Specific synthetic lethal killing of RAD54B-deficient human colorectal cancer cells by FEN1 silencing

Abstract

Mutations that cause chromosome instability (CIN) in cancer cells produce “sublethal” deficiencies in an essential process (chromosome segregation) and, therefore, may represent a major untapped resource that could be exploited for therapeutic benefit in the treatment of cancer. If second-site unlinked genes can be identified, that when knocked down, cause a synthetic lethal (SL) phenotype in combination with a somatic mutation in a CIN gene, novel candidate therapeutic targets will be identified. To test this idea, we took a cross species SL candidate gene approach by recapitulating a SL interaction observed between rad54 and rad27 mutations in yeast, via knockdown of the highly sequence- and functionally-related proteins RAD54B and FEN1 in a cancer cell line. We show that knockdown of RAD54B, a gene known to be somatically mutated in cancer, causes CIN in mammalian cells. Using high-content microscopy techniques, we demonstrate that RAD54B-deficient human colorectal cancer cells are sensitive to SL killing by reduced FEN1 expression, while isogenic RAD54B proficient cells are not. This conserved SL interaction suggests that extrapolating SL interactions observed in model organisms for homologous genes mutated in human cancers will aid in the identification of novel therapeutic targets for specific killing of cancerous cells exhibiting CIN.

Genomic instability is now widely recognized as an important factor in the evolution of cancer and arises through either of 2 mechanisms—increased mutation rate or chromosome instability (CIN). CIN correlates with ≈85% of solid tumors and is characterized by an increased error rate in the gain or loss of chromosomes during cell division (1). CIN is associated with numerous different tumor types including colon (2–5), ovarian (6, 7), and non-Hodgkin lymphoma (8–12), and it is believed to be an early event in the etiology of tumorigenesis (13–15). Conceptually, CIN promotes tumor heterogeneity by increasing or decreasing chromosome numbers (16), and directly affects the expression levels of both oncogenes and tumor suppressor genes encoded on the mis-segregated chromosomes. Most importantly to the work presented here, CIN gene mutations genetically distinguish tumor cells from normal cells and may therefore represent a genetic susceptibility that could be exploited for selective killing (see below). Consequently, identifying the gene products that regulate chromosome stability (CS) will not only provide insights into the molecular mechanisms of chromosome segregation and tumorigenesis, but it will also provide a list of candidate cancer CIN genes that may be exploited to identify novel therapeutic targets for the treatment of cancer.

In 1997, Hartwell and colleagues (17) posited that cancer cells harboring somatic mutations or deletions represent genetically sensitized cells, relative to normal surrounding cells, that may be susceptible to drug therapies selectively targeting a second unlinked gene product. They suggested that synthetic lethality (SL), which refers to the lethal combination of 2 independently viable mutations or deletions in 2 unlinked genes (Fig. 1A), could be used in model organisms such as yeast to identify candidate SL interactions that may be conserved in humans. Because chromosome segregation is an essential cellular process, we hypothesized that the CIN phenotype associated with tumors, but not normal cells, represents an excellent “Achilles' heel” that would allow for the selective killing of cancer cells. Presumably, if cross-species tests of candidate genes can be applied to identify second site targets that exacerbate the sublethal defect associated with a CIN-inducing mutation, a novel drug target will be identified. Furthermore, by generating a SL interaction network for the set of yeast CIN genes whose human homologs are somatically mutated in tumors (Fig. 1B), we can identify those yeast genes that are positioned as SL “interaction nodes” and whose human homologs would then represent candidate therapeutic targets for a broad spectrum of tumors (Fig. 1 C and D).

Fig. 1.

Synthetic lethality in model organisms and human cancer. (A) A SL interaction occurs when 2 independently viable gene mutations/deletions [i.e., yfg1 (e.g., rad54) and yfg2 (e.g., rad27)] are combined to produce a lethal phenotype. If slow growth is observed, a synthetic growth defect (SGD) is defined. (B) A representative example of a genetic interaction network generated from yeast data available in Biogrid (49), where circles identify genes and lines represent SL/SGD interactions. Note that rad27 intersects with both rad54 and rdh54. (C) Schematic representation of SL/SGD in a human cancer context. A mutation or deletion of yfg1 (e.g., RAD54B CIN mutation) genetically sensitizes a cancer cell to SL attack through down-regulation of a second unlinked gene product [yfg2 (e.g., FEN1)], while leaving the normal adjacent cell(s) unaffected. (D) The yeast network presented in (B) has been humanized by identifying the top hit human homolog for the respective yeast genes and is presented. Note that the lines only identify candidate interactions assuming evolutionary conservation. (E) Haploid rad54::URA3 and rad27::KanMX were mated and induced to undergo meiosis. The resulting tetrads were dissected on YPD and later replica plated to additional selection media to identify the genotypes indicated on the right. The combination of rad54::URA3 rad27::KanMX within the same spore resulted in SL (indicated by boxes).

