateral and ventral medial white matter, the dorsal corticospinal tract, and also occasionally in the contralateral ventral white matter. These results provide evidence that the right-sided complete hemisection at minimum approached midline and occasionally crossed midline, which was probably due to secondary damage after both types of transplants (data not shown). Nissl-stained sections of spinal cord tissue revealed that a few host cells had migrated into the injury sites of gelfoam alone recipients (Fig. 3C). In contrast, the injury sites of rats that had received transplants of Fb/GFP (Fig. 3B) or Fb/BDNF Fb/NT-3 (Fig. 3A) were filled with cells that exhibited a fibroblast-like morphology, as previously reported (Liu et al., 1999). These results show that Fb/BDNF Fb/NT-3 and Fb/GFP survived transplantation into a chronically injured spinal cord in these immune suppressed rats. The grafted region was evaluated with immunocytochemical techniques for the presence of host axons containing neurofilaments immunoreactive for RT-97 and GAP-43, a marker for growing axons, bulbospinal serotonergic axons, CGRP-containing primary afferent axons, and for host astrocytes with GFAP. Grafts of gelfoam alone contained a few host cells that had migrated into the injury site (Fig. 4A), and the grafts were surrounded by an astocytic scar at the borders (Fig. 4B). The gelfoam region contained no

8 Fig. 4. Dark-field montages of cervical spinal cord showing immunostaining (for host cellular structures) in parasagittal section 14 weeks after delayed transplant (20 weeks after initial injury) of gelfoam (A C) or Fb/GFP (D H) controls. (A) DAPI nuclear counterstained section of gelfoam transplant showing that a few host cells have migrated into the graft, (B) adjacent section showing strong GFAP immunostaining at the lesion border, (C) adjacent section showing only trace neurofilament (RT-97) immunostaining within the grafted region; (D) DAPI-counterstained section of Fb/GFP transplant showing cells filling the grafted region, (E) adjacent section showing borders of the lesion outlined by strong GFAP immunostaining, and (F) adjacent section showing little RT-97 immunostaining within the grafted region; (G) higher power view within the central portion of the Fb/GFP graft showing few RT-97-stained axons (arrows) and (H) few GAP-43-immunostained axons. A F are 100 montage reconstructions. Scale bars: A F, 1 mm; G H, 100 m.

9 Fig. 5. Dark-field montages showing immunostaining (for host cellular structures) in parasagittal sections 14 weeks after delayed transplant (20 weeks after initial injury) of Fb/BDNF Fb/NT-3. (A) DAPI nuclear counterstain shows the continuity of cellular structures across the damaged cord. (B) Adjacent section demonstrates demarcation of the lesion border by increased GFAP reactivity. (C) Adjacent section shows dense RT-97 immunostaining within the central area of grafted region, indicating robust growth of axons into the graft. (D G) Higher power views of the central region of a Fb/BDNF Fb/NT-3 transplant showing host axons immunoreactive for (D) CGRP, (E) RT-97, (F) 5-HT, and (G) GAP-43. Arrows indicate labeled axons. A C are 100 montage reconstructions. Scale bars: A C, 1 mm; D G, 100 m.

