Departamento de Fisiología Médica y Biofísica Universidad de Sevilla
Single‐cell transcriptomic and functional characterization of cortical parvalbumin interneurons in a novel conditional knock‐out mouse lacking CSPα/DNAJC5
Marina Valenzuela Villatoro
Advisors: Dr. Rafael Fernández Chacón Dr. Pablo García‐Junco Clemente
Doctoral Thesis
Sevilla, 2019
Departamento de Fisiología Médica y Biofísica
Universidad de Sevilla
Avd. Sánchez Pizjuán, 4
41009, Sevilla
España
Dr. Rafael Fernández Chacón, Catedrático del Departamento de Fisiología Médica y Biofísica de la Universidad de Sevilla y Dr. Pablo García‐Junco Clemente, Investigador Ramón y Cajal del Departamento de Fisiología Médica y Biofísica de la Universidad de Sevilla, CERTIFICAN que Marina Valenzuela Villatoro ha realizado bajo su dirección el trabajo titulado “Single‐cell transcriptomic and functional characterization of cortical parvalbumin interneurons in a novel conditional knock‐out mouse lacking CSPα/DNAJC5” que presenta para optar al grado de Doctor por la Universidad de Sevilla.
Fdo: Dr. Rafael Fernández Chacón Fdo: Dr. Pablo García‐Junco Clemente
Sevilla, 2019
TABLE OF CONTENTS
TABLE OF CONTENTS ...... I SUMMARY OF TABLES ...... II SUMMARY OF FIGURES ...... III SUMMARY OF BOXES ...... IV ABBREVIATIONS ...... V ABSTRACT ...... 1 INTRODUCTION ...... 3 1. Background: the synapse and the synaptic vesicle ...... 5 2. The synaptic vesicle cycle ...... 7 3. Membrane fusion by SNARE proteins ...... 8 4. Cysteine string protein alpha (CSPα) ...... 9 4.1. Localization and molecular structure ...... 9 4.2. The trimeric complex CSPα‐Hsc70‐SGT is essential for chaperone activity ...... 10 4.3. Phenotypic characterization of CSPα knock‐out (KO) mice ...... 11 4.4. SNAP25 is an important substrate of CSPα ...... 12 4.5. Role of CSPα in GABAergic synapses firing at high frequency ...... 13 4.6. Importance of CSPα in human disease ...... 14 5. The cerebral cortex ...... 15 5.1. Cortical organization ...... 15 5.2. Cortical glutamatergic neurons ...... 17 5.3. Cortical GABAergic interneurons ...... 17 5.3.1. Cortical interneuron classification ...... 17 5.4. Cortical connectivity: pyramidal neurons versus GABAergic interneurons ...... 19 6. Parvalbumin interneurons ...... 20 6.1. Morphological and functional properties of PV dendrites ...... 21 6.2. Morphological and functional properties of PV axon ...... 22 6.3. Specializations of PV cells for fast synaptic signaling ...... 23 6.4. PV‐derived feedforward and feedback inhibition ...... 24 6.5. Role of PV interneurons in modulating animal behavior ...... 24 6.6. Role of PV interneurons in neurological disorders ...... 25 7. Single‐cell RNA sequencing ...... 25 7.1. First milestones ...... 25 7.2. Cell isolation and sequencing methods ...... 26 7.3. Up‐to‐date PV classification by scRNA‐seq ...... 27
I 7.4. Transcriptomics in neurodegenerative diseases ...... 27 GOALS ...... 29 MATERIALS AND METHODS...... 33 1. Mice ...... 35 1.1. Generation of conditional Dnajc5flox/flox mice ...... 35 1.2. Generation of UBC‐Cre‐ERT2:Dnajc5flox mice ...... 36 1.3. Generation of PVcre:Ai27D:Dnajc5flox mice ...... 36 2. Fluorescent‐activated cell sorting (FACS) ...... 37 3. Validation of conditional Dnajc5flox/flox mice ...... 39 3.1. Genomic DNA (gDNA) validation ...... 40 3.1.1. gDNA extraction ...... 40 3.1.2. PCR for Dnajc5 gDNA ...... 41 3.2. RNA validation ...... 41 3.2.1. RNA extraction ...... 41 3.2.2. Reverse‐transcription (RT) to cDNA ...... 42 3.2.3. PCR for Dnajc5 cDNA ...... 43 3.3. Protein validation ...... 43 4. Real‐time quantitative reverse transcription‐PCR (real‐time qRT‐PCR) ...... 44 4.1. RNA extraction ...... 44 4.2. Reverse‐transcription and cDNA amplification ...... 44 4.3. Real‐time qRT‐PCR and quantification ...... 45 5. Survival analysis ...... 45 6. Body weight curve ...... 45 7. Open field ...... 46 8. Immunoblotting ...... 46 9. Quantification of protein levels ...... 47 10. Perfusion and sectioning using a cryostat ...... 48 11. Brigth field immunohistochemistry ...... 48 12. Quantification of PV somata number ...... 49 13. Immunofluorescence in brain slice sections ...... 49 14. Colocalization analysis ...... 49 15. Quantification of synaptic puncta ...... 50 16. Quantification of PV somata size...... 50 17. Electrophysiology ...... 51 17.1. Preparation of brain slices ...... 52 17.2. Electrophysiological recordings of parvalbumin interneurons ...... 52
I 17.2.1. Intrinsic properties ...... 52 17.2.2. Excitatory postsynaptic potentials (EPSPs)...... 53 17.3. Electrophysiological recordings of pyramidal neurons ...... 53 18. Neuron dissociation ...... 53 19. Single‐cell RNA sequencing protocol ...... 55 19.1. Steps 1‐2. FACS into a WaferGen 9600 microwell plate ...... 56 19.2. Steps 2‐3. Lysis and reverse transcription ...... 57 19.3. Steps 4‐5. Indexed PCR and extraction ...... 59 19.4. Step 6.Tagmentation and 5′ fragments isolation ...... 60 19.5. Step 7.Illumina sequencing ...... 60 19.6. Preliminary filtering of raw reads ...... 61 20. Single‐cell RNA sequencing data analysis ...... 63 21. Gene ontology (GO) analysis ...... 64 22. Statistics ...... 64 RESULTS ...... 65 1. Cortical synaptic dysfunctions in Dnajc5 conventional KO mice ...... 67 2. Specific Cre‐recombinase activity at the Dnajc5 locus in the PVcre:Ai27D:Dnajc5flox mouse line ...... 69 3. tdTomato expression is specific for the parvalbumin‐positive interneuron population ...... 74 4. The specific deletion of Dnajc5 in PV interneurons does not cause increased mortality but progressive body weight alterations in PVcre:Ai27D:Dnajc5flox/‐ mice ...... 77 5. Neurological phenotype upon deletion of Dnajc5 in PV interneurons ...... 79 6. Cortical expression of presynaptic CSPα/DNAJC5, synaptotagmin‐2 and Hsc70 proteins are progressively reduced in PVcre:Ai27D:Dnajc5flox/‐ mice ...... 84 7. Progressive loss of Syt2+ and PV+ presynaptic puncta in PVcre:Ai27D:Dnajc5flox/‐ mice with similar PV cell number and size ...... 85 7.1. Age‐dependent loss of PV+ and Syt2+ synaptic puncta in PVcre:Ai27D:Dnajc5flox/‐ mice ...... 85 7.2. The progressive degeneration of presynaptic terminals in PVcre:Ai27D:Dnajc5flox/‐ mice is not related to changes in the number of PV cells or soma size ...... 90 8. Electrophysiological analysis of PV interneurons shows similar intrinsic properties but changes in spontaneous synaptic responses in PVcre:Ai27D:Dnajc5flox/‐ animals ...... 91 8.1. Intrinsic properties of parvalbumin interneurons ...... 91 8.1.1. Passive properties of PV cells ...... 92 8.1.2. Active properties of PV cells ...... 93 8.1.3. Excitatory postsynaptic potentials (EPSPs) ...... 99 8.2. Miniature inhibitory postsynaptic currents (mIPSCs) on cortical pyramidal cells...... 100 9. Single‐cell RNA sequencing of cortical parvalbumin interneurons ...... 104 9.1. Robust average mapped mRNA reads per cell ...... 106
I 10. Analysis of transcriptomic data obtained from single cortical parvalbumin interneurons ...... 106 10.1. A readable matrix to visualize genes expressed by every single cell ...... 106 10.2. Integration of single‐cell transcriptomic data across different genotypes ...... 107 10.2.1. Cell filtering: quality control by filtering out data from low quality cells ...... 108 10.2.2. Normalization and noise reduction of individual genes expression in single cells ...... 109 10.2.3. Clustering analysis reveals different subpopulations of PV interneurons ...... 112 10.2.3.1. First clustering ...... 112 1) Selection of Pvalb mRNA expression ...... 112 2) Determination of High Variable Genes (HVG) ...... 113 3) Running multi‐set CCA and CC selection for cell clustering ...... 113 4) Alignment of CCA subspaces ...... 114 5) Cell clustering and non‐linear dimensional reduction (tSNE) ...... 115 6) Identification of conserved cell type markers ...... 117 10.2.3.2. Second clustering ...... 119 2) Determination of high variable genes (HVG) ...... 119 3) Running multi‐set CCA and CC selection ...... 120 4) Alignment of CCA subspaces...... 121 5) Cell clustering and non‐linear dimensional reduction (tSNE) ...... 122 6) Identification of conserved cell type markers ...... 122 10.2.3.3. Third clustering ...... 124 2) Determination of high variable genes (HVG) ...... 124 3) Running multi‐set CCA and CC selection ...... 124 4) Alignment of CCA subspaces ...... 125 5) Cell clustering and non‐linear dimensional reduction (tSNE) ...... 126 6) Identification of conserved cell type markers ...... 127 10.2.3.4. Fourth clustering ...... 129 2) Determination of high variable genes (HVG) ...... 129 3) Running multi‐set CCA and CC selection ...... 129 4) Alignment of CCA subspaces ...... 131 5) Cell clustering and non‐linear dimensional reduction (tSNE) ...... 132 6) Identification of conserved cell type markers ...... 132 10.2.4. Selection of cluster markers for the 8 detected PV identities ...... 132 10.2.5. Number of cells per genotype across PV identities ...... 145 10.2.6. Control for specific cell‐type markers across genotypes and PV clusters ...... 147 10.2.7. Dnajc5 expression in PV cells ...... 151 10.3. Differential expression (DE) analysis ...... 153
I 10.3.1. DE analysis including all PV cells ...... 153 10.3.2. DE analysis by PV clusters ...... 158 1) Comparison between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ PV clusters ..... 158 2) Comparison between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT PV clusters ...... 163 3) Comparison between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT PV clusters ...... 166 DISCUSSION ...... 173 1. The PVcre:Ai27D:Dnajc5flox/‐ as a mouse model to study CSPα/DNAJC5 in PV interneurons ...... 175 2. PVcre:Ai27D:Dnajc5flox/‐ mice suffer from a progressive age‐dependent neurological phenotype with hyperactivity and ataxia ...... 176 3. Selective reduction of presynaptic proteins in PVcre:Ai27D:Dnajc5flox/‐ mice ...... 177 4. Molecular signs of presynaptic degeneration in the cortex of PVcre:Ai27D:Dnajc5flox/‐ mice ...... 178 5. Subtle alterations of excitability in cortical PV interneurons lacking CSPα/DNAJC5 ...... 179 6. Normal excitatory synaptic inputs on PV interneurons lacking CSPα/DNAJC5 ...... 181 7. GABA release alterations in PV interneurons from PVcre:Ai27D:Dnajc5flox/‐ mice ...... 181 8. The use of UMIs and STRT‐seq‐2i/WaferGen technology as a novel and powerful method for sequencing ...... 183 9. Prior to data analysis: read depth, raw data classification and cell filtering by QC metrics ...... 184 10. The basis to support an integrated analysis and reasons for successive clustering steps ...... 186 11. The importance of an appropriate CC selection for clustering ...... 188 12. PV cells cluster into 8 different identities ...... 189 13. Dnajc5 expression in the PV population of different genotypes ...... 191 14. Differential expression analysis related to the absence of CSPα/DNAJC5 ...... 192 14.1. Gene ontology (GO) analysis reveals alterations in synaptic and metabolic pathways ...... 193 14.1.1. Dysregulation of synaptic processes in the absence of CSP/DNAJC5 ...... 193 14.1.2. Up‐regulation of "Glycolysis/Gluconeogenesis" in the absence of CSP/DNAJC5 ...... 193 14.2. Up‐regulation of the gene encoding the neuropeptide Y (Npy) ...... 194 14.3. Expression of sex‐specific genes ...... 196 14.4. Up‐regulated genes with consistent expression changes ...... 196 14.5. Up‐regulated genes potentially related to CSP/DNAJC5 ...... 197 14.6. Down‐regulated genes related to neuronal excitability and GABAergic transmission ...... 198 CONCLUSIONS ...... 201 BIBLIOGRAPHY ...... 205
I
SUMMARY OF TABLES
Table 1. Primers for mouse genotyping...... 40 Table 2. Primers for Dnajc5 gDNA and mRNA detection...... 43 Table 3. TaqMan Gene Expression Assay probes...... 45 Table 4. Primary antibodies...... 51 Table 5. Custom primers for STRT‐seq‐2i protocol...... 61 Table 6. Summary of cortical protein levels in Dnajc5 WT and KO mice by western blot...... 68 Table 7. Summary of colocalization values between parvalbumin and tdTomato expression in the experimental mice...... 76 Table 8. Summary of expression levels for PV mRNA in sorted cell fractions obtained from the experimental mice...... 77 Table 9. Summary of body weight values in the experimental mice (males and females)...... 79 Table 10. Summary of the open field assays performed in the experimental mice (males and females). . 81 Table 11. Summary of the open field assays performed in the experimental mice (all together)...... 82 Table 12. Summary of cortical protein levels in the experimental mice by western blot...... 85 Table 13. Summary of PV+ and Syt2+ synaptic puncta density in the experimental mice at 2 months. .... 87 Table 14. Summary of PV+ and Syt2+ synaptic puncta density in the experimental mice at 8 months. .... 88 Table 15. Summary of the quantitative analysis regarding of PV cell number and soma size in the experimental mice...... 90 Table 16. Summary of the passive properties of PV cells in the experimental mice...... 92 Table 17. Summary of AP properties of PV cells in the experimental mice...... 93 Table 18. Summary of AHP properties of PV cells in the experimental mice...... 95 Table 19. Summary of rheobase values of PV cells in the experimental mice...... 96 Table 20. Summary of EPSP properties of PV cells in the experimental mice...... 100 Table 21. Summary of mIPSC properties of pyramidal cells in the experimental mice...... 102 Table 22. Mapped reads from scRNA‐seq in WaferGen chips...... 106 Table 23. Summary showing the average numbers for nUMI, nGene and percent.mito per genotype. . 109 Table 24. Summary of conserved genes obtained after the final clustering analysis for PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV genotypes...... 142 Table 25. Number of cells across PV identities from PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice...... 146 Table 26. Average Dnajc5 expression for PV cells by genotype...... 151 Table 27. Summary of significant DE genes obtained by comparison of total PV cells between genotype pairs...... 154 Table 28. GO analysis for DE genes obtained by comparison of total PV cells between genotype pairs. 158 Table 29. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ mice...... 159
II Table 30. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ genotypes...... 163 Table 31. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT mice...... 164 Table 32. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT genotypes...... 166 Table 33. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT mice...... 168 Table 34. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT genotypes...... 172
II SUMMARY OF FIGURES
Figure 1. Trafficking proteins at the synaptic vesicle...... 6 Figure 2. The synaptic vesicle cycle...... 8 Figure 3. CSPα‐Hsc70‐SGT trimeric complex...... 11 Figure 4. CSPα KO mice phenotype...... 12 Figure 5. CSPα‐Hsc70‐SGT chaperone complex functioning...... 13 Figure 6. Predicted model of inhibitory and excitatory synaptic changes upon CSPα removal in hippocampal cultures...... 14 Figure 7. Diversity, classification and properties of neocortical GABAergic interneurons...... 19 Figure 8. Individual neuronal contributions among cortical interneurons...... 20 Figure 9. Axonal‐derived classification and electrophysiological properties of AP firing patterns in PV cells...... 23 Figure 10. Breeding strategy to generate the PVcre:Ai27D:Dnajc5flox mouse line...... 39 Figure 11. Mouse body identification and types of head movements analyzed during the open field experiments...... 46 Figure 12. Scheme of single‐cell RNA sequencing using a WaferGen plate...... 56 Figure 13. Graphic representation of well position patterns for cell sorting into a WaferGen platform. .. 59 Figure 14. Detailed STRT‐seq‐2i method using FACS to sort cells into a WaferGen plate...... 62 Figure 15. Cortical expression of presynaptic proteins from Dnajc5 WT and KO mice...... 68 Figure 16. Strategies to validate the deletion of Dnajc5 floxed gene upon Cre‐recombinase activity driven by Pvalb promoter at gDNA and mRNA levels...... 71 Figure 17. Validation of the floxed allele recombination by Cre activity at the Dnajc5 locus in tdTomato+ sorted cells...... 73 Figure 18. Analysis of CSPα/DNAJC5 protein in liver lysates from UBC‐cre‐ERT2:Dnajc5flox/flox mice...... 74 Figure 19. Analysis of CSPα/DNAJC5 protein in cortical lysates from both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice...... 76 Figure 20. Motor cortical neurons show high colocalization of PV and tdTomato labelling...... 78 Figure 21. Survival curve and body weight screening of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice until 8 months postnatally...... 80 Figure 22. Neurological phenotype upon deletion of Dnajc5 in PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months of age...... 82 Figure 23. Global expression of presynaptic proteins in cortical slices from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally...... 84 Figure 24. Cortical expression of presynaptic proteins from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally...... 86 Figure 25. Quantification of PV+ and Syt2+ synaptic puncta from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally...... 88
III Figure 26. Quantification of the number and soma size of PV interneurons from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally...... 90 Figure 27. Passive electrophysiological properties of layer II/III parvalbumin interneurons in motor cortical slices...... 95 Figure 28. Active electrophysiological properties of layer II/III parvalbumin interneurons in motor cortical slices...... 96 Figure 29. Excitability properties of layer II/III parvalbumin interneurons in motor cortical slices...... 99 Figure 30. Excitatory postsynaptic potentials (EPSPs) of layer II/III parvalbumin interneurons in motor cortical slices...... 101 Figure 31. Miniature inhibitory postsynaptic currents (mIPSCs) of layer II/III pyramidal cells in motor cortical slices...... 102 Figure 32. FAC‐sorting settings for neuron isolation during single‐cell RNA sequencing experiments. ... 105 Figure 33. Filtering out low quality cells from scRNA sequencing data...... 111 Figure 34. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the first clustering...... 115 Figure 35. First clustering of cells from scRNA sequencing data...... 117 Figure 36. Cluster markers defining PV interneuron populations after the first clustering step...... 118 Figure 37. Analysis of identities 3 and 8 obtained from the first clustering...... 120 Figure 38. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the second clustering...... 121 Figure 39. Second clustering of cells from scRNA sequencing data...... 122 Figure 40. Cluster markers defining PV interneuron populations after the second clustering step...... 123 Figure 41. Analysis of identity 8 obtained from the second clustering...... 124 Figure 42. Multi‐set CCA running, CCs selection and CCA subspaces alignmente during th third clustering...... 126 Figure 43. Third clustering of cells from scRNA sequencing data...... 127 Figure 44. Cluster markers identifying PV interneuron populations after the third clustering step...... 128 Figure 45. Analysis of identity 4 obtained from the third clustering...... 130 Figure 46. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the fourth clustering...... 131 Figure 47. Fourth clustering of cells from scRNA sequencing data...... 133 Figure 48. Cluster markers identifying PV interneuron populations after the fourth clustering step...... 134 Figures 49 & 50. Comparison of the PV clusters 0 and 1 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types...... 135 Figures 51 & 52. Comparison of the PV clusters 2 and 3 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types...... 136 Figures 53 & 54. Comparison of the PV clusters 4 and 5 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types...... 137
III Figures 55 & 56. Comparison of the PV clusters 6 and 7 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types...... 138 Figure 57. Expression of cluster markers defining the PV subclasses in the Allen Brain Atlas of mouse cell types into our experimental PV identities...... 140 Figure 58. Identifying PV interneuron populations after the fourth clustering step using the second cluster markers...... 141 Figure 59. Final identity names for PV clusters after sequential clustering steps...... 143 Figure 60. Expression of cluster markers that define PV identities within all PV populations using tSNE dimensional reduction...... 144 Figure 61. Top‐5 conserved markers defining PV populations across all PV identities...... 145 Figure 62. Quantification of PV cell number per cluster across genotypes...... 147 Figure 63. Control analysis to determine the expression of specific cell‐type markers within PV cells classified by genotype...... 149 Figure 64. Expression of Syt2 and Pvalb genes across genotypes and PV identities...... 150 Figure 65. Expression of Dnajc5 gene across genotypes and PV identities...... 152 Figure 66. Volcano plots showing significant DE genes found after comparison of all PV cells between genotype pairs...... 156 Figure 67. Volcano plots showing significant DE genes detected after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ genotypes...... 160 Figure 68. Volcano plots showing significant DE genes found after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT genotypes...... 164 Figure 69. Volcano plots showing significant DE genes detected after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT genotypes...... 168
III
SUMMARY OF BOXES
BOX 1 ...... 104 BOX 2 ...... 105 BOX 3 ...... 106 BOX 4 ...... 107 BOX 5 ...... 108 BOX 6 ...... 113 BOX 7 ...... 117 BOX 8 ...... 153
IV
ABBREVIATIONS
5HT3aR Ionotropic serotonin receptor aa Amino acids
ACSF Artificial cerebrospinal fluid
AD Alzheimer’s disease
ADP Adenosine diphosphate
AHP Afterhyperpolarization
AMPA α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid
ANCL Adult‐onset neuronal ceroid lipofuscinosis
AP Action potential
ATP Adenosine triphosphate bp Base pairs
BSA Bovine serum albumin
CC Correlation vector
CCA Canonical correlation analysis
CCK Cholecystokinin cDNA Complementary deoxyribonucleic acid
CEF Cell expression format
ChR2 Channelrhodopsin‐2
CNS Central nervous system
CSPα Cysteine string protein alpha
CSV Comma‐separated values
DE Differentially expressed
DNA Deoxyribonucleic acid dNTP Deoxynucleotides triphosphate
EDTA Ethylenediaminetetraacetic acid
EPSP Excitatory postsynaptic potential
V ER Endoplasmic reticulum
ERCCs External RNA controls consortium
ERT2 Estrogen receptor ligand binding domain
ES Embryonic stem
ESCG Eukaryotic single cell genomics
FACS Fluorescent‐activated cell sorting
FBS Fetal bovine serum
FC Fold change
FS Fast‐spiking
FSC Forward scatter
GABA Gamma‐aminobutyric acid gDNA Genomic deoxyribonucleic acid
GO Gene ontology
Hemi Hemizygous
Het Heterozygous
Homo Homozygous
HPD Histidine, proline and aspartic (acid residues) motif
Hsc70 Heat‐shock cognate 70 kDa protein
HVG High variable genes
IHC Immunohistochemistry
KI Karolinska Institute
KNN K‐Nearest neighbors
KO Knock‐out mEPSC Miniature excitatory postsynaptic current mIPSC Miniature inhibitory postsynaptic current mRNA Messenger ribonucleic acid
NGFC Neurogliaforms cells
V NMDA N‐methyl‐D‐aspartate
NMDG N‐methyl‐D‐glucamine
NSF N‐ethylmaleimide sensitive factor
NPY Neuropeptide Y ns Non‐significant
NT Neurotransmitter
O/N Over night
PB Phosphate buffer
PBS Phosphate buffered saline
PCA Principal components analysis
PCR Polymerase chain reaction
PFA Paraformaldehyde
PPT1 Palmitoyl‐protein thioesterase 1
PSD Postsynaptic density
PV Parvalbumin
PVDF Polyinylidene difluoride
QC Quality control qRT‐PCR Quantitative reverse transcription polymerase chain reaction
RIN RNA integrity number
RMP Resting membrane potential
RNA Ribonucleic acid
RPKM Reads per kilobase per million mapped reads rRNA Ribosomal ribonucleic acid
RT Reverse‐transcription
SBD Substrate‐binding domain scRNA‐seq Single‐cell RNA sequencing
SDS Sodium dodecyl sulfate
V SDS‐PAGE Sodium dodecyl sulfate polyacrylamide gel electrophoresis
SEM Standard error of the mean
SGT Small glutamine‐rich TRP‐containing protein
Smart Switching mechanism at the 5′ end of the RNA transcript
SNAP25 Synaptosomal‐associated protein 25 KDa
SNAPs Soluble NSF‐attachment proteins
SNARE Soluble NSF‐attachment protein receptors
SNN Shared nearest neighbor
SSC Side scatter
Sst Somatostatin
STRT Single‐cell tagged reverse transcription
SV Synaptic vesicle
TBS‐T Tris‐buffered saline and tween 20 tdT tdTomato
TMX Tamoxifen
TRP Tetratricopeptide repeat tSNE t‐distributed stochastic neighbor embedding
TTX Tetrodotoxin
UBC Ubiquitin C
UMI Unique molecular identifier
UV Ultraviolet
VGat Vesicular GABA transporter
VGlut Vesicular glutamate transporter
Vip Vasoactive intestinal peptide
WB Western blot
WT Wild type
V ABSTRACT
Neurodegenerative diseases are progressive disorders currently without cure. Indeed, the molecular mechanisms underlying synaptic and neuronal degeneration are poorly understood. Cysteine String Protein‐α/DNAJC5 (CSPα/DNAJC5) is a synaptic co‐chaperone associated to neurodegeneration in humans. Adult‐onset autosomal dominant neuronal ceroid lipofuscinosis is caused by mutations in the gene DNAJC5. In mice, the genetic removal of CSPα/DNAJC5 leads to activity‐dependent synaptic degeneration, specially evident in the fast‐spiking GABAergic interneurons that express parvalbumin (PV). The early death of these mice limits the investigation of the natural history of the synaptic and neuronal dysfunction of these cells. We have now generated a conditional KO mouse (PVcre:Ai27D:Dnajc5flox/‐) lacking CSPα/DNAJC5 specifically in parvalbumin‐positive GABAergic interneurons labeled with the optogenetic actuator channelrhodopsin‐2 fused to the fluorescent reporter tdTomato. These mice develop a progressive neurological phenotype characterized by hyperactivity and ataxia without early lethality. At the neocortex, those neurons form disorganized perisomatic synapses with a reduced number of puncta, that, unexpectedly do not compromise the number of PV somata up to 8 months of age. Those neurons are, however, more reluctant to initiate high frequency action potential trains compared to controls. The spontaneous release of GABA from those neurons is altered. The frequency of mIPSCs is decreased, consistent with a lower number of functional synapses. The amplitude of mIPSCs is strikingly reduced and the kinetics modified, perhaps caused by alterations in vesicular GABA content and/or postsynaptic changes in GABA receptors. Further characterization of ion currents and pharmacological and optogenetic dissection of synaptic currents is required to understand the changes in excitability and synaptic transmission. In order to get insight into the homeostatic mechanisms of gene expression associated to synaptic dysfunction in PV interneurons lacking CSPα/DNAJC5, we have carried out the analysis of single cell transcriptomes of tdTomato+ cells isolated from the whole cortex of three different genotypes: PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐. In collaboration with Dr. Hjerling‐ Leffler's group (Karolinska Institute), we have combined fluorescence‐activated cell sorting (FACS), the STRT‐seq‐2i method and the WaferGen 9600‐well platform as a novel and powerful strategy for single‐cell RNA sequencing. Next, we have conducted computational analysis using the R toolkit Seurat. We have integrated scRNA‐seq data from the three genotypes (2847 cells) and, based on common sources of variation, we have identified 8 different populations (also known as identities) of PV interneurons and carried out a downstream comparative analysis of gene expression. Three of our PV identities were highly similar to PV subclasses recently described in specific cortical areas and one of them corresponded to chandelier cells. The gene ontology (GO) study based on the differential gene expression analysis suggests genetic dysregulation of metabolism and synaptic function, among other processes, in PVcre:Ai27D:Dnajc5flox/‐ neurons. Amid a wide repertoire of expression changes, our study has identified
1 genes related to brain disorders and to the control of hyperexcitability. Those presumptive changes, however, still require validation with alternative methodologies before they can be used to explain the molecular mechanisms underlying the neuronal and the in vivo phenotypes found in the mice. Overall, the results presented in this thesisw thro light and open multiple new perspectives to advance the understanding of molecular and circuit mechanisms underlying synaptic alterations of PV interneurons and the role of CSPα/DNAJC5 in brain disorders.
Acknowledgments
This thesis was supported by the Spanish Ministry of Economy and Competitiveness (MINECO) (Grants BFU2013‐47493‐P and BFU2016‐76050‐P, FPI Fellowship BES2014‐070405, and short scientific visit EEBB‐I‐16‐10890), CIBERNED (Instituto de Salud Carlos III), ITRIBIS, Junta de Andalucía (Grants P12‐CTS‐ 2232 and CTS‐600) and Fondo Europeo de Desarrollo Regional. ES cells were obtained from EUCOMM.
2 INTRODUCTION
Introduction
1. Background: the synapse and the synaptic vesicle
Our current knowledge about the central nervous system derives from a summation of scientific observations performed through different research lines, even at distinct temporal windows that finally meet together. It is not appropriate to attribute the beginning of any discipline to a single author, but to all these talented and committed investigators that provided us with fruitful observations and discoveries, which at the end forged a collaborative vision of Science. Several researchers have contributed to the emergence of Neuroscience as an unique field of study; for example, Camillo Golgi, who published his microscopy technique showing a method to stain cells using chrome silver, opening the branch of physiology to other researchers; Santiago Ramón y Cajal, who postulated the “neuron doctrine” proposing the individuality of neurons in the brain, in contrast to the previous presumption as a continuous tissue (reticular theory), and who shared the Nobel Prize in 1906 with Golgi; and Charles Sherrington, who first published, together with Michael Foster in the seventh edition of A textbook of Physiology in 1897, the term “synapse” to name the “vibrating touch” of an axon terminal of one cell with a dendrite or body of another cell (Foster, 1897; Fulton, 1938; Tansey, 1997).
Synapses are junction sites that mediate the communication between neurons and/or excitable cells through the propagation of signals with high speed and precision. Although synapses are defined as electrical or chemical depending on how transmission occurs (direct propagation of electrical stimulus versus chemical substances as intermediaries, respectively), in the central nervous system the majority of synapses are chemical. This type of synapse involves chemical intermediaries that mediate the transmission of electric impulses, and was first postulated by Langley in 1921 (Cowan et al., 2001; Langley, 1921). The presynaptic electrical signal evokes the secretion of a chemical substance into the synaptic cleft, which is then reconverted postsynaptically into an electrical signal. Transmitter release applies to all presynaptic terminals where these chemical transmitters are synthesized, and where calcium develops an important role in the secretory process, at least in neuronal transmitter release (Katz, 1969).
Interestingly, it exists a huge distance from nerve terminals to the cell body, which may lead to problems to supply new components to the terminals when they get exhausted, because the bulk of protein synthesis happens at the perikaryal‐dendritic region (where RE and Golgi complexes are located). The neuron strategy to bridge this issue consists of a prolonged half‐life of axonal proteins and the use of a specialized secretory machine: the synaptic vesicle (Cowan et al., 2001). Synaptic vesicles were immediately related to chemical synapses since discovered by electron microscopy (Palay and Palade, 1955).
Synaptic vesicles are small organelles (normally between 35‐50 nm) that store neurotransmitters (NTs) such as glutamate, GABA (gamma‐aminobutyric acid), glycine or acetylcholine, and release their
5 Introduction kk content when fusing with the plasma membrane of the presynaptic terminal during neuronal depolarization (Cowan et al., 2001). They have a protein‐phospholipid composition of 1:3 ratio. Probably, among all proteins that have been associated with synaptic vesicles, a small number are constitutively present in spite of those which bind transiently or are present only in a subset of vesicles (Südhof, 2004). Among them, we highlight synaptobrevins, synaptotagmins, Rab3, synapsins, SCAMPs, synaptophysins, SV2A and CSPα, within others (Fernández‐Chacón and Südhof, 1999) (Figure 1). CSPα (Cysteine String Protein alpha) is associated to the external site of the synaptic vesicle membrane and involves the principal subject of this thesis; we will focus on it later.
Although the molecular machinery of NT release has been an extensive object of study in the last years for many groups – vesicular trafficking has been awarded with the Nobel Prize to J. E. Rothman, R. W. Schekman and T. C. Südhof in 2013 (Südhof, 2014) ‐, it still needs further investigations to completely elucidate the functioning of many proteins at the synapse.
Figure 1. Trafficking proteins at the synaptic vesicle. The structure of the major trafficking proteins of synaptic vesicles are shown in this scheme. These proteins have known functions in the synaptic vesicle cycle. Proteins involved in NT uptake are not included in the scheme. Many of the proteins shown are encoded by at least two genes, generating several isoforms. Only one isoform of each protein is displayed in the figure. The blue circle denotes cysteine string protein (CSP) including its DNA‐J domain. C, C‐terminus; N, N‐terminus; P, phosphate; ATP, adenosine triphosphate; ADP, adenosine diphosphate; SCAMPs, secretory carrier membrane proteins; SV(2A‐2B‐2C), synaptic vesicle glycoprotein; SVOP, SV2 related protein; Rab(3‐5), member RAS oncogene family (Modified from Fernández‐Chacón and Südhof, 1999).
6
Introduction
2. The synaptic vesicle cycle
Presynaptic terminals contain hundreds of synaptic vesicles filled with NT. When an action potential (AP) depolarizes the presynaptic plasma membrane, synaptic transmission is initiated by inducing the opening of voltage‐dependent Ca2+ channels, which allows calcium influx into the nerve terminal. The calcium influx finally triggers synaptic vesicle exocytosis through synaptic vesicle fusion with the presynaptic plasma membrane, thereby releasing NTs into the synaptic cleft (Südhof, 2004; Südhof and Rizo, 2011). This process is achieved by recruitment and docking of synaptic vesicles at release sites, their transformation into a fusion‐competent “primed” state (Ca2+ sensitive), and tethering of Ca2+ channels next to docking sites (Südhof, 2014). Synaptic vesicle fusion takes place in a specialized region called active zone, an electron‐dense and insoluble material located at the presynaptic membrane that organizes and ensures precision and speed of synaptic transmission. The active zone is precisely localized opposite to the postsynaptic density, where NT receptors are associated. This spatial organization allows NTs to bind to postsynaptic receptors, which drives activation or inactivation of downstream pathways. Among the identified active‐zone proteins, we highlight Munc18s, Munc13s, RIMs, liprins, bassoon, and piccolo, within others. After exocytosis, synaptic vesicles undergo an endocytic process in which synaptic terminals retake and reuse membranes that have been previously fused. These newly formed vesicles are then recycled and refilled with NT, so they are ready for a new round of exocytosis (Südhof, 2004; Südhof and Rizo, 2011).
In this way, synaptic vesicles follow a trafficking cycle at the presynaptic terminal that contains several steps: (1) NT uptake at high concentrations using active transport, driven by a proton pump and different transporters determined by the type of NT used by a neuron (glutamate: VGlut; GABA: VGat, etc.); (2) synaptic vesicles organization in front of the active zone; (3) synaptic vesicles docking at the active zone; (4) synaptic vesicles priming at the active zone; (5) synaptic vesicles transformation into a competent state to drive the fusion‐pore opening by Ca2+ (Figure 2). After NT release, synaptic vesicles are recycled (6) by three different mechanisms (Figure 2): (a) local reacidification and NT refilling without undocking, which is called “kiss‐and‐stay”, because the vesicles are maintained in the ready releasable pool at the active zone; (b) local recycling with undocking from the active zone, which is known as “kiss‐and‐run”, followed by reacidification and NT uptake (Alvarez de Toledo et al., 1993); and (c) full recycling via clathrin‐ mediated endocytosis, reacidification and NT refilling through an endosomal intermediate, a mechanism that takes longer to complete. All these steps and functions collaborate and converge in some way in the vesicle cycle in order to produce high‐speed, regulated and repeated rounds of NT release, which conforms the ultimate function of the presynaptic terminal (Südhof, 2004, 2014; Südhof and Rizo, 2011) (Figure 2).
7
Introduction kk
Figure 2. The synaptic vesicle cycle. The synaptic vesicle cycle consists of rounds of exocytosis (orange arrows) and endocytosis, recycling and refilling (yellow arrows). Synaptic vesicles are first filled with neurotransmitter (NT, red dots) by active transport and fueled by an electrochemical gradient (H+) provided by a proton pump that acidifies the inner part of the vesicle (green background). Then, synaptic vesicles are docked and posteriorly primed at the active zone in order to become stable and competent to respond to the calcium signal, once calcium channels open after membrane depolarization and enters the presynaptic terminal causing completion of fusion. NTs are then released into the presynaptic cleft and bind to receptors in the postsynaptic density (PSD). After fusion pore opening, synaptic vesicles probably recycle by three different mechanisms: local refilling without undocking (a), local recycling with undocking (b), and full clathrin‐mediated (pink) recycling via intermediate endosomal passage (c) (Taken from Südhof and Rizo, 2011).
3. Membrane fusion by SNARE proteins
Membrane fusion is the basis of all processes in eukaryotic cells. The first evidences about molecules involved in membrane fusion were observed at the synapse, where SNARE proteins were identified as targets of clostridial botulinum and tetanus toxins (Südhof and Rizo, 2011). Later, SNARE proteins were described as essential components of the membrane fusion machinery. Synaptic exocytosis is mediated by three SNARE proteins: synaptobrevin‐2/VAMP2 (vesicle‐associated membrane protein) located in the synaptic vesicle membrane, syntaxin‐1 and SNAP25 (synaptosomal‐associated protein 25 KDa), both located at the presynaptic plasma membrane (Söllner et al., 1993; Südhof, 2004, 2014; Südhof and Rizo, 2011).
8
Introduction
SNARE proteins are characterized by a SNARE motif (a homologous 70‐residue sequence), which is implicated in the core complex formation. These three proteins form a complex assembled into a parallel four‐helical bundle (formed by four SNARE motifs). SNAP25 contains two SNARE motifs and also a linker sequence involved in membrane attachment via palmitoylation. SNARE complexes are dissociated by NSF (N‐ethylmaleimide sensitive factor) that acts as an ATPase, and SNAPs (soluble NSF‐attachment proteins), which leads to the “SNARE” (soluble NSF‐attachment protein receptors) designation (Söllner et al., 1993). SNARE complex is resistant to SDS (sodium dodecyl sulfate), which reveals its high stability, and its assembly releases energy that is sufficient to trigger membrane fusion by forcing the synaptic vesicle membrane and the plasma membrane together (Südhof and Rizo, 2011). Insights in this mechanism were first provided by J. Heuser and R. Jahn, who demonstrated that SNARE complexes assemble in a similar way, bringing C‐terminal transmembrane regions of SNARE proteins near to each other (trans‐SNARE), generating a zipper effect from N‐ to C‐terminal direction, in order to force membranes to fusion (Hanson et al., 1997; Südhof, 2014). After the fusion‐pore opening, both membranes completely merge, and trans‐ SNARE complexes are turned into cis‐SNARE complexes (i.e. they are on a single membrane), which are then dissociated by NSF and SNAPs into monomers. SM (Sec1/Munc18‐like) proteins contribute to SNARE complex assembly initiation and are essential partners for SNARE proteins during fusion (Südhof and Rizo, 2011). Today, this is the accepted model in the field.
In summary, NT release is driven by repetitive cycles of SNARE proteins assembly and disassembly, which sometimes happen in a very high rate due to differential firing frequencies between neuron subtypes. Because of this, presynaptic terminals potentially produce misfolded proteins caused by an inappropriate formation of SNARE complexes, which makes not surprising that neurons have developed chaperone systems to maintain SNARE conformation and ensure a proper folding and assembly of SNARE complexes (Südhof, 2014; Südhof and Rizo, 2011). Two classes of chaperones have been identified: CSPα (cysteine string protein alpha) and synucleins. Special attention is worth to CSPα, whose relation to a specific subtype of GABAergic interneurons integrates the principal topic of this work.
4. Cysteine string protein alpha (CSPα) 4.1. Localization and molecular structure CSPα is a highly conserved synaptic vesicle protein that was first discovered in Drosophila melanogaster when using a neuronal‐specific monoclonal antibody to label nerve terminals (Zinsmaier et al., 1990). It has been then characterized in a variety of species, such as Torpedo californica, Caenorhabditis elegans, Xenopus and mammals; and in a wide range of secretory vesicle types, both in neuronal and non‐ neuronal tissues, such as chromaffin granules in chromaffin cells and zymogen granules in pancreatic cells. mRNA enconding for CSPα has also been found in heart, kidney, spleen, liver, lung, testis, muscle, and brain, among other localizations (Braun and Scheller, 1995; Chamberlain and Burgoyne, 1996; Chamberlain
9 Introduction kk et al., 1996; Coppola and Gundersen, 1996), suggesting a general function in secretion for CSP proteins. While C. elegans and Drosophila have only a single gene coding for CSP (dnj‐14 and Csp, respectively), mammals express three different CSP proteins (CSPα, CSPβ, CSPγ) encoded by DNAJC5a, b and g genes, presenting isoforms α and β high homology in their amino acid sequences, while CSPγ is less related. CSPα is the principal isoform in the brain, located in the synaptic vesicle membrane (Fernández‐Chacón et al., 2004; Schmitz and Fernández‐Chacón, 2009), whereas hair cells from the inner ear present both CSPα and CSPβ (Schmitz et al., 2006), and testis express the three CSP isoforms (Fernández‐Chacón et al., 2004).
The molecular structure of CSPs is characterized by distinct domains:
1. N‐terminal DNA‐J domain: this region is typical of the Hsp40 family of co‐chaperones, and presents homology with the bacterial DnaJ protein, which comprises a sequence of approximately 70 amino acids highly conserved within species (at least 78% identity). This domain contains a conserved HPD (histidine, proline, and aspartic acid residues) motif required for binding to Hsc70 protein to typically act in the refolding or disaggregation of client proteins (Braun et al., 1996; Burgoyne and Morgan, 2015; Chamberlain and Burgoyne, 2000; Schmitz and Fernández‐Chacón, 2009).
2. Linker region: this area, constituted of around 20 amino acids approximately, is highly conserved during evolution and is important for exocytosis (Zhang et al., 1999).
3. Central cysteine‐rich string: this region names the protein. The string contains 13‐15 cysteine residues within a total of 25 amino acids, being most of them palmitoylated, which is essential for membrane targeting of CSP to synaptic vesicles (Chamberlain and Burgoyne, 1998).
4. Variable C‐terminal domain: it is subjected to differential splicing and generates truncated proteins in Drosophila and mammals. Its function is not well‐understood yet, although it was hypothesized to participate in exocytosis and protein‐protein interactions (Boal et al., 2004).
4.2. The trimeric complex CSPα‐Hsc70‐SGT is essential for chaperone activity One of the best known functions for CSPα is its role as a co‐chaperone (Braun and Scheller, 1995; Chamberlain and Burgoyne, 2000; Tobaben et al., 2001; Zinsmaier et al., 1994). First observations determined that CSPα interacts with members of the heat‐shock protein family Hsp70 (mostly acting as chaperones), especially activating the Hsc70 ATPase (Braun et al., 1996; Chamberlain and Burgoyne, 1997; Stahl et al., 1999). Later, a novel CSPα binding partner was identified, SGT (small glutamine‐rich tetratricopeptide repeat‐containing protein), which consists of three tetratricopeptide repeat domains (TRPs) in tandem. Together, these three proteins constitute a tripartite complex that acts as a chaperone machinery on the synaptic vesicle surface (Figure 3), and was found to refold denatured luciferase in vitro (Tobaben et al., 2001).
10
Introduction
Specifically, the DNA‐J domain of CSPα mediates the binding to the constitutively expressed heat‐ shock cognate protein Hsc70, recruiting Hsc70 to the synaptic vesicle surface. Hsp70 proteins include a N‐ terminal adenosine triphosphate (ATPase) domain and a C‐terminal substrate‐binding domain (SBD). The ATPase domain binds ATP/ADP, but its ATPase activity stays low until the interaction with the DNA‐J domain of CSPα. Once ATP is bound and the DNA‐J domain of CSPα interacts with the Hsc70 ATPase domain, the hydrolysis of ATP switches the conformational state of Hsc70 into a high‐affinity locked state. Moreover, the ATPase activity is additionally enhanced by SGT, that simultaneously binds to the C‐terminal domain of Hsc70 and to the cysteine‐string region (and C‐terminus) of CSPα, acting as a molecular bridge between CSPα and Hsc70, and making the refolding of denatured proteins possible. A new binding of ATP promotes the release of renatured client proteins, leaving Hsc70 ready for a new round in the chaperone cycle (Schmitz and Fernández‐Chacón, 2009; Tobaben et al., 2001).
Figure 3. CSPα‐Hsc70‐SGT trimeric complex. The DNA‐J domain of CSPα binds to the ATPase region of Hsc70 while TPR repeats of SGT bind to the C‐terminus of Hsc70 as well as to the cysteine‐rich string and the C‐terminus of CSPα. Altogether conforms the tripartite chaperone complex. ATP, adenosine triphosphate; CSP, cysteine‐string protein; SBD, substrate‐binding domain; SGT, small glutamine‐rich tetratricopeptide; TPR, tetratricopeptide repeat (Taken from Schmitz and Fernández‐ Chacón, 2009).
4.3. Phenotypic characterization of CSPα knock‐out (KO) mice CSPα KO mice were generated by homologous recombination of a region including exon 2, which is replaced with a neomycin resistance gene cassette (Fernández‐Chacón et al., 2004; Tobaben et al., 2001) (see later Figure 16). Immediately after birth, KO animals do not show any obvious defects, but after two or three weeks of postnatal age, they begin to experience body‐weight loss, evident muscle weakness,
11 Introduction kk deficiency in locomotor activity and progressive blindness (Fernández‐Chacón et al., 2004; Schmitz and Fernández‐Chacón, 2009; Schmitz et al., 2006). Furthermore, CSPα KO mice present additional phenotypic characteristics, for example, a tendency to clasp hindlimbs when suspended by the tail, a poorer posture, and a leg position that simulates “inverted champagne bottles”. Deeper examinations show neurodegenerative changes and an age‐dependent deterioration in synaptic transmission of the neuromuscular junctions and Calyx of Held synapses. CSPα KO mice die around 1‐2 months after birth (Fernández‐Chacón et al., 2004) (Figure 4). A B C
Figure 4. CSPα KO mice phenotype. A. CSPα KO (CSP‐/‐) mice show a significant smaller size and severe motor deficits compared to controls (CSP+/+) (Taken from Schmitz and Fernández‐Chacón, 2009). B. Changes in body weight of WT (CSP+/+), heterozygous (CSP+/‐) and KO (CSP‐/‐) mice with age. C. Postnatal survival curve comparing the three different CSPα genotypes previously mentioned (Taken from Fernández‐Chacón et al., 2004).
4.4. SNAP25 is an important substrate of CSPα One of the client proteins for the CSPα‐Hsc70‐SGT complex is SNAP25, a SNARE complex‐forming protein (Chandra et al., 2005; Sharma et al., 2011). The trimeric complex binds directly to monomeric SNAP25 and prevents its degradation, allowing SNARE complex formation. Abnormal SNAP25 conformers that presumably arise spontaneously during synaptic activity inhibit the formation of SNARE complexes and are ubiquitylated for proteasomal degradation (Figure 5). Interestingly, this degradation is enhanced by deletion of CSPα and decreased if overexpressed, so the trimeric CSPα‐Hsc70‐SGT complex acts refolding SNAP25 to prevent its degradation and ensuring SNARE‐complex formation (Sharma et al., 2011). It has been already shown that SNAP25 levels are 40% decreased in mice lacking CSPα and that the impaired SNARE‐complex assembly (50% reduction) provokes neurodegeneration (Sharma et al., 2012). Interestingly, overexpression of α‐synuclein (mutations in α‐synuclein have been linked to Parkinson’s disease) abolishes the lethal phenotype caused by deletion of CSPα in mice by partially correcting SNARE complex assembly, suggesting a cooperating chaperone‐mechanism between CSPα and α‐synuclein to maintain SNARE complex (Chandra et al., 2005).
12
Introduction
These findings positioned CSPα‐Hsc70‐SGT as a chaperone complex that ensures the long‐term operation of the SNARE machinery, which seems especially crucial for nerve terminals that work at high‐ frequencies, since they suffer multiple rounds of exocytosis along the life of a neuron.
Figure 5. CSPα‐Hsc70‐SGT chaperone complex functioning. After SNAP25 is dissociated from the SNARE complex by NSF and SNAPs, there is a balance between SNARE‐ complex formation using competent conformers (SNAP25) and inhibition of SNARE‐complex assembly due to misfolded conformers (SNAP25*). Generated abnormal SNAP25* conformers are ubiquitylated (Ub) and degraded by the proteasome, however, the trimeric CSPα‐Hsc70‐SGT complex acts refolding SNAP25* into competent SNAP25, preventing its degradation and ensuring SNARE‐complex formation. ATP, adenosine triphosphate; ADP, adenosine diphosphate; Pi, inorganic phosphate; Syb2, synaptobrevin‐2; Synt‐1; syntaxin‐1 (Taken from Sharma, Burré and Südhof, 2011). 4.5. Role of CSPα in GABAergic synapses firing at high frequency Previous observations suggest that CSPα might be critically required at synapses subjected to heavy membrane trafficking due to intense synaptic vesicle cycling. Based on this, García‐Junco‐Clemente et al., 2010 studied CSPα function in small central GABAergic synapses that works under a high activity regime. They revealed that synaptotagmin‐2‐expressing (Syt2+) GABAergic synapses formed by fast‐ spiking ( ̴200 Hz) parvalbumin (PV) interneurons depended on CSPα to maintain their function and integrity along time. PV interneurons suffered progressive degeneration of presynaptic terminals upon CSPα removal, which was shown by an age‐dependent reduction in Syt2+ and PV+ synaptic puncta at the hippocampus of CSPα KO mice, but with no changes in PV somata number. Electrophysiological recordings in hippocampal cultured neurons from CSPα KO mice confirmed the previous data, showing a progressive decrease of mIPSCs frequency with time in culture, which agrees with a reduction in the GABAergic synaptic input. Interestingly, glutamatergic synapses were apparently not affected, confirmed by the absence of changes in mEPSCs frequency or glutamatergic puncta. Since glutamatergic neurons trigger APs at lower frequencies (1 Hz) compared to PV interneurons, the proposed hypothesis was that PV cells are more sensitive to network activity. Importantly, the inhibition of the excitatory transmission in cell cultures was sufficient to rescue Syt2+ synapses, suggesting a specific activity‐dependent degeneration of fast‐
13
Introduction kk spiking synaptic terminals. It is important to remark that the loss of inhibition was compensated by a homeostatic mechanism that consists of a downscaling of postsynaptically expressed AMPA receptors (García‐Junco‐Clemente et al., 2010) (Figure 6).
Figure 6. Predicted model of inhibitory and excitatory synaptic changes upon CSPα removal in hippocampal cultures. In CSPα WT cultures, there is a balance between excitatory and inhibitory inputs. Two types of inhibitory ones are represented: terminals coming from fast‐spiking interneurons (synaptotagmin‐2‐ (Syt2) and parvalbumin‐ positive (PV) basket cells) and terminals coming from less active interneurons (calretinin‐positive (CR)). Terminals of fast‐spiking neurons, which work under a high activity regime, need CSPα to overcome molecular stress derived from continuous activity, and might initiate a dying‐back degeneration process in an age dependent manner. CR+ interneurons and pyramidal neurons are less dependent on CSPα because they work under a low‐activity regime. The synaptic network seems to sense this inhibitory synaptic input reduction and triggers homeostatic plasticity mechanisms: excitatory synapses become weaker by postsynaptic functional downscaling of AMPA‐type receptors (Taken from García‐Junco Clemente et al., 2010).
4.6. Importance of CSPα in human disease Two specific mutations in human DNAJC5/CLN4 gene have been related to a human disease called autosomal‐dominant adult‐onset neuronal ceroid lipofuscinosis (ANCL), characterized by a lysosomal accumulation of misfolded and proteolysis‐resistant material (named lipofuscin) in neural tissues related to neurodegeneration. This disease starts during the adulthood, around 30 years old. The mutations in DNAJC5/CLN4 gene are caused by a genetic deletion (p.Leu116del) or substitution (p.Leu115Arg) into the cysteine‐string domain of CSPα, affecting the palmitoylation‐dependent sorting. Indeed, a single‐copy mutation of the gene produces a decrease in the amount of synaptic CSPα wild‐type protein in neuronal cells, because the mutations lead to aggregates composed of both mutant and WT proteins. Individuals
14 Introduction affected by ANCL develop seizures, movement disorders, cognitive deteriorations and progressive dementia, and also have a significant decrease in life span (Diez‐Ardanuy et al., 2017; Greaves et al., 2012; Nosková et al., 2011). Interestingly, PPT1 (palmitoyl‐protein thioesterase 1), a protein that removes acyl chains from palmitoylated proteins during degradation, was recently found to be increased and mis‐ localized in CSPα ANCL human cells, although showing dramatically reduced activity. Importantly, it has been demonstrated that CSPα is a substrate of PPT1, so it is depalmitoylated by PPT1 (Henderson et al., 2016).
5. The cerebral cortex
The increase in brain surface has been considered as the major development during evolution; for example, a significant variability is observed in brain size between mammalian species (DeFelipe, 2011). This size increment is in close relation with the huge expansion of the mammalian cortex regarding to the number of cells, layers and functional areas, compared to other brain regions. Moreover, the growth of cortical regions is associated with the acquisition of refined cognitive, sensory and motor functions. In large mammals and humans, the cerebral cortex has a folded shape that is constituted by fissures separated by elevated regions; it is thought that this folding happened to harbor the growing number of neurons. Strikingly, cortical thickness does not vary substantially among species and it is always around 2‐ 4 mm. It can reach up to 4.5 mm in humans, while in the fissures may have 1 mm thick. Nevertheless, in mammals with small brains (e.g. rats and mice), cortex is not folded but smooth (DeFelipe, 2011; Kandel et al., 2001). It was a crucial event during the evolution of the mammalian telencephalon the emergence of a highly complex multi‐laminated cortex: the neocortex. The activity of this structure is straightly related to the appearance of the abilities that distinguish humans from other mammals, e.g., perceptual awareness, thought, language or consciousness (DeFelipe, 2011).
5.1. Cortical organization Based on works about cortical cytoarchitecture, connectivity, functionality and gene expression (DeFelipe, 2011; Glasser et al., 2016; Ng et al., 2009; Tasic et al., 2018), approximately 200 cortical areas have been described in humans, and dozens in rodents. These are located within the four main cortical lobes: frontal, parietal, temporal and occipital. Although cortical areas have specific activity patterns and information processing roles (e.g. sensory or motor), the cortex shows similar layered structure containing precise neuronal populations. The number of layers and their organization vary between cortical areas, but the typical neocortical structure contains six layers that are named from the most superficial to the white matter. Besides, from a functional point of view, the layers of the cerebral cortex can be divided into three parts.
15
Introduction kk
1. The supragranular layers consist of layer I to III. They are the main origin and termination of intracortical connections, which are either associational (i.e., with other areas of the same hemisphere) or commissural (i.e., connections to the opposite hemisphere through the corpus callosum).
‐ Layer I is an acellular molecular stratum, which is occupied by dendrites coming from cells localized at higher depths, and by axons that make either synaptic connections in this layer or go through it.
‐ Layer II is basically composed by small spherical cells called granules, so it is named external granular stratum. It is normally described and analyzed together with layer III.
‐ Layer III comprises diverse cell types, including lots of pyramidal‐shape cells. In the deeper region of this layer, neurons are greater in size than those localized more superficially. It is named external pyramidal stratum. Layers II and III contain dendrites from cells localized in layers V and VI.
2. Layer IV is mainly composed by cells similar to granules of layer II, and hence it is called internal granular stratum. It receives thalamocortical connections, especially from specific thalamic nuclei, being more prominent in primary sensory cortices.
3. The infragranular layers, which primarily connect the cerebral cortex with subcortical regions. These layers are most developed in motor cortical areas.
‐ Layer V comprises cells with pyramidal‐like shape that are normally bigger than those from layer III. It is known as internal pyramidal stratum.
‐ Layer VI is a quite heterogeneous layer of neurons and it is known as polymorphic stratum. It is merge with the white matter, limiting the cortex and carrying axons from/to the cortex. Layers V and VI contain dendrites from neurons localized at layer III and IV.
There are many exceptions to the classical division of the neocortex into I‐VI layers. For example, in motor and premotor areas of many species (including mice) there is no layer IV; on the other hand, an extremely prominent layer VI is observed in primary visual cortex that could be subdivided in at least 3 more layers (DeFelipe, 2011; Kandel et al., 2001). It is also well‐known that the disposition of neurons in the barrel cortex of rodents is different compared to other cortical regions: the barrels consist of layer IV aggregates of small spiny neurons (DeFelipe, 2011; Feldmeyer, 2012).
Furthermore, in the 1950s, Mountcastle introduced the expression of “cortical columns” or “modules” to describe the concept of vertical information processing in cortex. The cortex is organized horizontally into six layers, but also vertically into groups of cells synaptically linked across the horizontal
16
Introduction layers. This structure is more obvious in sensory and motor areas of the neocortex (Feldmeyer, 2012; Mountcastle, 1997), and recently was also found in visual cortex. Nevertheless, columns are neither an obligatory nor unique cortical feature (DeFelipe, 2011).
5.2. Cortical glutamatergic neurons Glutamatergic (principal) neurons are the most abundant cortical neurons (around 80‐90%) and are distributed in all cortical layers except for layer I. They have a typical pyramidal‐like shape. Their axons make intracortical connections, involving the main source of excitatory synapses in the cortex, but they also represent the majority of projection neurons. Their dendritic spines are the principal postsynaptic target of excitatory synapses in the cortex. Nevertheless, the structure of these cells changes substantially among cortical areas and species, regarding several features: the size and complexity of dendritic arborization, the density of dendritic spines, and the total number of gspines, bein these differences critical for the functional processing of cortical information (DeFelipe, 2011). Importantly, cortical areas contain multiple highly structured subnetworks of interconnected principal neurons (Harris and Mrsic‐Flogel, 2013).
5.3. Cortical GABAergic interneurons GABAergic interneurons are presented in a lower proportion in cortex compared to pyramidal cells (10‐15% in rodents and 25% in monkeys). They are localized in almost all layers, and mostly make local connections. Although GABAergic interneurons represent a minority, their role is vital for brain functioning because they regulate the activity of principal neurons using GABA as NT (DeFelipe, 2011; Harris and Mrsic‐ Flogel, 2013; Tamamaki et al., 2003; Tremblay et al., 2016). Cortical GABAergic interneurons are born in the medial or caudal ganglionic eminences (MGE, CGE) and are subjected to specific transcription factors for cortical migration, such as Nkx2.1 and Lhx6 (Hu et al., 2014; Kelsom and Lu, 2013; Natalie et al., 2015; Wonders and Anderson, 2006).
5.3.1. Cortical interneuron classification The GABAergic population is highly heterogeneous, thus they can be described by several criteria, such as (1) morphology, (2) the expression of molecular markers like neuropeptides (somatostatin, cholecystokinin, vasoactive intestinal peptide (VIP) and neuropeptide Y (NPY)) and calcium‐binding proteins (parvalbumin, calretinin and calbindin), (3) the expression of receptors (acetylcholine, serotonin, noradrenaline and dopamine), (4) functional properties (concerning to the electrophysiological characteristics of the neurons), (5) connectivity and (6) in vivo patterns of activity (Tremblay, Lee and Rudy, 2016; Hu, Gan and Jonas, 2014).
Based on these observations, classical classification of neocortical interneurons divides them in three major groups: parvalbumin (PV), somatostatin (Sst) and 5HT3aR (ionotropic serotonin receptor)
17 Introduction kk interneurons, although they are also intrinsically heterogeneous (Figure 7). These three populations account for nearly 100% of all GABAergic cortical interneurons (Kelsom and Lu, 2013; Pfeffer et al., 2013; Tremblay et al., 2016).
1. PV interneurons. PV cells are the largest interneuron population in the neocortex. It includes fast‐ spiking (FS) basket cells and chandelier or axo‐axonic cells.
‐ FS basket cells: they preferentially make perisomatic “basket” synaptic connections on the soma and proximal dendrites (somato‐dendritic) of principal cells and interneurons. Indeed, FS basket cells are the largest population of interneurons in the neocortex. Basket cells have an impressive collection of molecular and cellular specializations to guarantee a fast, reliable, strong and temporally precise inhibition on their target cells. The speed and precision of their signaling is extraordinary (Tremblay et al., 2016).
‐ Chandelier cells: they form unique candlestick‐like synaptic terminal arrays to specifically target the axon initial segment (axo‐axonic) of pyramidal cells. Chandelier cells are also considered fast‐spiking but they show some differences in intrinsic electrophysiological properties compared to PV basket cells (Tremblay et al., 2016).
2. Sst interneurons. They are dendritic targeting interneurons. They can be divided into two main subgroups:
‐ Martinotti cells: they present an axonal plexus in LI and they target the tuft dendrites of pyramidal cells, making synapses on dendritic spines. Martinotti cells are mainly located in LII/III and LV/VI. Besides arborizing in LI, a significant proportion of their axonal arbor, probably contacting the basal dendrites of other neurons, is present in the layer where the soma is located (Tremblay et al., 2016).
‐ Non‐Martinotti cells: although sharing many properties with Martinotti cells, they lack significant axonal arborization in LI, making synapses onto different subcellular compartments or cell types (Tremblay et al., 2016).
3. 5HT3aR interneurons. This group represent around 30% of all neocortical interneurons and are thought to be more diverse than PV and Sst cells. They all express 5Ht3a and nicotinic receptors. They are enriched in supragranular layers and are divided into two subgroups based on whether or not they express the neuropeptide VIP.
‐ Vip interneurons: they represent about 40% of all 5HT3aR cells and they are located preferentially in LII/III. The majority of Vip interneurons have a vertically oriented, bipolar‐like
18
Introduction
dendritic morphology (the rest are multipolar). Their dendritic trees tend to be narrow and cross several layers (Tremblay et al., 2016).
‐ Non‐Vip 5HT3aR interneurons: they represent about 60% of all 5HT3aR cells and about 90% of all interneurons in LI. They include the neurogliaform cells (NGFC) (containing also multipolar neuropeptide Y (NPY)‐expressing cells, although NPY is also partially expressed in Sst and PV interneurons) (Karagiannis et al., 2009; Lee et al., 2010), cholecystokinin (CCK)‐ expressing interneurons and other less clearly defined types (Tremblay et al., 2016).
