Heightened Sensitivity to Paclitaxel in Class IVa β-Tubulin-transfected Cells Is Lost as Expression Increases*

Stably transfected Chinese hamster ovary cell lines expressing increasing levels of β4a, a class IV neuronal-specific β-tubulin, were compared for effects on microtubule organization, assembly, and sensitivity to antimitotic drugs. It was found that β4a reduced microtubule assembly in proportion to its abundance and thereby caused supersensitivity to microtubule disruptive drugs such as colcemid, vinblastine, and nocodazole. However, the response to paclitaxel was more complex. Low expression of β4a caused supersensitivity to paclitaxel, whereas higher expression resulted in the loss of supersensitivity. The results suggest that β4a may possess an enhanced ability to bind paclitaxel that increases sensitivity to the drug and acts substoichiometrically. At high levels of β4a expression, however, microtubule disruptive effects counteract the assembly promoting pressure exerted by paclitaxel binding, and drug supersensitivity is lost. β4a-Tubulin differs from the more ubiquitous β4b isotype at relatively few amino acid residues, yet β4b expression has little effect on microtubule assembly or drug response. To determine which amino acids mediate the effects of β4a expression, β4a and β4b were altered by site-directed mutagenesis and expressed in Chinese hamster ovary cells. The introduction of N332S or N335S mutations into β4b-tubulin was sufficient to confer microtubule disruption and increased colcemid sensitivity. On the other hand, mutation of Ala115 to serine in β4a-tubulin almost completely reversed heightened sensitivity to paclitaxel, but introduction of an S115A mutation into β4b had no effect, suggesting that a complex interaction of multiple amino acids are necessary to produce this phenotype.

Vertebrate tubulin is encoded by at least six ␣and seven ␤-tubulin genes that produce a highly conserved family of proteins (1). A region of especially high variability among the ␤-tubulins in any given species, however, occurs in the C-terminal 15 amino acids, and these variable sequences are conserved across species. This observation has led to the classification of ␤-tubulin into discreet classes or isotypes (2). Some of these isotypes are expressed in virtually all tissues of an organism, whereas expression of others is tissue-restricted (3). The pres-ence of multiple tubulin genes along with evidence of tissuespecific expression has long fueled speculation that specific tubulin isotypes subserve specific functions (4), but there has been little evidence to support this notion.
Early transfection studies in mammalian cells established that microtubules are composed from all of the available tubulin proteins but provided little evidence of unique roles for any of the examined isotypes (5,6). In vitro studies, however, gave hints that some isotypes of ␤-tubulin may possess unique properties that could potentially alter the behavior of the microtubules into which they incorporate. For example, differences in rates of assembly, dynamic behavior, and drug binding for some of the isotypes have been reported (3). More recently, several laboratories have observed overexpression of specific ␤-tubulin isotypes in cell lines selected for resistance to paclitaxel and have suggested that this overexpression was responsible for the drug resistance phenotype (7).
To explore how tubulin composition might affect microtubule assembly and drug response in living cells, we undertook a direct approach in which we transfected Chinese hamster ovary (CHO) 2 cells with cDNAs encoding various ␤-tubulin isotypes and placed the transcription of the cDNAs under the control of a tetracycline-regulated promoter to limit potential toxicity from their overexpression. In previous work, we reported that overexpression of ␤1-, ␤2-, and ␤4b-tubulin had no effect on microtubule assembly or response of the cells to antimitotic drugs (8). We later reported that overexpression of ␤3-tubulin, a brain-specific isotype, did cause microtubule disruption and weak paclitaxel resistance when expressed at high levels (9) and that overexpression of ␤5-tubulin, a minor ubiquitous isotype, produced major disruption of microtubules and paclitaxel resistance even at moderate levels of expression (10). In contradiction to previous studies linking its overexpression to paclitaxel resistance (11,12), we now report that low expression of ␤4a-tubulin, an isotype normally restricted to brain (13), produces increased paclitaxel sensitivity. Higher expression, however, inhibits microtubule assembly, causes increased sensitivity to microtubule disruptive drugs, and counteracts the paclitaxel-supersensitive phenotype.
