Abstract

   Circulating tumor cells (CTCs) break free from primary tumors and
   travel through the circulation system to seed metastatic tumors, which
   are the major cause of death from cancer. The identification of the
   major genetic factors that enhance production and persistence of CTCs
   in the bloodstream at a whole genome level would enable more
   comprehensive molecular mechanisms of metastasis to be elucidated and
   the identification of novel therapeutic targets, but this remains a
   challenging task due to the heterogeneity and extreme rarity of CTCs.
   Here, we describe an in vivo genome-wide CRISPR knockout screen using
   CTCs directly isolated from a mouse xenograft. This screen elucidated
   SLIT2—a gene encoding a secreted protein acting as a cellular migration
   cue—as the most significantly represented gene knockout in the CTC
   population. SLIT2 knockout cells are highly metastatic with
   hypermigratory and mesenchymal phenotype, resulting in enhanced cancer
   progression in xenograft models.
     __________________________________________________________________

   An in vivo genome-wide CRISPR knockout screen identifies SLIT2 as a
   regulator of circulating tumor cell production.

INTRODUCTION

   Metastasis accounts for more than 90% of deaths from cancer yet remains
   poorly understood and largely incurable ([46]1–[47]3). Despite advances
   in cancer treatment that substantially reduce morbidity and mortality
   of various cancer types, metastasis still leads to poor prognosis of
   patients with cancer and treating metastasis remains a tremendous
   challenge ([48]4, [49]5). To develop more effective therapies and
   improve patient outcomes, further progress elucidating the fundamental
   biology of metastasis is critical to reveal mechanisms at the molecular
   level and provide insights regarding new therapeutic targets. Comparing
   the genetic background between primary tumors and metastatic lesions
   using whole-exome sequencing (WES) and whole-genome analyses has
   facilitated the identification of genetic factors that drive tumor
   progression and dissemination ([50]6, [51]7). However, deciphering
   drivers of metastasis solely on the basis of the genetic information
   from solid tumors is limiting because of genetic divergence and tumor
   heterogeneity ([52]7–[53]9).

   As circulating tumor cells (CTCs) leave primary tumors and seed
   metastatic lesions, molecular characterization of CTCs is critical to
   facilitate comprehensive understanding of the metastatic processes.
   Progress in the area of CTC enrichment and single-cell sequencing
   technologies has enabled identification of CTC-specific mutations in
   patients with cancer at the genomic level ([54]10–[55]12). However,
   efficient capture of CTCs and their unbiased genomic amplification for
   WES are still challenging because of the rarity and fragility of CTCs
   ([56]11–[57]16). In addition, particular mutations observed in
   individual patient CTCs could be anecdotal, with low applicability and
   weak prognostic potential for large patient populations. Thus, new
   technologies and platforms are needed to comprehensively and
   systematically study the genetic factors underlying the CTC phenotypes
   that contribute to the metastatic process.

   Here, we report a genome-wide CRISPR knockout (KO) screen designed to
   identify genes contributing to CTC dissemination in vivo. Xenografted
   tumors were seeded with pools of CRISPR-edited cells with each cell
   having loss of function of one gene in the human genome. Using a
   high-performance approach for CTC capture directly from blood coupled
   with next-generation sequencing (NGS) of barcoded single-guided RNAs
   (sgRNAs), we identified gene KOs that promoted CTC abundance. This
   approach allowed the systematic identification of possible
   metastasis-promoting genetic factors at the CTC level with minimal
   bias. Previous genome-wide in vivo CRISPR screens identifying
   metastatic factors primarily used solid tumors or CTCs cultured ex vivo
   ([58]17, [59]18). Our approach, however, allowed genome-wide in vivo
   CRISPR screening to be performed on fresh CTCs directly collected from
   liquid biopsy without extensive culturing. Avoiding ex vivo culturing
   of CTCs is important to prevent possible changes in morphology and
   proliferation of CTC subpopulations ([60]19, [61]20), which could
   affect representations of subsets of sgRNAs in the library. Our in vivo
   CTC CRISPR KO screen elucidated Slit Guidance Ligand 2 (SLIT2) as a
   secreted factor that promoted the epithelial-to-mesenchymal transition
   (EMT) of tumor cells, resulting in enhanced intravasation. SLIT2 KO
   cells had increased expression of complex I and were hypersensitive to
   the treatment of complex I inhibitor rotenone. Loss of SLIT2 increased
   CTC numbers and the extent of metastasis of prostate cancer in an
   animal model.

RESULTS

Genome-wide in vivo CRISPR-Cas9 KO screen on CTCs

   Genome-wide CRISPR screens are powerful tools for the systematic and
   unbiased identification of genes involved in metastasis ([62]17,
   [63]18, [64]21). Combining a CRISPR-Cas9 genome-wide editing system
   (TKOv3: Toronto KnockOut version 3 CRISPR library) ([65]22) and a
   high-throughput immunomagnetic microfluidic device for rare cell
   enrichment ([66]23, [67]24), we systematically investigated genes
   promoting CTCs and tumor progression ([68]Fig. 1A and fig. S1A). The in
   vivo genome-wide loss-of-function CRISPR-Cas9 screen was conducted
   using the prostate cancer cell line PC-3M as a model of tumor
   progression. After subcutaneous injection of the TKOv3-edited cancer
   cells (n = 3.0 × 10^7 cells injected per animal) into the
   immunodeficient mice (n = 6 mice), tumor growth from the
   TKOv3-transduced PC-3M cells was monitored and compared to the tumor
   size resulted from the parental PC-3M cells (n = 3 mice, 3.0 × 10^7
   cells injected per animal) for 3 weeks. No significant differences in
   the size of the primary tumors between these two groups were observed
   ([69]Fig. 1B). At a predefined end point of 3 weeks, whole blood was
   collected from each animal to isolate CTCs for sgRNA enrichment
   analysis. Efficient isolation of CTCs from whole blood was accomplished
   by the immunomagnetic labeling of the epithelial marker epithelial cell
   adhesion molecule (EpCAM) on the cell surface and magnetic deflection
   of CTCs away from blood cells using metglas tracks embedded in a
   microfluidic chip (fig. S1, A and B) ([70]24). CTCs collected were
   lysed ([71]25) to obtain genomic DNA for the amplification of sgRNA
   regions using polymerase chain reaction (PCR) and the determination of
   each sgRNA abundance using NGS analysis (fig. S1C). CTC sgRNA counts
   were successfully obtained for all the six mice tested using our in
   vivo CTC CRISPR screening pipeline, with consistent total sgRNAs counts
   across replicates as detailed below.

Fig. 1. An in vivo genome-wide CRISPR-Cas9 KO screen identifying
CTC-promoting genetic factors.