In this study, we use both RAD54B knockout and RNAi-silenced cells, and isogenic controls, to demonstrate that decreases in human RAD54B expression in colorectal cancer cells correlates with increases in chromosome numbers. RAD54B was chosen because homozygous mutations at highly conserved positions have been identified in human primary lymphoma and colon cancers (18), although the functional status of these mutant alleles has not been directly tested in a mammalian context (19). Furthermore, RAD54B exhibits a significant degree of sequence and functional similarity with yeast Rdh54 and Rad54 which both exhibit strong CIN phenotypes in yeast (20). Using a cross-species candidate gene approach and high-content digital imaging microscopy techniques, we show that the synthetic lethality observed in yeast for rad54 rad27 double mutants (see Fig. 1E and ref. 21) is conserved within a human colorectal cell line (by simultaneous down-regulation of the corresponding human gene products, RAD54B and FEN1). Decreases in cell numbers with concomitant increases in cellular cytotoxicity were observed in RAD54B deficient/FEN1 depleted cells that were not apparent in isogenic RAD54B proficient/FEN1 depleted cells. These findings represent an example of a validated target for selective killing of mammalian cells as a prediction from a SL interaction between a CIN gene mutation and an unlinked gene mutation in yeast. We suggest that extrapolation of SL genetic interaction networks identified in yeast to a human context will provide a productive strategy to identify a broad range of novel cancer therapeutic targets.

Results

Diminished Rad54B Expression Causes Chromosome Instability in Human Tissue Culture.

Having previously demonstrated in yeast that both RAD54 and RDH54 play important roles in maintaining CS (20), we wished to determine if RAD54B exhibits a similar role in humans. Accordingly, we used genomic knockouts [RAD54B^−/−/− (see SI Materials and Methods for description)] generously provided by Dr. Miyagawa (22) and short-hairpinRNAs (shRNA) targeting RAD54B. HCT116 colorectal cells were specifically selected, because it is a near diploid cell line that does not inherently exhibit CIN (i.e., is chromosomally stable). Before examining CS, RAD54B expression was assessed at the protein level and determined to be absent or significantly diminished in the knockout or knockdown cells, respectively (Fig. 2A). Since most colorectal cancers exhibit increases in chromosome numbers (23, 24), we focused our attention to the proportion of cells with DNA contents in excess of the normal diploid G2/M peak (i.e., >4N). As RAD54B levels decreased, corresponding increases in the proportion of cells with DNA contents beyond the 4N peak were observed (Fig. 2B and Table 1). In addition, small discrete peaks corresponding to increases in polyploidy were observed in both the RAD54B knockout and knockdown cells (Fig. 2B). To determine if increases in chromosome number could account for the increases in DNA content, mitotic chromosome spreads were generated and total chromosome numbers were manually quantified (Fig. 2 C and D and Table 2). Although the modal number of chromosomes remained at 45, increases in total chromosome numbers for a small subset of cells were evident in the RAD54B depleted cells that were not evident in the controls (HCT116, Non-silencing, or eGFP) (Table 2). In agreement with the flow cytometry data, the fraction of cells with elevated chromosome numbers (i.e., >46) correlated with decreased RAD54B expression. Furthermore, the increases in the near polyploid populations observed by flow cytometry (Fig. 2B), were also apparent within the corresponding chromosome spreads (Fig. 2D). Subsequent Student's t tests comparing the mean chromosome number for each treatment and the isogenic HCT116 control revealed statistically significant differences for the RAD54B^−/−/− cells and the RAD54B-1- and RAD54B-2-treated cells (Fig. 2E and Table S1). Not surprisingly, these 3 conditions are those in which RAD54B expression has decreased the most (see Fig. 2A).

Fig. 2.

