Resolving cell state in iPSC-derived human neural samples with multiplexed fluorescence imaging

Human induced pluripotent stem cell-derived (iPSC) neural cultures offer clinically relevant models of human diseases, including Amyotrophic Lateral Sclerosis, Alzheimer’s, and Autism Spectrum Disorder. In situ characterization of the spatial-temporal evolution of cell state in 3D culture and subsequent 2D dissociated culture models based on protein expression levels and localizations is essential to understanding neural cell differentiation, disease state phenotypes, and sample-to-sample variability. Here, we apply PRobe-based Imaging for Sequential Multiplexing (PRISM) to facilitate multiplexed imaging with facile, rapid exchange of imaging probes to analyze iPSC-derived cortical and motor neuron cultures that are relevant to psychiatric and neurodegenerative disease models, using over ten protein targets. Our approach permits analysis of cell differentiation, cell composition, and functional marker expression in complex stem-cell derived neural cultures. Furthermore, our approach is amenable to automation, offering in principle the ability to scale-up to dozens of protein targets and samples.


Introduction 25
Induced pluripotent stem cell (iPSC)-derived cultures are increasingly becoming the principal 26 source of patient-specific and disease-specific cellular material for in vitro disease modeling. iPSC-27 derived cortical and motor neuron cultures have successfully been used to model 28 neurodevelopmental conditions including autism spectrum disorder (ASD; cortical neurons) [1-5] 29 and neurodegenerative conditions including spinal muscular atrophy and amyotrophic lateral 30 sclerosis (SMA, ALS; motor neurons) [6][7][8][9]. Such iPSC-based models are attractive because they 31 can generate the large numbers of neural cells needed for drug screening. They can also 32 recapitulate aspects of cortical and motor neuronal synaptic networks, which allow for functional 33 models of neurodevelopmental and neurodegenerative conditions to be developed in vitro. 34 However, phenotypic characterization of these cultures is challenging due to their complexity, 35 heterogeneity, and variability; this provides a clear opportunity for improved high content analysis 36 techniques, especially those that can produce multidimensional readouts from the same culture. 37 Fluorescence-based antibody staining of target markers is one approach to characterizing in situ 38 stem-cell derived neural cultures, but is limited by the conventional spectral limit of fluorophores 39 to imaging four targets simultaneously. One technique to overcome this limitation uses DNA-40 conjugated antibodies to fluorescently image more than 10 individual markers in the same sample 41 [10,11]. This procedure, called PRobe-based Imaging for Sequential Multiplexing (PRISM), 42 eliminates the need for antibody stripping or removal for multiplexing, a limiting factor in traditional 43 immunofluorescence/immunocytochemistry (IF/ICC)-based multiplexed imaging strategies [12,44 13], thereby allowing characterization using a greater number of molecular targets. Briefly, PRISM 45 antibodies use short, orthogonal oligonucleotide sequences that are complementary to either a 46 fluorophore-conjugated locked nucleic acid (LNA) or DNA strand that reversibly hybridizes to 47 produce a fluorescent readout, analogous to traditional IF/ICC imaging [11,14,15] and DNA-48 PAINT/EXCHANGE-PAINT [14]. Further, compared to standard antibody stripping procedures 49 [16][17][18], PRISM offers non-destructive imaging probe exchange, cycling fluorescent imaging 50 strands within several minutes permitting multiple rounds of imaging data acquisition [19][20][21] from 51 the same culture. These features allow the use of large panels of markers, consequently providing 52 higher content datasets. Generation of oligo pairs is relatively straightforward due to the use of 53 commercially available thiolated and fluorescently labelled oligo strands that can be readily 54 conjugated to a wide variety of commercially available antibodies [11,14,15]. 55 Here, we apply PRISM to 2D and 3D stem cell-derived cortical and motor neuron cultures to 56 characterize cell identity and population composition based on detection of structural, synaptic, 57 and transcription factor markers (Figure 1). Identification of multiple cell states within the same 58 iPSC-derived 2D or 3D sample helps spatially map the temporal evolution and heterogeneity of 59 targets in human samples, in addition to characterizing structural and synaptic features pertinent 60 to human disease phenotypes [22][23][24]. We also introduce a CellProfiler/FIJI computational pipeline 61 to analyze imaging data regarding cell state/phenotype in an automated, quantitative, unbiased 62 manner [23,24]. 