   [72]Fig. 1.
   [73]Open in a new tab

   (A) Experimental design of the genome-wide CRISPR-Cas9 KO screen
   focused on CTCs. (B) Primary tumor growth curves of the
   immunocompromised mice subcutaneously transplanted with either the
   TKOv3-transduced PC-3M cells or the control cells (n = 3 for each
   group). Error bars indicate SD. WT, wild type. (C) Box plot of the
   sgRNA counts from the TKOv3-transduced cell pool before transplantation
   (cells), primary tumors, and CTCs. (D) Cumulative probability
   distribution of sgRNA counts from the TKOv3-transduced cell pool before
   transplantation (cells, n = 2), primary tumors (n = 6), and CTCs (n =
   6). (E) Pearson correlation coefficient of the sgRNA counts from the
   TKOv3-transduced cell pool before transplantation (Tn), primary tumors
   (Pn), and CTCs (Cn).

   For sgRNA enrichment analysis, we first analyzed the TKOv3 library
   attrition across the initial CRISPR-edited PC-3M cells, the primary
   tumors, and the isolated CTCs. Compared to the sgRNA counts in the
   initial TKOv3-transduced PC-3M cell pool before injection, most of the
   sgRNAs from the library had significantly reduced or no representation
   in the isolated CTC samples ([74]Fig. 1, C and D). Given that 30
   million cells (each contained 1 sgRNA) were xenografted and a library
   size of 71,090 sgRNAs, an initial 400× representation of the TKOv3
   library was obtained. With a median capture of 100 CTCs per animal, a
   drastic library attrition at the CTC level was expected. Notably, the
   reduction of the sgRNA representation was much less in the primary
   tumor samples because substantially more cells and input materials were
   available for NGS ([75]Fig. 1, C and D, and fig. S2A). In addition, the
   overall distribution of sgRNAs in the CTC samples clustered tightly
   with each other and formed a clade distinct from the sgRNAs
   representing the CRISPR-edited PC-3M cells and primary tumor samples
   ([76]Fig. 1E). When individual sgRNA enrichment was examined for the
   primary tumors, a clear trend emerged with the top-ranked genes
   representing well-established tumor suppressor genes (TSGs) such as
   TSC1 (table S1) ([77]26, [78]27). Frequent loss-of-function mutations
   of the top gene hits were also observed in patient tumor data collected
   from the Catalogue Of Somatic Mutations In Cancer (COSMIC) with copy
   number variation analysis (fig. S3) ([79]17, [80]28, [81]29).

Selection of top enriched sgRNAs in CTCs

   To analyze the CTC-enriched sgRNAs, we developed two selection criteria
   for robust candidate hit identification ([82]Fig. 2, A and B). In the
   first selection criterion, we took into account the expected number of
   CTCs captured at the time of sample harvesting (median CTC count 100
   per mouse). As a result, only the top 100 sgRNA counts from each of the
   six replicates were pooled together (600 sgRNAs × 6 replicates) to
   generate a list of 162 unique sgRNAs ([83]Fig. 2A). The number of
   sgRNAs selected using this criterion was 162 instead of 600 due to the
   exclusion of the repetitive sgRNAs among the six replicates. The second
   criterion prioritized sgRNAs that were consistently enriched in CTC
   samples harvested across all the six mice injected compared to their
   tumor sample counterpart. sgRNAs that were enriched in five or less
   replicates were excluded from the selection. This resulted in the
   identification of 149 unique sgRNAs that were enriched across all the
   CTC samples compared to their corresponding primary tumors ([84]Fig.
   2B). After filtering for sgRNAs that satisfied both selection criteria
   (overlapping sgRNA from the two selection criteria), the final CTC
   sgRNA enrichment list consisted of 38 unique sgRNAs, and the genes
   targeted by these sgRNAs were selected for downstream analysis
   ([85]Fig. 2, C and D). The biological pathways represented by these 38
   genes were found to be highly related to metastatic processes such as
   the regulation of apoptosis and cell proliferation ([86]Fig. 2E).
   Because no more than one sgRNA was enriched for each of the 38 genes
   selected from the genome-wide screen, we designed a subpool CRISPR KO
   library to functionally validate the top hits from the initial
   genome-wide screen, and a true top hit must have multiple sgRNAs
   present in the top enriched sgRNA list generated from the subpool
   screen.

Fig. 2. Identification of top hits from the genome-wide CRISPR-Cas9 KO
screen.

   [87]Fig. 2.
   [88]Open in a new tab

   (A) Violin plots of the top 100 sgRNA counts from the six CTC samples
   (C1 to C6). The dashed boxes highlight the top 100 sgRNA counts from
   each CTC sample. (B) Frequency (percentage of mice) of each sgRNA being
   enriched in CTCs compared to the corresponding primary tumors. The
   dashed box highlights the 149 sgRNAs enriched in CTCs across all six
   mice. (C) Venn diagram showing selection of the top-enriched sgRNAs
   from the CTC samples. Thirty-eight sgRNAs satisfied both the “counts”
   criterion in (A) and the “replicates” criterion in (B), and their
   targeted genes were selected for downstream analysis. (D) List of the
   38 genes selected for the construction of the sublibrary. (E)
   Biological pathways being enriched from the 38 genes selected. Analysis
   and image preparation were conducted by Metascape.

Functional validation of top hits with sublibrary screen

   A customized subpool CRISPR KO library consisting of 400 sgRNAs
   specifically targeting the 38 genes of interest (10 sgRNAs per gene)
   plus 20 nontargeted control (NTC) sgRNAs was generated. CRISPR-edited
   PC-3M cells using the subpool library were xenografted using an
   analogous protocol to the original screen, including the number of mice
   used (n = 6 mice) and the number of cells injected (n = 3.0 × 10^7
   cells injected per animal) (fig. S1D). CTC capture was improved by
   viral transducing the Glycosylphosphatidylinositol (GPI)–anchored
   Myc-tag into PC-3M cells before the subpool screen, allowing Myc-based
   immunomagnetic capture of CTCs (fig. S1B). As observed previously,
   compared to the other sample groups (plasmid pool, infected cells
   before transplant, and primary tumors), the number of unique sgRNAs
   with significant representation in the total sgRNA reads was markedly
   reduced for the CTCs ([89]Fig. 3A and fig. S2B). Notably, reads for the
   sgRNAs targeting SLIT2 accounted for 10% of the total sgRNA counts in
   the CTC samples, with 4 of the 10 sgRNAs targeting SLIT2 highly
   represented in the CTC population ([90]Fig. 3B). Accordingly, SLIT2 KO
   was ranked to be the most significant loss-of-function mutation that
   appeared to increase the number of CTCs, with a much higher ranking
   score than the rest of the genes tested in the sublibrary ([91]Fig.
   3C). SLIT2 was thus selected for downstream analysis.

Fig. 3. Loss of SLIT2 was a key factor of CTC production in the subpool
screen and was associated with poor prognosis of patients with prostate
cancer.

