Extraction of nuclei from archived post-mortem tissues for single-nucleus sequencing applications

Single-cell and single-nucleus sequencing techniques are a burgeoning field with various biological, biomedical, and clinical applications. Numerous high and low-throughput methods have been developed for sequencing the RNA and DNA content of single cells. However, for all these methods the key requirement is high quality input of a single-cell or single-nucleus suspension. Preparing such a suspension is the limiting step when working with fragile, archived tissues of variable quality. This hurdle can prevent such tissues from being extensively investigated with single-cell technologies. We describe a protocol for preparing single-nucleus suspensions within the span of a few hours that reliably works for multiple post-mortem and archived tissues types using standard lab equipment. Moreover, these preparations are compatible with single-nucleus RNA-seq and ATAC-seq using the 10X Genomics’ Chromium system.


Introduction
Advancements in technology have allowed researchers to preform large-scale transcriptomic studies at the level of a single-cell. Droplet-based cell isolation has become a favorite in the field for its scalability and simplicity of use with either in-house fluidic set-ups 1,2 , or commercially available equipment (10x Genomics) 3 . This technique is particularly interesting for tissues with highly heterogeneous cellular compositions like intestine 4 , lung 5 , spinal cord 6 , and brain 7,8 . There has been particular interest in deconvoluting brain architecture and function, which at its base, starts by accurately identifying all the cells types present [9][10][11] . However, truly harnessing the power of individual cellular transcriptomes comes with assessing differences between those transcriptomes in different physiological states. This is of particular value for complex diseases where multiple genes contribute with additive effects, making it difficult to identify changes in tissues homogenates 12 . The chemical dissociation of tightly interconnected brain cells and other cell-types has been found to alter transcription profiles 13,14 . Given that nuclear transcriptomes closely reflect the cell's cytosolic profile 15,16 , isolating the nuclei from brain tissue has proven to be an excellent strategy for single-cell level studies. Likewise, other tissues that have either been frozen for long term storage or that are formed by syncytium, such as in skeletal muscles 17,18 , could benefit from this approach.

Development of the Protocol
Numerous protocols for isolating nuclei from brain cells have been published; some rely on the additional purification by fluorescence assisted cell sorting (FACS) 19 , which is costly and not readily available for all researchers, while others have made adjustments to the microfluidics component 10 , which can also be a limiting to many labs. Our protocol has been developed for use with the commercially available Chromium™ Single Cell Controller. This is a highly optimized system that allows scalable single-cell capture. We have adapted our protocol to allow the system to efficiently capture nuclei from archived tissue. Our preparation produces stable and easily quantifiable nuclear suspension even when using archived tissue. We have used this approach to successfully compare the gene expression differences in the post-mortem prefrontal cortex of depressed patients who died by suicide compared to psychiatrically healthy controls 20 .
The protocol has also been successfully applied to collect single-nucleus transcriptomic data from surgical samples of glioblastoma 21 .
Overview of the procedure The experimental workflow ( Fig. 1) begins with dounce-homogenization in low concentration detergent for cellular lysis. Integral to the protocol are numerous wash steps to reduce ambient nucleic acid contamination. Determination of the nuclei numbers and concentration are assessed by a cell counter or hemocytometer. The concentration of the nuclei suspension is important to reduce aggregation, particularly in tissues that have undergone longterm storage and are thus more likely to be damaged, fragile and inclined to aggregate.

Comparison with other methods
As previously mentioned, single-nucleus RNA sequencing (snRNA-seq) protocols exist using FACs which is harder to scale, as well as droplet based protocols which use in house setups. Early protocols used relatively fresh frozen tissue, which is not available from most tissue banks where samples are likely to have undergone long term storage [9][10][11] . Moreover, when studying specific phenotypes for which it is hard to obtain tissues, it is not always possible to select for short post-mortem intervals (PMIs) and archival times. Early protocols were also limitedly applied to the high-quality tissue which may not be an option for answering certain types of research questions.
As with several more recently published potocols 22-24 , we have been able to adapt our nuclear prep to be compatible with a 10X Chromium system which is more likely to be available as a service platform. Furthermore, the wet-lab aspect of the protocol will produce nuclei suitable for multiple post-nuclei capture application such as, whole genome sequencing for the study of somatic mutations or Assay for Transposase-Accessible Chromatin (ATAC-seq) as supported by preliminary results from our lab.
Each of the more recently developed protocols have their own strengths and weaknesses and, in some cases, adaptations for specific tissue types such as macro-dissections for white matter regions 24 . The strength of our protocol is that it is effective and mostly unaffected by variations in PMI or archival times of the samples (Fig 2). Some of these protocols also incorporate ultra-centrifugation 25 , which is time consuming and requires specialized equipment.
We are able to circumvent the additional challenges that arise with archived tissue such as, the fragility of the cells and organelles upon thawing which typically results in large amounts of debris and ambient RNA than can either, interfere with droplet formation, or be integrated into droplet increasing background sequencing noise. Here, we show that brain tissue which has been stored at -80 o C for as long as 25 years, can produce high quality single-nuclei suspensions.
Directly applying either the cell preparation protocol or the demonstration protocol for nuclei developed from 10x Genomics did not produce useable results (Fig. 3) , but is effective for samples from cell culture. The modifications made here are primarily for use with postmortem tissue that has been archived for long periods of time but can also be applied to any post mortem frozen sample. Similar to previous studies 10, 11 , we did apply a few modifications to the standard bioinformatic analysis to addresses a number of issues which arise with droplet-bases single-nucleus sequencing. First, we assembled a pre-mRNA reference to account for unprocessed transcripts found in the nucleus 26 . Second, given that previous studies have consistently shown fewer identifiable transcripts in glial cells 10,11 we performed customized barcode filtering to include cells with a wider range of UMIs but removed noise. With these minor modifications to the analysis our wet lab approach produced much improved data compared to the demonstration protocol from 10X Genomics.