Figure 7. Diversity, classification and properties of neocortical GABAergic interneurons. Almost all the interneurons in neocortex express one of the main three, non‐overlapping markers: Parvalbumin (PV, blue), somatostatin (Sst, red) and the ionotropic serotonin receptor 5HT3a (5HT3aR, green‐yellow). Additional subdivisions within each main population exist based on different features; here we show morphological structure and cellular and subcellular targeting biases (Modified from Tremblay et al., 2016). BC, basket cells; SBC, single bouquet cells; NGFC, neurogliaform cells; CCK, cholecystokinin; PV, parvalbumin; Sst, somatostatin; 5HT3aR, ionotropic serotonin receptor; L1, cortical layer I. 5.4. Cortical connectivity: pyramidal neurons versus GABAergic interneurons The neocortex receives afferent input from the thalamus, and from other cortical regions, besides strong neuromodulatory inputs from different brain nuclei. The efferent input from neocortex are driven to other regions of the encephalon, such as neocortical areas at both sides of the encephalon, the basal ganglia, and the spinal cord. The layered structure of neurons in cortex provides an efficient way to organize afferent‐efferent relations of neocortical neurons (Kandel et al., 2001).
The neuronal network at the brain functions through synaptic interactions between glutamatergic principal neurons and GABAergic interneurons. It has been classically considered that the connectivity between interneurons and principal cells may be nonspecific, although recent studies have defined unique interneuron targeting onto specific cell types and/or subcellular locations (García‐Junco‐Clemente et al., 2017, 2019; Pfeffer et al., 2013; Tremblay et al., 2016). Pfeffer et al., 2013 revealed that, indeed, there is a specific network of inhibitory synaptic connections in the cortex. This network is revealed not only by cortical inhibitory interneurons contacting principal cells but also interacting each other, in all cortical layers. For example, the connection probability to neighboring pyramidal cells is close to 100% for parvalbumin‐ and somatostatin‐expressing GABAergic interneurons (Harris and Mrsic‐Flogel, 2013), but
19
Introduction kk the inhibition contribution of PV cells onto principal cells is higher than the one from Sst or Vip interneurons (Figure 8). On the other hand, PV cells strongly inhibit one another (they are the main source of their own inhibition) but contribute for little inhibition to the other two populations (Sst and Vip). Sst interneurons, however, do not inhibit one another but robustly inhibit PV and Vip cells. Finally, Vip interneurons represented the main source of inhibition onto Sst cells (Figure 8). In conclusion, PV, Sst and Vip cells provide a highly specific and complementary network of connections in cortex (Pfeffer et al., 2013).
Understanding how multiple interneurons classes encode the information in the cortex is a field of active research. Nowadays, there is still an enormous work focused on cortical connectivity and circuit organization between pyramidal cells, interneurons and within each other that is being carried out basically thanks to new in vivo methods, such as stimulus‐based behavioral tasks, optogenetics, and two‐ photon microscopy (Chen et al., 2017b; DeFelipe, 2011; García‐Junco‐Clemente et al., 2017, 2019; Harris and Mrsic‐Flogel, 2013; Lim et al., 2013; Pfeffer et al., 2013). A B
Figure 8. Individual neuronal contributions among cortical interneurons. A. Heatmap of the normalized individual neuronal contributions of the three main presynaptic interneuron classes (Pvalb, Sst, Vip) onto the same three categories and pyramidal cells. B. Schematic of the inhibitory connections among the three largest classes of interneurons (Pvalb, Sst, Vip) and pyramidal cells. Pvalb, parvalbumin; Sst, somatostatin; Vip, vasoactive intestinal peptide; Pyr, pyramidal cell (Modified from Pfeffer et al., 2013).
6. Parvalbumin interneurons
Calcium‐binding proteins normally buffer Ca2+ ions or trigger the activity of some enzymes upon Ca2+ binding (Celio, 1986; Hippenmeyer et al., 2005). Parvalbumin is a low‐molecular weight protein (around 12 KDa) that contains EF‐hand (helix‐loop‐helix) calcium binding domains, thus it is involved in calcium signaling. Parvalbumin protein is structurally similar to calmodulin and is encoded by five exons in its genomic sequence, with exon 2 containing the start codon (ATG) and exon 5 holding the STOP codon at the mRNA level. Parvalbumin was first described as a marker for fast contracting muscle fibers (Celio
20 Introduction and Heizmann, 1982) and, in 1986, it was found in a subpopulation of GABAergic neurons that was actually large regarding total GABAergic interneurons (Celio, 1986). After that, parvalbumin was proved to be expressed selectively in fast‐spiking interneurons at the hippocampus (Kawaguchi et al., 1987). Nowadays, thanks to the extensive use of several techniques, such as patch‐clamp recordings, optogenetics, in vivo measurements and high‐throughput sequencing technologies, within others, the unique properties of PV interneurons have been intensively studied making our knowledge notably enriched.
Fast‐spiking, PV interneurons have been the focus of study for different laboratories since twenty years ago, constituting an interesting subtype of GABAergic interneurons. The PV interneuron population account 40% of total GABAergic cells in motor (M1 and M2), frontal and visual cortex (Condé et al., 1994; Gonchar et al., 2008; Hu et al., 2014; Tamamaki et al., 2003; Uematsu et al., 2008; Whissell et al., 2015). According to these numbers, PV interneurons comprise the 8% of the total cortical neuronal population. Specifically, the proportion of PV cells of total GABAergic interneurons in the different motor cortical layers consists of 0% in layer I, 44% in layer II/III, 44.5% in layer V and 42% in layer VI (Tamamaki et al., 2003).
The specific study of the PV interneuron population is possible based on its expression of parvalbumin, which can be detected through different experimental techniques. Moreover, they have unique electrophysiological properties, which make easy their detection, for example, the fast kinetic of their APs compared to other neuronal populations (AP frequency > 50 Hz at 22°C and > 150 Hz at 34°C) (Figure 9). Moreover, the PV gene promoter has been genetically exploited in mouse genetic engineering (Hippenmeyer et al., 2005; Madisen et al., 2010) for morphological and functional studies of these neurons, as well as for their specific manipulation through optogenetic techniques.
6.1. Morphological and functional properties of PV dendrites PV neurons contain multiple dendrites that frequently cross cortical layers. The total cumulative dendritic length varies from 3.1 to 9 mm in a single PV cell. This fact permits interneurons to receive numerous inputs from different afferent pathways and from many principal cells. PV somata and dendrites receive a high number of synaptic inputs, being mostly excitatory synapses compared to inhibitory ones (Gulyás et al., 1999). The excitatory input come from pyramidal neurons, which form synapses onto PV aspiny dendrites, facilitating the generation of fast excitatory postsynaptic potentials (EPSPs). As previously mentioned, most of the inhibitory input proceed from PV interneurons, although Vip‐ and Sst‐ expressing interneurons also provide inhibitory inputs onto PV cells.
Dendrites of PV interneurons are also characterized by several electrophysiological properties: (1) action potentials backpropagate into dendrites with decreasing amplitude and low activity dependence; (2) there is no dendritic spike initiation; (3) they contain a low density of voltage‐dependent Na+ channels; (4) dendrites enclose a high density of voltage‐dependent K+ channels, mainly the Kv3 type; these channels
21
Introduction kk display a high activation threshold, a fast activation/deactivation kinetic, and can be activated by local EPSPs, leading to a short EPSP decay time and summation, which facilitates the fast AP initiation; (5) PV dendrites are vastly interconnected by gap junctions, which accelerate EPSPs time course due to adjacent dendritic connections that share the excitatory charge. Moreover, gap junctions increase the spatial range of glutamatergic input detection affecting even the PV neurons non‐directly connected with glutamatergic cells. In this way, the efficacy of distal inputs detection is increased (Du et al., 1996; Hu et al., 2010, 2014).
6.2. Morphological and functional properties of PV axon PV axons are highly arborized and the total length can rise to 20‐50 mm in a single PV neuron (20‐ 24 mm in the frontal cortex) (Karube et al., 2004). PV arborized axons form “en passant” terminals mostly (approximately between 3200‐3800 synaptic boutons in the frontal cortex), providing an extensive divergent inhibitory output. Another important characteristic of PV axons is that they innervate postsynaptic cells at the perisomatic region and basal dendrites, or at the axon initial segment, depending on the PV subtype. As previously mentioned, the axon of basket cells forms a basket‐like shape around principal cells somata and proximal dendrites (somato‐dendritic), while the axon of axo‐axonic cells make synapse onto the axon initial segment of the glutamatergic neurons adopting a chandelier‐like shape (Figure 9). This synaptic distribution boost the inhibitory effect of PV interneurons since they innervate their target cells very close to the site of the AP initiation, strongly affecting the spiking pattern of the neurons (Freund and Buzsáki, 1996; Hu et al., 2014).
The physiology of PV axon highlights by several properties: (1) the AP initiation site is very close to the soma ( ̴ 20 µm); (2) failures in AP propagation are extremely rare; (3) AP propagation velocity is
‐1 + extremely fast ( ̴ 1.5 ms ); (4) density of Na channels (mainly Nav1.1 and Nav1.6) is gradually increased from the soma to the proximal axon, and finally to the distal axon, which contain almost 99% of all sodium channels, thus becoming PV excitability almost uniquely axonal; these channels contribute to a rapid signaling in PV cells and compensate for the unfavorable axon morphology (small diameter,t grea arborization and high boutons density) (Hu and Jonas, 2014); (5) PV axon harbors voltage‐gated K+ channels with similar properties to those on PV dendrites; they are principally of Kv3 subtype, which ensure a fast AP repolarization in the axon. However, Kv1 channels are present in the initial segment of the axon, which show a lower activation threshold and gating that could serve as a control mechanism, allowing PV cells to respond preferentially to large‐amplitude or synchronous inputs with rapid kinetics, while filtering out slow or out‐of‐phase inputs (Goldberg et al., 2008; Hu and Jonas, 2014; Hu et al., 2014). Remarkably, it was recently described that neocortical and hippocampal PV axons exhibit frequent myelination with a patchy distribution and short internodes when compared to glutamatergic axons. Further studies are needed to discern whether myelination is likely to regulate synapse formation in PV axons and/or to have implications in disease (Micheva et al., 2016; Stedehouder et al., 2017).
22 Introduction
A B
C
Figure 9. Axonal‐derived classification and electrophysiological properties of AP firing patterns in PV cells. A. Cortical PV interneurons differ in their axonal shape (basket‐like (PV BC) or chandelier‐like (PV ChC) shape) and localization (somato‐dendritic or axo‐axonic) when targeting pyramidal cells (Pyr) (Modified from Harris and Mrsic‐Flogel, 2013). B. Original drawing from the first work by Santiago Ramón y Cajal published in 1888 of a cross‐section of a lamina of cerebellum. The nest morphology of axon terminals from PV basket cells around Purkinje cells is shown in detail. C. Fast‐spiking AP properties of a representative neocortical PV interneuron in vitro. A long somatic current pulse evoked a high‐frequency train of action potentials in the intracellularly recorded neuron (Taken from Hu et al., 2014; originally from McCormick et al., 1985).
6.3. Specializations of PV cells for fast synaptic signaling PV interneurons are optimized for rapid output signaling due to several factors. For example, axonal action potentials are brief, and together with the “en passant” presynaptic terminals, lead to a fast and precise NT release. On the other hand, PV interneurons use P/Q‐type calcium channels, which have the fastest gating within calcium channels, contributing to short synaptic delay and to enhance precision of GABA release. The PV population also highlight for showing a tight coupling between presynaptic calcium channels and release sensors: GABA release at presynaptic terminals seems to be initiated by only two or three calcium channels; all these properties also give speed and efficacy to NT release (Bucurenciu et al., 2010; Hu et al., 2014). Furthermore, some subpopulations of PV interneurons express synaptotagmin‐2, which shows the fastest calcium‐binding kinetics among its family, as a calcium sensor for fast synchronic NT release, also contributing to signaling speed. Synaptotagmins are a large family of proteins with similar domains: an N‐terminal transmembrane region, a linker sequence, and two C‐
2+ terminal C2 domains that bind Ca in most synaptotagmins (Pang et al., 2006). Synaptotagmin‐1 (Syt1) functions as the widely Ca2+ sensor for synchronous release in the forebrain, however, synaptotagmin‐2
23
Introduction kk
(Syt2), which shares the highest homology with Syt1, is involved in fast, calcium dependent NT release and has been subjected to be specific to inhibitory neurons (Bouhours et al., 2017; Chen et al., 2017a; Kerr et al., 2008; Pang et al., 2006; Sommeijer and Levelt, 2012). Indeed, Syt2 expression has been used for PV synaptic labelling in several works (García‐Junco‐Clemente et al., 2010; Pang et al., 2006; Sommeijer and Levelt, 2012). Finally, PV expression is also relevant, mostly due to the EF‐hand domains that could bind both Ca2+ and Mg2+ ions, which may modulate NT release (Hu et al., 2014).
6.4. PV‐derived feedforward and feedback inhibition PV cells are involved in both feedforward and feedback inhibition mechanisms. During feedforward inhibition, a presynaptic principal cell excites an inhibitory interneuron, so this interneuron inhibits the next neuron, reducing its probability of firing, in a way of limiting excitation downstream in a neural circuit. Conversely, feedback inhibition involves in series activation of interneurons by principal cells that were previously activated by afferent glutamatergic axons; the inhibitory neurons then may inhibit a group of principal cells, including those that initially activated them. PV‐derived inhibition needs to be fast enough to ensure a considerable inhibitory conductance before action potentials are originated in glutamatergic neurons. PV‐mediated feedforward and feedback inhibition is neither uniform nor random, but depend on activity level and distance of target cells: inhibition from PV output synapses is stronger in target cells with heavy synaptic excitation and high activity (Freund and Buzsáki, 1996; Hu et al., 2014; Xue et al., 2014).
6.5. Role of PV interneurons in modulating animal behavior It is important to underline the role of PV interneurons in the control of complex animal behaviors, shaping higher brain functions. As previously mentioned, PV cells are responsible for controlling the spike pattern of principal glutamatergic cells, therefore modelling the activity of the neuronal network. Indeed, PV cells regulate the cortical and hippocampal network oscillations, responsible for the so called “Rhythms of the brain”. Specifically, PV engagement is essential for the generation of gamma oscillations during tasks that engage cognitive or complex perceptual functions in frontal cortex. Furthermore, PV interneurons drive enhancements in theta, delta, and ripple oscillations in the hours following learning, causing hippocampal memory consolidation. Importantly, PV cells control the gain of sensory responses. Their firing pattern present a broad tuning in response to sensory stimulation compared to pyramidal cells, which respond to the adequate stimulus. It is explained by the convergent input onto PV cells from a large number of principal neurons with a wide range of spatial or orientation preferences. Finally, PV interneurons are involved in the regulation of plasticity and learning, either in visual cortex, through the control of the ocular dominance plasticity, in auditory cortex, playing a critical role for associative fear learning, or in prefrontal cortex, controlling the reward seeking behavior. In all these events, the suppression of PV interneurons (i.e., disinhibition of pyramidal neurons) produces certain forms of
24
Introduction learning, whereas activation of PV cells promotes extinction. Nevertheless, also learning promotes plasticity modulation in PV interneurons (Donato et al., 2013; Hu et al., 2014; Ognjanovski et al., 2017)
6.6. Role of PV interneurons in neurological disorders A great number of works demonstrated the importance of a proper balance of excitation and inhibition (E/I) in cortical circuits to ensure an appropriate cortical information processing and regulation of behaviors. Multiple evidences implicate imbalances between excitatory and inhibitory activity as a common pathophysiological mechanism in many psychiatric disorders, such as schizophrenia, autism or epilepsy. Importantly, alterations in the inhibitory system are consistently identified in animal models of psychiatric disorders; specifically, mutations altering the function of PV interneurons play an essential role in the observed symptoms. Changes in PV interneurons can affect the input, ouput, intrinsic properties or cell density, which at the end tilt the E/I balance (Gao and Penzes, 2015; Lee et al., 2017; Selten et al., 2018). Also mutations or knock‐down of some genes in PV interneurons have been related to these diseases, for example, Scn1a with epilepsy or ErbB4 with schizophrenia (Benarroch, 2013; del Pino et al., 2013; Selten et al., 2018). Moreover, in patients with schizophrenia, it is consistently detected a reduction in the number or length of chandelier cell cartridges (the long linear axon terminals) and a misregulation of proteins associated with chandelier synapses (Selten et al., 2018). Finally, also other components are relevant for maintaining PV cell physiology, for example, a loss of the perineuronal net around PV interneurons is observed after an epileptic episode (Rankin‐Gee et al., 2015) and also in schizophrenia (Cabungcal et al., 2013; Enwright et al., 2016). In addition, a neurotrophic role for PV neurons has been proposed at the striatum (d’Anglemont de Tassigny et al., 2015).
7. Single‐cell RNA sequencing 7.1. First milestones Classically, cells in the nervous system have been classified attending to different features, including morphology, location, cell targeting, marker genes and electrophysiological properties. Indeed, many works have specifically focused on GABAergic interneurons classification, attempting to establish discrete cell types based on several of these features (Defelipe et al., 2013; Mihaljević et al., 2014, 2015). Although providing very qualified information, these characteristics do not generate a definitive cell type identification. Single‐cell RNA sequencing is a high‐throughput sequencing technology that is currently used to classify cells in different tissues, including spleen (Jaitin et al., 2014), pancreas (Enge et al., 2017) or lung epithelium (Treutlein et al., 2014). In the brain, a tremendous quantity of published works have revealed distinct neuronal and non‐neuronal populations even at specific brain areas (Harris et al., 2018; Hochgerner et al., 2018; Johnson et al., 2015; Luo et al., 2015; Marques et al., 2016; Mayer et al., 2018; Paul et al., 2017; Pollen et al., 2014; Poulin et al., 2016; Tasic et al., 2018, 2016; The Tabula Muris
25 Introduction kk
Consortium, 2018; Usoskin et al., 2015; Zeisel et al., 2015). The first transcriptomic experiment produced by single‐cell RNA sequencing was published by Tang et al., in 2009 using a mouse blastomere, after what a wide range of studies began in this field and were published only few years later.
7.2. Cell isolation and sequencing methods To sequence the whole mRNA from a single cell, two principal challenges need to be overcome: (1) isolation/capturing of single cells and (2) amplification of a minimal amount of mRNA per cell. For the first goal, several approaches have been established, such as the widely‐used fluorescent‐activated cell sorting (FACS). Moreover, some companies have developed advanced cell dispenser technologies to capture individual cells before sequencing. For example, the Fluidigm C1 platform was designed to isolate cells using a microfluidic chip where cells remain arrested in capture sites; later, through imaging techniques, it is possible to select fluorescent/healthy cells (Islam et al., 2014; Marques et al., 2016; Muñoz‐Manchado et al., 2018; Pollen et al., 2014; Zeisel et al., 2015). There are also microdoplet methods, such those used for Drop‐Seq or 10X Genomics, which involve cell isolation using aqueous droplets in a non‐aqueous suspension (Grün and Van Oudenaarden, 2015; Kolodziejczyk et al., 2015; Macosko et al., 2015; Poulin et al., 2016). Other methods also used for cell isolation are laser capture microdissection, cytoplasmic aspiration (such as Patch‐Seq), and manual or automated micropipetting (Grün and Van Oudenaarden, 2015; Kolodziejczyk et al., 2015; Muñoz‐Manchado et al., 2018; Poulin et al., 2016). More recently, the WaferGen ICELL8 platform has emerged to increase the throughput of microfluidic cell capture (Fluidigm C1) by increasing the number of cells per chip (9600 wells).
Different methods for RNA sequencing have been developed, being Smart‐seq (or the improved method Smart‐seq2 by Picelli et al., 2013) and STRT‐seq (Islam et al., 2011) (or the advanced version STRT‐ seq‐2i by Hochgerner et al., 2017) the most used protocols (Grün and Van Oudenaarden, 2015; Kolodziejczyk et al., 2015; Poulin et al., 2016). These techniques employ different approaches to label and count mRNA molecules, including UMIs (unique molecular identifiers) or RPKM (reads per kilobase per million mapped reads) for STRT‐seq and Smart‐seq2 protocols, respectively. Nowadays, both methods are usually run on Illumina HiSeq technology.
Isolating whole single‐cells might result difficult in highly interconnected tissues, such as the brain. Indeed, some cell types are more vulnerable to tissue dissociation than others, including parvalbumin interneurons, which adds the need to selectively enrich this population using Cre‐driver lines to ensure a sufficient sampling (Bakken et al., 2018; Lake et al., 2017). scRNA‐seq includes more likely the cytoplasmic RNA content, but also nuclear RNA might be sequenced since using the whole cell. Besides, nucleolar, mitochondrial and other organelle‐specific RNAs might be also caught. Nowadays, there are emerging works showing a comparable result using single‐nucleus transcriptomics, which are apparently more resistant to cell isolation‐based transcriptional artifacts derived from whole cells (Abdelmoez et al., 2018;
26
Introduction
Bakken et al., 2018; Grindberg et al., 2013; Lake et al., 2017). Nevertheless, scRNA‐seq can reveal RNA abundancy with high quantitative precision but, so far, a limitation is that we only capture a static snapshot at a time point in a cell (La Manno et al., 2018).
7.3. Up‐to‐date PV classification by scRNA‐seq Normal brain function is maintained by a varied set of differentiated cell types that develop specialized roles to support neural processes. In cortex, neuronal network integration is essential to discern activity and behavior patterns, involving a high diversity of GABAergic interneurons. Several classes of GABAergic interneurons have been described based on single‐cell RNA sequencing in specific cortical, hippocampal or striatal regions by multiple published works (Harris et al., 2018; Hochgerner et al., 2017; Muñoz‐Manchado et al., 2018; Paul et al., 2017; Poulin et al., 2016; Tasic et al., 2018; Zeisel et al., 2015). Interestingly, also several subpopulations of PV interneurons have been defined by some of these publications: e.g., Harris et al., 2018 differentiated between 7 subclasses of hippocampal CA1 PV cells, and Tasic et al., 2018 described 10 PV populations in visual and anterior lateral motor cortex (they can be also differentiated by cortical layer origin). These works are attempting to define discrete PV interneurons classification, although demonstrating a variability that depends on read depth, sample size, sequencing method and clustering algorithms, which should be considered for a complete understanding. PV basket and chandelier cells expression patterns based on synapse specificity during development (Favuzzi et al., 2019) have been also compared with defined clusters from scRNA‐seq data (Paul et al., 2017; Tasic et al., 2016; Zeisel et al., 2015). Hopefully, future investigations will reveal new insights into region‐, layer‐ and synapse‐specific PV identities.
7.4. Transcriptomics in neurodegenerative diseases Transcriptome studies about neurodegenerative disorders are not currently very abundant. Gene expression profilings have been used to quantify expression levels of dozens to thousands of genes between control and patient brains to uncover the molecular biology of neurodegenerative diseases, including Alzheimer’s disease, Parkinson’s disease and schizophrenia (Cooper‐Knock et al., 2012; Labadorf et al., 2018; Mufson et al., 2006). However, these approaches analyze a limited number of genes and thus provide only sets of differentially expressed (DE) genes that may be involve in some biological pathways. The single‐cell RNA seq technology provides a high‐throughput tool to exhaustively identify DE genes along the whole transcriptome that could arise as powerful candidate markers of the neurodegenerative condition. Nevertheless, the majority of scRNA‐seq publications are not focused on the comparison between two conditions (e.g. control versus injured/knock‐out/treated mice, human tissues or cells) but on cell type classification. Few works are now emerging on this topic, where they compare a specific cell type population, finding differentially expressed genes in the impaired condition and/or the emergence or disappearance of certain subpopulations (Butler et al., 2018; Munguba et al., 2019; Tay et al., 2018). We
27
Introduction kk consider that scRNA sequencing provides a robust method that can be applied to discern gene expression patterns in neurodegenerative disorders.
28
GOALS
Goals
Our knowledge of the molecular mechanisms underlying neurodegeneration is still poor, at least to be able to design therapies to ameliorate neuronal and synaptic dysfunction. Neurodegeneration is a multistep progressive process that does not start as a generalized synchronized neuronal phenomena within the brain. Sick neurons co‐exist with healthy neurons and, likely, the healthy brain reacts upon the neurodegeneration of damaged neuronal subpopulations. On the other hand, at the single cell level, neuronal degeneration is probable not a mere passive “wear and tear” phenomenon but, instead, an active process involving progressive homeostatic molecular changes in gene expression. Indeed, it is currently unknown how just synaptic degeneration or even long term synaptic dysfunction impact on the dynamics of the neuronal transcriptome. As described in the Introduction, previous studies have identified CSPα/DNAJC5 as a synaptic vesicle protein that, by so far not well understood mechanisms, promotes the maintenance of the synapses of highly active neurons such as fast‐spiking GABAergic interneurons. Genetically modified KO mice lacking CSPα/DNAJC5 die around one month of age and, therefore, present limitations to study molecular mechanisms of neurodegeneration beyond early adulthood. This thesis pursues the investigation of the neural and molecular mechanisms along the eadult lif underlying synaptic degeneration upon the genetic deletion of CSPα/DNAJC5 specifically in PV GABAergic interneurons.
Accordingly, the main goals of this thesis are:
1. Generation and validation of a novel conditional KO mouse specifically lacking CSPα/DNAJC5 in PV GABAergic interneurons expressing, in the same neurons, a fluorescent marker and an optogenetic actuator suitable for light‐mediated stimulation.
2. Phenotypic characterization of the mice, including neurological, biochemical, immunohistochemical and electrophysiological analysis of PV GABAergic neurons and synapses.
3. Investigation of gene expression changes associated to synaptic and neuronal dysfunction in the absence of CSPα/DNAJC5 based on single‐cell transcriptomics comparative analysis.
31
MATERIALS AND METHODS
Materials and Methods
1. Mice
To fulfil the experimental tasks proposed in this thesis, several mouse lines were required. For initial experiments, we used conventional Dnajc5 knockout (KO) animals (The Jackson laboratory, stock no. 006392, Thomas C. Südhof) (Fernández‐Chacón et al., 2004), although conditional mouse lines carrying Dnajc5 floxed alleles were generated for the majority of the procedures described in this work. For experiments involving conditional Dnajc5flox/flox mice, two Cre‐recombinase‐expressing mouse lines were purchased for breeding: UBC‐Cre‐ERT2 (The Jackson Laboratory, stock no: 007001, Eric Brown) and PVcre (The Jackson Laboratory, stock no: 008069, Silvia Arber) (Hippenmeyer et al., 2005). Finally, we have also taken advantage of the Ai27D mouse line (The Jackson Laboratory, stock no: 012567, Hongkui Zeng) (Madisen et al., 2012) to introduce the channelrhodopsin‐2/tdTomato (ChR2/tdTomato) fusion protein in the experimental PVcre mouse line, and, then, being able to label PV cells using tdTomato as reporter. The strategies for the generation of the Dnajc5flox/flox and the two Cre‐recombinase‐expressing mouse lines are described in detail below.
All procedures involving animal research were performed in accordance with the European Union Directive 2010/63/EU about the protection of animals used for scientific purposes and approved by the Committee of Animal Use for Research at the University of Seville. Animals were kept in animal facilities (IBiS and campus Macarena) with a 12/12‐h light/dark cycle and unrestricted access to food and water.
1.1. Generation of the conditional Dnajc5flox/flox mouse line Embryonic stem (ES) cells were acquired from the European Conditional Mouse Mutagenesis Program (EUCOMM). The mutant cells were generated by recombinant insertion of a knock‐out first mutant construct containing loxP sites flanking exon 3 of the mouse Dnajc5 gene (ENSMUSE00000170833 chr2: 181282045‐181282258). The full sequence of the mouse targeted KO‐first, conditional ready, lacZ‐ tagged mutant allele Dnajc5:tm1a (EUCOMM) from the Wellcome Trust Sanger Institute is available at GenBank (accession number JN952788). Three different ES cell clones were examined based on the karyotype and their morphological and growing properties in culture. One of them (JM8A3.N1‐E01) was chosen and injected into C57BL/6JOlaHsd mouse blastocysts at the Unitat d’Animals Transgènics (Universitat Autònoma de Barcelona). A total of 12 chimeric mice were obtained with variable degree of chimerism and among them one male with high chimerism (90%) was used for breeding with C57BL/6NHsd female mice and gave germline transmission. Interbred of resultant germline mutant mice (Tm1a heterozygous mice) yielded Tm1a homozygous mice that lacked Dnajc5 and developed the typical early neurological phenotype described for conventional Dnajc5 KO mice. Next, germline mutant mice were crossbred with ROSA26:FLPe knock‐in mice (The Jackson Laboratory, Stock no. 003946) to remove the FRT‐ flanked splice acceptor‐ site, β gal reporter, and neomycin resistance cassette. Resultant Dnajc5flox/+ mice were interbred to generate homozygous mutant mice with conditional knockout potential (Dnajc5flox/flox).
35 Materials and Methods
Dnajc5flox/flox mice were genotyped by PCR with the primers FMS‐2014‐14 and FMS‐2014‐15, generating a wild‐type product of 678 bp and a mutant product of 900 bp (Table 1). We have recently described this mouse line in Nieto‐González et al., 2019.
1.2. Generation of the UBC‐Cre‐ERT2:Dnajc5flox mouse line We also generated a second mouse line for our study, which was mostly used as control sample of Dnajc5‐floxed recombination after Cre activity. This mouse line was called UBC‐Cre‐ERT2:Dnajc5flox because it expresses a fusion gene of Cre‐recombinase with the estrogen receptor ligand binding domain (ERT2), under control of the ubiquitin C promoter, which restricts Cre‐activity to tamoxifen exposition; and also contains the Dnajc5 floxed allele (from the conditional strain). First, Dnajc5flox/flox mice were crossed with hemizygous UBC‐Cre‐ERT2 animals and crossbred always against Dnajc5flox/flox mice to obtain UBC‐ Cre‐ERT2hemi:Dnajc5flox/flox mice (first parental). Finally, these mice were again crossed with Dnajc5flox/flox animals (second parental) to generate experimental mice: UBC‐Cre‐ERT2hemi:Dnajc5flox/flox mutant mice (carrying Cre‐recombinase) and Dnajc5flox/flox littermate controls (without Cre‐recombinase). At P60, all experimental mice were fed with tamoxifen‐enriched diet (TAM400Cre/ER, Envigo) for 30 days to induce genetic removal of the Dnajc5 locus in every tissue, and samples were collected 10 days post‐tamoxifen (P100). The nomenclature used in this work is UBC‐Cre‐ERT2:Dnajc5flox/flox (for mutant mice) and Dnajc5flox/flox (for control mice). UBC‐Cre‐ERT2 transgene was detected with the primers oIMR1084 and oIMR1085 from The Jackson Laboratory, while an internal positive control was obtained using the primers oIMR7338 and oIMR7339 from The Jackson Laboratory. The internal positive control product was 324 bp and the transgene produced a band of 100 bp (Table 1).
1.3. Generation of the PVcre:Ai27D:Dnajc5flox mouse line The principal mouse line employed in this study was generated using a breeding strategy involving some of the strains mentioned above. This line was called PVcre:Ai27D:Dnajc5flox, because it contains a Cre‐ recombinase under control of the parvalbumin promoter (PVcre knockin animals), a ChR2‐tdTomato construct inserted into the ROSA26 locus (Ai27D animals) and the Dnajc5 floxed allele (Dnajc5flox/flox conditional mice). First, homozygous PVcre mice were crossed with heterozygous conventional Dnajc5 mice to introduce the Dnajc5 knock‐out allele in the strain until we got the PVcre allele in homozygosis. Secondly, PVcre homo:Dnajc5het mice were crossed with homozygous Ai27D animals to generate mice carrying the ChR2/tdTomato fusion protein. Once we obtained both PVcre and Ai27D alleles in homozygosis using crossbreeding strategies, PVcre homo:Ai27Dhomo:Dnajc5het animals (first parental) were crossed with Dnajc5flox/flox mice (second parental) to generate experimental mice: PVcre hemi:Ai27Dhemi:Dnajc5flox/‐ mutant mice and PVcre hemi:Ai27Dhemi:Dnajc5flox/+ littermate controls. The simplified nomenclature used in this work is PVcre:Ai27D:Dnajc5flox/+ or controls, and PVcre :Ai27D:Dnajc5flox/‐ or mutants, because both PVcre and Ai27D alleles were always in hemizygosis (hemi) (Figure 10). Nevertheless, in some occasions, we have even
36 Materials and Methods shortened names to Dnajc5flox/‐ or F/‐ for mutant mice, and Dnajc5flox/+ or F/+ for control mice, due to available space in figures and legends. During single‐cell sequencing analysis, we named the mice as follows: WT for PVcre:Ai27D:Dnajc5WT, Flox+ for PVcre:Ai27D:Dnajc5flox/+ and Flox‐ for PVcre:Ai27D:Dnajc5flox/‐ due to nomenclature limitations in R language. PVcre knockin allele was detected using the primers oIMR1084 and oIMR1085 from The Jackson Laboratory, while the wild‐type allele was recognized with the primers oIMR8291 and oIMR8290 from The Jackson Laboratory, allowing the identification of homozygous (102 bp), hemizygous (102 and 500 bp) and wild‐type (500 bp) mice. Regarding the Ai27D construct, the primers oIMR9103 and oIMR9105 identified the mutant allele and primers oIMR9120 and oIMR9121 (all from The Jackson Laboratory) detected the wild‐type allele, leading to distinguish between homozygous (315 bp), hemizygous (315 and 297 bp) and wild‐type (297 bp) animals. To discriminate between Dnajc5flox/+ and Dnajc5flox/‐, we identified the floxed allele with primers FMS‐2014‐14 and FMS‐2014‐15 (as in Dnajc5flox/flox conditional mice), but we also genotyped the wild‐type and knock‐out alleles for Dnajc5. It was done using a single‐tube PCR that shared a common reverse primer 25939, while forward primer 25938 was used for wild‐type allele detection and forward primer oIMR8574 for the mutant allele. It produced a homozygous product of 150 pb, a heterozygous product with 150 and 336 bp, and a wild‐type product of 336 bp. This PCR was also used to genotype the conventional Dnajc5 mouse line (Table 1).
2. Fluorescent‐activated cell sorting (FACS)
For Cre‐recombinase validation experiments (genomic DNA and mRNA) and real‐time qRT‐PCR, FACS was performed in a BD FACSJazz cell sorter, previously prepared for RNAse free conditions (10 min RNase‐ExitusPlus, A7153‐1000RF, Panreac; 10 min DEPC water), using a 100 µm nozzle and a 585/29 filter for a 561 nm laser, achieving proper tdTomato‐fluorescence detection (maximum excitation at 554 nm and maximum emission at 582 nm). Gating strategy was set as follows: (1) selection of cells with forward (FSC‐H) and side (SSC‐H) scatters, (2) selection of singlets based on FSC‐H and Trigger Pulse Width (FSC‐ W), (3) selection of tdTomato+ (585/29 [561nm]) cells. Sorted cells were kept on ice until processing. The solution in which cells were sorted depended on downstream application, hence it is specified in each section.
FACS performed for single‐cell RNA seq experiments is detailed later in the 19.Single‐cell RNA sequencing protocol section.
37 Materials and Methods
38 Materials and Methods
3. Validation of conditional Dnajc5flox/flox mice
According to the International Mouse Phenotyping Consortium (IMPC), Ensemble and NCBI, the Dnajc5 gene generates four possible RNA splice variants: ENSMUST00000108796, ENSMUST00000108797, ENSMUST00000072334, ENSMUST00000116365. Two additional variants are also reported by IMPC and Ensembl (ENSMUST00000152578, ENSMUST00000141523), but the first composed an incomplete transcript (it has a coding sequence incomplete in 3’ end) and the later does not contain an open reading frame. The first three variants comprise mRNA transcripts that encode the same protein of 198 aa (NP_001258513.1, NP_001258514.1, NP_058055.1); this protein is exclusively encoded by exons 2‐5 in each of the three transcripts (consensus coding sequence: CCDS17215.1) (S1). The fourth variant is supposed to encode a peptide of 167 aa, but it is represented as non‐coding because it contains an additional exon between exons 4 and 5 in 3’ end that produces a nonsense‐mediated mRNA decay. After removal of Dnajc5, these four variants loss the floxed exon 3, generating four shorter mRNA derived transcripts. In the validation strategies for genomic DNA and mRNA detection of Dnajc5, the primers were not designed to detect the non‐coding variant 4, while the other three variants were undistinguishable detected.
______
Figure 10. Breeding strategy to generate the PVcre:Ai27D:Dnajc5flox mouse line. Scheme of the strategy to establish both parental and experimental mouse colonies for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes. PVcre locus is displayed as described in The Jackson Laboratory documentation, as well as the Rosa26 locus, which is modified for channelrodhopsin‐2/tdTomato fusion expression. Briefly, PVcre locus was designed containing Cre coding sequence, internal ribosome entry site (IRES), and polyadenylation site sequence inserted into the 3' UTR of exon 5. Ai27D locus comprised a CMV‐IE enhancer/chicken beta‐actin/rabbit beta‐globin hybrid promoter (CAG), an FRT site, a loxP‐flanked STOP cassette (with stop codons in all 3 reading frames and a triple polyA signal), a mammalianized ChR2(H134R)‐tdTomato fusion gene, a woodchuck hepatitis virus post‐transcriptional regulatory element (WPRE; to enhance protein expression), a BGH polyA signal, and a hybrid AttL site resulting from the removal with a site‐specific recombinase (PhiC31o) of AttB/AttP‐flanked PGK‐TFR ‐Neo‐polyA cassette. Generation of the fusion gene was performed using the first 315 amino acids of channelrhodopsin‐2 (derived from the green alga Chlamydomonas reinhardtii) that were modified to harbor codons optimized for mammalian expression and a gain‐of‐function H134R substitution (CAC to CGC) to enhance stationary photocurrents. This sequence was fused in‐frame to the amino terminus of a tdTomato sequence (two copies). Final construct was inserted into intron 2 of the Gt(ROSA)26Sor locus. Dnajc5 floxed locus was generated by EUCOMM, containing a FRT site and two loxP sites flanking exon 3. Note the asterisk specifying that PVcre line was already carrying Dnajc5 in heterozygosis, allowing the generation of Dnajc5flox/+ and Dnajc5flox/‐ locus. Only a single allele is shown in the picture, and the WT and KO sequences for Dnacj5 (carried by the two experimental mice) are not displayed.
39 Materials and Methods
Locus Primer Sequence 5’ 3’ Primer Size name type Wild‐type 25938 CAAGAATGCAACCTCAGATGAC Forward Homozygous = 150 bp Conventional 25939 CTTTTAAGTGTGTTTACTTTTTGGTG Common Heterozygous = 150 & 336 bp Dnajc5 KO Mutant Wild‐type = 336 bp oIMR8574 GAGCGCGCGCGGCGGAGTTGTTGAC Forward FMS‐2014‐ Transgene Homozygous = 900 bp Conditional TATCGGTAAGCAGCCGTGTTAACC 14 Forward Hemizygous = 900 & 678 bp Dnajc5flox/flox FMS‐2014‐ Transgene Wild‐type = 678 bp TATAGCATTCACTCCTGCCAACCC 15 Reverse Mutant oIMR1084 GCGGTCTGGCAGTAAAAACTATC Forward Mutant Homozygous = 102 bp oIMR1085 GTGAAACAGCATTGCTGTCACTT PVcre Reverse Hemizygous = 102 & 500 bp Wild‐type Wild‐type = 500 bp oIMR8291 AGTACCAAGCAGGCAGGAGA Forward Wild‐type oIMR8290 CAGAGCAGGCATGGTGACTA Reverse Wild type oIMR9020 AAGGGAGCTGCAGTGGAGTA Forward Wild type Homozygous = 315 bp oIMR9021 CCGAAAATCTGTGGGAAGTC Ai27D Reverse Hemizygous = 315 & 297 bp Mutant Wild‐type = 297 bp oIMR9103 GGCATTAAAGCAGCGTATCC Forward Mutant oIMR9105 CTGTTCCTGTACGGCATGG Reverse Transgene oIMR1084 GCGGTCTGGCAGTAAAAACTATC Forward Transgene oIMR1085 GTGAAACAGCATTGCTGTCACTT Reverse UBC‐Cre‐ERT2 Internal Transgene = 100 bp oIMR7338 CTAGGCCACAGAATTGAAAGATCT Control Internal control = 324 bp Forward Internal oIMR7339 GTAGGTGGAAATTCTAGCATCATCC Control Reverse Table 1. Primers for mouse genotyping. Brief information about names, sequences, types and product sizes for all primers used during genotyping (genomic DNA) of each mouse line generated and/or employed in this experimental work.
3.1. Genomic DNA (gDNA) validation 3.1.1. gDNA extraction tdTomato+ (tdT+) and tdTomato‐ (tdT‐) cells from PVcre:Ai27D:Dnajc5flox/+ (n=3) and PVcre:Ai27D:Dnajc5flox/‐ (n=3) littermates were directly sorted into a solution containing Proteinase K and RNase A in Buffer CL, which was the first step of protocol C (for cells) from G‐spin™ Total DNA Extraction Kit (17046, Intron). For band‐size control, gDNA was also extracted from tail samples from a littermate pair of Dnajc5 WT and KO mice (P30), and a littermate pair of Dnajc5flox/flox and UBC‐cre‐ERT2:Dnajc5flox/flox animals (P100). For these samples, protocol B (for tissue) of the same kit was followed, which also started
40 Materials and Methods by adding the same Proteinase K and RNase A solution to samples. After that, we immediately followed manufacturer’s instructions. Briefly, this solution was vortexed vigorously and incubated at 56°C for 15 min; then Buffer BL was added, samples mixed and incubated at 70°C for 5 min. In case of using tail samples, before continuing, an additional centrifugation step was performed to remove non‐lysed tissue, and supernatant was transferred to a new tube. Absolute ethanol was added into lysates, mixed gently by inverting and centrifuged to collect drops from walls. Mixture was then transferred to a spin column contained in a 2 ml tube, closed and centrifuged at 15700 g for 1 min. Filtrate was discarded, Buffer WA added to the column and again centrifuged for 1 min at 15700 g. Filtrate was removed again, Buffer WB added to the column, centrifuged for 1 min at 15700 g and filtrate discarded. Column was additionally centrifuged for 1 min to dry the membrane before placing it into a new tube. Autoclaved miliQ water was used for final elution of gDNA from the membrane by incubation for 1 min at room temperature and centrifugation for 1 min at 15700 g. DNA concentration and quality was assessed by NanoDrop 2000 (Thermo Scientific) using the same water for blank subtraction. All samples contained around 3 ng/µl of gDNA and were stored at ‐20°C until use.
3.1.2. PCR for Dnajc5 gDNA PCR conditions were set up using different amounts of starting material from control samples, and a temperature range based on the annealing properties was established for both SLB‐2017‐01 and SLB‐ 2017‐02 primers. Final PCR mix was fixed to 15 µl (8.3 µl miliQ water, 2.5 µl Buffer PCR 10X (Tris‐HCl 670 mM, ammonium sulfate 160 mM, Tween 20 0.1 mM, pH 8.8), 1 µl MgCl2 50 mM, 1 µl SLB‐2017‐01 10 µM, 1 µl SLB‐2017‐02 10 µM (Table 2), 0.2 µl dNTPs mix 100 mM, 1 µl handmade Taq polymerase). 10 µl gDNA was added in case of sorted samples (≈30 ng), or 6 µl gDNA (≈20 ng) plus 4 µl autoclaved miliQ water in case of tail control samples, reaching a final volume of 25 µl per reaction. An increased amount of sorted samples was added to the PCR mix because a poor DNA integrity and quality was found after extraction of the samples, due to sorting stress and low starting material. Finally, PCR was run using the following parameters: 95°C 5min; 35 cycles: 95°C 30 sec, 60°C 30 sec, 72°C 1 min 30 sec (to favor elongation of 1029 bp band); 72°C 5 min; 10°C hold. A 2%‐agarose gel containing ethidium bromide was used to separate the amplified samples supplemented with loading buffer 5X (bromophenol blue 0.05% w/v, sucrose 40% w/v, EDTA 0.1 M pH 8), using PowerPac 200 (Bio‐Rad) at 100mV during 40 min for better resolution. Gel was revealed and image taken using the BioDoc‐It™ 220 Imaging System (UVP) with an M‐26 ultraviolet (UV) transilluminator.
3.2. RNA validation 3.2.1. RNA extraction Similar control samples as for gDNA validation were used: the whole brain from a littermate pair of Dnajc5 WT and KO mice (P30), and the cerebral cortex from a littermate pair of post‐TMX Dnajc5flox/flox
41
Materials and Methods and UBC‐cre‐ERT2:Dnajc5flox/flox animals (P100). The RNA was extracted by RNeasy Mini Kit (74104, Qiagen) following manufacturer’s instructions. Few modifications were made to this protocol, such as using TRIzol reagent (15596026, Ambion), chloroform (22711.260, AnalaR NORMAPUR) and 70% ethanol (1009832500, EMSURE) to efficiently lysate, separate and clear, respectively, samples before transferring them to columns and continuing with the custom Qiagen protocol. On the other hand, tdT+ and tdT‐ cells from PVcre:Ai27D:Dnajc5flox/+ (n=3) and PVcre:Ai27D:Dnajc5flox/‐ (n=3) littermates were directly sorted into 200 µl of RNAlaterTM Solution (AM7020, Invitrogen) to improve RNA storage and stabilization at ‐80°C until processing. For sorted cells, we carried out a specific protocol that improved RNA obtaining. First, frozen cells were thawed on ice, centrifuged at 2600 g for 1 min at 4°C and supernatant carefully discarded using a pipette, leaving a minimal volume in the tube due to not visible pellet. 500 µl of TRIzol was added to samples and homogenized by vortex for 1 min, then left to rest at room temperature at least 5 min. Chloroform was then supplemented, and samples vigorously shaken by hand for 15 sec and then left to incubate at room temperature for 2 min. A centrifugation step at 12000 g for 15 min at 4°C was performed to help phenol/chloroform‐based phase separation. Upper aqueous phase was collected in a new tube, and 250 µl 2‐Propanol 100% (I9516, Sigma) and 0.5 µl glycogen (stock 20mg/ml, 10901393001, Roche) were added before shaking by hand, and incubated O/N at ‐20°C to facilitate RNA precipitation. Next day, samples were centrifuged at 12000 g for 10 min at 4°C, supernatant discarded and pellet washed with 1 ml 75% ethanol by vortexing. Samples were again centrifuged at 7500 g for 10 min at 4°C, supernatant removed and pellet was air dried during 10‐15 min; then, RNA was resuspended in 15 µl RNase‐free water and incubated at 55°C during 15 min. 2 µl RNA was used to test RNA quality and quantity by 2100 Bioanalyzer (Eukaryote Total RNA Pico, Agilent) before finally kept at ‐80°C until reverse‐transcription. RNA Integrity Number (RIN) for all samples were around 7‐8, which were quite optimum concerning sorted cells, and concentrations ranged from 1‐5 ng/µl, hence no corrections were made to equalize sample concentration due to low quantity.
3.2.2. Reverse‐transcription (RT) to cDNA The RNA reverse‐transcription (RT) was performed using a two‐step protocol based on the Transcriptor First Strand cDNA synthesis kit (04379012001, Roche). Starting material for the first reaction was 1 µg of RNA. Because we could not obtain such amount in our samples, the first reaction mix was changed to include as much starting material as possible; in this case, the total volume of our remaining RNA (12 µl RNA), was mixed to 1 µl of Random hexamer primer 600 µM to obtain the maximum cDNA quantity. This mixture was incubated in a thermal cycler for 10 min at 65°C for denaturation and cooled down on ice. The second reaction was then prepared by mixing 4 µl Transcriptor RT Reaction Buffer 5X, 0.5 µl RNase Inhibitor 40 U/µl, 2 µl dNTP Mix 10mM each and 0.5 µl Transcriptor Reverse Transcriptase 20 U/µl, and finally 7 µl of this mixture was added to the primer‐template mixture tube. Reagents were mixed
42
Materials and Methods and centrifuged to collect samples at the bottom. Reverse‐transcription was run using the following program settings: 25°C 10 min; 50°C 60 min; 85°C 5 min to inactivate transcriptase; 4°C hold to stop reaction. Samples were directly frozen at ‐20°C until use, presumably containing the same amount of cDNA as starting RNA (from 12 to 60 ng in a total of 20 µl).
3.2.3. PCR for Dnajc5 cDNA PCR conditions and mix were first set up with control samples and an extra sorted sample. PCR mix was prepared, including 17.3 µl miliQ water, 2.5 µl Buffer PCR 10X, 1 µl MgCl2 50 mM, 0.5 µl MLM‐2005‐ 01 10 µM, 0.5 µl MLM‐2005‐02 10 µM (Table 2; S1), 0.2 µl dNTPs mix 100 mM and 1 µl handmade Taq polymerase. 2 µl of cDNA samples was added to 23 µl of PCR mix, reaching a total volume of 25 µl per reaction. Reactions were run according to these program settings: 95°C 5min; 35 cycles: 95°C 30 sec, 60°C 30 sec, 72°C 30 sec; 72°C 2 min; 10°C hold. Samples were then supplemented with loading buffer 5X and separated in a 2%‐agarose gel with ethidium bromide by PowerPac 200 (Bio‐Rad) at 100mV during 40 min for better resolution. Gel was visualized using the BioDoc‐It™ 220 Imaging System (UVP) with an M‐26 ultraviolet (UV) transilluminator and pictures were taken.
Primer Primer Locus Sequence 5’ 3’ Template Size name type SLB‐ CACGTCAGATAGTCTTGGGGTCT Forward gDNA Wild‐type = 765 bp 2017‐01 Tm1c = 1029 bp SLB‐ Conditional GCAAATGGTTGTGCTCCATGTACG Reverse gDNA Tm1d = 245 bp 2017‐02 Dnajc5flox/flox MLM‐ mRNA ACCATGTTCTTGGACTGGACA Forward Wild‐type or floxed 2005‐01 (cDNA) without recombination MLM‐ mRNA = 348 bp AGCAGCAGTTAAAGCAACAGC Reverse 2005‐02 (cDNA) Mutant = 134 bp Table 2. Primers for Dnajc5 gDNA and mRNA detection. Summary including name, sequence, type, template and product size for primers used to detect Dnajc5 genomic DNA (gDNA) and mRNA (cDNA), to validate Cre‐dependent recombination at this locus.
3.3. Protein validation According to the International Mouse Phenotyping Consortium (IMPC), the analysis of the mutant allele Dnajc5tm1a(EUCOMM)Wtsi predicts that the four mRNA transcript variants for Dnajc5 would generate two truncated protein products after removal of the floxed region. The three mRNA transcripts coding the same wild‐type protein containing 198 aa (ENSMUST00000072334, ENSMUST00000108797, ENSMUST00000108796), after removal of Dnajc5, would produce the following peptide with 112 aa: MADQRQRSLSTSGESLYHVLGLDKNATSDDIKKSYRRCLLFVASSPAATAAAVCAVALTAAVGNASPRHLRVRRQNSTY PLKTWRHSCSLMKGRLQTHRSSYSQHLPQRPPS*, with an estimated molecular weight of 12.38 kDa (S1). The fourth wild‐type non‐coding transcript (ENSMUST00000116365), after removal of Dnajc5, would generate the following peptide consisting of 112 aa:
43
Materials and Methods
MADQRQRSLSTSGESLYHVLGLDKNATSDDIKKSYRRCLLFVASSPAATAAAVCAVALTAAVGNASPRHLRVRRQNSTY PLKTWRHSCSLMKGEGTDTVPRVFVVASGTVEV*, with an estimated molecular weight of 12.07 KDa (S1). For both predicted truncated proteins, the first 36 aa would correspond to the N‐terminus of CSPα/DNAJC5 while the rest would be an aberrant polypeptide. If those polypeptides were stable, the DnaJ domain (aminoacids 15 to 84) would be disrupted because only the first 22 aminoacids of such a domain would be preserved, according to the Conserved Domains tool from NCBI (S1). Hence, in order to check if indeed any truncated version of CSPα/DNAJC5 was produced, we analyzed protein expression around 12 KDa in the UBC‐Cre‐ERT2:Dnajc5flox/flox and PVcre:Ai27D:Dnajc5flox mouse lines using immunoblots with R807, an antibody raised against a GST‐fusion protein of CSPα/DNAJC5 that binds to the full‐length version of the protein (Fernández‐Chacón et al., 2004; Tobaben et al., 2001) (Table 4). See 8.Immunoblotting section for methodology description.
4. Real‐time quantitative reverse transcription‐PCR (real‐time qRT‐PCR) 4.1. RNA extraction The same protocol for RNA extraction on tdT+ and tdT‐ sorted cells was carried out for real‐time qRT‐PCR in littermate pairs of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals (n=3). RNA quality and concentration were assessed by 2100 Bioanalyzer, and RINs were found optimal (around 7‐8.5) for all samples, except for two samples where the RIN was not possible to be determined. Anyway, all samples were stored at ‐80 °C and used for downstream analysis.
4.2. Reverse‐transcription and cDNA amplification Because we previously observed very low amount (1‐5 ng/μl) of starting RNA material from sorted cells, and samples were depleted after a couple of tests, we decided to maximize pre‐amplified RNA before reverse‐transcription using the QuantiTect Whole Transcriptome Kit (207043, Qiagen) and following manufacturer’s instructions. Briefly, 10 ng of each sample was prepared in a total volume of 5 µl nuclease‐ free water, and 5 µl of RT mix (4 µl T‐Script Buffer, 1 µl T‐Script Enzyme) was added by pipetting and then centrifuged. This mixture was incubated in a thermocycler for reverse‐transcription: 37°C 30 min; 95°C 5 min; 22°C hold. Secondly, the addition of 10 µl ligation mix (6 µl Ligation Buffer, 2 µl Ligation Reagent, 1 µl Ligation Enzyme 1, 1 µl Ligation Enzyme 2) to the RT reaction and later incubation at 22°C during 2h produced cDNA ligation. Finally, 30 µl of amplification mix (29 µl REPLI‐g Midi Reaction Buffer, 1 µl REPLI‐ g Midi DNA Polymerase) was added to the ligation reaction, and then incubated at 30°C during 8 h for a high‐yield reaction (up to 40 µg cDNA), at 95°C for 5 min to inactivate all enzymes and 10°C hold. Every sample was run in an 2%‐agarose gel to check for the presence of amplified cDNA before performing the real‐time qRT‐PCR, and measured by NanoDrop 2000, which revealed concentrations between 1700‐1900
44 Materials and Methods ng/µl and good quality ratios. Samples were dilute ½ in TE 2X (20 mM Tris‐HCl pH 8, 2 mM EDTA) and stored at ‐20°C until use.
4.3. Real‐time qRT‐PCR and quantification The real‐time qRT‐PCR was carried out in a 384‐well plate using reduced volume per reaction: 2.5 µl TaqMan Fast Advanced Master Mix (4444557, Applied Biosystems), 0.25 µl TaqMan Gene Expression Assay (Table 3) and 2.25 µl cDNA of a 1/250 dilution from the original concentration. Triplicate reactions were run using the 7900HT Fast Real‐Time PCR System (Applied Biosystems) for 45 cycles: 95°C 15 sec, 60°C 1 min. Average Ct (cycle threshold) values for Pvalb from triplicates were standardized with a housekeeping gene (beta‐actin) for each sample (ΔCt) and both sorted fractions were compensated by substraction of the mean of tdTomato‐negative fractions values (ΔΔCt). Data were finally represented using the mean 2‐ΔΔCt values normalized against the mean 2‐ΔΔCt value from all tdTomato‐negative fractions, separately for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice, using GraphPad Prism 6 software.
Probe Gene Exon boundary Reference 4453320, Applied Mm00443100_m1 Parvalbumin Probe spans exons 1‐2 Biosystems 4334482, Applied Mm02619580_g1 Actin, beta Probe maps within exon 3 Biosystems Table 3. TaqMan Gene Expression Assay probes. Description of Taqman probes used for real‐time qRT‐PCR experiments.
5. Survival analysis
In our mouse colony, we maintained a reasonable number of mice for both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes until 8 months, which were used for distinct experiments during this work and therefore making it possible to elaborate a survival curve from 0 to 8 months. 29 males and 29 females were used for that purpose. Data were analyzed by GraphPad Prism 6 using a Kaplan‐Meier survival curve and the statistics provided by this test (Log‐rank Mantel‐Cox test).
6. Body weight curve
A total of 5 males and 5 females for PVcre:Ai27D:Dnajc5flox/+ mice, and 5 males and 4 females for PVcre:Ai27D:Dnajc5flox/‐ were reserved to perform a body weight study. Because the phenotypical differences between control and mutant mice became clearly visible around P30, this assay was done by weighting animals once a week from P30 to P79 (around 3 months), and once a month from P79 to P240 (8 months). Analysis was carried out with GraphPad Prism 6, where each time point was compared between control and mutant mice, separating data by sex due to physiological differences in body weight in males and females.
45
Materials and Methods
7. Open field
For the analysis of locomotor activity, we utilized the open field test. Before starting, it was needed to set up several parameters on Biobserve software to optimize motion identification (http://www.biobserve.com/behavioralresearch/products/viewer/): (1) an appropriate mouse body detection (Figure 11), (2) calibration of cage dimensions (conversion: pixels to cm, square boxes), (3) configuration of the working area (the whole cage), (4) establishment of zones (in this case, only the working area) and (5) selection of a degree of sensitivity to avoid background contamination. All recordings were re‐analyzed on a lab computer by Viewer3 to define the same parameters and make track corrections in case of noisy signals. We carried out this experiment on 44 mice (23 females: n=9 PVcre:Ai27D:Dnajc5flox/+ and n=14 PVcre:Ai27D:Dnajc5flox/‐; and 21 males: n=11 PVcre:Ai27D:Dnajc5flox/+ and n=10 PVcre:Ai27D:Dnajc5flox/‐) during 30 min in dark. UV cameras located on top of four cages recorded four mice per experiment. Automatically, Biobserve software generated recording data and a summary of behavioral parameters. Among them, we selected velocity (cm/s), activity (percentage (%) of time the mouse overcame the activity threshold of 1 cm/s), track length (cm), wall distance (cm), and number of headbobs, headstretches (Figure 11), tailmoves and ambulations (spontaneous short‐term accelerations that overcame the ambulation threshold of 60 cm/s). All parameters were analyzed using GraphPad Prism 6, separately for males and females, and all together.
Figure 11. Mouse body identification and types of head movements analyzed during the open field experiments. Drawings extracted from the Biobserve open field documentation showing mouse identification (head, body, tail) (left) and natural head movements (right) that were analyzed during video recording.
8. Immunoblotting
Cortex and/or liver samples were extracted from (1) Dnajc5 WT and KO littermate mice at P15 and P30, (2) PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ littermate mice at 2‐month (adult mice) and 8‐ month (old mice) postnatal age, and (3) Dnajc5flox/flox and UBC‐Cre‐ERT2:Dnajc5flox/flox littermate mice at P100.
Mice were anesthetized with 2% tribromoethanol, and fresh tissues, flash‐frozen in liquid nitrogen, were kept at ‐80°C until processing. Tissue lysis and homogenization were carried out in RIPA buffer (20 mM Tris‐HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% IGEPAL, 0.1% SDS) supplemented with protease and
46
Materials and Methods phosphatase inhibitors (05892791001 and 04906837001, Roche). Samples were then maintained in continuous agitation for 2 hours on ice and vortexed every 15‐30 minutes. After that, samples were centrifuged at maximum speed (14000 rpm) for 15 minutes at 4°C, and supernatants were collected as lysates. Protein concentration of lysates was estimated using Pierce BCA Protein Assay Kit (23227, Thermo Scientific), following manufacturer’s instructions. Lysates were then processed with Laemmli buffer 4X supplemented with 0.4 M beta‐mercaptoethanol, reaching a final concentration of Laemmli at 1X (50 mM Tris‐HCl pH 6.8, 10% glycerol, 2% SDS, 0.0067% bromophenol blue, 0.1 M beta‐mercaptoethanol), and heated at 95°C for 10‐15 minutes in a thermomixer without agitation. Percentage of acrylamide for gels were chosen depending on desired protein molecular weights. We used 10%‐acrylamide gels for separation of medium‐molecular weight proteins among 25‐75 KDa and 14% for separation of small proteins among 10‐20 KDa. Equivalent amounts of protein (40 µg) and volume (30 µl) were analyzed by SDS‐PAGE using Running Buffer 1X (25.5 mM glycine, 2.5 mM Tris, 0.1% SDS, pH 8.4 ± 0.2). Samples were run first at 50 mV for approximately 30 min and finally at 100 mV for 1 hour and a half more. After that, samples in acrylamide gels were transferred to PVDF membranes (1620177, Biorad) into cold Transfer Buffer 1X (25.5 mM glycine, 2.5 mM Tris, 20% methanol, with 0.1% SDS in case of medium‐molecular weight proteins, or without SDS if small proteins) at 200 mA for two hours if medium‐proteins or two hours and a half if small proteins. Membranes were then washed three times in Tris‐Buffered Saline and Tween 20 (TBS‐T) 1X (13.7 mM NaCl, 10 mM Tris, 0.1% Tween 20, pH 7.5) and blocked with 3% BSA or 3% milk in TBS‐T during 1 hour at room temperature while mildly agitating. Then, membranes were immunoblotted with primary antibodies (Table 4) in the same blocking solution O/N at 4°C while mildly agitating. Posteriorly, they were washed three times with TBS‐T and incubated during 1 hour at room temperature while mildly agitating with appropriate horseradish peroxidase (HRP)‐conjugated secondary antibodies in the same blocking solution: goat anti‐rabbit IgG (1:10000; 111‐035‐144, Jackson ImmunoResearch) and goat anti‐mouse IgG (1:10000; 115‐035‐166, Jackson ImmunoResearch). Mouse anti‐β‐actin was used as loading control. ClarityTM Western ECL Substrate (170‐5060, BIORAD) was used for chemiluminescent western blot detection. Images were taken using the Chemidoc Touch Imaging System (ThermoFisher Scientific).
9. Quantification of protein levels
After immunoblotting, protein quantification was performed. Images were acquired using the same exposure times and bands were analyzed using ImageJ (https://imagej.nih.gov/ij/download.html), an open source image processing program. Briefly, bands were selected with a rectangular box (Ctrl+1) and lanes were plotted (Ctrl+3). We used straight lines to separate between band areas and from the background, and we quantified each band area using the wand (tracing) tool. Data were collected and
47
Materials and Methods compensated with the housekeeping protein (β‐actin) expression level for each well. Data were represented in bar diagrams using normalized data versus control mice in GraphPad Prism 6 software.
10. Perfusion and sectioning using a cryostat
Anesthesia used was 2% tribromoethanol on (1) Dnajc5 WT and KO littermate mice at P30 and (2) PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ littermate mice at 2‐month and 8‐month postnatal age. Mice were transcardially perfused with 1X Phosphate Buffered Saline (PBS) (137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4) to eliminate blood from tissues, and then with 4% paraformaldehyde
(PFA) in Phosphate Buffer 1X (PB) (19.98 mM Na2HPO4, 79.59 mM KH2PO4, pH 7.4) to fix tissues. Brains were removed and post‐fixed O/N in 4% PFA at 4°C. Brains were then cryoprotected in 30% sucrose in PB (with 0.01% sodium azide) and a Leica CM 1950 cryostat was utilized for serial sectioning using coronal planes; one out of every sixth brain slices (40 µm thickness) was analyzed for immunohistochemistry and immunofluorescence experiments. Slices were stored at ‐20°C (50% glycerol in PBS 1X) until use.