2D cultures can be used for high-throughput characterization of stem cell-derived 63 neurons, which lends itself well to automation assays, whereas 3D organoid sections retain the 64 higher order structure of cellular organization, especially in neural cultures, which can be critical 65 for disease modeling in vitro. Because PRISM is a direct analog to IF/ICC imaging, our approach 66 complements assays in model culture systems currently used to evaluate stem cell differentiation, 67 compound and drug candidate screening, tissue engineering, and potentially in vitro disease 68 diagnostics development. 69

RESULTS 70
We built 12 antibodies into a PRISM panel to characterize the cellular composition of iPSC-derived 71 CN and MN cultures. Specificity of target imaging was confirmed using traditional IF and compared 72 to respective PRISM markers. We used established stem cell-derived astrocyte, CN, and MN 73 cultures to screen a large panel of neural markers (SI Table 1)-before narrowing down to a panel 74 of 12 PRISM-compatible antibodies (SI Table 2, Figure 2a). Based on staining intensity, signal 75 specificity, and co-localizations between target markers, we were able to use 3+ specific markers 76 to define a cell's identity (SI Table 2 and Figure 2a). For each selected antibody, we further 77 compared the staining signal between IF/ICC and PRISM before and after PRISM conjugation 78 (Figure 2b and SI Figure 4) to ensure each was still highly selective for its target and that the 79 fluorescence pattern was consistent and unchanged between IF/ICC control and PRISM signal 80 (SI Figure 7). 81 Newly generated DNA imaging strands were supplemented with published LNA imaging strands 82 [11] for an initial round of 10 DNA-PRISM pairs (SI Table 7 and SI Figure 4 and SI Figure 5).

83
Fluorescence signals were suitable for both manual and automated quantification of all 10 PRISM 84 markers and 2 control IF/ICC markers (SI Figure 6). In order to confirm that PRISM signal is 85 significantly above background, and thus suitable to imaging, we compared DNA-PRISM with 86 traditional IF using cross-correlation analysis to map specific PRISM signals to subsequent IF 87 signals in the same culture validated all 12 markers in human iPSC-derived CN (SI Figure 7) and 88 in rat hippocampal neurons validated the tested markers (SI Figure 8). and minimal off-target binding and non-specific signal between the validated oligos and any 97 present cell DNA or RNA transcripts, which is necessary for imaging nuclear transcription factors 98 and to reduce non-specific binding to native DNA and/or RNA even after salmon sperm DNA 99 blocking and RNAse treatment. 100 PRISM enables high content imaging of over ten neural markers in 2D and 3D iPSC-derived 101 CN and MN cultures. 3D iPSC-derived CN and MN cultures were dissociated, plated in standard 102 2D culture wells, maintained for 14 days, and then fixed and stained for analysis. First, BJ-SiPs 103 iPSCs were differentiated into CN (SI Table 4) and characterized via multiple markers by 104 traditional IF/ICC to confirm their cortical nature (SI Figure 1). Representative images of the 2D 105 CN culture illustrate multiple cell subtypes (Figure 3a). We further characterized in-depth two time 106 points during differentiation of cortical cultures in relation to cell identity (Figure 3b). At day 55 of 107 cortical differentiation, there were 84 cells that fulfilled our analysis criteria (Figure 2a) where we 108 were able to positively identify 30% of the analyzed cells as neurons. Synaptic marker staining 109 characterized 100% of identified neurons as excitatory in this earlier culture (VGLUT1+). 110 Immature/inactive astrocytes (CD44+/Vimentin+) made up 18% of the culture and 7% of the cells 111 expressed markers for mature/activated astrocytes (CD44+/GFAP+). We were also able to 112 positively identify both neural progenitor cells (2%, Pax6+) and a significant percentage of radial 113 glial cells (36%, Pax6+/Vimentin+). Cells that were present but could not be positively identified 114 represented 7% of the total population. We then followed up with a day 85 differentiation timepoint, 115 where we identified 161 cells that fulfilled the criteria for analysis as outlined in the Methods 116 section. Of these, 58% were positively identified as neurons (Tuj1+/MAP2+) and 26% as 117 astrocytes (CD44+/Vimentin+/GFAP+). Neurons were successfully further sub-typed based on 118 synaptic marker expression into either excitatory (52%, VGLUT1+) or inhibitory (3%, VGAT+). The 119 remaining 45% of identified neurons were not strongly associated with either type. Astrocytes were 120 further characterized as either immature/inactive (19%, Pax6-/Vimentin+/CD44-) or mature/active 121 (7%, Pax6-/GFAP+/CD44+). We observed no significant presence of either neural progenitor 122 (Pax6+) or radial glial cells (Pax6+/Vimentin+). We were unable to classify 16% of cells present, 123 based on our characterization criteria. 