10 106 C.A. Tobias et al. / Experimental Neurology 184 (2003) Dorsal corticospinal axons express GAP-43 rostral to neurotrophin secreting grafts Fig. 6. Bar graph showing total axon length of GAP-43 immunoreactivity in combination (Fb/BDNF Fb/NT-3) and control treated (Fb/GFP or gelfoam alone) animals in the central transplant region 14 weeks after delayed grafting (20 weeks after initial injury). The mean axon length SEM was significantly greater in the experimental combination treatment (Fb/BDNF Fb/NT-3; N 6) than in the operated control treatment (Fb/GFP or gelfoam alone; N 6); (*** P , unpaired Student s t test). CGRP or 5-HT staining (data not shown) and only very small amounts of RT-97 (Fig. 4C) and GAP-43 immunoreactivity (data not shown). Grafts of Fb/GFP were full of cells in the transplant region (Fig. 4D) with an astrocytic scar at the borders (Fig. 4E). The grafted region contained some RT-97 (Figs. 4F and G) and GAP-43-positive axons (Fig. 4H), but neither 5-HT nor CGRP immunoreactivity was observed. Transplants consisting of a combination of Fb/BDNF Fb/NT-3 were filled with cells (Fig. 5A) that were surrounded by an astrocytic scar (Fig. 5B). The combination grafts contained numerous axons, including fibers immunopositive for RT-97 (Figs. 5C and E), GAP-43 (Fig. 5G), 5-HT (Fig. 5F), and CGRP (Fig. 5D). Quantification of the total length of GAP-43-immunostained axons present in the central portion of the graft showed a 23-fold increase in mean axon length in Fb/BDNF Fb/NT-3 grafts vs gelfoam alone or Fb/GFP grafts. These differences were highly significant (P ; Fig. 6). These results provide quantitative evidence that transplants of Fb/BDNF Fb/ NT-3 formed an environment for host axon growth that was more permissive than that provided by either gelfoam alone or Fb/GFP grafts, as indicated by the abundance of centrally located GAP-43 axons, even when the cells were transplanted 6 weeks after the original injury. The axon length measured within these grafts did not distinguish between a few long axons and/or numerous short axons, but overall it is a measure of the permissiveness and attractiveness of the neurotrophin secreting transplants. The glial scar, as shown by GFAP immunoreactivity, was at least qualitatively similar among groups, suggesting that the trophic factors acted primarily on axons. We cannot rule out the possibility that the trophic factors also modified the environment more directly, e.g., the inhibitory environment of the scar, to make a more permissive environment. GAP-43 staining in unoperated control rats shows modest immunoreactivity in the dorsal horn, the dorsal commissure, and the corticospinal tract, as described in detail elsewhere (Vizzard, 1999). In experimental animals, staining in the dorsal horn and dorsal commissure appeared to be similar in intensity to that seen normally. GAP-43 immunoreactivity was, however, dramatically increased bilaterally and symmetrically in the dorsal corticospinal tracts rostral to Fb/BDNF Fb/NT-3 transplants (Fig. 7A), but not caudal to the injury site (Fig. 7C). Immunostaining for GAP-43 in the upper cervical dorsal corticospinal tracts was minimal or absent rostral to either Fb/GFP control (Fig. 7B) or gelfoam (Fig. 7D) grafts and also in uninjured rats (Fig. 7E). These results suggest that, 3 months after grafting Fb/BDNF Fb/NT-3, dorsal corticospinal tract axons continue to express growth-associated molecules at their distal ends, whereas control grafts do not have this effect. The unilateral complete hemisection model injures the dorsal corticospinal tract ipsilateral to the lesion while leaving the contralateral corticospinal tract intact. The pattern of increased staining ipsilateral to the lesion could represent a regenerative response (see Tuszynksi et al., 2003) and the staining contralateral to the lesion could represent sprouting by undamaged CST axons. Corticospinal tract axons presumably contributed to the robust GAP-43 immunostaining within the neurotrophin secreting grafts but did not extend caudal to the graft; there was no staining in the area of the