RAD54B depletion underlies CIN. (A) Western blots depicting RAD54B expression levels in knockout (RAD54B^+/+/− and RAD54B^−/−/−), knockdown (RAD54B-1, RAD54B-2, and RAD54B-3), and control (Untransfected, Non-silencing, and eGFP) cells. An α-tubulin loading control has been included. (B) DNA content analysis of RAD54B knockout and knockdown cells. Asynchronous cells were PI-labeled and subjected to flow cytometry. The diploid 2N (G0/G1), 4N (G2/M) and >4N (aneuploid/polyploid) populations have been identified. The various cell lines/conditions are indicated in the legend. (C) Representative images of DAPI counterstained chromosomes found in mitotic spreads generated from untransfected HCT116 (top left), RAD54B-1 transfected (top right and bottom left) and RAD54B^−/−/− (bottom right) cells. The total chromosome numbers are indicated. (D) Scatter plots depicting the total chromosome distribution for cells RAD54B knockout and knockdown cells and controls. (E) Graphical representation of the mean chromosomes numbers determined for each of the conditions indicated on the x axis as quantified from the mitotic spreads (± SEM). Student's t tests were performed between the mean chromosome number of the untransfected HCT and each of the conditions. Conditions with statistically significant differences in means are identified by *, P <0.05 and ***, P <0.001.

Table 1.

DNA content analysis of RAD54B depleted colon cancer cells by flow cytometry

Cell Line	Percentage of total cells					Fold increase over Untransfected (>4N)
	<2N	2N (G0/G1)	S-phase	4N (G2/M)	>4N
Untransfected	1.0	31.4	39.9	21.8	5.9	NA
Non-Silencing	1.2	33.3	36.8	21.2	7.5	1.3×
eGFP	1.2	30.1	37.9	24.3	6.5	1.1×
RAD54B-1	1.5	27.8	36.2	16.5	18.0	3.1×
RAD54B-2	1.5	27.5	33.4	22.3	15.3	2.6×
RAD54B-3	1.4	29.3	34.4	21.6	13.3	2.3×
P37 (RAD54B+/+/−)	0.9	30.5	36.4	22.0	10.2	1.8×
HPHB (RAD54B−/−/−)	1.0	28.8	31.6	21.5	17.1	2.9×

Only a single set of values from a single representative dataset are shown; experiments were conducted in triplicate. NA, not applicable

Table 2.

Increased Chromosome Numbers Following Diminished RAD54B Expression

Condition	n	Percentage of Mitotic Spreads								FI>46
		≤43	44	45	46	47	48	49	≥50
Untransfected	322	4.0	16.1	63.8	12.7	2.2	0.0	0.0	1.2	NA
Nonsilencing	221	6.3	16.7	58.8	12.7	1.8	0.0	0.5	3.2	1.59×
eGFP	201	5.5	18.9	56.2	13.9	3.5	1.0	0.0	1.0	1.61×
RAD54B-1	213	7.0	11.7	52.6	12.2	7.0	2.8	1.4	5.2	4.83×
RAD54B-2	215	9.3	13.4	50.5	12.5	5.1	3.2	0.9	5.1	4.21×
RAD54B-3	207	9.7	14.5	53.1	11.6	4.8	1.9	0.5	3.9	3.26×
RAD54B+/+/−	203	5.9	16.3	55.7	12.8	3.9	1.5	1.5	2.5	2.75×
RAD54B−/−/−	606	7.3	15.7	56.3	10.4	3.8	0.7	0.3	5.6	3.05×

n = number of mitotic spreads quantified. FI_>46 identifies the fold increase in frequency of cells harboring > 46 chromosomes (sum of the percentages for those mitotic spreads with 47, 48, 49 and ≥ 50 chromosomes) relative to the Untransfected control HCT116 (first line). NA, not applicable

Ectopic Expression of RAD54B Suppresses the CIN Phenotype.