11. Brigth field immunohistochemistry
Coronal slices of the motor cortex from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice were washed with PBS 1X and incubated during 20 minutes at room temperature with 3% H2O2 and 10% methanol in PBS to block the endogenous peroxidase. Slices were then washed with PBS and blocked in a solution containing 3% FBS and 0.3% Triton X‐100 in PBS for 1 hour at room temperature, while mildly agitating. Sections were incubated O/N with primary antibodies (Table 4) at 4°C using the same blocking solution. Several washes in PBS were performed before incubation using appropriate secondary biotinylated antibodies (1:1000 dilution in 0.3 % Triton/PBS) for 1 hour at room temperature while mildly agitating: donkey anti‐rabbit IgG (711‐065‐152, Jackson ImmunoResearch) and donkey anti‐mouse IgG (711‐065‐151, Jackson ImmunoResearch). Slices were washed several times using PBS, and then incubated in an avidin‐biotin‐peroxidase solution (ABC kit, VECTASTAIN, Vector Laboratories) for 1 hour while mildly agitating at room temperature. After washing, a developing solution containing 0.02% diaminobenzidine as chromogen, 0.04% nickel ammonium sulfate and 0.01% H2O2 in PB was added to reveal the immunoperoxidase labeling, one maker each time. Finally, PB was added to stop the reaction when appropriate, followed by several washes in PB. Sections were mounted on previously gelatinized glass slides and let dry. Slides were then passed through an alcohol battery to dehydrate all sections (30 sec, water; 30 sec, ethanol 70%; 30 sec, ethanol 80%; 30 sec, ethanol 90%; 30 sec, ethanol 100% twice) and through xylene to clear the samples (30 sec, xylene; 10 min, xylene). Slides were coverslipped with DPX (Sigma), and let dry. Images were taken using a NewCAST BX ‐61 microscope (Olympus) and a 4X objective lens (Olympus).
48 Materials and Methods
12. Quantification of PV somata number
Images taken from immunohistochemistry experiments were used for PV somata number quantification. A total of three XY planes with same dimensions were cropped from each bright field image selecting regions that contained only layer II/III neurons from motor cortex. PV somata was manually quantified by the “multi‐point” tool from ImageJ, and mean values (n=27 fields per genotype, from 3 PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice) were plotted and analyzed using GraphPad Prism 6. Number of PV cells were represented per mm2.
13. Immunofluorescence in brain slice sections
Approximately fifteen serial slices were selected from each mouse to analyze the whole motor cortex (AP: since 2.58 mm to 1.42 mm, from bregma) through immunofluorescence techniques. After several washes with PBS, sections were incubated in blocking solution (3% FBS and 0.3% Triton X‐100 in PBS) for 1h at room temperature while agitating. Primary antibodies (Table 4) were prepared in blocking solution and slices incubated O/N at 4°C. Sections were rinsed three times with PBS, and incubated in dark with appropriate fluorescently labelled secondary antibodies in 0.3% Triton X‐100 in PBS for 2h at room temperature, while agitating: Rhodamine Red donkey anti‐chicken IgG (1:300, 703‐295‐155, Jackson ImmunoResearch), Alexa Fluor 488 donkey anti‐rabbit IgG (1:500, 711‐545‐152, Jackson ImmunoResearch), Cy3 donkey anti‐rabbit IgG (1:500, 711‐165‐152, Jackson ImmunoResearch), Alexa Fluor 488 donkey anti‐mouse IgG (1:500, 715‐545‐151, Jackson ImmunoResearch) and/or Alexa Fluor 647 goat anti‐rabbit IgG (1:250, 111‐605‐144, Jackson ImmunoReseach). Slices were then washed, mounted and coverslipped with Dako Fluorescence Mounting Medium (S3023, Dako), and let dry at 4°C in dark for 1‐2 days. Images were taken with a Nikon A1R+ confocal microscope.
14. Colocalization analysis
Confocal images of motor cortical regions were taken using a 20X glycerol immersion objective lens (Nikon) to study the degree of colocalization between PV+ and tdTomato+ neurons. The same laser intensity was set up for imaging acquisition in all mice. ImageJ was used for quantification. Briefly, regions having the same XY dimensions were chosen to identify motor cortical neurons of layer II/III, for every image. Then, from a total of 15‐16 serial slices per mouse, we selected two Z‐stacks of 1 µm for quantification, obtaining around 30‐32 fields per mouse. At the end, n≈90 fields per genotype (from a total of 3 PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/+ mice) were separately and manually quantified for PV‐positive (PV+) somata (green channel) and tdTomato‐positive (tdTomato+) perisomal labelling (red channel), using the “multi‐point” tool. Points detected in both green and red channels were considered as colocalization values. By this way, we obtained a total average number of PV+ and tdTomato+ cells, and therefore we were able to calculate the percentage of PV‐tdTomato colocalization, and the percentages
49
Materials and Methods of PV+ cells and tdTomato+ cells that did not colocalize. We used RStudio 1.1.442 and R free software version 3.4.4 to plot the data with a Venn diagram using the function draw.pairwise.venn from the package VennDiagram.
15. Quantification of synaptic puncta
Confocal images of motor cortical regions were acquired using a 60X oil immersion objective lens (Nikon) to quantify synaptic puncta labelled by Syt2 and PV. Laser intensity was always the same for each age (2 months or 8 months) and Z‐plane distance was established at 1 µm. Two planes with the same XY dimensions were acquired from 16‐30 motor cortex slices (40 µm‐thickness) per mouse and separately quantified for PV+ (red channel) and Syt2+ (green channel) puncta (PV: n=82 or 67 images from PVcre:Ai27D:Dnajc5flox/+ and n=79 or 72 images from PVcre:Ai27D:Dnajc5flox/‐ animals at 2 and 8 months, respectively; Syt2: n=76 or 67 images from PVcre:Ai27D:Dnajc5flox/+ and n=79 or 72 images from PVcre:Ai27D:Dnajc5flox/‐ animals at 2 and 8 months, respectively; from a total of 3 mice per genotype). For this quantification, we used ImageJ to create two different macros for red and green channels. Both macros contained the following steps and were run in every image: (1) selection of a total of 1‐30 stacks in Z‐plane, (2) images were converted to 8‐bits, (3) a threshold level was adjusted for each channel for binary conversion and (4) puncta density was calculated with “Analyze particles” module using the thresholded image of every Z‐stack. We excluded PV somas setting the maximum puncta size to 200 pixels, and the background contamination setting the minimum puncta size to 10 pixels. For Syt2+ puncta, no maximum puncta size was stablished. Collected data were analyzed using GraphPad Prism 6. Average puncta density was represented per 100 µm2.
16. Quantification of PV somata size
Using the red channel (PV) of the confocal 60X images used for synaptic puncta quantification, we measured PV somata size (n=104 PV cells from PVcre:Ai27D:Dnajc5flox/+ and n=181 PV cells from PVcre:Ai27D:Dnajc5flox/‐ animals at 2 months; n=152 PV cells from PVcre:Ai27D:Dnajc5flox/+ and n=142 PV cells from PVcre:Ai27D:Dnajc5flox/‐ at 8 months, from a total of 3 mice per genotype). Two Z‐stacks of 1 µm from 4‐5 images per mouse were quantified with Bitplane Imaris. Minimum and maximum size detection parameters were adjusted separately for 2 months and 8 months, to recognize only the neuron somata labelling (not dendritic) and ensure the separation of somas in case they appeared side by side. PV somata area data were determined in µm2 and analyzed with GraphPad Prism 6.
50
Materials and Methods
Host Protein Primary antibody Method Concentration Reference species 1:2000 Enzo Cat. No. ADI‐VAP‐ CSPα anti‐CSPα (C‐terminus) Rabbit (BSA) SV003‐E (Fernández‐Chacón et anti‐GST‐fusion protein 1:1000 CSPα Rabbit al., 2004; Tobaben et of CSPα (full‐length) (BSA) al., 2001) anti‐synaptotagmin‐1 1:2000 Synaptic Systems Cat. Synaptotagmin‐1 cytoplasmic tail, clone Mouse (BSA) No. 105 011 41.1 1:750 Developmental Studies Synaptotagmin‐2 znp‐1 Mouse (Milk) Hybridoma Bank WB anti‐Uncoating ATPase 1:1500 Synaptic Systems Cat. Hsc70 Mouse (hsc70), clone 3C5 (BSA) No. 149 011 anti‐SNAP25, clone 1:1000 SNAP25 Mouse Biolegend SMI 81 (BSA) anti‐parvalbumin, PV 1:750 Parvalbumin Rabbit Swant 27 (Milk) anti‐synaptobrevin‐2 / 1:2000 Synaptic Systems, Cat. Synaptobrevin‐2 Mouse VAMP 2, clone 69.1 (Milk) No. 104 211 anti‐β‐actin, clone AC‐ 1:5000 β‐actin Mouse Sigma‐Aldrich, A2228 74 (BSA or milk) 1:1000 Enzo Cat. No. ADI‐VAP‐ CSPα anti‐CSPα Rabbit SV003‐E anti‐parvalbumin, PV 1:1000 Parvalbumin Rabbit Swant, PV 27 27 IHC 1:500 Developmental Studies Synaptotagmin‐2 znp‐1 Mouse Hybridoma Bank anti‐SNAP25, clone 1:5000 SNAP25 Mouse Biolegend SMI 81 anti‐parvalbumin, PV Parvalbumin Rabbit 1:1000 Swant 27 Encor Biotechonology tdTomato CPCA‐mCherry Chicken 1:1000 IF Inc. Developmental Studies Synaptotagmin‐2 znp‐1 Mouse 1:500 Hybridoma Bank
Table 4. Primary antibodies. Summary of primary antibodies employed during this work in western blot (WB), immunohistochemistry (IHC) and immunofluorescence (IF).
17. Electrophysiology
Electrophysiological recordings were performed at two different postnatal ages, 2 months and 8 months. For each time lapse, PVcre:Ai27D:Dnajc5flox/+ (n=3) and PVcre:Ai27D:Dnajc5flox/‐ (n=3, with the exception of n=4 for intrinsic properties at 2 months) animals were used. Intrinsic electrophysiological properties and excitatory postsynaptic potentials (EPSPs) of parvalbumin interneurons, and miniature inhibitory postsynaptic currents (mIPSCs) on glutamatergic cells were analyzed at both ages to evaluate the degree of inhibition in the motor cortex of our mouse model.
51
Materials and Methods
17.1. Preparation of brain slices Mice were anesthetized with 2% tribromoethanol, decapitated, and brains were removed and placed into ice‐cold cutting solution: 222 mM sucrose, 11 mM D‐glucose, 3 mM KCl, 1 mM NaH2PO4, 26 mM NaHCO3, 0.5 mM CaCl2, 7 mM MgCl2 (Osmolarity: 300‐310 mOsm/l); the pH was 7.4 once the solution was aerated with 95% O2, 5% CO2. 250 µm coronal slices including motor cortex (AP: since 2.58 mm to 1.42 mm, from bregma) were cut with a Leica VT1200S vibratome. Slices were allowed to recover at 37 °C in artificial cerebrospinal fluid (ACSF): 124 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 10 mM
D‐glucose, 2.5 mM CaCl2, 2 MgCl2, and 4 mM sucrose (Osmolarity: 300‐310 mOsm/l); the pH was 7.4 once the solution was bubbled with carbogen (95% O2, 5% CO2). Afterwards, slices were kept at room temperature in ACSF with continuous carbogen bubbling until recording.
17.2. Electrophysiological recordings of parvalbumin interneurons Brain slices were located into a recording chamber with a continuous perfusion of bubbled ACSF at 33 ± 1°C (2‐3 ml/min). Layer II/III motor cortical parvalbumin interneurons were visualized based on its tdTomato expression pattern detected using tetramethylrhodamine‐isothiocyanate (TRITC) filters incorporated into a Nikon Eclipse FN1 microscope, with a 40X water immersion objective lens (Nikon) and a USB 2.0 monochrome camera (DMK 31BU03.H, TheImagingSource). For whole‐cell patch‐clamp experiments, a double patch‐clamp EPC10 plus amplifier (HEKA) was used. Recordings were low‐pass filtered at 2.9 kHz, digitized at 20 kHz and acquired by PatchMaster software (HEKA). For current‐clamp protocols, pipettes (2.5‐3.5 MΩ) contained 120 mM K‐gluconate, 10 mM KCl, 10 mM phosphocreatine disodium salt, 2 mM Mg2+‐ATP, 0.3 mM Na+‐GTP, 0.1 mM EGTA, 10 mM HEPES; pH 7.2 adjusted with KOH; Osmolarity: 280‐290 mOsm/l adjusted with sucrose. Series resistance was below 20 MΩ and fully compensated in current clamp mode.
17.2.1. Intrinsic properties Excitability was calculated counting the number of action potentials (APs) during depolarizing injection protocols using 20 pA/500 ms square pulse steps (from 20 pA to 600pA). To analyze all AP parameters, we injected the minimum current intensity to evoke one AP with 50% likelihood (rheobase). Standard analysis of AP amplitude, rise time (20‐80%) and half‐width were done using Axograph X software (Molecular Devices). Resting membrane potential (RMP) was measured as the membrane potential average for 30 seconds recording in absence of injected current. Time constant was calculated through single exponential fitting of the membrane potential change after a 50 pA/500ms pulse of negative (hyperpolarizing) current. The AP threshold was calculated as the first sample where the rising membrane potential exceeded a slope of 20 mV/ms. Input resistance was calculated as the slope of the linear fit of the voltage–current plot generated from a family of positive and negative current injections (−50 to +50 pA at 10 pA intervals, 500 ms in duration). The voltage value was acquired during the last 100 ms of the
52 Materials and Methods pulse because of the presence of inward rectification or sag. Capacitance was calculated as the quotient of the time constant and the input resistance. Afterhyperpolarization (AHP) amplitude was measured as the maximum membrane potential drop below the normal resting potential. Normalization was done (each one to itself) to compare kinetics in both genotypes (0‐1). AHP decay was measured using a double exponential fitting. Data were analyzed using GraphPad Prism 6.
17.2.2. Excitatory postsynaptic potentials (EPSPs) For each PV cell recorded, we estimated the amplitude, frequency and kinetic of the excitatory postsynaptic potentials measured for 30 seconds recording in absence of injected current. The isolation and analysis of EPSPs was performed using Axograph X, based on the detection method of spontaneous synaptic events designed by J.D Clements and J.M Bekkers (Clements and Bekkers, 1997). Data were analyzed using GraphPad Prism 6.
17.3. Electrophysiological recordings of pyramidal neurons Whole‐cell recordings of layer II/III cortical pyramidal neurons were made in voltage clamp mode under visual guidance using differential interference contrast (DIC) optics. Excitatory neurons were identified based on typical pyramidal shaped soma and by thick tufted apical dendrites projecting towards upper cortical layers. Patch‐pipettes (2.0‐3.0 MΩ) for miniature inhibitory postsynaptic currents (mIPSCs) recording contained 140 mM CsCl, 2 mM MgCl2, 0.05 mM EGTA, and 10 mM HEPES; adjusted to pH 7.2 with CsOH (Osmolarity: 280–290 mOsm/l); mIPSCs were recorded in the presence of tetrodotoxin (1 µM) and kynurenic acid (2.5 mM) to block sodium channels and ionotropic glutamate receptors, respectively.
Cells were held in voltage‐clamp mode at a holding potential (Vhold) of ‐70 mV and resistance was compensated by 70%. Recordings were discontinued whether series resistance increased by above 50% or exceeded 20 MΩ. The isolation and analysis of mIPSCs was performed using Axograph X, based on the detection method of spontaneous synaptic events designed by J.D Clements and J.M Bekkers (Clements and Bekkers, 1997). Briefly, recordings were first digitally filtered at 1 kHz and a synaptic template was generated averaging hand‐picked synaptic events. A detection criterion was established to optimize the signal/noise ratio for event detection: 2.5 X noise standard deviation. Amplitude, frequency, rise‐time (20‐ 80%) and half‐width were calculated based on averaged mIPSCs detected for each cell. mIPSCs decay was calculated through double exponential fitting and with normalized values (each one to itself) to compare kinetics (0‐1). Data were analyzed using GraphPad Prism 6.
18. Neuron dissociation
Single‐cell suspensions were obtained as previously described in Marques et al., 2016 with some modifications for P60 animals as performed in Harris et al., 2018. We used the NMDG‐HEPES recovery solution (Tanaka et al., 2008) in all steps of the protocol, instead of the commonly known artificial
53
Materials and Methods cerebrospinal fluid (ACSF), to ensure an improved recovery of aged cells. The NMDG‐HEPES solution contained 93 mM N‐Methyl‐D‐Glucamine (NMDG), 2.5 mM KCl, 1.2 mM NaH2PO4, 30 mM NaHCO3, 20 mM HEPES, 25 mM glucose, 5 mM sodium ascorbate, 2 mM thiourea, 3 mM sodium pyruvate, 10 mM
MgSO4*7H2O and 0.5 mM CaCl2*2H2O; and it was adjusted to pH 7.4 with 10 N HCl. This solution was prepared each experimental day. It was then separated into two bottles: one to be cool down at ‐80°C for approximately 30 min, and another warmed at 37°C until use. Mice were anesthetized with 2% tribromoethanol (in case of experiments at IBiS) or with a mixture of ketamine/xylazine (80 mg/kg; 10 mg/kg) (in case of experiments at KI), and transcardially perfused with 20 ml of oxygenated (95% O2, 5%
CO2) NMDG‐HEPES solution cooled down at ‐80°C. Brains were removed and sectioned into 300 µm slices using a Vibratome (Leica VT1200S) filled with the oxygenated, cold NMDG‐HEPES solution. During sectioning, slices were placed and maintained in an incubation‐chamber containing the pre‐warmed NMDG‐HEPES solution, which progressively reached room temperature. Whole cortex was dissected from every slice and incubated with papain at 37°C during 30‐45 minutes for enzymatic digestion, using the Papain Dissociation System (PDS, LK003150, Worthington); NMDG‐HEPES solution was employed for reagents resuspension instead of the provided EBSS (Earle’s Balanced Salt Solution). Mechanically trituration with 3‐4 fire‐polished Pasteur pipettes with decreasing diameter was performed as quickly as possible, taking care of avoiding bubble generation. No albumin density gradient was carried out after proving that eluding it, cell viability and PV cell number at FACS were improved, probably due to time and cell stress reduction. Then, an only centrifugation step (5 min at 200g) after filling a 15 ml‐falcon with oxygenated NMDG‐HEPES at room temperature was performed to pellet cells. Supernatant was carefully discarded, and cells resuspended into cold FACS buffer (NMDG‐HEPES with 0.2% BSA). Finally, samples were filtered before FACS with a pre‐wet CellTrics 30 µm filter (04‐0042‐2316, Sysmex‐Partec) in case of continuing with RNA sequencing, or with Cup filcons 50 µm filter (340632, BD) in case of gDNA/mRNA PCR validations and real‐time qRT‐PCR. Cells were maintained in ice until FACS.
NMDG‐HEPES solution was employed in all experiments of neuron dissociation except for the first scRNA‐seq WG17002 chip, where we used ACSF (80 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 26 mM
NaHCO3, 75 mM sacarose, 10 mM D‐glucose, 1 mM CaCl2, 2 MgCl2). Besides, the albumin density gradient step included in the Papain Dissociation System was also only performed in the WG17002 chip. Because the percentage of cells obtained at FACS and the cDNA amount were poor in this first chip, we switched to NMDG‐HEPES solution and eluded the density gradient for the second chip. Percentage of cells at FACS and cDNA quantity were improved after these modifications and, therefore, we established the above mentioned protocol for the rest of experiments.
54
Materials and Methods
19. Single‐cell RNA sequencing protocol
The single‐cell RNA sequencing (scRNA‐seq) strategy followed in this work is based on the single‐ cell tagged reverse transcription (STRT) sequencing method, a quantitative technique characterized by three main advantages: (1) the use of unique molecular identifiers (UMIs) for absolute quantification, (2) the use of single‐read instead ofd paire ‐end sequencing, and (3) the generation of 5’‐end reads that provide information about transcription starting sites (Islam et al., 2014). This method, which was originally performed on Fluidigm C1 96‐well platform, has been improved with the introduction of a second index (2i) previous to sequencing and with the use of a WaferGen 9600 micro‐well chip; all these improvements named this new method as STRT‐seq‐2i. Samples can be dispensed inside the chip either from a single‐cell suspension, using a dispenser (up to 8 samples), or by FACS into the wells (as many samples as desired). In our case, we used FACS directly into wells containing lysis buffer, discarding the possibility of performing a posterior imaging step per well that is possible after using the WaferGen dispenser (imaging can give us information about doublets and/or cell fluorescence per well). Up to 2400 wells per chip can dbe fille by FACS due to the current set of 32 unique index primers. During library preparation, each sample is tagged with a unique combination of indexes (UMI, and first and second indexes) before they are pooled together for sequencing. This method carries out a 45‐50 bp single‐read sequencing and the expression data is obtained by UMI counts for each mRNA (http://escg.se/methods/strt‐seq‐2i/; Hochgerner et al., 2017). Hence, sequencing with full‐length read coverage is not possible as detected in other applications such as Smart‐seq2 (Picelli et al., 2013).
The WaferGen 9600 micro‐well chip (Dual Index Chip, P/N:310‐000030, WaferGen Biosystems; now available at Takara as ICELL8 chip) is a nanoliter‐volume microwell aluminum platform composed by 9600 wells that are divided in 96 subarrays (12 columns x 8 rows) with 100 wells each (10 columns x 10 rows), which allows the sequencing of several samples using one chip. Each well permits a maximum well volume of 1 microliter, facilitating multistep protocols as the STRT‐seq‐2i (Figure 12).
Detailed steps for STRT‐seq‐2i method were first published in Hochgerner et al., 2017 and are described below step‐by‐step with no important modifications from the original paper. Cell‐sorting was performed by the FACS facility in SciLifeLab and all sequencing steps were carried out by the ESCG (Eukaryotic Single Cell Genomics) facility also in SciLifeLab at the Karolinska Institute (KI).
55 Materials and Methods
Figure 12. Scheme of single‐cell RNA sequencing using a WaferGen plate. A. Picture showing summarized protocol for scRNA‐seq. The whole cortex is dissociated and neurons are sorted by FACS into a WaferGen plate for downstream sequencing and analysis. B. Detailed structure of the aluminum WaferGen platform, a nanoliter‐volume microwell plate composed by 9600 wells and divided into 96 subarrays (12 columns x 8 rows). Each subarray contains 100 wells (10 columns x 10 rows).
19.1. Steps 1‐2. FACS into a WaferGen 9600 microwell plate Four experiments were performed in four distinct days using PVcre:Ai27D:Dnajc5WT (n=3), PVcre:Ai27D:Dnajc5flox/+ (n=4) and PVcre:Ai27D:Dnajc5flox/‐ (n=5) mice at 2 months. Three mice were used per day, each one with a different genotype (when possible) and sorted into a WaferGen plate in a different order each day, to avoid potential batch effect. Order, number and genotype for each sorted mouse per day is specified below:
Day 1 (2017‐05‐03): 449 (WT), 451 (Flox‐), 465 (Flox+).
Day 2 (2017‐05‐08): 458 (Flox‐), 466 (Flox+), 485 (WT).
Day 3 (2017‐05‐11): 486 (Flox+), 457 (Flox‐), 464 (Flox‐).
Day 4 (2017‐05‐12): 493 (Flox‐), 491 (WT), 488 (Flox+).
These samples were run into four WaferGen plates with the following ID#: WG17002 (day 1), WG17003 (day 2), WG17004 (day 3) and WG17005 (day 4).
56 Materials and Methods
In order to improve the yield per microwell array (WaferGen), we used FACS to sort cells directly into the wells using optimal sorting parameters. A nozzle of 140 µm and a 585/29 filter for a 561 nm laser were used for accurate tdTomato detection from cortical dissociated cells in a BD Influx cell sorter. The gating strategy was performed by (1) selection of cells with forward (FSC‐H) and side (SSC‐H) scatters, (2) selection of singlets based on FSC‐H and FSC‐W, (3) re‐selection of singlets by FSC‐H and FSC‐A, and (4) selection of tdTomato+ (585/29[561nm]) cells. As previously described in Hochgerner et al., 2017, prior to FACS and only into the wells to be used, the plate was dispensed with 50 nl of lysis buffer (500 nM STRT‐ P1‐T31 (Table 5), 4.5 nM dNTP, 2% Triton‐X‐100, 20 mM DTT, 1.5 U/μl TaKaRa RNase Inhibitor) and kept on dry ice until sorting.
Although the plate contained 9600 wells, only a quarter of the full chip (exactly 2392 wells) could be used at a time due to software memory limitations during setting up the sorter, and because it was essential to sort in a reasonable time to maintain cells alive. Anycase, full plate sequencing is indeed not possible with the current set of 32 unique index primers. Furthermore, due to nanodispenser physical limitations by which some wells are not accessible by the tips during automated dispensing, at the end only 25 wells became usable for each 10x10 quadrant, except for the H12 quadrant, which followed a different sorting pattern (only 17 usable wells). To set up the filling layout (Figure 13), we took advantage of the symmetric design of the plate; two quarter layouts were firstly created, and then the plate was turned 180° to generate the remaining two quarters, thus designing the whole WaferGen plate filling layout. Before any experiment, the layout needed to be aligned using an empty plate covered with a thin film, and Accudrop fluorescent beads were used to check for drop sorting positions. The time we took to fill the entire plate was approximately 1.5 hours, which was a reasonable time, because also sample changing was carried out to avoid the batch effect (approximately 800 cells per mouse genotype were sorted each day; in total, 2392 wells). The plate was then frozen on dry ice and stored at ‐80°C for later processing (Figure 14: steps 1, 2).
19.2. Steps 2‐3. Lysis and reverse transcription The plate kept at ‐80°C was first thawed until it reached room temperature. Next, 85 nl of reverse transcription (RT) mix (2.1X SuperScript II First‐Strand Buffer, 12.6 mM MgCl2, 1.79 M betaine, 14.7 U/μl SuperScript II, 1.58 U/μl TaKaRa RNase Inhibitor, 10.5 μM P1B‐UMI‐RNA‐TSO (Table 5)) were dispensed and reverse transcription carried out at 42°C for 90 min. During this step, the STRT‐P1‐T31 primer containing 31 thymine bases (supplemented in the lysis buffer) annealed with the polyA tail of mRNA transcripts. RT was then initiated by the SuperScript II reverse transcriptase (derived from Moloney Murine Leukemia Virus) and, at the end of retro‐transcription, it caused the incorporation of 3‐6 non‐templated nucleotides at the 3’ end of cDNA, with preference for cytosine (Zajac et al., 2013). This enzyme is capable to generate cDNA up to 12.3 kb (Invitrogen), although the typical average cDNA size obtained was
57
Materials and Methods
58
Materials and Methods around 2‐3Kb (ESCG data). After that, the P1B‐UMI‐RNA‐TSO primer, containing 3 guanines at the 3’ end, annealed to the 3 incorporated cytosines at the cDNA, and made a template switching used to introduce a distinct barcode (composed of 6‐bp random sequence (UMI)) to every synthetized cDNA molecule. In this way, the 6‐bp UMI is capable of distinguishing up to 4096 different mRNA molecules, which is sufficient for a single mammalian cell (Islam et al., 2014). The plate was centrifuged for 1 minute at maximum speed (>2000g) after any dispense and incubation step to guarantee accurate mixing of reagents. MicroSeal A film (BioRad) was employed to seal the array when necessary (Figure 14: steps 2, 3).
19.3. Steps 4‐5. Indexed PCR and extraction 25 index primers (DI‐P1A‐idx[1‐25]‐P1B) (Table 5) were added following the reverse transcription to carry out the first PCR. In this way, each well from every 10x10 quadrant of the chip got a unique barcode (BC1). These indexes were later employed to identify to which cell/well each transcript belongs. Each primer was dispensed in 100 nl into a different well (from 1 to 25 wells within each 96 quadrant), using washes with 2% bleach between each set of primers, and reaching a final primer concentration of 200 nM in the PCR reaction. Finally, 570 nl of PCR mix were added to each sample (final concentration of 1X KAPA HiFi Ready Mix, supplemented with 0.2 mM dNTP and 100 nM DI‐PCR‐P1A), and PCR reaction was run with the following program: 95°C 3 min; 5 cycles: 98°C 30 sec, 67°C 1 min, 72°C 6 min; 15 cycles: 98°C 30 sec, 68°C 30 sec, 72°C 6 min; 72°C 5 min; 10°C hold. Each pair of primers (the same common DI‐PCR‐P1A primer with each one of the DI‐P1A‐idx[1‐25]‐P1B primers) amplified all UMI‐tagged cDNA molecules from a single cell/well. After that, a cleaning step with AMPure XP beads at 0.7:1 ratio was performed to purify cDNA from primers, and cDNA was eluted in EB buffer (Qiagen). Finally, an assembly between the WaferGen plate and a clean 96‐well plate was mounted and centrifuged during 5 min at maximum speed (>3000g) to transfer the WaferGen plate content into the 96‐well plate. At the end, each well from the 96‐well plate contained pooled indexed amplified cDNA from 25 wells belonging to an entire 10x10 quadrant from the original chip. This plate was analyzed by Bioanalyzer (High Sensitivity DNA Assay, Agilent) to determine cDNA quantity and quality, and finally sealed and kept at ‐20°C (Figure 14: steps 4, 5).
______
Figure 13. Graphic representation of well position patterns for cell sorting into a WaferGen platform. Picture of a matrix simulating the WaferGen plate composed by 9600 microwells. It is formed by 96 quadrants (12 column quadrants x 8 row quadrants), and each quadrant is in turn constituted by 100 wells (10 columns x 10 rows). Only 25 positions can be used due to sequencing barcode limitation, leading to a total of 2392 useable wells (orange positions). Note the different sorting template in H12 quadrant (17 suitable wells). Purple wells represent positions where sample changing took place at FACS; the right well eis th last position sorted from the previous sample and the left well corresponds to the first position sorted from the new sample. The eleven highlighted black positions (row 4, in row C quadrant; row 2, in row F quadrant; and H12‐W01 well) indicate wells labelled as “Undetermined” during ReadData.R script and later removed from downstream analysis to avoid potential contamination. Sorting row direction is specified on the right.
59 Materials and Methods
19.4. Step 6. Tagmentation and 5′ fragments isolation Once cDNA was amplified, it was simultaneously fragmented and barcoded by “tagmentation”. The tagmentation step was performed using 96 transposomes, each one containing a distinct index barcode (BC2) to label cDNA from each subarray (96 in total). To assemble transposome stocks, barcoded adapters (6.25 μM) (STRT‐Tn5‐Idx[1–96] + STRT‐TN5‐U) (Table 5), Tn5 transposase (6.25 μM) and glycerol (40%) were mixed during 1h at 37°C. For Tn5 reactions, 3 µl of the previous transposome mix was added to 2 µl of amplified cDNA, in a total volume of 20 µl of 1X CutSmart Buffer (New England Biolabs), and then incubated during 20 min at 55°C. During the tagmentation step, the Tn5 transposase (assembled with the barcoded adapters) randomly cut the amplified double‐stranded cDNA and pasted the annealed barcoded adapters at the place where the cDNA molecule was cleaved. Two possible products are generated, but only the one containing Ithe UM and BC1 at the 5’ end, and the BC2 at the 3’ end will be amplified during the second PCR step.
After tagmentation, 100 µl of Dynabeads MyOne Streptavidin C1 (ratio 1:20 in BB buffer: 10 mM Tris HCl pH 7.5, 5 mM EDTA, 250 mM NaCl, 0.5% SDS) were added to the tagmentation reaction (ratio 1:1) and incubated at room temperature during 15 min. Then, all samples were pooled in a single tube, washed twice in TNT (20 mM Tris HCl pH 7.5, 50 mM NaCl, 0.02% Tween‐20) and finally resuspended in 50 µl of TNT. ExoSAP IT (Affymetrix) was used to clean the remaining adapters using an incubation step of 15 min at 37°C, and then washed twice in TNT and once in EB buffer. The single‐stranded library was eluted in 50 µl of nuclease‐free water by incubation during 10 min at 70°C, after what the supernatant was collected in a clean tube. Next, 1.5X volumes of AMPure XP beads (Beckman) were employed to bind to the library, the supernatant discarded and beads resuspended in the second PCR mix (1X KAPA HiFi Ready Mix supplemented with 200 mM P1_2nd_PCR, 200 nM P2_2nd_PCR) (Table 5) to eperform th following cycle: 95 °C 2 min; 8 cycles: 98 °C 30 sec, 65 °C 10 sec, 72 °C 20 sec; 72 °C 5 min; 10 °C hold. Three consecutive steps to clean and bind the cDNA with AMPure XP beads (0.7X, 0.5X and 1X volumes, respectively) were carried out to finally elute the library in EB buffer (Figure 14: step 6).
19.5. Step 7. Illumina sequencing Before proceeding with sequencing, library concentration and quality was assessed using the Bioanalyzer (High Sensitivity DNA Assay, Agilent). Illumina HiSeq 2000 or 2500 equipment was used for high‐throughput sequencing using a Single‐End 50 cycle kit, and DI‐Read1‐Seq (Read1), STRT‐Tn5‐U (Index 1) and DI_idxP1A‐Seq (Index 2) primers (Table 5). 45 bp‐reads were produced by STRT‐seq‐2i, which were composed by a 6‐bp UMI followed by 3 guanidines and finally the 5’ transcript. The two index reads contained 8 and 5 bp, corresponding to Index 1 (subarray barcode for each 96 pooled well from a 10x10 quadrant) and Index 2 (well barcode (1‐25) within each 10x10 quadrant) (Figure 14: step 7). Samples were
60 Materials and Methods sequenced once in an only run. One lane of the Illumina sequencing flow cell was used per chip; in total, 4 lanes were used.
Process Primer Sequence 5’ 3’ 5′Bio‐AATGATACGGCGACCACCGATCG‐ STRT‐P1‐T31 Lysis/reverse TTTTTTTTTTTTTTTTTTTTTTTTTTTTTTT transcription 5′Bio‐rCrTrArCrArCrGrArCrGrCrTrCrTrTrCrCrGrArTrCrT‐ P1B‐UMI‐RNA‐TSO rNrNrNrNrNrN‐rGrGrG DI‐PCR‐P1A 5′Bio‐AATGATACGGCGACCACCGA PCR 1 5′Bio‐AATGATACGGCGACCACCGAGATCTACAC‐XXXXX‐ DI‐P1A‐idx[1–25]‐P1B CTACACGACGCTCTTCCGATC CAAGCAGAAGACGGCATACGA‐YYYYYYYY‐ STRT‐Tn5‐Idx[1–96] Tagmentation GCGTCAGATGTGTATAAGAGACAG STRT‐TN5‐U 5′PHO‐CTGTCTCTTATACACATCTGACGC P1_2nd_PCR AATGATACGGCGACCACCGAGATC PCR 2 P2_2nd_PCR CAAGCAGAAGACGGCATACGAGAT ATGATACGGCGACCACCGAGATCTACAC‐NNNNNN‐ DI‐Read1‐Seq CTACACGACGCTCTTCCGATCT Sequencing STRT‐Tn5‐U 5′PHO‐CTGTCTCTTATACACATCTGACGC DI_idxP1A‐Seq AATGATACGGCGACCACCGAGATCTACAC Table 5. Custom primers for STRT‐seq‐2i protocol. Primers used for full STRT‐seq‐2i method are displayed in this table as described in Hochgerner et al. 2017. Numbers between [ ] represent a total of 25 primers needed for each well identification in a 10x10 quadrant, and a total of 96 primers required to identify the 96 subarrays. UMI, BC1 and BC2 sequences are denoted in green, orange and red, respectively, according to Figure 14. Bio, biotin; PHO, phosphorylation of 5’ end.
19.6. Preliminary filtering of raw reads First, Illumina HiSeq software performed an intrinsic control annotating reads as invalid. Second, 3’ bases having a bad quality score were discarded. Third, all the six UMI bases in the 5’ end had to satisfy a score that saved the read after cutting away the UMI, else the read was removed. Fourth, a control checking whether the three following bases were guanidines (GGG) was carried out to keep the read. These three G were removed, besides any additional G up to nine, as potential template‐switch derived. Fifth, the read was discarded according to the following assumptions: (1) if the remaining sequence of the read ended in a polyA sequence leaving < 25 bases, (2) if it contained less than six non‐A bases or (3) if it contained a dinucleotide repeat with less than six other bases at either end. Finally, reads were organized by well/cell, given by the two index barcodes. After quality control, Bowtie aligner was used for alignment, only permitting up to 3 base‐mismatches and up to 24 alternative mappings. When a read found no alignment, it was realigned to an artificial chromosome that included every splice junction existent for an exon (UCSC, Genome Browser). The coordinates of aligned splice junctions were translated back to the corresponding genomic positions. . At this point, we estimated the read/sequencing depth by average mRNA total mapped reads per cell.
61 Materials and Methods
Figure 14. Detailed STRT‐seq‐2i method using FACS to sort cells into a WaferGen plate. Schematic STRT‐seq‐2i method modified from Hochgerner et al. 2017, including pictures for visual description of the full‐step single‐cell RNA sequencing process (described in the 19.Single‐cell RNA sequencing protocol section).
62 Materials and Methods
UMI counts were calculated by the number of reads containing each UMI sequence for each genomic position and strand combination, separately for each well. Before that, the exons of each locus with all transcript variants were merged to a combined model representing total expression of the locus. If any multiread (a read that maps to several distinct locations) mapped to some repeat outside exons, it was annotated as one of these repeats by random and it was not considered as an exon‐derived transcript. However, whether the multiread mapped only to some exons, it was assigned to the most similar exon to the transcript model 5’ end. If it did not map to any exon, a random assignation to any mapping was made. After those assignations, the number of distinct UMIs was counted to generate the total number of molecules at each mapping position; nevertheless, if only a single read was produced from any UMI, this UMI was discarded with the purpose of minimizing the probability of finding false molecules derived from PCR or sequencing mistakes. Raw UMI counts were finally corrected for collision probability.
20. Single‐cell RNA sequencing data analysis
Single‐cell RNA sequencing generates huge datasets that need to be analyzed with bioinformatics. Packages for single‐cell sequencing analysis are predominantly developed in R programming language, thus we analyzed our sequencing data using R free software (https://www.r‐project.org/). We installed R for Windows (version 3.4.4) (https://cran.r‐project.org/bin/windows/base/old/3.4.4/) and also RStudio (version 1.1.442) (https://www.rstudio.com/products/rstudio/download/), an interface that integrates a set of tools designed to help with R programming. It contains an editor to write R script before execution, a console, and tools to manage different parameters such as history, environment (where variables are), installed packages, files, plots and help, within others. We preferred to work with RStudio because it generates an easy and intuitive platform that helps beginners to work with R, instead of using a console.
The main R package utilized for our single‐cell sequencing data analysis was the Seurat package (version 2.3.4), developed by Satija lab (‘https://github.com/satijalab/seurat’), in addition to well‐known typical packages already contained in R such as base and graphics, within others. Anyway, all used packages that were not installed by default in R were properly installed and called when necessary at the beginning of any R script.
It usually exists some genes that can be identified in different locations (loci) at the genome; the provided gene list contained by default all these known locations, which were designated by “_locN” after the gene name. In our analysis, few of the detected cluster markers and differentially expressed genes (DE) corresponded to genes with specific locations, so the gene name contained a locus suffix that appeared in the generated figures (e.g. volcano plots, violin plots or DE tables). However, compiled bibliographic information regarding those genes was done without taking into account the exact gene location, using just the single gene name, and this is how it appears written down in S11 and S15.
63 Materials and Methods
The scRNA sequencing analysis was entirely developed during this thesis and involved a new and laborious study for us that consumed an important time of this work. For this reason, the bioinformatics analysis of the scRNA‐seq study has been included in the Results section in addition to the materials and methodology for the dissociation and genomic protocols described here.
21. Gene ontology (GO) analysis
ClueGO v2.5.1 plugin of Cytoescape v3.6.1 software was used for Gene Ontology (GO) analysis (Bindea et al., 2009), using only the genes significantly (p‐value < 0.05) detected as up‐ or down‐regulated in each group for both clustered or total PV cell studies (Supp. Data: R scripts and files > Seurat integrated analysis > Up‐ and down‐regulated genes for GO analysis.R; Supp. Data: Up‐ and down‐regulated genes for GO (CSV)). Databases employed for GO analysis involved Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology Consortium (GO), which include pathways, functions, processes and reactions (named terms) already. known We set up the analysis to find pathways with a p‐value ≤ 0.05 (significant) to which at least 3 genes were found related. Kappa scores (≥ 0.04) were used for functional grouping. We selected the GO term fusion option to keep only the most interesting term for GO parent‐child relation to simplify results. Moreover, term repetitions (the same term can be obtained from more than one database) were discarded to avoid for redundancy and only the first result for a term was maintained. Statistical significance was measured by Enrichment/Depletion (two‐sided hypergeometric test) with Bonferroni step down p‐value correction, the default option in ClueGO. Finally, in every GO analysis, the group leading term (the most significant term within a group) was selected as the most representative term for each group. Those leading terms were then ordered based on their GO group p‐value (from the most significant to the less significant). No exact gene location (loci) was taken into account neither for GO analysis.
22. Statistics
Statistical significance was measured based on the result of the normality test applied to each pair of compared samples. The normality test used for the majority of analyses was D’Agostino‐Pearson (the recommended option in GraphPad Prism 6). When it was not possible to use this test due to the sample size (n<8), the Shapiro‐Wilk test (for the body weight curve analysis) or Anderson‐Darling test (for immunoblotting and qRT‐PCR analyses) were applied to test for normality distribution. Depending on the result of the normality test, statistical significance was measured using Student’s t‐test or Mann‐Whitney U test in each pair of compared samples.
64 RESULTS
Results
1. Cortical synaptic dysfunctions in Dnajc5 conventional KO mice
Previous work of our group showed that the complete deletion of CSP/DNAJC5 (Dnajc5 conventional KO mouse) produced an activity‐dependent degeneration of GABAergic synapses, mainly in PV interneurons (García‐Junco‐Clemente et al., 2010). This work was performed in hippocampus, a brain structure composed by well‐characterized and easily distinguishable layers that present defined neuronal connectivity. One plausible hypothesis is that during the neurodegeneration of GABAergic terminals could exist differential gene expression in PV cells lacking CSP/DNAJC5 compared to control mice, so using the high‐throughput single‐cell RNA sequencing technique we could detect candidate genes for neurodegeneration. When we first started with hippocampal preparations to sort PV cells, we found the limitation of collecting very low yield of cells, which made sorting time very long, compromising cell viability and making subsequent experiments very difficult to optimize, such as gDNA and RNA extraction. Furthermore, once we realized that sorting efficiency would be extremely harmed and cell viability highly compromised, ‐ even more, it would have repercussions in an important objective of this thesis, which was the single‐cell RNA sequencing ‐, we took the decision to switch to the cerebral cortex, with improved sorting results. In perspective, a major advantage of the cortex is the possibility of analyzing circuit dynamics of GABAergic neurons using in vivo two‐photon microscopy in awake behaving mice. In addition, the alterations that we might find could help, at least partially, to understand the motor phenotype found in these mutant mice (see later). With the idea of performing a more detailed study centered in a specific cortical region, we focused on the motor cortical layer II/III, so all the immunohistochemistry and immunofluorescence quantifications, as well as the electrophysiological recordings, were carried out in this cortical region. Layer II/III was chosen due to the highest percentage of PV interneurons (together with layer V) in this particular area (Tamamaki et al., 2003).
Several publications regarding CSP/DNAJC5 have also focused on cortical neurons, even on the whole brain (Sharma et al., 2011, 2012), supporting our decision. Although a neurodegenerative cortical dysfunction has been already reported in these publications, we aimed to confirm it in our hands, looking for a similar dysfunctional synaptic phenotype compared to the hippocampus that would support to focus our efforts in cortical PV neurons during this work.
To assess this question, we first analyzed by immunohistochemistry different presynaptic proteins involved in synaptic vesicle fusion during NTe releas at the presynaptic terminals, using P30 Dnajc5 WT and KO mice (n=2 mice per genotype). With this technique, we aimed to have a global view of CSPα/DNAJC5, PV, SNAP25 and Syt2 expression in the motor cortical region using brain slices. We observed a general reduction in SNAP25 and Syt2 puncta, as well as a reduction in PV axonal labelling of Dnajc5 KO mice, but not in overall PV somata. Moreover, KO mice completely lacked CSPα, as expected (Figure 15A) (Fernández‐Chacón et al., 2004; Tobaben et al., 2001). To confirm these observations, we performed
67 Results immunoblots in Dnajc5 WT and KO animals at two different ages, P15 (before mice develop the neurodegenerative phenotype) and P30 (after impairments are evident), using the same proteins but also including other presynaptic proteins. Protein levels of Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV were examined (n=3 rmice pe genotype and age). Interestingly, we observed significantly reduced levels of SNAP25 in KO animals, both at P15 (**p=0.0079, Student’s t‐test) and P30 (***p=0.0003, Student’s t‐test) compared to WT animals, and also a significant decrease in Hsc70 in KO mice at P30 compared to WT mice (*p=0.024, Student’s t‐test) (Table 6). Furthermore, although not significant, a detectable decrease was observed in Syt2 (P15: p=0.0554, ns, Student’s t‐test; P30: p=0.1261, ns, Student’s t‐test) and PV (P15: p=0.2596, ns, Student’s t‐test; P30: p=0.2262, ns, Student’s t‐test) protein expression at both ages, which possibly corresponded to the loss of PV terminals observed by immunohistochemistry. As expected, CSPα expression was found abolished in KO animals at both ages (P15: ***p=0.0007, Student’s t‐test; P30: ****p<0.0001, Student’s t‐test) (Table 6; Figure 15B, C). These results confirmed that the lack of CSPα affected the cerebral cortex and the hippocampus in a similar way (García‐Junco‐Clemente et al., 2010).
Age Genotype Hsc70 Syt2 Syt1 CSPα Dnajc5 WT 1.00 ± 0.01 1.00 ± 0.11 1.00 ± 0.10 1.00 ± 0.10 P15 Dnajc5 KO 0.93 ± 0.06 0.56 ± 0.13 0.82 ± 0.06 0.04 ± 0.01 Dnajc5 WT 1.00 ± 0.02 1.00 ± 0.09 1.00 ± 0.09 1.00 ± 0.04 P30 Dnajc5 KO 0.80 ± 0.05 0.74 ± 0.12 0.96 ± 0.02 0.03 ± 0.01 Age Genotype SNAP25 Syb2 PV Dnajc5 WT 1.00 ± 0.07 1.00 ± 0.06 1.00 ± 0.10 P15 Dnajc5 KO 0.65 ± 0.03 1.07 ± 0.07 0.83 ± 0.08 Dnajc5 WT 1.00 ± 0.02 1.00 ± 0.12 1.00 ± 0.04 P30 Dnajc5 KO 0.56 ± 0.30 1.12 ± 0.17 0.90 ± 0.05 Table 6. Summary of cortical protein levels in Dnajc5 WT and KO mice by western blot. Normalized mean values ± SEM of cortical protein levels for Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV in Dnajc5 WT and KO mice at P15 and P30. Values are rounded using two decimals. Referred to Fig. 15B, C.
______
Figure 15. Cortical expression of presynaptic proteins from Dnajc5 WT and KO mice. A. Immunohistochemistry experiment (bright field) showing general expression of CSPα, PV, SNAP25 and Syt2 in slices including motor cortical regions (M1 and M2) from P30 animals. Boxes display magnifications of layer II/III from motor cortex. Qualitative reductions in PV, SNAP25 and Syt2 are found in Dnajc5 KO mice, besides a total absence of CSPα. B. Immunoblots showing molecular weights and protein bands of Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV in P15 and P30 mice. C. Normalized quantifications of presynaptic protein levels from immunoblots at two experimental ages (normalized against WT animals). Significant decreased expression of Hsc70 and SNAP25 is detected in Dnajc5 KO mice compared to WT, besides a complete absence of CSPα in KO mice. Detectable reductions in Syt2 and PV are also found in the KO genotype. β‐actin is used as loading control. Error bars indicate SEM. Student’s t‐test is used for statistical significance determination. *p=0.05‐0.01; **p=0.01‐0.001; ***p=0.001‐0.0001; ****p<0.0001.
68 Results
2. Specific Cre‐recombinase activity at the Dnajc5 locus in the PVcre:Ai27D:Dnajc5flox mouse line
Because a relevant synaptic degeneration of PV interneurons was observed at the hippocampus and cortex of Dnajc5 KO mice, we aimed to further investigate these cells deeply. For that purpose, we generated a mouse model that specifically lacked Dnajc5 only in PV cells and also carried the ChR2‐ tdTomato fusion protein for FACS separation (PVcre:Ai27D:Dnajc5flox mouse line). With this strategy, we were able to study how the absence of CSPα/DNAJC5 affected the viability and functionality of PV cells over time, since the conventional Dnajc5 KO mouse has a lifespan of 30‐40 days, impairing this kind of studies. This strategy also opened the opportunity to investigate if cell autonomous neurodegeneration and early death would occur due to a general PV cell dysfunction.
69 Results
First of all, in order to validate our working model and confirm the specificity of Cre‐recombinase activity at the Dnajc5 locus, we checked the novel PVcre:Ai27D:Dnajc5flox mouse line. The deletion of the floxed‐sequence at Dnajc5 gene upon Pvalb‐promoter‐driven Cre activity was first studied at two levels, genomic DNA (gDNA) and mRNA (cDNA). As previously described, EUCOMM generated a strategy by which the exon 3 of the Dnajc5 locus remained flanked by two loxP sites and one residual FRT site (tm1c) (Figure 16A). Taking advantage of this approach, we produced separated strategies to detect the removal of exon 3 at the Dnajc5 locus, using both gDNA and mRNA. For band recognition at gDNA level, we designed the forward primer SLB‐2017‐01 and the reverse primer SLB‐2017‐02 surrounding the exon 3, which were able to produce three band sizes: a band of 750 bp from both Dnajc5 WT and KO locus, a band of 1029 bp from Dnajc5 floxed‐allele (tm1c) and a band of 245 bp from the recombined Dnacj5 gene without the exon 3 (tm1d). The same band size was obtained from Dnajc5 WT and KO genes because the production strategy of the KO allele does not affect the region where primers hybridized (Figure 16A). For mRNA bands detection, we used the forward primer MLM‐2005‐01 within exon 2 and the reverse primer MLM‐2005‐02 within exon 4, which generated two different PCR products: a band of 348 bp from Dnajc5 WT gene and Dnajc5 floxed‐allele (tm1c), and a band of 134 bp from the recombined Dnajc5 allele (tm1d). Because the forward primer hybridized in exon 2, which was the one deleted during the production of the KO allele, there was no PCR product from KO mRNA (Figure 16B; S1).
Although the Dnajc5 floxed deletion was already appreciated by PCR at 1 month (not shown), there was no early mortality or evident morbidity in mutant mice. Because the motor phenotype – that will be described in the next section‐ was more evident at 2 months old than at 1 month, we decided to carry out all the experiments at 2 months. Cortical FAC‐sorted tdTomato‐positive and negative fractions from 2‐ month old mice (n=3 for both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice) were properly collected (Figure 17A) and gDNA or cDNA was obtained as described. As control samples, in order to corroborate all band sizes obtained by PCR, we used one littermate pair of Dnajc5 WT and KO conventional mice, as well as one littermate pair of Dnajc5flox/flox mice treated with tamoxifen, with or without a transgene driving the ubiquitous expression of Cre‐recombinase (UBC‐cre‐ERT2). After Cre‐recombinase activity, it was observed a gDNA band of 245 bp in tdTomato‐positive sorted cells that were not detected in tdTomato‐negative sorted fractions, both in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice, confirming the specific deletion of the floxed Dnajc5 gene by Cre‐recombinase in PV cells (Figure 17B). We also appreciated the Dnajc5 WT and KO bands of 750 bp presented in both control and mutant animals (Figure 17B), as expected based on our gDNA strategy (PVcre:Ai27D:Dnajc5flox/+ mice carried a Dnajc5 WT allele and PVcre:Ai27D:Dnajc5flox/‐ animals contained a Dnajc5 KO allele) (Figure 16A). We additionally observed in tdTomato+ cells from both genotypes a 1029 bp‐band that could potentially correspond to
70
Results residual Dnajc5 floxed alleles that did not suffer Cre‐dependent recombination, which was expected to be found only in tdTomato‐negative fractions (Figure 17B).
Figure 16. Strategies to validate the deletion of Dnajc5 floxed gene upon Cre‐recombinase activity driven by Pvalb promoter at gDNA and mRNA levels. A. Genomic DNA strategy displaying the Dnajc5 locus for WT, KO, non‐recombined floxed allele (tm1c) and Cre‐ deleted floxed allele (tm1d). The intronic region between exon 1 and 2 is not fully represented due to space restriction. B. RNA strategy showing the Dnajc5 mRNA for WT, KO, tm1c and tm1d alleles. Red arrows represent forward and reverse primers located at the annealing site. PCR product sizes are shown on the right. All pictures are scaled except for FRT and loxP sites, primers length and the β‐actin + neomycin cassette in the KO allele. Introns and exons are represented with black lines and green boxes, respectively.
On the other hand, Cre‐recombinase activity at the Dnajc5 locus was also confirmed at the mRNA level. The deletion of the floxed region in gDNA generated a mRNA transcript of 134 bp only observed in tdTomato+ sorted fractions from both genotypes (Figure 17C). However, another band of 348 bp was also detected in tdTomato+ cells of PVcre:Ai27D:Dnajc5flox/+animals, which might proceed from the Dnajc5 WT alleles that also carried these mice, but even from residual non‐recombined Dnajc5 floxed alleles (Figure 17C), as similarly detected at gDNA (Figure 17B). This assumption was also suggested by the presence of the same band of 348 bp in PV cells from PVcre:Ai27D:Dnajc5flox/‐ mice; in this case, the Dnacj5 KO allele could not be amplified by our mRNA strategy (Figure 16B), so the band should only proceed from residual non‐recombined Dnajc5 floxed alleles (Figure 17C).
71
Results
In conclusion, although Dnajc5 Cre‐dependent deletion seemed to be specific to PV cells (tdTomato+ sorted fractions), residual Dnajc5 floxed alleles that did not suffer recombination were still present in these fractions, suggesting a partial Cre‐recombinase activity or even an impure PV cell isolation by FACS, although our sorting strategy was quite restrictive (Figure 17A).
Once we confirmed the specific Cre‐recombinase activity at Dnajc5 locus, we checked whether the mRNA product obtained from the deletion of Dnajc5 (134 bp‐band) may be somehow translated and may generate a truncated protein. While the wild type CSPα/DNAJC5 protein consists of 198 amino acids and has a molecular weight of 22 KDa approximately, the aberrant protein would contain 112 amino acids and
72
Results a molecular weight close to 12 KDa (Figure 13). Thereby, we checked the presence of CSPα/DNAJC5 proteins in liver lysates from Dnajc5flox/flox and UBC‐cre‐ERT2:Dnajc5flox/flox mice treated with tamoxifen (n=3 for each genotype), a tissue where we previously detected that the Dnajc5 floxed allele was acutely removed by Cre‐recombinase activity (previous laboratory data). If any aberrant protein would exist, it should be found in this tissue. We used cortical and liver samples from Dnajc5 WT and KO mice at P30 as positive and negative controls, respectively (n=2 for liver samples and n=1 for cortical samples per genotype). First of all, we used an antibody against CSPα/DNAJC5 (ENZO) that recognizes the C‐terminus of the protein to obtain a clear and general detection of CSPα/DNAJC5 in these tissues. We obtained a ̴34 KDa band in all liver samples from Dnajc5 WT and Dnajc5flox/flox mice (Figure 18A), although this protein could not be detected in liver samples from Dnajc5 KO and UBC‐cre‐ERT2:Dnajc5flox/flox mice, suggesting a complete abolishment of wild type CSPα/DNAJC5 expression after Cre‐activity in those mice. Cortical KO control sample showed no CSPα/DNAJC5 protein expression as previously described (Fernández‐Chacón et al., 2004; Tobaben et al., 2001). Nevertheless, if any aberrant protein is generated, it could not be detected using this antibody due to its C‐terminus recognition. The potential truncated protein would suffer a switch in the reading frame after exon 3 removal, leading to a different amino acid composition from the 37th position to the end (C‐terminus), thus making this antibody inappropriate for aberrant protein detection.
______
Figure 17. Validation of the floxed allele recombination by Cre activity at the Dnajc5 locus in tdTomato+ sorted cells. A. Representative sorting of cortical tdTomato‐positive (tdT+) and tdTomato‐negative (tdT‐) cells belonging to PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice using the BD FACSJazz sorter. “Cells” selects only alive cells; “Singlets” selects unique cells, discarding events with high size (potential doublets); “tdT‐“ shows cells selected as negative control (tdTomato‐negative cells); “tdT+” represents cells that express tdTomato (561 nm laser). Table below shows specific parameters obtained for 100000 events from “Singlets” for this representative example. Note that the percentage of tdT+ cells is 0.37% of total events and 0.96% of parent, which makes sorting efficiency slow. This percentage was similar in all FACS experiments, both for control and mutant mice (around 0.1‐0.4% of 100000 total events). B. Gel electrophoresis showing equivalent amounts of gDNA. The gDNA PCR shows the deleted Dnajc5 floxed allele (245 pb, lower arrow) upon Cre‐recombinase activity in tdT+ sorted fractions from both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. These fractions also contain the wild type or the KO Dnajc5 allele (750 bp, middle arrow), and a residual non‐recombined Dnajc5 floxed allele (1029 bp, upper arrow). tdT‐ sorted fractions only show the WT or KO Dnajc5 allele (750 bp) and the non‐recombined Dnajc5 floxed allele (1029 bp). C. Gel electrophoresis showing equivalent amounts of cDNA. The cDNA PCR shows the deleted Dnajc5 floxed allele (134 pb, lower arrow) upon Cre‐recombinase activity in tdT+ sorted fractions from both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes; tdT+ fractions belonging to control animals also contain the Dnajc5 WT and/or residual non‐recombined Dnajc5 floxed fragments (348 bp, upper arrow); tdT+ fractions from mutant mice show a 348 bp band corresponding only to the residual non‐recombined Dnajc5 allele (KO allele cannot be amplified using our mRNA strategy). tdT‐ sorted fractions only show the non‐ recombined Dnajc5 floxed allele (348 bp). SSC, Side Scatter (internal cell complexity); FSC, Forward Scatter (cell size/volume); FITC, Fluorescein IsoTioCyanate (488 nm laser).
73 Results
Our laboratory disposes of an antibody (R807) that recognizes the full‐length CSPα/DNAJC5 protein (Tobaben et al., 2001), enabling a proper detection of a potentially CSPα/DNAJC5 truncated protein. Nonetheless, although the R807 antibody did recognize the wild type protein in livers from Dnajc5 WT and Dnajc5flox/flox mice ( ̴34 KDa), there were no differences in expression patterns between Dnajc5flox/flox and UBC‐cre‐ERT2:Dnajc5flox/flox samples at the lower‐molecular‐weight section of the blot, neither when we isolated and highly exposed that region (from 10 to 19 KDa approximately) (n=3 mice per genotype) (Figure 18B).
Once we confirmed the absence of aberrant versions of CSPα proteins with low molecular weight in liver samples, a tissue that previously showed a strong deletion of floxed Dnajc5 mRNA, we carried out the same test using the R807 antibody in cortical lysates from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice (n=3 mice per genotype). A band with a molecular weight close to 30 KDa, corresponding to the wild type protein, was found in all samples except for the Dnajc5 KO negative control. We also observed a small reduction in CSPα/DNAJC5 expression in PVcre:Ai27D:Dnajc5flox/‐ samples, a surprising result because the Dnajc5 floxed allele was only removed from PV cells. Nevertheless, we did not find signs of truncated proteins, even upon long time exposures, at the molecular range between 10‐ 19 KDa (Figure 19).
3. tdTomato expression is specific for the parvalbumin‐positive interneuron population
In the previous section we confirmed a correct Cre‐recombinase activity at Dnajc5 locus, so the next step was to validate the specificity of the tdTomato+ labelling on PV interneurons. For this purpose, we carried out immunofluorescence experiments to quantify PV‐tdTomato colocalization in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old (n=90 fields per genotype from a total of 3 mice per genotype). PV+ somata and tdTomato+ perisomal labelling were considered for ______
Figure 18. Analysis of CSPα/DNAJC5 protein in liver lysates from UBC‐cre‐ERT2:Dnajc5flox/flox mice. Immunoblots with equivalent amounts of protein from cortical brain or liver tissue identifying CSPα/DNAJC5 with two different antibodies. Dnajc5 WT and KO are used as positive and negative controls, respectively. Samples are obtained from Dnajc5 WT mice (WT), Dnajc5 conventional KO mice (KO) and Dnajc5flox/flox mice (F/F) treated with tamoxifen, without (‐) or with (+) a transgene driving the ubiquitous expression of Cre‐recombinase (UBC‐cre‐ ERT2). A. Blot analyzed with an antibody raised against the C‐terminus of CSPα; it shows that the liver from UBC‐ cre‐ERT2:Dnajc5flox/flox mice treated with tamoxifen lacks CSPα. B. Blot analyzed with an antibody raised against a GST‐fusion protein of CSPα (R807); it shows that the liver from UBC‐cre‐ERT2:Dnajc5flox/flox mice treated with tamoxifen lacks CSPα (arrow). The right panel represents a higher exposure image of the same blot showing no evidence for the existence of a truncated lower molecular weight version of CSPα upon Cre‐recombinase mediated genomic deletion. Note that the liver CSPα/DNAJC5 presents a higher molecular weight compared to the cortical protein. β‐actin is used as loading control.
74 Results
75 Results
Figure 19. Analysis of CSPα/DNAJC5 protein in cortical lysates from both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. Immunoblot containing equivalent amounts of protein from cortical brain tissue analyzed with an antibody that recognizes the GST‐fusion protein of CSPα (R807). Dnajc5 WT and KO are used as positive and negative controls, respectively. Samples are obtained from Dnajc5 WT mice (WT), Dnajc5 conventional KO mice (KO), PVcre:Ai27D:Dnajc5flox/+ mice (F/+) and PVcre:Ai27D:Dnajc5flox/‐ mice (F/‐). This blot shows a small reduction in CSPα expression (arrow) in cortical brain tissue of PVcre:Ai27D:Dnajc5flox/‐ mice. Right panel displays a higher exposure image of the same blot showing no evidence for the existence of a truncated lower molecular weight version of CSPα upon Cre‐recombinase mediated genomic deletion. β‐actin is used as loading control. quantification. We obtained around 90% of colocalization for both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes,while percentages of non‐colocalization were 7 and 10% for PV+ neurons and 4 and 1% for tdTomato+ cells in controls and mutants, respectively (Table 7; Figure 20A). No differences were found between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice in total PV‐ tdTomato colocalization (p=0.3369, ns, Mann‐Whitney U test), as expected from the same promotor‐ driven Cre‐activity (PVcre).