corticospinal tracts caudal to the injury. Axonal growth of descending spinal tracts (VST, RST) We used BDA as an anterograde tracer to test for the presence of regenerating and sprouting supraspinal axons in grafts. We injected BDA into the left red nucleus to label regenerating axons, because 99% of these neurons were axotomized by the right cervical hemisection. We injected the left lateral vestibular nucleus to label sprouting axons because the vestibulospinal tract projects primarily ipsilaterally (Rubertone et al., 1995) and very few of these axons would have been injured by a right-sided hemisection. Anterograde tracing of the rubrospinal (RST) tract provided evidence of limited regeneration into delayed transplants, but no labeled RST axons extending through or caudal to the transplant. While small numbers of rubrospinal axons penetrated into Fb/BDNF Fb/NT-3 transplants (Figs. 8A and B), none entered into Fb/GFP control (Fig. 8C) or gelfoam alone transplants (data not shown). Some rubrospinal axons appeared to form retraction bulbs rostral to the Fb/GFP control transplants (Fig. 8D) and Fb/BDNF Fb/NT-3 transplants. These results suggest that transplants of modified fibroblasts that secrete BDNF NT-3 elicit only modest regeneration of RST when these trans-

11 C.A. Tobias et al. / Experimental Neurology 184 (2003) plants are delayed for 6 weeks after a complete unilateral cervical hemisection. Anterograde tracing of vestibulospinal (VST) axons revealed numerous labeled processes within Fb/BDNF Fb/ NT-3 transplants (Fig. 8F), but few if any within Fb/GFP control (data not shown) or gelfoam alone grafts (Fig. 8E). Since the VST contralateral to the injury is primarily uncrossed and projects to the intact side of the spinal cord, the bulk of the BDA-labeled axons observed in Fb/BDNF Fb/NT-3 transplants is likely to be intact VST axons undergoing collateral sprouting in response to neurotrophins secreted by the graft. We cannot rule out the possibility, however, that some labeled axons in the grafts are regenerating and arise from those axons that normally cross the midline rostral to the injury site (Huisman et al., 1984; Bacskai et al., 2002; Matesz et al., 2002). Delayed transplants provide partial protection for axotomized red nucleus neurons Counts of red nucleus neurons with a recognizable nucleus in Nissl-stained sections using stereological methods revealed no statistical difference in the number of red nucleus neurons on the uninjured side of the combinationtreated (Fb/BDNF Fb/NT-3) and control treated (Fb/GFP or gelfoam alone) groups (P 0.05, Table 1). Since there are no statistical differences in the total number of red nucleus neurons between the right and left sides of intact adult rats (Mori et al., 1997; Shibata et al., 2000; Kwon et al., 2002a, Liu et al., 2002), we compared the ratio of recognizable injured red nucleus neurons to uninjured red nucleus neurons in the combination-treated (Fb/BDNF Fb/NT-3) group and the control (Fb/GFP or gelfoam alone) groups. Fourteen weeks after grafting (20 weeks after initial injury) we found the mean loss of detectable cells within the injured red nucleus of Fb/GFP or gelfoam-treated animals to be 27% (73% survival) (Table 1, Fig. 9A). Animals that received combination transplants (Fb/BDNF Fb/NT-3) had a mean detectable cell loss within the injured red nucleus of 18.3% (81.7% survival) (Table 1, Fig. 9A). The difference between these values was statistically significant (P 0.05) and represents a 32% rescue of neurons that would otherwise have been lost or unrecognizable. Grafts of BDNF NT-3 secreting fibroblasts therefore produce a small but significant rescue of axotomized red nucleus neurons, even though the graft is delayed until 6 weeks after injury and the rubrospinal axons have been subjected to a second injury. Because the mean cross-sectional area of uninjured red nucleus neurons in combination-treated (Fb/BDNF Fb/ NT-3) and control (Fb/GFP or gelfoam alone) animals was not statistically different (P 0.05, Table 2), we compared the ratio of injured to uninjured red nucleus neuron mean