Transfection and Selection of Stable Cell Lines-CHO/tTA 6.6a cells, which express the tetracycline-regulated transactivator (16), were maintained in ␣ modification of minimum essential medium (␣MEM; Sigma-Aldrich), supplemented with 5% fetal bovine serum (Atlanta Biologicals, Lawrenceville, GA), 50 units/ml penicillin, and 50 g/ml streptomycin (Invitrogen). Transfections were performed using Lipofectamine (Invitrogen) according to the manufacturer's instructions. After transfection, the cells were trypsinized and seeded into 60-mm dishes containing ␣MEM with 1 g/ml tetracycline (Sigma-Aldrich) and 2 mg/ml G418 (Invitrogen) for the selection of stable cell lines. After 7 days, G418-resistant colonies were isolated and screened by immunofluorescence and Western blots for tetracycline-regulated expression of exogenous tubulin. Stable cell lines were maintained in ␣MEM containing G418 and tetracycline to repress the expression of the cDNAs until the time of analysis.
Electrophoretic Techniques-The cells were grown in 24-well dishes for 24 h with or without tetracycline and lysed in 1% SDS. The proteins were precipitated with 5 volume of acetone, resuspended in SDS loading buffer (50 mM Tris-HCl, pH 6.8, 2% SDS, 5% 2-mercaptoethanol, 10% glycerol), resolved in 8% SDS-PAGE gels, and transferred to nitrocellulose (Pall Life Sciences, Ann Arbor, MI). The membranes were stained with a mixture of TUB 2.1 (1:2,000 dilution; Sigma-Aldrich) that recognizes both endogenous and transfected ␤-tubulin, and actin antibody C4 (1:30,000; Chemicon International Inc., Temecula, CA) as a loading control. Chemiluminescence was carried out by incubation with horseradish peroxidase-conjugated goat anti-mouse IgG (Bethyl Laboratories, Montgomery, TX), followed by SuperSignal chemiluminescence detection (Pierce) and exposure to x-ray film.
Immunofluorescence Microscopy-The cells were grown on sterile glass coverslips for 48 h to ϳ50% confluence. They were briefly washed with PBS and gently lysed in a microtubule-stabilizing buffer (MTB; 20 mM Tris-HCl, pH 6.8, 1 mM MgCl 2 , 2 mM EGTA, 0.5% Triton X-100) containing 4 g/ml paclitaxel (all from Sigma-Aldrich) for 90 s at 4°C and fixed in methanol at Ϫ20°C for at least 30 min. The fixed cells were rehydrated in PBS for 15 min and incubated in affinity purified rabbit HA antibody (Bethyl Laboratories; 1:100 dilution) for 1 h at 37°C followed by Alexa 488-conjugated goat anti-rabbit IgG (Invitrogen, 1:50 dilution) and 1 g/ml 4Ј,6Ј-diamino-2-phenylindole. After washing in PBS, the coverslips were inverted onto 5 l of Gel/Mount (BioMeda Corp., Foster City, CA) and viewed with an Optiphot microscope equipped with a Plan Apochromat 60ϫ, 1.4 numerical aperture oil objective (Nikon Inc., Melville, NY). The images were acquired using a MagnaFire digital camera (Optronics, Goleta, CA).
Measurement of Drug Sensitivity-A colony formation assay was used to determine the sensitivities of stable cell lines to various drugs. Approximately 200 cells were seeded into replicate wells of 24-well dishes containing increasing concentrations of drug in ␣MEM with or without tetracycline and incubated for 7 days until visible colonies appeared. The cells were then stained with 0.5% methylene blue in water for 2 h. The excess dye was washed away, and the dishes were dried. The dye from each well was extracted with 200 l of 1% SDS, and optical density was measured at 630 nm using an Emax plate reader (Molecular Dynamics Inc., Sunnyvale, CA).