124 A parallel PRISM assay of iPSC-derived 2D MN cultures validated the full panel of conjugated 125 antibodies (Figure 4) to characterize the MN cultures in a similar manner. To validate that the 126 generated neuronal cells are indeed motor neurons, we stained separate wells on the same plate 127 with the motor neurons markers Islet 1 and 200kD neuroflament protein (SI Figure 2) 128 Subsequently, PRISM antibodies generated highly specific signals, and the images were then 129 overlaid to generate high-content imaging datasets. We show here that DNA-PRISM antibodies 130 can characterize complex stem cell-derived cultures in a multi-dimensional manner with the added 131 ability to preserve the spatial relationships between the markers used in our characterization 132 pipeline. Further optimization to improve staining quality by testing antibodies that are generated 133 to produce high specificity and optimized signal-to-noise ratios in the putative MN cultures in 134 particular is ongoing. 135

Automated pipeline to pair PRISM high-content data analysis with high throughput data 136
generation. Due to the high volume of raw data that is generated in a PRISM imaging assay, we 137 endeavored to develop an automated platform that could manage staining, imaging, and data 138 analysis in the same pipeline. We first identified key points in our manual PRISM assay data 139 acquisition process, namely buffer exchanges, imager strand incubation, and IF/ICC imaging, and 140 automated them using a BRAVO liquid handler and a PerkinElmer Phenix spinning disk confocal 141 microscope. Next, points of adaptation within the physical assay steps such as buffer volume, 142 aspiration and dispensing speed, and incubation times were streamlined to improve automation. 143 Slowing buffer aspiration and addition into the wells significantly reduced cell detachment and 144 damage during the twenty consecutive buffer exchanges that were necessary for a complete 145 PRISM assay. Representative PRISM staining performed with automation (SI Figure 10a) was 146 comparable to manual PRISM staining and imaging (Figure 3a). While the average time for data 147 acquisition was similar between the manual and automated assays, incorporating automated 148 CellProfiler and FIJI analysis pipelines (available upon request) reduced the analysis time by 149 almost 3-fold resulting in an overall reduction of assay duration, while increasing throughput (SI 150 Figure 10b). 151 Recently, high content analysis has been used to characterize multiple neurological cell types and 154 their interactions in vivo and in vitro [25,26] Table 3 for detailed protocol). We further chose DNA oligo pairs between 11-12 nucleotides with 170 GC content between 30-40%, which we found to be optimal for reducing background fluorescence 171 while still allowing complete washout or exchange of imaging strands post-acquisition. High 172 melting temperature of the DNA/LNA oligos may push PRISM into a range where samples need 173 to be heated, even under low salt concentration, in order to release imaging strands. High melting 174 temperature also can cause fluorescence artifacts such as residual signal and non-specific binding 175 and increased cell detachment due to the need for heating and cooling of the sample, thereby 176 complicating assay automation [21,34,35]. While fluorescence from the PRISM staining was 177 considerably lower than the corresponding signal from conventional secondary antibodies (SI 178 Figure 4c), it was still well above backround levels and this potential limitation coul be overcome 179 with further development of the technique to increase number of fluorophores on the imager 180 strands, or with longer exposure times. 181 The 12 markers that we used in our assay allowed in depth analysis suggesting that disease-182 specific morphological and gene expression differences could be elucidated. These findings open 183 the door to using PRISM antibodies for drug screening and characterizing in vitro disease models.