cross-sectional areas between combination-treated and control treated groups. At 14 weeks after grafting (20 weeks after initial injury) the mean cross-sectional area of red nucleus neurons in Fb/GFP or gelfoam-treated rats had shrunk to 50.8% of its original size (Table 2, Fig. 9B). The mean cross-sectional area of red nucleus neurons of Fb/ BDNF Fb/NT-3 animals was reduced to 58.7% of its original cell size (Table 2, Fig. 9B). These differences were significantly different (P 0.05; Table 2, Fig. 9B) and represent a 16% amelioration of atrophy, showing that the combination (Fb/BDNF Fb/NT-3) grafts reduces, but does not prevent, atrophy in red nucleus neurons after chronic injury. To examine the distribution of mean cross-sectional area, red nucleus neurons were classified into three categories: small neurons ( m 2 ); medium neurons ( m 2 ); and large neurons ( m 2 ). The size distribution of uninjured red nucleus neurons was approximately 33% small, 42% medium, and 25% large, with no statistical difference between control (Fb/GFP or gelfoam alone) and experimental (Fb/BDNF Fb/NT-3) groups (P 0.05; Fig. 10A). In the injured red nucleus we detected a shift to smaller neurons in all groups (Fig. 10B). The injured red nucleus of control treated animals (Fb/GFP or gelfoam) contained 90% small neurons and 10% medium neurons, with no large cells detected; the injured red nucleus of combination-treated (Fb/BDNF Fb/NT-3) animals contained 79% small, 20% medium, and 1% large neurons. The differences between the number of small and medium-sized neurons present in the Fb/BDNF Fb/NT-3 and Fb/GFP (or gelfoam alone) groups were statistically different (P 0.05, for both), suggesting partial prevention or reversal of atrophy. We cannot exclude the possibility, however, that there was a preferential survival of smaller neurons compared to medium or large neurons. Discussion The results of this study show that a transplant of BDNF and NT-3 producing fibroblasts can stimulate limited growth from some intact descending and segmental host axons and significantly decrease detectable cell loss and atrophy of red nucleus neurons when the graft is delayed for 6 weeks after the initial spinal cord injury. We found, however, considerably less regeneration of rubrospinal axons into combination grafts than when neurotrophin secreting fibroblasts were grafted acutely into a lateral funiculus lesion (Liu et al., 1999) or after a 4-week delay into a complete hemisection lesion (Jin et al., 2002). Axonal growth Fibroblasts modified to produce neurotrophins have been reported to attract numerous types of host axons when placed acutely into a spinal cord injury site (Nakahara et al., 1996; Tuszynski et al., 1994, 1996; Grill et al., 1997b; Liu et al., 1999; Lu et al., 2001; Tobias et al., 2001). Among these populations are CGRP containing axons originating

12 108 C.A. Tobias et al. / Experimental Neurology 184 (2003) Fig. 7. Dark-field photomicrographs of GAP-43 immunostaining in transverse cervical spinal cord sections 14 weeks after delayed transplant (20 weeks after initial injury). GAP-43 immunostaining in (A) a high cervical segment (C2) rostral to the transplant and (C) a low cervical segment (C6) caudal to the transplant in sections from animals that received Fb/BDNF Fb/NT-3 transplants. GAP-43 immunostaining in high cervical (C2) sections of animals that received (B) Fb/GFP and (D) gelfoam alone as transplants and (E) the normal uninjured cord. Note the robust GAP-43 expression bilaterally and symmetrically in the dorsal corticospinal tracts rostral to a lesion site containing Fb/BDNF Fb/NT-3 transplant and decreased or absent staining in the normal uninjured cord, or rostral to Fb/GFP or gelfoam alone transplants, or caudal to Fb/BDNF NT-3 grafts. A E are DAPI counterstained. Scale bars: A E, 500 m. from dorsal root ganglion neurons, 5-HT containing axons originating from brainstem raphe neurons, and noradrenergic axons originating from locus coeruleus. The present results show ingrowth into the neurotrophin secreting grafts of CGRP-containing and 5-HT-containing axons which