   [92]Fig. 3.
   [93]Open in a new tab

   (A) Bar graph of the number of significant sgRNAs (counts > 0.1% of
   total reads) from the plasmid library (plasmid), the initial cell pool
   before transplant (Ln), primary tumors (Pn), and CTCs (Cn) of the
   sublibrary CRISPR KO screen. (B) Pie chart showing gene enrichment
   (percentage of total reads) of the initial cell pool, primary tumor,
   and CTC samples from the subpool CRISPR KO screen. (C) Gene ranking
   from the subpool CRISPR KO screen. SLIT2 KO was found to be the most
   abundant gene KO present in CTCs. (D) SLIT2 expression levels in tumor
   and normal tissues of patients with prostate adenocarcinoma (PRAD) were
   compared using RNA sequencing data collected from TCGA and prepared by
   GEPIA2 (tumor, n = 492; normal, n = 152). P value cutoff was set to be
   0.01, *P < 0.01. (E) Survival plot of patients with PRAD with low or
   high expression of SLIT2. Data were collected from TCGA, with an
   expression cutoff at 2.45 FPKM (high expression, n = 189; low
   expression, n = 305).

Loss of SLIT2 is associated with poor prognosis of patients with prostate
cancer and induced metastasis in vivo

   SLIT2 belongs to a family of secreted proteins with a known role in
   axon guidance through interaction with the Roundabout family of
   receptors ([94]30, [95]31). SLIT2 was also found to be expressed in
   various cancer tissues with context-dependent promotion or inhibition
   of cancer progression ([96]32–[97]34). However, no previous studies had
   focused on the expression of SLIT2 in CTCs and the effect of this
   genetic factor on generation of CTCs, especially in a context of
   prostate cancer. In the light of our CTC CRISPR KO screen conducted
   with the prostate cancer cell line PC-3M, we hypothesized that SLIT2
   negatively regulates CTC production and thereby prevents metastatsis.

   On the basis of patient data collected from The Cancer Genome Atlas
   (TCGA), SLIT2 expression level is significantly reduced in tumor
   tissues compared to normal prostate tissues ([98]Fig. 3D). Low SLIT2
   expression is also associated with poor prognosis and overall survival
   of patients with prostate cancer ([99]Fig. 3E). In addition, reduced
   expression of SLIT2 in tumor tissues compared to the corresponding
   normal tissues is observed in 26 of the 31 human cancers recorded in
   TCGA (fig. S4), indicating the generality of the negative correlation
   between SLIT2 expression and development of tumors.

   To investigate the role of SLIT2 in a prostate cancer model, we
   generated SLIT2 KO PC-3M cells using CRISPR editing to probe the
   effects of losing this gene in vivo (fig. S5A). Equal numbers of NTC or
   the SLIT2 KO PC-3M cells (1 × 10^6) were orthotopically injected into
   the prostate of immunodeficient mice. At the end point of tumor growth,
   the number of CTCs present in the whole blood of the mice injected with
   SLIT2 KO cells was increased up to fivefold compared to the group
   injected with the NTC cells ([100]Fig. 4A and fig. S5, D and E). The
   xenograft generated from SLIT2 KO cells also exhibited enhanced
   proliferation of cancer cells, resulting in significantly increased
   tumor size compared to the control group (fig. S5, B and C). To assess
   whether the enhanced CTC production in SLIT2 KO model was solely
   because of elevated tumor size, SLIT2 KO xenografts with comparable
   tumor sizes to the control were generated with a decreased number of
   SLIT2 KO cells at injection ([101]Fig. 4B). In this case, SLIT2 KO
   again exhibited a significantly increased number of CTCs in whole blood
   compared to the control group ([102]Fig. 4B), indicating enhanced
   intravasation and dissemination of cells lacking SLIT2. Because less
   cells were used for injection and blood samples were harvested earlier
   in this case to ensure comparable tumor sizes between control and SLIT2
   KO cells, overall CTC counts were lower compared to the previous
   injection. In addition, severe infiltration of neoplastic cells was
   observed inside the lymph nodes of the SLIT2 KO models, with cases of
   subcapsular and intravascular invasion found in multiple hematoxylin
   and eosin (H&E)–stained histology sections ([103]Fig. 4, C and D, and
   fig. S5F). No invasion or peripheral infiltration of neoplastic cells
   was observed for the prostate cancer model generated from the control
   cells at the time of harvest. Lymph node was used as a representation
   of metastasis model based on our previous observation that PC-3M had
   more chances to develop metastasis at the lymph nodes at late
   metastatic stage compared to other prostate cancer cell lines such as
   LNCaP ([104]35). We concluded that KO of SLIT2 promotes CTC numbers and
   thereby increases metastatic burden in a prostate cancer model.

Fig. 4. SLIT2 KO promoted cancer progression in vivo.

   [105]Fig. 4.
   [106]Open in a new tab

   (A) Bar graph of the number of CTCs counted from the whole blood of the
   immunocompromised mice after transplant of NTC, genome-wide CRISPR KO
   library (TKOv3), TSG-enriched CRISPR KO sublibrary, or SLIT2 KO cells
   (n = 3 for each group). Error bars indicate SD. P value was calculated
   by two-tailed unpaired t test. *P = 0.045. (B) Bar graph comparing CTC
   number resulted from similar tumor size of NTC and SLIT2 KO cells (n =
   3 for each group). KO group had one additional sample. Error bars
   indicate SD. P value was calculated by two-tailed unpaired t test. **P
   = 0.0072. (C) Bar graph of the number of lymph nodes with normal,
   neoplastic cell–infiltrated, and severely infiltrated tissues.
   Histology slides were prepared from lymph nodes collected from the
   immunocompromised mice after orthotopic injection of either NTC or the
   two SLIT2 KO cell lines. (D) Representative images of the H&E-stained
   sections showing normal lymph node tissues (upper left) (Ctrl), as well
   as severe (upper right), subcapsular (bottom left), and intravascular
   (bottom right) neoplastic cell infiltration (SLIT2 KO).

Loss of SLIT2 results in enhanced migratory and invasive behavior of prostate
cancer cells

   To understand mechanisms by which SLIT2 modulates the metastatic
   process, we studied SLIT2 KO and NTC cells using a variety of
   phenotypic assays to assess cell proliferation, deformability, and
   invasion. Three-dimensional (3D) spheroids cultured from the SLIT2 KO
   and the NTC cells were comparable in size and composed of similar
   number of cells, but the SLIT2 KO spheroids appeared to be poorly
   organized with cells on the outer surface of the spheroids being
   loosely attached ([107]Fig. 5A and fig. S6). To evaluate the impact of
   SLIT2 loss of function on the migratory potential of PC-3M cells, we
   designed a specialized microfluidic device with migration channels
   connecting the cell-loading and the nutrient-holding pools (fig. S7).
   The channels had a constant height of 5 μm but decreasing width,
   thereby requiring the cells to undergo remodeling to squeeze through
   narrowing channels toward the nutrients (fig. S7). While a comparable
   number of SLIT2 KO and NTC cells traveled through the migration
   channels wider than 10 μm, significantly more SLIT2 KO cells were able
   to fit into the channels of width 8 μm or less ([108]Fig. 5B and fig.
   S7B).