Measurement of Steady State Microtubule Assembly-Stable cell lines were plated in triplicate wells of a 24-well dish and incubated in ␣MEM with or without tetracycline for 24 h. The cells were gently washed with PBS and scraped into 200 l of MTB containing 0.14 M NaCl and 4 g/ml paclitaxel. This buffer prevents the depolymerization of microtubules but does not promote further assembly (17). The lysates were transferred to 1.5-ml microcentrifuge tubes, briefly vortexed, and centrifuged at 12,000 ϫ g for 10 min at 4°C. The supernatant fractions containing free tubulin dimers were transferred to a new tube, and the pellet fractions containing microtubules were resuspended in 50 l of water. To dissolve any residual cellular material remaining in the dish, 100 l of 1% SDS was added to each well and transferred to the corresponding pellet fraction. An equal volume of bacterial lysate containing glutathione S-transferase fused to ␣-tubulin (GST-␣) was added to each fraction as an external loading control, and the proteins were precipitated with 5 volumes of acetone. The precipitated proteins were dissolved in 50 l of SDS loading buffer. The proteins were resolved on 8% SDS-PAGE gels, transferred to nitrocellulose, and blotted with a mouse monoclonal antibody to either ␣-tubulin (DM1A; Sigma-Aldrich) or ␤-tubulin (TUB 2.1; Sigma-Aldrich), followed by an Alexa 647-conjugated goat anti-mouse IgG (Invitrogen). The fluorescence emission from Alexa 647 was captured using a STORM 860 imager (Molecular Dynamics Inc.) and quantified with NIH Image analysis software. The percentage of total tubulin polymerized into microtubules was calculated by dividing tubulin in the pellet fraction by the total tubulin in pellet and supernatant fractions after first normalizing each value to the amount of GST-␣ in the corresponding fraction.
Statistics-Individual experiments were repeated three to five times, and the significance of any differences between samples was analyzed using Student's t test.

RESULTS
Isolation of Stable Cell Lines-CHO microtubules are composed of ␣␤ heterodimers containing three ␤-tubulin isotypes (␤1, ␤4b, and ␤5) in a ratio of 75:25:5 (18,19). To assess the effects of incorporating ␤4a-tubulin, stably transfected CHO cell lines expressing an HA-tagged ␤4a cDNA under the control of a tetracycline-regulated promoter were isolated, and four individual cell lines with different expression levels were cho-sen for further study. Each cell line was screened by immunofluorescence microscopy to ensure that at least 95% of the cells in each population were positive for HA␤4a expression. The amount of HA␤4a-tubulin produced in each cell line was determined by Western blot analysis with an antibody that recognizes both endogenous ␤-tubulin and the ectopic HA␤4a-tubulin ( Fig. 1). The cell lines were named clones 1-4 in order of increasing HA␤4a expression, which ranged from 18 to 89% of total ␤-tubulin produced in the cells. In all four cell lines, tetracycline completely repressed the expression of transfected HA␤4a-tubulin (Fig. 1C). An HA␤1 (Fig. 1A) and two HA␤4b cell lines (Fig. 1B) were also isolated to serve as controls in later experiments. Although the ␤4a and ␤4b cDNAs we used were derived from mouse rather than hamster, it should be noted that mammalian species have identical amino acid sequences for these isotypes. Also, we previously showed that the presence of a C-terminal HA tag on ␤-tubulin has no effect on microtubule assembly or the cellular response to antimitotic drugs (16).