is comparable to that observed into neurotrophin grafts placed acutely (Liu et al., 1999). This suggests that several types of axons retain an undiminished ability to respond to grafts even when the grafts are delayed 6 weeks after injury. This conclusion is further supported by our qualitative and quantitative evidence of robust ingrowth of RT-97 and GAP-43 fibers into neurotrophin secreting grafts. There was no obvious difference among groups in the density of GFAP at the lesion borders, suggesting that scar formation was similar. Interestingly, robust GAP-43 immunoreactivity in the dorsal corticospinal tracts rostral to Fb/BDNF Fb/NT-3 grafts but not in the CST rostral to control grafts (Fb/GFP or gelfoam alone) suggested that delayed neurotrophin secreting transplants stimulated descending spinal tracts that were

13 C.A. Tobias et al. / Experimental Neurology 184 (2003) Fig. 8. Photomicrographs of rubrospinal and vestibulospinal tract axons labeled by BDA injections in parasagittal cervical spinal cord sections 14 weeks after delayed grafting of Fb/BDNF Fb/NT-3 or control transplants (20 weeks after initial injury). (A) A few rubrospinal axons have regenerated into the Fb/BDNT Fb/NT-3 transplant (arrows); (B) is a higher power view of the labeled axons (arrows). (C) Rubrospinal axons are present at the lesion border but do not enter the Fb/GFP graft. (D) A higher power view shows apparent retraction bulbs on some of the axons shown in C (arrows). (E) Vestibulospinal axons are absent in gelfoam grafts, but numerous axons (F) are present in Fb/BDNF Fb/NT-3 grafts. H is host; TP is transplant. Scale bars: A C, E, F, 100 m; D, 50 m. chronically injured (CST ipsilateral to the lesion) or intact (CST contralateral to the lesion). The regenerative capacity of chronically injured corticospinal axons is supported by a recent report using a fibroblast transplant similar to ours placed into chronic dorsal hemisection lesions demonstrating regenerative growth of anterogradely labeled CST axons (Tuszynski et al., 2003). Robust sprouting by undamaged corticospinal axons also has been noted following unilateral lesions to the corticospinal tract. Thallmair et al. (1998) reported that the intact corticospinal tract underwent sprouting in response to a lesion to the contralateral corticospinal tract (at the pyramid) in combination with the application of the IN-1 antibody, and Zhou et al. (2003) described growth of axons from the intact corticospinal tract to the denervated side after a similar lesion following transduction of spinal motoneurons with NT-3. Thus corticospinal axons are capable of regenerating and of sprouting in response to a lesion in the presence of an intervention that modifies the environment. Since GAP-43 expression is associated with active axonal growth, the results suggest that the CST axons remain responsive to neurotrophins even several months after injury and also that the grafts continue to secrete neurotrophins during this time. Anterograde tracing of the uninjured VST also revealed growth into delayed neurotrophin producing grafts, but not into control grafts. BDA-labeled VST axons in these grafts could include both previously uninjured axons sprouting from the contralateral VST or previously injured and there-

14 110 C.A. Tobias et al. / Experimental Neurology 184 (2003) Table 1 Total red nucleus neuron numbers Group Mean number of red nucleus neurons Left Right Ratio (L/R) Control treatment Experimental * Abbreviations used: Control treatment, C4 hemisection with either gelfoam alone or Fb/GFP as the transplants (N 6); experimental, C4 hemisection with Fb/BDNF Fb/NT-3 producing transplants (N 8). Values are mean numbers of neurons in the red nucleus SEM in Nisslstained tissue. Fb/BDNF Fb/NT-3 treatment rescued 32% of retrograde neuron loss in the axotomized red nucleus. The asterisk indicates significant difference from the control treated (injured) red nucleus using an unpaired Student s t test (P 0.05). fore regenerating axons that normally cross the midline at segmental levels within the spinal cord (Huisman et al., 1984; Bacskai