184
We further expanded our fluorophore detection lines from one [11] to two, halving the time 185 necessary for imaging and reducing the number of wash steps, which decreased cell detachment, 186 thus improving our multi-dimensional analysis reproducibility. To take full advantage of the multi-187 dimensional data set that was generated, we incorporated in-depth analysis capabilities using a 188 custom commonly used neural culture-specific antibodies (SI Table 1), we chose a set of twelve markers 204 to build the PRISM antibody panel (Figure 2). The antibody panel features housekeeping and 205 structural targets (α-Tubulin, Actin), markers for canonical neural culture characterization (Tuj1, 206 Map2, Synapsin I, GFAP, Vimentin, CD44), as well as cell-specific markers, including those for 207 identifying cortical (VGAT, VGLUT1) and motor neurons (Islet1), glial cell subtypes (Vimentin, 208 GFAP, CD44), neural progenitor cells (Pax6), and residual pluripotent cells (Oct4A) that might 209 have been retained post-differentiation (SI Table 2 and all secondary fluorophore-conjugated antibodies at 1:1000 dilution for IF. Secondary PRISM-224 conjugated antibodies were used at 1:200 dilution. We used highly cross-adsorbed secondary 225 antibodies raised in donkey for PRISM, to minimize any possible signal cross-talk and fluorophore 226 conjugated secondary antibodies also raised in donkey as negative/positive controls. The full list 227 of antibodies is provided in SI were grown in mTeSR1 supplemented with the above factors for the initial 24 hours, then the cells 257 were maintained in Knockout Serum Replacement (15% KSR; Thermo Fisher Scientific) medium 258 for the next 3 days, again supplemented with the above factors. Between days 5 to 11, we 259 gradually transitioned the cells into NIM as follows. On day 5, 75% KSR Pulse medium was mixed 260 with 25% NIM medium supplemented with SB431542 (10 µM) and LDN193189 (1 µM) and cells 261 were incubated for 2 days. On day 7, 50% KSR medium and 50% NIM medium were 262 supplemented with LDN193189 (1 µM) for another 2 days. On day 9, cells were kept in 25% KSR 263 mixed with 75% NIM medium with LDN193189 (1 µM) for another 2 days. On day 11, the 3D 264 cultures were fully transitioned to NIM medium for 9 days with full medium changes every 3 days.