HA␤4a Assembles into the Microtubule Network-The ␤4a isotype is not normally found in CHO cells, raising the possibility that it might not assemble with the endogenous tubulin. To test this possibility, we removed tetracycline from the culture medium for 2 days to induce expression and stained the cells with antibodies to the HA tag and to ␣-tubulin. Immunofluorescence microscopy revealed that HA␤4a was produced in essentially all of the cells, and it colocalized with all of the microtubules identified by ␣-tubulin staining (Fig. 2). (Note that only the HA staining is shown in the figure because the ␣-tubulin-stained images were identical.) Cells expressing HA␤1 ( Fig. 2A), HA␤4b (Fig. 2B), or moderate amounts of HA␤4a (i.e. clones 1-3; e.g. Fig. 2C), exhibited normal looking cytoplasmic microtubules and bipolar spindles, suggesting that incorporation of HA␤4a did not grossly affect microtubule organization or spindle morphology. Microtubule function was also likely still intact given that the cells had normal proliferation and produced single symmetrical interphase nuclei (e.g. Fig. 2E). Cells with very high expression of HA␤4a (clone 4), however, displayed an altered morphology in which the cells became much larger than normal (Fig. 2D) and had multi-lobed nuclei with evidence of mis-segregated chromosomes that went on to form small restitution nuclei separated from the main nucleus (Fig. 2F, arrows). These morphological changes are similar to those we and others have previously reported for CHO cells with mitotic defects that cause failures in chromosome segregation and an exit from mitosis without cytokinesis (20 -22). Similar to what was reported in those previous studies, cells with high HA␤4a expression produced multipolar spindles with weak spindle fibers (Fig. 2D, inset). Thus, cells can tolerate the lower expression of HA␤4a found in clones 1-3 (18 -71% of total ␤-tubulin), but the very high expression seen in clone 4 (89% of total ␤-tubulin) interferes with mitosis.
Although immunofluorescence microscopy indicated that HA␤4a assembled into all microtubules, the possibility  remained that it might assemble with a lower efficiency compared with the endogenous tubulin. To test this possibility, we induced expression for 24 h, lysed the cells with a microtubulestabilizing buffer, centrifuged to separate nonassembled tubulin (supernatant) from the polymer (pellet), and compared the relative content of endogenous and ectopic tubulin in the two fractions using Western blots with an antibody that recognizes all the ␤-tubulins. As shown in Fig. 3, the ratio of HA␤4a to endogenous ␤-tubulin was similar in the supernatant and pellet fractions for all of the clones, arguing that the ectopic tubulin assembles into microtubules with the same efficiency as the endogenous tubulin. Thus, at this level of analysis, HA␤4a appears to be used interchangeably with the endogenous tubulin in microtubule assembly. As we reported previously, transfected HA␤4b and HA␤1 isotypes are also used interchangeably as assessed by similar assays (8).
Expression of HA␤4a Inhibits Microtubule Assembly in CHO Cells-Although immunofluorescence did not indicate gross changes in microtubule organization for cells with low to moderate expression, the mitotic defects seen with high overexpression of HA␤4a suggested that the isotype must be having an effect on microtubule properties. To examine this possibility, the extent of cellular microtubule assembly was analyzed by lysing cells with a microtubule-stabilizing buffer and centrifuging the lysate to separate polymerized from nonpolymerized tubulin. Western blot analysis of the two fractions using an antibody to ␣-tubulin is shown in Fig. 4. In agreement with previous studies (9,10,16,17,23,24), ϳ38% of the cellular tubulin was found in the microtubule fraction for both HA␤1and HA␤4b-transfected cells. HA␤4a, however, caused a small but significant reduction in microtubule assembly in an expression-dependent manner, i.e. the percentage of total tubulin in the microtubule fraction decreased progressively from 38% to 26% as the proportion of HA␤4a in the cell increased up to 89%.
CHO Cells Expressing HA␤4a Have Altered Sensitivity to Microtubule Drugs-In extensive previous studies into the mechanisms of drug resistance among CHO cells with mutations in tubulin, we have noted that paclitaxel resistance and increased sensitivity to colcemid are almost always found in cells with reduced tubulin assembly (see Ref. 25 for review). Given the decrease in microtubule assembly that we observed in HA␤4a-expressing cells, we therefore expected them to exhibit paclitaxel resistance, a result that would have confirmed the correlation between paclitaxel resistance and elevated ␤4a expression reported by others (11,12). Instead we found that cells with low levels of HA␤4a expression (e.g. clone 1) were more sensitive to paclitaxel than nontransfected cells ( Fig. 5A and Table 1). Moreover, this increased sensitivity was lost as the level of HA␤4a expression increased (e.g. clones 2 and 3). It is unlikely that the increased sensitivity to paclitaxel exhibited by clone 1 is due to some other random change in the cell line because multiple clones with low expression possessed the phenotype, and the phenotype was lost when the same cells were cultured in the presence of tetracycline to inhibit HA␤4a expression (data not shown). Also, the phenotype was specific to paclitaxel because it did not extend to another microtubule-stabilizing drug, epothilone A (Fig. 5B), even though the two drugs share the same binding site and mechanism of action (26,27).