Fig. 5. SLIT2 KO facilitated metastasis with activation of EMT.

   [109]Fig. 5.
   [110]Open in a new tab

   (A) Representative images of the spheroids formed from NTC or SLIT2 KO
   PC-3 M cells. The scale bar represents length of 100 μm. (B) Bar graph
   comparing the number of deformed NTC versus SLIT2 KO cells at different
   migration channel widths (n = 3 for each group). (C) Box plot comparing
   the number of NTC versus SLIT2 KO cells invaded across the Transwell
   membrane with or without the presence of the chemoattractant or N-SLIT2
   protein in the culture media (n = 8 for each group). (D) Proteins
   ranked on the basis of levels of differential expression in SLIT2 KO
   compared to NTC cells as detected by the oncology antibody array. (E)
   Immunoblots detecting E-cadherin, N-cadherin, Snail, and Vimentin
   expression in SLIT2 KO and NTC cells. GAPDH, glyceraldehyde-3-phosphate
   dehydrogenase. (F) Immunoblots detecting Vimentin protein levels in
   control and SLIT2 KO PC-3M, as well as MDA-231 cells, either with or
   without supplementation of N-SLIT2 protein in culture medium. Error
   bars indicate SD. P values were calculated by two-way analysis of
   variance (ANOVA), except for (C) which was by two-tailed unpaired t
   test. ***P < 0.001; **P < 0.01; *P < 0.05.

   The invasive behavior of SLIT2 KO and control cells was also tested
   using a Transwell cell invasion assay. A larger number of SLIT2 KO
   cells migrated from the top to the bottom of the membrane compared to
   the control group, with or without the presence of the chemoattractant
   in the lower chamber ([111]Fig. 5C). The number of invading SLIT2 KO
   cells was similar to that of control cells when the media was
   supplemented with exogenous SLIT2 protein ([112]Fig. 5C). Results of
   cell deformability and invasion assays were independent of cell
   proliferation rates, as no significant difference between the cell
   numbers of control and SLIT2 KO was found after 2D culturing for 3 days
   (fig. S8).

Loss of SLIT2 results in activation of EMT

   The interrogation of a panel of cancer-related proteins using an
   oncology antibody array revealed several strong hits that connected the
   migratory and invasive phenotypes observed in SLIT2 KO cells to the EMT
   ([113]Fig. 5D and fig. S9). Considering the most elevated protein
   family or biological pathways identified from the array, protein levels
   of both mesenchymal markers Vimentin and Snail were enhanced in SLIT2
   KO cells. Thus, we hypothesized that the KO produced a more mesenchymal
   phenotype. Reduced expression of the epithelial marker (E-cadherin) and
   increased expression of mesenchymal markers (Vimentin, N-cadherin, and
   Snail) were observed for SLIT2 KO cells compared to NTC cells,
   indicating that EMT may be linked to the metastatic phenotype of cells
   lacking SLIT2 ([114]Fig. 5E and fig. S10). Expression level of the
   epithelial marker EpCAM on the cell surface was also checked as it was
   used for the CTC capture during the screening process, and no
   significant difference was found comparing control and SLIT2 KO cells
   (fig. S11).

   Reversal of the mesenchymal phenotype was explored in a rescue
   experiment where exogenous SLIT2 was added to culture media, and the
   level of Vimentin was analyzed. Addition of SLIT2 did not have a
   significant effect on Vimentin expression in control PC-3M cells
   ([115]Fig. 5F and fig. S12). However, the level of Vimentin in SLIT2 KO
   PC-3M cells, which was originally doubled compared to that of the
   control cells, was significantly reduced with the addition of exogenous
   SLIT2 ([116]Fig. 5F and fig. S12). In addition, we also tested Vimentin
   levels for a metastatic breast cancer cell line (MDA-MB-231), and
   significant decrease of Vimentin expression was observed with the
   presence of SLIT2 ([117]Fig. 5F and fig. S12), extending the link
   between SLIT2 and EMT in a different context.

Loss of SLIT2 resulted in elevated complex I expression with hypersensitivity
to rotenone

   To elucidate additional drivers of the invasive phenotype of SLIT2 KO
   cells, we also performed gene expression profiling of NTC and SLIT2 KO
   cells ([118]Fig. 6, A and B). Notably, SLIT2 KO cells exhibited
   increased expression of one component of the ATP synthase (ATP6) and
   multiple subunits of the complex I (MT-ND1, MT-ND2, and MT-ND5)
   belonging to the mitochondrial electron transport chain (ETC), making
   metabolic process and oxidative phosphorylation the most enriched gene
   ontology term ([119]Fig. 6B and fig. S13). Elevated expression of ETC
   components suggested a possible dependency of SLIT2 KO cells on
   mitochondrial activity and oxidative metabolism. Confirming this
   prediction, SLIT2 KO cells were significantly more sensitive to
   treatment with the complex I inhibitor rotenone when compared to the
   control cells ([120]Fig. 6C). To test whether the elevated activity of
   complex I contributed to the increased migratory potential of the PC-3M
   cells lacking SLIT2, we tested the ability of rotenone to block
   migration of SLIT2 KO and control cells using the microfluidic device
   described above (fig. S7). Upon treatment of rotenone, significant
   reduction of the number of deformed cells was observed in both groups
   at wider channel width ([121]Fig. 6D). However, for the narrower
   channels, rotenone was more effective in inhibiting the migration of
   the SLIT2 KO cells than on the NTC cells ([122]Fig. 6D). Other two
   prostate cancer cell lines LNCaP and PC3 also showed enhanced
   sensitivity to rotenone with SLIT2 being knocked out (figs. S14 and
   S15). Treatment of rotenone significantly reduced the enhanced
   invasiveness of LNCaP and PC3 with SLIT2 KO (fig. S16).

Fig. 6. SLIT2 KO resulted in elevated complex I expression and
hypersensitivity to rotenone.

   [123]Fig. 6.
   [124]Open in a new tab

   (A) Heatmap of the top 1000 differentially expressed genes in SLIT2 KO
   cells compared to the NTC control. (B) Volcano plot highlighting (in
   red circles) the most differentially expressed genes of SLIT2 KO cells
   compared to NTC. (C) Bar graph showing viability of the SLIT2 KO
   compared to NTC cells under titration of rotenone (n = 3 for each
   group). (D) Bar graph showing the number of SLIT2 KO cells migrated
   across various channel widths compared to the NTC cells, with or
   without the presence of rotenone at 20% inhibitory concentration
   (IC[20]) (n = 3 for each group). Error bars indicate SD. P values were
   calculated by two-way ANOVA. ***P < 0.001; **P < 0.01; *P < 0.05; ns, P
   > 0.05.