et al., 2002; Matesz et al., 2002). Neurons in the lateral vestibular nucleus are known to express both the TrkB and TrkC receptors (King et al., 1999), which would enable them to respond to both BDNF and NT-3 secreted by the grafts. When we looked specifically for regeneration in the rubrospinal tract we found that few BDA-labeled fibers had penetrated into neurotrophin secreting transplants, but that none extended around or through the graft. Liu et al. (1999) found much greater regeneration into BDNF producing fibroblast grafts provided acutely in a lateral funiculus lesion and Jin et al. (2002) found regeneration into a similar graft placed into a hemisection after a 4-week delay that was greater than what we report but less extensive than that shown acutely. In both our study and the experiments described by Jin et al., graft placement into the injury site required a debridement with the extension of the lesion. The sparse regeneration suggests that some chronically injured CNS neurons have a poor capacity to respond to grafts provided 6 weeks after injury, even when they are stimulated by a second injury. Additional factors may therefore be required to induce these chronically injured axons to regenerate. It is possible that the decline in regenerative capacity after injury can be overcome by greater amounts of neurotrophins. For example, Coumans et al. (2001) supplemented fetal spinal cord transplants with BDNF or NT-3 at concentrations more than 100 times greater than the concentration per day provided by our grafts and reported regeneration of corticospinal tract axons through a thoracic transection site into lumbar segments when the grafts were inserted 2 or 4 weeks after the initial injury, while Kwon et al. (2002a) applied similarly large amounts of BDNF directly to the perikarya of injured red nucleus neurons and reported regeneration of rubrospinal tract axons into a peripheral nerve graft 1 year after initial injury. Another explanation for the limited regeneration in the present study could be the lack of an appropriate substrate on which regenerating axons could grow. Previous reports utilizing more limited injuries have shown patterns of regenerating growth of supraspinal axons that suggest preferential elongation along the gray white matter interface (Grill et al., 1997b; Liu et al., 1999; Brosamle et al., 2000). In the present study, the unilateral complete hemisection left, in addition to the grafts themselves, only the contralateral host spinal cord for regenerating axons to traverse. Jin et al. (2002), however, showed regenerative growth 4 weeks after transplantation into a hemisection lesion made 4 weeks previously. Thus the hemisection lesion itself does not prevent regeneration. Together these results suggest that some axons may have metabolic requirements that are greater after chronic than acute injury. Neuroprotection Cervical spinal cord injury produces retrograde effects in the red nucleus that include death or very severe atrophy of neurons and shrinkage of those that survive (Feringa et al., Fig. 9. Bar graphs showing mean red nucleus neuron survival and neuron size as represented by a ratio of injured (left) to uninjured (right) RN. (A) Mean number of surviving red nucleus neurons SEM in control treated (Fb/GFP or gelfoam alone; N 6) vs combination-treated (Fb/BDNF Fb/NT-3; N 8) animals. Counts were made on Nissl-stained sections. There was a 27% loss of detectable neurons in control treated (Fb/GFP or gelfoam alone) animals compared to 18% cell loss in combination-treated (Fb/BDNF Fb/NT-3) animals. These values were significantly different (P 0.05, Student s t test), indicating partial rescue by the Fb/BDNF Fb/NT-3 grafts. (B) Mean red nucleus neuron size SEM in control treated vs combination-treated animals determined from Nissl-stained sections. There was a 49% atrophy of neurons in control treated animals and 41% atrophy in combination-treated animals. These values were significantly different (P 0.05, Student s t test), suggesting partial prevention or reversal of atrophy by the Fb/BDNF Fb/NT-3 grafts.