On the other hand, the response to microtubule-depolymerizing agents was exactly as predicted for cell lines with decreased tubulin assembly, i.e. the cells were ϳ2-fold more sensitive to colcemid and vinblastine (Figs. 5, C and D), even though the two drugs bind to different sites (28,29). Similar results were obtained with two additional microtubule-destabilizing drugs, nocodazole and CI-980, which bind to the colce-  mid site (data not shown). In contrast with the results obtained with HA␤4a, cells overexpressing HA␤1 or HA␤4b had normal sensitivity to the microtubule-depolymerizing drugs colcemid and vinblastine, and they had normal sensitivity to the microtubule-stabilizing drug paclitaxel. HA␤4b-expressing cells did, however, exhibit a slight increase in sensitivity to epothilone A (Table 1). We conclude that HA␤4a expression has a small disruptive effect on microtubules that causes increased sensitivity to drugs that inhibit microtubule assembly. The response to paclitaxel, however, is more complex. Low expression causes a shift in the dose-response curve to the left, i.e. toward higher sensitivity to paclitaxel. We propose that this response is caused by an increased binding affinity of HA␤4a for paclitaxel, a mechanism that is known to act in a substoichiometric manner (30).
As the level of HA␤4a expression increases, however, the microtubule disruptive effect shifts the dose-response curve to the right, i.e. toward paclitaxel resistance, thereby counteracting the supersensitivity caused by enhanced paclitaxel binding.
Site-directed Mutagenesis of ␤4a and ␤4b-In contrast to ␤4a, overexpression of ␤1 and ␤4b did not alter microtubule assembly or affect sensitivity to paclitaxel or colcemid. To identify the amino acid residues in ␤4a that might be important in mediating its unique properties, we reasoned that amino acids in ␤4a, which differed from their counterparts in both ␤4b and ␤1, would be the most likely candidates. The amino acid sequences of all three isotypes are compared in Fig. 6. There are multiple differences in sequence between ␤1 and ␤4b near the C terminus, whereas ␤4a and ␤4b differ by a single amino acid; thus, we discarded this region as one of prime importance.
Within the remaining sequence, we identified five amino acids (highlighted residues in Fig. 6 that span all three sequences) that could potentially account for the differences in behavior between ␤4a and the other two isotypes. Each of these amino acids in HA␤4a was converted, one at a time, to its ␤4b counterpart to determine whether the amino acid was essential for conferring supersensitivity to paclitaxel or colcemid. Similarly, each amino acid in HA␤4b was converted to its ␤4a counterpart to determine whether the change was sufficient to confer supersensitivity to those drugs. Because supersensitivity to paclitaxel was only seen at low levels of expression, we used transfected cell lines with low expression levels to examine this phenotype. For colcemid supersensitivity, however, we used cell lines with higher expression because this phenotype required microtubule disruption that was expressiondependent. The mutations that were analyzed and their locations in the tubulin structure are summarized in Table 2.

TABLE 1 IC 50 (nM) of nontransfected and HA␤4a-and HA␤4b-expressing CHO cells to microtubule drugs
The numbers in parentheses represent fold resistance (positive numbers) or fold sensitivity (negative numbers) compared with nontransfected cells. Following transfection and selection of stable cell lines, immunofluorescence was used to ensure that at least 95% of the cells in the population had similar expression of the transfected mutant cDNA, and Western blot analysis was used to determine the proportion of exogenous and endogenous tubulin for each of the mutant cell lines. IC 50 values for paclitaxel and colcemid were determined, and they are summarized in Table 3. In the case of supersensitivity to paclitaxel, all five amino acid substitutions in HA␤4a produced at least some reversal of the phenotype, implying that a complex interaction among the amino acids is involved, but the amino acid change with the greatest effect in reversing the phenotype was A115S. The fact that no single amino acid substitution was found to be uniquely required for paclitaxel supersensitivity was reinforced by the observation that none of the amino acid substitutions introduced into HA␤4b was sufficient to recreate the phenotype, although N332S did appear to partially increase sensitivity to the drug.