Age Genotype PV‐tdTomato PV+ cells with no tdTomato+ cells with colocalization (%) colocalization (%) no colocalization (%) Dnajc5flox/+ 88.73 ± 0.97 7.41 ± 0.52 4.52 ± 0.94 2 months Dnajc5flox/‐ 88.78 ± 0.64 10.42 ± 0.66 0.95 ± 0.19 Table 7. Summary of colocalization values between parvalbumin and tdTomato expression in the experimental mice. Mean colocalization values (%) ± SEM between parvalbumin and tdTomato in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old. The table also shows the percentage of PV+ cells and tdTomato+ cells that did not colocalize. Values are rounded using two decimals. Referred to Fig. 20A.
76
Results
Furthermore, we also investigated the specificity of colocalization in vivo by sorting tdTomato+ cells, namely whether we were obtaining an enriched population of PV neurons when sorting tdTomato+ cells compared to tdTomato‐ cells. We performed a real‐time quantitative PCR to study mRNA levels of parvalbumin both in tdTomato‐positive and negative sorted fractions from control and mutant animals (n=3 for each genotype). Results were normalized versus negative fractions from each genotype and represented as 2(‐ΔΔCt), normalized mRNA levels. In both genotypes, parvalbumin mRNA was strongly increased in tdTomato+ fractions compared to tdTomato‐ fractions, although an important variability within samples of the same genotype was found, generating large standard error of the mean (SEM) values and non significant results (p=0.29, ns, Student’s t‐test for controls; p=0.30, ns, Student’s t‐test for mutants) (Table 8). Differences were not significant either between tdTomato‐ fractions (p>0.99, ns, Student’s t‐ test) or between tdTomato+ fractions (p=0.29, ns, Student’s t‐test) of controls versus mutants. These data proved that we were not losing many PV cells in tdTomato‐negative fractions, but instead we were obtaining a highly enriched population of PV neurons when sorting tdTomato+ cells, confirming a specific tdTomato+ labelling in PV cells (Figure 20B).
Age Genotype tdTomato‐ sorted cells tdTomato+ sorted cells Dnajc5flox/+ 1.00 ± 0.32 991440 ± 810214 2 months Dnajc5flox/‐ 1.00 ± 0.84 1971 ± 1638 Table 8. Summary of expression levels for PV mRNA in sorted cell fractions obtained from the experimental mice. Normalized mean values ± SEM of mRNA expression in tdTomato‐negative (tdTomato‐ cells) and tdTomato‐ positive (tdTomato+ cells) sorted cells in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old. Values are rounded using two decimals. Referred to Fig. 20B.
4. The specific deletion of Dnajc5 in PV interneurons does not cause increased mortality but progressive body weight alterations in PVcre:Ai27D:Dnajc5flox/‐ mice
PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice did not show neither early lethality nor a distinguishable phenotype immediately after birth. We developed a continuous survival curve from P0 until 8 months postnatally. No significant differences in mortality were detected between mutant and control mice (96.552% survival at 8 months for PVcre:Ai27D:Dnajc5flox/+ mice; 93.103% survival at 8 months for PVcre:Ai27D:Dnajc5flox/‐ mice; p=0.5614, ns, Log‐rank Mandel‐Cox test; n=29 per genotype) (Figure 21A). Furthermore, we observed that mutant animals could survive even for longer periods (at least until 15 months) without showing dramatic lethality compared to controls; however, it was frequently necessary to provide food and water on the cage floor to facilitate feeding access due to hindlimb locomotor deterioration (see next section).
77
Results
Figure 20. Motor cortical neurons show high colocalization of PV and tdTomato labelling. A. Left, representative confocal images showing motor cortical neurons of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice labelled with anti‐parvalbumin (green) and anti‐mCherry (red) antibodies. White squares identify magnified images of cortical layer II/III neurons. Colocalization between PV and tdTomato is pointed out using white arrows in the merge panel. The fusion of tdTomato with ChR2 generates the expected perisomal and axonal labelling localized to the membrane. Right, Venn diagrams displaying the percentage of colocalization between PV and tdTomato in motor cortical layer II/III for both genotypes, and also the percentages of PV+ cells and tdTomato+ cells that do not colocalize (n≈90 fields per genotype, 3 mice per genotype). B. Levels of parvalbumin mRNA (2(‐∆∆Ct)) obtained from real‐time qPCR of FAC‐sorted tdTomato+ and tdTomato‐ cells from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. As expected, there is a predominance of PV cells in positive fractions of both genotypes. Error bars indicate SEM and statistical significance is determined by Mann‐ Whitney U test for the colocalization analysis and Student’s t‐test for qRT‐PCR.
During growth, certain phenotypic characteristics became notorious in mutant mice. Around P30, we realized that PVcre:Ai27D:Dnajc5flox/‐ mice apparently had reduced body size compared to their littermate controls, so we decided to carry out a body weight curve from P30 (1 month) to P240 (8 months). Male and female body weights are naturally different, so the study was made separately for each sex (n=5
78
Results for male and female control mice, n=5 for male mutant mice and n=4 for female mutant mice). Data showed that male PVcre:Ai27D:Dnajc5flox/‐ mice experienced body weight decrease compared to male control mice, which became statistically significant from P37 to P240 (e.g. P30: p=0.6994, ns, Student’s t‐test; P128: ***p=0.001, Student’s t‐test; P240: *p=0.0159, Mann‐Whitney U test) (Table 9; Figure 21B, D). Female PVcre:Ai27D:Dnajc5flox/‐ mice also showed a decrease in body weight similar to mutant males. Compared to female control mice, they exhibited statistically significance from P51 to P240 (e.g. P30: p=0.3576, ns, Student’s t‐test; P128: ****p=0.0001, Student’s t‐test; P240: **p=0.0017, Student’s t‐test) (Table 9; Figure 21C, D).
Males (g) Females (g) Age Dnajc5flox/+ Dnajc5flox/‐ Dnajc5flox/+ Dnajc5flox/‐ P30 15.66 ± 1.15 15.15 ± 0.59 14.97 ± 0.46 13.89 ± 1.01 P37 19.38 ± 0.69 16.39 ± 0.17 16.67 ± 0.51 15.43 ± 0.65 P44 21.92 ± 0.40 18.67 ± 0.37 17.85 ± 0.62 16.26 ± 0.49 P51 24.19 ± 0.43 20.68 ± 0.52 19.72 ± 0.69 17.00 ± 0.48 P58 25.18 ± 0.54 22.32 ± 0.36 20.74 ± 0.65 18.24 ± 0.62 P65 26.45 ± 0.80 22.36 ± 0.65 20.94 ± 0.68 17.27 ± 0.36 P72 26.83 ± 0.75 22.37 ± 0.42 21.83 ± 0.70 17.98 ± 0.50 P79 28.48 ± 0.86 23.48 ± 0.76 22.88 ± 0.58 18.78 ± 0.45 P100 32.26 ± 1.56 24.51 ± 1.09 24.10 ± 0.68 19.03 ± 0.32 P128 35.46 ± 1.67 24.78 ± 1.28 25.69 ± 0.54 20.24 ± 0.10 P156 37.89 ± 2.25 23.70 ± 0.74 27.88 ± 1.53 19.80 ± 0.45 P184 38.21 ± 2.42 23.09 ± 0.91 28.05 ± 1.20 19.19 ± 0.20 P212 39.41 ± 2.66 23.48 ± 1.72 28.56 ± 1.62 19.02 ± 0.33 P240 41.93 ± 2.40 24.88 ± 0.69 29.05 ± 1.65 19.55 ± 0.47 Table 9. Summary of body weight values in the experimental mice (males and females). Mean values ± SEM of body weight (g) in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ male and female mice from P30 to P240 (1 month to 8 months old). Values are rounded using two decimals. Referred to Fig. 21B, C, D.
5. Neurological phenotype upon deletion of Dnajc5 in PV interneurons
After birth, control and mutant mice were indistinguishable. However, long term observations during growth revealed a characteristic neurological phenotype in mutant mice. Around P30, the phenotype was mildly observed in PVcre:Ai27D:Dnajc5flox/‐, but it was at P60 when it became noticeable and easily detected, so we established 2 months (adult mice) as a suitable age to investigate phenotypes. At this time point, mutant mice presented the following phenotypical characteristics: (1) a high locomotor activity level, including disorganized movements throughout the cage composed of time‐variable patterned running episodes; (2) a hindlimb deterioration that generated a characteristic ataxic gait and hyperreflexia, with tendency to stretch hindlimbs during tail suspension; (3) a slight back hump; (4) lumbar lordosis, which was revealed by their ability to contort easily; and (5) abnormal head movements, both at rest and during locomotion, which included spontaneous head tossing movements and upward gazes that sometimes were sustained during seconds (called stargazing), similar to the movements reported for
79 Results
Figure 21. Survival curve and body weight screening of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice until 8 months postnatally. A. Analysis of late survival shows the absence of differences in mortality between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice over 8 months. Statistical significance is calculated using the Log‐Rank (Mantel‐Cox) test (p=0.5614). B. Body weight curve for male mice from P30 to P240 (8 months) between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes. Significant changes are observed from P37. C. Body weight curve for female mice from P30 to P240 (8 months) between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes. Significant changes are observed from P51. D. Plot representing body weight curves for male and female mice from both upper plots. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. *p=0.05‐0.01; **p=0.01‐0.001; ***p=0.001‐0.0001.
80 Results stargazer knock‐out mice (Letts, 2005; Letts et al., 1998; Noebels et al., 1990; Osten and Stern‐Bach, 2006). Moreover, in certain litters, it was also observed (6) a repeated backward somersaulting activity. These observations led us to conclude that PVcre:Ai27D:Dnajc5flox/‐ mice developed a neurological phenotype, easily observed at P60 (Supp. Data: Neurological phenotype (videos)).
As impaired movements were observed in mutants, we quantified locomotor activity in these mice. To evaluate this behavior, we performed the open field test separately on males and females (n=11 control males, n=10 mutant males, n=9 control females, n=14 mutant females). The locomotor behavior of both male and female PVcre:Ai27D:Dnajc5flox/‐ mice produced a non‐stop motion track around the whole cage compared to controls (Figure 22A). The data analyzed from video recordings revealed that, compare to controls, male and female PVcre:Ai27D:Dnajc5flox/‐ animals displayed significant increases in velocity (males: ****p<0.0001, Mann‐Whitney U test; females: ***p=0.0001, Student’s t‐test), track length (males: ****p<0.0001, Mann‐Whitney U test; females: ***p=0.0001, Student’s t‐test) and activity (males: ****p<0.0001, Mann‐Whitney U test; females: ****p<0.0001, Student’s t‐test). Only mutant males showed significantly increased number of ambulations (spontaneous short‐term acceleration) (males: ***p=0.0003, Mann‐Whitney U test; females: p=0.0558, ns, Student’s t‐test). On the other hand, normal head movements, such as headbobs (males: ****p<0.0001, Student’s t‐test; females: ****p<0.0001, Student’s t‐test) and headstretches (males: ****p<0.0001, Student’s t‐test; females: ****p<0.0001, Student’s t‐test), and also regular tailmoves (males: ****p<0.0001, Student’s t‐test; females: ***p<0.0004, Student’s t‐test) were significantly reduced, both in PVcre:Ai27D:Dnajc5flox/‐ males and females compared to control animals. Nevertheless, we did not find apparent differences in anxiety between genotypes, a behavior that is mildly assessed in the open field assay by wall distance, neither in males (p=0.1905, ns, Mann‐Whitney U test) nor females (p=0.279, ns, Student’s t‐test) (Table 10; Figure 22B).
Males Females Age Dnajc5flox/+ Dnajc5flox/‐ Dnajc5flox/+ Dnajc5flox/‐ Velocity (cm/s) 6.29 ± 0.37 17.16 ± 1.93 5.93 ± 0.48 13.59 ± 1.28 Track length (cm) 11335 ± 664.60 30881 ± 3474 10698 ± 866.90 24233 ± 2293 Activity (%) 48.95 ± 1.77 81.28 ± 2.19 45.96 ± 1.92 71.92 ± 2.99 # Ambulations 2.18 ± 0.87 41 ± 14.27 11.22 ± 3.64 90.36 ± 30.98 Wall distance (cm) 5.01 ± 0.31 5.40 ± 0.41 4.37 ± 0.31 4.96 ± 0.38 # Headbobs 1827 ± 73.99 791.10 ± 88.81 2226 ± 97.40 1113 ± 106 # Headstretches 2652 ± 94.02 986.40 ± 111.70 2669 ± 140.40 1317 ± 124.70 # Tailmoves 559.90 ± 36.19 284.20 ± 36.30 776.40 ± 68.19 422.70 ± 50.84 Table 10. Summary of the open field assays performed in the experimental mice (males and females). Mean values ± SEM for velocity, track length, activity, ambulations, wall distance, headbobs, headstretches and tailmoves in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old, separating males and females. Values are rounded using two decimals. Referred to Fig. 22B.
81
Results
Male and female results were very similar, so they were also analyzed together by mixing control and mutant raw data: increments in velocity (****p<0.0001, Mann‐Whitney U test), track length (****p<0.0001, Mann‐Whitney U test), activity (****p<0.0001, Mann‐Whitney U test) and ambulations (***p=0.0007, Mann‐Whitney U test) were detected in PVcre:Ai27D:Dnajc5flox/‐ mice. Decrements in headbobs (****p<0.0001, Student’s t‐test), headstretches (****p<0.0001, Student’s t‐test) and tailmoves (****p<0.0001, Student’s t‐test) were found in total PVcre:Ai27D:Dnajc5flox/‐ mice compared to total PVcre:Ai27D:Dnajc5flox/+ animals (Table 11; Figure 22C). No changes were obtained in wall distance between genotypes using total mice (p=0.255, ns, Student’s t‐test) (Figure 22C).
Total mice Age Dnajc5flox/+ Dnajc5flox/‐ Velocity (cm/s) 6.13 ± 0.29 15.08 ± 1.13 Track length (cm) 11048 ± 524.90 27120 ± 2035 Activity (%) 47.60 ± 1.31 75.82 ± 2.16 # Ambulations 6.25 ± 1.95 69.79 ± 19.38 Wall distance (cm) 4.72 ± 0.23 5.15 ± 0.28 # Headbobs 2007 ± 73.88 979.10 ± 78.04 # Headstretches 2660 ± 79.37 1179 ± 91.18 # Tailmoves 657.40 ± 43.25 365 ± 35.64 Table 11. Summary of the open field assays performed in the experimental mice (all together). Mean values ± SEM for velocity, track length, activity, ambulations, wall distance, headbobs, headstretches and tailmoves in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old, mixing males and females. Values are rounded using two decimals. Referred to Fig. 22C.
Although the mutant mice lived long (Figure 21A), they were progressively losing mobility along time. At 8 months, few PVcre:Ai27D:Dnajc5flox/‐ animals needed to be humanly sacrificed due to high mobility loss, especially in hindlimbs. It was also distinctive to observe 8‐month old mutant mice reclined on their stretched hindlimbs and, as they were not able to fully control their own body balance, often falling down, even during locomotion (Supp. Data: Neurological phenotype (videos)). Because we observed this time‐dependent progressive phenotype, we decided to establish two temporal intervals for future experiments (P60: 2 months, and P240: 8 months), in order to study both, early and late stages of the neurological phenotype.
______
Figure 22. Neurological phenotype upon deletion of Dnajc5 in PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months of age. A. Representative track examples from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice during a 30 minute‐recording open field experiment. Mutant mice show a motion track characterized by non‐stop movements throughout the whole cage. B. PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice are analyzed in open field experiments during 30 minutes, in dark, and quantification of velocity (cm/s), track length (cm), activity (%), ambulations (spontaneous short‐term acceleration), wall distance (cm), and number of headbobs, headstretches and tailmoves are performed. C. Same analysis as B but mixing males and females due to their data similarity. Error bars indicate SEM and statistical significance is obtained by Student’s t‐test and Mann‐Whitney U test. ***p=0.001‐0.0001; ****p<0.0001.
82
Results
83 Results
6. Cortical expression of presynaptic CSPα/DNAJC5, synaptotagmin‐2 and Hsc70 proteins are progressively reduced in PVcre:Ai27D:Dnajc5flox/‐ mice
Due to the clear neurological phenotype described in PVcre:Ai27D:Dnajc5flox/‐ animals, we aimed to analyze in depth how the lack of CSPα/DNAJC5 affected the cortical regions of these mice. We first performed immunohistochemistry experiments to obtain a general view of the expression patterns for those presynaptic proteins (CSPα/DNAJC5, PV, SNAP25 and Syt2) previously studied in the conventional Dnajc5 KO mouse. In this case, differences between controls and mutants were quite difficult to observe by eye, although we could say that CSPα, PV and Syt2 labelling seemed slightly decreased in motor cortical regions of PVcre:Ai27D:Dnajc5flox/‐ mice at least at 8 months (n=3 mice per genotype and age) (Figure 23).
Figure 23. Global expression of presynaptic proteins in cortical slices from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally. Immunohistochemistry experiment (bright field) showing general expression of CSPα, PV, SNAP25 and Syt2 in slices including motor cortical regions. Boxes display magnifications of layer II/III from motor cortex. A mildly decrease in CSPα, PV and Syt2 expression is appreciated at least at 8 months in mutant animals compared to controls.
Because the removal of the Dnajc5 floxed gene occurred only in PV cells, it was difficult to detect differences between genotypes using immunohistochemistry observations at a qualitative level. This is, however, not surprising because peroxidase‐based staining is poorly quantitative compared, for example,
84
Results with western blot or immunofluorescence. Thus, we further examined protein levels by western blot. Following the same approach as in conventional Dnajc5 KO mice, presynaptic proteins were studied in cortex from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals at 2 and 8 months old (n=4 for both genotypes and ages). Surprisingly, a significant decrease in CSPα/DNAJC5 expression was observed in mutants at 2 months (*p=0.0318, Student’s t‐test) and this reduction was greater at 8 months (**p=0.005, Student’s t‐test). On the other hand, it was also appreciated a mildly fall in Syt2 expression at 2 months (p=0.17, ns, Student’s t‐test) that became significant at 8 months (*p=0.0475, Student’s t‐test) in PVcre:Ai27D:Dnajc5flox/‐ animals compared to controls. Furthermore, the expression of Hsc70 was found significantly reduced (***p=0.0002, Student’s t‐test) at 8 months in PVcre:Ai27D:Dnajc5flox/‐ animals. We also detected a progressive reduction in SNAP25 expression, although it was not significant (2 months: p=0.6891, ns, Student’s t‐test; 8 months: p=0.1405, ns, Student’s t‐test) (Table 12; Figure 24A, B). Somehow, the deletion of Dnajc5 floxed gene specifically in PV interneurons affected the expression levels of presynaptic proteins that might potentially lead to neurological dysfunctions. These changes were detected when analyzing the whole cortex. We, therefore, assumed that the overall reduction in CSPα/DNAJC5 represented a specific decrease of CSPα/DNAJC5 protein levels in PV interneurons.
Age Genotype Hsc70 Syt2 Syt1 CSPα Dnajc5flox/+ 1.00 ± 0.04 1.00 ± 0.07 1.00 ± 0.08 1.00 ± 0.07 2 months Dnajc5flox/‐ 0.96 ± 0.03 0.88 ± 0.03 1.06 ± 0.10 0.76 ± 0.05 Dnajc5flox/+ 1.00 ± 0.04 1.00 ± 0.02 1.00 ± 0.04 1.00 ± 0.07 8 months Dnajc5flox/‐ 0.64 ± 0.03 0.91 ± 0.03 0.94 ± 0.03 0.62 ± 0.05 Age Genotype SNAP25 Syb2 PV Dnajc5flox/+ 1.00 ± 0.06 1.00 ± 0.04 1.00 ± 0.02 2 months Dnajc5flox/‐ 0.97 ± 0.03 1.05 ± 0.06 0.95 ± 0.06 Dnajc5flox/+ 1.00 ± 0.04 1.00 ± 0.06 1.00 ± 0.05 8 months Dnajc5flox/‐ 0.88 ± 0.05 1.04 ± 0.08 1.15 ± 0.08 Table 12. Summary of cortical protein levels in the experimental mice by western blot. Normalized mean values ± SEM of cortical protein levels for Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months old. Values are rounded using two decimals. Referred to Fig. 24.
7. Progressive loss of Syt2+ and PV+ presynaptic puncta in PVcre:Ai27D:Dnajc5flox/‐ mice with similar PV cell number and size 7.1. Age‐dependent loss of PV+ and Syt2+ synaptic puncta in PVcre:Ai27D:Dnajc5flox/‐ mice Synaptotagmin‐2 has been previously described as a specific presynaptic maker of PV cells in hippocampal cultures (García‐Junco‐Clemente et al., 2010) and in different cortical areas (Sommeijer and Levelt, 2012). To check if the unique absence of CSPα/DNAJC5 in PV interneurons affected the integrity of their presynaptic terminals, we examined the presence of Syt2+ synaptic puncta in motor cortical layer II/III of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. To confirm that Syt2 is mainly restricted
85
Results to PV terminals, we also quantified the presence of PV+ synaptic puncta to evaluate both markers. Using 60X confocal images, we observed by eye a general reduction in synaptic puncta, both for parvalbumin and synaptotagmin‐2, in PVcre:Ai27D:Dnajc5flox/‐ compared to PVcre:Ai27D:Dnajc5flox/+ animals at both ages (Figure 25A). In detail, quantifications revealed significant reductions at 2 months in both PV+ and Syt2+ synaptic puncta in mutants (PV: *p=0.0124, Student’s t‐test; Syt2: *p=0.0114, Student’s t‐test) when taking all data together (n=82 fields for PV+ labelling and n=76 fields for Syt2+ labelling from 3 control mice; n=79 fields for both PV+ and Syt2+ labelling from 3 mutant mice) (Table 27; Figure 25B).
Figure 24. Cortical expression of presynaptic proteins from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally. A. Upper, immunoblots showing molecular weights and protein levels of Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months. Lower, normalized quantification of protein levels for each presynaptic marker. B. Upper, immunoblots showing molecular weights and protein levels of Hsc70, Syt2, Syt1, CSPα, SNAP25, Syb2 and PV from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 8 months. Lower, normalized quantification of protein levels for each presynaptic marker. A significant decrease in CSPα level is detected in mutant mice when compared to controls at both ages. Besides, significant reductions in Hsc70 and Syt2 are found in mutant mice at 8 months. β‐actin is used as loading control. Error bars indicate SEM and statistical significance is determined by Student’s t‐test and Mann‐Whitney U test. *p=0.05‐0.01; **p=0.01‐0.001; ***p=0.001‐0.0001. Protein levels are normalized against PVcre:Ai27D:Dnajc5flox/+ animals.
86 Results
Nevertheless, we found high variability when analyzing each littermate pair separately at 2 months, revealing significant decreases in both PV+ and Syt2+ puncta in the first mouse pair (PV: **p=0.0011, Student’s t‐test; Syt2: **p=0.0089, Student’s t‐test; n=30 fields for PV+ and Syt2+ labelling from both control and mutant mice), no significant changes in the second mouse pair (PV: p=0.4719, ns, Student’s t‐test; Syt2: p=0.2954, ns, Student’s t‐test; n=30 fields for both PV+ and Syt2+ labelling and n=33 fields for both makers from the control and the mutant, respectively), and only a significant PV+ puncta reduction in the third pair of mice (PV: *p=0.0122, Student’s t‐test; Syt2: p=0.2547, ns, Student’s t‐test; n=22 fields for PV+ labelling and n=16 for Syt2+ labelling from the control, and n=16 fields for both PV+ and Syt2+ labelling from the mutant) (Table 13; Figure 25C).
2 months Genotype PV+ puncta #/100µm Syt2+ puncta #/100µm Dnajc5flox/+ 79.02 ± 2.55 47.44 ± 1.86 Mixed pairs Dnajc5flox/‐ 70.23 ± 2.35 40.95 ± 1.73 Dnajc5flox/+ 81.32 ± 3.39 45.03 ± 2.27 1st pair Dnajc5flox/‐ 66.86 ± 2.49 36.30 ± 2.29 Dnajc5flox/+ 86.75 ± 5.24 49.79 ± 3.77 2nd pair Dnajc5flox/‐ 82.30 ± 3.40 44.78 ± 2.96 3rd pair Dnajc5flox/+ 65.36 ± 2.66 47.57 ± 3.25 Dnajc5flox/‐ 51.65 ± 4.88 41.77 ± 3.79 Table 13. Summary of PV+ and Syt2+ synaptic puncta density in the experimental mice at 2 months. Mean values ± SEM of PV+ and Syt2+ synaptic puncta density in layer II/III cortical slices for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months old. Values are rounded using two decimals. Referred to Fig. 25B, C.
The variability detected between mouse pairs at the same postnatal age (2 months) suggested that loss of PV+ and Syt2+ synaptic puncta may be progressive, with differential evolution between mice across time. For that reason, we also quantified PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 8 months, when the phenotype should be more evident. In old mice, we found significant decreases in both PV+ and Syt2+ synaptic puncta in mutant mice compare to controls (PV: ****p<0.0001, Student’s t‐test; Syt2: ****p<0.0001, Student’s t‐test) when mixing all mouse pairs (n=67 fields for both PV+ and Syt2+ labelling from 3 control mice; n=72 fields for both PV+ and Syt2+ labelling from 3 mutant mice) (Table 28; Figure 25D). When we analyzed each mouse pair separately, the results revealed significant decreases in both PV+ and Syt2+ puncta in the second mouse pair (PV: ***p=0.0002, Student’s t‐test; Syt2: ***p=0.0017, Mann‐Whitney U test; n=22 fields for PV+ and n=26 fields for Syt2+ labelling from both control and mutant mice) and the third mouse pair (PV: ****p<0.0001, Student’s t‐test; Syt2: ***p=0.0002, Man‐Whitney U test; n=30 fields for both PV+ and Syt2+ labelling and n=33 fields for both makers, from controls and mutants, respectively), but no significant differences were found in the first mouse pair, although there was a clear decreasing tendency (PV: p=0.06, ns, Student’s t‐test; Syt2: p=0.1145, ns,
87
Results
Student’s t‐test; n=22 fields for PV+ labelling and n=16 for Syt2+ labelling from the control, and n=16 fields for both PV+ and Syt2+ labelling from the mutant) (Table 14; Figure 25E).
PV interneurons exert a strong control of spike pattern generation in pyramidal cells through axo‐ somatic synapses. Importantly, we detected a remarkable disorganization in perisomatic labelling with both PV+ and Syt2+ stainings in all PVcre:Ai27D:Dnajc5flox/‐ mice analyzed at 2 and 8 months (n=3 mice per genotype and age). Intriguingly, large spreading structures of unknown origin, better detected with PV+ labelling, were also observed in all PVcre:Ai27D:Dnajc5flox/‐ mice compared to controls at both ages (white boxes in Figure 25A).
These results demonstrated that the specific deletion of Dnajc5 in PV interneurons may result in progressive, age‐dependent alterations of synaptic terminals, which may become substantial across time depending on the individual.
8 months Genotype PV+ puncta #/100µm Syt2+ puncta #/100µm Dnajc5flox/+ 100.90 ± 3.27 52.44 ± 2.76 Mixed pairs Dnajc5flox/‐ 78.15 ± 1.81 36.79 ± 1.45 Dnajc5flox/+ 90.52 ± 6.11 45.01 ± 4.16 1st pair Dnajc5flox/‐ 75.89 ± 3.96 36.39 ± 3.11 Dnajc5flox/+ 100.20 ± 4.01 51.91 ± 3.75 2nd pair Dnajc5flox/‐ 80.36 ± 2.90 38.30 ± 1.79 3rd pair Dnajc5flox/+ 116.10 ± 4.98 63.23 ± 5.97 Dnajc5flox/‐ 77.82 ± 2.62 35.53 ± 2.74 Table 14. Summary of PV+ and Syt2+ synaptic puncta density in the experimental mice at 8 months. Mean values ± SEM of PV+ and Syt2+ synaptic puncta density in layer II/III cortical slices for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 8 months old. Values are rounded using two decimals. Referred to Fig. 25D, E.
______
Figure 25. Quantification of PV+ and Syt2+ synaptic puncta from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally. A. Right, confocal images stained with anti‐synaptotagmin‐2 and anti‐parvalbumin antibodies, showing a qualitative decrease of PV+ and Syt2+ labelling in motor cortical layer II/III from PVcre:Ai27D:Dnajc5flox/‐ mice at both 2 and 8 months old. Note also a remarkable disorganization of perisomatic PV+ and Syt2+ labelling and the appearance of large spreading structures (putative axons) only detected in mutant animals at both ages (dashed boxes). Left, immunohistochemistry images showing in detail the reduction and the perisomatic disorganization in Syt2+ labelling in PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months. B. Bar plots displaying the quantification of PV+ and Syt2+ synaptic puncta at 2 months after grouping the three mouse pairs. It shows statistically significant decreases of PV+ and Syt2+ puncta in mutants compared to controls. C. Bar plots showing the quantification of PV+ and Syt2+ synaptic puncta at 2 months, distinguishing by mouse pairs. Note that there is variability within mouse pairs. D. Bar plots displaying the quantification of PV+ and Syt2+ synaptic puncta at 8 months after grouping the three mouse pairs. Significant decreases are observed in both PV+ and Syt2+ puncta in mutants compared to controls. E. Bar plots showing the quantification of PV+ and Syt2+ synaptic puncta at 8 months, distinguishing by mouse pairs. Note that there is also variability within mouse pairs. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. *p=0.05‐0.01, **p=0.01‐0.001, ***p=0.001‐0.0001, ****p<0.0001. n refers to the number of quantified fields.
88
Results
89 Results
7.2. The progressive degeneration of presynaptic terminals in PVcre:Ai27D:Dnajc5flox/‐ mice is not related to changes in the number of PV cells or soma size After the global screening in expression patterns of presynaptic proteins, we directed our attention specifically into PV neurons, to figure out if the degenerative phenotype is only restricted to synapses or if, in addition, it is accompanied by somatic alterations, i.e., in the number or size of PV interneurons. We used the immunohistochemistry images labelled with parvalbumin antibodies to quantify PV somata number in motor cortical layer II/III (n=27 fields from a total of 3 mice per genotype and age). Unexpectedly, we found an increased number of PV cells in PVcre:Ai27D:Dnajc5flox/‐ mice at 2 months compared to controls (*p=0.0216, Student’s t‐test), but most importantly, no changes were observed at 8 months between genotypes (p=0.5583, ns, Mann‐Whitney U test) (Table 15; Figure 26A). That suggested that neurodegeneration was affecting just the presynaptic terminals, not reaching the PV somata. Next, to test if there were morphological alterations involving PV somata size, the area of PV somas was measured using confocal images stained with anti‐parvalbumin antibodies (n=104 PV somata from 19 images collected from 3 control mice and n=181 PV somata from 19 images collected from 3 mutant mice at 2 months; n=152 PV somata from 21 images collected from 3 control mice and n=142 PV somata from 21 images collected from 3 mutant mice at 8 months). We found no changes in PV somata size between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ mice, neither at 2 months (p=0.6344, ns, Mann‐Whitney U test) nor at 8 months (p=0.5472, ns, Mann‐Whitney U test) (Table 15; Figure 26B).
Age Genotype # PV somas/mm2 PV soma size (µm2) Dnajc5flox/+ 215.70 ± 12.74 696.20 ± 24.07 2 months Dnajc5flox/‐ 262 ± 14.82 672.10 ± 16.90 Dnajc5flox/+ 196.30 ± 13.26 892.50 ± 40.38 8 months Dnajc5flox/‐ 191.70 ± 16.12 915.50 ± 40.51 Table 15. Summary of the quantitative analysis regarding PV cell number and soma size in the experimental mice. Mean values ± SEM indicating the number of PV somas and PV soma size in layer II/III cortical slices for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months old. Values are rounded using two decimals. Referred to Fig. 26.
______
Figure 26. Quantification of the number and soma size of PV interneurons from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months postnatally. A. Bar plots showing the number of PV somas per mm2 measured at 2 (left) and 8 (right) months, using immunohistochemistry images of motor cortical layer II/III labelled with anti‐parvalbumin antibodies in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals. B. Bar plots displaying the quantification of PV soma size (µm2) at 2 (left) and 8 (right) months, using confocal images of immunofluorescence of motor cortical layer II/III labelled with anti‐parvalbumin antibodies from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals. Error bars indicate SEM. Statistical significance is obtained using Student’s t‐test and Mann‐Whitney U test. *p=0.05‐0.01.
90
Results
8. Electrophysiological analysis of PV interneurons shows similar intrinsic properties but changes in spontaneous synaptic responses in PVcre:Ai27D:Dnajc5flox/‐ animals
Our previous results have shown a significant loss of PV+ and Syt2+ synaptic puncta that may suggest a degeneration of parvalbumin presynaptic terminals in vivo. To investigate the functional meaning of these results, we aimed to study the intrinsic properties of this type of neurons, as well as their inhibitory effect onto excitatory neurons through electrophysiological recordings in motor cortical layer II/III neurons, using brain slices from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at both 2 and 8 months postnatal age.
The electrophysiological experiments have been entirely carried out by Dr. Pablo García‐Junco‐ Clemente and Dr. José Luis Nieto‐González.
8.1. Intrinsic properties of parvalbumin interneurons To determine functional intrinsic properties of cortical parvalbumin interneurons, both passive and active properties were analyzed.
91
Results
8.1.1. Passive properties of PV cells The study of the passive properties of neurons informs about how transient changes in membrane potential dissipate as they propagate in space and time (local potential). According to this, passive properties of PV cells were examined. Resting membrane potential (RMP) was similar in both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at 2 and 8 months (2 months: p=0.4449, ns, Mann‐Whitney U test; 8 months: p=0.2175, ns, Student’s t‐test), even when we compared between controls and mutants at different ages (controls: p=0.6681, ns, Student’s t‐test; mutants: p=0.9715, ns, Mann‐Whitney U test). After that, we calculated the time constant (tau, τ) applying single 50 pA hyperpolarizing current injection to determine the velocity of the membrane potential change (see Materials and Methods section). We observed no differences between control and mutant animals neither at 2 and 8 months of age (2 months: p=0.1385, ns, Student’s t‐test; 8 months: p=0.0764, ns, Student’s t‐ test), nor when comparing controls and mutants at different ages (controls: p=0.0641, ns, Student’s t‐test; mutants: p=0.0779, ns, Student’s t‐test) (Figure 27A). Next, we applied both hyperpolarizing and subthreshold depolarizing current injections using 10 pA steps (see Materials and Methods section) to determine the input resistance of the PV cells (Figure 27B). Intensity‐voltage plots were represented for each neuron, and input resistance was calculated according to Ohm’s law: R=V/I. No changes in PV input resistance was observed at 2 months between controls and mutants (p=0.4075, ns, Mann‐Whitney U test); however, PV cells from 8‐month PVcre:Ai27D:Dnajc5flox/‐ mice underwent a significant reduction (*p=0.0463, Mann‐Whitney U test) of input resistance compared to PVcre:Ai27D:Dnajc5flox/+ mice (Figure 27C). After comparing controls and mutants at different ages, we could not observe any variation in input resistance (controls: p=0.25, ns, Mann‐Whitney U test; mutants: p=0.6554, ns, Mann‐Whitney U test). Moreover, the capacitance values of PV cells (time constant/input resistance) did not display variations between genotypes or age (2 months: p=0.6483, ns, Mann‐Whitney U test; 8 months: p=0.4914, ns, Student’s t‐test; controls: p=0.8777, ns, Student’s t‐test; mutants: p=0.8402, ns, Mann‐Whitney U test) (Figure 27D). Mean values for passive properties of PV cells are shown in Table 16 (2 months: n=30 control PV cellsd an n=43 mutant PV cells; 8 months: n=35 control PV cells and n=34 mutant PV cells).
Time constant Input resistance Age Genotype RMP (mV) Capacitance (pF) (ms) (MΩ) Dnajc5flox/+ ‐71.14 ± 1.50 5.03 ± 0.21 144.50 ± 12.74 39.49 ± 2.31 2 months Dnajc5flox/‐ ‐70.21 ± 0.88 4.66 ± 0.15 125.20 ± 7.46 41.83 ± 2.25 Dnajc5flox/+ ‐71.88 ± 0.68 5.55 ± 0.19 151 ± 9.00 39.97 ± 2.1 8 months Dnajc5flox/‐ ‐70.42 ± 0.95 5.08 ± 0.19 126.60 ± 6.07 41.90 ± 1.82 Table 16. Summary of the passive properties of PV cells in the experimental mice. Mean values ± SEM of resting membrane potential (RMP), time constant, input resistance and capacitance in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells from motor cortical layer II/III at 2 and 8 months old. Values are rounded using two decimals. Referred to Fig. 27.
92 Results
8.1.2. Active properties of PV cells An active neuronal signal involves a change in membrane potential that is maintained over a long distance (action potential). According to this, the study of the action potential (AP) kinetic can provide essential evidence about how neurons propagate the electrical information. Based on this premise, several parameters were measured from AP of PV cells: amplitude, rise‐time, half‐width and after hyperpolarization (AHP) (Figure 28A), which will be explained in detail during the analysis.
First, we analyzed the AP parameters for both genotypes and ages. For that purpose, we applied the minimum current injection (see Materials and Methods section) to evoke a single action potential in every cell, and afterwards we calculated an average from all APs produced within all PV cells. AP threshold showed no variations between genotypes or age (2 months: p=0.4408, ns, Student’s t‐test; 8 months: p=0.7813, ns, Student’s t‐test; controls: p=0.9672, ns, Student’s t‐test; mutants: p=0.3288, ns, Student’s t‐ test). After that, AP amplitude was calculated from threshold peak (Figure 28A). We could not observe any difference in AP amplitude between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes, neither at 2 nor 8 months (2 months: p=0.4176, ns, Student’s t‐test; 8 months: p=0.9869, ns, Student’s t‐test). Interestingly, the AP amplitude was decreased at 8 months compared to 2 months, showing significant reductions when comparing both controls (**p=0.0011, Student’s t‐test) and both mutants (****p<0.0001, Student’s t‐test) (Figure 28B). Regarding AP rise‐time and half‐width, both parameters were very similar in all analyzed cases, although showing significant changes only between controls at the two different ages (AP rise‐time: **p=0.0074, Mann‐Whitney U test; AP half‐width: **p=0.0034, Mann‐ Whitney U test). Mean values for AP parameters are shown in Table 17 (2 months: n=30 control PV cells and n=43 mutant PV cells; 8 months: n=35 control PV cells and n=34 mutant PV cells).
AP threshold AP amplitude AP rise‐time AP half‐width Age Genotype (mV) (mV) (ms) (ms) Dnajc5flox/+ ‐33.91 ± 1.16 85.33 ± 2.67 0.086 ± 0.003 0.149 ± 0.005 2 months Dnajc5flox/‐ ‐35.63 ± 1.67 87.81 ± 1.72 0.091 ± 0.003 0.158 ± 0.006 Dnajc5flox/+ ‐33.97 ± 1.04 74.21 ± 1.94 0.095 ± 0.003 0.174 ± 0.006 8 months Dnajc5flox/‐ ‐33.55 ± 1.10 74.25 ± 1.92 0.097 ± 0.003 0.162 ± 0.005 Table 17. Summary of AP properties of PV cells in the experimental mice. Mean values ± SEM of AP threshold, amplitude, rise‐time and half‐width in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells from motor cortical layer II/III at 2 and 8 months old. Values are rounded using two or three decimals. Referred to Fig. 28B.
The afterhyperpolarization phase (AHP) of a neuron’s action potential provides information about the membrane potential fall below the RMP during the repolarization step (also known as undershoot phase) (Figure 28A). For each PV neuron, the AHP amplitude was calculated from RMP to hyperpolarization peak of an action potential, and the AHP decay was measured using a normalized AP for every parvalbumin
93
Results
94
Results interneuron. We observed no changes in AHP amplitude between control and mutant PV cells neither at 2 nor 8 months (2 months: p=0.2913, ns, Student’s t‐test; 8 months: p=0.4425, ns, Student’s t‐test), or even when comparing between controls and mutants at different ages (controls: p=0.4305, ns, Student’s t‐test; mutants: p=0.788, ns, Student’s t‐test). Nevertheless, it was detected a significant decrease in AHP decay (calculated by a double exponential fitting, using a weighted tau) at 8 months in parvalbumin cells from PVcre:Ai27D:Dnajc5flox/‐ mice compared to controls (*p=0.0149, Mann‐Whitney U test), although a decreasing tendency at 2 months (p=0.8265, ns, Mann‐Whitney U test) was also evident. Based on this, mutant PV cells repolarize faster than control cells, thus they might be preferably ready to fire APs again (Figure 28C). Indeed, when comparing PVcre:Ai27D:Dnajc5flox/‐ PV cells at both ages, we found a significant decrease at 8 month compared to 2 months (**p=0.0085, Mann‐Whitney U test). Control neurons showed similar AHP weighted tau (p=0.9954, ns, Mann‐Whitney U test). Mean values for AHP parameters are displayed in Table 18 (2 months: n=30 control PV cells and n=43 mutant PV cells; 8 months: n=35 control PV cells and n=34 mutant PV cells).
Age Genotype AHP amplitude (mV) AHP weighted tau (ms) Dnajc5flox/+ 23.68 ± 0.75 14.76 ± 5.56 2 months Dnajc5flox/‐ 22.49 ± 0.78 12.22 ± 4.07 Dnajc5flox/+ 22.94 ± 0.58 12.56 ± 2.63 8 months Dnajc5flox/‐ 22.18 ± 0.80 4.99 ± 0.91 Table 18. Summary of AHP properties of PV cells in the experimental mice. Mean values ± SEM of AHP amplitude and AHP weighted tau in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells from motor cortical layer II/III at 2 and 8 months old. Values are rounded using two decimals. Referred to Fig. 28C.
Finally, studied the excitability of PV cells, that is, their capacity to generate action potentials. We followed a step protocol consisting of 20 pA depolarizing current injections (see Materials and Methods section) to induce action potentials, and number of evoked APs were measured at each current pulse ______
Figure 27. Passive electrophysiological properties of layer II/III parvalbumin interneurons in motor cortical slices. A. Left, representative change of the cell membrane potential during hyperpolarizing current injections in current‐clamp experiments of a PVcre:Ai27D:Dnajc5flox/+ mouse. The exponential fitting (red line) to the mean membrane potential is used to calculate the time constant. Right, bar plots from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice showing similar membrane time constants at 2 and 8 months old. B. Left, current step protocol to calculate input resistance using the current‐clamp configuration. Right, representative example of membrane potential changes during the application of the previous protocol in a PVcre:Ai27D:Dnajc5flox/+ mouse. C. Left, example of an intensity‐voltage (I‐V) plot in a PVcre:Ai27D:Dnajc5flox/+ genotype, where linear fitting (dashed line) is used to calculate the input resistance according to Ohm’s law. Right, bar plots showing the input resistance quantification for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at 2 and 8 months old. D. Bar plots showing the capacitance values of PV interneurons for both genotypes at the two postnatal ages. No changes in capacitance are detected. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. *p=0.05‐0.01.
95 Results
(Figure 29A). We found similar firing patterns in control PV cells at both ages (Figure 29C). However, when compared to their respective mutant cells, a slightly curve displacement was observed at 8 months in PVcre:Ai27D:Dnajc5flox/‐ cells, showing a reduction in APs number at each current injection, while no variations seemed to happen at 2 months, with the exception that mutant PV cells seemed to saturate first than control neurons (accommodation) (Figure 29B). Interestingly, when we compared mutant PV cells at the two ages, we found that 2 month‐old PV interneurons fired an increasing number of APs per injected current step, but 8 month‐old PV interneurons fired a single AP after the same injected current, although they finally reached a similar firing pattern after strong depolarizing steps (Figure 29C). These observations were confirmed by the rheobase parameter, which describes the minimal current injection needed to generate at least one action potential with 50% likelihood. Figure 29D showed a significant increase in rheobase of parvalbumin cells from PVcre:Ai27D:Dnajc5flox/‐ animals compared to control cells at 8 months (*p=0.0489, Mann‐Whitney U test), indicating they needed a stronger depolarizing current to fire the first AP, so they were less excitable. No changes were found in rheobase neither at 2 months (p=0.4805, ns, Mann‐Whitney U test) nor when comparing control or mutant cells at different ages (controls: p=0.7156, ns, Student’s t‐test; mutants: p=0.1081, ns, Mann‐Whitney U test). Mean values for rheobase are shown in Table 19 (2 months: n=30 control PV cells and n=43 mutant PV cells; 8 months: n=35 control PV cells and n=34 mutant PV cells).
Age Genotype Rheobase (pA) Dnajc5flox/+ 219.60 ± 19.23 2 months Dnajc5flox/‐ 254.50 ± 23.67 Dnajc5flox/+ 228.90 ± 17.06 8 months Dnajc5flox/‐ 288.50 ± 21.15 Table 19. Summary of rheobase values of PV cells in the experimental mice. Mean values ± SEM of rheobase for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells at 2 and 8 months old. Values are rounded using two decimals. Referred to Fig. 29D.
______
Figure 28. Active electrophysiological properties of layer II/III parvalbumin interneurons in motor cortical slices. A. Representative trace of a PV action potential indicating all the electrophysiological measurements. B. Left, averaged action potential obtained for both for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes, at 2 and 8 months old. Right, bar plots showing mean AP amplitude of PV interneurons for both genotypes at 2 and 8 months old. No differences between genotypes are found; a time‐dependent decrease in amplitude is observed in both genotypes. C. Upper, afterhyperpolarization (AHP) trace obtained after normalizing the averaged AP for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes, at 2 and 8 months old. Lower left, bar plots indicating mean AHP amplitude of PV neurons for both genotypes, at 2 and 8 months old. No differences in AHP amplitude are detected. Lower right, bar plots showing weighted tau of PV cells for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals at 2 and 8 months. Statistically significant differences are found at 8 months in mutant mice. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. *p=0.05‐0.01, **p=0.01‐0.001, ****p<0.0001.
96
Results
97
Results
98 Results
8.1.3. Excitatory postsynaptic potentials (EPSPs) Excitatory postsynaptic potentials (EPSPs) are depolarizing local potentials that make the postsynaptic neuron more likely to fire an action potential. This temporary depolarization of the postsynaptic membrane is a result of opening glutamate‐gated ion channels. The EPSPs were recorded to study the excitatory inputs onto the cortical PV cells, so current‐clamp recordings were performed in absence of injected current (at RMP). No spontaneous APs were detected in PV cells from neither control nor mutant mice at any age, but we did observe EPSPs in both genotypes and ages (Figure 30A). For each PV cell, we recorded the spontaneous EPSPs during 30 seconds. We calculated the frequency of events per cell, and then we averaged all EPSPs to obtain a template to calculate the amplitude and kinetic of the EPSPs per genotype and age (see Materials and Methods section). We could not detect differences in EPSP frequency between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes neither at 2 nor 8 months (2 months: p=0.3965, ns, Mann‐Whitney U test; 8 months: p=0.1788, ns, Mann‐Whitney U test), but it was observed a significant age‐dependent frequency increase both in controls and mutants (****p<0.0001 for both analysis, Mann‐Whitney U test). Regarding EPSP amplitude, no changes were detected in PVcre:Ai27D:Dnajc5flox/‐ mice neither at 2 or 8 months (2 months: p=0.3809, ns, Student’s t‐test; 8 months: p=0.1772, ns, Mann‐Whitney U test) nor when comparing between controls and mutants at different ages (controls: p=0.8286, ns, Mann‐Whitney U test; mutants: p=0.8608, ns, Mann‐Whitney U test) (Figure 30B). EPSP rise‐time and half‐width displayed similar values for both genotypes and ages (EPSP rise‐time at 2 months: p=0.2523, ns, Student’s t‐test; 8 months: p=0.8721, ns, Mann‐Whitney U test; controls: p=0.0930, ns, Mann‐Whitney U test; mutants: p=0.2635, ns, Mann‐Whitney U test; EPSP half‐width at 2 months: p=0.4894, ns, Mann‐Whitney U test; 8 months: p=0.2476, ns, Student’s t‐test; controls: p=0.2616, Mann‐ Whitney U test; mutants: p=0.5364, ns, Student’s t‐test). Averaged values for EPSP properties are shown in Table 20 (2 months: n=28 control PV cells and n=37 mutant PV cells; 8 months: n=35 control PV cells and n=34 mutant PV cells).
______
Figure 29. Excitability properties of layer II/III parvalbumin interneurons in motor cortical slices. A. Scheme of the depolarizing current step protocol (gray trace) and AP firing response (black trace) of PV cells. The dashed red boxes mark magnifications of AP responses to specific current steps. B. Number of action potentials produced following injection of depolarizing current steps, comparing PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells at 2 (left) and 8 (right) months. C. Number of action potentials produced following injection of depolarizing current steps, comparing control cells at 2 and 8 months (left) and mutant cells at 2 and 8 months (right). D. Rheobase values of individual PV cells and mean showing the minimal current injection needed to generate at least one action potential for both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at 2 and 8 months. A statistically significant increase is detected in mutant mice at 8 months compared to controls. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. *p=0.05‐0.01.
99
Results
EPSP amplitude EPSP frequency EPSP rise‐time EPSP half‐width Age Genotype (mV) (Hz) (ms) (ms) Dnajc5flox/+ 0.48 ± 0.02 25.88 ± 0.87 0.348 ± 0.01 3.308 ± 0.21 2 months Dnajc5flox/‐ 0.45 ± 0.02 26.40 ± 0.61 0.365 ± 0.01 3.396 ± 0.15 Dnajc5flox/+ 0.48 ± 0.03 30.62 ± 0.53 0.379 ± 0.01 3.497 ± 0.14 8 months Dnajc5flox/‐ 0.43 ± 0.03 29.96 ± 0.24 0.375 ± 0.01 3.271 ± 0.13 Table 20. Summary of EPSP properties of PV cells in the experimental mice. Mean values ± SEM of EPSP amplitude, frequency, rise‐time and half‐width in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV cells from motor cortical layer II/III at 2 and 8 months old. Values are rounded using two or three decimals. Referred to Fig. 30.
8.2. Miniature inhibitory postsynaptic currents (mIPSCs) on cortical pyramidal cells The possible presynaptic degeneration of PV terminals shown in previous experiments predicts a functional defect in the inhibitory input coming from PV cells, which control the spike pattern of the pyramidal cells through axo‐somatic inhibition. We performed voltage‐clamp electrophysiological recordings of miniature inhibitory postsynaptic currents (mIPSCs) on cortical pyramidal cells, using tetrodotoxin (TTX) to avoid AP generation in any cell. Additionally, kynurenic acid was added to block ionotropic glutamate receptors, ensuring that all postsynaptic currents detected were entirely GABAergic. As spontaneous events, it is assumed that any mIPSC generated will come from a single vesicle fusion, each of them presumably having the same neurotransmitter content. For this analysis, every individual event was extracted from each recording, and amplitude histograms were obtained by genotype and age. At the same time, averaged mIPSCs values per cell were used to calculate several parameters by genotype and age (see Materials and Methods section) (Figure 31). At 2 months, mIPSC amplitude was significantly decreased in the PVcre:Ai27D:Dnajc5flox/‐ genotype compared to the control one (****p<0.0001, Mann‐ Whitney U test), as well as the mIPSC frequency (****p<0.0001, Student’s t‐test). In contrast, we found a significant increase in the mIPSC rise‐time (**p=0.0068, Mann‐Whitney U test), decay (***p=0.0001, Student’s t‐test) and half‐width (****p<0.0001, Mann‐Whitney U test) in the PVcre:Ai27D:Dnajc5flox/‐ genotype compared to the control one (Figure 31A).
Interestingly, at 8 months, results reproduced the same tendency observed at 2 months but slightly aggravated. mIPSC amplitude (****p<0.0001, Mann‐Whitney U test) and frequency (****p<0.0001, Mann‐Whitney U test) were significantly reduced in the mutant genotype compared to the control one, while half‐width (****p<0.0001, Student’s t‐test) and decay (****p<0.0001, Student’s t‐ test) were both significantly increased. Rise‐time was not significantly changed at 8 months (p=0.0717, ns, Mann‐Whitney U test), although an increased tendency in mutant neurons persisted as at 2 months (Figure 31B).
100
Results
Figure 30. Excitatory postsynaptic potentials (EPSPs) of layer II/III parvalbumin interneurons in motor cortical slices. A. Representative current‐clamp recordings showing excitatory postsynaptic potentials (EPSPs) of PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at 2 and 8 months old; the dashed boxes mark magnified regions in both recordings. B. Left, averaged EPSP obtained for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes at 2 and 8 months old. Right upper, bar plots showing mean EPSP frequency in PV neurons for both genotypes at 2 and 8 months. No differences between genotypes are found; a time‐ dependent increase in frequency is observed in both genotypes. Right lower, bar plots showing mean EPSP amplitude in PV cells from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ animals at 2 and 8 months. No changes in amplitude are detected. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. ****p<0.0001.
101
Results
Finally, when comparing controls and mutants genotypes at different ages, no significant differences were observed in mIPSC amplitude (controls: p=0.6014, ns, Mann‐Whitney U test; mutants: p=0.8612, ns, Mann‐Whitney U test), rise‐time (controls: p=0.5498, ns, Mann‐Whitney U test; mutants: p=0.5518, ns, Mann‐Whitney U test), half‐width (controls: p=0.5391, ns, Student’s t‐test; mutants: p=0.7196, ns, Mann‐Whitney U test) or decay (controls: p=0.8178, ns, Student’s t‐test; mutants: p=0.9874, ns, Student’s t‐test). In contrast, mIPSC frequency was significantly reduced at 8 months when comparing both control and mutant genotypes (controls: *p=0.0269, Mann‐Whitney U test; mutants: ***p=0.0001, Student’s t‐test). Mean mIPSC values are shown in Table 21 (2 months: n=32 control PV cells and n=35 mutant PV cells; 8 months: n=31 control PV cells and n=39 mutant PV cells).
Age Genotype mIPSC amplitude (pA) mIPSC frequency (Hz) mIPSC rise‐time (ms) Dnajc5flox/+ ‐38.03 ± 2.33 36.97 ± 1.38 0.30 ± 0.009 2 months Dnajc5flox/‐ ‐24.53 ± 1.35 26.40 ± 1.24 0.32 ± 0.008 Dnajc5flox/+ ‐35.73 ± 1.32 33.17 ± 1.32 0.30 ± 0.005 8 months Dnajc5flox/‐ ‐24.47 ± 1.13 19.14 ± 1.26 0.32 ± 0.01 Age Genotype mIPSC half‐width (ms) mIPSC weighted tau (ms) Dnajc5flox/+ 2.28 ± 0.11 3.35 ± 0.15 2 months Dnajc5flox/‐ 3.03 ± 0.15 4.24 ± 0.16 Dnajc5flox/+ 2.36 ± 0.09 3.40 ± 0.13 8 months Dnajc5flox/‐ 3.13 ± 0.15 4.23 ± 0.15 Table 21. Summary of mIPSC properties of pyramidal cells in the experimental mice. Mean values ± SEM of mIPSC amplitude, frequency, rise‐time, half‐width and weighted tau in PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ pyramidal cells from motor cortical layer II/III at 2 and 8 months old. Values are rounded using two or three decimals. Referred to Fig. 31.
______
Figure 31. Miniature inhibitory postsynaptic currents (mIPSCs) of layer II/III pyramidal cells in motor cortical slices. A. Upper, representative voltage‐clamp recordings showing miniature inhibitory postsynaptic currents (mIPSCs) in PVcre:Ai27D:Dnajc5flox/+ (left) and PVcre:Ai27D:Dnajc5flox/‐ (right) pyramidal cells at 2 months; the dashed boxes mark magnified regions in both recordings. Middle, averaged mIPSC traces (left), histograms and cumulative distributions (center) and bar plots representing mIPSC amplitude and frequency (right) from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ pyramidal cells at 2 months. There is a significant reduction in mIPSC frequency and size in mutant mice. Lower (from left to right), bar plots showing mean values for rise‐time, half‐width, normalized mIPSCs and decay from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ pyramidal cells at 2 months. mIPSC kinetics is significantly affected in mutant mice. B. The same description as in A but for PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ pyramidal cells from animals at 8 months. Again, significant differences in mIPSC amplitude, frequency and kinetics are detected in mutant mice. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. **p=0.01‐0.001, ***p=0.001‐0.0001, ****p<0.0001.
102
Results
103
Results
9. Single‐cell RNA sequencing of cortical parvalbumin interneurons The beginning of this work coincided with a peak of publications related to single‐cell RNA sequencing, a novel technique that allows for a complete transcriptome analysis in single cells using high‐ throughput sequencing technology (Grün and Van Oudenaarden, 2015; Islam et al., 2014; Luo et al., 2015; Marques et al., 2016; Qiu et al., 2012; Tasic et al., 2016; Zeisel et al., 2015). One of our main goals during this thesis was to examine in depth potential changes in gene expression that could shed light on the molecular mechanisms involved in the presynaptic neurodegeneration in absence of Dnajc5, likely triggering the neurological phenotype.
The scRNA‐seq experiments were carried out in collaboration with Dra. Ana Belén Muñoz Manchado (Prof. Jens Hjerling‐Leffler, Karolinska Institute). We performed series of programmed single‐ cell RNA sequencing experiments in PVcre:Ai27D:Dnajc5flox/+ (n=4) and PVcre:Ai27D:Dnajc5flox/‐ mice (n=5), including also PVcre:Ai27D:Dnajc5WT (n=3) animals as an extra control (BOX 1).
BOX 1
We run the protocol on 3 mice per day (4 days in total) to minimize the cell death due to prolonged experimental periods, and mixing one mouse of each genotype per day (when possible) to avoid the batch effect. It took around 6‐7 hours per day to carry out the whole experiment, starting with the dissociation protocol, followed by FACS isolation of tdTomato+ cells and the delivery of the 9600‐microwell array platform to the Eukaryotic Single Cell Genomics Facility (ESCG at SciLifeLab). Details of dissociation, FACS and RNA sequencing protocols are described in Materials and Methods section.
FACS was chosen to select the specific subpopulation of cells fluorescently labelled (tdTomato+ cells). Cells were directly sorted into the WaferGen plate instead of using the WaferGen cell dispenser, a faster procedure to avoid cell stress. In our case, since the PV population scarce regarding the whole cortex, we obtained percentages of tdTomato+ cells by FACS that ranged from 0.1 to 0.5% (out of 50000 events of total cortical cells) during all experimental days, which made it difficult to gain velocity and to reduce time during sorting (Figure 32). These numbers were similar to the percentages obtained in our FACS facility (Figure 17A). After FACS and before the sequencing, two cDNA quality controls per chip were performed (BOX 2).
104
Results
Figure 32. FAC‐sorting settings for neuron isolation during single‐cell RNA sequencing experiments. Representative example of sorting settings and gates using the BD Influx cell sorter. “Cells” indicates the first gate selecting only alive cells; “Singlets” selects single cells by discarding events with high size (potential doublets) using FSC width; “Singlets 2” is employed to re‐select unique cells using FSC area, discarding escaping doublets; “tdT+” represents cells that expressed tdTomato (561 nm laser). Table shows specific parameters obtained for 50000 events from total cells for this representative example. Note that the percentage of tdT+ cells is 0.38% of total and 1.30% of parent, which makes sorting efficiency slow. tdT+ percentage varied from 0.1 to 0.5% of 50000 total cells in all sorting experiments for sequencing using PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. SSC, Side Scatter (internal cell complexity); FSC, Forward Scatter (cell size/volume).
BOX 2
Before sequencing, the genomic facility carried out two cDNA quality controls per chip based on: (1) cDNA obtained from 20‐22 randomly selected subarrays containing pooled wells (25 cells), and (2) cDNA obtained from prepared library of the full plate (see Materials and Methods section). The first control was chosen to make the decision to continue sequencing or not. S2A shows representative examples of purified and amplified cDNA from 20‐22 randomly selected quadrants belonging to every plate. Although we found some samples with low concentration of cDNA – most of them from the WG17002 chip (where we used the aCSF instead of the NMDG‐ Hepes recovery solution and the albumin density gradient) ‐, the majority of samples contained a substantial amount of cDNA and showed a normal histogram distribution in the Bioanalyzer (S2A). The average cDNA yield was around 200pg/μl for the WG17002 plate, 700pg/μl for the WG17003 plate, 1ng/μl for the WG17004 plate, and 1.8ng/μl for the WG17005 plate. STRT‐based sequencing studies showed a typical yield of around 1ng/μl (Muñoz‐Manchado et al., 2018; Zeisel et al., 2015). Besides, the average cDNA size was around 1kb in the four plates. Therefore, we made the decision to continue with RNA sequencing using the four plates. Moreover, immediately before sequencing, the second control was performed on every chip‐derived library containing all 5’ fragments; again, all of them satisfied the quality control and reinforced our decision to run the sequencing (S2B). The average library yield was 1.6ng/μl for the WG17002 plate, 1.7ng/μl for the WG17003 plate, 2.2ng/μl for the WG17004 plate, and 5.3ng/μl for the WG17005 plate; and the average cDNA size was around 500bp in each of the four libraries.
105
Results
9.1. Robust average mapped mRNA reads per cell The first analysis we carried out after receiving our scRNA‐seq data was to calculate the read depth (BOX 3). The average mapped mRNA reads obtained per cell for all chips were similar to that found in Hochgerner et al., 2017 (41000 mapped mRNA reads/cell) (Table 22).
BOX 3
After scRNA‐seq, the ESCG facility provided us with raw data to calculate sequencing/read depth, which defines the average number of reads that align to a known reference genome. Using the number of mapped reads per well, we estimated the average of mapped mRNA reads per cell: we summed the mapped reads along all wells and divided it per total wells sequenced. 2392 wells were sequenced, however, H12‐W01 well position from every plate was discarded before calculation, so in our case we divided by 2391 sequenced wells/cells, according to sorting and sequencing limitations for a WaferGen chip (see Materials and Methods section; Figure 13).
Mapped mRNA reads Mapped mRNA reads/cell WG17002 93140173 38954 WG17003 82832269 34643 WG17004 78612820 32879 WG17005 83241102 34814
Table 22. Mapped reads from scRNA‐seq in WaferGen chips. Table displaying total mapped mRNA reads obtained from each of the four sequenced WaferGen chips (WG17002, WG17003, WG17004, WG17005), and the estimated average mapped mRNA reads per cell (total mapped genes divide by 2391 sequenced cells in each chip) (see BOX 3). Numbers are rounded to units.