To more conclusively demonstrate that RAD54B expression is causally linked to CIN, phenotypic rescue experiments were performed in RAD54B^−/−/− cells. Briefly, V5 or EmGFP tagged versions of RAD54B were ectopically expressed in RAD54B^−/−/− cells. After a brief selection process, cells were divided into 2 groups; 1 group was used for protein quantification and the second group was harvested for DNA content analysis by flow cytometry as above. Total RAD54B expression was assayed by either standard western blot analysis (V5-RAD54B expression level) or quantitative imaging microscopy (QIM; EmGFP-RAD54B expression levels) which quantify expression levels in cell populations, or single cell levels, respectively (Fig. 3A). In both cases, ectopic RAD54B expression levels were determined to be slightly elevated over wild-type RAD54B levels, but still remained predominantly within the endogenous expression range (Fig. 3B). In fact, QIM demonstrated that the predominant proportion (i.e., >80%) of cells ectopically expressing EmGFP-RAD54B had expression levels within the normal distribution range of isogenic HCT116 cells expressing endogenous RAD54B levels. Next, we wished to determine if ectopic RAD54B expression could rescue the CIN phenotype. RAD54B expressing cells and controls were subjected to FACS analysis as detailed above, and DNA content profiles are presented in Fig. 3C. Interestingly, the DNA content profiles for the RAD54B rescued cells more closely resembled the profiles of the wild-type (RAD54B proficient) HCT116 cells than the parental RAD54B^−/−/− into which RAD54B was reintroduced (Table S2). More specifically, the near polyploid populations previously present within the RAD54B^−/−/− parental line are visually diminished (Fig. 3C). Since it is highly unlikely that ectopic RAD54B expression reverts karyotypically abnormal cells to karyotypically normal cells, these observations are most likely attributable to either natural apoptotic mechanisms affecting the near polyploid populations or through diminished cell cycle progression and/or proliferation which would effectively dilute those near polyploid cells within the actively growing normal diploid cells.

Fig. 3.

Ectopic RAD54B expression rescues CIN. (A) Western blot analysis of RAD54B expression in the RAD54B^−/−/− cells ectopically expressing V5-RAD54B was determined to be near wild-type levels (HCT116) at the populations level. An α-tubulin loading control is included. (B) GFP-RAD54B expression levels at single cell resolution as determined by QIM. Note that the entire range of the normalized RAD54B signal intensities are shown for both the wild-type HCT116 cells and the isogenic RAD54B^−/−/− cells ectopically expressing EmGFP-RAD54B. Although the distribution range is larger in the transfected cells, the regions indicated in the boxes [25^th percentile (bottom line), mean (middle line), and 75^th percentile (top line)] overlap to a large degree, indicating that protein expression levels are similar, albeit slightly elevated. (C) Asynchronous and sub-confluent cells were PI-labeled and subjected to flow cytometry. The DNA content profiles were determined for wild-type HCT116 cells (red) and RAD54B^−/−/− cells (green) ectopically expressing V5-RAD54B (blue), EmGFP-RAD54B (brown), or empty EmGFP vector alone (purple). The arrows highlight the near polyploid populations that exist within the RAD54B-deficient cells.

RAD54B Deficient Cells Exhibit Proliferation Defects when FEN1 Expression Is Reduced.

Yeast rdh54 and rad54 have each previously been shown to exhibit SL/SGD interactions with rad27 (21, 25–27) (Fig. S1A). To determine if a similar genetic interaction is conserved in human cells, RAD54B proficient and deficient cells were transiently transfected with siRNA duplexes specifically targeting FEN1, the homolog of yeast rad27. All FEN1 duplexes (including the pools) specifically target unique non-overlapping regions within the FEN1 coding region and cause reduced FEN1 expression 24 h to at least 7 days post-transfection (Fig. S1B). The 2 most effective independent siRNA duplexes (FEN1–2 and FEN1–3) were used to demonstrate the specificity of the phenotype, while a FEN1-pool was used to decrease the number of off-target effects that are potentially observed when using a single duplex. The mitotic kinase, PLK1, was included as a positive control as it is an essential mitotic kinase known to decrease cellular proliferation through increased cytotoxicity that is independent of any known SL interaction (28, 29). High- content digital imaging microscopy (HC-DIM) was performed on fixed cells and the total numbers of Hoechst positive cells imaged were calculated (Table S3). The percentages of cells relative to GAPD were determined for each of the conditions, and are presented in Fig. 4A. As anticipated, PLK1 silencing diminished the relative total number of cells significantly, irrespective of RAD54B status. However, visually striking decreases in relative cell numbers were also apparent in RAD54B-deficient cells in which FEN1 expression had been diminished, that were not apparent in RAD54B-proficient cells. Moreover, the large difference in relative cell numbers observed in the RAD54B-deficient cells for the various FEN1 conditions appears to reflect the efficiency of FEN1 knock-down in general (Fig. S1B). For example, the FEN1-pool was visually the most effective down-regulator and exhibited the greatest decrease in relative cell numbers (23.5% ± 10.7% of GAPD-silenced total cell numbers). FEN1–3 however, was visually less efficient at silencing and exhibited a less profound, but still significant, effect on relative cell numbers (47.6% ± 5.5%), while the FEN1–2 knockdown efficiency was intermediate, as was its effect on relative cell numbers (40.5% ± 6.4%). Student's t tests (Table S4) comparing mean relative cell numbers identified highly statistically significant differences (P <0.0001) for PLK1 in both RAD54B-proficient and deficient cells, while highly significant differences were only observed after FEN1 silencing (i.e., FEN1–2, FEN1–3, and FEN1-pool) in RAD54B deficient cells.