Cell line
The results for colcemid supersensitivity were more straightforward. The A115S mutation in HA␤4a, which completely reversed paclitaxel supersensitivity, had essentially no effect on reversing supersensitivity to colcemid. All other substitutions had very small effects in reversing the phenotype, with only the S335N substitution rising to statistical significance. Thus, no single amino acid change in HA␤4a completely abrogated the ability of that isotype to confer colcemid supersensitivity. How-ever, three amino acid substitutions introduced into HA␤4b (S115A, N332S, and N335S) significantly increased the ability of that isotype to confer supersensitivity to colcemid. Interestingly, the degree of colcemid supersensitivity conferred by each of the latter two substitutions was greater than that observed for the wild-type HA␤4a isotype, suggesting that even though some of the substitutions are capable of causing supersensitivity to colcemid by themselves, their combined effect is somewhat antagonistic in producing the final phenotype.

DISCUSSION
In previous studies we found that most cells selected for resistance to paclitaxel are cross-resistant to other microtubule-stabilizing drugs such as epothilone A, are more sensitive to microtubule-disrupting drugs, and have reduced microtubule assembly. Conversely, the cells selected for resistance to colcemid or vinblastine are cross-resistant to other microtubule-disrupting drugs, are more sensitive to microtubule-stabilizing drugs, and have increased microtubule assembly. Thus, the tubulin mutations causing resistance appear to act by altering microtubule assembly in a direction that opposes the action of the selecting drug (reviewed in Ref. 25). Many of these changes in microtubule assembly are small enough that they do not alter the ability of the mutant cells to proliferate, but some mutations cause larger perturbations in microtubule assembly that interfere with cell division. Examination of a large number of mutants has established that normal proliferation of CHO cells occurs when ϳ22-58% of the total cellular tubulin is assembled into microtubules (31). Outside of this range, spindle function is compromised, chromosomes mis-segregate, and cells fail to complete cytokinesis (22). The effect of these mutations on microtubule assembly and drug resistance is dose-dependent, i.e. the magnitude of the change in microtubule assembly and the degree of drug resistance both increase as the amount of mutant tubulin increases in transfected cells (24).  In related studies we made a number of amino acid substitutions of Leu 215 by site-directed mutagenesis of ␤1-tubulin cDNA and tested for drug sensitivity in transfected cells. Most of the mutations produced phenotypes similar to the paclitaxel resistance mutations already described, i.e. they decreased microtubule assembly, conferred resistance to paclitaxel and epothilone A, and made the cells more sensitive to colcemid. However, one mutation, L215I, behaved differently. This mutation did not affect microtubule assembly, sensitivity to epothilone A, or sensitivity to microtubule-destabilizing drugs. It did, however, confer increased sensitivity to paclitaxel. Moreover, the increased sensitivity was seen in cells with very low expression of L215I mutant tubulin, and it did not change as the level of expression increased (24). We proposed that heightened sensitivity to paclitaxel occurred because the L215I mutation increased the affinity of tubulin for paclitaxel, a mechanism consistent with the absence of observable effects on microtubule assembly or changes in response to drugs other than paclitaxel. Because drugs like paclitaxel are known to affect microtubule assembly at substoichiometric concentrations relative to tubulin (30), this mechanism also explained why effects on paclitaxel sensitivity were seen at low levels of mutant tubulin expression and why those effects did not change as the expression of mutant tubulin increased.
HA␤4a transfected cells appear to have a composite of the two mutant phenotypes. Like tubulin with the L215I mutation, ␤4a may have an increased affinity for paclitaxel that causes cells to exhibit increased sensitivity to the drug at low levels of expression. Superimposed on this, however, ␤4a also acts like the majority of tubulin mutations in that it increasingly inhibits microtubule assembly as its expression increases. Thus, at low expression, there is little if any microtubule disruption, and only the effects of increased paclitaxel affinity are seen, thereby producing cells that have enhanced sensitivity to paclitaxel. As the expression of ␤4a increases, however, there is a progressive decrease in microtubule assembly that produces a progressive increase in sensitivity to colcemid and vinblastine and a corre-sponding progressive resistance to paclitaxel that offsets the effects of enhanced paclitaxel binding.