265
The NIM medium at this point was not supplemented with any additional small molecules. On day 266 20, spheroids were transitioned to NB medium, supplemented with B27 Supplement (1x; catalog 267 #A3582801 ThermoFisher), N2 Supplement (1x; catalog #17502048 ThermoFisher), brain-268 derived neurotrophic factor (BDNF; 10 ng/mL; catalog #248-BD-010 Tocris) and glial cell-derived 269 neurotrophic factor (GDNF; 10 ng/mL; catalog #212-GD-010 Tocris) and maintained with full 270 medium changes every 4 days until day 40. At this point, mitotically active cortical progenitors 271 could be cryo-stored or maintained as spheroids for a further 14 days or more (up to 100 days) to 272 generate mature CN for functional analysis, with 50% medium changes every 3 days. For staining 273 and analysis, spheroids were dissociated in 0.25% Trypsin-EDTA (catalog #15575020 274 Thermofisher) to single cells. The cell suspension was plated into 96-well plates that were 275 previously coated with poly-D-lysine (25 µg/mL) and poly-L-ornithine (25 µg/mL) overnight at 37°C, 276 and then further coated with Laminin ((10 µg/mL; catalog #11243217001 Sigma-Aldrich) for 3 277 hours at 37°C. These 2D cultures were then matured for an additional 14 days prior to DNA-278 PRISM marker analysis, with 50% medium changes every 3 days. See SI Table 4 for detailed 279 protocol steps and SI Figure 1 for representative images of cortical cultures 280 Human iPSC-derived motor neuron differentiation and maintenance. Motor neurons (MN) were 281 generated using a modified version of previously established protocols [45,46]. iPSC colonies 282 were dissociated into single cells using Accutase. Cells were then seeded into ultra-low attachment dishes in mTESR1 medium supplemented with ROCK inhibitor, Y-27632 (10 µM; 284 catalog #1254 Tocris) and basic fibroblast growth factor (FGF-2; 20 ng/mL; catalog #233-FB-010 285 Tocris) for the first 24 hours to allow for embryoid body (EB) formation. The following day, ROCK 286 inhibitor was removed, and fresh mTESR1 was added to the cultures. Forty-eight hours after EB 287 aggregation, cells were switched to MN Differentiation Medium: Advanced DMEM/F-12 (catalog 288 #12634010 ThermoFisher) & NB Medium (50:50 v/v), 1x N-2 supplement, 1x B27+Insulin, 1x 289 GlutaMAX (catalog #A1286001 ThermoFisher), 1x pen/strep, and 0.1 mM 2-mercaptoethanol 290 (catalog #31350010 ThermoFisher). To specify neural patterning, dual SMAD inhibition was used 291 with small molecules SB431542 (10 µM) and LDN193189 (100 nM) from day 0 to day 6 of 292 differentiation. From day 0 to day 4 the glycogen synthase kinase 3 inhibitor, CHIR99021 (3 µM; 293 catalog #4423 Tocris) was added to increase the population of Olig2 positive MN progenitors. 294 Beginning on day 2, MN specification was induced with 1 µM All-trans Retinoic Acid ( #18047019 ThermoFisher), and plated onto poly-L-ornithine (25 µg/mL; catalog #P2533-10MG 302 Sigma-Aldrich), Fibronectin (10 µg/mL; catalog #11051407001 Sigma-Aldrich), and Laminin (10 303 µg/mL)-coated plates in medium supplemented with BDNF and GDNF (10 ng/mL each). See SI 304 Table 5 for detailed protocol steps and refer to SI Figure 2 for differentiation protocol outline and 305 representative images of MN cultures. 306 Human iPSC-derived astrocyte differentiation and maintenance. Human NPCs were expanded as 307 progenitors and then seeded at ~20% confluence onto Matrigel-coated tissue culture plates. 308 Commercial astrocyte medium (catalog #1801 Sciencell) was used to differentiate NPCs as 309 previously described [47,48]. Briefly, starting with ~40% confluent NPC cultures, astrocyte 310 medium was changed every 3-4 days and cells were passaged at 1:10 ratio when they reached 311 ~80-90% confluence for the first 28 days. Immature astrocytes generated by this method express 312 multiple canonical glial markers and were further matured via small molecules or exposure to 313 FBS/Matrigel into GFAP+ cells [49]. See SI Figure 3 and SI Table 6 for detailed protocol and 314 representative images. 315 Design of imaging strands for PRISM antibody conjugation. We designed new DNA imaging 316 strands and used previously published LNA sequences to generate stable PRISM pairs by varying 317 their length and altering the GC content of the oligos (SI Table 7). 