Experimental design
The most important factors to take into consideration while designing single-cell or singlenucleus RNA-seq experiments is the potential batch effects. Given that the Chromium system only allows for the capture of 8 samples at a time and that for many experiments that total number of samples to be analyzed may be greater than eight, it may be preferable to create a balanced experimental design if possible. This will help limit the effects of batch to batch variability. For example, if two phenotypic or treatment groups are to be compared, it would be ideal to include equal numbers of samples from each group in every batch. Moreover, other potential co-variates to take into consideration include age, PMI, and sex. It may be possible to account for the effects of these variables by matching samples by these parameters within each batch.
In cases where cell-type specific gene-expression data has been previously published, or singlecell or nucleus gene expression datasets are available, these data can be used for comparison to help determine whether that cell-types identified, and single-nucleus transcriptomic profiles detected reflect, previously published literature. In cases where such datasets are not available it may be informative to prepare bulk samples in parallel or to perform sequencing for fluorescence assisted nuclei sorting (FANS) purified populations of expected cell-types based on known markers to validate the cell-type identification from the single-nucleus transcriptomic data 27 .
High-throughput in situ hybridization (ISH) 28 and ISH based nuclei sorting 27 have also been used to confirm experimentally determined cell-types from snRNA-seq. In the case of complex tissues, it can be useful to perform careful dissection and even to cryosection the tissue before preparing nuclei to ensure that the approximate cell-type composition for each sample will be comparable 11,24 .
Another strategy which has been recently applied to increase cost-effectiveness as well as to aid in batch effect correction is combining male and female samples in a single capture followed by using the expression of sex-specific, X-chromosome genes such as XIST and Y-chromosome genes such as SRY 29,30 , or the chromosome accessibility ratios for sex-chromosome versus autosomes 31 to separate the cells from each sample. Since both samples are captured on the same lane of the microfluidic chip, it may be possible to account for lane to lane variability using this approach. Moreover, the use of cryosections of histological grade tissue blocks may be a good strategy for accounting for uniform input from a micro-anatomicallly heterogeneous region such as the cerebral cortex 11,23 .
Expertise needed to implement the protocol This protocol requires access to a 10X Genomics' Chromium system and corresponding reagents or an in-house droplet based single-nucleus sequencing system. A hemocytometer or cellcounting microscope will be required for determining proper loading concentration. Wet-lab work will require familiarity with standard molecular biology approaches such as cDNA synthesis, and sequencing library preparation.

Advantages and Limitations
We are unable to get information about cytoplasmic transcription. Some tissue types may require additional processing such as FACS. Representation of all cell types may not be uniform as different cell types are differentially susceptible to lysis during the isolation procedure.

REAGENTS:
• Tissue samples: This protocol was successfully applied for processing frozen archived postmortem prefrontal cortex tissue from the Douglas Bell Canada Brain Bank, post-mortem intestinal tissue (with modifications), and surgical samples of tumor tissue 21

EQUIPMENT SETUP
• Precool both centrifuges to 4 o C.
• Set up EVOS FL Auto microscope with 10X magnification, bright field and DAPI channels.

Computational requirements
Linux OS with minimum requirements for running CellRanger as described on the 10X Genomics' webpage must be met.