The decrease in microtubule assembly caused by ␤4a expression is relatively mild and causes no problems in growth until ␤4a accumulates to very high levels (89% of total tubulin). At this level of overexpression, the extent of microtubule assembly drops from 38% (in wild-type cells) to 26% (in HA␤4a clone 4) of total cellular tubulin, a level that is only slightly higher than the 22% at which growth problems are encountered in cells with mutant forms of ␤1-tubulin (see Ref. 31). The reasons for this small difference are unclear, but the toxic end points for ␤4a and mutant ␤1 transfections are similar: defective spindle function that results in mis-segregated chromosomes and a failure of cytokinesis. How the decrease in microtubule assembly produces defects in spindle function is unclear, but it is unlikely to involve alterations in tubulin synthesis. Previous studies have demonstrated that overexpression ␤-tubulin in CHO cells has only minor effects on steady state ␣and ␤-tubulin levels because any overproduced ␤-tubulin that cannot find an ␣-tubulin partner for heterodimer formation is degraded (16). It is also unlikely that toxicity is due to an inherent inability of ␤4a to assemble into microtubules because in vitro studies have indicated that purified ␤4 tubulin subunits actually assemble faster than heterogeneous brain tubulin (32). On the other hand, it has been reported that inhibition of microtubule dynamics can inhibit mitotic spindle function (33). It is therefore possible that particular ratios of tubulin isotypes may cause changes in microtubule structure or alter the binding of microtubule-interacting proteins and thereby affect microtubule dynamics or other behavior essential for spindle function.
HA␤4a and HA␤4b are very similar in amino acid sequence, yet only HA␤4a produces a phenotype when overexpressed. We therefore attempted to identify the amino acid residues that are responsible for differences in their behavior and focused on five residues that are distinct in ␤4a from their counterparts in ␤1 and ␤4b (Fig. 7). For both paclitaxel supersensitivity (because of altered drug binding) and colcemid supersensitivity (because of microtubule disruption), it was not possible to find single amino acid residues that were uniquely responsible for the phenotypes, suggesting that interactions among multiple residues are involved. Some insights, however, were obtained. It was surprising, for example, that changes at N-loop residue 58 had relatively little effect on sensitivity to either paclitaxel or colcemid. Although much of the N-loop (indicated in red in Fig.  7) appears to face the microtubule lumen, at least part of it has been predicted to form lateral interactions with the M-loop of ␤-tubulin (Fig. 7, light blue) in adjacent protofilaments (34). In agreement with this prediction, mutations have been identified in the N-loop that affect microtubule assembly and confer resistance to several antimitotic drugs including paclitaxel and colcemid (23,35,36). The observation that alterations at residue 58 in HA␤4a and HA␤4b had little effect on drug sensitivity implies that this part of the loop may not be involved in forming lateral interactions or that the amino acid substitutions that we made did not sufficiently alter the structure. In contrast, changing residue Ala 115 in ␤4a to its counterpart in ␤4b almost completely reversed the ability of ␤4a to confer supersensitivity to paclitaxel, a phenotype that is likely due to increased binding of the drug. This is again surprising given that this residue is distant from the paclitaxel-binding site (Fig.  7). However, it is known that amino acid 115 resides in the H3 loop, which in turn forms lateral interactions with the M-loop and with the H6-H7 loop of ␤-tubulin in the adjacent protofilament (34). Given that paclitaxel binds preferentially to assembled rather than dimeric tubulin (37), that it binds to an area close to both the M-loop and the H6-H7 loop, and that it likely acts by strengthening lateral associations between protofilaments (38), it is plausible that residue 115 participates in an H3/M-loop or H3/H6-H7 loop interaction that favors pacli-taxel binding. An alternative hypothesis, that the mutation alters tubulin structure in such a way as to make it more susceptible to the action of microtubule-stabilizing drugs, is less likely because HA␤4a-expressing cells exhibited little change in their sensitivity to epothilone A, even though the drug binds to the same site and has a similar mechanism of action to paclitaxel (26,27). Mutagenesis revealed that other residues also appeared to participate in conferring supersensitivity to paclitaxel, but none of them alone, including Ala 115 , was able to recapitulate the phenotype when introduced into ␤4b, suggesting that a complex interaction among the amino acids is required.