Fig. 4.

FEN1 down-regulation underlies synthetic lethality in RAD54B-deficient human cells. (A) Graphical representation of the percentages of cells relative to GAPD knockdown (± SEM) are shown for the isogenic RAD54B^+/+/+ and RAD54B^−/−/− cells treated with the various siRNAs indicated (x-axis). A single representative data series collected in sextuplet and compiled from 1 of 3 experiments is shown (see Table S3). Highly statistically significant differences (P <0.0001) in the mean percentage of cells relative to GAPD knockdown as determined by Student's t- test (see Table S4) are identified (***). (B) Live cell imaging coupled with PI incorporation into dead/dying cells reveals an increase in death in RAD54B deficient (black) cells treated with FEN1 siRNAs versus RAD54B-proficient (gray) cells treated similarly. Five non-overlapping images from each well were collected every 2 h for 48 h and the total number of PI-positive nuclei were scored. All data were normalized to the first time-point (t = 0) to permit easy comparisons between the respective siRNA treatments indicated at the top. Each graph depicts a single representative experiment performed in triplicate and repeated at least once. Note that the relative death index (y-axis) scale is different for the PLK1 positive control.

Because HCT116 cells are MLH1-deficient, there is a possibility that a second-site gene mutation (unrelated to RAD54B knockout) was clonally fixed in the background of the RAD54 knockout cell line that is actually responsible for the SL interaction with FEN1 knockdown. Accordingly, we performed dual RNAi against RAD54B and FEN1 in the parental HCT116 cell line. In an analogous fashion to that described above, diminished RAD54B and FEN1 expression in concert resulted in a highly statistically significant decrease in total cell numbers relative to a GAPD-silenced control (Fig. S2 and Table S5) that is not observed for either of the independent knockdowns (RAD54B-pool or FEN1-pool). The extent of the decrease in total cell numbers for the dual RNAi system was not as great as with the RAD54B knockout cell line, and is likely due to the residual RAD54B and FEN1 expression levels.

To further assess the specificity of the RAD54B/FEN1 synthetic interaction, 12 randomly selected human gene targets (see Table S6) were subjected to silencing in the RAD54B-deficient background. None of the 12 silenced targets produced a statistically significant decrease in overall cell numbers relative to the GAPD-silenced control that was dependent on RAD54B expression status (Fig. S3 and Table S7). These results strongly suggest that diminished FEN1 expression in a RAD54B deficient background decreases cellular proliferation and/or increases cellular cytotoxicity in a manner analogous to the yeast SL interactions described for the homologous yeast gene mutations, rdh54/rad54 and rad27.

Increased Cellular Cytotoxicity Underlies the RAD54B/FEN1 Genetic Interaction.

To determine if an increase in cellular cytotoxicity could account for the decreased cell numbers identified above, HC-DIM was performed on live RAD54B-proficient and deficient cells treated with FEN1 or control siRNAs. Cells were transfected as above, however, medium was supplemented with propidium iodide (PI). Since PI is normally membrane-impermeable, only nuclei with compromised biological membranes (e.g., necrotic or late stage apoptotic cells) will become fluorescently labeled (30). Therefore, an increase in the number of PI-stained nuclei over time was used as a metric for cellular death (30–32). HC-DIM was streamlined by only including FEN1–2 and FEN1-pool as they produce the greatest degree of FEN1 silencing (Fig. S1) and cellular proliferation defects (Fig. 4A). Live cell images were acquired every 2 h for a total of 48 h and the total number of PI-positive nuclei were determined and normalized to 1 at t = 0 h for each treatment. PLK1 was selected as a positive cytotoxicity control (see above) and under these conditions it consistently exhibited a 12- to 19-fold increase in PI-staining nuclei over the course of the experiment (Fig. 4B). In agreement with the fixed HC-DIM presented above, increases in PI-labeled cells over time were readily apparent for RAD54B-deficient cells (3- to 5-fold) treated with FEN1 siRNAs, but not for similarly treated RAD54B-proficient cells. In fact, the FEN1-silenced RAD54B-proficient cells exhibited similar kinetics to those of the negative GAPD controls (<2-fold). Of particular note is the observation that the greatest increases in cytotoxicity generally occur within the first 24 h of imaging (≈24–48 h post-transfection), with only slight increases, or a plateau effect, observed from t = 24 to 48 h.