DISCUSSION

   Previous studies of the role of SLIT2 in metastasis have produced
   contradictory observations in different experimental systems, with
   limited studies on prostate cancer models ([125]33, [126]34,
   [127]36–[128]39). Promoter hypermethylation was previously found to be
   responsible for the repression of SLIT2 and was associated with
   aggressive prostate, breast, and lung cancers ([129]33, [130]34). A
   recent study showed enhanced expression of SLIT2 in endothelial cells
   but repressed expression due to hypermethylation in the tumor
   compartment, suggesting the differential expression profiles of SLIT2
   at the tumor microenvironment as a driver for the release of tumor
   cells into the blood vessels ([131]33). As a top hit in our genome-wide
   CTC CRISPR KO screen and in a sublibrary screen, SLIT2 was identified
   as a strong modulator of invasiveness and metastasis in prostate
   cancer. The metastasis-promoting effect of SLIT2 KO was also confirmed
   in a prostate cancer mouse model in our study. Subcutaneous injection
   of SLIT2 KO cells resulted in increased tumor size accompanying marked
   increase in the number of CTCs in the whole blood compared to the same
   number of control cells injected. However, a SLIT2 KO model with
   similar tumor size as the control group still resulted in significantly
   increased number of CTCs in the bloodstream, suggesting a possible
   combined effect of SLIT2 on dissemination and proliferation. In
   addition, our study showed enhanced invasive and migratory behavior of
   SLIT2 KO cells compared to the control group, suggesting the potential
   inhibiting effect of SLIT2 on intravasation and dissemination. However,
   because metastasis is a very complicated process, the potential roles
   of SLIT2 on other pathways that lead to increase in CTC numbers
   including extravasation and cell viability in the bloodstream could be
   further studied to comprehensively understand the effect of SLIT2 on
   the metastatic process.

   No pathways have previously been connected with the invasive behavior
   of prostate tumor cells carrying a loss-of-function mutation in SLIT2,
   except that the expression of SLIT2 is regulated by the transcriptional
   repressor Enhancer of zeste homolog 2 (EZH2) ([132]34). Our EMT and
   transcriptome studies revealed aberrant expression of multiple EMT
   markers and ETC components in SLIT2 KO cells that potentially
   contribute to metastasis. Mitochondrial activity has been linked to
   cancer progression in various studies ([133]40–[134]44), with
   inhibitors of complex I reported to be promising antineoplastic
   therapies ([135]45, [136]46). With the validation of two additional
   prostate cancer cell lines LNCaP and PC-3, our studies suggest that
   prostate cancer cells lacking SLIT2 expression exhibit increased
   sensitivity toward complex I inhibition, revealing a potential
   therapeutic strategy for patients with cancer with SLIT2
   loss-of-function mutations.

   Bulk sequencing of tumors has uncovered a number of cancer-causing
   mutations, some of which were harnessed for the development of new
   targeted therapeutic agents. The identification of gene mutations that
   are functionally implicated in tumor progression is paramount to
   nominate new targets for cancer. One important limitation is assessing
   the function of genes in rare cell populations, such as in CTCs or
   subsets of immune cells present in the tumor microenvironment. In this
   study, we performed an in vivo genome-wide CRISPR-Cas9 KO screen
   targeting CTCs with prostate cancer origin in mouse models. Because the
   extreme rarity and fragility of patient CTCs resulted in poor
   feasibility of performing a genome-wide CRISPR screen without extensive
   ex vivo culturing of the CTCs, we injected a high number of tumor cells
   into the mice to generate a model to mimic the process of CTC
   production and to obtain sufficient number of tumor cells from whole
   mice blood for NGS. Development of this experimental pipeline is
   transferrable and can be easily applied to search for the unique
   CTC-promoting factors in other cancer types. Upon establishment of
   appropriate animal models, this platform can also be further applied to
   explore genetic factors that affect the behavior of immune cells or
   other rare cell types in the tumor microenvironment contributing to the
   metastatic process. Building on this study, we anticipate that
   combining efficient capture of modeled rare cells with gene editing
   methodologies will provide new insights into the progression of various
   diseases, including cancer, and will lead to the imagination of new
   therapies.

MATERIALS AND METHODS

Cell culture

   Prostate cancer cell line PC-3M was obtained from A. Allan at London
   Health Sciences, London, ON, Canada and was cultured in RPMI 1640
   (Wisent). LNCaP and PC-3 were obtained from American Type Culture
   Collection (ATCC) and cultured in RPMI 1640 and Dulbecco’s modified
   Eagle’s medium (DMEM)/F12 (Wisent), respectively. MDA-MB-231 (ATCC) and
   lenti-X 293T cells (Takara Bio) were grown in DMEM (Sigma-Aldrich)
   supplemented with 4 mM stable glutamine (Sigma-Aldrich). All culture
   media were supplemented with 10% fetal bovine serum (FBS) (Wisent) and
   1% penicillin/streptomycin (Wisent), and cells were kept at 37°C, 5%
   CO[2].

Lentiviral production

   All lentivirus was produced using lenti-X 293T cells. For TKOv3
   genome-wide library production, 8.0 μg of TKOv3 pooled plasmid
   (Addgene, #90294) ([137]22) was cotransfected into 80% confluent
   lenti-X 293T cells with 4.8 μg of psPAX2 (Addgene, #12260) and 3.2 μg
   of pMD2.G (Addgene, #12259) per 15-cm plate (Sarstedt). Plasmids psPAX2
   and pMD2.G were gifts from Didier Trono. The mixture of plasmids (16.0
   μg in 50 μl of ddH[2]O) was added to 48 μl of X-tremegene 9 DNA
   transfection reagents (Roche) that preincubated for 5 min in 750 μl of
   OptiMEM (Gibco). The mixture of plasmid transfection reagent–OptiMEM
   was further incubated for 25 min before adding dropwise to lenti-X
   cells. Eighteen hours later, medium was changed to fresh DMEM. Viral
   particles were harvested 48 hours after the media exchange, snap-frozen
   in liquid nitrogen, and stocked at −80°C. The customized sublibrary
   (plasmid pool synthesized at GenScript) lentivirus was prepared using
   the same protocol as the TKOv3 library. Individual SLIT2 CRISPR KO
   lentivirus was generated following the polyethyenimine transfection
   protocol (Addgene). sgRNAs targeting the SLIT2 gene were cloned into
   the lentiCRISPRv2 (Addgene, #52961) ([138]47), and the 20–base pair
   (bp) sgRNA sequences used were CGGGTTGGTCTGTACACTCA for KO 1 and
   ACGGAAAGCTTTCCTTGGGA for KO 2. For the NTC, the sgRNA sequence used was
   CCCGAATCTCTATCGTGCGG.