Multiple amino acid changes also appear to be needed to completely recapitulate the effects of HA␤4a transfection on cellular response to colcemid. In this case, no single amino acid change in HA␤4a was able to completely reverse its effects in transfected cells. However, two amino acids clearly stood out in their ability to confer supersensitivity to colcemid when introduced into HA␤4b. Introduction of an N332S or N335S substitution caused clear disruption of microtubules and supersensitivity to colcemid that was equal to, or greater than, the effects seen in HA␤4a-transfected cells. The location of residues 332 and 335 in the H10 helix of ␤-tubulin are consistent with these effects (Fig. 7). H10 sits at the intradimer interface between ␣and ␤-tubulin, a position where it can influence the ability of tubulin heterodimers to assume a "straight" assembly enhancing versus a "curved" assembly inhibiting conformation (see Ref. 39 for a review of the role of tubulin conformation in microtubule assembly). S115A and Y159F substitutions in HA␤4b also increased sensitivity to colcemid, but the effects were much smaller in magnitude.
Our results highlight at least two important considerations when assessing the functional significance of structural differences among ␤-tubulin isotypes. First, they point out the danger of forming conclusions from correlations observed in complex systems such as drug-resistant cell lines derived through multiple rounds of selection. Whereas previous studies predicted that expression of ␤4a in non-neuronal cells would increase their resistance to paclitaxel, we showed instead that expression of this isotype increases sensitivity to the drug, most likely through increased binding, and we showed that the increased sensitivity is reversed at higher levels of expression because of a competing mechanism based on microtubule disruption. We further demonstrated that at still higher levels of expression, ␤4a becomes lethal and is therefore incapable of conferring resistance to paclitaxel absent any other cellular alterations. A second lesson is that functional differences among the isotypes may be related to changes anywhere within the coding sequence. This may seem obvious, but most attention has previously been focused on the C-terminal 15 amino acids that have been used to define the isotypic classes of ␤-tubulin. Although there may well be functional significance to those sequences as evidenced by recent reports indicating that they are involved in binding microtubule motor proteins (40,41), we believe that inherent microtubule properties are more likely to be influenced by amino acid differences in other regions that affect the interactions of tubulin subunits in forming microtubule structures. Finally, the results of this study reinforce the idea that some ␤-tubulin isotypes act like "mutant" tubulins that affect microtubule assembly and sensitivity to antimitotic drugs. Based on an alignment of vertebrate ␤-tubulin sequences, Sullivan (1) previously proposed two groups of tubulin: one highly conserved group (outside of the C-terminal 15 amino acids) included ␤1, ␤2, and ␤4; the other more divergent group included ␤3, ␤5, and ␤6. Results from our laboratory showing that changes in expression of members from the second, but not the first, group produce distinctive changes in microtubule assembly and drug resistance support Sullivan's observation, the lone exception being ␤4a, which is very similar to ␤4b yet produces significant effects in transfected CHO cells. Thus, differences in cellular isotype composition can impart subtle to rather dramatic effects on the properties of microtubules, leading to potential functional consequences. Exactly how expression of specific tubulin isotypes subserves microtubule function in any particular cell such as a neuron remains a mystery, but it could potentially involve the need to control microtubule dynamics differently than might be needed for a dividing cell, or it might involve a need for the interaction of brain-specific proteins with the microtubule network. In addition to their potential role in mediating some aspects of cellular behavior, the existence of distinct tubulin isotypes in specific cells opens up an opportunity to exploit microtubule isotype composition for therapeutic intervention.