10. Analysis of transcriptomic data obtained from single cortical parvalbumin interneurons
During our RNA sequencing data analysis, we created several R scripts including all steps, annotations and explanations of functions and processes used. The scRNA‐seq analysis will be separated in three main sections: the creation of a readable matrix for data analysis, the integration of all the data for clustering cells, and the differential expression analysis between genotype pairs.
10.1. A readable matrix to visualize genes expressed by every single cell Based on the well distributions already mentioned (Figure 13), we created the ReadData.R script (Supp. Data: R scripts and files > Read data) for reading raw data, labelling wells and combining datasets. We combined wells belonging to WT, Flox+ and Flox‐ genotypes from every chip previous to downstream analysis. Briefly, the same process was applied to the four raw datasets to obtain: (1) attributes and gene expression data saved into two different dataframes per chip, (2) a matrix containing well positions to fill in with “genotype” and “animal number” labels (BOX 4), and finally (3) the merge of the four attributes datasets and the four gene expression datasets to generate the files AttributesFinal.csv and ExpressionFinal.csv (Supp. Data: R scripts and files > Read Data). These files were necessary for the next
106
Results steps. ExpressionFinal.csv contained information of a total of 9568 cells (2392 sequenced wells per chip) and 24487 genes (Figure 33A).
As explained in BOX 4, because three different genotypes (see Materials and Methods section) were collected in the chips WG17002, WG17003 and WG17005, a total of 11 wells (+5 from C quadrants, +5 from F quadrants, +1 from HA12‐W01 well) were denoted as “Undetermined” per chip and later discarded. On the other hand, we only sorted two genotypes in the WG17004 array (see Materials and Methods section). This distribution evaded potential contamination after sample changing between the two PVcre:Ai27D:Dnajc5flox/‐ mice and only 6 wells (only +5 from F quadrants, +1 from HA12‐W01) were denoted as “Undetermined” in this chip and later removed. In conclusion, from 2392 sequenced wells per chip, 2381 (2392‐11) remained as valid cells from the WG17002, WG17003 and WG17005 chips, and 2386 (2392‐6) remained as valid cells from the WG17004 chip.
BOX 4
Starting with our four raw datasets from each WaferGen microwell‐array platform (WG17002, WG17003, WG17004 and WG17005), we wrote the ReadData script to read this raw data from files in “cell expression format” (.cef), as provided by the ESCG facility (Supp. Data: R scripts and files > ReadData). This prototype of file contains single cells in columns and genes in rows, but it also has additional information, such as well names, plate names, barcodes, etc. Our purpose was to create a single dataset called ExpressionFinal, where all single cells from each array were localized in columns, and all genes (they were always the same in each dataset) appeared in rows. Extra information from each well was combined in a dataset named AttributesFinal.
As the CEF format has a well‐defined structure (https://github.com/linnarsson‐lab/ceftools), we wrote the script according to known positions relative to the information required (we could also know these positions by opening the .cef file with Excel). Moreover, thanks to the sorting pattern provided by ESCG (Figure 13; Supp. data: ReadData > WG Wells IDs.csv), we were able to know the identity of the mouse contained in each position. This file included very important and useful information to recognize where every single cell came from, thus allowing us to put a “genotype” label (“WT” for PVcre:Ai27D:Dnajc5WT animals, “Flox+” for PVcre:Ai27D:Dnajc5flox/+ animals, or “Flox‐” for PVcre:Ai27D:Dnajc5flox/‐ animals; onwards, WT, Flox+ and Flox‐ due to nomenclature limitations with R language) and an “animal number” label to each well/column. Because our strategy aimed to sort around 800 cells from each mouse per day, it was quite ingenious to write a script distinguishing between adjacent wells that belonged to different animals. In fact, two particular rows from the plate were specially treated: row 4, in row C quadrants, and row 2, in row F quadrants (only in WG17002, WG17003 and WG17005 chips). These two row quadrants contained wells with two different genotypes, due to sample changing at FACS. To avoid potential contamination when changing samples during FACS, we decided to discard five wells around these locations: two wells from the previous sample and three wells from the new sample after changing (Figure 13). We rejected an extra well from the new sample because sorting contamination seemed more likely to happen in this sample. These wells were denoted as “Undetermined” and were not used for downstream analysis. Furthermore, H12‐W01 well was also named as “Undetermined” and removed from the analysis, due to a misunderstanding with the special sorting pattern in H12 quadrant that left this well empty (Figure 13).
10.2. Integration of single‐cell transcriptomic data across different genotypes We tested several packages for the study of our scRNA‐seq data, such as the SCE (Lun et al., 2016) and SCDE (Kharchenko et al., 2014) packages. We finally decided to follow Seurat package instructions for
107
Results an integrated analysis (Butler et al., 2018). This is a novel package designed to (1) explore first two or more datasets by examining data separately (in this case by genotype) followed by (2) powerful data clustering based on combining (integrating) the different datasets. Our integrated dataset, containing a high number of cells upon combining different genotypes were, therefore, especially suitable to yield a fruitful clustering based on integrated analysis. This method is described in Butler et al., 2018 and explained in Satija Lab website (https://satijalab.org/seurat/immune_alignment.html).
10.2.1. Cell filtering: quality control by filtering out data from low quality cells By default, our datasets contained several groups of genes not suitable for our analysis strategy that we eliminated. These genes included fluorescent markers or RNA spike‐in and ERCC (External RNA Controls Consortium) genes (BOX 5). A total of 109 genes were removed, leaving 24378 genes (Figure 33A). In addition, we excluded “Undetermined” wells, which eliminated 39 cells (11+11+11+6) and led to a total of 9529 cells (2381 cells per three chips and 2386 cells per one chip) (Figure 33A). After this initial filtering, we split ExpressionFinal.csv and AttributesFinal.csv by genotype into three separated dataframes (2381 WT cells, 3169 Flox+ cells, 3979 Flox‐ cells) (Figure 33A), which were prepared for individual quality control.
BOX 5
Spike‐in and ERCC features are RNA transcripts of known sequence and quantity that are used as control genes to calibrate sequencing measurements (Lun et al., 2016; Kharchenko et al., 2014). ERCCs are usually used to calculate the fitting curve in some analyses (for example, SCE‐ or SCDE‐based tutorials), however, the Seurat package includes other methods to avoid noise that do not require ERCCs. Instead, Seurat calculates high variable genes (see later 2.Determination of High Variable Genes (HVG) during the clustering steps).
Next, we proceeded with cell filtering, an imperative initial step to remove low‐quality cells in order to ensure that technical noise do not impair the downstream analysis (Butler et al., 2018; Lun et al., 2016; Mayer et al., 2018; Zeisel et al., 2015). R code for cell filtering, normalization and scaling of data is described in Seurat integrated analysis ‐ Cell filtering.R script (Supp. Data: R scripts and files > Seurat integrated analaysis). Common quality control (QC) metrics involve: (1) the number of total UMIs (nUMI), which is defined as the sum of counts (in this case, UMIs) through all genes for a single cell, (2) the number of expressed genes (nGene), which corresponds to the number of genes with non‐zero counts for a single cell, and (3) the percentage of mitochondrial genes detected per single cell (percent.mito). All QC parameters provide a general view of cell quality. Cells with low nUMI or nGene are considered low quality cells, probably because mRNA has not been well captured to be converted to cDNA and amplified, or because the diverse population of transcripts have not been captured equally, respectively. Conversely, high proportions of mitochondrial genes indicate poor cell quality, possibly due to increased apoptosis and/or loss of cytoplasmic RNA upon cell lysis (Lun et al., 2016).
108
Results
First of all, we created a Seurat object (CreateSeuratObject function) of each genotype‐based dataframe, a useful tool to store big data in an organized variable. This function automatically discarded cells that presented a total count sum equal to zero across all genes, which left 2190 WT cells, 2833 Flox+ cells and 3434 Flox‐ cells (Figure 33A). Once the Seurat object was created, we represented distributions of nUMI, nGene and percent.mito for WT, Flox+ and Flox‐ cells, using the VlnPlot function from Seurat to draw violin plots, useful graphics that show probability density. By examining QC metrics with violin plots, we observed that cell density looked preferentially at the bottom in nUMI and nGene plots in all genotypes (Figure 33B). Cell density in percent.mito plots looked quite high in all genotypes, and even some cells reached to 100% (Figure 33B). These three results suggested that our data contained cells with poor quality. To further evaluate QC metrics, we calculated the average number for nUMI, nGene and percentage of mitochondrial genes before filtering; they showed similar results among genotypes (Table 23).
According to recent publications using UMIs (Butler et al., 2018; Hochgerner et al., 2017; Zeisel et al., 2015), we established accurate thresholds for nUMI, nGene and percent.mito to ensure an appropriate cell quality control for our data: (1) cells with more than 1000 detected UMIs, (2) more than 350 detected genes and (3) less than 20% expressed mitochondrial genes were retained using the FilterCells function from Seurat (Figure 33B, C). This step left 946 WT cells, 1470 Flox+ cells and 1600 Flox‐ cells (Figure 33A) and higher new mean values for nUMI, nGene and percent.mito (Table 23; Figure 33C), which were again very similar between genotypes and now comparable to that obtained in Hochgerner et al., 2017 (3686 genes, 8706 molecules).
Before cell filtering After cell filtering nUMI nGene Percent.mito nUMI nGene Percent.mito WT 3122 1106 19 6734 2331 12 Flox+ 3750 1286 15 6872 2323 11 Flox‐ 3565 1141 15 7373 2312 11 Table 23. Summary showing the average numbers for nUMI, nGene and percent.mito per genotype. Mean values of the number of molecules (nUMI), the number of genes (nGene) and the percentage of mitochondrial genes (percent.mito, %) detected before and after cell filtering were calculated by mean function from the base package in R for each genotype. Values are rounded to units.
10.2.2. Normalization and noise reduction of individual genes expression in single cells After cell filtering, the expression of every gene at every single cell needs to be normalized to the total gene expression in every single cell, so that each cell can be compared against each other. NormalizeData is the global normalization function employed in Seurat. It normalizes gene expression for every single cell by its total gene expression, then uses a default factor (10000) to multiply to and makes a log‐transformation of the product. After normalization, data were scaled. Data scaling is an individual gene
109
Results
110
Results
Figure 33. Filtering out low quality cells from scRNA sequencing data. A. Sequential scheme displaying step by step cell selection. In green, the number of cells removed; in red, cells selected during each step (blue) along the bioinformatics study. The first part of the analysis is detailed in the ReadData.R script and the second part, in Cell filtering.R script. B. Violin plots showing number of genes (nGene), number of UMIs (nUMI) and percentage of mitochondrial genes (percent.mito) detected in WT, Flox+ and Flox‐ single cells by scRNA sequencing before cell filtering. Black dots represent single cells. The orange areas indicate distribution and density of the analyzed data, being thickened where most of cells are located. VlnPlot function is used to make the plots. Minimal thresholds for nGene and nUMI (350 and 1000, respectively) and a maximal threshold for percent.mito (20%) are established for cell quality control (red lines); cells with less than 350 nGene and 1000 nUMI, and with more than 20% percent.mito are discarded as indicative of poor cell quality. C. Violin plots showing number of genes (nGene), number of UMIs (nUMI) and percentage of mitochondrial genes (percent.mito) detected in WT, Flox+ and Flox‐ single cells by scRNA sequencing after cell filtering. Again, black dots represent single cells, and the orange areas indicate distribution and density of analyzed data. VlnPlot function is used to make the plots, representing cells that pass the cell quality control described in B. Note that density in all parameters is now displaced to higher values.
111
Results treatment to eliminate sources of variation that could affect downstream dimensionality reduction and clustering to infer the correct gene expression patterns. Sources of variation might come from technical noise, batch effect (in our case, the variability due to different experimental days) or noise of biological origin. ScaleData is the Seurat function used for scaling, which constructs lineal models to predict gene expression based on the variables we included (to regress on), and then centers (the expression of each gene in every single cell is centered by subtracting the average expression for that gene) and scales (the centered expression value for each gene in every single cell is divided by its standard deviation) the resulting values in the dataset. We tested the effect on clustering of regressing out genes according to nUMI, nGene, the percentage of mitochondrial genes and the number of plate (for batch effect). We found useful to regress out genes based on nUMI as a potential source of variation, but we observed that regressing out the other mentioned features did not alter results substantially. Filtered, normalized and scaled data from separated genotypes were saved for downstream analysis.
10.2.3. Clustering analysis reveals different subpopulations of PV interneurons A number of recent studies based on single cell transcriptomics have advanced the molecular taxonomy of neurons and revealed the existence of different subpopulations (clusters) of PV interneurons (Harris et al., 2018; Tasic et al., 2016, 2018). Therefore, we proceeded to investigate the existence of PV clusters among our selected cells. Once we obtained datasets containing “healthy” cells and separated by genotype, we carried out a sequential clustering analysis.
Overall, we performed four consecutive clustering steps to finally achieve a rational and consistent PV clustering. After the first three clustering steps, we got expression of cell‐damage‐related genes or non‐ neuronal genes as cluster markers, so we excluded these cell populations and started again a new clustering analysis. This subparagraph has been organized in four sections/clustering analyses, where each clustering consisted of five main processes. Because from the 2nd to the 6th step were repeated per clustering analysis, they will be only fully described in the first one to avoid redundancy.
10.2.3.1. First clustering We applied several criteria to build up a first set of PV neuron clusters. As explained below, we attended first to the expression of Pvalb, we calculated high variable genes (HVG), and then applied canonical correlation analysis (CCA) for clustering. The corresponding code is in Seurat integrated analysis ‐ PV clustering 1.R script (Supp. Data: R scripts and files > Seurat integrated analysis).
1) Selection of Pvalb mRNA expression
We evaluated Pvalb mRNA expression and removed non‐Pvalb‐expressing cells. We found that the majority of the remaining cells per genotype expressed Pvalb mRNA (Pvalb expression > 0 UMI; WT: 86%,
112
Results
Flox+: 85%, Flox‐ 83%) while only some cells did not. After removing cells that did not expressed Pvalb, we obtained 817 WT cells, 1257 Flox+ cells and 1331 Flox‐ cells (Figure 35A).
2) Determination of High Variable Genes (HVG)
The second step involved the detection of high variable genes (HVG). These are genes that contribute strongly to cell‐to‐cell transcriptomic variation within a homogeneous cell population and therefore are useful to identify clusters. The detection of HVG is required to apply the CCA (Canonical Correlation Analysis). The R function in Seurat that identifies outlier genes is FindVariableGenes (BOX 6).
BOX 6
FindVariableGenes uses a plot that displays in logarithmic scale the average expression (we used mean) versus a dispersion measurement (we employed variance to mean ratio, that is, the standard deviation) for each gene across cells. This function provides a useful method to identify genes that show a strong relation between variability and average expression within cells, based on predefined thresholds for each parameter. These limitations were established according to recommended settings for UMI data normalized to 10000 molecules (https://satijalab.org/seurat/pbmc3k_tutorial.html), which comprised 0.0125 and 3 for minimal and maximal cutoff, respectively, of the mean expression (X axis), and 0.5 for minimal cutoff of the standard deviation (Y axis). We were only interested in identifying genes which expression varies significantly within cells of each genotype and therefore they show a high standard deviation.
We obtained HVG separately per genotype (S3): 6074 genes in WT, 5728 genes in Flox+ and 5063 genes in Flox‐ cells. However, only HVG present in at least two of the datasets were selected within the first 1000 HVG from each genotype, which finally yielded 703 HVG in total. An additional requirement for a gen to be considered HVG was to have a variance value different to zero in every dataset. Accordingly, the RunMultiCCA (see 3.Running multi‐set CCA and CC selection for cell clustering section) function applied to the previous HVG selection, detected and filtered out any gene with zero variance in any of the datasets, reducing the HVG final number from 703 to 698 HVG (Figure 35A).
3) Running multi‐set CCA and CC selection for cell clustering
The dimension of our dataset at that moment was 24378 genes (with 698 HVG) by 3405 cells (Figure 35A), but many genes and single cells were probably not interesting because they did not vary a lot. The goal of Principal Component Analysis (PCA), a dimensionality reduction procedure, is to find the vectors (usually the first two principal components) in which there is most variance. In other words, the aim of dimensionality reduction is to remove redundant information to describe data only by features with high variance, i.e., finding a low‐dimension set of axes to define our data (https://towardsdatascience.com/dimensionality‐reduction‐does‐pca‐really‐improve‐classification‐ outcome‐6e9ba21f0a32). In this way, cells with high correlation (similar cells) will appear together, while distinct cells will be separated.
113
Results
After selecting HVG, the three datasets were finally integrated in one dataset (so called “Integrated analysis”) in order to run a multi‐set dimensionality reduction with the selected high variable genes. RunMultiCCA function was used to perform dimensionality reduction with the three datasets (similar to PCA, which uses only one dataset) for 30 canonical vectors (CCs) or subspaces, taking genotype‐mixed cells and finding common sources of variation. Each CC would basically represent a “metagene” set that describes a specific group of cells in our data. In Figure 34A, B, we can observe cell distribution using a CC1 violin plot, and also by plotting CC1 against CC2, the two first canonical correlation vectors (CCs), after running CCA. Although two distinguishable groups were easily identified in both plots, single cells overlapped between genotypes and no shifted populations were detected ‐ except for a mild displacement of Flox‐ cells in the smaller group ‐, suggesting the absence of unique populations in any of them, at least using the first two CCs.
Normally, the first CCs explain most of the variability in the data, making the rest of dimensions somehow dispensable. Nevertheless, it is very important to choose CCs correctly for downstream analysis to overcome the technical noise in any single gene, because Seurat employs dimensionality reduction scores for clustering. For CC selection, we took advantage of two approaches. We used the MetageneBicorPlot function to plot the measurement of correlation strength for the 30 CCs (bicor‐plot), in order to find when the curve saturates, which gave us the stop point for CCs selection since next CCs would not give extra information (Figure 34B). In addition, we also examined heatmaps (a useful visualization of the above mentioned “metagene”) of the top 10 genes driving each one of the 30 CCs, using DimHeatmap function, but using only 500 cells to speed up the process. Considering these two visualizations, we decided to reasonably select the first 17 CCs, because we observed that Flox+ and Flox‐ curves in the bicor‐plot extinguished at CC17 (Figure 34B), and heatmaps became undistinguishable also after CC17 (S7). In addition, we detected that heatmaps from CC18 to CC30 displayed the same genes, confirming that these dimensions were not providing additional information as they came only from WT data (S7).
4) Alignment of CCA subspaces
As a final process before clustering, data were subjected to a CCA subspaces alignment to correct scale and density changes between genotypes that could lead to differences among CCs. It was done by the AlignSubspace function, which returned a new dimensional reduction aligned, in our case, for the 17 selected CCs, that were used for clustering. Aligned CC1 and CC2 were shown by violin plots in Figure 34C.
114
Results
Figure 34. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the first clustering. A. Violin plot displaying WT, Flox+ and Flox‐ cell distribution described by the first subspace or canonical vector, CC1. Black dots represent single cells, and colors indicate genotype. B. Dot plot showing cell distribution represented by the two first canonical vectors, CC1 versus CC2, after running CCA dimensional reduction. DimPlot is used to create this plot. Again, each dot represents a single cell, and colors indicate genotype. Cells are separated in two distinguishable groups, both containing cells from all genotypes. C. MetageneBicorPlot function generates a bicor‐plot showing measures of correlation strength (Y axis) for each CC (X axis). Colors indicate genotype. CC selection is established at CC17 as Flox+ and Flox‐ curves extinguished after this CC. D. Violin plots showing cell distribution described by CC1 (left) and CC2 (right) scores after CCA subspaces alignment (ACC). Note that all genotypes are successfully aligned.
5) Cell clustering and non‐linear dimensional reduction (tSNE)
Aligned data were used as input for FindClusters, the Seurat function involved in grouping cells by clusters/identities. Briefly, it first calculates k‐nearest neighbors (KNN) based on Euclidean distance in CCA space and constructs a shared nearest neighbor (SNN) plot; then it adjusts the modularity function to define clusters. That is, FindClusters incorporates a resolution parameter that establishes the clustering granularity, where high levels could lead to a greater number of identities in larger datasets. For a dataset
115
Results around 3000 cells, Seurat team recommends to adjust this value between 0.6‐1.2 (https://satijalab.org/seurat/pbmc3k_tutorial.html). Because our integrated cell population was composed by 3405 cells, we established the resolution parameter in 0.6 after several tests, which finally revealed 9 different cell clusters (Figure 35A). We next used t‐distributed stochastic neighbor embedding (tSNE), a dimensionality reduction technique employed for clustering visualization of high‐dimensional data. It uses a Student’s‐t distribution to calculate the similarity between two points in a low‐dimensional space (Van der Maaten and Hinton, 2008), constructing a probability distribution where similar points appear closer than distinct points (in this case, points mean cells). RunTSNE has been developed for running tSNE dimensionality reduction in Seurat objects, and visualization in two dimensions (2D) can be plotted by TSNEPlot. Using tSNE plots (tSNE 1 versus tSNE 2), we observed cell populations colored according to the 9 clusters found, and also by genotype (Figure 35B). The tSNE plot confirmed non‐unique subpopulations of PV cells exclusively coming from any specific genotype (PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ nor PVcre:Ai27D:Dnajc5flox/‐ mice); indeed the 9 identities were all found in the three different genotypes (Figure 35B).
116
Results
6) Identification of conserved cell type markers
Seurat developed the FindConservedMarkers function to identify markers that are conserved between groups (genotypes) in each identity/cluster. This process has to be applied individually for each identity because it carries out a differential expression analysis between cells in a given cluster and cells in the rest of clusters, without differentiating by genotype. This function generates a CSV file containing several parameters (BOX 7).
BOX 7
The FindConservedMarkers function generates a CSV file containing several parameters for each condition (genotype): p_val (the significance of the difference found in the average expression of a marker in a cluster compared to the rest of clusters, calculated by Wilcoxon rank sum test), avg_logFC (log fold‐change of the average expression of a marker in a cluster compared to the rest of clusters, being a positive value if increased expression and a negative value if decreased expression in that cluster), pct.1 (percentage of cells expressing the marker in a cluster), pct.2 (percentage of cells expressing the marker in the rest of clusters), p_val_adj (adjusted p‐value based on Bonferroni correction), max_pval (maximum p‐value) and minimump_p_val (minimum p‐ value).
Because CSV files comprised separated results by genotype, it was necessary an appropriate tool to decide which genes were conserved in each cluster across genotypes, according to p‐values and fold‐ changes. We decided to calculate the mean of log fold‐change (avg_logFC) from all genotypes for every marker and named it as mean_avg_logFC, which was incorporated in a new column at the end of the file. Selection for cluster markers were done by ordering this new calculated parameter from higher expressed to lower expressed genes per cluster (Supp. Data: Conserved markers (CSV)) (advice from https://github.com/satijalab/seurat/issues).
We represented the first up‐regulated marker from each identity using violin plots to have a global graphical view and check specificity of cluster markers. First observations revealed that markers for clusters 0 to 4 were also expressed in the other populations, while markers for identities 5 to 8 were quite more specific (Figure 36). It is necessary to confirm that p‐values for selected cluster markers were below 0.05 (significant) within the three genotypes, thus we checked p‐values at the end of all clustering steps before selecting candidate genes to be plotted by violin plots. ______
Figure 35. First clustering of cells from scRNA sequencing data. A. Sequential scheme displaying step by step selection of cells along the first clustering event; in green, number of cells removed; in red, number of cells, genes and CCs maintained or selected during each step (1‐5, blue) along the Seurat integrated analysis ‐ PV clustering 1.R script. Nine clusters are found after the first clustering step. The final table shows the number of cells per cluster (0‐8). Red crosses indicate later removal of clusters 3 and 8. B. Left, tSNE plot showing cell populations with a color pattern based on the 9 clusters obtained after the first clustering. Right, the same tSNE representation colored by genotype. The three genotypes are spread and present in all PV clusters. Clusters are ordered (0‐8) by higher number of cells. HVG, high variable genes; CCA, canonical correlation analysis; CCs, canonical vectors.
117
Results
Figure 36. Cluster markers defining PV interneuron populations after the first clustering step. Violin plots displaying cluster marker expression (Y axis) across clusters (X axis) based on probability distributions. All genes shown in this figure correspond to the first significant conserved marker obtained for each PV cluster. Markers are selected by computing the mean of avg_logFC values from WT, Flox+ and Flox‐ cells per population. The red arrow indicates which identity defines each conserved marker.
118
Results
Surprisingly, we observed two striking factors when looking at cluster markers: (1) markers for cluster 3 were composed basically by ribosomal RNA (such as Rn4.5s) and pseudogenes containing prefix “Gm” (such as Gm1821), and (2) markers in cluster 8 contained oligodendrocyte‐derived genes as the most up‐regulated markers (Plp1, Mobp, Mal, Cryab, Cnp, Opalin, etc) (Marques et al., 2016; Tasic et al., 2016, 2018). These results suggested that identity 3 may contain lysed cells that lost their cytoplasmic RNA content, so ribosomal RNA might have been preferentially sequenced (as involving a huge percentage of total RNA (Zhao et al., 2018)). On the other hand, cluster 8 comprised cells contaminated by oligodendrocytes that probably were pulled together with PV neurons during FAC‐sorting or might express tdTomato nonspecifically (see DISCUSSION). Anyway, we explored nUMI in every identity to check if potential doublets could be the cause of finding oligodendrocyte‐related genes in cluster 8, but we observed that neither cluster 3 nor 8 showed the higher nUMI values (Figure 37A). We analyzed in depth nUMI values in the identities 3 and 8 to discard batch effect or a genotype origin predominance. No correlation with genotype or mouse number was found in cluster 3 (Figure 37B, C). In contrast, Flox+ cells were predominant and presented higher nUMI compared to WT and Flox‐ cells in cluster 8 (Figure 37E). Moreover, we also observed the absence of cells belonging to several mouse numbers in cluster 8 (Figure 37D).
Based on the above explorations, our decision was to remove cells belonging to identities 3 and 8 from the analysis, as they were impeding our clustering ‐ they were not “healthy” or “pure” PV neurons ‐, and this would interfere with our downstream analysis. After subtraction of clusters 3 and 8, integrated data were again split into WT (734 cells), Flox+ (1090 cells) and Flox‐ (1167 cells) (Figure 39A), because a new clustering analysis was necessary.
10.2.3.2. Second clustering In this second clustering round, we repeated the previously described process (paragraph 10.2.3.1) from the 2nd to the 6th step. The detailed code for the second clustering is written in Seurat integrated analysis ‐ PV clustering 2. R script (Supp. Data: R scripts and files > Seurat integrated analysis).
2) Determination of high variable genes (HVG)
It was necessary to recalculate the HVG because we discarded cells in the previous round, so new values for average expression per gene and dispersion (standard deviation) were expected. The same parameters for FindVariableGenes were used this time. We identified 6486 HVG for WT, 6202 HVG for Flox+ and 5380 HVG for Flox‐ (S4), but only 614 genes were finally selected as HVG following the same selection approach than previously described (Figure 39A).
119
Results
Figure 37. Analysis of identities 3 and 8 obtained from the first clustering. A. Violin plots examining the number of molecules (nUMI) per cell (black dot) in all clusters (0‐8). The red arrows indicate identities subjected to further examination. No potential doublets are found in clusters 3 and 8. B. Violin plots showing nUMI per cell in cluster 3 based on mouse origin number. No batch effect suggested according to the analysis. C. Violin plots showing nUMI per cell in cluster 3 based on the different genotypes (PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐). Not a clear genotype predominance is observed. D. Violin plots showing nUMI per cell in cluster 8 based on mouse origin number. There is absence of cells from some mice (9 mice out of n=12 total mice). E. Violin plots showing nUMI per cell in cluster 8 based on genotype (PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐). Flox+ cells (red arrow) are predominant and present higher nUMI compared to WT and Flox‐ cells. The same color pattern is maintained for each genotype.
3) Running multi‐set CCA and CC selection
After running RunMultiCCA function, we generated CCs plots that were very similar to those found in the first clustering. Two distinct populations were described by CC1 and CC2 subspaces, but again, cells overlapped in each of them and no populations exclusive from any of the genotypes were detected using these CCs (Figure 38A, B). Based on the bicor‐plot and the heatmaps, we chose 22 CCs for downstream
120
Results analysis. We observed that WT, Flox+ and Flox‐ curves in the bicor‐plot went up around CC20 after a continuous falling, but WT curve suffered a huge decrease again at CC22. This was concordant with the heatmaps, since they became similar in gene expression and indistinguishable after CC22 (Figure 38C; S8).
4) Alignment of CCA subspaces.
Alignment with 22 CCs produced a new dimensionality reduction and aligned subspaces for all genotypes, showed by CC1 vs CC2 in Figure 38D.
Figure 38. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the second clustering. A. Violin plot displaying WT, Flox+ and Flox‐ cell distribution described by the first subspace or canonical vector, CC1. Black dots represent single cells, and colors indicate genotype. B. Dot plot showing cell distribution represented by the two first canonical vectors, CC1 versus CC2, after running CCA dimensional reduction. DimPlot is used to create this plot. Again, each dot represents a single cell, and colors indicate genotype. Cells are separated in two distinguishable groups, both containing cells from all genotypes. C. MetageneBicorPlot function generates a bicor‐plot showing measures of correlation strength (Y axis) for each CC (X axis). Colors indicate genotype. CC selection is established according to the accentuated fall in the WT curve at CC22. D. Violin plots showing cell distribution described by CC1 (left) and CC2 (right) scores after CCA subspaces alignment (ACC). Note that all genotypes are successfully aligned.
121
Results
5) Cell clustering and non‐linear dimensional reduction (tSNE)
Next, FindClusters was run using the aligned CCA subspaces, with 22 CCs. A resolution parameter of 0.6 revealed 9 clusters (0‐8), where cells from all genotypes were again spread within the 9 identities (Figure 39A, B).
Figure 39. Second clustering of cells from scRNA sequencing data. A. Sequential scheme displaying step by step selection of cells along the second clustering event; in red, number of cells, genes and CCs maintained or selected during each step (2‐5, blue) along the Seurat integrated analaysis ‐ PV clustering 2.R script. Nine clusters are found after the second clustering step. The final table shows the number of cells per cluster (0‐8). The red cross indicates later removal of cluster 8. B. Left, tSNE plot showing cell populations with a color pattern based on the 9 clusters obtained after the second clustering. Right, the same tSNE representation colored by genotype. The three genotypes are spread and present in all PV clusters. Identities are ranked (0‐8) according to higher number of cells. HVG, high variable genes; CCA, canonical correlation analysis; CCs, canonical vectors.
6) Identification of conserved cell type markers
Cluster markers detected in this second clustering are available at Supp. Data: Conserved markers (CSV). Violin representations of cluster markers revealed that identity markers for populations 0 to 4 were higher expressed in these groups, although they were also detected in the other clusters; while markers for populations 5 to 8 seemed to be more specific. All of them showed a p‐value < 0.05 for the three genotypes (Figure 40).
122
Results
Figure 40. Cluster markers defining PV interneuron populations after the second clustering step. Violin plots displaying cluster marker expression (Y axis) across clusters (X axis) based on probability distributions. All genes shown in this figure correspond to the first significant conserved marker obtained for each PV cluster. Markers are selected by computing the mean of avg_logFC values from WT, Flox+ and Flox‐ cells per population. The red arrow indicates which identity defines each conserved marker.
123
Results
Nevertheless, after exploring cluster markers, we realized that identity 8 contained genes (Apoe, Sepp1, Itm2a, Fcer1g, Dbi, Crip1, Cst3, B2m, C1qc, Ctss, etc.) belonging to non‐neuronal cell populations, such as astrocytes, oligodendrocytes, and macrophages, within others. Those genes were barely or not expressed in PV subpopulations, according to mouse cell types from the Allen Brain Atlas (http://celltypes.brain‐map.org/rnaseq/mouse; Tasic et al., 2016, 2018) (Figure 41A). When we looked at nUMI in all clusters, we found no potential doublets in identity 8 (Figure 41B), neither an evident batch effect (Figure 41C) nor a clear genotypic predominance (Figure 41D). Therefore, we removed cluster 8 due to non‐neuronal cell contamination that might have remained masked during the first clustering. Data were again separated into cells from PVcre:Ai27D:Dnajc5WT (726 cells), PVcre:Ai27D:Dnajc5flox/+ (1074 cells) and PVcre:Ai27D:Dnajc5flox/‐ (1150 cells) mice (Figure 43A).
10.2.3.3. Third clustering The third clustering started again from the 2nd to the 6th step of the clustering process. Seurat integrated analysis ‐ PV clustering 3.R script describes the R code for clustering 3 (Supp. Data: R scripts and files > Seurat integrated analysis).
2) Determination of high variable genes (HVG)
FindVariableGenes was run again, identifying 6801 HVG for WT, 6788 HVG for Flox+ and 6590 HVG for Flox‐ cells (S5). 649 genes were selected as HVG taking into account the first 1000 genes from the three datasets (Figure 43A).
3) Running multi‐set CCA and CC selection
One more time, we observed two groups of cells after plotting CC1 and CC2 dimensions, with cells overlapping in both groups from all genotypes (Figure 42A, B). WT and Flox‐ curves in the bicor‐plot followed a normal decrease with a visible elbow driving to saturation after CC20, while Flox+ line experimented a high increase before CC20 (Figure 42C). We followed the same guidelines for CC selection as before, so heatmaps were also examined. CC dimensions visualized by heatmaps became messy and ______
Figure 41. Analysis of identity 8 obtained from the second clustering. A. Heatmap generated from the Allen Brain Atlas for mouse cell types. It shows the expression levels of the first significant markers defining cluster 8 within their established PV populations and some non‐neuronal subclasses, including astrocytes, oligodendrocytes, VLMC and macrophages. This comparison demonstrates that the first highly expressed genes found in the identity 8 belong to non‐neuronal subclasses (yellow box) and not to PV ones. B. Violin plots examining the number of molecules (nUMI) per cell (black dot) in all clusters (0‐8). The red arrow indicates the identity subjected to further examination. C. Violin plots showing nUMI per cell in cluster 8 based on mouse origin number. D. Violin plots showing nUMI per cell in cluster 8 based on the different genotypes (PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐). The same colors denoting genotypes are maintained. Data suggest no potential doublets, batch effect or a clear genotype predominance in cluster 8. Note the absence of cells from some mice in cluster 8 (10 mice out of n=12 total mice). OPC, oligodendrocyte progenitor cells; VLMC, vascular and leptomeningeal cells; PVM, perivascular macrophages.
124
Results showed the same gene expression after CC23 (S9). Thus, 23 CCs were chosen for alignment.
4) Alignment of CCA subspaces
Figure 42D showed aligned CC1 and CC2 after alignment of subspaces using 23 CCs.
125
Results
Figure 42. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the third clustering. A. Violin plot displaying WT, Flox+ and Flox‐ cell distribution described by the first subspace or canonical vector, CC1. Black dots represent single cells, and colors indicate genotype. B. Dot plot showing cell distribution represented by the two first canonical vectors, CC1 versus CC2, after running CCA dimensional reduction. DimPlot is used to create this plot. Again, each dot represents a single cell, and colors indicate genotype. Cells are separated in two distinguishable groups, both containing cells from all genotypes. C. Bicor‐plot generated by MetageneBicorPlot function showing measures of correlation strength (Y axis) for each CC (X axis). Colors indicate genotype. CC selection is established at CC23. D. Violin plots showing cell distribution described by CC1 (left) and CC2 (right) scores after CCA alignment (ACC). Note that all genotypes are properly aligned.
5) Cell clustering and non‐linear dimensional reduction (tSNE)
Next, cluster markers detection was carried out using the first 23 CCs and a resolution parameter of 0.6, which identified 8 clusters (0‐7) from the remaining cells. One more time, all genotypes were present in each of the identities (Figure 43A, B).
126
Results
Figure 43. Third clustering of cells from scRNA sequencing data. A. Sequential scheme displaying step by step selection of cells along the third clustering event; in red, number of cells, genes and CCs maintained or selected during each step (2‐5, blue) along the Seurat integrated analysis ‐ PV clustering 3.R script. Eight clusters are found after the third clustering step. The final table shows the number of cells per cluster (0‐7). The red cross indicates later removal of cluster 4. B. Left, tSNE plot showing cell populations with a color pattern based on the 8 clusters obtained after the third clustering. Right, the same tSNE representation colored by genotype. The three genotypes are spread and present in all PV clusters. Identities are ranked (0‐7) according to higher number of cells. HVG, high variable genes; CCA, canonical correlation analysis; CCs, canonical vectors.
6) Identification of conserved cell type markers
Identification of cluster markers during this third clustering was presented at Supp. Data: Conserved markers (CSV). When looking at violin plots showing the first identity markers with p‐value < 0.05 for all genotypes, we observed that markers for clusters 0 to 3 were also expressed in the other populations. However, cluster markers for populations 4 to 7 were mainly detected in their own identities (Figure 44).
127
Results
Figure 44. Cluster markers identifying PV interneuron populations after the third clustering step. Violin plots displaying cluster marker expression (Y axis) across clusters (X axis) based on probability distributions. All genes shown in this figure correspond to the first significant conserved marker obtained for each PV cluster. Markers are selected by computing the mean of avg_logFC values from WT, Flox+ and Flox‐ cells per population. The red arrow indicates which identity defines each conserved marker.
128
Results
The routine examination of cluster genes demonstrated that identity 4 surprisingly contained ribosomal RNA (Rn4.5s) and predicted “Gm” genes (such as Gm1821), indicating a potential cell lysis. This result was similar to that found in identity 3 during the first clustering, but this time cluster 4 also enclosed several neuronal genes that were detected in other identities, for example, Snca, Kcna1 and Pthlh, suggesting that cluster 4 englobed damaged PV cells – probably previously masked among different clusters ‐ that emerged as a consequence of running a new clustering analysis. In that way, re‐clustering using the same resolution parameter (0.6) might had collected these cells and had put them together because their expression patterns of damage‐related genes were very similar. Because these genes have been detected twice in our data (see First clustering ‐ step 6), we decided to confirm that they were actually not expressed in PV interneurons using the Allen Brain Atlas for mouse cell types (http://celltypes.brain‐ map.org/rnaseq/mouse). Indeed, it was found no expression of ribosomal RNA Rn4.5s in parvalbumin cells, but the Gm1821 gene (an ubiquitin pseudogene) was randomly detected among PV subclasses (Figure 45A). We also detected that identity 4 comprised cells with the highest number of UMIs (Figure 45B), suggesting that this cluster might harbor some doublets, although no batch effect seemed to happen (Figure 45C) or a genotypic predominance in cell origin (Figure 45D). All these observations concluded with cluster 4 removal for downstream analysis, leading to a fourth re‐clustering withe th remaining cells: 712 cells from PVcre:Ai27D:Dnajc5WT mice, 1027 cells from PVcre:Ai27D:Dnajc5flox/+ mice and 1108 cells from PVcre:Ai27D:Dnajc5flox/‐ mice (Figure 47A).
10.2.3.4. Fourth clustering Detailed R code for the fourth clustering is described in Seurat integrated analysis ‐ PV clustering 4.R script (Supp. Data: R scripts and files > Seurat integrated analysis), including from the 2nd to 6th step.
2) Determination of high variable genes (HVG)
High variable genes in this occasion involved 6822 genes for WT, 6857 genes for Flox+ and 6564 genes for Flox‐ cells (S6). 635 genes were selected as HVG after the evaluation of the first 1000 genes from the three datasets (Figure 47A).
3) Running multi‐set CCA and CC selection
As before, CC1 and CC2 described two distinct groups, both containing cells overlapping from all genotypes (Figure 46A, B). Similar to the third clustering, WT and Flox‐ curves in the bicor‐plot followed a normal decrease with a detectable saturation tendency after CC20, but Flox+ curve showed a big rise around CC15 (Figure 46C). Also, although heatmaps looked pretty indistinguishable almost since CC5, it was after CC21 when we detected the same gene expression patterns (S10). This fact directed our decision to take 21 CCs for downstream analysis.
129 Results
Figure 45. Analysis of identity 4 obtained from the third clustering. A. Heatmap generated from the Allen Brain Atlas for mouse cell types. It shows the expression levels of the first two significant markers defining cluster 4 (Rn4.5s and Gm1821) within their established PV subpopulations. No expression is detected for Rn4.5s, and random expression of Gm1821 is found among PV subclasses. B. Violin plots examining the number of molecules (nUMI) per cell (black dot) in all clusters (0‐7). The red arrow indicates the identity subjected to further examination. C. Violin plots showing nUMI per cell in cluster 4 based on mouse origin number. D. Violin plots showing nUMI per cell in cluster 4 based on the different genotypes (PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐). The same colors denoting genotypes are maintained. Data show that cluster 4 includes cells with the highest nUMI values, but suggest no batch effect or a clear genotype predominance.
130 Results
4) Alignment of CCA subspaces
AlignSubspace for 21 CCs produced the new dimensional reduction that was shown by CC1 and CC2 in Figure 46D.
Figure 46. Multi‐set CCA running, CCs selection and CCA subspaces alignment during the fourth clustering. A. Violin plot displaying WT, Flox+ and Flox‐ cell distribution described by the first subspace or canonical vector, CC1. Black dots represent single cells, and colors indicate genotype. B. Dot plot showing cell distribution represented by the two first canonical vectors, CC1 versus CC2, after running CCA dimensional reduction. DimPlot is used to create this plot. Again, each dot represents a single cell, and colors indicate genotype. Cells are separated in two distinguishable groups, both containing cells from all genotypes. C. Bicor‐plot generated by MetageneBicorPlot function showing measures of correlation strength (Y axis) for each CC (X axis). Colors indicate genotype. CC selection is established at CC21. D. Violin plots showing cell distribution described by CC1 (left) and CC2 (right) scores after CCA alignment (ACC). Note that all genotypes are properly aligned.
131
Results
5) Cell clustering and non‐linear dimensional reduction (tSNE)
For cell clustering, we used the first 21 CCs and a resolution parameter of 0.8. It revealed 8 clusters (0‐7) from the remaining 2847 cells. As previously found, WT, Flox+ and Flox‐ genotypes were spread within all identities (Figure 47A, B). Furthermore, nUMIs were similar within identities, except for cluster 3, which contained cells with the lower nUMI (Figure 47C).
6) Identification of conserved cell type markers
Cluster genes for every identity from the fourth clustering are available at Supp. Data: Conserved markers (CSV). The first genes found as conserved markers involved BC002163, Kcna1, Gapdh, Usmg5, Rbp4, SSt, Pcp4 and Snca genes for 0 to 7 clusters, respectively. We used violin plots to represent these markers (p‐value < 0.05 for all genotypes) throughout all identities (Figure 48). We could appreciate that markers for clusters 0 to 4 were also expressed in the other identities although with higher expression in their own clusters, while cluster markers for populations 5 to 7 were more specific to their particular identities.
10.2.4. Selection of cluster markers for the 8 detected PV identities The first conclusion we reached after the analysis of conserved markers was that none of the identities showed ribosomal Rn4.5s gene or Gm1821 pseudogene expression as the first conserved genes anymore. When we looked at violin plots (Figure 48), we also noticed that the first cluster markers found for identities 0 to 2 were maintained during all clustering rounds (BC006123, Kcna1 and Gapdh), with the only difference lying in cluster cell size (Figures 36, 40, 44, 48). Moreover, after the second clustering, we also detected another identity (identified by the expression of the Usmg5 gene) that was maintained during the third and fourth clustering steps too (Figures 40, 44, 48). Surprisingly, these four cluster genes presented low specificity (as shown by their expression in other populations), although they might be sufficiently relevant as they have been observed through all clustering analysis. On the other hand, other genes such as Snca or Pcp4 were also detected as cluster markers during all clustering rounds, suggesting that they might remain as potential candidates for cluster identification (Figures 36, 40, 44, 48), as we have now confirmed.
The second conclusion was that the first up‐regulated marker in each cluster was expressed in PV cells (at least in some populations), based on PV identities from the Allen Brain Atlas of mouse cell types (Tasic et al., 2018; http://celltypes.brain‐map.org/rnaseq/mouse). We reached to this conclusion by further analysis of PV cluster markers: we chose the top‐10 significantly (p‐value < 0.05) conserved genes per cluster and checked whether they were related to the PV classification obtained from the Allen Brain Atlas of mouse cell types (Figures 49‐56).
132
Results
Figure 47. Fourth clustering of cells from scRNA sequencing data. A. Sequential scheme displaying step by step selection of cells along the fourth clustering event; in red, number of cells, genes and CCs maintained or selected during each step (2‐5, blue) along the Seurat integrated analaysis ‐ PV clustering 4.R script.t Eigh clusters are found after the fourth clustering step. The final table shows number of cells per cluster (0‐7). B. Left, tSNE plot showing cell populations with a color pattern based on the 8 clusters obtained after the fourth clustering. Right, the same tSNE representation colored by genotype. The three genotypes are spread and present in all PV clusters. Identities are ranked (0‐7) according to higher number of cells. C. Violin plot showing the number of molecules (nUMI) per cell (black dot) in all clusters (0‐7). Cluster colors are maintained. HVG, high variable genes; CCA, canonical correlation analysis; CCs, canonical vectors.
133 Results
Figure 48. Cluster markers identifying PV interneuron populations after the fourth clustering step. Violin plots displaying cluster marker expression (Y axis) across clusters (X axis) based on probability distributions. All genes shown here correspond to the first significant conserved marker obtained for each PV cluster. Markers are selected by computing the mean of avg_logFC values from WT, Flox+ and Flox‐ cells per population. The red arrow indicates which identity defines each conserved marker.
134
Results
Figures 49 & 50. Comparison of the PV clusters 0 and 1 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types. Heatmap showing expression of the top‐10 conserved markers (Y axis) in cluster 0 (Fig. 49, upper) and cluster 1 (Fig. 50, lower) within PV subclasses defined in The Allen Brain Atlas (X axis). Generated from http://celltypes.brain‐map.org/rnaseq/mouse. CPM, counts per million reads.
135
Results
Figures 51 & 52. Comparison of the PV clusters 2 and 3 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types. Heatmap showing expression of the top‐10 conserved markers (Y axis) in cluster 2 (Fig. 51, upper) and cluster 3 (Fig. 52, lower) within PV subclasses defined in The Allen Brain Atlas (X axis). Generated from http://celltypes.brain‐map.org/rnaseq/mouse. CPM, counts per million reads.
136 Results
Figures 53 & 54. Comparison of the PV clusters 4 and 5 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types. Heatmap showing expression of the top‐10 conserved markers (Y axis) in cluster 4 (Fig. 53, upper) and cluster 5 (Fig. 54, lower) within PV subclasses defined in The Allen Brain Atlas (X axis). The yellow boxes highlight the gene pattern similarity of the PV population 4 with the Pvalb Gabrg1 identity, and the PV population 5 obtained with two PV identities (Pvalb Th Sst and Pvalb Calb1 Sst) obtained from The Allen Brain Atlas. Generated from http://celltypes.brain‐map.org/rnaseq/mouse. CPM, counts per million reads.
137 Results
Figures 55 & 56. Comparison of the PV clusters 6 and 7 versus the PV identities obtained from The Allen Brain Atlas of mouse cell types. Heatmap showing expression of the top‐10 conserved markers (Y axis) in cluster 6 (Fig. 55, upper) and cluster 7 (Fig. 56, lower) within PV subclasses defined in The Allen Brain Atlas (X axis). The yellow box highlights the gene pattern similarity of the PV population 7 with the Pvalb Vipr2 identity obtained from The Allen Brain Atlas. Generated from http://celltypes.brain‐map.org/rnaseq/mouse. CPM, counts per million reads.
138
Results
Top‐10 up‐regulated genes from identities 0 to 3 were broadly expressed in all PV identities established at the Allen Brain Atlas, suggesting a general expression of these genes within PV interneurons (Figures 49, 50, 51, 52), although being significantly (p_val < 0.05 in all genotypes) detected as cluster markers in our study (Supp. Data: Conserved markers (CSV)). The exception was cluster 0, which showed some genes that were not detected in any PV cluster from the Allen Brain Atlas (Figure 49). In contrast, top‐10 up‐regulated genes from identities 4 to 7 were found distinctly expressed along PV subpopulations from the Allen Brain Atlas (Figures 53, 54, 55, 56). Indeed, we found some similarities in several clusters. For example, genes in cluster 7 were all detected in Pvalb Vipr2 identity (Figure 56), which represents chandelier cells (Tasic et al., 2018), although Vipr2 gene was only found in a single cell in our study (Figure 57). Moreover, cluster 4 was comparable to Pvalb Gabrg1 identity (Figure 53), and cluster 5 showed more similarity with both Sst‐expressing identities: Pvalb Th Sst and Pvalb Calb1 Sst (Figure 54). Conversely, cluster 6 contained genes that were not pretty much detected in PV identities from the Allen Brain Atlas (Lamp5, Nrn1, Hsd11b1, etc.) (Figure 55). These results were also supported in the other direction, i.e., by looking in our data genes that define PV clusters at the Allen Brain Atlas (Figure 57). In this case, we could not achieve any direct correlation between clusters. The main observations involved Gabrg1 and Th expression in our cluster 4, highly detection of Sst in cluster 5, the expression of Reln in our 0, 1 and 2 clusters, and a single Vipr2‐expressing cell in cluster 7 (Figure 57).
After that, we first examined p‐values and mean_avg_logFC (mean of avg_logFC values from all genotypes) of the first 2‐3 conserved markers from every PV population (Table 24). Then, we also elaborated a summary with basic information regarding gene/protein function of these genes, using NCBI, Ensemble and Genecards databases (S11). This material, together with the expression distribution of the first (Figure 48) and the second (Figure 58) conserved genes across clusters, was useful to establish final cluster names, since we realized that the first up‐regulated gene was not always the appropriate option.
Indeed, we made three changes: (1) we established the second up‐regulated gene (Atp5k) to name the cluster 0 because the first gene (BC002163) was a pseudogene that does not code for any protein; (2) we took the two first genes (Rbp4 and Th) to differentiate cluster 4 from cluster 5, since Rbp4 gene was detected as the fourth up‐regulated conserved marker in identity 5, being Th gene unique for identity 4; (3) Snca gene was the first conserved marker for cluster 7, but it was also detected in other clusters such as cluster 6, so we selected the two first genes (Snca and Pthlh) to name cluster 7, where Pthlh was highly expressed compared to the other identities. Regarding clusters 1, 2 and 3, they were defined by the first up‐regulated gene (Kcna1, Gapdh and Usmg5) although also detectable in the rest of the groups, because the second and successive markers were similarly expressed in the other identities. Conversely, clusters 5 and 6 were specifically described by their first detected gene, Sst and Pcp4, respectively. Although we
139
Results
Figure 57. Expression of cluster markers defining the PV subclasses in the Allen Brain Atlas of mouse cell types into our experimental PV identities. Violin plots displaying expression of PV cluster markers from The Allen Brain Atlas within our detected PV populations (0‐7). There are no considerable similarities regarding the expression of their cluster markers into our PV identities, although correlations of Gabrg1 and Th (cluster 4), Sst (cluster 5), Reln (clusters 0, 1, 2) and Vipr2 (cluster 7) are found.
140
Results
Figure 58. Identifying PV interneuron populations after the fourth clustering step using the second cluster markers. Violin plots displaying probability distributions of cluster marker expression (Y axis) across clusters (X axis). All the genes shown here correspond to the second significant conserved marker obtained for each PV cluster. Markers are selected by computing the mean of avg_logFC values from WT, Flox+ and Flox‐ cells per population. The red arrow indicates which identity defines each conserved marker.
141
Results
Conserved markers CLUSTER Gene mean_avg_logFC p‐value WT p‐value Flox+ p‐value Flox‐ BC002163 0.87 3.55E‐75 6.15E‐69 3.43E‐80 0 Atp5k 0.70 3.63E‐75 2.71E‐65 4.34E‐76 Kcna1 1.65 1.17E‐86 2.98E‐98 6.70E‐58 1 Atp2a2 1.46 9.02E‐92 4.05E‐105 2.86E‐59 Gabra1 1.46 7.96E‐89 2.89E‐106 8.37E‐59 Gapdh 1.01 3.95E‐50 2.50E‐52 1.21E‐80 2 Zwint 0.76 9.10E‐47 1.54E‐44 8.74E‐54 Tuba1b 0.73 8.60E‐29 3.12E‐30 3.45E‐59 Usmg5 0.61 8.80E‐10 9.76E‐19 3.41E‐30 3 Cox7b 0.60 8.17E‐11 1.48E‐17 2.24E‐33 Rbp4 1.17 6.41E‐11 2.50E‐25 1.08E‐30 4 Th 0.91 2.31E‐56 3.44E‐30 4.03E‐50 Sst 3.23 1.29E‐08 2.56E‐08 2.34E‐16 5 Npy 1.80 3.04E‐10 5.82E‐13 3.23E‐19 Pcp4 2.81 1.33E‐12 2.53E‐33 5.35E‐26 6 Tmsb10 2.61 1.19E‐14 2.70E‐24 2.13E‐15 Snca 3.08 4.58E‐14 9.01E‐23 1.48E‐22 7 Pthlh 1.87 1.32E‐07 8.43E‐16 6.78E‐19 Table 24. Summary of conserved genes obtained after the final clustering analysis for PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ PV genotypes. First 2‐3 conserved genes found after PV cell clustering within the three genotypes (WT, Flox+ and Flox‐) are shown in order of decreasing mean log fold change (mean_avg_logFC), with a p‐value < 0.05 in all conditions. Mean log fold change refers to the mean of the log fold changes found in the three genotypes per gene in the selected cluster compared with the rest of clusters. Values are rounded using two decimals. Data are extracted from CSV files (clustering 4) directly obtained by FindConservedMarkers function from the Seurat package (Supp. Data: Conserved Markers (CSV)).
obtained several ubiquitous genes as cluster markers (Gapdh, Usmg5, Atp5k), we established our final clustering as mentioned since our data significantly revealed those results. Indeed, RNA sequencing is an open field continuously supplemented with new data, so we cannot discard any finding, although it has to be carefully interpreted. Once we decided how to name each cluster, we substituted identity numbers (from 0 to 7) with the final names in the tSNE plot (Pvalb.Atp5k, Pvalb.Kcna1, Pvalb.Gapdh, Pvalb.Usmg5, Pvalb.Rbp4.Th, Pvalb.Sst, Pvalb.Pcp4, Pvalb.Snca.Pthlh) (Figure 59).
Then we employed another visualization function from Seurat package, FeaturePlot, to annotate the final cell type markers. It used dimensional reduction to plot cells colored by the expression level of a feature of interest. In this case, we plotted cells based on precomputed tSNE (RunTSNE result), using dimensions 1 and 2, and coloring them by cluster genes: Atp5k, Kcna1, Gapdh, Usmg5, Rbp4, Th, Sst, Pcp4, Snca and Pthlh (Figure 60). Feature graphics showed similar results as those from violin plots (Figures 48, 58) but with the additional information of PV clusters probability distances in two dimensions. We could confirm that genes identifying the first clusters (Pvalb.Atp5k, Pvalb.Kcna1, Pvalb.Gapdh and Pvalb.Usmg5) were not very specific to their own identities as they were expressed in other PV populations too. Conversely, genes from the rest of clusters seemed to be more specific, such as Snca and Pthlh genes,
142
Results which almost uniquely described Pvalb.Snca.Pthlh identity. As mentioned before, the 10 genes analyzed in this cluster were all expressed in Pvalb.Vipr2 cluster from The Allen Brain Atlas of mouse cell types, thus suggesting they might represent PV chandelier cells.
Moreover, clusters of PV interneurons were also graphically represented by a complete heatmap (DoHeatmap function) showing expression of the top‐5 conserved genes from all identities (Figure 61), which reproduced the results mentioned above in an easier visual representation.
Figure 59. Final identity names for PV clusters after sequential clustering steps. The tSNE plot generated after the fourth clustering step is now displaying the selected cluster names for all PV identities (0‐7). Cells are colored according to the 8 identities obtained after the fourth clustering; to provide an easy visualization of the identity location and size; each PV cluster is outlined using the same color than cells belonging to that cluster.
143
Results
Figure 60. Expression of cluster markers that define PV identities within all PV populations using tSNE dimensional reduction. Feature plots showing expression level of the different cluster genes previously selected as PV cluster names within all PV identities. tSNE dimensional reduction is used to plot these graphics using the FeaturePlot function from Seurat package. Note that genes from the first clusters (Atp5k, Gapdh or Usmg5) are also detected in other populations, in contrast to genes from Pvalb.Pcp4 or Pvalb.Snca.Pthlh clusters, which are more specific. Purple intensity indicates expression level.
144
Results
Figure 61. Top‐5 conserved markers defining PV populations across all PV identities. Heatmap showing expression of the first top‐5 conserved genes found in each cluster across PV identities. This function (DoHeatmap) cannot display genes a second time in case they have already appeared in the plot once; for this reason, the Pvalb.Sst cluster only shows 3 genes since Lgals1 and Rbp4, which go between Npy and Synpr genes, cannot be re‐plotted. In any case, the expression of these two genes can be observed immediately above, within the top‐5 genes from Pvalb.Rbp4.Th cluster. Cluster names are specified on the top, keeping the same colors as previously. The red boxes highlight the group of cells defined by the top‐5 conserved genes corresponding to each cluster.
10.2.5. Number of cells per genotype across PV identities Interestingly, when we looked at cell numbers for each genotype across PV identities, we identified several differences. Although Pvalb.Pcp4 and Pvalb.Snca.Pthlh clusters contained similar percentage of cells from WT, Flox+ and Flox‐ mice, small alterations were found regarding Pvalb.Gapdh, Pvalb.Usmg5, Pvalb.Rbp4.Th and Pvalb.Sst clusters. In these groups were found few more cells belonging to Flox‐ than to Flox+ or WT animals (Table 25). Remarkably, the two remaining clusters revealed interesting changes in
145
Results cell percentage origin: Pvalb.Atp5k cluster was made up of 40% of total Flox+ cells, 38% of total Flox‐ cells and 30% of total WT cells, so there were approximately 1.3 times more Flox+ and Flox‐ cells than WT cells; Pvalb.Kcna1 identity showed an evident reduction in Flox‐ cells (11% of total), which was a third part compared to WT cells (34% of total) and decreased by half compared to Flox+ cells (23% of total), the last genotype also showing a slight decrease compare to WT cells (Table 25). However, these differences were not significant when analyzing PV cell number per cluster and mouse within each genotype (S12; Figure 62). But, interestingly, the tendencies mentioned above were appreciated in all populations, showing also an evident reduction of the number of PV cells in the Pvalb.Kcna1 population from Flox‐ mice, compared to both controls. This high variability was due to the differences found in the number of cells that remained from each chip after the analysis: 34 cells from the WG17002 plate, 773 cells from the WG17003 plate, 829 cells from the WG17004 plate, and 1211 cells from the WG17005 plate (S12).
WT Flox+ Flox‐ (712 cells) (1027 cells) (1108 cells) Pvalb.Atp5k 217 (30%) 408 (40%) 423 (38%) Pvalb.Kcna1 240 (34%) 241 (23%) 118 (11%) Pvalb.Gapdh 130 (18%) 157 (15%) 265 (24%) Pvalb.Usmg5 43 (6%) 69 (7%) 115 (10%) Pvalb.Rbp4.Th 23 (3%) 56 (5%) 80 (7%) Pvalb.Sst 30 (4%) 49 (5%) 67 (6%) Pvalb.Pcp4 18 (3%) 30 (3%) 20 (2%) Pvalb.Snca.Pthlh 11 (2%) 17 (2%) 20 (2%) Table 25. Number of cells across PV identities from PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice. Number of cells belonging to each genotype are presented for every PV identity. Besides, percentage of cells in each cluster was also added due to differences in total number of cells between genotypes (712, 1027 and 1108 cells for WT, Flox+ and Flox‐ mice, respectively), reaching an improved result for comparison (number of cells in a genotype‐cluster divided by the total number of cells in that genotype). Percentage values are rounded to units.
146 Results
10.2.6. Control for specific cell‐type markers across genotypes and PV clusters As a final control test before concluding our clustering, we checked the expression of some genes described as specific cell‐type markers, such as the pan‐neuronal markers Stmn2, Thy1 and Syt1, which might be expressed in PV cells, but also markers specific for oligodendrocytes (Mog), astrocytes (Gfap, Aqp4), endothelial cells (Fn1) and microglia (C1qc), which might not be detected within PV populations (Hochgerner et al., 2017, 2018; Pollen et al., 2014). As shown in Figure 63A, the pan‐neuronal markers were expressed together with Pvalb within WT, Flox+ and Flox‐ PV cells, acting as positive control markers; in contrast, the non‐neuronal genes were not found in any genotype, except for few cells. Moreover, we also distinguished between glutamatergic and GABAergic markers to deeply check our cells, finding few cells that expressed Slc17a7 gene encoding Vglut, but abundantly expressing Gad1, Gad2 (also known as Gad67 and Gad65, respectively) and Slc32a1 (encoding VGat) in all genotypes. Syt2 was also found expressed in PV cells but not in all of them (Figure 63B). The average expression of normalized data for all these genes across genotypes and PV identities was calculated using the AverageExpression function from Seurat package and compiled in S13.
Syt2 and Pvalb mRNA expression was further analyzed across PV identities per genotype (Figure 64). Syt2 was present in PV interneurons, however, not in all PV populations, as it has been reported at the transcriptomic level to label PV basket, but not chandelier cells (Harris et al., 2018; Kerr et al., 2008; Sommeijer and Levelt, 2012; http://celltypes.brain‐map.org/rnaseq/mouse). We found that the Pvalb.Kcna1 cluster was the population that showed the higher Syt2 expression level in all genotypes; besides, we confirmed in our data that the chandelier population (Pvalb.Snca.Pthlh) did not express Syt2 as previously reported (average expression equal to zero) (Figure 64; S13). Moreover, although Syt2 expression was lower in the Flox‐ genotype and also in several PV populations from the Flox‐, those differences were not significant compared to the other genotypes since Syt2 gene did not emerge as a DE gene (see next chapter 10.3.Differential expression analysis). On the other hand, regarding Pvalb, Pvalb.Usmg5 identity showed the higher Pvalb expression in all genotypes, followed by Pvalb.Atp5k cluster in second place. The order for Pvalb expression in the remaining PV populations varied per genotype, although Pvalb.Pcp4 and Pvalb.Snca.Pthlh always occupied the seventh or last position (Figure 64; S13). However, Pvalb expression was neither significantly changed across genotypes (see next chapter 10.3.Differential expression analysis). ______
Figure 62. Quantification of PV cell number per cluster across genotypes. Bar plots showing the mean ± SEM values for PV cell number that belong to each cluster (X axis) across genotypes (WT, Flox+ and Flox‐). No statistically significant differences are found due to the high variability within each genotype. However, a decreasing tendency is appreciated in the Pvalb.Kcna1 identity of the Flox‐ genotype compared to both controls. Error bars indicate SEM. Student’s t‐test and Mann‐Whitney U test are used for statistical significance determination. Represented from S12. 147
Results
148 Results
Figure 63. Control analysis to determine the expression of specific cell‐type markers within PV cells classified by genotype. A. Violin plots showing the distribution (Y axis) across genotypes (X axis) of well‐known cell‐type markers expression. The genes displayed are specific for neurons (Stmn2, Thy1 and Syt1), oligodendrocytes (Mog), astrocytes (Gfap, Aqp4), endothelial cells (Fn1) and microglia (C1qc). B. Violin plots showing the distribution (Y axis) across genotypes (X axis) of well‐known markers expression for PV interneurons (Pvalb, Syt2), GABAergic cells (Gad1/Gad67, Gad2/Gad65, Slc32a1) and glutamatergic cells (Slc17a7).