Discussion

The concept of using model organism genetics (in particular, synthetic lethality screens) to predict evolutionarily conserved proteins that could be targeted to selectively kill cancer cells was first articulated by Hartwell and Friend in 1997 (17). However, over the past decade, very little experimental evidence in favor or against this concept has appeared. In this regard, enhancing the phenotype of a somatic mutation causing genetic instability to a lethal phenotype through the specific down-regulation of a second unlinked gene product predicted from a yeast SL genetic interaction has not previously been reported in human cells. In fact, very few examples of SL in humans have been described (33–35) and were primarily based on characterized biology involving DNA repair pathways rather than cross-species candidate approaches. Here, we report that human RAD54B (18) exhibits a role in maintaining CS in human colorectal cancer cells. We demonstrate that diminished RAD54B expression correlates with increasing DNA content and chromosome numbers. We also show that reexpression of an epitope-tagged version of RAD54B in RAD54B-deficient cells is sufficient to restore DNA content profiles back to those of wild-type RAD54B-proficient cells. Most importantly, we provide mammalian data demonstrating a conserved SL/SGD interaction initially observed in yeast. Specifically, using HC-DIM we demonstrate that a RAD54B-deficient background genetically sensitizes cells to selective killing in combination with diminished FEN1 expression.

In this work, we describe an experimental paradigm using the HCT116 colon cancer cell line for assessing synthetic lethal interactions between candidate CIN genes mutated in tumors and candidate synthetic lethal partner genes predicted from yeast genetic network analysis. Because the HCT116 cell line is MLH1-deficient, we cannot exclude the formal possibility that the synthetic lethality observed when FEN1 and RAD54B are simultaneously knocked down in HCT116 could be dependent on the defective MLH1 allele in the HCT116 background (formally a 3-way synthetic lethal interaction between FEN1, RAD54B, and MLH1). However, because no known synthetic growth defects have been characterized between yeast mlh1 and rad54, rdh54, or rad27, we believe that the increased cellular cytotoxicity observed after FEN1 down-regulation in RAD54B-silenced HCT116 cells (Fig. S2) and in the RAD54B-deficient cells (Fig. 4), results from a 2-way SL interaction (i.e., RAD54B/FEN1) rather than a 3-way interaction (i.e., MLH1/RAD54B/FEN1).

Somatic CIN mutations underlie aberrant chromosome segregation and therefore represent ‘sublethal’ hits on an essential process. Conceivably, if this genetic sensitization could be subsequently exploited by enhancing the ‘sublethal’ phenotype to a ‘lethal’ phenotype, then selective killing could be invoked. Following this logic, any cells (i.e., normal cells) not harboring the genetically sensitizing CIN mutations would be left unaffected or relatively unaffected. Therefore, uncovering the genetic vulnerabilities, or the known CIN mutational spectrum, for a given tumor type could conceivably lead to the identification of potential therapeutic targets through the identification of unlinked gene product (SL/SGD) interaction networks. Conceptually, SL/SGD is of particular interest in the treatment of cancers, because it would only adversely affect tumor cells harboring the primary sensitizing somatic mutations, and leave the normal tissue unaffected.