Tissue preparation (Timing 1-4 hours)
1. Cut using a scalpel and weigh 30-50mg of frozen tissue per sample. Keep tissue on dry ice while cutting to minimize degradation. Transfer to a 1.5mL microcentrifuge tube using a spatula and place back on dry ice. Clean scalpel and spatula with 70% ethanol between samples. Use a fresh weigh boat for each sample. Alternatively, this step could be replaced by cryosectioning a fresh frozen histology grade dissection of tissue and collecting several sections such that the total weight is between 30-50 mg. CAUTION: Post-mortem human tissue can contain pathogens. Take precautions including wearing PPE and seek medical attention if the scalpel breaks your skin.

Nuclei extraction (Timing 2-3 hours)
2. Transfer tissue using spatula to douncing tube on ice. Add 3mL of ice-cold lysis buffer and dounce with loose pestle 10 times and 5 more times with the tight pestle. CRITICAL: Use proper douncing technique to ensure proper mechanical breakdown of tissue. Proceed slowly and avoid bubbles. Grind tissue against the bottom of the tube using the douncer with each stroke. 3. Transfer homogenized tissue to a 15mL centrifuge tube by pouring and add 2 mL of chilled lysis buffer. Incubate on ice for 5 minutes gently swirling to mix 2 times during incubation. 4. Add 5 ml of chilled wash buffer to lysed tissue to quench lysis. Swirl to mix. 5. Place 30 μm MACs SmartStrainer on a 15mL centrifuge tube. Pipette lysed tissue suspension on top of filter to remove cell debris and large clumps. In case of blocked flow through the filter, tap filter gently to encourage the suspension to flow through. 6. In the precooled Allegra-14X centrifuge, spin down the lysed tissue suspensions at 500g for 5 minutes at 4°C. 7. Decant supernatant into a waste beaker without disrupting the nuclei pellet.
CAUTION: The supernatant should be treated as biohazardous waste and treated with bleach before disposal. CRITICAL: Pour out supernatant in a single motion as repeated pouring motions can dislodge the pellet. If the pellet dislodges during decanting slowly remove the supernatant using a pipette. 8. Using a 10 mL serological pipette add 10mL of nuclei wash buffer to the pelleted nuclei and gently pipet 8-10 times to mix. 9. Repeat step 5-7 using the resuspended nuclei. 10. Using a 10 mL serological pipette add 5mL of nuclei wash buffer to the pelleted nuclei and gently pipet 8-10 times to mix. 11. Repeat steps 6-7 using the resuspended nuclei. 12. Using a 1000 μL pipette tip, add 1mL of nuclei wash buffer to pelleted nuclei and gently pipet 8-10 times to mix. 13. Add 1 mL of 50% working solution of Optiprep™ and mix well to make 2 mL of 25% iodixanol solution containing nuclei. 14. Prepare Optiprep™ cushion of 2ml of 29% Optiprep™ solution in a 5ml Eppendorf centrifuge tube. 15. Gently add the 2ml nuclei suspension on top of the Optiprep™ cushion by pipetting slowly against the wall of the tube to avoid mixing. 16. In the precooled Eppendorf centrifuge, spin the tubes containing nuclei layered over Optiprep™ cushion at 10,000g for 30 minutes at 4°C. 17. Carefully pour out the supernatant leaving the least amount of volume in the tube without disrupting the pellet. ?TROUBLESHOOTING 18. Using a 1000 μL pipette tip, resuspend the nuclei pellets in 500 μL or less of nuclei wash buffer. Gently pipette 8-10 times or until nuclei are resuspended. 19. For a quick estimate of nuclei concentration, mix 10 μL of the nuclei suspension with 10 μL of Trypan blue in a separate tube. Load 10 μL of the mixture onto a Countess hemocytometer slide. Count nuclei on the Countess hemocytometer and measure range of sizes. For human neurons the average diameter of the nuclei should be around 10 μm. Trypan blue is a live dead stain and properly isolated nuclei should have be marked as dead cells. 20. Using the estimated count from the Countess dilute cells to around 500,000 cells/ mL or 500 cells/ μL by adding an appropriate volume of nuclei wash buffer. It may be possible to increase these concentrations for capturing more nuclei. CRITICAL: If the concentration of nuclei is too high it can result in aggregation which will prevent efficient capture of single nuclei in subsequent steps. 21. Add Hoechst stain to the resuspended nuclei at a 1:2000 dilution.

Nuclei concentration assessment and capture (Timing 45 mins)
Immediately prior to loading the Hoechst stained nuclei on the Chromium system, check to make sure the nuclei are well segregated (nuclei may clump) and recheck sample concentration using a fluorescent microscope like the Evos FL Auto (Thermofisher) 22. OPTIONAL: Use a 1mL pipette to take a minimum of 200 μL of the sample and filter it through a 40μm Flowmi pipette tip filter before counting and loading. This will get rid of the clumped nuclei and large debris that can clog the microfluidics of Chromium chips. 23. Load 10μL of sample onto a disposable hemocytometer slide and into the Evos.
CRITICAL: Make sure to pipette the full volume of the nuclei suspension several times before loading for accurate counts. 24. Set the Evos to 10X magnification and image the entire hemocytometer grid field of view in DAPI and bright field. This image makes counting nuclei easier and serves as a record as well (Fig. 3g). 25. In parallel to steps 22-24, make the RT Master Mix and aliquot appropriate volumes of master mix and water into PCR tubes according to the number of nuclei to be targeted for capture, referring to the Chromium protocol CG00052 Rev. D or later.