149 Results
Figure 64. Expression of Syt2 and Pvalb genes across genotypes and PV identities. A. Violin plots showing Syt2 expression distribution across PV populations from PVcre:Ai27D:Dnajc5WT (left), PVcre:Ai27D:Dnajc5flox/+ (middle) and PVcre:Ai27D:Dnajc5flox/‐ (right) mice. Clusters are ranked by decreasing Syt2 average expression per genotype. Pvalb.Kcna1 comprises the cluster with the highest Syt2 expression, and Pvalb.Snca.Pthlh shows no expression for Syt2 in any genotype. B. Violin plots showing Pvalb expression distribution across PV populations from PVcre:Ai27D:Dnajc5WT (left), PVcre:Ai27D:Dnajc5flox/+ (middle) and PVcre:Ai27D:Dnajc5flox/‐ (right) mice. Clusters are ranked by decreasing Pvalb average expression per genotype. Pvalb.Usmg5 and Pvalb.Atp5k comprise the clusters with the highest Pvalb expression, while Pvalb.Pcp4 and Pvalb.Snca.Pthlh are the clusters with the lowest Pvalb expression in all genotypes. Cluster colors are maintained during the analysis.
150
Results
10.2.7. Dnajc5 expression in PV cells Despite the restrictions of STRT‐based methods (the method does not provide full‐length information, so the Cre‐recombination in exon 3 at the Dnajc5 floxed gene might be difficult to detect), we further checked for Dnajc5 expression across genotypes and clusters. Detailed code for this analysis was compiled in Seurat integrated analysis ‐ Dnajc5 expression.R script (Supp. Data: R scripts and files > Seurat integrated analysis). In general, Dnajc5 expression level seemed low across genotypes (Table 26) if compared to the expression of genes coding other synaptic vesicle proteins, such as Sv2a or Vamp2 (8.56 and 4.62, respectively). No expression of the wild‐type version of Dnajc5 would be expected in PVcre:Ai27D:Dnajc5flox/‐ cells; surprisingly, Dnajc5 was detected in PV cells belonging to these animals although approximately reduced by half compared to controls (Table 26). Moreover, we found 1.23 times higher average expression of Dnajc5 in PVcre:Ai27D:Dnajc5flox/+ compared to PVcre:Ai27D:Dnajc5WT mice, which was unexpected because a reduction in expression levels by half (heterozygous) would be predicted (Figure 65A; Table 26).
Flox‐ Flox+ WT Dnajc5 0.85 1.58 1.28 Table 26. Average Dnajc5 expression for PV cells by genotype. Average expression (AverageExpression function from Seurat package) of Dnajc5 in WT, Flox+ and Flox‐ PV cells. Values are rounded using two decimals.
When we looked at PV clusters per genotype, we also obtained low Dnajc5 expression levels in general (S14). However, Dnajc5 expression was detected the highest in Pvalb.Kcna1 identity in the three genotypes, but randomly detected in the rest of clusters, showing no preserved ordering among genotypes. The second exception was ethat th Pvalb.Usmg5 population showed the lowest expression in all cases (Figure 65B, C; S14).
Nevertheless, the STRT‐seq‐2i method carries out the sequencing step using a single primer (DI‐ Read1‐Seq) that is able to provide only reads of 45 bp‐long from 5’ cDNA fragments, so it did not offer full‐ length mRNA information (see Materials and Methods section). Because of that, our results showed that the detected Dnajc5 expression in PVcre:Ai27D:Dnajc5flox/‐ genotype arose from the Dnajc5 mRNA transcribed from a recombined version of the Dnajc5 floxed allele (a transcript lacking only the floxed‐ exon 3), which can be actually detected by this method due to its intact 5’ sequence (Figure 16).
Finally, data were split into three new datasets, containing clustered PV cells from (1) PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ mice (2135 cells), (2) PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT mice (1820 cells) and (3) PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT mice (1739 cells).
151 Results
Figure 65. Expression of Dnajc5 gene across genotypes and PV identities. A. Violin plots showing distribution of Dnajc5 expression in PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ genotypes, ordered by the higher average expression from left to right. Genotype colors are maintained during data analysis. The Flox+ genotype shows the highest average expression for Dnajc5. B. FeatureHeatmap plot displaying scaled expression of Dnajc5 across PV identities from WT, Flox+ and Flox‐ genotypes using 2D tSNE plots. The Pvalb.Kcna1 cluster shows the highest expression of Dnajc5 in all genotypes. C. Violin plots showing Dnajc5 expression distribution across PV populations from PVcre:Ai27D:Dnajc5WT (left), PVcre:Ai27D:Dnajc5flox/+ (middle) and PVcre:Ai27D:Dnajc5flox/‐ (right) genotypes. Clusters are ranked by decreasing Dnajc5 average expression per genotype. Cluster colors are maintained during the analysis. Pvalb.Kcna1 comprises the cluster with the highest Dnajc5 expression, and Pvalb.Usmg5 is the cluster with the lowest Dnajc5 expression in all genotypes.
152 Results
10.3. Differential expression (DE) analysis After PV clustering, we also used the Seurat package to apply differential expression analysis, which aims to determine differentially expressed (DE) genes between two given groups of cells. We first looked for DE genes among all PV cells together, a major initial goal. However, the clustering of PV cells opened the opportunity to carry out this analysis on each PV identity to find potential cluster‐specific changes in gene expression. Both types of analysis were done comparing genotype pairs: PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5flox/+ mice, PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5WT mice, and PVcre:Ai27D:Dnajc5flox/+ versus PVcre:Ai27D:Dnajc5WT mice.
FindMarkers is the function in Seurat that finds DE genes between two groups of cells. We established the same considerations (BOX 8) to run FindMarkers in both types of analysis (using all PV cells and by clusters). The parameters generated by this function are also described in BOX 8.
BOX 8
We applied FindMarkers using the Wilcoxon rank sum test (by default). An expression requirement in at least 10% of cells in the two groups (by default) was also established. We did not consider a minimum fold‐difference between groups of cells (by default), because we wanted to detect total changes.
FindMarkers produces a CSV file with the following parameters (similar to FindConservedMarkers): p_val shows the significance of the difference found in the average expression of a gene between the two groups, while avg_logFC gives the log fold‐change of the average expression of a gene between the two groups, which comprise values that were either positive, if the gene was up‐regulated in the first group, or negative if down‐ regulated in the first group. The parameter pct.1 represents the percentage of cells where the gene is detected in the first group (depending on the analysis, all PV cells or a cspecifi PV cluster from Flox‐ or Flox+ cells). The parameter pct.2 comprises the percentage of cells where the gene is detected in the second group (depending on the analysis, all PV cells or a specific PV cluster from Flox+ or WT cells). p_val_adj indicates an adjusted p‐ value based on Bonferroni correction.
10.3.1. DE analysis including all PV cells During this analysis, we looked for gene expression changes in PV interneurons between two conditions, Flox‐ versus Flox+, Flox‐ versus WT, and Flox+ versus WT, without considering different PV identities. Analysis between controls (Flox+ and WT) was performed as an additional control in case of detecting relevant gene differences. The R script developed for this study was named Seurat integrated analysis ‐ DE genes – All PV.R (Supp. Data: R scripts and files > Seurat integrated analysis) and contains the three comparisons.
The parameters generated by FindMarkers (BOX 8) were always referred to differences found in Flox‐ cells when we compared the mutant to any of the controls (Flox+ or WT); and differences found in Flox+ cells when we compared both controls. In addition, to complement logarithmic scale values (avg_logFC), we calculated the fold change in a new column that was added at the end of the file (named as FoldChange). FoldChange helped to visualize better the actual number of times a gene is up‐ or down‐
153 Results regulated. As avg_logFC referred to the natural (Neperian) logarithm, the fold change was obtained by the Euler number raised to avg_logFC for every gene (eavg_logFC).
After running FindMarkers (Supp. Data: DE genes (CSV)), we performed an appropriate selection for significant DE genes. Threshold values for both p_val and avg_logFC were established to decide whether a gene should be selected as differently expressed. We considered a restriction value for p_val below 0.05 to be significant, but setting a proper threshold for FoldChange remained tough. Indeed, to determine the minimum threshold value to consider a fold change as meaningful or not may result undefinable. For that reason, and because it was out of the scope of this initial study to analyze the differences in the total number of genes found, we considered approximately the first significant 10 genes in both directions (up‐ and down‐regulated) as significant DE genes (Table 27).
Up‐regulated genes Down‐regulated genes Comparison Gene FC p‐value Gene FC p‐value Npy 2.27 4.12E‐08 Meg3 0.64 1.49E‐18 Gapdh 1.68 1.83E‐30 Slc24a4 0.69 3.59E‐20 Cort 1.60 1.67E‐07 Scn1a 0.70 1.45E‐11 Xist 1.56 5.62E‐53 Gabrb2 0.71 2.47E‐12 Tuba1b 1.42 1.56E‐18 Dnajc5 0.715 1.97E‐13 Flox‐ vs Flox+ Lgals1 1.38 2.65E‐07 Kcna2 0.720 5.71E‐10 Mustn1 1.33 3.86E‐16 AI593442 0.722 3.93E‐10 Pfn1 1.32 1.37E‐19 Snhg11 0.723 1.75E‐14 Tecr 1.32 1.09E‐21 Rpph1 0.727 2.67E‐09 Aldoa 1.30 6.83E‐23 App 0.73 2.40E‐11 Lgals1 1.61 1.60E‐08 Sv2a 0.566 1.06E‐17 Xist 1.59 1.85E‐43 Atp1a3 0.569 8.44E‐16 Cort 1.35 0.008966 Kcna1 0.588 4.02E‐15 Flox‐ vs WT Gapdh 1.34 4.72E‐11 Pcsk1n 0.593 1.59E‐12 Swi5 1.33 4.74E‐21 Nat8l 0.594 2.02E‐13 Dbi 1.32 3.93E‐13 Slc6a17 0.60 1.44E‐13 Ppia 1.31 1.75E‐28 Slc22a17 0.62 2.14E‐16 Rpph1 1.49 3.50E‐12 Atp1a3 0.67 1.09E‐10 Malat1 1.38 2.22E‐10 Npy 0.68 0.001305 Gm14305 1.29 2.01E‐09 Clstn3 0.68 2.91E‐11 Flox+ vs WT Grina 0.70 1.37E‐12 Slc22a17 0.71 8.23E‐10 Slc6a17 0.711 2.62E‐09 Syp 0.714 3.23E‐11 Table 27. Summary of significant DE genes obtained by comparison of total PV cells between genotype pairs. Approximately 10 DE genes are shown in order of decreasing fold change (FC) and with a p‐value < 0.05 for up‐ and down‐regulated genes. The comparison analysis was performed of Flox‐ versus Flox+ PV cells, Flox‐ versus WT PV cells, and Flox+ versus WT PV cells. Values are rounded using two/three decimals for FC and six decimals for p‐value (when needed). These results are extracted from CSV files obtained directly by the FindMarkers function from the Seurat package (Supp. Data: DE genes (CSV)).
A useful tool to visualize DE genes is the volcano plot. These scatter plots uses both, fold change and significance to represent large datasets and identify significant variations. Volcano plots are usually represented using the negative log of p‐value (normally base 10) in Y axis, and the log of fold change in X
154
Results axis. Then, genes with lower p‐values appear at the top of the plot, while the log of fold change is used to separate gene alterations in both directions (right, up‐regulated; left, down‐regulated), with no change at the center. Here, we employed the basic plot function from the graphics package to draw volcano plots, using the pre‐computed avg_logFC values versus the negative log (base 10) of p_val values of all obtained DE genes per comparison of genotype pairs. Moreover, restriction parameters for p_val and avg_logFC were also used for an easy and quick visualization of candidate genes. We labeled in red genes with p_val < 0.05. Approximately the first 10 genes with the highest avg_logFC in both directions (up‐ and down‐ regulated genes) were colored in orange. We labeled in green the genes that satisfied both restrictions. Genes that did not pass any filter remained black. Finally, we added name labels for the significant DE genes obtained in green. Volcano plots in Figure 66 show the main detected DE genes when we compared total PV cells between the following genotype pairs: PVcre:Ai27D:Dnajc5flox/‐ to PVcre:Ai27D:Dnajc5flox/+ mice, PVcre:Ai27D:Dnajc5flox/‐ to PVcre:Ai27D:Dnajc5WT mice, and PVcre:Ai27D:Dnajc5flox/+ to PVcre:Ai27D:Dnajc5WT mice.
To understand our results, we elaborated an overview containing brain‐related information of these genes using NCBI, Genecards and Uniprot databases, and also bibliography (S15). It allowed us to extract some useful information. For example, Xist, a X‐chromosome related gene, was found up‐regulated in Flox‐ cells compared to both, Flox+ and WT cells. This finding, however, could be due to the use of a larger number of females with PVcre:Ai27D:Dnajc5flox/‐ genotype (4 out of 5) than with PVcre:Ai27D:Dnajc5flox/+ (2 out of 4) or PVcre:Ai27D:Dnajc5WT (2 out of 3) genotypes, thus we did not prioritize Xist an interesting DE gene to be further investigated in Flox‐ PV cells. On the other hand, the expression levels of Gapdh, Cort and Lgals1 genes appeared up‐regulated in the two comparisons (Flox‐ versus Flox+ and Flox‐ versus WT) and, therefore, it strongly suggests that a biologically relevant cause linked to Flox‐ genotype could underlie those genetic differences. About these three genes, Cort is a neuropeptide that regulates cortical activity and may produce deficits in synaptic plasticity and learning if over‐expressed (Ibáñez‐Costa et al., 2017; de Lecea et al., 1997; Méndez‐Díaz et al., 2005). A pro‐apoptotic role has been described for Lgals1 in neurons, and it might participate in the elimination of neuronal processes, preventing the neurodegeneration (Bischoff et al., 2012; Plachta et al., 2007; Starossom et al., 2013). Finally, Gapdh is involved in neuronal death initiation (Ishitani et al., 1996; Tanaka et al., 2002). Importantly, neuropeptide Y (Npy) was detected as the first up‐regulated gene in Flox‐ PV cells compared to Flox+. Previous publications about NPY pointed out its relevant neuroprotective role in epilepsy, stress, anxiety and/or neurodegeneration (Erickson et al., 1996; Gøtzsche and Woldbye, 2016; Li et al., 2016; Noè et al., 2008; Reichmann and Holzer, 2016). On the other hand, cytoskeletal‐related genes, such as Tuba1b
155
Results
Figure 66. Volcano plots showing significant DE genes found after comparison of all PV cells between genotype pairs. Significant up‐ and down‐regulated genes are selected after PV cells comparison between genotypes and represented using volcano plots. Upper left, comparison of PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5flox/+; upper right, comparison of PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5WT; down, comparison of PVcre:Ai27D:Dnajc5flox/+ versus PVcre:Ai27D:Dnajc5WT. The X axis shows average log (natural) of fold change and the Y axis displays negative log (base 10) of p‐value. Genes are colored in red if p‐value<0.05; in orange, if passing a distinct avg_logFC threshold for each genotype; in green, if satisfying both restrictions (gene labels). Black dots represent the remaining genes. Up‐regulated genes appear on the right side of the plot due to their positive values for avg_logFC; down‐regulated genes are located on the left side because they have negative values for avg_logFC.
156 Results and Pfn1, were also found increased in Flox‐ compared to Flox+ animals. Moreover, we also detected an increment in metabolism‐related genes in Flox‐ cells, such as Tecr and Aldoa if compared to Flox+ cells, and Dbi and Ppia when compared to WT PV cells (Table 27; S15; Supp. Data: DE genes (CSV)).
Regarding to down‐regulated genes, we especially highlighted those genes which, upon mutations or knock‐down, cause impaired protein functions associated with epilepsy (Scn1a, Gabrb2, Kcna2 if compared to Flox+ cells; Sv2a, Atp1a3 when compared to WT), schizophrenia (Gabrb2 compared to Flox+; Atp1a3, Kcna1 when compared to WT cells), Alzheimer’s disease (Slc24a4, App when compared to Flox+) or impaired locomotor‐activity (Scn1a, Kcna2 if compared to Flox+; Atp1a3, Kcna1 in comparison to WT cells). Solute carrier family members Slc6a17 (a sodium‐dependent transporter for proline, glycine, leucine and alanine) and Slc22a17 (related to iron transport and homeostasis) were also reduced in Flox‐ cells compared to WT PV cells. As expected, it was found a down‐regulation in Dnajc5 gene in Flox‐ cells compared to Flox+ (Table 27; S15; Supp. Data: DE genes (CSV)).
The comparison between both controls (Flox+ versus WT) did not provide substantial information involving up‐regulated genes, besides, essentially, non‐coding RNAs. However, several down‐regulated genes were detected: Atp1a3, Npy (down‐regulated in Flox+ compared to WT, in contrast to the increase found in Flox‐ compared to), Flox+ mice solute carrier family members, (Slc22a17 and Slc6a17) or the synaptic vesicle protein synaptophysin (Syp) (Table 27; S15; Supp. Data: DE genes (CSV)). Nevertheless, we found a list of just 24 genes differentially expressed between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT genotypes. When we counted the DE genes detected when comparing Flox‐ PV cells versus Flox+ cells, we obtained 60 genes, and 167 genes if we compared Flox‐ versus WT PV cells (Supp. Data: DE genes (CSV)).
To finish our study we carried out Gene Ontology (GO) analysis. GO analysis is a useful exploratory tool to determine processes and pathways (called terms) related to a given set of genes. We utilized ClueGO software to collect information regarding only significant up‐ and down‐regulated genes detected in PV cells from both Flox‐ versus Flox+, Flox‐ versus WT, and Flox+ versus WT comparisons (Supp. Data: Up‐ and down‐regulated genes for GO (CSV); previously selected by Up‐ and down‐regulated genes for GO analysis.R script at Supp. Data: R scripts and files > Seurat integrated analysis). Significant terms were found related to up‐regulated genes and/or down‐regulated genes in several of the studies, which provided interesting information (Table 28). Among them, when we compared Flox‐ versus WT cells, we observed terms related to several of the significant down‐regulated genes, such as synaptic transmission, regulation of membrane potential, neurotransmitter release, synaptic vesicle cycle, or lysosome. Less terms were found when comparing down‐regulated genes from Flox‐ versus Flox+ PV cells, but showing special interest the Glycolysis/Gluconeogenesis term obtained from up‐regulated genes (Table 28). No
157 Results related terms were found through the comparison of Flox+ versus WT cells, neither regarding up‐ nor down‐regulated genes (Table 28).
Comparison Terms (up‐regulated genes) Terms (down‐regulated genes) ‐ Regulation of neurotransmitter transport. Flox‐ vs Flox+ ‐ Glycolysis/Gluconeogenesis. ‐ Action potential. ‐ Neuromuscular process. ‐ GABAergic synapse. ‐ Chemical synaptic transmission. ‐ Regulation of membrane potential. ‐ Neurotransmitter secretion. ‐ Positive regulation of synaptic transmission. ‐ Neuromuscular process. ‐ Response to metal ion. ‐ Learning. ‐ Ionotropic glutamate receptor signaling pathway. Flox‐ vs WT ‐ No functions related. ‐ Synaptic vesicle cycle. ‐ Lysosome. ‐ Regulation of G protein‐coupled receptor signaling pathway. ‐ Secretory granule localization. ‐ Response to nerve growth factor. ‐ Glutamate binding. ‐ Cell adhesion molecules (CAMs). ‐ Regulation of cardiac muscle cell membrane potential.
Flox+ vs WT ‐ No functions related. ‐ No functions related.
Table 28. GO analysis for DE genes obtained by comparison of total PV cells between genotype pairs. All significant up‐ and down‐regulated genes found after comparing Flox‐ vs Flox+ and WT cells, and Flox+ vs WT cells were selected for Gene Ontology analysis using the ClueGO plugin from Cytoscape software. Only processes and pathways (terms) with p‐value≤0.05 were chosen by ClueGO, which created GO groups that englobed associated terms. These GO groups are ordered by the lowest GO group p‐value (the most significant p‐value corrected with Bonferroni step down) and only the term with the highest significance (term p‐value corrected with Bonferroni step down) within each GO group is shown in this table. Data extracted from raw result files at Supp. Data: ClueGO analysis (XLS).
10.3.2. DE analysis by PV clusters After the DE analysis using all PV cells, we looked for genes that were differentially expressed within a same PV identity from two distinct genotypes: PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5flox/+ mice, PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5WT mice, and PVcre:Ai27D:Dnajc5flox/+ versus PVcre:Ai27D:Dnajc5WT mice.
1) Comparison between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ PV clusters
We compared PV cell identities between Flox‐ and Flox+ genotypes. Detailed R code for this analysis is described in Seurat integrated analysis ‐ PV differential expression (DE) – Flox‐ vs Flox+.R script
158 Results
(Supp. Data: R scripts and files > Seurat integrated analysis). The procedure was identical to that performed for total PV cells (BOX 8), with the exception that differentially expressed genes were computed independently for every PV identity. The fold change was also calculated for easier interpretation of differences. Generated CSV files were ordered by decreasing avg_logFC (Supp. Data: DE genes (CSV)).
To select for significant DE genes, we established a p_val threshold below 0.05. Setting a consensual threshold for avg_logFC remained also difficult due to differences in avg_logFC values within PV identities. For example, setting avg_logFC > 0.5, which represented a x1.65 fold change, left only one gene in some clusters but retained many genes in others, which resulted rather inconvenient. We decided to apply the identical strategy previously used, but reducing the number of genes to be studied. Only the first 3‐4 up‐ and down‐regulated genes were selected according to ranked avg_logFC (Table 29).
Up‐regulated genes Down‐regulated genes CLUSTER Gene FC p‐value Gene FC p‐value Npy 3.25 2.72E‐06 Rpph1 0.69 1.65E‐05 Cort 1.71 1.98E‐06 Ccl27a_loc1 0.76 2.59E‐05 Pvalb.Atp5k Lgals1 1.63 1.75E‐07 Mustn1 1.55 5.42E‐13 Xist 1.88 1.46E‐13 Pld3 0.70 2.73E‐06 Pvalb.Kcna1 Cacybp 1.34 0.00620 Srrm2 0.71 4.58E‐05 Stmn4 1.33 0.00313 Dnajc5 0.72 7.80E‐07 Xist 2.69 8.49E‐36 Dynlt1b 0.64 2.31E‐10 Arf5 1.45 1.36E‐09 Eif2s3y 0.67 3.54E‐18 Pvalb.Gapdh Rnf187 1.44 1.61E‐10 Malat1 0.69 1.12E‐05 Gapdh 1.43 9.61E‐17 D10Jhu81e 1.77 0.000437 Dynlt1b 0.48 0.000273 Pvalb.Usmg5 Fbxo2 1.69 0.000801 Rpph1 0.51 0.029372 Ttc9b 1.65 0.002719 Ubb 0.60 0.005319 Npy 2.69 0.026419 mt‐Rnr2 0.60 1.03E‐10 Pvalb.Rbp4.Th Gapdh 1.82 2.30E‐07 Ntm 0.604 0.001297 Spp1 1.68 0.009860 Amd2 0.62 0.011879 Npy 2.01 0.023762 Tcap 0.57 0.000403 Ifitm10 1.63 0.000124 mt‐Rnr2 0.619 7.12E‐08 Pvalb.Sst Vamp1 1.61 0.000245 Malat1 0.620 0.000856 Gapdh 1.59 0.000497 Cox6a2 3.22 0.002206 Sostdc1 0.35 0.015013 Pvalb.Pcp4 Gad1 2.59 0.009797 Mt1 0.43 0.001333 Ap1s1 2.45 2.67E‐05 2310036O22Rik 0.44 0.002066 M6pr 2.14 0.0002 Luc7l2 0.43 0.003563 Spryd7 2.10 0.013354 Nucks1 0.46 0.006068 Pvalb.Snca.Pthlh Bccip 1.948 0.005788 Lrrc4c 0.49 0.009166 Anxa5 1.946 0.006212 Table 29. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ mice. The first 3‐4 genes were selected in order of decreasing fold change (FC), with a p‐value < 0.05 for up‐ and down‐ regulated genes among PV populations using Flox‐ and Flox+ animals. Values are rounded with two/three decimals for FC and with six decimals for p‐value when necessary. The results are extracted from CSV files obtained directly by FindMarkers function from Seurat package (Supp. Data: DE genes (CSV)). Changes are referred to Flox‐ cells.
159 Results
To improve visualization of selected DE genes within each population between Flox‐ and Flox+ PV cells, we took advantage of volcano plots. Cells were colored according to the following parameters: red for genes with p_val<0.05, orange for the first significant 3‐4 genes (which was our threshold for avg_logFC), green for genes that fulfilled both restrictions, and black for the remaining genes. Finally, name labels for significant DE genes in green were added to the plots (Figure 67).
Functional information of these genes is shown at S15. Among them, we could find several genes also detected when comparing alls PV cell together: e.g. Xist (which was discarded as a DE gene), Npy, Cort, Lgals1, Mustn1 and Gapdh as up‐regulated genes, and Dnajc5, Rpph1 as down‐regulated ones. Npy and Gapdh must be highlighted because they were up‐regulated in several PV clusters of Flox‐ mice, but its wa Npy which showed the highest fold changes (around 2‐3 times) in general. It was also necessary to spotlight up‐regulated genes related to ubiquitination (Rnf187, Cacybp, Fbxo2), microtubules (Stmn4, Bccip), mitochondria (Cox6a2), endo/exocytosis (Ap1s1, Anxa5), lysosomes (M6pr), and synaptic vesicles (Vamp1). Regarding downd‐regulate genes, the variations looked milder compared to fold changes from up‐regulated genes, and only Dynlt1b (dynein), Rpph1, mt‐Rnr2 (both RNA components; the last one being rRNA from mitochondria) and Malat1 (nuclear non‐coding RNA) were repeated within some identities. Similar to Xist, the Y chromosome‐linked gene Eif2s3y appeared as a down‐regulated gene due to different number of males used from Flox+ (2 out of 4) than from Flox‐ (1 out of 4) mice. Apart from that, some genes spotlighted because of their association with lysosomes (Pld3), neurite outgrowth (Ntm, Lrrc4c), and spliceosome (Srrm2, Luc7l2). It is remarkable that down‐regulation of Dnajc5 (CSPα/DNAJC5) was significantly detected in third place among down‐regulated genes in Pvalb.Kcna1 cluster, although it has been also found in a farther position in the Pvalb.Snca1.Pthlh identity (Table 29; S15; Supp. Data: DE genes (CSV)).
Finally, the GO analysis revealed several terms associated with the significant DE genes detected across PV clusters, while other identities remained unrelated to any known process or pathway loaded in the databases we used (Table 30). The significant DE genes were previously selected by Up‐ and down‐ regulated genes for GO analysis.R script (Supp. Data: R scripts and files > Seurat integrated analysis) and are stored at Supp. Data: Up‐ and down‐regulated genes for GO (CSV). ______
Figure 67. Volcano plots showing significant DE genes detected after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ genotypes. Significant up‐ and down‐regulated genes are selected after PV clusters comparison of Flox‐ versus Flox+ genotypes and represented using volcano plots. The X axis shows average log (natural) of fold change and the Y axis displays negative log (base 10) of p‐value. Genes are colored in red if p‐value<0.05; in orange, if passing a distinct avg_logFC threshold for each cluster; in green, if satisfying both restrictions (gene labels). Black dots represent the remaining genes. Up‐regulated genes appear on the right side of the plot due to their positive values for avg_logFC; down‐regulated genes are located on the left side because they have negative avg_logFC.
160 Results
161 Results
Cluster Terms (up‐regulated genes) Terms (down‐regulated genes) Pvalb.Atp5k ‐ No functions related. ‐ No functions related.
Pvalb.Kcna1 ‐ No functions related. ‐ No functions related.
Pvalb.Gapdh ‐ Glycolysis/Gluconeogenesis. ‐ No functions related.
‐ Membrane assembly. ‐ Intracellular protein transmembrane Pvalb.Usmg5 ‐ Proteasome. transport. ‐ Mineral absorption. ‐ Glycolysis/Gluconeogenesis. ‐ Cellular response to thyroid hormone stimulus. ‐ Endocrine and other factor‐regulated calcium reabsorption. ‐ Gap junction. ‐ Regulation of ATPase activity. ‐ Synaptic vesicle endocytosis. ‐ Endopeptidase activator activity. Pvalb.Rbp4.Th ‐ VEGF signaling pathway. ‐ Lysosome.
‐ Negative regulation of protein polymerization. ‐ Central carbon metabolism in cancer. ‐ Cellular response to epidermal growth factor stimulus. ‐ Cation‐transporting ATPase activity. ‐ Legionellosis. ‐ Phagosome. ‐ Response to thyroid hormone. ‐ Regulation of synaptic vesicle fusion to ‐ No functions related. Pvalb.Sst presynaptic active zone membrane. ‐ Myosin V binding. ‐ Glycolysis/Gluconeogenesis. ‐ Modulation of chemical synaptic transmission. ‐ Regulation of supramolecular fiber organization. ‐ Synapse organization. ‐ Regulation of protein localization to cell periphery. ‐ Oxidative phosphorylation. Pvalb.Pcp4 ‐ Mitochondrion organization. ‐ Mineral absorption. ‐ GTP binding. ‐ Positive regulation of ATPase activity. ‐ Oligodendrocyte differentiation. ‐ Chaperone‐mediated protein folding. ‐ Macroautophagy. ‐ Dopaminergic synapse. ‐ Ubiquitin‐like protein ligase binding. ‐ Mitochondrial gene expression. ‐ Regulation of protein complex ‐ Positive regulation of mRNA splicing, via disassembly. Pvalb.Snca.Pthlh spliceosome. ‐ Regulation of actin filament bundle ‐ Intracellular protein transmembrane assembly. transport. ‐ GABAergic synapse.
162
Results
2) Comparison between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT PV clusters
We similarly studied DE genes in PV cell identities between Flox‐ and WT genotypes. R code used for this analysis is described in Seurat integrated analysis ‐PV differential expression (DE) – Flox‐ vs WT.R script (Supp. Data: R scripts and files > Seurat integrated analysis).
We followed an identical procedure that the one employed for Flox‐ and Flox+ PV clusters comparison, but in this case comparing Flox‐ versus WT PV clusters. P‐value threshold was also established below 0.05 and the first significant 3‐4 genes with the highest fold change in both directions (up‐ and down‐regulated) were selected to be studied (Table 31; Figure 68).
A compilation of brain‐related functional information regarding these significant DE genes can be found at S15. As in the previous analysis between Flox‐ and Flox+ PV clusters, we observed that Xist was again found up‐regulated in some clusters of Flox‐ cells, because 4 out of 5 PVcre:Ai27D:Dnajc5flox/‐ mice were females, compared to 1 out of 2 mice in WT. Similar results were obtained regarding Eif2s3y gene expression, which presented decreased levels in the same clusters as before (Pvalb.Gapdh and Pvalb.Pcp4).
Remarkably, there was an important up‐regulation of Npy in Pvalb.Atp5k identity, a gene that presents increased expression levels after epileptic seizures. Among up‐regulated genes, we also observed those associated to schizophrenia (Rpl30), mitochondria (Coq7, Immp1l), axonal growth and synaptic plasticity (Serpini1), clearance of endosomes/autophagic vacuoles (Rab24), regulation of mTOR signaling (Tirpl), neurogenesis (Coq7, Psip1) and chaperones (Pfdn1, Pfdn2).g Concernin down‐regulated genes, we found genes related to endo/exocytosis (Rab3c, S100a10, 6330403K07Rik, Rnf128), lysosomes (Gaa) synapses (Syp, Syn1, Rab3c, Sv2a), neurotransmission (Sst, Kcna6), mitochondria (Mrpl13, Mrpl36, Sfxn3), and genes whose decrease is associated with epilepsy (Sst, Syn1), schizophrenia (Sst, Cntn5) and neurodegenerative disorders (Sst, Cntn5, S100a10, App). Finally, Dnajc5 was reduced in Pvalb.Pcp4 and Pvalb.Sst populations (Table 31; S15; Supp. Data: DE genes (CSV)).
______
Table 30. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5flox/+ genotypes. All significant up‐ and down‐regulated genes found after comparing PV identities in Flox‐ vs Flox+ cells were used for Gene Ontology analysis using ClueGO plugin from Cytoscape software. Only processes and pathways (terms) with p‐value≤0.05 were chosen by ClueGO, which created GO groups that englobed associated terms. These GO groups are ordered by the lowest GO group p‐value (the most significant p‐value corrected with Bonferroni step down) and only the term with the highest significance (term p‐value corrected with Bonferroni step down) within each GO group is shown in this table. Data extracted from raw result files at Supp. Data: ClueGO analysis (XLS).
163 Results
Up‐regulated genes Down‐regulated genes CLUSTER Gene FC p‐value Gene FC p‐value Npy 3.60 0.000955 A930005H10Rik 0.67 1.27E‐08 Lgals1 1.72 0.001412 Snrpf 0.68 7.71E‐08 Pvalb.Atp5k Serpini1 1.65 2.65E‐15 4933409K07Rik 0.69 5.75E‐07 Mustn1 1.58 1.31E‐09 Ccl27a_loc1 0.70 9.24E‐06 Cort 1.57 0.007188 Dynlt1b 0.72 7.11E‐06 Xist 2.63 1.74E‐25 Pld3 0.61 5.57E‐09 BC002163_loc1 1.68 2.28E‐06 Syp 0.73 8.25E‐08 Pvalb.Kcna1 Pcsk2 1.48 0.000360 Tecr 0.736 8.47E‐07 Cacybp 1.43 0.000545 Rpl18 0.737 1.88E‐05 Xist 2.48 3.58E‐26 Sst 0.64 0.008740 Eif2s3y 0.69 1.31E‐17 Pvalb.Gapdh Rpph1 0.745 2.90E‐05 Cntn5 0.748 0.003395 Immp1l 2.07 0.000103 Lrrc4b 0.55 0.019075 1110058L19Rik 1.78 0.000745 Lrrc59 0.58 0.001098 Pvalb.Usmg5 Hsd17b12 1.73 0.004826 Sfxn3 0.59 0.003012 Rab24 1.67 0.021377 Syn1 0.59 0.011252 Coq7 1.81 0.000149 App 0.53 0.032184 Pvalb.Rbp4.Th Tiprl 1.75 0.000121 6330403K07Rik 0.56 0.005845 Xist 1.72 0.000294 Th 0.56 0.029210 Pfdn2 1.66 8.44E‐05 Gaa 0.56 0.016000 Sc4mol 1.52 0.005522 Rab3c 0.58 0.000494 Pvalb.Sst Psip1 1.50 0.009557 Pcsk1n 0.59 0.039577 Slc22a17 0.60 0.005844 Sv2a 0.63 0.000413 Rpl30 2.23 0.000732 S100a10 0.30 0.008700 Hspa12a 2.15 0.002040 Kcna6 0.39 0.006266 Pvalb.Pcp4 Kars 1.97 0.012389 Tmbs10 0.45 0.028202 Nrip3 1.96 0.004840 Ankrd34b 2.06 0.006778 Mrpl13 0.481 0.013538 Pvalb.Snca.Pthlh Pfdn1 2.04 0.015359 Mrpl36 0.484 0.003519 Chrac1 2.02 0.014380 Rnf128 0.49 0.000361 Table 31. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT mice. The first 3‐4 genes were selected in order of decreasing fold change (FC), with a p‐value < 0.05 for up‐ and down‐ regulated genes among PV populations using Flox‐ and WT animals. Values are rounded with two/three decimals for FC and with six decimals for p‐value when needed. The results are extracted from CSV files obtained directly by FindMarkers function from Seurat package (Supp. Data: DE genes (CSV)). Changes are referred to Flox‐ cells.
______
Figure 68. Volcano plots showing significant DE genes found after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT genotypes. Significant up‐ and down‐regulated genes are selected after PV clusters comparison of Flox‐ versus WT genotypes and represented using volcano plots. The X axis shows average log (natural) of fold change and the Y axis displays negative log (base 10) of p‐value. Genes are colored in red if p‐value<0.05; in orange, if passing a distinct avg_logFC threshold for each cluster; in green, if satisfying both restrictions (gene labels). Black dots represent the remaining genes. Up‐regulated genes appear on the right side of the plot due to their positive values for avg_logFC, while down‐regulated genes are located on the left side because they have negative values for avg_logFC.
164
Results
165 Results
To conclude, we carried out a GO analysis on significant DE genes found across PV identities, finding several terms in some PV populations while others remained not associated to any process or function contained in the databases used (Table 32). The significant DE genes were previously selected by Up‐ and down‐regulated genes for GO analysis.R script (Supp. Data: R scripts and files > Seurat integrated analysis) and are stored at Supp. Data: Up‐ and down‐regulated genes for GO (CSV).
It was striking that several of the genes detected during Flox‐ versus WT comparison by clusters were also found as DE genes when comparing Flox‐ and Flox+ PV populations, even showing the same tendency (up‐ or down‐regulated), which clearly reinforced the evidence of these variations. For example, Npy, Lgals1, Mustn1, Cort and Cacybp were up‐regulated, and Ccl27a, Dynltb1, Rpph1 and Pld3 were down‐ regulated. Indeed, some genes were presented since the previous analysis using all PV cells. For instance, comparison of Flox‐ versus Flox+ cells (all PV cells together or by clusters) preserved Npy, Lgals1, Mustn1, Cort and Gapdh among up‐regulated genes, and Dnajc5 and Rpph1 as down‐regulated ones. On the other hand, analysis regarding all PV cells or by identities between Flox‐ and WT mice maintained Npy, Cort, Lgals1 and Musnt1 as up‐regulated genes while Slc22a17, Pcsk1n and Sv2a stayed as down‐regulated (Tables 27, 29, 31).
3) Comparison between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT PV clusters
Finally, we examined DE genes in PV cell identities between Flox+ and WT genotypes. The R code used for this analysis is described in Seurat integrated analysis ‐PV differential expression (DE) – Flox+ vs WT.R script (Supp. Data: R scripts and files > Seurat integrated analysis).
We followed the same methodology as mentioned before during the two previous analyses. P‐ value threshold was also established below 0.05 and the first significant 3‐4 genes with the highest fold change in both directions (up‐ and down‐regulated) were selected for the study (Table 33; Figure 69).
______
Table 32. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/‐ and PVcre:Ai27D:Dnajc5WT genotypes. Total up‐ and down‐regulated genes found after comparing PV identities in Flox‐ vs WT cells were used for Gene Ontology analysis using ClueGO plugin from Cytoscape software. Only processes and pathways (terms) with p‐ value≤0.05 were chosen by ClueGO, which created GO groups that englobed associated terms. These GO groups are ordered by the lowest GO group p‐value (the most significant p‐value corrected with Bonferroni step down) and only the term with the highest significance (term p‐value corrected with Bonferroni step down) within each GO group is shown in this table. Data extracted from raw result files at Supp. Data: ClueGO analysis (XLS).
166
Results
Cluster Terms (up‐regulated genes) Terms (down‐regulated genes) Pvalb.Atp5k ‐ No functions related. ‐ No functions related. ‐ Oxidative phosphorylation. Pvalb.Kcna1 ‐ Mitochondrial respiratory chain complex ‐ No functions related. assembly. Pvalb.Gapdh ‐ No functions related. ‐ No functions related. ‐ Negative regulation of stress fiber assembly. Pvalb.Usmg5 ‐ Ribosome. ‐ Negative regulation of Ras protein signal transduction. ‐ Positive regulation of phospholipase ‐ Synaptic transmission, GABAergic. Pvalb.Rbp4.Th activity. ‐ Axo‐dendritic transport.
‐ Regulation of synaptic vesicle cycle. ‐ Positive regulation of phospholipase Pvalb.Sst ‐ No functions related. activity. ‐ Bacterial invasion of epithelial cells. ‐ Ubiquitin mediated proteolysis. ‐ “De novo” protein folding. ‐ Microtubule polymerization or depolymerization. ‐ Negative regulation of supramolecular fiber organization. Pvalb.Pcp4 ‐ Ribosomal large subunit assembly. ‐ Negative regulation of binding. ‐ RNA transport. ‐ Positive regulation of protein ubiquitination. ‐ Arrhythmogenic right ventricular cardiomyopathy (ARVC). ‐ ncRNA metabolic process. ‐ Ribosome. ‐ Nucleotidyltransferase activity. ‐ Negative regulation of protein homooligomerization. ‐ Positive regulation of viral life cycle. ‐ Negative regulation of response to endoplasmic reticulum stress. ‐ Cellular response to nerve growth factor ‐ Mitochondrial gene expression. Pvalb.Snca.Pthlh stimulus. ‐ Autophagy of mitochondrion. ‐ Negative regulation of phosphoprotein phosphatase activity. ‐ Glycolipid metabolic process. ‐ Cortical actin cytoskeleton organization. ‐ Positive regulation of protein dephosphorylation. ‐ Amino sugar and nucleotide sugar metabolism.
167 Results
Up‐regulated genes Down‐regulated genes CLUSTER Gene FC p‐value Gene FC p‐value Rpph1 1.72 3.87E‐08 Gm13826 0.75 0.000112 Pvalb.Atp5k Serpini1 1.65 4.65E‐15 Erdr1 0.76 5.00E‐06 Spock3 1.59 6.34E‐11 Xist 0.76 2.81E‐05 BC002163_loc1 1.61 7.20E‐09 Gapdh 0.65 3.28E‐05 Pvalb.Kcna1 Dnajc5 1.59 3.84E‐15 Cort 0.67 0.013587 Malat1 1.53 9.24E‐07 Scg5 0.70 1.20E‐16 Malat1 1.53 6.89E‐06 Tubb3 0.70 7.25E‐07 Pvalb.Gapdh Dynlt1b 1.44 5.20E‐06 Ap1s1 0.72 8.78E‐06 Tmem181a 1.34 0.000272 Syp 0.74 1.59E‐06 mt‐Tt 2.14 0.000162 Lrrc4b 0.49 0.004648 Pvalb.Usmg5 Mettl5 1.83 0.028336 Ttc9b 0.54 0.002556 Immp1l 1.76 0.028336 Aplp2 0.55 0.014544 Calb1 2.56 0.038978 Th 0.42 0.008419 Fopnl 1.83 0.000505 Ctsd 0.43 9.24E‐07 Pvalb.Rbp4.Th Snw1 1.74 0.000564 6330403K07Rik 0.46 0.001169 Npy 0.48 0.044619 Tcap 1.72 0.015348 Pcsk1n 0.41 6.48E‐06 Pvalb.Sst Tmsb10 1.63 0.014899 Slc22a17 0.54 0.000647 Gm14305 1.61 0.008130 Atp1a3 0.55 0.014268 Tuba1c 2.43 0.030157 Cox6a2 0.36 0.038233 Pvalb.Pcp4 Ubb 2.16 0.017317 Cort 0.37 0.011728 Adnp 2.14 0.000214 Kcnc2 0.42 0.034784 Tmed10 2.40 0.013533 Npy 0.28 0.001178 Pvalb.Snca.Pthlh Ywhag 2.11 0.017335 Fbxo44 0.43 0.004574 Nucks1 2.09 0.030474 Map1lc3a 0.45 0.026799 Table 33. Summary of significant DE genes found during the comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT mice. The first 3‐4 genes were selected in order of decreasing fold change (FC), with a p‐value < 0.05 for up‐ and down‐ regulated genes among PV populations using Flox+ and WT animals. Values are rounded with two decimals for FC and with six decimals for p‐value when needed. The results are extracted from CSV files obtained directly by FindMarkers function from Seurat package (Supp. Data: DE genes (CSV)). Changes are referred to Flox+ cells.
All bibliographic information regarding these significant DE genes, in a brain‐related context, is disclosed in S15. The results of this analysis were in concordance with our previous results using all PV cells between the same genotypes. For example, Malat1, Rpph1 and Gm14305 were the only up‐regulated genes found using all PV cells, and the increase in those genes were also observed in some clusters. On the other hand, Npy, Atp1a3, Slc22a17 and Syp coincided as down‐regulated genes in both analyses ______
Figure 69. Volcano plots showing significant DE genes detected after comparison of PV identities between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT genotypes. Significant up‐ and down‐regulated genes are selected after PV clusters comparison of Flox+ versus WT genotypes and represented using volcano plots. The X axis shows average log (natural) of fold change and the Y axis displays negative log (base 10) of p‐value. Genes are colored in red if p‐value<0.05; in orange, if passing a distinct avg_logFC threshold for each cluster; in green, if satisfying both restrictions (gene labels). Black dots represent the remaining genes. Up‐regulated genes appear on the right side of the plot due to their positive values for avg_logFC, while down‐regulated genes are located on the left side because they have negative values for avg_logFC.
168
Results
169 Results
(Tables 27, 33). Interestingly, we observed that some genes were already found during the two previous PV cluster comparisons, although sometimes in different directions. For example, Rpph1 and Dynlt1b were down‐regulated in both Flox‐ versus Flox+ or WT analyses by clusters, which reinforced the result; instead, they were up‐regulated in the comparison between both controls by clusters. Ap1s1, Cort and Gapdh looked completely different, they were up‐regulated in both Flox‐ versus Flox+ or WT analyses using all PV cells (and also by clusters between Flox‐ and Flox+ genotypes), confirming the result; but they were down‐ regulated in the comparison between both controls by clusters. On the other hand, Immp1l remained up‐ regulated in the same identity (Pvalb.Usmg5) in both Flox+ and Flox‐ genotypes when compared to WT PV cells; conversely, Th and Lrrc4b were down‐regulated in the same population (Pvalb.Rbp4.Th and Pvalb.Usmg5, respectively) in Flox‐ and Flox+ genotypes compared to WT. However, we should be cautious when interpreting some previous results, because several genes that were found as DE in Flox‐ cells after comparison with the Flox+ genotype, they looked in the other direction after comparing both controls. This is the case of Cox6a2, Ttc9b or Tcap, within others, that showed a down‐regulation in Flox+ genotype compared to WT, and therefore, they appeared up‐regulated in Flox‐ cells compared to Flox+ genotype, which might be due to the increase in the latter. Intriguingly, Dnajc5 was increased in Pvalb.Kcna1 population in Flox+ cells compared to WT, which also compromised the down‐regulation of Dnajc5 in the same identity in Flox‐ cells compared to Flox+ (Table 27, 29, 31, 33).
In the same way, Npy, the most up‐regulated gene in Flox‐ cells when compared to the Flox+ genotype, looked down‐regulated in Flox+ cells compared to WT, which might justify this increase. This result, then, might be taken carefully. In detail, when comparing all PV cells, Npy was significantly up‐ regulated in Flox‐ cells compared to Flox+, but down‐regulated in Flox+ cells when compared to WT. No significant differences were found in Flox‐ versus WT genotypes when taking all PV cells (Table 27). On the other hand, in the analysis by PV clusters, Npy was significantly up‐regulated in Flox‐ vs Flox+ cells in several PV identities (Pvalb.Atp5k, Pvalb.Rbp4.Th, Pvalb.Sst and Pvalb.Snca.Pthlh), only significantly up‐regulated in Pvalb.Atp5k in Flox‐ vs WT PV cells, but significantly down‐regulated in Flox+ vs WT genotypes, importantly, in two of the same populations that was up‐regulated in Flox‐ cells compared to Flox+ (Pvalb.Rbp4.Th and Pvalb.Snca.Pthlh). These results compromised the relevance of the Npy increase in these two latter populations, since it might be due to a decrease of the gene in the heterozygous Flox+ mouse. Conversely, this fact reinforces the finding of the Npy up‐regulation in Pvalb.Atp5k identity, which coincided in the two Flox‐ analyses compared to both controls and, importantly, this PV population composed the biggest PV group (Table 25, 29, 31, 33; Supp. Data: DE genes (CSV)).
In addition, other DE genes were revealed by this analysis. We highlighted microtubule‐associated genes (Tuba1c, up‐regulated; Tubb3, Map1lc3a, down‐regulated), genes related to ubiquitination (Ubb, up‐regulated; Fbxo44, down‐regulated), to neuroprotection (Adnp, up‐regulated), to protein trafficking
170
Results and lysosomal functions (Tmed10, Ywhag, up‐regulated; Ctsd, down‐regulated), a voltage‐gated potassium channel related to seizures (Kcnc2, down‐regulated), a chaperone (Scg5, down‐regulated), and a member of the amyloid precursor protein (Aplp2, down‐regulated), within others (Table 33; S15; Supp. Data: DE genes (CSV)).
Finally, we performed a GO analysis on significant DE genes found across PV identities. Several terms were related to some PV populations while others remained not associated to any process or function (Table 34). The significant DE genes were previously selected by Up‐ and down‐regulated genes for GO analysis.R script (Supp. Data: R scripts and files > Seurat integrated analysis) and stored at Supp. Data: Up‐ and down‐regulated genes for GO (CSV).
Cluster Terms (up‐regulated genes) Terms (down‐regulated genes) Pvalb.Atp5k ‐ Cation‐transporting ATPase activity. ‐ No functions related.
Pvalb.Kcna1 ‐ No functions related. ‐ No functions related.
Pvalb.Gapdh ‐ No functions related. ‐ No functions related. ‐ Nicotinamide nucleotide metabolic process. ‐ Modification of postsynaptic actin Pvalb.Usmg5 ‐ Protein export. cytoskeleton. ‐ Postsynaptic density organization. ‐ Autophagy of mitochondrion. ‐ Ubiquitin protein ligase binding. ‐ Regulation of neurotransmitter levels. ‐ Protein localization to membrane. ‐ Adrenergic signaling in cardiomyocytes. ‐ ATP metabolic process. ‐ Gap junction. ‐ Glycine transport. ‐ Maternal process involved in female pregnancy. ‐ Oxidative phosphorylation. ‐ Lysosome. Pvalb.Rbp4.Th ‐ Respiratory electron transport chain. ‐ Protein processing in endoplasmic ‐ Renal cell carcinoma. reticulum. ‐ Vacuole organization. ‐ Response to ketone. ‐ MHC class I protein binding. ‐ Negative regulation of lipid catabolic process. ‐ Neuron recognition. ‐ Response to ammonium ion. ‐ GABAergic synapse.
171 Results
‐ Establishment of vesicle localization. ‐ ATPase activity, coupled to transmembrane movement of substances. ‐ Central carbon metabolism in cancer. ‐ NADH metabolic process. ‐ Axonal transport. ‐ Arginine and proline metabolism. ‐ Regulation of protein localization to cell surface. ‐ Regulation of tumor necrosis factor ‐ Ribosome. secretion. Pvalb.Sst ‐ Nucleosome binding. ‐ Regulation of ruffle assembly. ‐ Cysteine and methionine metabolism. ‐ Vasopressin‐regulated water reabsorption. ‐ Lysosome. ‐ Cerebellar cortex development. ‐ Central nervous system neuron development. ‐ Regulation of response to endoplasmic reticulum stress. ‐ Regulation of insulin receptor signaling pathway. ‐ Cofactor biosynthetic process. ‐ Lamellipodium organization. ‐ Protein export. ‐ Dopaminergic synapse. Pvalb.Pcp4 ‐ Aminoacyl‐tRNA ligase activity. ‐ Positive regulation of ATPase activity. ‐ Renal cell carcinoma. ‐ Regulation of type B pancreatic cell apoptotic process. ‐ Insulin secretion. ‐ ncRNA metabolic process. ‐ Proteasomal protein catabolic process. ‐ Mitochondrial gene expression. Pvalb.Snca.Pthlh ‐ No functions related. ‐ Response to endoplasmic reticulum stress. ‐ Cellular response to starvation. ‐ Protein localization to chromosome. Table 34. GO analysis for DE genes obtained by comparison of PV clusters between PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5WT genotypes. All significant up‐ and down‐regulated genes found after comparing PV identities in Flox+ vs WT cells were used for Gene Ontology analysis using ClueGO plugin from Cytoscape software. Only processes and pathways (terms) with p‐value≤0.05 were chosen by ClueGO, which created GO groups that englobed associated terms. These GO groups are ordered by the lowest GO group p‐value (the most significant p‐value corrected with Bonferroni step down) and only term with the highest significance (term p‐value corrected with Bonferroni step down) within each GO group is shown in this table. Data extracted from raw result files at Supp. Data: ClueGO analysis (XLS).
172 DISCUSSION
Discussion
1. The PVcre:Ai27D:Dnajc5flox/‐ as a mouse model to study CSPα/DNAJC5 in PV interneurons
We carried out a set of experiments to validate our PVcre:Ai27D:Dnajc5flox mouse line as a useful model to study the function of CSPα/DNAJC5 in PV neurons. We have found that the analysis of genomic DNA and mRNA supported Cre‐induced recombination at the floxed Dnajc5 locus (Figure 17B, C). We found a 245 bp gDNA band only in tdTomato+ cells from both PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice as expected for a successful Cre‐recombinase activity at the Dnajc5 floxed allele. No gDNA band of 245 bp was observed in tdTomato‐negative cells, confirming that Dnajc5 floxed gene was only deleted in Cre‐expressing cells, that is, in PV interneurons. On the other hand, the cDNA analysis after RT‐PCR of mRNA revealed the expected band of 134 bp only in tdTomato+ cells. In contrast, in tdTomato‐negative cells only the 348 bp band appeared, as expected, for the non‐recombined floxed allele. We, nevertheless, also found a residual band corresponding to the non‐recombined Dnajc5 floxed allele in sorted tdTomato+ fractions at gDNA level. At the level of mRNA, we also observed a high molecular‐weight fragment (348 bp) present in tdTomato+ cells from PVcre:Ai27D:Dnajc5flox/+ animals, however, mRNAs from the Dnajc5 WT and the non‐recombined Dnajc5 floxed alleles are suitable templates to generate this band. The presence of this band in tdTomato+ cells from PVcre:Ai27D:Dnajc5flox/‐ mice would not be expected upon isolating a pure population of Cre‐recombinase expressing cells having the recombined Dnajc5 floxed allele in all of them. Nevertheless, we still found a genomic 1029 bp‐band and a mRNA 348 bp‐band originating from the non‐recombined Dnajc5 floxed allele (Figure 16; S1). Several possible explanations could account for these results:
1) tdTomato+ labelling occurs in PV‐negative cells that contain the non‐recombined Dnajc5 floxed allele. This is unlikely because our PV‐tdTomato colocalization analysis revealed that almost 90% of PV cells were also positive for tdTomato and only few tdTomato+ cells did not colocalize with PV interneurons (4% in controls and 1% in mutants) (Figure 6); similar values haven been previously reported (Hippenmeyer et al., 2005).
2) FACS tdTomato+ cell isolation was not restrictive enough to ensure the selection of only cells with high expression of tdTomato. This is also unlikely becauser ou sorting strategy was actually highly restrictive for tdTomato expression (above 102 for tdTomato detection) (Figure 3A). Even if we had collected few tdTomato low‐expressing cells that might had not comprised PV interneurons, the amount of non‐ recombined Dnajc5 floxed alleles should have been too low to be detected by PCR.
3) tdTomato+ cells might be attached to tdTomato‐negative cells. We cannot rule out that, for example, oligodendrocytes attached to PV positive cells (Micheva et al., 2016) might be co‐purified in tdTomato+ fractions.
175 Discussion
4) There are tdTomato+ Cre‐expressing cells in which the Dnajc5 floxed allele was not recombined. We cannot rule out that the Cre‐recombinase is more efficient at the ROSA26 than at the Dnajc5 locus.
In any case, our western blot analysis in PVcre:Ai27D:Dnajc5flox/‐ animals indicates a reduction in the levels of CSP/DNAJC5 (Figure 24) that appears only as a partial reduction because CSP/DNAJC5 is a pan‐ neuronal protein. Furthermore, there are no signs of any truncated protein version (Figure 19), supporting the notion that a complete loss in protein translation occurs. Furthermore, the validity of genomic strategy is also supported by the western blot analysis in UBC‐cre‐ERT2:Dnajc5flox/flox mice (Figure 18).
2. PVcre:Ai27D:Dnajc5flox/‐ mice suffer from a progressive age‐dependent neurological phenotype with hyperactivity and ataxia
PVcre:Ai27D:Dnajc5flox/‐ animals and their littermate controls (PVcre:Ai27D:Dnajc5flox/+) were analyzed from 0‐8 months for early and late mortality, finding no significant differences between genotypes (Figure 21A). Conversely, an obvious reduction in body size was observed in mutant mice since P30, which led us to measure mouse body weight from 1 to 8 months. Both male and female PVcre:Ai27D:Dnajc5flox/‐ animals showed a significant loss in body weight from initial time points (P37 for males and P51 for females) compared to PVcre:Ai27D:Dnajc5flox/+ controls (Figure 21B, D), likely secondary to the neurological phenotype.
PVcre:Ai27D:Dnajc5flox/‐ animals displayed a characteristic phenotype easily detectable by visual inspection, mainly characterized by hyperactivity, lumbar lordosis, back hump and impaired hindlimb movements leading to ataxia (Supp. Data: Neurological phenotype (videos)). Open field measurements confirmed hyperactivity: velocity, track length, activity and number of ambulations (spontaneous short‐ term accelerations) were increased in PVcre:Ai27D:Dnajc5flox/‐ animals compared to PVcre:Ai27D:Dnajc5flox/+ controls, both in males and females (Figure 22). In contrast, the Biobserve software detected a reduction in normal head and tail movements in mutant mice. This is not surprising because, indeed, these mice presented irregular head and tail movements that were obvious by visual inspection. The abnormal head movements including upward gazes resembled the previously described phenotype of stargazer mutants. Stargazer mice were initially recognized for a distinctive head‐tossing, ataxic gait and recurrent spike‐wave seizures characteristic of absence seizures in humans (Letts, 2005; Letts et al., 1998; Noebels et al., 1990). Stargazer mice bears a single, recessive mutation in Cacng2 gene, which encodes stargazin protein, a subunit of voltage‐dependent calcium channel (VDCC). Stargazin disruption is linked to a reduction in GABA release and less GABAergic synapses at cerebellum (Letts, 2005; Osten and Stern‐Bach, 2006). The observation of the stargazer‐like phenotype in PVcre:Ai27D:Dnajc5flox/‐ animals could then be consistent with a dysfunction in GABAergic synaptic transmission and perhaps a cerebellar phenotype that deserves to be analyzed in future studies. The normal wall‐distance parameter analyzed during the open field could
176 Discussion indicate that there is no increase in anxiety, however, the analysis of complex behaviors in these mice might not be trivial. Indeed, a more detailed behavioral characterization, including the investigation of epileptic phenotypes, is currently in progress in collaboration with Prof. Agnès Gruart and Prof. José M. Delgado at the University Pablo de Olavide.
In any case, the major features of the neurological phenotype are age‐dependent. At 8 months of age, hindlimb ataxic deterioration increased and the mobility loss became accentuated in mutant mice.
It is interesting to highlight that changes in locomotor behavior, such as hyperactivity or ataxia, are often linked to epileptic seizures in several mouse models and humans. There are studies that have revealed candidate genes that, upon loss‐of‐function mutations or deletion, directly lead to seizures and locomotor deficits in mice. Genes coding for ion channels are representative examples: Scn1a (Dutton et al., 2014; Tatsukawa et al., 2018), Kcna1 (Eunson et al., 2000; Jiang et al., 2016; Li et al., 2012; Smart et al., 1998) and Kcna2 (Robbins and Tempel, 2012; Syrbe et al., 2015). These observations suggested that potential changes in gene expression could be happening in this mouse model, and therefore, single‐cell RNA sequencing of PV interneurons could be of particular interest in these mice.
3. Selective reduction of presynaptic proteins in PVcre:Ai27D:Dnajc5flox/‐ mice
Analysis of protein levels in total cortex by western blot revealed reduced levels in presynaptic proteins in PVcre:Ai27D:Dnajc5flox/‐ mice. At 8 months, synaptotagmin‐2 and Hsc70 were significantly decreased, and a reduction in SNAP25 was also appreciated (Figure 24B). SNAP25 decrease is not surprising. SNAP25 was discovered as a substrate of CSPα/DNAJC5, and its expression is significantly reduced in Dnajc5 KO animals around P30‐40 (Chandra et al., 2005; Rozas et al., 2012; Sharma et al., 2011, 2012). A decrease in Hsc70 levels (Sharma et al., 2011, 2012), a component of the trimeric complex CSPα‐ SGT‐Hsc70 located at the synaptic vesicle surface (Tobaben et al., 2001), has been also previously reported (Chandra et al., 2005); as well as a reduction in Syt2 in Dnajc5 KO conventional mice at the hippocampus (García‐Junco‐Clemente et al., 2010), where fast‐spiking Syt2‐expressing PV interneurons are highly dependent on CSPα/DNAJC5 to maintain their function and integrity. Decreases in Syt2, Hsc70 and SNAP25 in the cortex of Dnajc5 KO mice were also detected in our work by western blot (Figure 15), according to those studies. A remarkable reduction in the levels of CSPα/DNAJC5 was found by western blot in PVcre:Ai27D:Dnajc5flox/‐ animals compared to PVcre:Ai27D:Dnajc5flox/+ controls, which was in principle unexpected given that Dnajc5 gene was only deleted from PV interneurons (Figures 19, 24). This reduction might be consistent with high protein levels of CSPα/DNAJC5 in PV interneurons, which normally form complex axo‐somatic synapses (baskets) on multiple postsynaptic targets (Hu et al., 2014).
177 Discussion
4. Molecular signs of presynaptic degeneration in the cortex of PVcre:Ai27D:Dnajc5flox/‐ mice
A goal of this thesis is to investigate the role of CSPα/DNAJC5 in the maintenance of synaptic connections established by PV interneurons. We focused our attention on layer II/III of motor cortex. With perspective for complementary studies, this area is suitable to analyze in vivo the role of PV neurons in motor behavior using two‐photon microscopy as previously described (García‐Junco‐Clemente et al., 2017, 2019). Besides motor cortex, alterations of PV interneurons in other brain areas such as the striatum and the cerebellum could be contributing to the observed motor phenotype. Regulation of motor signal transmission is coordinated by cerebellum (Arshavsky et al., 1983), and walking ataxia comprises a characteristic signal of cerebellar damage (Morton and Bastian, 2004). Indeed, the study of movement control mechanisms is an area of great interest (Goulding, 2009; Muzzu et al., 2018). The future analysis of presynaptic dysfunction in the cerebellum in PVcre:Ai27D:Dnajc5flox/‐ might provide interesting findings and we plan to explore the function of PV+ Purkinje cells, which might be compromised upon genetic removal of CSPα/DNAJC5.