15 C.A. Tobias et al. / Experimental Neurology 184 (2003) Table 2 Mean red nucleus neuron size Group Mean cell size of red nucleus neurons ( m 2 ) Left Right Ratio (L/R) Control Treatment Experimental * Abbreviations used: Control treatment, C4 hemisection with either gelfoam alone or Fb/GFP as the transplants (N 6); experimental, C4 hemisection with Fb/BDNF Fb/NT-3 producing transplants (N 8). Values are mean sizes ( m 2 ) of red nucleus neurons SEM in Nisslstained tissue. Fb/BDNF Fb/NT-3 prevented 16% of the atrophy in the axotomized red nucleus. An asterisk indicates significant difference from the control treated (injured) red nucleus using an unpaired Student s t test (P 0.05). 1988; Fukuoka et al., 1997; Kobayashi et al., 1997; Mori et al., 1997; Bregman et al., 1998; Novikova et al., 2000; Shibata et al., 2000; Liu et al., 2002). The present results show that fibroblasts modified to express BDNF and NT-3 transplanted 6 weeks after SCI produce a statistically significant rescue of red nucleus neurons that disappears when treatment consists of control fibroblasts or gelfoam alone and a significant preservation of the mean cell area of those that survived. Neuroprotective effects were less than those seen after acute administration of neurotrophin producing fibroblasts into a cervical lateral funiculus injury site (Liu et al., 2002). Previous reports have described increased survival and protection from atrophy in axotomized red nucleus neurons after delayed treatment with exogenously administered neurotrophins (Houle and Ye, 1999; Novikova et al., 2000; Kwon et al., 2002a). While the basic findings are similar to ours, some of the details differ. Differences in the percentages of rescue or loss in the red nucleus neuron number and cell size among these studies can be explained by the differences in counting methods, sites and amounts of neurotrophin delivery, delays following neurotrophin treatments, and the presence of a second axotomy. In addition, we counted red nucleus neurons from Nissl-stained sections using stereological counting methods, while two different retrograde tracing methods and counting protocols were used in the Houle and Ye (1999), and Novikova et al. (2000) studies. Aside from the differences in counting methods, there is also the possibility that inefficient retrograde transport of the tracer in chronically injured rubrospinal axons (Goshgarian et al., 1983; Tseng et al., 1995) may have affected the total number of cells that could effectively transport the tracer. Poor efficiency of retrograde tracing in injured axons could lead to undercounting of surviving red nucleus neurons. Our counting methods, based on unbiased stereological techniques and clear criteria for identifying neurons in Nissl-stained sections, demonstrate that delayed transplants of BDNF and NT-3 producing fibroblasts significantly reduce the atrophy and loss of red nucleus neurons after complete cervical hemisection. We do not rule out the possibility that some injured red nucleus neurons may have atrophied so severely that they were not identified as neurons using our counting method. Our results show that a combination of BDNF and NT-3 producing fibroblasts rescued axotomized red nucleus cells from atrophy and/or loss and induced growth of host axons when the transplant was delayed for 6 weeks after spinal cord injury. The companion paper shows that these effects were associated with only a modest recovery of function (Shumsky et al., 2003). Regeneration and recovery were less robust than with neurotrophin producing fibroblasts transplanted acutely (Liu et al., 1999) or 4 weeks after injury (Jin et al., 2002) or with fetal spinal cord transplanted 2 or 4 weeks after injury supplemented with exogenous neurotrophins (Coumans et al., 2001). In general these results favor early intervention and suggest that additional combination strategies will be required to achieve levels of Fig. 10. Bar graphs showing mean red nucleus neuron size distribution of uninjured (A) and injured (B) red nucleus neurons in control (Fb/GFP or gelfoam) and combination-treated (Fb/BDNF Fb/NT-3) animals. The uninjured control red nucleus (right) is evenly distributed into small ( m 2 ), medium ( m 2 ), and large ( m 2 ) neurons. (A) The mean percentages of small, medium, and large neurons present in the intact red nucleus of rats that received Fb/GFP (or gelfoam alone) or Fb/BDNF Fb/NT-3 were not statistically different, P 0.05, unpaired Student s t test, error bars SEM. (B) After injury and delayed grafting the distribution shifted toward smaller cells. Combination-treated (Fb/ BDNF Fb/NT-3) animals had significantly fewer neurons present in the small size range and a significantly greater number of neurons in the medium range compared to control treated (Fb/GFP or gelfoam alone) animals (P 0.05, unpaired Student s t test, error bars SEM). These results are consistent with partial prevention or rescue from atrophy.