Lentiviral transduction and multiplicity of infection determination

   Lentivirus was titrated into 2 × 10^5 PC-3M cells on six-well plates
   with 2 ml of complete RPMI 1640 supplemented with polybrene (8 μg/ml;
   Sigma-Aldrich). After overnight infection, medium was exchanged for
   fresh RPMI supplemented with puromycin (2 μg/ml; Sigma-Aldrich). After
   48 hours of antibiotic selection, multiplicity of infection (MOI) was
   determined by comparing the number of cells remained at specific viral
   load with or without puromycin selection using 0.1% crystal violet
   staining (Sigma-Aldrich) and absorbance reading at 590 nm. A MOI of 0.3
   was used for the transduction of CRISPR libraries into the PC-3M cells.

Genome-wide CTC CRISPR KO screen

   PC-3M cells (1.0 × 10^8) were transduced with TKOv3 library (71,090
   sgRNAs) at a MOI of 0.3 using polybrene as transduction reagent as
   described in the previous section. After two rounds of puromycin
   selection, 3 × 10^7 transduced cells were subcutaneously injected into
   the flank of each of the six male athymic nude (nu/nu) mice of 6 to 8
   weeks old (1.5 × 10^7 each side per mouse). Three additional mice were
   injected with unmodified parental PC-3M cells for tumor size
   comparison. Tumor volume was measured twice a week using a caliper and
   calculated using the formula length × width^2/2. Mice were euthanized 3
   weeks after transplant to collect tumor and whole blood samples
   (cardiac puncture). All animal work was conducted following the
   protocol approved by University of Toronto Animal Care Committee under
   the guidelines of Canadian Council on Animal Care.

Prism chip design and fabrication

   An initial iteration of the Prism chip ([139]24) was modified to
   increase throughput to 3 × 10^7 cells/hour per chip and sort cells into
   zero/low-, medium-, and high-expression outlets. Magnetically labeled
   cells are deflected laterally on the basis of their level of
   expression. Deflection is facilitated using guides made from highly
   magnetically permeable alloy set at an angle from the fluid flow (5°
   for medium and 20° for high). In the presence of a magnetic field from
   a permanent magnet, localized regions of high magnetic flux are found
   at guide edges causing the attraction of magnetic nanoparticles
   ([140]48). A balancing of the magnetic force and the Stokes drag force
   from fluid flow causes magnetically labeled cells to follow the
   deflection guides until the angle changes or the guides end at the
   lateral edges of the device. Angle changes of the guides cause an
   increase in the component of the drag force and medium-expression cells
   to detach. Chips were fabricated on a soda lime glass wafer
   (UniversityWafer). Guides were patterned using photolithography and wet
   etching, and channels were created using SU-8 photo resist (Kayaku AM).
   A polydimethylsiloxane (PDMS) slab (Dow Sylgard) was bonded to
   encapsulate the device according to an existing protocol ([141]49).
   Multiple chips were set up in parallel to simultaneously isolate CTCs
   from multiple blood samples. To prepare the Prism chips for CTC
   isolation, the chips were primed with sterile water containing 0.1%
   Pluronic F68 (Sigma-Aldrich) under constant hydraulic pressure for
   overnight before use. The nonionic surfactant Pluronic F68 prevented
   nonspecific capture of cells in the microchannel, and the constant
   hydraulic pressure removed small air bubbles. For immunomagnetic-based
   cell sorting, the Prism chip was first mounted on a permanent magnet
   (BY088-N52, K&J Magnetics). Before sample loading, a buffer solution
   made of Hanks’ balanced salt solution (HBSS; GIBCO) supplemented with
   2% bovine serum albumin (BSA; Sigma-Aldrich) and 5 mM EDTA (BioShop)
   was introduced to the Prism chip to remove the pluronic water.

CTC collection and cell lysis

   To isolate the CTCs, each blood sample was diluted in a 1:1 ratio with
   the HBSS buffer solution and incubated with 10% EpCAM MicroBeads
   (Miltenyi Biotec) with shaking at room temperature for an hour.
   Afterward, equal volumes of buffer solution and the blood samples
   containing the magnetically labeled CTCs were loaded onto the buffer
   inlet and sample inlet, respectively, before withdrawn from the chip at
   a flow rate of 2 ml/hour. After sample processing, 300 μl of buffer
   solution was loaded to both the sample and buffer inlets and withdrawn
   at the same flow rate to wash out remaining samples in the
   microchannels. Solution collected from the high-expression outlet (~500
   μl) of each chip was injected into a 1.5 ml centrifuge microtube
   (Axygen) and set on the DynaMag-2 magnet (Thermo Fisher Scientific) for
   an hour. After the cells settled down, supernatant was pipetted out
   slowly without touching the bottom or the wall of the tube, leaving
   minimal solution (5 to 10 μl) containing CTCs. Alkaline lysis method
   adapted from a previous publication was used to lyse CTCs ([142]25).
   Briefly, 10 μl of 200 mM KOH with 50 mM dithiothreitol was added to CTC
   solution and heated at 65°C for 10 min in a thermocycler. Lysate was
   then neutralized by 10 μl of 900 mM tris-HCl (pH 8.3) with 300 mM KCl
   and 200 mM HCl.

Genomic DNA extraction and PCR amplification of sgRNA

   To extract gDNA from the CRISPR library–transduced cell pool (3 × 10^7
   to 5 × 10^7 cells) and the tumor tissues (100 to 200 mg), the QIAamp
   DNA Blood Maxi Kit (Qiagen) was used following the manufacturer’s
   instructions. For gDNA extraction from tumor tissues, tumors were
   chopped into small pieces using a razor blade and incubated with 1.8-ml
   tissue lysis buffer ATL (Qiagen) and 0.2 ml of proteinase K (Qiagen) at
   56°C overnight before processed using the QIAamp DNA Blood Maxi Kit.
   gDNA extracted from the initial cell pool, and the primary tumor
   samples were checked for 230:280 and 260:280 ratios before sending to
   the Princess Margret Genomics Center (PMGC; Toronto, ON, Canada) for
   PCR amplification of the sgRNA region and generation of the barcoded
   sgRNA library. For the CTC samples, gDNA extraction was not necessary,
   and 10 μl of the neutralized lysate from each CTC sample was directly
   set for PCR amplification of the sgRNA region. Each 100 μl of PCR was
   set as follows: 10 μl of 10× buffer (supplied with Ex Taq), 8 μl of
   deoxynucleotide triphosphate (supplied with Ex Taq), 5 μl of Amp1
   primer mix (10 μM), 10 μl of neutralized CTC lysate, 65.5 μl of water,
   and 1.5 μl of Ex Taq DNA polymerase (Takara Bio).

   The thermocycling parameters used were as follows: Step 1: 95°C for 1
   min; step 2: 95°C for 30 s; step 3: 52°C for 30 s; step 4: 72°C for 20
   s (repeat steps 2 to 4 for 34 additional times); step 5: 72°C for 2
   min; step 6: 4°C hold. PCR products were run on a 1% agarose gel to
   confirm the size of the product (expected size: 235 bp) before sent to
   PMGC for barcoded sgRNA library generation. NGS was done using Illumina
   NovaSeq 6000 with 1 million reads per sample.