318 Automated pipeline for antibody staining, imaging, and characterization of cortical cultures. To 319 validate PRISM antibodies for automation suitability, we performed IF imaging on two high content 320 plate confocal microscopes, the PerkinElmer Opera Phenix and the Molecular Devices 321 ImageXpress. Once antibody signal was confirmed, IF staining of cortical cultures (SI Figure 10) 322 was adapted to 96-well plates and was partially automated using a BRAVO liquid handler (Agilent); 323 dispensing 100 µl of either buffer, antibody, or PRISM reagent per well (See SI Table 3 for details 324 on buffer compositions and dilutions). The system was programmed to perform an initial rinse with 325 Wash Buffer, then each well was aspirated and 100 µl of fresh Imaging Buffer with PRISM imaging 326 strands was added and incubated at room temperature for 10 min. Plates were then rinsed three 327 times with Imaging Buffer and imaged on a PerkinElmer Opera Phenix at 20X, with the 405 nm 328 laser for Hoechst nuclear stain (IF), 488 nm laser for Actin (IF), and 590 nm and 647 nm lasers for 329 PRISM antibodies. Exposures were either 300 ms (PRISM antibodies) or 150 ms (ICC controls). 330 Post-imaging, each well was washed three times with Wash Buffer and incubated for 5 min 331 between washes to eliminate residual PRISM imaging strand signal. Control imaging was 332 performed after the third rinse to confirm removal of imaging strands prior to subsequent imaging 333 strand addition and imaging rounds. These steps were repeated until all PRISM antibodies in the 334 panel were imaged. Further automation was achieved by incorporating robotics to move the plates 335 for confocal imaging to the PerkinElmer Opera Phenix high content platform at 60X magnification, 336 where each well was imaged in a 4x4 tile pattern, covering roughly 10% of the well area. Based 337 on cell type-specific markers that have been validated in the field (SI Table 2), we then built up a 338 querry algorithm to identify each cell subtype we were interested in within the differentiated 339 cultures. Only cells that were positive for all the required markers were included in the analysis 340 groups within each sub-type. A detailed breakdown on cellular characterization is shown in Figure  341 2a. 342 Data were analyzed using custom CellProfiler (cellprofiler.org) and FIJI (https://fiji.sc/) pipelines 343 for post-processing and image analysis of high-content PRISM data (reference code is available 344 upon request). Each image from the PRISM assay was aligned in sequential imaging rounds using 345 the nuclear (Hoechst 33342; 405 nm) and actin (Phalloidin; 488 nm) channels. The aligned signal 346 from each marker was then overlaid in a composite image that was then analyzed for protein co-347 localization within the cells, expression patterns of specific markers in the culture and levels of 348 protein expression. Only cells that were present in all image sequences of an experiment, based 349 on the overlays, were fed to the CellProfiler analysis pipeline to characterize the neural cultures. 350 Sampling data from 12 markers, i.e., 5 probe addition/imaging sequences, allowed us to generate 351 multi-stain identity of cellular populations that were designated as neurons (cortical or motor), 352 astrocytes, neural stem cells, etc., building cellular composition profiles of the tested iPSC-derived 353 cortical and motor neuron cultures. 354

DATA AVAILABILITY 356
The data that support the findings of this study are available from the corresponding author upon 357 reasonable request. The flow chart illustrates a representative readout of the breakdown of cell populations in cortical 482 cultures. (b) ICC versus PRISM images of the markers used to characterize neural cultures. The 483 Oct4A marker was imaged in the undifferentiated pluripotent state cell culture (iPCS), and the 484 PAX6 marker was imaged in day 14 neural progenitor state cell culture (NPCs) for validation 485 purposes. 486 as a 3D spheroid, then dissociated into 2D and maintained for 14 days. 498