SL/SGD interactions have been studied extensively in yeast (26, 27, 36, 37) and much data are currently available for yeast genes whose (putative) human orthologs are known to be somatically mutated in human CIN tumors. Although several groups have used rad54 as a gene query (21, 25–27), rdh54 has never been used as a query, but rather has only been identified as a hit (26). Fig. S1 graphically depicts all of the known SL/SGD interactions for both yeast rad54 and rdh54. Of note, only 3 of those characterized SL/SGD interactions are shared in common, namely rad27, pop2, and ccr4. Of those, rad27 is of particular interest because it has a known functional human ortholog, FEN1, which exhibits a high degree of sequence identity (58%) and similarity (73%) and has a BLAST value of e⁻¹⁰⁴. FEN1 is an essential protein that is required for both DNA synthesis and repair. In addition to its flap endonuclease and nick exonuclease activities, it also exhibits gap endonuclease activity. During S-phase, FEN1 processes the 5′ ends of Okazaki fragments in lagging strand synthesis and following DNA damage, it is involved in both base excision and homologous recombination repair where it removes the 5′ overhanging flaps [reviewed in (38, 39)]. Most recently, FEN1 was shown to be involved in telomere stability through its contribution to lagging strand DNA replication at the telomeres which has direct implications on CIN (40). Of particular interest, RAD54B also exhibits roles in DNA repair (22, 41) and CS (this manuscript). Since SL/SGD interactions frequently identify genes whose products impinge on the same essential biological process, we reasoned that diminished FEN1 expression/activity in a RAD54B-deficient background would represent an excellent test candidate.

To determine if SL/SGD interactions are conserved between species and potentially identify a new therapeutic target for cancer therapy, we specifically targeted FEN1 for RNAi- mediated silencing in isogenic RAD54B-proficient and deficient cells. To eliminate the characterization of artifacts arising through off-target effects, 3 independent FEN1 silencing conditions were used—2 independent siRNA duplexes (FEN1–2 and FEN1–3) and a FEN1-pool comprised of 4 independent siRNA duplexes (effectively quartering the concentration of each duplex), that has been shown to greatly diminish off-target effects (42–46). Using fixed and live cell HC-DIM, we showed that diminished FEN1 expression adversely affects overall cell numbers, which presumably occurs through the corresponding increases in cellular cytotoxicity. The underlying reason for the apparent diminishment in relative death from t = 24 to 48 h is currently unknown and under investigation. However, it may simply reflect a decrease in the efficiency of silencing produced by the siRNA duplexes, perhaps by degradation and/or dilution effects. Alternatively, it may signal the presence of a subpopulation of cells in which initial doses of siRNA duplexes were not sufficient to diminish protein expression past a specific required threshold value, or, the existence of a subpopulation of cells that are refractory to siRNA treatment. In any case, FEN1 depletion was demonstrated to significantly enhance cellular cytotoxicity in a synthetic genetic manner analogous to that of rad54/rdh54 and rad27.

The results presented here suggest that in the context of a RAD54B-deficient cancer cell, diminished FEN1 activity through either siRNA-mediated silencing or a small molecule inhibitor could adversely affect the proliferation of cancer cells, while leaving the normal surrounding cells relatively unaffected. Importantly, these data support the conservation in mammalian cells, at least in part, of SL/SGD networks identified in model organisms such as S. cerevisiae. Combining these data with an increased understanding of the mutational spectrum for any CIN tumor type and the continually expanding SL/SGD data emerging from yeast screens (26, 27) could provide critical insights into the identification of therapeutic targets in which human CIN tumors are efficiently and exclusively eliminated through therapeutic intervention. Integration of knowledge among emerging high-throughput datasets in model organisms such as S. cerevisiae and Caenorhabditis elegans, will stimulate new research directions and solutions to current challenges in combating human cancer.

Materials and Methods

Retroviral shRNAs or siRNA pools and independent duplexes were purchased from Open Biosystems or Dharmacon, respectively, and transfected with RNAiMax (Invitrogen). Western blots were conducted on proteins extracted from asynchronous and sub-confluent cells 5 days post-transfection, essentially as described elsewhere (47). Flow cytometry, mitotic chromosome spreads, and microscopy were performed as described in ref. 48. Fixed high-content imaging (HCI) was performed with a Cellomics ArrayScan V HCS Reader equipped with a 20× dry lens. Three days post-transfection all nuclei were stained with Hoechst, and 10 images per well were collected. Total nuclear counts per well were summed and normalized to a GAPD-silenced control. Live cell HCI was performed on a Cellomics KineticScan equipped with a live cell chamber and a 10× dry lens. To visualize dead and/or dying cells, complete growth media was supplemented with PI. All HCI experiments were conducted in sextuplet and repeated at least once. Further details can be found in SI Materials and Methods.