Strikingly, we found PV somata to be normal in number and size at any age (Figure 26). The exception was a significant increase in the number of PV somata in PVcre:Ai27D:Dnajc5flox/‐ animals at 2 months, which was not expected. We do not have a biological explanation for that, so we decided not to explore that observation further in this thesis. In any case, our observations indicate that PV cells remain alive at the cortex and that their viability is not compromised upon CSPα/DNAJC5 loss. We further analyzed the number of PV+ and Syt2+ synaptic puncta in layer II/III of motor cortex to determine whether PV terminals were affected, finding a significant reduction in PV boutons labelled by both markers in PVcre:Ai27D:Dnajc5flox/‐ mice at 2 and 8 months (Figure 25). Furthermore, it was detected that this PV synaptic loss followed an age‐dependent progressive alteration (Figure 25B, D) that may become relevant at different time points for each individual, as shown by differences in mouse pairs (Figure 25C, E). In addition, we found a notable thinning and disorganization in perisomal PV+ and Syt2+ labeling in mutant mice (Figure 12A), which may be reflecting PV terminals degeneration. Furthermore, only in the mutant, it was evident to find synaptic PV+ puncta often distributed along linear structures of unknown origin. These structures somehow resemble cartridges found at axo‐axonic synapses formed by chandelier cells (Chiu et al., 2002; Fish et al., 2013; Lewis and Lund, 1990). In the future, we will explore this intriguing finding and investigate if, indeed, any relationship with chandelier cell exists.
In summary, our results indicate that PV cells without CSP/DNAJC5 are present at cortex, at least up to 8 months of age, even after their presynaptic terminals present signs of progressive age‐dependent degeneration. This finding is somehow surprising, because we would expect a decrease in neuronal
178 Discussion viability secondary to synaptic degeneration that could compromise the uptake and retrograde transport of trophic factors (Gu et al., 2018; McAllister et al., 1999). One possibility is that the reduction of synaptic puncta is not total and PV neurons could still receive trophic factors from spared synapses. Resistance to neurodegeneration of PV neurons with synaptic loss has been studied in the context of human brain disorders. It is known that a loss in PV synaptic puncta without changes in PV cell number occurs in some neurodegenerative disorders. For example, a 30‐35% decrease in PV chandelier terminals is found in layer II/III of neocortex in human brains of Alzheimer’s disease patients compared to control brains (Fonseca et al., 1993), while no reduction in PV somata number is observed (Ferrer et al., 1991). These findings propose that PV interneurons are quite resistant to neurodegeneration in AD, which may also happen in other neurodegenerative scenarios, such as in our conditional KO mouse model. In contrast, it is controversial that other works report a significant loss of PV neurons in the hippocampus and cortex in AD (Takahashi et al., 2010), and even find a significant reduction in PV cell size (Arai et al., 1987), suggesting that PV neurons are not that resistant to insults underlying neurodegeneration. Reduced density of PV neurons and lower PV levels have been also described in schizophrenia (Cabungcal et al., 2013; Enwright et al., 2016). Moreover, an age‐dependent increase in PV cell number in an AD mouse model has been also reported, which looks unexpected, and was proposed as an alteration in PV differentiation (Lemmens et al., 2011). In any case, although we could not detect reduction in PV neurons, we considered further electrophysiological analysis to evaluate neuronal and synaptic functionality.
5. Subtle alterations of excitability in cortical PV interneurons lacking CSPα/DNAJC5
To assess in more depth the apparent degeneration of PV terminals in vivo, we used electrophysiological recordings to study PV functional properties. Our results show that there is no variation in resting membrane potential (RMP) nor time constant (π) between control and PVcre:Ai27D:Dnajc5flox/‐ PV cells at 2 or 8 months (Table 16; Figure 27A). However, when analyzing input resistance of PV cells, we found a significant decrease of this parameter in PVcre:Ai27D:Dnajc5flox/‐ PV cells at 8 months compared to control neurons (Figure 27B, C). Input resistance decrease could indicate an increase in the number of open channels in the cell membrane. Hence, one possibility to explain our results is that, at 8 months, PV cells from PVcre:Ai27D:Dnajc5flox/‐ animals overexpress ion channels. Increased size of cell body, even with a normal ion channel density, could also explain a lower input resistance; however, this hypothesis was not supported because in average PV interneurons capacitance was similar in control and mutant PV cells (Figure 27D). We, however, detected some few cells with higher capacitance in PVcre:Ai27D:Dnajc5flox/‐ mice but could not confirm changes in PV somata size assessed by morphological measurements (Figure 26B) and, therefore, concluded that mutant cells were not bigger than control cells. Our results probably indicate that, in the absence of CSPα/DNAJC5, the expression of channels in the
179 Discussion membrane of PV cells increases. This would be consistent with previous studies proposing that CSP/DNAJC5 limits the protein levels of the large conductance, calcium‐activated K+ (BK) channel by unknown mechanisms in which CSP/DNAJC5 could be involved in the membrane retrieving and/or lysosomal mediated degradation of those channels (Ahrendt et al., 2014; Donnelier et al., 2015; Kyle et al., 2013). Although these channels have not been specifically reported to be expressed in PV interneurons, other channels (Pelkey et al., 2017) could potentially also be regulated by CSP/DNAJC5 and, therefore, be increased in the interneurons of PVcre:Ai27D:Dnajc5flox/‐ mice. We nevertheless detected a faster recovery at the end of action potentials following afterhyperpolarization (AHP) at 8 months old mice (Figure 28C). This observation cannot be simply explained by an increased number of potassium channels. A faster closing rate and/or the existance of different K+ channels subunits could also explain that phenomenon. On the other hand, if the number of open channels was increased, we should detect a change in resting membrane potential that we did not detect (Table 16). The amplitude of the afterhyperpolarization peak was also normal (Figure 28C). In order to clarify this point, ionic currents in PV interneurons need to be studied in future experiments.
We did not find differences in AP amplitude between control and mutant cells. We, however, found a reduction in AP amplitude when comparing PV cells at 8 versus 2 months common to both genotypes, suggesting that a general age‐dependent decrease in AP amplitude in PV cells occurs (Figure 28B). In any case, AP amplitude values are consistent with previous publications analyzing the PV fast‐ spiking phenotype (Hu et al., 2014; McCormick et al., 1985). We next analyzed the ability of mutant PV neurons to fire APs at the typical high frequencies of fast‐spiking neurons. Indeed, the faster repolarization following AHP (Figure 27C) could predict quicker recovery to fire again earlier than a control neuron, but this did not happen. PV neurons lacking CSP/DNAJC5 were able to reliably fire at high frequencies ( ̴300 Hz, Figure 29B). Indeed, we found that, at 8 but not at 2 months, PV cells lacking CSPα/DNAJC5 were less excitable: they needed a greater current stimulus to fire the first action potential. Hence, they presented increased rheobase (the minimal current injection needed to generate at least one action potential with 50% likelihood) (Figure 29D). Interestingly, this feature is also consistent with a lower input resistance as stated above. Furthermore, a careful analysis in 8‐month mutant mice of the relationship between the number of action potentials generated versus injected current showed that, upon reaching rheobase, there is still a lag during which only one single action potential is triggered even for increasing levels of injected current (Figure 29B, C). Interestingly, such a behavior could be explained if Kv1.1 channels were increased in the mutant neurons, because these channels located at the axon initial segment dampen near‐threshold excitability of neocortical fast‐spiking GABAergic interneurons (Goldberg et al., 2008).
In conclusion, intrinsic and active properties of PV interneurons lacking CSPα/DNAJC5 reveal that they maintained their functional properties and were able to fire actions potentials. Nonetheless, these
180 Discussion cells presented (1) a lower membrane input resistance, that may be a consequence of an increased number of channels expressed in their membrane; (2) a reduced AHP recovery time, indicating that PV cells repolarize faster; and (3) a higher rheobase, which means that these cells are less excitable because they need a higher current stimulus to fire the first action potential. The lower input resistance, the higher rheobase and the lag to reach the high firing regime are intrinsically consistent with an increase in K+ conductance, not translated into changes in resting membrane potential though. This phenotype progresses between 2 and 8 months and suggests that CSPα/DNAJC5 might have a role to control membrane trafficking of ion channels that deserves further investigations, including the characterization of ionic currents using the patch‐clamp technique.
6. Normal excitatory synaptic inputs on PV interneurons lacking CSPα/DNAJC5
It has been previously shown how the pattern of synaptic inputs that a particular neuron receives is sensitive to the activity levels of that neuron (Burrone et al., 2002). We recorded EPSPs to measure the functional synaptic input on PV neurons. EPSPs on PV cells were apparently normal in size and frequency PVcre:Ai27D:Dnajc5flox/‐ mice. We found an age‐dependent progressive increase in EPSP frequency in both control and mutant PV neurons, perhaps indicating a higher number of excitatory synapses at 8 months (Figure 30).
7. GABA release alterations in PV interneurons from PVcre:Ai27D:Dnajc5flox/‐ mice
We detected two major defects in mIPSCs evoked by PV cells on pyramidal neurons at layer II/III of motor cortex in PVcre:Ai27D:Dnajc5flox/‐ mice: a decrease in frequency and a decrease in amplitude (Figure 31). We assume that mIPSCs recorded on pyramidal neuron somata proceed from local (layer II/III) PV cells, given that PV interneuron innervations of postsynaptic cells are mostly perisomatic (Freund and Buzsáki, 1996; Hu et al., 2014; Pelkey et al., 2017), in contrast to Sst‐expressing interneurons that target cells at dendrites (Harris and Mrsic‐Flogel, 2013). Parvalbumin‐expressing GABAergic cells have approximately a 100% connection probability to neighboring pyramidal cells (Harris and Mrsic‐Flogel, 2013). mIPSCs, also termed “miniature”, were recorded upon AP blocking with tetrodotoxin (TTX) to measure the spontaneous GABA release of a single synaptic vesicle.
The mIPSC frequency recorded in our control mice was twice higher compared to previous measurements in layer V of visual cortex (Nieto‐Gonzalez et al., 2011). Those apparently high values were maintained in low‐calcium conditions (reduction from 2.5 to 1 mM CaCl2) (data not shown). In any case, the decrease in mIPSC frequency that we found in PVcre:Ai27D:Dnajc5flox/‐ mice (Figure 31) is consistent with a progressive reduction in the number of GABAergic synapses and expected according to the reduction in the number of puncta expressing synaptotagmin‐2 and PV (Figure 25). The reduction in mIPSC amplitude is, in contrast, surprising. The kinetic analysis of mIPSCs revealed a remarkable slowdown of the
181 Discussion decay time (weighted tau) and increased half‐width at 2 and 8 months when compared to control PV cells (Figure 31). Several explanations are considered to account for these results:
1) A presynaptic mechanism: The vesicular content of GABA is reduced. It has been proposed that CSP/DNAJC5 participates with Hsc70 to anchor L‐glutamic acid decarboxylase to the synaptic vesicles and to facilitate the coupling of GABA‐synthesis with GABA‐transport into the vesicles mediated by the vesicular transporter VGat (Hsu et al., 2000). Accordingly, the absence of CSP/DNAJC5 would uncouple the synthesis and the vesicular loading of GABA resulting in a lower amount of GABA per vesicle. This explanation, however, would not be supported by our previous work on hippocampal neuronal cultures from CSP/DNAJC5 KO mice, in which the mIPSC amplitude was normal (García‐ Junco‐Clemente et al., 2010), although it is difficult to compare these results using different techniques (cortical slices versus hippocampal cultures). Indeed, the phenotype could be progressive and becomes only evident after 2 months of age. In order to clarify this issue, it would be required to analyze spontaneous GABA release in cortical slices of CSP/DNAJC5 KO mice as old as possible (normally they die around P30) and compare it with recordings in PVcre:Ai27D:Dnajc5flox/‐ at the same age.
2) A postsynaptic mechanism: The number of postsynaptic GABA receptors is reduced and/other GABA subunits are expressed. This could happen as secondary effect to presynaptic degeneration and, if that was the case, it would be an interesting opportunity to study molecular mechanisms of synaptic remodeling and trans‐synaptic signaling in the context of degeneration (Krueger‐Burg et al., 2017). Interestingly, deletion of the cell adhesion molecule neuroligin‐2 selectively decreases mIPSC amplitude originating from fast‐spiking but not from somatostatin‐positive interneurons (Gibson et al., 2009). The slower decay (weighted tau) could be consistent with slower kinetics of GABA channels to close, either by the intrinsic properties of the channel or because the higher affinity for GABA. This scenario would be compatible with a change in subunit composition of GABA receptors in PVcre:Ai27D:Dnajc5flox/‐ pyramidal neurons that would modify the biophysical properties of the channel and, therefore, the opening/closing kinetics (Enna and Möhler, 2007). A total of 19 different genes code for GABA receptor subunits that assemble to form pentameric receptors (Sigel and Steinmann, 2012). The rules by which pentamers compositions are determined and how they depend on scaffold and cell adhesion molecules are starting to be understood (Krueger‐Burg et al., 2017; Martenson et al., 2017). The subunit composition of GABA‐A receptors involved in IPSC kinetics has been broadly studied, however, not only a specific subunit but several subunits composing this channel have nbee related to slow IPSC kinetics. For example, the presence of the δ‐subunit in a GABA‐A channel elongates the channel opening duration and the recovery of GABA‐evoked currents in transfected mouse fibroblast cells, proposing this subunit to play a role in preventing seizures (Saxena and
182 Discussion
Macdonald, 1994); on the other hand, in amygdala, the combination of α2‐ and γ1‐subunits slow both the activation and deactivation rates of GABA‐A receptors (Dixon et al., 2014). Studies using diazepam have also suggested that receptors containing the α4‐, α5‐ or α6‐subunits may contribute to slowly
rising and decaying IPSCs (GABA‐A,slow) (Pearce, 1993), whereas receptors containing the γ2‐subunit in association with α1‐, α2‐, or α3‐subunits preferentially mediate fast synaptic transmission (Enna and Möhler, 2007).
3) The smaller mIPSCs are not coming from PV interneurons. In case the presynaptic PV terminals were suffering from a severe degeneration, implicating synaptic silencing, the minis originated at dendrites upon GABA release from non‐PV neurons (e.g. somatostatin neurons) could be detected. The smaller amplitude could be explained because of the electrotonic decay (Marlin and Carter, 2014; Pouille et al., 2013; Stuart et al., 2008).
In conclusion, taking all together, GABA release in PV cells lacking CSPα/DNAJC5 is significantly impaired and the phenotype that we have found open a very interesting experimental scenario to investigate the role of CSPα/DNAJC5 in these synapses and to get insights into the molecular mechanisms of trans‐synaptic signaling under synaptic degeneration.
8. The use of UMIs and STRT‐seq‐2i/WaferGen technology as a novel and powerful method for sequencing
Single‐cell RNA sequencing emerged as a necessity for a high‐throughput technology able to massively decipher gene expression patterns of individual cell populations when only a small amount of material is available. The first transcriptomes produced by scRNA‐seq date from 2009 (Kolodziejczyk et al., 2015; Tang et al., 2009).
Several techniques can be used for isolation of cells before sequencing. In our case, we used FACS to specifically separate PV neurons by their tdTomato‐expression from the whole cortex. We directly FAC‐ sorted into the WaferGen plate to reduce collection time for better cell preservation. Furthermore, the FACS facility at KI had already efficiently set up the FAC‐sorting strategy into the 9600‐well plate (Hochgerner et al., 2017). In our experiments, we processed 3 animals per day to shorten the whole processing time: 6 hours per experiment (from mouse anesthesia to FACS into plate). Around 1.5 hours were spent at FACS, because of the low sorting efficiency (0.1‐0.5% measured from 50000 total events) (Figure 32) of a poorly enriched cell population, since the PV population account approximately for an 8% of the total cortex (Condé et al., 1994; Gonchar et al., 2008; Hu et al., 2014; Tamamaki et al., 2003; Uematsu et al., 2008; Whissell et al., 2015). Although we found variations in cDNA quantity of randomly selected quadrants from the plates ‐ some of them even contained very poor cDNA ‐, the majority of tested
183 Discussion samples were good enough to move to the next step: sequencing. Indeed, the final four cDNA libraries, each one prepared from an independent chip, yielded good scores (around 1‐5ng/μl) (BOX 2).
The STRT protocol employs UMIs as molecular identifiers, supporting an advance method to directly measure gene expression. The use of UMIs allows to correct for amplification bias due to PCR‐ derived artifacts and it offers an entire scale of measurement with a defined zero level. The UMI‐method is based on the addition of a random sequence (UMI) to each cDNA molecule during reverse transcription and then in the enumeration of the total number of distinct UMIs aligned to each position (Islam et al., 2014). In contrast, the relative measures obtained using RPKM‐related methods could mask changes in total mRNA content, providing results that are not real (i.e. a gene up‐regulation in terms of RPKM does not necessarily mean an increase in the absolute expression level, because total mRNA content may also has changed) (Islam et al., 2014; Picelli et al., 2013). As the initial STRT method, the novel STRT‐seq‐2i in WaferGen technology uses UMIs for quantification (Hochgerner et al., 2017).
9. Prior to data analysis: read depth, raw data classification and cell filtering by QC metrics
Regarding the first results from sequencing, we found a read depth per cell (around 32.000‐38.000 mapped mRNA reads/cell) (Table 22) comparable to that found in the original publication of the STRT‐seq‐ 2i and WaferGen technology (Hochgerner et al., 2017), in which an average read depth of 41000 mapped mRNA reads per cell was detected.
Creating the ReadData.R script to classify wells by genotype and other variables (such as animal number), and to discard potential contaminated wells (named as “Undetermined”) was laborious. This was mandatory due to software memory limitations of the sorter and physical limitations of the reagent nanodispenser, which made necessary to develop a specific sorting template (Figure 13; Supp. data: ReadData > WG Wells IDs.csv). So, we created a script to order, mix and discard specific wells. This was required because each WaferGen plate included cells from 3 distinct mice with different genotypes, trying to neutralize differences due to batch effect.
There are some scRNA‐seq analysis packages with functions to remove potential noise caused by batch effect. For example, in the ScaleData function from Seurat package (https://satijalab.org/seurat/pbmc3k_tutorial.html) we can introduce specific parameters to regress out uninteresting sources of heterogeneity, such as batch effect. Nevertheless, during the initial phase of analysis of our data, we did not find relevant differences in the results when regressing out the parameter “Plate” (WG17002, WG17003, WG17004 and WG17005). This suggests that our data were relatively robust in terms of batch effect noise, perhaps because of the processing of 3 mice per day.
184 Discussion
After discarding “Undetermined” wells and some not useful genes (ERCCs, spikes and fluorescent markers), the ExpressionFinal.csv file contained 9529 cells and 24378 genes, but this will not be so far our final number of cells (Figure 33A). We skipped to use ERCC genes as a QC metric or to calculate a fitting curve for technical noise. In Seurat, QC is based on other parameters and also employs another method to measure variable genes (FindVariableGenes function identifies outlier genes) (https://satijalab.org/seurat/pbmc3k_tutorial.html). Nonetheless, ERCCs have been previously used (Brennecke et al., 2013; Kharchenko et al., 2014; Lun et al., 2016; Tasic et al., 2016). We evaluated the quality of cells by three distinct parameters: (1) the number of total molecules detected in each cell (nUMI), (2) the number of genes expressed in each cell (nGene), and (3) the percentage of mitochondrial genes detected in each cell (percent.mito). These are three typical and broadly used metrics to evaluate cell quality (Butler et al., 2018; Lun et al., 2016; McCarthy et al., 2017; Muñoz‐Manchado et al., 2018; Poulin et al., 2016; Tasic et al., 2018). Our QC metrics distributions looked rather poor considering all the initial 9529 cells. Although mean values for each parameter showed quite good results (Table 23), many of our cells presented values close to zero for nUMI and nGene, and apparently high values for mitochondrial genes (Figure 33B). Since those values were not comparable to previously published values using the STRT method, we determined proper thresholds to select for “healthy” cells. The STRT‐seq‐2i method using WaferGen platform was pioneered by Hochgerner et al., 2017. They found great variability in average nUMI and nGene values depending on cell type. For example, with an average read depth of 41000 mapped mRNA reads per cell, they obtained a mean value around 4300 detected molecules (nUMI) and 2250 detected genes (nGene) in the interneuron cell class from mouse somatosensory cortex. Those numbers were higher than ours that might be due, however, to their slightly higher read depth values. These values were not comparable to those obtained in Zeisel et al., 2015 using Fluidigm C1: with an average read depth of 500000 mapped reads per cell, they detected around 5000 genes and 17000 molecules on average in the same cell type (interneurons). As expected, one sequencing lane in WaferGen chip with 2392 valid wells will produce fewer reads per well compared to Fluidigm C1 platform (Zeisel et al., 2015), which contains only 96 wells per chip. Thus, based on those publications and according to our read depth (around 36000 mapped mRNA reads per cell) we established that a valid cell must contain: >1000 nUMI, >350 nGene and <20% mitochondrial genes (Figure 33C). Upon applying those thresholds, our data improved and became comparable to Hochgerner et al., 2017 data. The average values remained similar between genotypes, providing additional proof of lack of batch effect‐dependent heterogeneity (Table 23). At this stage, upon cell filtering by QC metrics, the total number of cells fell down to 4016 cells from the initial 9529 cells (Figure 35A). This mandatory loss of cells is required to secure quality control in scRNA‐seq analysis (Butler et al., 2018; Harris et al., 2018; Hochgerner et al., 2018; Lun et al., 2016; McCarthy et al., 2017; Muñoz‐Manchado et al., 2018; Poulin et al., 2016; Tasic et al., 2016, 2018; Zeisel et al., 2015).
185
Discussion
10. The basis to support an integrated analysis and reasons for successive clustering steps
Seurat proposes two different methods to analyze scRNA‐seq data that depend on the number of datasets to be compared. In our case, having three distinct cell datasets (corresponding to PVcre:Ai27D:Dnajc5WT, PVcre:Ai27D:Dnajc5flox/+, PVcre:Ai27D:Dnajc5flox/‐ mice), the integrated analysis would produce a more powerful and robust clustering than the standard protocol analyzing each dataset separately. The integrated analysis is useful to uncover shared populations present across datasets that, otherwise, would be difficult to identify using the standard workflow. The combination of datasets increases the total number of cells, a key factor to improve cell clustering (https://satijalab.org/seurat/immune_alignment.html; Butler et al., 2018). Nevertheless, if any non‐ overlapping population had existed in only one of the genotypes, it would have been identified by a workflow implemented in Seurat that we, indeed, run after the fourth clustering step (data not shown). Upon establishing the clustering of cells, the datasets were separated by genotype to investigate differential expression analysis.
In order to reduce noise, some previous studies recommend, prior to analysis, to remove useless genes: predicted genes (e.g. gene names that start with “Gm”), pseudogenes, genes from mitochondria, ribosomal genes and sex‐specific genes (Tasic et al., 2018). Although we did not perform that step before analysis, we have discarded those genes when detected as cluster makers (as in the case of the final population 0, where we chose Atp5k gene as cluster marker instead of BC002163, which is a pseudogene) and when those genes were found as up‐ or down‐regulated genes. By performing this previous gene elimination, we would have saved time and effort by probably avoiding some of the re‐clusterings, since several of them were performed because of finding the expression of ribosomal RNA genes, predicted genes and pseudogenes in some clusters. In any case we now think that our strategy was correct: those genes turned out to be useful reporters of low‐quality cells to be eliminated (Figures 36, 37, 40, 41, 44, 45). In the first three clustering rounds, we found and eliminated clusters containing higly expression of: oligodendrocyte‐related genes (Plp1, Mobp, etc.) or ribosomal RNA and pseudogenes (Rn4.5s, prefix “Gm”). Those clusters, though, contained a high number of cells (Figures 35A, 39A, 43A, 47A) in the three different genotypes. Because of that, the total number of cells at the end of the final (fourth) clustering became again reduced from 3405 to 2847 cells (Figure 47A). In recent studies, it has been reported that neuronal cell‐type expression obtained using single‐nuclei RNA is in agreement with conventional whole single‐cell sequencing, in which heterogeneity in gene expression is due to other RNA sources, such as mitochondrial transcription, splicing, nuclear export rates or post‐transcriptional mechanisms (Bakken et al., 2018; Lake et al., 2017). Indeed, most protein‐coding transcripts are expressed in both the nucleus and cytoplasm, however, many genes have transcripts restricted to the cytoplasm, including mitochondrial
186
Discussion genes (such as mitochondrial rRNA) and, surprisingly, pseudogenes (Bakken et al., 2018; Lake et al., 2017). Moreover, Bakken et al., 2018 also demonstrated that non‐coding genes (ncRNA) and pseudogenes were better markers of cell types than protein‐coding genes, and even long ncRNAs have specific expression among neuronal types in the mouse cortex. This is in agreement with our results involving the emergence of several ncRNAs, pseudogenes and mitochondrial rRNA genes as markers for some PV populations or as differentially expressed genes (S15).
According to the expertise of our collaborators (Dr. Hjerling‐Leffler’s group), the expression of ribosomal RNAs, found as the first expressed genes in some clusters, might indicate cell lysis, damage or apoptosis. This hypothesis might be supported by the notion that all these processes may implicate the loss of cytoplasmic RNA content from the cell. Ribosomal RNA (rRNA) accounts for more than the 80% of the total RNA in a mammalian cell (Kampers et al., 1996; Zhao et al., 2018). Residual rRNA is likely released from the nucleolus in non‐healthy lysed cells, becoming preferentially captured, sequenced and over‐ represented in comparison to a healthy cell. This is a flag to consider further improvements of our neuron isolation protocol, such as a reduction in the time spent for neuron isolation and FACS by using a lower number of mice per experiment or by securing a gentler mechanical tissue dissociation. On the other side, the genomic facility also provided useful information about the presence of these rRNA genes in our data. As they previously carried out a selection of polyA‐containing sequenced transcripts to guarantee only the presence of mRNA, it was quite unusual to find ribosomal RNA genes expressed in our data, although these genes are contained in their large list of genes to carry reads alignment. They suggested that these rRNA genes might be found as a consequence of an internal priming to polyA regions during retrotranscription and library construction that can be either stochastic or really reflecting the number of such molecules. In any case, we decided to discard the identities containing those genes to really guarantee the presence of healthy cells and reliable read data for our downstream analysis. Nonetheless, in those clusters in some cases, other informative genes were detected together with rRNAs genes (Supp. Data: Conserved markers (CSV)) and they were removed only if rRNA and pseudogenes were the majority of genes in the list. For example, cluster 4 from the second clustering contains around 16 rRNA genes and pseudogenes out of 180 total conserved markers, which comprise other genes including Scna and Pthlh that are well‐known markers for chandelier cells. Thus, to correctly separate these cells into distinct populations during the next clustering, we retained those clusters in which we found not only damage‐related genes but also a high number of other useful genes. Following that strategy, we finally reached a reasonable, “healthy”, final cell clustering (Figure 47).
On the other hand, genes related to the glial lineage, especially related to oligodendrocytes, defined a cluster after the first and the second clustering steps. We proposed two different explanations to this result: (1) tdTomato expression is not restricted to PV neurons and therefore it is labelling other
187 Discussion cell types such as oligodendrocytes, or (2) oligodendrocytes might have been pulled together with PV interneurons during FACS. Plausibility of the first hypothesis is low because we have reported a 90% correlation between PV+ and tdTomato‐expressing neurons (Figure 20A). The lower percentage of tdTomato cells that does not colocalize to PV cells (1% in mutants and 4% in controls) could still comprise glial‐type cells, however, we found the second option more likely to explain the results. It has been shown that PV axons are highly ensheathed by myelin at cortex and hippocampus, most of them associated to the PV basket cell class and myelinated closely to the axon initial segment (Micheva et al., 2016; Stedehouder et al., 2017). Having into account that oligodendrocytes are the myelin‐producing cells at the CNS, it is probable that they were sorted attached to PV cells. Indeed, in pilot experiments previous to sequencing, we had also detected mRNA for Plp1 and Mog by RT‐PCR and qRT‐PCR together with Pvalb expression in tdT+ and tdT‐ sorted fractions from PVcre:Ai27D:Dnajc5flox/+ and PVcre:Ai27D:Dnajc5flox/‐ mice (data not shown). These observations support our strategy of cell clustering as an advantageous method to remove damaged/impure cells that would have possibly interfered with the downstream analysis.
11. The importance of an appropriate CC selection for clustering
The analysis of principal components (PCA) is a tool widely used to visualize relationships between cells in a low‐dimensional space. This allows cells with similar expression profiles to be located together in a plot while cells that differ in their expression pattern get located far apart from each other (https://towardsdatascience.com/dimensionality‐reduction‐does‐pca‐really‐improve‐classification‐ outcome‐6e9ba21f0a32; Lun, McCarthy and Marioni, 2016 https://satijalab.org/seurat/pbmc3k_tutorial.html). Since we have cells belonging to three different genotypes (WT, Flox+ and Flox‐), in our analysis, we ran a function to visualize cells that has been developed to be used for more than two different datasets. This tool is called canonical correlation analysis (CCA) and performs dimensionality reduction by finding common sources of variation between all datasets, and stores the canonical correlation vectors (CCs) – the vectors that project each dataset into the maximally correlated subspaces ‐ into a single object (https://satijalab.org/seurat/immune_alignment.html; Butler et al., 2018). It has been reported that running dimensionality reduction on high variable genes (HVG) obtained from these datasets (as we did) can improve clustering performance (https://satijalab.org/seurat/pbmc3k_tutorial.html).
Normally, the first components/vectors explain most of the variability of the data, because they are sustained by strong differences between specific populations. However, if we selected only few first components, we could be losing resolution to detect additional subtle differences between other subpopulations (Lun et al., 2016). When we plotted CC1 vs CC2 – and even when only plotting CC1 ‐ after the four clustering steps (Figures 34A, B; 38A, B; 42A, B; 46A,, B) we observed a clear separation of cells in two different groups. This suggests that the first correlation vector (CC1) is already providing a lot of
188 Discussion information by differentiating two groups of cells with high variances between them and a dissimilar expression pattern. Each CC provide specific information to differentiate cells. Therefore, it is crucial to make a proper CCs selection for downstream analysis since additional subtle differences may not be detected only by the first CCs. Seurat performs clustering using a list of scores given by the selected CCs (https://satijalab.org/seurat/immune_alignment.html; Butler et al., 2018), so the inclusion of more CCs, beyond the first ones, provided useful information to separate cells into 8 PV subpopulations (Figure 47).
CCs selection is mainly carried out by looking for a saturation in the relationship between the number of CCs and the percentage of variance explained by each CC, and the gene expression patterns defyning each subspace, which was assessed in this work by metagene‐bicor plots and heatmaps, respectively (Figures 34C, 38C, 42C, 46C; S7‐10). It was difficult to find some dissimilarities in the curve pattern that followed our bicor‐plots compared to those from the Seurat tutorial for an integrated analysis (https://satijalab.org/seurat/immune_alignment.html). For instance, every genotype curve followed the typical falling‐like shape, however, some of them did no end in saturation but began again to grow (and fell down again in some cases) (Figures 34C, 38C, 42C, 46C). This happened in distinct genotypes in each clustering step, but, although curves rised up, heatmaps proved that the curves, indeed, did not give extra information concerning gene expression patterns. Heatmaps provided the evidence that from a given CC in each clustering step (CC17, CC22, CC23, CC21, respectively), the remaining genes explaining the rest of CCs persevered as the same, thus, no degenerating additional information (S7‐10). Overall and facing such unexpected difficulties, we believe that our four different CC selections have been reasonably justified and carried out (based on both bicor‐plots and heatmaps) in a way that ensure a powerful clustering.
12. PV cells cluster into 8 different identities
For the selection of the final cluster markers, we established the following limitations: (1) to be a flag marker, the gene has to occupy one of the first positions (according to fold change) within the generated, ranked list of conserved markers between genotypes (Supp. Data: Conserved Markers (CSV)); (2) the gene needs to satisfy a p‐value < 0.05 as a conserved marker for each of the genotypes (WT, Flox+ and Flox‐) (Table 24); (3) the gene must not involve a predicted gene nor pseudogene (S11); (4) in the case that the gene selected by the first three parameters was also highly expressed in other clusters, then a second extra gene was selected as a flag marker that really differentiates that population (these are the cases of Pvalb.Rbp4.Th and Pvalb.Scna.Pthlh clusters) (Figures 48, 58). Notably, we found several clusters in which the gene satisfying all these characteristics turned out to be an ubiquitously‐expressed gene (i.e. Gapdh, Usmg5, Atp5k) which, accordingly, was also present in other clusters (Figures 48, 58). However, we rely on the function FindConservedMarkers that we think has been correctly applied to our data and detected genes that fulfill our requirements, therefore, likely represent relevant results. It is worth to
189 Discussion notice that in this type of studies pursuing a classification of cell types, unexpected results should not be underestimated.
Before the selection of cluster markers, to set up a proper resolution value for clustering, we consulted the Allen Brain Atlas that contains a repertoire of mouse brain cell types recently reported (http://celltypes.brain‐map.org/rnaseq/mouse; Tasic et al., 2018). The first 10‐ranked genes in each of our PV populations were studied for their similarity to the 10 reported mouse PV cell subclasses at the Allen Brain Atlas (Figures 49‐56). We found that our Pvalb.Scna.Pthlh cluster fitted very well with the chandelier group of PV cells at The Allen Brain Atlas (Figure 56) (Harris et al., 2018; Tasic et al., 2016, 2018). In addition, although not that strong as with chandelier cells, we detected some similarities with other PV populations. Our Pvalb.Rbp4.Th identity fitted with the Pvalb.Gabrg1 subclass (Figure 53), and our Pvalb.Sst cluster overlapped with Pvalb.Th.Sst and Pvalb.Calb1.Sst identities (Figure 54). Certainly, our PV populations do not entirely match with those from Tasic et al., 2018. An explanation could be that our analysis focused on the whole cortex, while Tasic et al., 2018 focused on the primary visual cortex and the anterior lateral motor cortex. It has been reported that a resolution value between 0.6‐1.2 provides good results for clustering datasets containing around 3000 cells (https://satijalab.org/seurat/pbmc3k_tutorial.html; Butler et al., 2018). We think that our comparative analysis with the Allen Brain Atlas data, supports all the resolution values we selected for each clustering and, especially, with the value chosen for our fourth clustering (0.8), which revealed the final 8 PV identities and final cluster markers (Figure 59). However, future investigations could be performed on this “granularity” parameter in order to potentially find out more PV subpopulations or even search for sub‐populations within a PV cluster (Butler et al., 2018). Although not focusing on a specific cortical region, we believe that our detected PV populations might provide new insights into the current PV classification (Harris et al., 2018; Muñoz‐Manchado et al., 2018; Poulin et al., 2016; Tasic et al., 2016, 2018; Zeisel et al., 2015).
It is worthy to pay attention to the differences in the number of cells from each genotype included within the different PV clusters (Table 25). Remarkably, the proportion of cells within the Pvalb.Kcna1 cluster was different among genotypes: 34% in PVcre:Ai27D:Dnajc5WT cells, 23% in PVcre:Ai27D:Dnajc5flox/+ cells and 11% in PVcre:Ai27D:Dnajc5flox/‐ cells. These differences might be of interest to investigate the causes of the neurodegenerative phenotype in mutant mice, because this cluster is characterized by a high expression of the potassium channel Kv1.1 (Kcna1). Kv1‐type channels are expressed at the axon initial segment of PV neurons and importantly regulate their fast‐spiking properties ‐ besides Kv3‐type channels ‐ by providing a crucial counterbalance to the rapid‐response characteristic of fast‐spiking cells (Goldberg et al., 2008; Hu et al., 2014). However, our results might be taken with caution, because no significant differences have been appreciated (Figure 62; S12). Some technical aspects could influence the number of cells in every cluster: the neuron dissociation protocol, the capture of cells by FACS, the use of aCSF instead
190
Discussion of NMDG‐Hepes solution and the albumin density gradient in the first chip, and/or the recurrent discarding of cells along the sequential clustering steps. It is important to remark that our first experiment, in which neuron isolation was carried out using the conventional aCSF and the albumin density gradient, led to an extremelly poor yield in terms of collected cells and, accordingly, the 3 mice used in that experiment contributed only with a very low number of cells (34 cells in total) (S12). Under these circumstances, we can only use these results as a path for further exploration using other techniques, such as determination of transcripts or proteins in brain tissue in situ, more suitable to detect potential reduction in the number of cells within specific subpopulations.
The additional control step involving specific cell‐type makers included another common strategy to eliminate potential doublets (Hochgerner et al., 2017; Muñoz‐Manchado et al., 2018; Tasic et al., 2018). In our case, we decided to check for the expression of non‐GABAergic markers after we completely established our final clustering. We found few cells expressing non‐neuronal genes (Figure 63) that could indicate the existence of doublets (one well containing a PV neuron plus another contaminant cell). Removing the cells expressing those genes from the analysis before continuing would have been a reasonable option to completely guarantee the absence of doublets. However, we believe that leaving those sparse potential “contaminated” cells did not interfere with the differential expression analysis, as they only involved a small number of cells with genes being expressed at low level (Gfap, Mog, Aqp4, Fn1, C1qc, Slc17a7) (S13). Indeed, in agreement with our results, a few random PV cells expressed those genes, at the Allen Brain Atlas collection that were obtained after applying a procedure to exclude doublets (Tasic et al., 2018). On the other hand, GABAergic‐ and PV‐specific markers were all found in our PV cells (Figure 63, 64; S13). However, Syt2 and Pvalb genes were not significantly changed between genotypes or clusters as supported by the fact that they were not found as DE genes after applying the Seurat specific package (Supp. Data: DE genes (CSV)).
13. Dnajc5 expression in the PV population of different genotypes
We have paid special attention to the expression levels of Dnajc5 as summarized in the following points:
1) According to the results obtained in the WT genotype, the general expression of Dnajc5 (1.28) was apparently low (Table 26; S14) in contrast to other genes encoding for synaptic vesicle proteins such as Sv2a (8.56) or Vamp2 (4.62). Takamori et al., 2006 found that the number of CSP/DNAJC5 molecules per vesicle was among the lowest compared to the rest of the SV proteins, for example, with SV2A and VAMP2. However, explanations of direct relationships between expression and translation of these proteins are, however, not trivial. Looking at particular identities, Dnajc5 seems to be mostly expressed in the Pvalb.Kcna1 identity, but less expressed in the Pvalb.Usmg5 cluster in
191
Discussion
all the genotypes (Figure 65B, C). Notably, the number of cells of Pvalb.Kcna1 in PVcre:Ai27D:Dnajc5flox/‐ mice was apparently smaller compared to control mice (Figure 62; Table 25), however, at the moment, if this reduction in cell number is a consequence of Dnajc5 knock‐down is uncertain.
2) Dnajc5 expression was detected in PVcre:Ai27D:Dnajc5flox/‐ mice. Indeed, the levels of Dnajc5 expression in the three genotypes was similar (1.28 for PVcre:Ai27D:Dnajc5WT, 1.58 for PVcre:Ai27D:Dnajc5flox/+ and 0.85 for PVcre:Ai27D:Dnajc5flox/‐ genotypes). This apparently conflicting result was interpreted as a false positive result generated by the detection of a partial mRNA. The STRT‐seq‐2i method does not provide full‐length read coverage (Hochgerner et al., 2017). This method generates 45 bp‐long reads (UMIs) starting at the 5' end of the transcript that serves to report the expression of that gene. Accordingly, the Dnajc5 mRNA transcribed from a recombined version of the Dnajc5 floxed allele (a transcript lacking only the floxed‐exon 3) would be detected and counted as Dnajc5 expression, even being an unproductive transcript that will not yield any protein (Figures 18, 19). In any case, ultimately, we can clearly detect the reduction in protein levels of CSP/DNAJC5 in the brain of PVcre:Ai27D:Dnajc5flox/‐ mice (Figure 24). On the other hand, a deeper analysis of the sequence of every amplified transcript from Dnajc5 is possible by further analysis of raw data carrying out the alignment of UMI‐derived Dnajc5 counts against the mouse Dnajc5 cDNA sequence. This complementary analysis will be carried out in the near future.
14. Differential expression analysis related to the absence of CSPα/DNAJC5
Once we have the clusters separated and identified for every genotype, we investigated which genes are specifically expressed in each group. Seurat method testes for differential expression analysis the same genes (HVG) used for clustering. This method, although useful, has been discussed by other authors (Aaron T. Lun et al.) that emphasize that p‐values should be carefully interpreted. For our study, we only selected genes that presented a p‐value < 0.05. A threshold determination for fold‐change was not straight forward to determine, because: (1) we did not establish a minimum fold‐difference between groups of cells when running the FindMarkers function, then obtaining a large gene list with all found changes that show a high heterogeneity in fold‐change values between genes; (2) the vast majority of genes show a fold‐change value particularly low (only few of them reaching close to 4 “times”) (Tables 27, 29, 31, 33). Therefore, we made the decision to take some top‐ranked up‐ and down‐regulated significant genes as DE genes (Figures 66‐69) in order to investigate their potential relationship with the neurodegenerative phenotype reported in our PVcre:Ai27D:Dnajc5flox/‐ mouse model. The first general conclusion we made from this selection was that fold‐changes were slightly higher in the up‐regulated group than in the down‐regulated genes. A thorough bibliographic support is provided about all the DE genes obtained after all differential expression analyses (S15).
192 Discussion
14.1. Gene ontology (GO) analysis reveals alterations in synaptic and metabolic pathways Interestingly, the GO analysis revealed several changes in pathways related to different up‐ and down‐regulated genes. The changes were found in the analysis of all the PV cells and in the analysis of PV clusters (Tables 28, 30, 32, 34). The GO analysis did not reveal changes in any pathway when comparing global gene expression changes between PVcre:Ai27D:Dnajc5WT and PVcre:Ai27D:Dnajc5flox/+ mice (Table 28). Nevertheless, we can still find differences between those two controls by examining GO changes at every PV cluster. This observation is interesting and deserves further consideration in a future follow‐up study of this thesis.
14.1.1. Dysregulation of synaptic processes in the absence of CSP/DNAJC5 The most reiterative GO terms along comparisons included the regulation of synaptic processes (synapse, action potential, neurotransmitter secretion, transmission, synaptic vesicle cycle, and GABAergic synapse, within others), preferentially indicating genetic down‐regulation. In addition, the glycolysis/gluconeogenesis pathway turned out to be up‐regulated in Flox‐ versus Flox+ cells. Other terms involving lysosomal, intracellular protein transport, ATPase activity, ribosomal, actin‐related, protein ubiquitination/proteasome/folding, and mitochondrial functions, among others, were also highlighted by their repetitive occurrence during comparative analyses, but, in those cases, they were found either from up‐ or down‐regulated genes (Tables 27, 29, 31, 33; Supp. Data: ClueGO analysis (XLS)). These results suggest that CSP/DNAJC5 may have roles, not only related to the synaptic vesicles, but also to other types of organelles, such as lysosomes and mitochondria.
14.1.2. Up‐regulation of "Glycolysis/Gluconeogenesis" in the absence of CSP/DNAJC5 Interestingly, the comparison between all PV cells from PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5flox/+ genotypes revealed up‐regulation of genes related to the term "Glycolysis/Gluconeogenesis" (Table 28). This term was also found in several clusters when comparing these two genotypes (Pvalb.Gapdh, Pvalb.Rbp4.Th, Pvalb.Sst) (Table 30). However, the comparison between PVcre:Ai27D:Dnajc5flox/‐ versus PVcre:Ai27D:Dnajc5WT genotypes did not detect this pathway in any of the analyses (Tables 28, 32). The up‐regulation of genes related to gluconeogenesis is intriguing and deserves to be followed up. Although gluconeogenesis exists in brain astrocytes, it has not yet been convincingly demonstrated in neurons, so our results could suggest a pathological metabolic transformation (Yip et al., 2017). Glycolysis is a key process in neuronal and synaptic energetics (Ashrafi and Ryan, 2017; Dienel, 2018). Our finding related to the upregulation of the glycolysis pathway is particularly relevant in our context of activity‐dependent presynaptic degeneration. Several enzymes that are part ofe th glycolytic machinery are specifically enriched on the synaptic vesicles (Ikemoto et al., 2003;
193 Discussion
Takamori et al., 2006). This biochemical organization is perhaps advantageous for the local production of ATP required by ATP consuming enzymes, such as the V‐type ATPase proton pump involved in the loading of NTs into synaptic vesicles and other enzymes involved in the synaptic vesicle cycle (Ashrafi and Ryan, 2017). Ryan's group has demonstrated that presynaptic glycolysis is required for synaptic vesicle endocytosis, indicating that the membrane fission depends on the rapid ATP generation (Rangaraju et al., 2014). Interestingly, synaptic vesicle recycling is compromised in motorneurons lacking CSPα/DNAJC5 (Rozas et al., 2012). On the other hand, in a genetic screen pursuing the identification of the protein machinery involved in the maintaining of presynaptic terminals in C. elegans, a key role of glycolysis has been revealed. It has been shown that, during periods of energy stress, a local metabolon composed of glycolytic enzymes is formed at presynaptic terminals and it is required for synaptic vesicle recycling (Jang et al., 2016). The glycolytic enzymes found in the metabolon include pyruvate kinase, glyceraldehyde‐3‐ phosphate dehydrogenase, phosphofructokinase and aldolase (Jang et al., 2016). On the other hand, the up‐regulation of genes involved in glycolysis has been recently related to synaptic activity as a mechanism to supply lipids for neurite growth (Segarra‐Mondejar et al., 2018). This study proposes that synaptic activity induces Ca2+ transients that activate the transcription factor CREB. As a consequence, the expression of the glucose‐transporter Glut3 is up‐regulated and the glucose uptake by neurons increases. CREB also induces the expression of the ubiquitin‐ligase Siah2 that, together with lactate‐dehydrogenase (LDH) activity, promotes the stabilization and activation of HIF‐1which activates the expression of genes encoding glycolytic enzymes. Enhanced glycolysis increases the levels of acetyl‐CoA that is used for lipid synthesis required for membrane extension during neurite growth (Segarra‐Mondejar et al., 2018). Altogether, those observations set‐up a hypothetical scenario relating CSP/DNAJC5 with glycolysis and neurodegeneration. Hypothetically, CSP/DNAJC5 would be required as a co‐chaperone to stabilize the glycolytic metabolon. In the absence of CSP/DNAJC5, synaptic glycolysis would be deficient due to metabolon instability. A failure in glycolysis could compromise the energetic requirements of the synaptic vesicle cycle and the production of lipids for neurite growth and stability. This phenotype would be more evident in neurons with a demanding ATP consumption as in fast‐spiking neurons which, consequently, might suffer from early preferential degeneration. In addition, the neuron might try to compensate the deficit in glucose metabolism by the up‐regulation of genes encoding key glycolytic enzymes. This hypothesis will be investigated in future experiments.
14.2. Up‐regulation of the gene encoding the neuropeptide Y (Npy) Our scRNA‐seq study provided evidences of Npy gene as one of the highest up‐regulated genes in the absence of CSP/DNAJC5. Npy expression levels in the total pool of cells of Flox/‐ mice is highly up‐ regulated in comparison with Flox/+ (Table 27), but unchanged in comparison with WT cells. In contrast, Npy is down‐regulated in PV cells from Flox+ vs WT genotypes, which is intriguing having into account the
194 Discussion opposite tendency in the Flox/‐ genotype (Table 27). To get deeper into this issue, we analyzed Npy expression levels in specific identities and found the following: (1) Npy was up‐regulated in Flox/‐ vs Flox/+ cells in four PV identities (Pvalb.Atp5k, Pvalb.Rbp4.Th, Pvalb.Sst and Pvalb.Snca.Pthlh) (Table 29; Supp. Data: DE genes (CSV)), (2) Npy was up‐regulated in Flox/‐ vs WT cells in one PV identity (Pvalb.Atp5k) (Table 31; Supp. Data: DE genes (CSV)), and (3) Npy was down‐regulated in Flox/+ vs WT cells in two PV identities (Pvalb.Rbp4.Th and Pvalb.Snca.Pthlh) (Table 33; Supp. Data: DE genes (CSV)). Altogether, this suggests that, in the absence of CSP/DNAJC5, the gene Npy is particularly upregulated by the cells from the Pvalb.Atp5k identity. This finding is relevant because of many key functions that have been described for NPY. NPY is implicated in synaptic transmission, food intake, cognition and learning (Gøtzsche and Woldbye, 2016; Reichmann and Holzer, 2016) and it is predominantly expressed in and released by GABAergic interneurons upon sustained neuronal activity (Colmers and Bahh, 2003; Gøtzsche and Woldbye, 2016; Karagiannis et al., 2009). Moreover, NPY has been reported to be particularly expressed in a subset of parvalbumin‐expressing basket cells, correlating with a longer spike latency than other fast‐ spiking PV neurons (Karagiannis et al., 2009). NPY is also crucial for the stress adaptation process, having an anxiolytic effect on mice that seems to be primarily mediated by its union to the Y1 receptor (Reichmann and Holzer, 2016). One of the most important functions of NPY is that it may be involved in the response to epileptogenesis as its expression is increased in mouse models of epilepsy, functioning as an endogenous anticonvulsant by regulating neuronal excitability (Colmers and Bahh, 2003; Erickson et al., 1996; Li et al., 2016; Noè et al., 2008). Furthermore, the injection of NPY in mice suppresses seizure susceptibility, thus it is being studied as an anti‐epileptogenic, anti‐excitatory promising pharmacological treatment for epilepsy (Noè et al., 2008) and other neurodegenerative disorders (Reichmann and Holzer, 2016). For instance, NPY is found decreased in Alzheimer’s disease, but increased in Parkinson’s and Huntington’s diseases, suggesting a neuroprotective role also in those pathologies (Reichmann and Holzer, 2016). These functions propose NPY as a response system to elevated activity, that can be considered adaptive, since NPY and its receptors expression may be modulated with high neuronal activity: NPY and Y2 receptor would become up‐regulated while the proconvulsant Y1 receptor expression would be reduced (Colmers and Bahh, 2003). In this context, an adaptation by up‐regulating the NPY‐mediated response might be a plausible mechanism underlying the increase in Npy levels by dysfunctional PV cells in PVcre:Ai27D:Dnajc5flox/‐ mice. The up‐regulation of Npy expression may reflect an attempt to balance the loss of inhibition by PV cells and to compensate for a progressive neurodegenerative process, as suggested by decreased levels in Syt2+ and PV+ synaptic boutons and a reduction in mIPSC amplitude and frequency in PVcre:Ai27D:Dnajc5flox/‐ animals (Figures 25, 31). One possibility is that Npy up‐regulation may comprise a self‐neuroprotective response triggered by PV interneurons, perhaps as part of homeostatic mechanisms to compensate the excess of circuit hyperexcitability. In summary, these observations related to changes in Npy need to be validated with quantitative measurements of protein levels and immunohistochemical
195 Discussion analysis. One scenario would be the in situ identification of the neurons overexpressing Npy to further analyze the synaptic consequences in specific neural circuits using electrophysiology and imaging.
14.3. Expression of sex‐specific genes Regarding sex‐specific genes, we have found in multiple occasions the gene Xist, which is only expressed from the inactive X chromosome, as an up‐regulated gene when comparing Flox‐ cells to both control cells (Tables 27, 29, 31). That is because we have used more females of PVcre:Ai27D:Dnajc5flox/‐ animals than in the other two groups, making the analysis to detect this gene as an DE gene although being actually an artifact. Other sex‐linked genes have also emerged as DE genes (Eif2s3y). To avoid this sex‐ derived differences, it would be required to only run the scRNA‐ seq experiments in either males or females. However, the complicated logistic and planning of these experiments to be performed in Stockholm in a specific period and with the same age made it extremely difficult to restrict the animal selection to only mice from the same sex.
14.4. Up‐regulated genes with consistent expression changes Some genes appeared consistently up‐regulated along the different comparisons in (1) all PV cells from Flox‐ vs Flox+ and Flox‐ vs WT (Table 27), and in (2) some identities when comparing clustered Flox‐ PV cells to both control PV neurons (Tables 29, 31), supporting robustness to our results. Some of these genes are:
1) Cort. It encodes for cortistatin, a neuropeptide very similar to somatostatin but with different roles. It has been reported to be expressed in GABAergic neurons, mostly in somatostatin cells but also significantly in parvalbumin interneurons in the cortex and hippocampus (de Lecea et al., 1997). It has been related to decrease neuronal excitability by inhibiting acetylcholine excitatory actions, but also to cause an impairment in memory consolidation if overexpressed in rats and mice (Ibáñez‐Costa et al., 2017; Jiang et al., 2017; de Lecea et al., 1997; Méndez‐Díaz et al., 2005; Tallent et al., 2005). However, although increased in Flox‐ cells compared to Flox+ and WT cells, it was decreased in few populations of Flox+ cells compared to WT cells.
2) Gapdh. The protein GAPDH has been reported to be increased in culture when cerebellar and cortical neurons suffer age‐induced apoptosis, and also after brain ischemic damage, both suggesting a role for GAPDH in the initiation of cell death (Ishitani et al., 1996; Tanaka et al., 2002). The up‐regulation of this gene in Flox‐ PV cells may indicate the initiation of apoptosis, but other general functions have been also described for Gapdh, such as to be involved in enzymatic metabolic conversion processes and in the regulation of mRNA stability (NCB, Genecards databases).
196 Discussion
3) Lgals1. This gene encodes for the lectin galectin‐1. It has been proposed to participate in the elimination of neuronal processes, to suppress neuronal death in epilepsy when it is absent (Lgals1 KO mice) and to protect from microglia action and inflammation induced by neurodegeneration (Bischoff et al., 2012; Plachta et al., 2007; Starossom et al., 2013). Galectin‐1 may have a neuroprotective role in the brain. In a mutant PV cell where cell death might be initiated, similarly, galectin‐1 could be providing a protective mechanism in the cortex by also inducing apoptosis under a context of neurodegeneration.
14.5. Up‐regulated genes potentially related to CSP/DNAJC5 Taking into account other genes that are relevant in relation to pathways in which CSPα/DNAJC5 might be involved, we highlighted the up‐regulation of Cox6a2 in the Pvalb.Pcp4 cluster and M6pr in the Pvalb.Snca.Pthlh cluster, both in the comparison of Flox‐ to Flox+ clustered PV neurons (Table 29).
1) Cox6a2. It encodes for a subunit of cytochrome c oxidase, which acts in the mitochondria, and has been suggested as a novel candidate marker for PV fast‐spiking due to an 80% reported colocalization with PV interneurons (Mancarci et al., 2017; Rossier et al., 2015). The high activity regime of PV neurons explains the high abundance of mitochondria and cytochrome c in comparison with other interneurons (Gulyás et al., 2006). However, although increased in Pvalb.Pcp4 cluster from Flox‐ cells compared to Flox+ and WT cells, it was decreased in the same population of Flox+ cells compared to WT cells.
2) M6pr. This gene encodes for a receptor of mannose‐6‐phosphate that has a key role in lysosomal transport. It is involved in the best understood direct pathway of lysosomal hydrolases (Matrone et al., 2016; Saftig and Klumperman, 2009). The up‐regulation of M6pr in mutant PV cells may be relevant in the study of the lysosomal function of CSPα/DNAJC5 in the pathological context of the autosomal dominant neuronal ceroid lipofuscinosis (Diez‐Ardanuy et al., 2017; Greaves et al., 2012; Nosková et al., 2011; Sambri et al., 2017; Xu et al., 2018).
Regarding the comparison between clustered PV interneurons from Flox‐ vs WT mice (Table 31), it is worth mentioning the up‐regulation of Tiprl gene in the Pvalb.Rbp4.Th cluster:
3) Tiprl. It may be involved in the regulation of mTOR signaling, inhibition of apoptosis and biogenesis and recycling of PP2A phosphatase (Nakashima et al., 2013; Rosales et al., 2015; Scorsato et al., 2016; Song et al., 2012). This finding might be along the line of evidence relating CSPα/DNAJC5 with the mTOR pathway, recently reported showing the loss of postnatal quiescence of neural stem cells through mTOR activation upon genetic removal of CSPα/DNAJC5 (Nieto‐González et al., 2019). The
197 Discussion
precise molecular mechanism involving CSPα/DNAJC5‐mTOR signaling interaction still needs further investigations to be clarified.
14.6. Down‐regulated genes related to neuronal excitability and GABAergic transmission Among down‐regulated genes, the detected fold‐changes are pretty lower than those from up‐ regulated genes. In any case, we found differences that may help to throw light into important questions. The majority of the relevant information could be extracted from the comparisons of all PV cells from Flox‐ versus dFlox+ an WT mice. Notably, several down‐regulated genes encode voltage‐dependent ion channels and GABA receptors relevant for PV cell functional properties:
1) Potassium channels. For example, Kv1‐type channels (both Kcna1 and Kcna2, which encode for Kv1.1 and Kv1.2, respectively) have been previously mentioned to be expressed at the axon initial segment of PV interneurons and to be of key relevance in the regulation of the firing behavior by damping PV excitability near to AP threshold (Goldberg et al., 2008). The absence of Kcna1 and Kcna2 genes is linked to increased seizure susceptibility, probably caused by an impaired PV excitability (Brew et al., 2007; Jiang et al., 2016). Interestingly, the genetic expression of these channels is highly regulated during development (Brew et al., 2007). Therefore, the expression changes that we found in PVcre:Ai27D:Dnajc5flox/‐ mice, although apparently contradictory with our findings (Figure 29), might represent an homeostatic mechanism related to excitability alterations detected in mutant PV cells. Nevertheless, to further clarify this issue, the study of K+ currents in PV interneurons is mandatory.
2) Sodium channels. The absence, inactivation or impaired function of Scn1a (coding for Nav1.1 channel) have been associated to the emergence of epileptic seizures, since this channel is essential for the rapid speed of action potential propagation and the maximal action potential frequency in PV neurons (Dutton et al., 2014; Hu and Jonas, 2014; Noh et al., 2012). Nevertheless, AP depolarizing kinetics of PV interneurons were normal in PVcre:Ai27D:Dnajc5flox/‐ mice (Table 17; Figure 28B).
3) GABA‐A receptor. Interestingly, down‐regulation of a gene coding for a GABA‐A receptor, Gabrb2, has been also detected in Flox‐ cells, being also reported as a risk allele for schizophrenia and epilepsy (Enna and Möhler, 2007; Mueller et al., 2015; Srivastava et al., 2015; Yeung et al., 2018). This finding might suggest a downscaling of GABA‐A receptors also in PV cells, a potential regulatory mechanism that might comprise a consequence from a fail or reduction in the number PV terminals arriving to the own PV interneurons, similarly to what happened in glutamatergic ecells (Figur 31).
4) Genes associated to epilepsy and other neurodegenerative disorders. Sv2a, the gene coding for the synaptic vesicle protein SV2A (De Groot et al., 2010; Janz et al., 1999; Tokudome et al., 2016; Van Vliet
198 Discussion
et al., 2009) was also found down‐regulated in Flox‐ versus WT PV interneurons. Atp1a3 was also down‐regulated in both Flox‐ and Flox+ cells versus the WT genotype. This is an ATPase that maintains the electrochemical gradients of sodium and potassium ions across the plasma membrane (Aguiar et al., 2004; Paciorkowski et al., 2015; Smedemark‐Margulies et al., 2016). Further studies into the potential epileptic disorders that may take place in PVcre:Ai27D:Dnajc5flox/‐ mice will provide extra information about this question.
In conclusion, our study opens promising perspectives to investigate the molecular mechanisms of presynaptic dysfunction in PV cells from PVcre:Ai27D:Dnajc5flox/‐ mice based on the substantial number of genes that appeared up‐ and down‐regulated within this single‐cell RNA sequencing study. Nevertheless, a complete validation of our scRNA‐seq results requires the use of complementary approaches, such as the novel and reliable RNAscope® in situ hybridization to test our results further.
199
CONCLUSIONS
Conclusions
1. PVcre:Ai27D:Dnajc5flox/‐ mice is a novel mouse model in which the Dnajc5 gene is deleted specifically in PV interneurons by Cre‐recombinase.
2. PVcre:Ai27D:Dnajc5flox/‐ mice do not suffer from early lethality. They show neurological signs such as hyperactivity and progressive ataxia from 2 to 8 months of age.
3. Perisomal disorganization and reduction of PV+ and Syt2+ synaptic puncta at layer II/III of motor cortex in PVcre:Ai27D:Dnajc5flox/‐ suggests structural degeneration of presynaptic terminals. In contrast, the number and size of PV somata at 2 and 8 months of age was not decreased, indicating that the integrity of PV neurons is preserved.
4. Electrophysiological analysis of PV (tdTomato+) interneurons without CSPα/DNAJC5 reveals subtle but progressive alterations compared to controls such as: (1) lower input resistance,) (2 action potentials with faster recovery after hyperpolarization and (3) higher rheobase. Excitatory postsynaptic potentials (EPSPs) were normal. The molecular mechanism underlying these alterations are unknown, however, ion channel alterations might be involved.
5. A reduction in the frequency of spontaneous miniature inhibitory postsynaptic currents (mIPSC) recorded in cortical pyramidal neurons is consistent with a lower number of PV synapses in PVcre:Ai27D:Dnajc5flox/‐ due to structural degeneration of presynaptic terminals.
6. The amplitude of mIPSCs is reduced while the decay time is increased in PVcre:Ai27D:Dnajc5flox/‐ mice compared to controls. Future experiments are required to investigate the mechanism underlying this phenotype.
7. The combination of FACS, the STRT‐seq‐2i method and the WaferGen 9600‐well platform provided a novel and powerful strategy for single‐cell RNA sequencing suitable to analyze individual transcriptomes from thousands of cortical PV interneurons collected from the whole mouse cortex.
8. Upon applying an analytical strategy for integrating scRNA‐seq data sets to identify shared populations of PV cells across genotypes, we collected a total of 712 WT cells, 1027 PVcre:Ai27D:Dnajc5flox/+ cells and 1108 PVcre:Ai27D:Dnajc5flox/‐ cells.
9. The integrated analysis of 2847 PV cells from the three genotypes revealed 8 different PV identities.
10. Three of our PV identities were highly similar to PV subclasses recently described in specific cortical areas. These identities are Pvalb.Snca.Pthlh, Pvalb.Sst and Pvalb.Rbp4.Th.
203 Conclusions
11. The gene ontology (GO) study based on the differential gene expression analysis suggests genetic dysregulation of glycolysis and synaptic function, among other processes, in PVcre:Ai27D:Dnajc5flox/‐ cells.
12. PVcre:Ai27D:Dnajc5flox/‐ mice present down‐regulated expression of genes encoding for synaptic proteins previously associated with brain disorders.
13. PVcre:Ai27D:Dnajc5flox/‐ mice present up‐regulated expression of genes encoding for the neuropeptide NPY and proteins involved in cell death, the reduction of neuronal excitability, mitochondrial oxidative functions, lysosomal transport and mTOR pathway, among others.
14. Up‐regulation of Npy in PVcre:Ai27D:Dnajc5flox/‐ cells might be reflecting homeostatic mechanisms to ameliorate network hyperexcitability specially driven by the Pvalb.Atp5k subpopulation.
15. This work opens multiple new perspectives to further investigate molecular and circuit mechanisms underlying synaptic alterations of PV interneurons and the role of CSPα/DNAJC5 in brain disorders.