?TROUBLESHOOTING
CRITICAL: The Chromium capture rate for nuclei from archived tissue is lower (~20% less) than the capture rate for cells. To account for this, it is necessary to adjust the count used to determine loading volume. We empirically determined that choosing the loading volume by using a count that is 30% less than observed count worked best for our samples. This adjustment may vary from tissue type to tissue type. For example, if the sample has a concentration of 500 nuclei/μL, the sample volume should be loaded as if it has 350 cells/μL (70% of 500 cells) in order to get the targeted nuclei to recover.
CRITICAL: Resuspend nuclei by pipette mixing the full volume several times immediately before loading to prevent aggregation of nuclei.  36 . Unfiltered gene barcode matrices for each sample were loaded into R using the Read10X function. At this step, cell names can be modified such that the subject name, batch, and biological condition are appended to them. Seurat objects were created corresponding to each sample using the CreateSeuratObject function with the imported unfiltered gene-barcode matrices provided as the raw data. Individual Seurat objects for each sample were combined sequentially using the MergeSeurat function. No filtering or normalization was performed up to this step. Since we were working with a single nucleus dataset, all mitochondrial genes that are transcribed from the mitochondrial genome were removed, along with genes not detected in any cell. More recently several methods have been developed to align multiple datasets of snRNAseq and other single-cell level data 28,37 which can be used for combining the data from individual subjects if inter-individual variability or batch effects are deemed to have a large influence on the results.
For preliminary filtering, some nuclei with very low number of genes detected (>110) and nuclei with very high numbers of UMIs detected (in the top 0.5%) were removed as low-quality nuclei and potential multiplets respectively. These cut-offs are arbitrary but can be based upon the distribution of the data. For example, in our dataset there was a sharp increase in the number of UMIs from 16,393 at the 99.5 th percentile to 102,583 at the maximum which probably represents the multiplets in the dataset.
If the dataset contains multiple cell-types which are expected to be heterogenous in terms of the

?TROUBLESHOOTING
Troubleshooting advice can be found in Table 1. The capture rate for nuclei may not be the same as that for cells.
Empirically determine the difference between the capture rate expected and observed and adjust loading volume accordingly.
TIMING number of molecules of RNA present per nucleus (such as when the nuclei of different cell-types are known to be of very different size), the following approach can be used for removing low quality cells without unduly biasing the filtering against nuclei which biologically contain fewer molecules . For our dataset, given the known trend for higher number of RNA molecules in neuronal nuclei compared to glial nulcei 10,11,38 , the distribution of number of UMIs was fit with three normal distributions using the normalmixEM function from the mixtools 39 package. The rationale is that the filtered barcodes contain a population of low quality "noise" barcodes that have a very low numbers of UMIs on average, a population of non-neuronal cells that have an intermediate numbers of UMIs and a population of neuronal cells that have a high numbers of UMIs. After fitting the normal distributions, only the barcodes with a high probability (> 0.95) of belonging to either the putative "non-neuronal" or putative "neuronal" distributions, and a low probability (<0.05) of belonging to the "noise" distribution were retained for further analysis. As an example, in our publication, for a subset of 20 subjects, applying our custom filtering approximately doubled the total number of cells but increased the number of non-neuronal cells by almost 6 times 20 .
Step 1, 1 hour for 8 samples if cutting pieces using a scalpel, up to 4 hours if collecting cryosections Step 2-21, 2-3 hours Step 22-26, 45 minutes Step 27, can be split into two 4 hours blocks on 2 days

Anticipated Results
We expect our single-nuclei extraction protocol to produce high quality single-nucleus suspensions (Fig. 3) from frozen archived post-mortem tissues. The nuclear suspensions are relatively free from debris and do not show substantial aggregation of nuclei even after 16 hours of refrigeration, upon visual inspection (Fig. 3-f). Capture of single-nuclei using these nuclear suspensions on a microfluidic device for droplet-based snRNA-seq or snATAC-seq reproducibly produces high-quality libraries with sufficient cDNA yield for sequencing (Fig. 4). Moreover, the variability in sample parameters such as PMI, archival time, pH and RIN did not affect most of the quality metrics of snRNA-seq results with this nuclei extraction protocol (Fig. 2).