CTC subpool CRISPR KO screen

   Sublibrary (400 sgRNAs) was synthesized and cloned into the same vector
   as the TKOv3 genome-wide library (lentiCRISPRv2) by GenScript. Cloned
   sublibrary was deep-sequenced to verify sgRNA representation in the
   plasmid pool. Lentivirus production, MOI determination, and in vivo
   screen were conducted in the same manner as the TKOv3 library, except
   that GPI-anchored tag was lentivirally introduced to the PC-3M cells
   for displaying Myc on the cell surface before introducing the
   sublibrary to the cells. Thus, CTCs in blood samples were labeled with
   1% biotinylated anti-Myc antibody (Abcam, ab197139) for an hour and 5%
   anti-biotin microbeads (Miltenyi Biotech) for an hour before being
   isolated on the Prism chip. Lysis and sgRNA PCR amplification of CTCs
   in the sublibrary screen used the same protocol as the genome-wide
   screen.

NGS data and sgRNA enrichment analyses

   The genome-wide screen data were mapped by PMGC, and the customized
   sublibrary data were mapped following the MAGeCK pipeline ([143]50).
   Number of reads in each sample was normalized by converting raw sgRNA
   counts to reads per million (rpm). The rpm values were then subject to
   log[2] transformation for downstream analyses. To generate correlation
   heatmaps, Nonnegative matrix factorization (NMF) R package was used
   ([144]51), and Pearson correlations between individual samples were
   calculated using log[2] rpm counts. Empirical cumulative distribution
   plot was generated using the Python libraries Pandas, Seaborn, and
   Matplotlib. Biological pathway enrichment analysis was performed using
   Metascape ([145]52). For sgRNA enrichment analysis, fold change was
   obtained using MAGeCK and two criteria were used to identify the top
   candidate genes from the genome-wide screen based on CTC sgRNAs: (i)
   counts: if a sgRNA was counted as the top 100 in the normalized mapping
   results of any CTC sample (100 sgRNAs × 6 replicates), resulting in 162
   unique sgRNAs instead of 600 due to the exclusion of the overlapping
   sgRNAs among the six replicates; (ii) replicates: if a CTC sgRNA was
   enriched compared to its counterpart in the corresponding primary tumor
   sample (fold change ≥ 1.0) across all the samples (n = 6). For the
   sublibrary data, sgRNA counts were normalized and analyzed for
   enrichment using DrugZ ([146]53).

In vivo validation of screen top hit

   PC-3M NTC or SLIT2 KO cells (1.0 × 10^6 each) were orthotopically
   injected into the prostate via the right dorsolateral lobe of 6- to
   8-week-old male athymic nude (nu/nu) mice. After 3 weeks, mice were
   euthanized to collect primary tumor (size comparison), blood (CTC
   counting), and lymph nodes (metastasis histology). Peripheral lymph
   nodes from the mice were collected, including two inguinal, two
   axillary, and two brachial lymph nodes from each mouse. Lymph nodes for
   histology were formalin-fixed, paraffin-embedded, sectioned, and
   stained with H&E (prepared by Pathology Research Program Laboratory at
   University Healthy Network). Prepared tissue sections were imaged and
   examined by a pathologist at Charles River Laboratories for the
   assessment and grading of the metastatic lesions. The metastatic status
   of the lymph nodes was categorized into three groups: no neoplastic
   cells observed inside the lymph nodes or at the peripheral tissues
   (none); neoplastic cells observed inside the lymph nodes or at the
   peripheral tissues (partial); and neoplastic cells observed inside the
   lymph nodes and constituted greater than half of the cross-sectional
   area (severe).

Fabrication and performance of the eight-zone CTC capture device

   Eight-zone profiling chips were fabricated using 3D stereolithography
   for the 3D-printed molds and standard soft lithographic techniques for
   subsequent steps as described previously ([147]54–[148]56). Before
   usage, the microfluidic chips were treated with 0.1% Pluronic F-68
   overnight. Devices were assembled with two rectangular arrays of N52
   neodymium magnets above and below before connected to a syringe pump
   (Chemyx) set to withdraw for the duration of the sample processing.
   Blood samples were prepared following the same protocol as for the
   Prism chips before loaded onto the eight-zone device, and CTCs were
   captured and profiled for EpCAM expression. Phosphate-buffered saline
   (PBS; 500 μl) was withdrawn through the chip to displace the Pluronic
   F-68 solution before sample loading. The samples were processed at a
   flow rate of 750 μl/hour. The samples were then fixed with 150 μl of 4%
   paraformaldehyde (Sigma-Aldrich) in PBS, permeabilized with 150 μl of
   0.2% Triton X-100 (Sigma-Aldrich) in PBS. The chips were washed in
   between each step with 125 μl of CliniMACS PBS-EDTA solution
   (Miltenyi). Captured cells were immunostained by 200 μl of an antibody
   cocktail containing anti-pan cytokeratin AF488 (30 μg/ml; Abcam,
   ab277270), anti-cytokeratin 18 fluorescein isothiocyanate (6 μg/ml;
   LifeSpan BioSciences, LS-[149]C46335), anti-cytokeratin 19 AF488 (30
   μg/ml; BioLegend, 628508), and anti-mouse CD45 fluorophore
   Allophycocyanin (3 μg/ml; BD Bioscience, 559864) in PBS supplemented
   with 1% BSA and 0.1% Tween 20 (Sigma-Aldrich) at a flow rate of 200
   μl/hour. The captured cells were then stained with
   4′,6-diamidino-2-phenylindole (Thermo Fisher Scientific, [150]R37606)
   before fluorescently imaged and scanned using a Nikon Ti-E Eclipse
   microscope with Andor’s Neo sCMOS camera. CTCs were identified as
   nucleus present and EpCAM^+ CK^+ CD45^− cells.

Spheroid assay

   3D spheroids consisting of NTC or SLIT2 KO cells were grown following a
   protocol described previously ([151]57). Briefly, RPMI 1640 was
   supplemented with 0.75% methylcellulose (Sigma-Aldrich), and the
   spheroids were cultured in pretreated ultralow attachment plates
   (Corning). To determine the amount of cells consisting the spheroids in
   each well, CellTiter-Glo 3D Cell Viability Assay (Promega) was used
   following the manufacturer’s instructions. Briefly, assay reagent was
   mixed vigorously with cell culture medium at a 1-to-1 ratio after the
   plate was brought to room temperature. After 25 min of incubation at
   room temperature, luminescence was recorded for each well.