204 BIBLIOGRAPHY
Bibliography
Abdelmoez, M.N., Iida, K., Oguchi, Y., Nishikii, H., Yokokawa, R., Kotera, H., Uemura, S., Santiago, J.G., and Shintaku, H. (2018). SINC‐seq: Correlation of transient gene expressions between nucleus and cytoplasm reflects single‐cell physiology. Genome Biol. 19, 1–11.
Aguiar, P.D.C., Sweadner, K.J., Penniston, J.T., Zaremba, J., Liu, L., Caton, M., Linazasoro, G., Borg, M., Tijssen, M.A.J., Bressman, S.B., et al. (2004). Mutations in the Na+/K+ ‐ATPaseα3 Gene ATP1A3 Are Associated with Rapid‐Onset Dystonia Parkinsonism. Neuron 43, 169–175.
Ahrendt, E., Kyle, B., Braun, A.P., and Braun, J.E.A. (2014). Cysteine string protein limits expression of the large conductance, calcium‐activated K+ (BK) channel. PLoS One 9, 1–9.
Alvarez de Toledo, G., Fernández‐Chacón, R., and Fernández, J.M. (1993). Release of secretory products during transient vesicle fusion. Nature 363, 554–558.
Arai, H., Emson, P.C., Mountjoy, C.Q., Carassco, L.H., and Heizmann, C.W. (1987). Loss of parvalbumin‐ immunoreactive neurones from cortex in Alzheimer‐type dementia. Brain Res. 418, 164–169.
Arshavsky, Y.I., Gelfand, I.M., and Orlovsky, G.N. (1983). The cerebellum and control of rhythmical movements. Trends Neurosci. 6, 417–422.
Ashrafi, G., and Ryan, T.A. (2017). Glucose metabolism in nerve terminals. Curr. Opin. Neurobiol. 45, 156–161.
Bakken, T.E., Hodge, R.D., Miller, J.A., Yao, Z., Nguyen, T.N., Aevermann, B., Barkan, E., Bertagnolli, D., Casper, T., Dee, N., et al. (2018). Single‐nucleus and single‐cell transcriptomes compared in matched cortical cell types. PLoS One 13, 1–24.
Benarroch, E.E. (2013). Neocortical interneurons: functional diversity and clinical correlations. Neurology 81, 273–280.
Bindea, G., Mlecnik, B., Hackl, H., Charoentong, P., Tosolini, M., Kirilovsky, A., Fridman, W.H., Pagès, F., Trajanoski, Z., and Galon, J. (2009). ClueGO: A Cytoscape plug‐in to decipher functionally grouped gene ontology and pathway annotation networks. Bioinformatics 25, 1091–1093.
Bischoff, V., Deogracias, R., Poirier, F., and Barde, Y.‐A. (2012). Seizure‐Induced Neuronal Death Is Suppressed in the Absence of the Endogenous Lectin Galectin‐1. J. Neurosci. 32, 15590–15600.
Boal, F., Zhang, H., Tessier, C., Scotti, P., and Lang, J. (2004). The variable C‐terminus of cysteine string proteins modulates exocytosis and protein‐protein interactions. Biochemistry 43, 16212–16223.
Bouhours, B., Gjoni, E., Kochubey, O., and Schneggenburger, R. (2017). Synaptotagmin2 (Syt2) Drives Fast Release Redundantly with Syt1 at the Output Synapses of Parvalbumin‐Expressing Inhibitory Neurons. J. Neurosci. 37, 4604–4617.
207 Bibliography
Braun, J.E.A., and Scheller, R.H. (1995). Cysteine string protein, a DnaJ family member, is present on diverse secretory vesicles. Neuropharmacology 34, 1361–1369.
Braun, J.E.A., Wilbanks, S.M., and Scheller, R.H. (1996). The cysteine string secretory vesicle protein activates Hsc70 ATPase. J. Biol. Chem. 271, 25989–25993.
Brennecke, P., Anders, S., Kim, J.K., Kołodziejczyk, A.A., Zhang, X., Proserpio, V., Baying, B., Benes, V., Teichmann, S.A., Marioni, J.C., et al. (2013). Accounting for technical noise in single‐cell RNA‐seq experiments. Nat. Methods 10, 1093–1098.
Brew, H.M., Gittelman, J.X., Silverstein, R.S., Hanks, T.D., Demas, V.P., Robinson, L.C., Robbins, C.A., McKee‐Johnson, J., Chiu, S.Y., Messing, A., et al. (2007). Seizures and Reduced Life Span in Mice Lacking the Potassium Channel Subunit Kv1.2, but Hypoexcitability and Enlarged Kv1 Currents in Auditory Neurons. J. Neurophysiol. 98, 1501–1525.
Bucurenciu, I., Bischofberger, J., and Jonas, P. (2010). A small number of open Ca2+ channels trigger transmitter release at a central GABAergic synapse. Nat. Neurosci. 13, 19–21.
Burgoyne, R.D., and Morgan, A. (2015). Cysteine string protein (CSP) and its role in preventing neurodegeneration. Semin. Cell Dev. Biol. 40, 153–159.
Burrone, J., O’Byrne, M., and Murthy, V.N. (2002). Multiple forms of synaptic plasticity. Nature 420, 414– 418.
Butler, A., Hoffman, P., Smibert, P., Papalexi, E., and Satija, R. (2018). Integrating single‐cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420.
Cabungcal, J.‐H., Steullet, P., Morishita, H., Kraftsik, R., Cuenod, M., Hensch, T.K., and Do, K.Q. (2013). Perineuronal nets protect fast‐spiking interneurons against oxidative stress. Proc. Natl. Acad. Sci. 110, 9130–9135.
Celio, M.R. (1986). Parvalbumin in Most y‐Aminobutyric Acid‐ Containing Neurons of the Rat Cerebral Cortex. Science (80‐. ). 492, 984–986.
Celio, M.R., and Heizmann, C.W. (1982). Calcium‐binding protein parvalbumin is associated with fast contracting muscle fibres. Nature 297, 504–506.
Chamberlain, L.H., and Burgoyne, R.D. (1996). Identification of a novel cysteine string protein variant and expression of cysteine string proteins in non‐neuronal cells. J. Biol. Chem. 271, 7320–7323.
Chamberlain, L.H., and Burgoyne, R.D. (1997). Activation of the ATPase activity of heat‐shock proteins Hsc70/Hsp70 by cysteine‐string. protein Biochem. J. 322, 853–858.
Chamberlain, L.H., and Burgoyne, R.D. (1998). The cysteine‐string domain of the secretory vesicle
208 Bibliography cysteine‐string protein is required for membrane targeting. Biochem. J. 335, 205–209.
Chamberlain, L.H., and Burgoyne, R.D. (2000). Cysteine‐string protein: The chaperone at the synapse. J. Neurochem. 74, 1781–1789.
Chamberlain, L.H., Henry, J., and Burgoyne, R.D. (1996). Cysteine string proteins are associated with chromaffin granules. J. Biol. Chem. 271, 19514–19517.
Chandra, S., Gallardo, G., Fernández‐Chacón, R., Schlüter, O.M., and Südhof, T.C. (2005). α‐Synuclein cooperates with CSPα in preventing neurodegeneration. Cell 123, 383–396.
Chen, C., Arai, I., Satterfield, R., Young, S.M., and Jonas, P. (2017a). Synaptotagmin 2 Is the Fast Ca2+ Sensor at a Central Inhibitory Synapse. Cell Rep. 18, 723–736.
Chen, T.W., Li, N., Daie, K., and Svoboda, K. (2017b). A Map of Anticipatory Activity in Mouse Motor Cortex. Neuron 94, 866–879.e4.
Chiu, C.‐S., Jensen, K., Sokolova, I., Wang, D., Li, M., Deshpande, P., Davidson, N., Mody, I., Quick, M.W., Quake, S.R., et al. (2002). Number, density, and surface/cytoplasmic distribution of GABA transporters at presynaptic structures of knock‐in mice carrying GABA transporter subtype 1‐green fluorescent protein fusions. J Neurosci 22, 10251–10266.
Clements, J.D., and Bekkers, J.M. (1997). Detection of spontaneous synaptic events with an optimally scaled template. Biophys. J. 73, 220–229.
Colmers, W.F., and Bahh, B. El (2003). Neuropeptide Y and Epilepsy. Epilepsy Curr. 3, 53–58.
Condé, F., Lund, J.S., Jacobowitz, D.M., Baimbridge, K.G., and Lewis, D.A. (1994). Local circuit neurons immunoreactive for calretinin, calbindin D‐28k or parvalbumin in monkey prefronatal cortex: Distribution and morphology. J. Comp. Neurol. 341, 95–116.
Cooper‐Knock, J., Kirby, J., Ferraiuolo, L., Heath, P.R., Rattray, M., and Shaw, P.J. (2012). Gene expression profiling in human neurodegenerative disease. Nat. Rev. Neurol. 8, 518–530.
Coppola, T., and Gundersen, C. (1996). Widespread expression of human cysteine string proteins. FEBS Lett. 391, 269–272.
Cowan, W.M., Südhof, T.C., and Stevens, C.F. (2001). Synapses (The Johns Hopkins University Press). d’Anglemont de Tassigny, X., Pascual, A., and López‐Barneo, J. (2015). GDNF‐based therapies, GDNF‐ producing interneurons, and trophic support of the dopaminergic nigrostriatal pathway. Implications for Parkinson’s disease. Front. Neuroanat. 9, 1–15.
Defelipe, J., López‐Cruz, P.L., Benavides‐Piccione, R., Bielza, C., Larrañaga, P., Anderson, S., Burkhalter, A.,
209
Bibliography
Cauli, B., Fairén, A., Feldmeyer, D., et al. (2013). New insights into the classification and nomenclature of cortical GABAergic interneurons. Nat. Rev. Neurosci. 14, 202–216.
DeFelipe, J. (2011). The evolution of the brain, the human nature of cortical circuits, and intellectual creativity. Front. Neuroanat. 5, 1–17.
Dienel, G.A. (2018). Brain Glucose Metabolism: Integration of Energetics with Function. Physiol. Rev. 99, 949–1045.
Diez‐Ardanuy, C., Greaves, J., Munro, K.R., Tomkinson, N.C.O., and Chamberlain, L.H. (2017). A cluster of palmitoylated cysteines are essential for aggregation of cysteine‐string protein mutants that cause neuronal ceroid lipofuscinosis. Sci. Rep. 7, 1–10.
Dixon,, C., Sah P., Lynch, J.W., and Keramidas, A. (2014). GABAa receptor α and γ subunits shape synaptic currents via different mechanisms. J. Biol. Chem. 289, 5399–5411.
Donato, F., Rompani, S.B., and Caroni, P. (2013). Parvalbumin‐expressing basket‐cell network plasticity induced by experience regulates adult learning. Nature 504, 272–276.
Donnelier, J., Braun, S.T., Dolzhanskaya, N., Ahrendt, E., Braun, A.P., Velinov, M., and Braun, J.E.A. (2015). Increased expression of the large conductance, calcium‐activated K+ (BK) channel in adult‐onset neuronal ceroid lipofuscinosis. PLoS One 10, 1–17.
Du, J., Zhang, L., Weiser, M., Rudy, B., and McBain, C.J. (1996). Developmental expression and functional characterization of the potassium‐channel subunit Kv3. 1b in parvalbumin‐containing interneurons of the rat hippocampus. J. Neurosci. 16, 506–518.
Dutton, S.B., Makinson, C.D., Papale, L.A., Shankar, A., Balakrishnan, B., Nakazawa, K., and Escayg, A. (2014). Preferential inactivation of Scn1a in parvalbumin interneurons increases seizure susceptibility. Neurobiol. Dis. 8, 4017–4018.
Enge, M., Arda, H.E., Mignardi, M., Beausang, J., Bottino, R., Kim, S.K., and Quake, S.R. (2017). Single‐Cell Analysis of Human Pancreas Reveals Transcriptional Signatures of Aging and Somatic Mutation Patterns. Cell 171, 321–330.e14.
Enna, S.J., and Möhler, H. (2007). The GABA Receptors (Humana Press).
Enwright, J.F., Sanapala, S., Foglio, A., Berry, R., Fish, K.N., and Lewis, D.A. (2016). Reduced labeling of parvalbumin neurons and perineuronal nets in the dorsolateral prefrontal cortex of subjects with schizophrenia. Neuropsychopharmacology 41, 2206–2214.
Erickson, J.C., Clegg, K.E., and Palmiter, R.D. (1996). Sensitivity to leptin and susceptibility to seizures of mice lacking neuropeptide Y. Nature 381, 415–418.
210 Bibliography
Eunson, L.H., Rea, R., Zuberi, S.M., Youroukos, S., Panayiotopoulos, C.P., Liguori, R., Avoni, P., McWilliam, R.C., Stephenson, J.B.P., Hanna, M.G., et al. (2000). Clinical, genetic, and expression studies of mutations in the potassium channel gene KCNA1 reveal new phenotypic variability. Ann. Neurol. 48, 647–656.
Favuzzi, E., Deogracias, R., Marques‐Smith, A., Maeso, P., Jezequel, J., Exposito‐Alonso, D., Balia, M., Kroon, T., Hinojosa, A.J., F. Maraver, E., et al. (2019). Distinct molecular programs regulate synapse specificity in cortical inhibitory circuits. Science (80‐. ). 363, 413–417.
Feldmeyer, D. (2012). Excitatory neuronal connectivity in the barrel cortex. Front. Neuroanat. 6, 1–22.
Fernández‐Chacón, R., and Südhof, T.C. (1999). GENETICS OF SYNAPTIC VESICLE FUNCTION: Toward the Complete Functional Anatomy of an Organelle. Annu. Rev. Physiol. 61, 753–776.
Fernández‐Chacón, R., Wölfel, M., Nishimune, H., Tabares, L., Schmitz, F., Castellano‐Muñoz, M., Rosenmund, C., Montesinos, M.L., Sanes, J.R., Schneggenburger, R., et al. (2004). The synaptic vesicle protein CSPα prevents presynaptic degeneration. Neuron 42, 237–251.
Ferrer, I., Soriano, E., Tuñón, T., Fonseca, M., and Guionnet, N. (1991). Parvalbumin immunoreactive neurons in normal human temporal neocortex and in patients with Alzheimer’s disease. J. Neurol. Sci. 106, 135–141.
Fish, K.N., Hoftman, G.D., Sheikh, W., Kitchens, M., and Lewis, D.A. (2013). Parvalbumin‐Containing Chandelier and Basket Cell Boutons Have Distinctive Modes of Maturation in Monkey Prefrontal Cortex. J. Neurosci. 33, 8352–8358.
Fonseca, M., Soriano, E., Ferrer, I., Martínez, A., and Tuñón, T. (1993). Chandelier cell axons identified by parvalbumin‐immunoreactivity in the normal human temporal cortex and in Alzheimer’s disease. Neuroscience 55, 1107–1116.
Foster, M. (1897). An address on the progress of physiology during the last thirteen years. Br. Med. J. 445–450.
Freund, T.F., and Buzsáki, G. (1996). Interneurons of the hippocampus. Hippocampus 6, 347–470.
Fulton, J.F. (1938). Physiology of the Nervous System (Oxford University Press).
Gao, R., and Penzes, P. (2015). Common Mechanisms of Excitatory and Inhibitory Imbalance in Schizophrenia and Autism Spectrum Disorders. Curr. Mol. Med. 15, 146–167.
García‐Junco‐Clemente, P., Cantero, G., Gomez‐Sanchez, L., Linares‐Clemente, P., Martinez‐Lopez, J.A., Lujan, R., and Fernandez‐Chacon, R. (2010). Cysteine String Protein‐alfa Prevents Activity‐Dependent Degeneration in GABAergic Synapses. J. Neurosci. 30, 7377–7391.
García‐Junco‐Clemente, P., Ikrar, T., Tring, E., Xu, X., Ringach, D.L., and Trachtenberg, J.T. (2017). An
211 Bibliography inhibitory pull–push circuit in frontal cortex. Nat. Neurosci. 20, 389–392.
García‐Junco‐Clemente, P., Tring, E., Ringach, D.L., and Trachtenberg, J.T. (2019). State‐Dependent Subnetworks of Parvalbumin‐Expressing Interneurons in Neocortex. Cell Rep. 26, 2282–2288.e3.
Gibson, J.R., Huber, K.M., and Sudhof, T.C. (2009). Neuroligin‐2 Deletion Selectively Decreases Inhibitory Synaptic Transmission Originating from Fast‐Spiking but Not from Somatostatin‐Positive Interneurons. J. Neurosci. 29, 13883–13897.
Glasser, M.F., Coalson, T.S., Robinson, E.C., Hacker, C.D., Harwell, J., Yacoub, E., Ugurbil, K., Andersson, J., Beckmann, C.F., Jenkinson, M., et al. (2016). A multi‐modal parcellation of human cerebral cortex. Nature 536, 171–178.
Goldberg, E.M., Clark, B.D., Zagha, E., Nahmani, M., Erisir, A., and Rudy, B. (2008). K+ Channels at the Axon Initial Segment Dampen Near‐Threshold Excitability of Neocortical Fast‐Spiking GABAergic Interneurons. Neuron 58, 387–400.
Gonchar, Y., Wang, Q., and Burkhalter, A. (2008). Multiple distinct subtypes of GABAergic neurons in mouse visual cortex identified by triple immunostaining. Front. Neuroanat. 1.
Gøtzsche, C.R., and Woldbye, D.P.D. (2016). The role of NPY in learning and memory. Neuropeptides 55, 79–89.
Goulding, M. (2009). Circuits controlling vertebrate locomotion: Moving in a new direction. Nat. Rev. Neurosci. 10, 507–518.
Greaves, J., Lemonidis, K., Gorleku,, O.A. Cruchaga, C., Grefen, C., and Chamberlain, L.H. (2012). Palmitoylation‐induced aggregation of cysteine‐string protein mutants that cause neuronal ceroid lipofuscinosis. J. Biol. Chem. 287, 37330–37339.
Grindberg, R. V., Yee‐Greenbaum, J.L., McConnell, M.J., Novotny, M., O’Shaughnessy, A.L., Lambert, G.M., Arauzo‐Bravo, M.J., Lee, J., Fishman, M., Robbins, G.E., et al. (2013). RNA‐sequencing from single nuclei. Proc. Natl. Acad. Sci. 110, 19802–19807.
De Groot, M., Toering, S.T., Boer, K., Spliet, W.G.M., Heimans, J.J., Aronica, E., and Reijneveld, J.C. (2010). Expression of synaptic vesicle protein 2A in epilepsy‐associated brain tumors and in the peritumoral cortex. Neuro. Oncol. 12, 265–273.
Grün, D., and Van Oudenaarden, A. (2015). Design and Analysis of Single‐Cell Sequencing Experiments. Cell 163, 799–810.
Gu, F., Parada, I., Yang, T., Longo, F.M., and Prince, D.A. (2018). Partial TrkB receptor activation suppresses cortical epileptogenesis through actions on parvalbumin interneurons. Neurobiol. Dis. 113,
212 Bibliography
45–58.
Gulyás, A.I., Megıas,́ M., Emri, Z., and Freund, T.F. (1999). Total Number and Ratio of Excitatory and Inhibitory Synapses Converging onto Single Interneurons of Different Types in the CA1 Area of the Rat Hippocampus. J. Neurosci. 19, 10082–10097.
Gulyás, A.I., Buzsáki, G., Freund, T.F., and Hirase, H. (2006). Populations of hippocampal inhibitory neurons express different levels of cytochrome c. Eur. J. Neurosci. 23, 2581–2594.
Hanson, P.I., Roth, R., Morisaki, H., Jahn, R., and Heuser, J.E. (1997). Structure and conformational changes in NSF and its membrane receptor complexes visualized by quick‐freeze/deep‐etch electron microscopy. Cell 90, 523–535.
Harris, K.D., and Mrsic‐Flogel, T.D. (2013). Cortical connectivity and sensory coding. Nature 503, 51–58.
Harris, K.D., Hochgerner, H., Skene, N.G., Magno, L., Katona, L., Bengtsson Gonzales, C., Somogyi, P., Kessaris, N., Linnarsson, S., and Hjerling‐Leffler, J. (2018). Classes and continua of hippocampal CA1 inhibitory neurons revealed by single‐cell transcriptomics.
Henderson, M.X., Wirak, G.S., Zhang, Y., Dai, F., Ginsberg, S.D., Dolzhanskaya, N., Staropoli, J.F., Nijssen, P.C., Lam, T.T., Roth, A.F., et al. (2016). Neuronal ceroid lipofuscinosis with DNAJC5/CSP mutations have PPT1 pathology and exhibit aberrant protein palmitoylation. Acta Neuropathol. 15, 477–491.
Hippenmeyer, S., Vrieseling, E., Sigrist, M., Portmann, T., Laengle, C., Ladle, D.R., and Arber, S. (2005). A developmental switch in the response of DRG neurons to ETS transcription factor signaling. PLoS Biol. 3, 0878–0890.
Hochgerner, H., Lönnerberg, P., Hodge, R., Mikes, J., Heskol, A., Hubschle, H., Lin, P., Picelli, S., La Manno, G., Ratz, M., et al. (2017). STRT‐seq‐2i: Dual‐index 5′ single cell and nucleus RNA‐seq on an addressable microwell array. Sci. Rep. 7, 1–8.
Hochgerner, H., Zeisel, A., Lönnerberg, P., and Linnarsson, S. (2018). Conserved properties of dentate gyrus neurogenesis across postnatal development revealed by single‐cell RNA sequencing. Nat. Neurosci. 21, 290–299.
Hsu, C.C., Davis, K.M., Jin, H., Foos, T., Floor, E., Chen, W., Tyburski, J.B., Yang, C.Y., Schloss, J. V., and Wu, J.Y. (2000). Association of L‐glutamic acid decarboxylase to the 70‐kDa heat shock protein as a potential anchoring mechanism to synaptic vesicles. J. Biol. Chem. 275, 20822–20828.
Hu, H., and Jonas, P. (2014). A supercritical density of Na+channels ensures fast signaling in GABAergic interneuron axons. Nat. Neurosci. 17, 686–693.
Hu, H., Martina, M., and Jonas, P. (2010). Dendritic mechanisms underlying rapid synaptic activation of
213
Bibliography fast‐spiking hippocampal interneurons. Science (80‐. ). 327, 52–58.
Hu, H., Gan, J., and Jonas, P. (2014). Fast‐spiking, parvalbumin+ GABAergic interneurons: From cellular design to microcircuit function. Science (80‐. ). 345, 529‐.
Ibáñez‐Costa, A., Luque, R.M., and Castaño, J.P. (2017). Cortistatin: A new link between the growth hormone/prolactin axis, stress, and metabolism. Growth Horm. IGF Res. 33, 23–27.
Ikemoto, A., Bole, D.G., and Ueda, T. (2003). Glycolysis and Glutamate Accumulation into Synaptic Vesicles. J. Biol. Chem. 278, 5929–5940.
Ishitani, R., Kimura, M., Sunaga, K., Katsube, N., Tanaka, M., and Chuang, D.M. (1996). An antisense oligodeoxynucleotide to glyceraldehyde‐3‐phosphate dehydrogenase blocks age‐induced apoptosis of mature cerebrocortical neurons in culture. J. Pharmacol. Exp. Ther. 278, 447–454.
Islam, S., Kjällquist, U., Moliner, A., Zajac, P., Fan, J.‐B., Lönnerberg, P., and Linnarsson, S. (2011). Characterization of the single‐cell transcriptional landscape by highly multiplex RNA‐seq. Genome Res. 21, 1160–1167.
Islam, S., Zeisel, A., Joost, S., La Manno, G., Zajac, P., Kasper, M., Lönnerberg, P., and Linnarsson, S. (2014). Quantitative single‐cell RNA‐seq with unique molecular identifiers. Nat. Methods 11, 163–166.
Jaitin, D.A., Kenigsberg, E., Keren‐Shaul, H., Elefant, N., Paul, F., Zaretsky, I., Mildner, A., Cohen, N., Jung, S., Tanay, A., et al. (2014). Massively Parallel Single‐Cell RNA‐Seq for Marker‐Free Decomposition of Tissues into Cell Types. Science (80‐. ). 343, 776–779.
Jang, S.R., Nelson, J.C., Bend, E.G., Rodríguez‐Laureano, L., Tueros, F.G., Cartagenova, L., Underwood, K., Jorgensen, E.M., and Colón‐Ramos, D.A. (2016). Glycolytic Enzymes Localize to Synapses under Energy Stress to Support Synaptic Function. Neuron 90, 278–291.
Janz, R., Goda, Y., Geppert, M., Missler, M., and Sü, T.C. (1999). SV2A and SV2B Function as Redundant Ca2+ Regulators in Neurotransmitter Release. Neuron 24, 1003–1016.
Jiang, J., Peng, Y., He, Z., Wei, L., Jin, W., Wang, X., and Chang, M. (2017). Intrahippocampal injection of Cortistatin‐14 impairs recognition memory consolidation in mice through activation of sst2, ghrelin and GABAA/Breceptors. Brain Res. 1666, 38–47.
Jiang, X., Lachance, M., and Rossignol, E. (2016). Involvement of cortical fast‐spiking parvalbumin‐ positive basket cells in epilepsy. Prog. Brain Res. 226, 81–126.
Johnson, M.B., Wang, P.P., Atabay, K.D., Murphy, E.A., Doan, R.N., Hecht, J.L., and Walsh, C.A. (2015). Single‐cell analysis reveals transcriptional heterogeneity of neural progenitors in human cortex. Nat. Neurosci. 18, 637–646.
214 Bibliography
Kampers, T., Friedhoff, P., Biernat, J., Mandelkow, E.M., and Mandelkow, E. (1996). RNA stimulates aggregation of microtubule‐associated protein tau into Alzheimer‐like paired helical filaments. FEBS Lett. 399, 344–349.
Kandel, E.R., Schwartz, J.H., and Jessell, T.M. (2001). Principios de Neurociencia (McGraw‐Hill).
Karagiannis, A., Gallopin, T., David, C., Battaglia, D., Geoffroy, H., Rossier, J., Hillman, E.M.C., Staiger, J.F., and Cauli, B. (2009). Classification of NPY‐Expressing Neocortical Interneurons. J. Neurosci. 29, 3642– 3659.
Karube, F., Kubota, Y., and Kawaguchi, Y. (2004). Axon Branching and Synaptic Bouton Phenotypes in GABAergic Nonpyramidal Cell Subtypes. J. Neurosci. 24, 2853–2865.
Katz, B.). (1969 The release of neural transmitter substances (Liverpool University Press).
Kawaguchi, Y., Katsumaru, H., Kosaka, T., Heizmann, C.W., and Hama, K. (1987). Fast spiking cells in rat hippocampus (CA1 region) contain the calcium‐binding protein parvalbumin. Brain Res. 416, 369–374.
Kelsom, C., and Lu, W. (2013). Development and specification ofc GABAergi cortical interneurons. Cell Biosci. 3, 1–19.
Kerr, A.M., Reisinger, E., and Jonas, P. (2008). Differential dependence of phasic transmitter release on synaptotagmin 1 at GABAergic and glutamatergic hippocampal synapses. Proc. Natl. Acad. Sci. 105, 15581–15586.
Kharchenko, P. V., Silberstein, L., and Scadden, D.T. (2014). Bayesian approach to single‐cell differential expression analysis. Nat. Methods 11, 740–742.
Kolodziejczyk, A.A., Kim, J.K., Svensson, V., Marioni, J.C., and Teichmann, S.A. (2015). The Technology and Biology of Single‐Cell RNA Sequencing. Mol. Cell 58, 610–620.
Krueger‐Burg, D., Papadopoulos, T., and Brose, N. (2017). Organizers of inhibitory synapses come of age. Curr. Opin. Neurobiol. 45, 66–77.
Kyle, B.D., Ahrendt, E., Braun, A.P., and Braun, J.E.A. (2013). The large conductance, calcium‐activated K+ (BK) Channel is regulated by Cysteine String Protein. Sci. Rep. 3, 29–32.
Labadorf, A., Choi, S.H., and Myers, R.H. (2018). Evidence for a Pan‐Neurodegenerative Disease Response in Huntington’s and Parkinson’s Disease Expression Profiles. Front. Mol. Neurosci. 10, 1–12.
Lake, B.B., Codeluppi, S., Yung, Y.C., Gao, D., Chun, J., Kharchenko, P. V., Linnarsson, S., and Kun, Z. (2017). A comparative strategy for single‐nucleus and single‐cell transcriptomes confirms accuracy in predicted cell‐type expression from nuclear RNA. Sci. Rep. 7, 1–8.
215 Bibliography
Langley, J.N. (1921). The autonomic nervous system (Cambridge: W. Heffer & Sons LTD). de Lecea, L., del Rio, J.A., Criado, J.R., Alcántara, S., Morales, M., Danielson, P.E., Henriksen, S.J., Soriano, E., and Sutcliffe, J.G. (1997). Cortistatin is expressed in a distinct subset of cortical interneurons. J. Neurosci. 17, 5868–5880.
Lee, E., Lee, J., and Kim, E. (2017). Excitation/Inhibition Imbalance in Animal Models of Autism Spectrum Disorders. Biol. Psychiatry 81, 838–847.
Lee, S., Hjerling‐leffler, J., Zagha, E., and Fishell, G. (2010). The largest group of superficial neocortical GABAergic interneurons expresses ionotropic serotonin receptors. J. Neurosci. 30, 16796–16808.
Lemmens, M.A.M., Sierksma, A.S.R., Rutten, B.P.F., Dennissen, F., Steinbusch, H.W.M., Lucassen, P.J., and Schmitz, C. (2011). Age‐related changes of neuron numbers in the frontal cortex of a transgenic mouse model of Alzheimer’s disease. Brain Struct. Funct. 216, 227–237.
Letts, V.A. (2005). Stargazer‐‐a mouse to seize! Epilepsy Curr. 5, 161–165.
Letts, V.A., Felix, R., Biddlecome, G.H., Arikkath, J., Mahaffey, C.L., Valenzuela, A., Bartlett, F.S., Mori, Y., Campbell, K.P., and Frankel, W.N. (1998). The mouse stargazer gene encodes a neuronal Ca2+‐channel γ subunit. Nat. Genet. 19, 340–347.
Lewis, D.A., and Lund, J.S. (1990). Heterogeneity of Chandelier Neurons in Monkey Neocortex : Corticotropin‐Releasing Factor‐ and Parvalbumin‐Immunoreactive Populations. J. Comp. Neurol. 293, 599–615.
Li, K.X., Lu, Y.M., Xu, Z.H., Zhang, J., Zhu, J.M., Zhang, J.M., Cao, S.X., Chen, X.J., Chen, Z., Luo, J.H., et al. (2012). Neuregulin 1 regulates excitability of fast‐spiking neurons through Kv1.1 and acts in epilepsy. Nat. Neurosci. 15, 267–273.
Li, L., Deng, J., Liu, C., Luo, H., and Guan, Y. (2016). Altered expression of neuropeptide Y receptors caused by focal cortical dysplasia in human intractable epilepsy. Oncotarget 7, 15329–15338.
Lim, D.H., LeDue, J., Mohajerani, M.H., Vanni, M.P., and Murphy, T.H. (2013). Optogenetic approaches for functional mouse brain mapping. Front. Neurosci. 7, 1–15.
Lun, A.T.L., McCarthy, D.J., and Marioni, J.C. (2016). A step‐by‐step workflow for low‐level analysis of single‐cell RNA‐seq data with Bioconductor. F1000Research 5, 2122.
Luo, Y., Coskun, V., Liang, A., Yu, J., Cheng, L., Ge, W., Shi, Z., Zhang, K., Li, C., Cui, Y., et al. (2015). Single‐ cell transcriptome analyses reveal signals to activate dormant neural stem cells. Cell 161, 1175–1188.
Van der Maaten, L., and Hinton, G. (2008). Visualizing Data using t‐SNE. J. Mach. Learn. Res. 9, 2579– 2605.
216
Bibliography
Macosko, E.Z., Basu, A., Satija, R., Nemesh, J., Shekhar, K., Goldman, M., Tirosh, I., Bialas, A.R., Kamitaki, N., Martersteck, E.M., et al. (2015). Highly parallel genome‐wide expression profiling of individual cells using nanoliter droplets. Cell 161, 1202–1214.
Madisen, L., Zwingman, T.A., Sunkin, S.M., Oh, S.W., Zariwala, H.A., Gu, H., Ng, L.L., Palmiter, R.D., Hawrylycz, M.J., Jones, A.R., et al. (2010). A robust and high‐throughput Cre reporting and characterization system for the whole mouse brain. Nat. Neurosci. 187, 14–17.
Madisen, L., Mao, T., Koch, H., Zhuo, J., Berenyi, A., Fujisawa, S., Hsu, Y.A., Iii, A.J.G., Gu, X., Zanella, S., et al. (2012). A toolbox of Cre‐dependent optogenetic transgenic mice for light‐induced activation and silencing. Nat. Neurosci. 15, 793–802.
Mancarci, B.O., Toker, L., Tripathy, S.J., Li, B., Rocco, B., Sibille, E., and Pavlidis, P. (2017). Cross‐ Laboratory Analysis of Brain Cell Type Transcriptomes with Applications to Interpretation of Bulk Tissue Data. Eneuro 4, 1–20.
La Manno, G., Soldatov, R., Zeisel, A., Braun, E., Hochgerner, H., Petukhov, V., Lidschreiber, K., Kastriti, M. e., Lönnerberg, P., Furlan, A., et al. (2018). RNA velocity of single cells. Nature 560, 494–498.
Marlin, J.J., and Carter, A.G. (2014). GABA‐A Receptor Inhibition of Local Calcium Signaling in Spines and Dendrites. J. Neurosci. 34, 15898–15911.
Marques, S., Zeisel, A., Codeluppi, S., Bruggen, D. Van, Falcão, A.M., Xiao, L., Li, H., Häring, M., Hochgerner, H., Romanov, R.A., et al. (2016). Oligodendrocyte heterogeneity in the mouse juvenile and adult central nervous system. Science (80‐. ). 352, 1326–1329.
Martenson, J.S., Yamasaki, T., Chaudhury, N.H., Albrecht, D., and Tomita, S. (2017). Assembly rules for GABAA receptor complexes in the brain. Elife 6, 1–18.
Matrone, C., Dzamko, N., Madsen, P., Nyegaard, M., Pohlmann, R., Søndergaard, R. V., Lassen, L.B., Andresen, T.L., Halliday, G.M., Jensen, P.H., et al. (2016). Mannose 6‐phosphate receptor is reduced in ‐ synuclein overexpressing models of parkinsons disease. PLoS One 11, 1–21.
Mayer, C., Hafemeister, C., Bandler, R.C., Machold, R., Batista Brito, R., Jaglin, X., Allaway, K., Butler, A., Fishell, G., and Satija, R. (2018). Developmental diversification of cortical inhibitory interneurons. Nature 555, 457–462.
McAllister, A.K., Katz, L.C., and Lo, D.C. (1999). Neurotrophins and Synaptic Plasticity. Annu. Rev. Neurosci. 22, 295–318.
McCarthy, D.J., Campbell, K.R., Lun, A.T.L., and Wills, Q.F. (2017). Scater: Pre‐processing, quality control, normalization and visualization of single‐cell RNA‐seq data in R. Bioinformatics 33, 1179–1186.
217
Bibliography
McCormick, D.A., Connors, B.W., Lighthall, J.W., and Prince, D.A. (1985). Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. J. Neurophysiol. 54, 782–806.
Méndez‐Díaz, M., Irwin, L., Gómez‐Chavarín, M., Jiménez‐Anguiano, A., Cabeza, R., Murillo‐Rodríguez, E., and Prospéro‐García, O. (2005). Cortistatin modulates memory evocation in rats. Eur. J. Pharmacol. 507, 21–28.
Micheva, K.D., Wolman, D., Mensh, B.D., Pax, E., Buchanan, J., Smith, S.J., and Bock, D.D. (2016). A large fraction of neocortical myelin ensheathes axons of local inhibitory neurons. Elife 5, 1–29.
Mihaljević, B., Bielza, C., Benavides‐Piccione, R., DeFelipe, J., and Larrañaga, P. (2014). Multi‐dimensional classification of GABAergic interneurons with Bayesian network‐modeled label uncertainty. Front. Comput. Neurosci. 8, 1–13.
Mihaljević, B., Benavides‐Piccione, R., Bielza, C., DeFelipe, J., and Larrañaga, P. (2015). Bayesian Network Classifiers for Categorizing Cortical GABAergic Interneurons. Neuroinformatics 13, 193–208.
Morton, S.M., and Bastian, A.J. (2004). Cerebellar control of balance and locomotion. Neuroscientist 10, 247–259.
Mountcastle, V.B. (1997). The columnar organization of the neocortex. Brain 120, 701–722.
Mueller, T., Remedies, C., Haroutunian, V., and Meador‐Woodruff, J. (2015). Abnormal subcellular localization of GABAA receptor subunits in schizophrenia brain. Transl. Psychiatry 16, 91–98.
Mufson, E.J., Counts, S.E., Che, S., and Ginsberg, S.D. (2006). Neuronal gene expression profiling: uncovering the molecular biology of neurodegenerative disease. Prog. Brain Res. 158, 197–222.
Munguba, H., Nikouei, K., Hochgerner, H., Oberst, P., Kouznetsova, A., Ryge, J., Batista‐brito, R., Munoz‐, A.B., Close, J., Linnarsson, S., et al. (2019). Transcriptional maintenance of cortical somatostatin interneuron subtype identity during migration. Bioarxive.
Muñoz‐Manchado, A.B., Bengtsson Gonzales, C., Zeisel, A., Munguba, H., Bekkouche, B., Skene, N.G., Lönnerberg, P., Ryge, J., Harris, K.D., Linnarsson, S., et al. (2018). Diversity of Interneurons in the Dorsal Striatum Revealed by Single‐Cell RNA Sequencing and PatchSeq. Cell Rep. 24, 2179–2190.
Muzzu, T., Mitolo, S., Gava, G.P., and Schultz, S.R. (2018). Encoding of locomotion kinematics in the mouse cerebellum. PLoS One 13, 1–26.
Nakashima, A., Tanimura‐Ito, K., Oshiro, N., Eguchi, S., Miyamoto, T., Momonami, A., Kamada, S., Yonezawa, K., and Kikkawa, U. (2013). A positive role of mammalian Tip41‐like protein, TIPRL, in the amino‐acid dependent mTORC1‐signaling pathway through interaction with PP2A. FEBS Lett. 587, 2924– 2929.
218 Bibliography
Natalie, D., Ciceri, G., Bartolini, G., Lim, L., del Pino, I., and Marín, O. (2015). Tuning of fast‐spiking interneuron properties by an activity‐dependent transcriptional switch. Science (80‐. ). 349, 1216–1220.
Ng, L., Bernard, A., Lau, C., Overly, C.C., Dong, H.W., Kuan, C., Pathak, S., Sunkin, S.M., Dang, C., Bohland, J.W., et al. (2009). An anatomic gene expression atlas of the adult mouse brain. Nat. Neurosci. 12, 356– 362.
Nieto‐Gonzalez, J.L., Moser, J., Lauritzen, M., Schmitt‐John, T., and Jensen, K. (2011). Reduced GABAergic inhibition explains cortical hyperexcitability in the wobbler mouse model of ALS. Cereb. Cortex 21, 625– 635.
Nieto‐González, J.L., Gómez‐Sánchez, L., Mavillard, F., Linares‐Clemente, P., Rivero, M.C., Valenzuela‐ Villatoro, M., Muñoz‐Bravo, J.L., Pardal, R., and Fernández‐Chacón, R. (2019). Loss of postnatal quiescence of neural stem cells through mTOR activation upon genetic removal of cysteine string protein‐α. Proc. Natl. Acad. Sci. 116, 8000–8009.
Noè, F., Pool, A.H., Nissinen, J., Gobbi, M., Bland, R., Rizzi, M., Balducci, C., Ferraguti, F., Sperk, G., During, M.J., et al. (2008). Neuropeptide Y gene therapy decreases chronic spontaneous seizures in a rat model of temporal lobe epilepsy. Brain 131, 1506–1515.
Noebels, J.L., Qiao, X., Bronson, R.T., Spencer, C., and Davisson, M.T. (1990). Stargazer: a new neurological mutant on chromosome 15 in the mouse with prolonged cortical seizures. Epilepsy Res. 7, 129–135.
Noh, G.J., Jane Tavyev Asher, Y., and Graham, J.M. (2012). Clinical review of genetic epileptic encephalopathies. Eur. J. Med. Genet. 55, 281–298.
Nosková, L., Stránecký, V., Hartmannová, H., Přistoupilová, A., Barešová, V., Ivánek, R., Hlková, H., Jahnová, H., Van Der Zee, J., Staropoli, J.F., et al. (2011). Mutations in DNAJC5, encoding cysteine‐string protein alpha, cause autosomal‐dominant adult‐onset neuronal ceroid lipofuscinosis. Am. J. Hum. Genet. 89, 241–252.
Ognjanovski, N., Schaeffer, S., Wu, J., Mofakham, S., Maruyama, D., Zochowski, M., and Aton, S.J. (2017). Parvalbumin‐expressing interneurons coordinate hippocampal network dynamics required for memory consolidation. Nat. Commun. 8, 16120.
Osten, P., and Stern‐Bach, Y. (2006). Learning from stargazin: the mouse, the phenotype and the unexpected. Curr. Opin. Neurobiol. 16, 275–280.
Paciorkowski, A.R., McDaniel, S.S., Jansen, L.A., Tully, H., Tuttle, E., Ghoneim, D.H., Tupal, S., Gunter, S.A., Vasta, V., Zhang, Q., et al. (2015). Novel mutations in ATP1A3 associated with catastrophic early life epilepsy, episodic prolonged apnea, and postnatal microcephaly. Epilepsia 56, 422–430.
219
Bibliography
Palay, S.L., and Palade, G.E. (1955). The fine structure of neurons. J. Biophys. Biochem. Cytol. 1, 69–88.
Pang, Z.P., Melicoff, E., Padgett, D., Liu, Y., Teich, A.F., Dickey, B.F., Lin, W., Adachi, R., and Sudhof, T.C. (2006). Synaptotagmin‐2 Is Essential for Survival and Contributes to Ca2+ Triggering of Neurotransmitter Release in Central and Neuromuscular Synapses. J. Neurosci. 26, 13493–13504.
Paul, A., Crow, M., Raudales, R., He, M., Gillis, J., and Huang, Z.J. (2017). Transcriptional Architecture of Synaptic Communication Delineates GABAergic Neuron Identity. Cell 171, 522–539.
Pearce, R.A. (1993). Physiological evidence for two distinct GABAA responses in rat hippocampus. Neuron 10, 189–200.
Pelkey, K.A., Chittajallu, R., Craig, M.T., Tricoire, L., Wester, J.C., and McBain, C.J. (2017). Hippocampal GABAergic Inhibitory Interneurons. Physiol. Rev. 97, 1619–1747.
Pfeffer, C.K., Xue, M., He, M., Huang, Z.J., and Scanziani, M. (2013). Inhibition of inhibition in visual cortex: The logic of connections between molecularly distinct interneurons. Nat. Neurosci. 16, 1068– 1076.
Picelli, S., Björklund, Å.K., Faridani, O.R., Sagasser, S., Winberg, G., and Sandberg, R. (2013). Smart‐seq2 for sensitive full‐length transcriptome profiling in single cells. Nat. Methods 10, 1096–1100. del Pino, I., García‐Frigola, C., Dehorter, N., Brotons‐Mas, J.R., Alvarez‐Salvado, E., MartínezdeLagrán, M., Ciceri, G., Gabaldón, M.V., Moratal, D., Dierssen, M., et al. (2013). Erbb4 Deletion from Fast‐Spiking Interneurons Causes Schizophrenia‐like Phenotypes. Neuron 79, 1152–1168.
Plachta, N., Annaheim, C., Bissière, S., Lin, S., Rüegg, M., Hoving, S., Müller, D., Poirier, F., Bibel, M., and Barde, Y.A. (2007). Identification of a lectin causing the degeneration of neuronal processes using engineered embryonic stem cells. Nat. Neurosci. 10, 712–719.
Pollen, A.A., Nowakowski, T.J., Shuga, J., Wang, X., Leyrat, A.A., Lui, J.H., Li, N., Szpankowski, L., Fowler, B., Chen, P., et al. (2014). Low‐coverage single‐cell mRNA sequencing reveals cellular heterogeneity and activated signaling pathways in developing cerebral cortex. Nat. Biotechnol. 32, 1053–1058.
Pouille, F., Watkinson, O., Scanziani, M., and Trevelyan, A.J. (2013). The contribution of synaptic location to inhibitory gain control in pyramidal cells. Physiol. Rep. 1, 1–9.
Poulin, J.F., Tasic, B., Hjerling‐Leffler, J., Trimarchi, J.M., and Awatramani, R. (2016). Disentangling neural cell diversity using single‐cell transcriptomics. Nat. Neurosci. 19, 1131–1141.
Qiu, S., Luo, S., Evgrafov, O., Li, R., Schroth, G.P., Levitt, P., Knowles, J.A., and Wang, K. (2012). Single‐ neuron RNA‐Seq: Technical feasibility and reproducibility. Front. Genet. 3, 1–8.
Rangaraju, V., Calloway, N., and Ryan, T.A. (2014). Activity‐driven local ATP synthesis is required for
220
Bibliography synaptic function. Cell 156, 825–835.
Rankin‐Gee, E.K., McRae, P.A., Baranov, E., Rogers, S., Wandrey, L., and Porter, B.E. (2015). Perineuronal net degradation in epilepsy. Epilepsia 56, 1124–1133.
Reichmann, F., and Holzer, P. (2016). Neuropeptide Y: A stressful review. Neuropeptides 55, 99–109.
Robbins, C.A., and Tempel, B.L. (2012).1 Kv1. and Kv1.2: Similar channels, different seizure models. Epilepsia 53, 134–141.
Rosales, K.R., Reid, M.A., Yang, Y., Tran, T.Q., Wang, W.I., Lowman, X., Pan, M., and Kong, M. (2015). TIPRL inhibits protein phosphatase 4 activity and promotes H2AX Phosphorylation in the DNA Damage Response. PLoS One 10, 1–13.
Rossier, J., Bernard, A., Cabungcal, J.H., Perrenoud, Q., Savoye, A., Gallopin, T., Hawrylycz, M., Cuénod, M., Do, K., Urban, A., et al. (2015). Cortical fast‐spiking parvalbumin interneurons enwrapped in the perineuronal net express the metallopeptidases Adamts8, Adamts15 and Neprilysin. Mol. Psychiatry 20, 154–161.
Rozas, J.L., Gómez‐Sánchez, L., Mircheski, J., Linares‐Clemente, P., Nieto‐González, J.L., Vázquez, E.M., Luján, R., and Fernández‐Chacón, R. (2012). Motorneurons Require Cysteine String Protein‐α to Maintain the Readily Releasable Vesicular Pool and Synaptic Vesicle Recycling. Neuron 74, 151–165.
Saftig, P., and Klumperman, J. (2009). Lysosome biogenesis and lysosomal membrane proteins: Trafficking meets function. Nat. Rev. Mol. Cell Biol. 10, 623–635.
Sambri, I., D’Alessio, R., Ezhova, Y., Giuliano, T., Sorrentino, N.C., Cacace, V., De Risi, M., Cataldi, M., Annunziato, L., De Leonibus, E., et al. (2017). Lysosomal dysfunction disrupts presynaptic maintenance and restoration of presynaptic function prevents neurodegeneration in lysosomal storage diseases. EMBO Mol. Med. 9, 112–132.
Saxena, N.C., and Macdonald, R.L. (1994). Assembly of GABA, receptor subunits: Role of the delta subunit. J. Neurosci. 14, 7077–7088.
Schmitz, F., and Fernández‐Chacón, R. (2009). Cysteine‐String Proteins (CSPs). Encycl. Neurosci. 3, 285– 292.
Schmitz, F., Tabares, L., Khimich, D., Strenzke, N., Villa‐Polo, P. d. l., Castellano‐Munoz, M., Bulankina, A., Moser, T., Fernandez‐Chacon, R., and Sudhof, T.C. (2006). CSP ‐deficiency causes massive and rapid photoreceptor degeneration. Proc. Natl. Acad. Sci. 103, 2926–2931.
Scorsato, V., Lima, T.B., Righetto, G.L., Zanchin, N.I.T., Brandão‐Neto, J., Sandy, J., Pereira, H.D.M., Ferrari, Á.J.R., Gozzo, F.C., Smetana, J.H.C., et al. (2016). Crystal structure of the human Tip41
221 Bibliography orthologue, TIPRL, reveals a novel fold and a binding site for the PP2Ac C‐terminus. Sci. Rep. 6, 1–12.
Segarra‐Mondejar, M., Casellas‐Díaz, S., Ramiro‐Pareta, M., Müller‐Sánchez, C., Martorell‐Riera, A., Hermelo, I., Reina, M., Aragonés, J., Martínez‐Estrada, O.M., and Soriano, F.X. (2018). Synaptic activity‐ induced glycolysis facilitates membrane lipid provision and neurite outgrowth. EMBO J. 37, e97368.
Selten, M., van Bokhoven, H., and Nadif Kasri, N. (2018). Inhibitory control of the excitatory/inhibitory balance in psychiatric disorders. F1000Research 7, 23.
Sharma, M., Burré, J., and Südhof, T.C. (2011). CSPα promotes SNARE‐complex assembly by chaperoning SNAP‐25 during synaptic activity. Nat. Cell Biol. 13, 30–39.
Sharma, M., Burré, J., Bronk, P., Zhang, Y., Xu, W., and Südhof, T.C. (2012). CSPα knockout causes neurodegeneration by impairing SNAP‐25 function. EMBO J. 31, 829–841.
Sigel, E., and Steinmann, M.E. (2012). Structure, function, and modulation of GABAA receptors. J. Biol. Chem. 287, 40224–40231.
Smart, S.L., Lopantsev, V., Zhang, C.L., Robbins, C.A., Wang, H., Chiu, S.Y., Schwartzkroin, P.A., Messing, A., and Tempel, B.L. (1998). Deletion of the Kv1.1 Potassium channel causes epilepsy in mice. Neuron 20, 809–819.
Smedemark‐Margulies, N., Brownstein, C.A., Vargas, S., Tembulkar, S.K., Towne, M.C., Shi, J., Gonzalez‐ Cuevas, E., Liu, K.X., Bilguvar, K., Kleiman, R.J., et al. (2016). A novel de novo mutation in ATP1A3 and childhood‐onset schizophrenia. Cold Spring Harb. Mol. Case Stud. 2, a001008.
Söllner, T., Whiteheart, S.W., Brunner, M., Erdjument‐Bromage, H., Geromanos, S., Tempst, P., and Rothman, J.E. (1993). SNAP receptors implicated in vesicle targeting and fusion. Nature 362, 318–324.
Sommeijer, J.P., and Levelt, C.N. (2012). Synaptotagmin‐2 is a reliable marker for parvalbumin positive inhibitory boutons in the mouse visual cortex. PLoS One 7, 1–12.
Song, I.S., Jun, S.Y., Na, H.J., Kim, H.T., Jung, S.Y., Ha, G.H., Park, Y.H., Long, L.Z., Yu, D.Y., Kim, J.M., et al. (2012). Inhibition of MKK7‐JNK by the TOR signaling pathway regulator‐like protein contributes to resistance of HCC cells to TRAIL‐induced apoptosis. Gastroenterology 143, 1341–1351.
Srivastava, S., Cohen, J., Pevsner, J., Aradhya, S., McKnight, D., Butler, E., Johnston, M., and Fatemi, A. (2015). A Novel Variantin GABRB2 Associated with Intellectual Disability and Epilepsy. Am. J. Med. Genet. 18, 1089–1098.
Stahl, B., Tobaben, S., and Südhof, T.C. (1999). Two distinct domains in hsc70 are essential for the interaction with the synaptic vesicle cysteine string protein. Eur. J. Cell Biol. 78, 375–381.
Starossom, S.C., Mascanfroni, I.D., Imitola, J., Cao, L., Raddassi, K., Hernandez, S.F., Bassil, R., Croci, D.O.,
222
Bibliography
Cerliani, J.P., Delacour, D., et al. (2013). Galectin‐1 Deactivates Classically‐Activated Microglia and Protects from Inflammation‐Induced Neurodegeneration. Immunity 37, 249–263.
Stedehouder, J., Couey, J.J., Brizee, D., Hosseini, B., Slotman, J.A., Dirven, C.M.F., Shpak, G., Houtsmuller, A.B., and Kushner, S.A. (2017). Fast‐spiking Parvalbumin Interneurons are Frequently Myelinated in the Cerebral Cortex of Mice and Humans. Cereb. Cortex 27, 5001–5013.
Stuart, G., Spruston, N., and Häusser, M. (2008). Dendrites (Oxford University Press).
Südhof, T.C. (2004). The Synaptic Vesicle Cycle. Annu. Rev. Neurosci. 27, 509–547.
Südhof, T.C. (2014). The molecular machinery of neurotransmitter release (nobel lecture). Angew. Chemie ‐ Int. Ed. 53, 12696–12717.
Südhof, T.C., and Rizo, J. (2011). Synaptic Vesicle Exocytosis. Cold Spring Harb. Perspect. Biol. 3, a005637.
Syrbe, S., Hedrich, U.B.S., Riesch, E., Djémié, T., Müller, S., Møller, R.S., Maher, B., Hernandez‐Hernandez, L., Synofzik, M., Caglayan, H.S., et al. (2015). De novo loss‐or gain‐of‐function mutations in KCNA2 cause epileptic encephalopathy. Nat. Genet. 47, 393–399.
Takahashi, H., Brasnjevic, I., Rutten, B.P.F., Van Der Kolk, N., Perl, D.P., Bouras, C., Steinbusch, H.W.M., Schmitz, C., Hof, P.R., and Dickstein, D.L. (2010). Hippocampal interneuron loss in an APP/PS1 double mutant mouse and in Alzheimer’s disease. Brain Struct. Funct. 214, 145–160.
Takamori, S., Holt, M., Stenius, K., Lemke, E.A., Grønborg, M., Riedel, D., Urlaub, H., Schenck, S., Brügger, B., Ringler, P., et al. (2006). Molecular Anatomy of a Trafficking Organelle. Cell 127, 831–846.
Tallent, M.K., Fabre, V., Qiu, C., Calbet, M., Lamp, T., Baratta, M. V., Suzuki, C., Levy, C.L., Siggins, G.R., Henriksen, S.J., et al. (2005). Cortistatin overexpression in transgenic mice produces deficits in synaptic plasticity and learning. Mol. Cell. Neurosci. 30, 465–475.
Tamamaki, N., Yanagawa, Y., Tomioka, R., Miyazaki, J.I., Obata, K., and Kaneko, T. (2003). Green Fluorescent Protein Expression and Colocalization with Calretinin, Parvalbumin, and Somatostatin in the GAD67‐GFP Knock‐In Mouse. J. Comp. Neurol. 467, 60–79.
Tanaka, R., Mochizuki, H., Suzuki, A., Katsube, N., Ishitani, R., Mizuno, Y., and Urabe, T. (2002). Induction of glyceraldehyde‐3‐phosphate dehydrogenase (GAPDH) expression in rat brain after focal ischemia/reperfusion. J. Cereb. Blood Flow Metab. 22, 280–288.
Tanaka, Y., Tanaka, Y., Furuta, T., Yanagawa, Y., and Kaneko, T. (2008). The effects of cutting solutions on the viability of GABAergic interneurons in cerebral cortical slices of adult mice. J. Neurosci. Methods 171, 118–125.
Tang, F., Barbacioru, C., Wang, Y., Nordman, E., Lee, C., Xu, N., Wang, X., Bodeau, J., Tuch, B.B., Siddiqui,
223 Bibliography
A., et al. (2009). mRNA‐sequencing whole transcriptome analysis of a single cell. Nat. Methods 6, 377– 382.
Tansey, E.. (1997). Not Committing Barbarisms: Sherrington and the Synapse, 1897. Brain Res. Bull. 44, 211–212.
Tasic, B., Menon, V., Nguyen, T.N., Kim, T.K., Jarsky, T., Yao, Z., Levi, B., Gray, L.T., Sorensen, S.A., Dolbeare, T., et al. (2016). Adult mouse cortical cell taxonomy revealed by single cell transcriptomics. Nat. Neurosci. 19, 335–346.
Tasic, B., Yao, Z., Smith, K.A., Graybuck, L., Nguyen, T.N., Bertagnolli, D., Goldy, J., Garren, E., Economo, M.N., Viswanathan, S., et al. (2018). Shared and distinct transcriptomic cell types across neocortical areas. Nature 563, 72–78.
Tatsukawa, T., Ogiwara, I., Mazaki, E., Shimohata, A., and Yamakawa, K. (2018). Impairments in social novelty recognition and spatial memory in mice with conditional deletion of Scn1a in parvalbumin‐ expressing cells. Neurobiol. Dis. 112, 24–34.
Tay, T.L., Sagar, Dautzenberg, J., Grün, D., and Prinz, M. (2018). Unique microglia recovery population revealed by single‐cell RNAseq following neurodegeneration. Acta Neuropathol. Commun. 6, 87.
The Tabula Muris Consortium (2018). Single‐cell transcriptomics of 20 mouse organs creates a Tabula Muris. Nature 562, 367–372.
Tobaben, S., Thakur, P., Fernández‐Chacón, R., Südhof, T.C., Rettig, J., and Stahl, B. (2001). A trimeric protein complex functions as a synaptic chaperone machine. Neuron 31, 987–999.
Tokudome, K., Okumura, T., Shimizu, S., Mashimo, T., Takizawa, A., Serikawa, T., Terada, R., Ishihara, S., Kunisawa, N., Sasa, M., et al. (2016). Synaptic vesicle glycoprotein 2A (SV2A) regulates kindling epileptogenesis via GABAergic neurotransmission. Sci. Rep. 6, 1–11.
Tremblay, R., Lee, S., and Rudy, B. (2016). GABAergic interneurons in the neocortex: From cellular properties to circuits. Neuron 91, 260–292.
Treutlein, B., Brownfield, D.G., Wu, A.R., Neff, N.F., Mantalas, G.L., Espinoza, F.H., Desai, T.J., Krasnow, M.A., and Quake, S.R. (2014). Reconstructing lineage hierarchies of the distal lung epithelium using single‐cell RNA‐seq. Nature 509, 371–375.
Uematsu, M., Hirai, Y., Karube, F., Ebihara, S., Kato, M., Abe, K., Obata, K., Yoshida, S., Hirabayashi, M., Yanagawa, Y., et al. (2008). Quantitative Chemical Composition of Cortical GABAergic Neurons Revealed in Transgenic Venus‐Expressing Rats. Cereb. Cortex 18, 315–330.
Usoskin, D., Furlan, A., Islam, S., Abdo, H., Lönnerberg, P., Lou, D., Hjerling‐Leffler, J., Haeggström, J.,
224
Bibliography
Kharchenko, O., Kharchenko, P. V., et al. (2015). Unbiased classification of sensory neuron types by large‐ scale single‐cell RNA sequencing. Nat. Neurosci. 18, 145–153.
Van Vliet, E.A., Aronica, E., Redeker, S., Boer, K., and Gorter, J.A. (2009). Decreased expression of synaptic vesicle protein 2A, the binding site for levetiracetam, during epileptogenesis and chronic epilepsy. Epilepsia 50, 422–433.
Whissell, P.D., Cajanding, J.D., Fogel, N., and Kim, J.C. (2015). Comparative density of CCK‐ and PV‐GABA cells within the cortex and hippocampus. Front. Neuroanat. 9, 1–16.
Wonders, C.P., and Anderson, S.A. (2006). The origin and specification of cortical interneurons. Nat. Rev. Neurosci. 7, 687–696.
Xu, Y., Cui, L., Dibello, A., Wang, L., Lee, J., Saidi, L., Lee, J.G., and Ye, Y. (2018). DNAJC5 facilitates USP19‐ dependent unconventional secretion of misfolded cytosolic proteins. Cell Discov. 4, 11.
Xue, M., Atallah, B. V., and Scanziani, M. (2014). Equalizing excitation‐inhibition ratios across visual cortical neurons. Nature 511, 596–600.
Yeung, R.K., Xiang, Z.H., Tsang, S.Y., Li, R., Ho, T.Y.C., Li, Q., Hui, C.K., Sham, P.C., Qiao, M.Q., and Xue, H. (2018). Gabrb2‐knockout mice displayed schizophrenia‐like and comorbid phenotypes with interneuron‐ astrocyte‐microglia dysregulation. Transl. Psychiatry 8, 128.
Yip, J., Geng, X., Shen, J., and Ding, Y. (2017). Cerebral gluconeogenesis and diseases. Front. Pharmacol. 7, 1–12.
Zajac, P., Islam, S., Hochgerner, H., Lönnerberg, P., and Linnarsson, S. (2013). Base preferences in non‐ templated nucleotide incorporation by MMLV‐derived reverse transcriptases. PLoS One 8, 1–13.
Zeisel, A., Muñoz‐Manchado, A.B., Codeluppi,, S. Lönnerberg, P., Manno, G. La, Juréus, A., Marques, S., Munguba, H., He, L., Betsholtz, C., et al. (2015). Cell types in the mouse cortex and hippocampus revealed by single‐cell RNA‐seq. Science (80‐. ). 347, 1138–1142.
Zhang, H., Kelley, W.L., Chamberlain, L.H., Burgoyne, R.D., and Lang, J. (1999). Mutational analysis of cysteine‐string protein function in insulin exocytosis. J. Cell Sci. 112, 1345–1351.
Zhao, S., Zhang, Y., Gamini, R., Zhang, B., and Von Schack, D. (2018). Evaluation of two main RNA‐seq approaches for gene quantification in clinical RNA sequencing: PolyA+ selection versus rRNA depletion. Sci. Rep. 8, 1–12.
Zinsmaier, K.E., Hofbauer, A., Heimbeck, G., Pflugfelder, G.O., Buchner, S., and Buchner, E. (1990). A cysteine‐string protein is expressed in retina and brain of drosophila. J. Neurogenet. 7, 15–29.
Zinsmaier, K.E., Eberle, K.K., Buchner, E., Walter, N., and Benzer, S. (1994). Paralysis and Early Death in
225 Bibliography
Cysteine String Protein Mutants of Drosophila. Science (80‐. ). 263, 977–980. http://www.biobserve.com/behavioralresearch/products/viewer/. https://imagej.nih.gov/ij/download.html. http://escg.se/methods/strt‐seq‐2i/. https://www.r‐project.org/. https://cran.r‐project.org/bin/windows/base/old/3.4.4/. https://www.rstudio.com/products/rstudio/download/. https://github.com/satijalab/seurat. https://satijalab.org/seurat/immune_alignment.html. https://towardsdatascience.com/dimensionality‐reduction‐does‐pca‐really‐improve‐classification‐ outcome‐6e9ba21f0a32. https://satijalab.org/seurat/pbmc3k_tutorial.html. https://github.com/satijalab/seurat. http://celltypes.brain‐map.org/rnaseq/mouse. https://github.com/linnarsson‐lab/ceftools.
226