Migration chip fabrication and cell migration/deformability assay

   The migration chip includes a sample-loading channel in the middle and
   two stimulus-loading channels on each side. There are numerous narrow
   migration channels (constant height at 5 μm) with varying widths
   ranging from 6 to 20 μm connecting the sample channel and stimulation
   channels for cell migration. The chip consists of a PDMS substrate with
   microchannels and a glass cover. The PDMS substrate was fabricated
   using soft lithography. Briefly, the liquid base and reagent of PDMS
   were fully mixed at a ratio of 10:1, casted on a SU8-negative
   photoresist-patterned silicon mold, and then incubated at 70°C for 2
   hours for polymerization. The solidified PDMS was peeled off the mold
   and bonded with the glass cover upon plasma treatment. For cell
   deformability assay, the migration chip was first filled with 70%
   ethanol and ultraviolet-treated. The degassed and sterilized chip was
   then washed with PBS followed by cell culture medium. PC-3M NTC (3 ×
   10^6) or SLIT2 KO cells in 500 μl of complete medium were loaded to the
   sample channel and incubated at 37°C for overnight to allow cell
   attachment. The next day, the sample channel was washed with PBS and
   loaded with serum-free medium. The stimulus channels were filled with
   fresh cell culture medium with 10% FBS to produce nutrient gradient
   along the narrow migration channels. After 6 hours of incubation, cells
   across the entire chip were fluorescently labeled by loading 5 μM SYTO
   24 (Invitrogen) to both the sample and stimulation channels and
   incubated for 15 min. Migration behavior of the cells were inspected
   under a fluorescence microscope. Cells that entered the migration
   channels or reached the stimulus channels were considered as deformed
   and migrated cells. The number of migrated cells was counted and
   compared for the deformability and migration potential of the cells
   from different samples. For testing the effect of rotenone on
   deformability and migration of the cells, rotenone at 20% inhibitory
   concentration (IC[20]) was added to the medium loaded onto the chip.

Transwell cell invasion assay

   To prepare the invasion assay chambers, 750 μl of 10% FBS-supplemented
   RPMI 1640 was added to each of the 24 wells of the Transwell plate
   (Corning), and 100 μl of Matrigel (Corning) was added to the cell
   culture inserts (8 μm). The plate was incubated in a humidified
   incubator at 37°C with 5% CO[2] for an hour. Five hundred microliters
   of PC-3M NTC or SLIT2 KO cells in serum-free RPMI 1640 was added to the
   Matrigel-coated inserts at 1 × 10^5 cells/ml. On the basis of the
   conditions tested, medium in the wells was either with or without 10%
   FBS and secreted N-SLIT2 (1 μg/ml; R&D Systems, 8616-SL-050). The cells
   were incubated in the Matrigel-coated inserts for overnight in the
   incubator. The next day, medium was removed from the inserts and wells.
   The nonmigrated cells on the upper side of the membrane were removed
   using a cotton swab and washed away with PBS. Cells that remained on
   the membrane were fixed by submerging the membranes in the wells filled
   with cold methanol for 20 min and air-dried for 30 min. The membranes
   were then submerged in 750 μl of 0.1% crystal violet (Sigma-Aldrich) in
   wells for 30 min at room temperature to stain the cells before washed
   thoroughly with distilled water. After air-drying, membranes were
   submerged in 750 μl of 10% acetic acid (VWR) with shaking until the
   stain was completely dissolved. Optical density of the dissolved stain
   in each well was determined at 590 nm using a microplate reader.

Oncology proteome antibody array

   Oncology antibody array (R&D Systems) was performed following the
   manufacturer’s instructions. Briefly, NTC or SLIT2 KO cell lysates were
   diluted and incubated overnight with the nitrocellulose membranes that
   had been prespotted in duplicate with antibodies targeting 84
   oncology-related proteins. The membranes were washed before incubated
   with a cocktail of biotinylated detection antibodies, followed by
   addition of streptavidin–horseradish peroxidase. The amount of protein
   bound at each capture spot was detected using a chemiluminescent
   substrate, and the signal was captured using a ChemiDoc imaging system
   (Bio-Rad). Intensity of the capture spots was quantified using ImageJ.

Immunoblot analysis

   Whole-cell extracts were prepared with radioimmunoprecipitation assay
   buffer (Thermo Fisher Scientific) supplemented with protease inhibitor
   cocktail (Sigma-Aldrich). Proteins were separated on 4 to 15% precast
   protein gels (Bio-Rad) and transferred onto polyvinylidene difluoride
   membranes. Primary antibodies used were anti-SLIT2 (Abcam, ab134166),
   anti-Vimentin (Abcam, ab92547), anti–N-cadherin (BD Bioscience,
   610920), anti–E-cadherin (R&D Systems, AF648), and anti-Snail (NEB,
   3879S). Secondary antibodies were purchased from Cell Signaling
   Technology. Blots were visualized with chemiluminescent substrate
   (Thermo Fisher Scientific) using a ChemiDoc imaging system. Intensity
   of the protein bands were quantified using ImageJ.

RNA sequencing and data analysis

   RNA was extracted using the RNeasy Plus Kit (Qiagen) following the
   manufacturer’s instructions and submitted to PMGC for complementary DNA
   library preparation and RNA sequencing using Illumina NovaSeq 6000. RNA
   sequencing data was processed for sgRNA representation using A*STAR
   procedure. Figures were generated using the normalized read counts in R
   and RStudio (R project, Revolution Analytics), Metascape, and the Gene
   Set Enrichment Analysis tools (Broad Institute).

Cell viability assay

   NTC or SLIT2 KO PC-3M cells were seeded on 96-well plates in triplicate
   at a density of 2 × 10^4 cells per well. The next day, rotenone was
   titrated into the 96-well plate at twofold serial dilution with
   concentration ranging from 0 to 10 μM. Viable cell population was
   determined 48 hours later using the Cell Counting Kit-8 (Dojindo)
   following the manufacturer’s instructions.

Clinical and patient data analysis

   RNA sequencing and clinical data were retrieved from portals of TCGA,
   the Human Protein Atlas, GEPIA2, and COSMIC. SLIT2 expression median of
   2.45 Fragments Per Kilobase Million (FPKM) was used as the threshold to
   define low or high expression of SLIT2 in patients with prostate cancer
   (prostate adenocarcinoma). SLIT2 expression levels in tumor and normal
   tissues were retrieved for 31 cancer types to assess fold change.
   Information on the loss-of-function mutations in the top screen hits
   occurred in patients with cancer was retrieved from Cancer Mutation
   Census (CMC) at COSMIC to compare with the level of loss-of-function
   mutations in well-known TSGs and OGs ([152]29).

Data collection and analysis

   Counts of sgRNAs were collected by PMGC Toronto and analyzed using
   MAGeCK or DrugZ as described above. Screen data quality control
   analysis and figure preparations were performed using software packages
   as specified in the corresponding sections above. Microscopy images
   were acquired using Zeiss Axio Observer and prepared using Zeiss Zen
   Blue. Protein band intensities were analyzed and quantified using
   ImageJ. Biological pathway analysis was performed using Metascape. Data
   were analyzed using the statistical tests with specified sample sizes
   indicated in each figure legend. Statistical tests and graphic
   representation of the data were conducted using Graphpad Prism v7.0.
   All data were presented as the mean with SD indicated by error bars.
   Statistical significance was set at P < 0.05 with two-sided tests.

Acknowledgments