Attachment Theory and BPD

Bowlby's (1973, 1977, 1980) attachment theory, and the subsequent research it generated, provides a comprehensive developmental perspective for conceptualizing and understanding BPD. Attachment theory offers a cogent theory regarding the development and maintenance of the interpersonal difficulties and adaptations that characterize personality pathology, while simultaneously explaining the concomitant development of self-concept and the problems of self-definition and self-regulation. Embedded in a developmental psychopathology perspective, attachment theory provides a perspective for understanding atypical development in the context of typical development (Cicchetti & Cohen, 1995). Both attachment theory and developmental psychopathology share a common interest in uncovering the developmental course of psychological disorders of childhood and adulthood. Attachment theory, although traditionally regarded as a dilution of psychoanalytic ideas, has developed significantly over the recent years, combining objects relations theory with empirical research in developmental psychopathology. The integration of a developmental psychopathology framework with attachment theory offers a unique window for exploring the development and maintenance of the behaviors, symptoms, and dynamics that characterize borderline pathology.

This article reviews attachment theory and research as a means of providing a developmental psychopathology perspective on BPD. Following a brief review of Bowlby's theory of attachment, and an overview of the evidence with respect to the major claims of attachment theory, I discuss individual differences, the evidence that these differences are rooted in patterns of interaction with caregivers, and how these patterns have important implications for evolving adaptations and development. Following this discussion, I present recent work linking attachment theory and BPD, focusing on the implications for understanding the etiology and treatment of BPD. In conclusion, I address some of the salient issues that point to the direction for future research efforts.

Fundamentals of Attachment Theory

Attachment theory posits that the affective bond that develops between the child and caregiver has consequences for the child's emerging self-concept and developing view of the social world. The theory emerged from John Bowlby's observations of the pervasive disruptive consequences of maternal deprivation in children temporarily separated from their primary caregiver (usually mother) during World War II. His observations suggested that: “the young child's hunger for his mother's love and presence is as great as his hunger for food,” and that her absence inevitably generates “a powerful sense of loss and anger” (Bowlby, 1969, p. xiii).

Based on ethological theory, John Bowlby conceptualized human motivation in terms of “behavioral systems,” and noted that attachment-related behavior in infancy (e.g., clinging, crying, smiling, monitoring caregivers, and developing a preference for a few reliable caregivers or “attachment figures”) is part of an evolution-based functional biological system that increases the likelihood of protection from dangers and predation, and comfort during times of stress. The attachment system also enhances the infant's chances for survival by allowing the immature brain to use the parents' mature functions to organize its own life processes. In fact, the fundamental survival gain of attachment lies not only in eliciting a protective caregiver response, but also in the experience of psychological containment of aversive affect states required for the development of a coherent and symbolizing self (Fonagy, 2001).

Bowlby (1973) proposed that through repeated transactions with their attachment figures, infants form mental representations or affective–cognitive schemata of the self and others and develop expectations about interpersonal relations, which he called “internal working models.” These “internal working mental models” (Bowlby, 1973) are believed to organize personality development, and subsequently direct and shape future relationships. The continuity of these mental models over time is regarded as rooted in the complementary nature of working models of self, other, and concomitant expectations regarding one's role in interpersonal relationships. For example, it is hypothesized that an infant whose needs are typically left unmet may develop a model of others as unreliable and uncaring. Consequently, the neglected infant and child may believe, as an adult, that each new person in his or her life will likewise prove to be inaccessible, uncaring, and unresponsive. Conversely, the child whose needs have been addressed in a consistent loving and supportive manner may subsequently regard others as dependable and trustworthy. Working models are thought to be initially encoded in procedural memory as expectations that help the infant feel secure. These working models are hypothesized to subsequently act as heuristics in relationships, organizing personality development and the regulation of affect. It is thought that parental responses serve both to amplify and reinforce the infant's positive emotional state and attenuate the infant's negative emotional states by giving the infant secure protection when upset. As procedural memories, these early experiences are believed to be often implicit and thus remain unconscious or out of one's immediate awareness.

Based on Bowlby's attachment theory, Mary Ainsworth conducted a seminal study to observe the effects of child rearing techniques and the development of attachment patterns. Ainsworth, Blehar, Waters, and Wall (1978) developed a laboratory procedure called the Strange Situation, which was designed to assess the quality and organization of infant attachment and exploratory behavior in the context of incrementally increasing environmental stress. The Strange Situation consists of a series of infant–caregiver separations and reunions, and the behavior that the infant manifests during the procedure serves as the basis for Ainsworth's attachment classifications. Based on observations of infants and caretakers, Ainsworth et al. (1978) identified three distinct patterns or styles of infant–mother attachment: secure (63% of the dyads tested), avoidant (21%), and anxious–ambivalent (16%).

All three types of infants are attached to their mothers, yet there are significant individual differences in the quality of these attachment relationships, and these differences can be reliably measured. The avoidant dyad is characterized by quiet distance in the mother's presence, often acting unaware of the mother's departure, and avoiding the mother upon reunion. The anxious–ambivalent dyad is characterized by emotional protest and anger on the part of the infant, who becomes extremely distressed upon the mothers' departure, and often continues crying long after the mother returns. These reunions are also characterized by the infant's seeking attention, yet being unable to experience the mother's ministrations as soothing and comforting. The secure dyad is characterized by the confident use of the mother as a “secure base” to explore the playroom with considerable ease and comfort in the mother's presence. Secure infants may experience distress and reduce their exploration upon the mother's departure; however, upon the mother's return, secure babies greet the mother with enthusiasm, accept comfort readily, seek proximity and interaction with the mother, and then resume their exploration of the environment. Later, a fourth category, disorganized–disoriented, was added (Ainsworth & Eichberg, 1991; Hesse & Main, 2000; Main & Solomon, 1986, 1990). The disorganized baby displays disorganized and/or disoriented behaviors in the parent's presence, suggesting a temporary “collapse” of a behavioral strategy. For example, the infant may freeze with a trancelike expression and hands in the air, or may approach the parent but then fall prone and huddled on the floor. Because these behaviors are seen as a temporary collapse of one of the other three attachment patterns, the disorganized classification is not used on its own. Instead, infants classified as disorganized are assigned to either the secure, avoidant or anxious–ambivalent classifications as a secondary attachment pattern.

Bowlby strongly considered attachment as a life-span construct that indelibly influenced human relationships “from the cradle to the grave.” Accordingly, Main, Kaplan, and Cassidy (1985) employed Ainsworth's typology of attachment patterns in the development of the Adult Attachment Interview (AAI), an instrument that assesses aspects of adults' internal working models of attachment with regard to their parents. The AAI is a semistructured interview designed to elicit thoughts, feelings, and memories about early attachment experiences, and to assess the individual's state of mind with regard to early attachment relationships (George, Kaplan, & Main, 1985). Consisting of a set series of 20 questions, the AAI requires the interviewees to reflect on their parents' styles of parenting and how their childhood experiences with their parents influenced their lives. The technique has been described as having the effect of “surprising the unconscious” (George et al., 1985), and allowing numerous opportunities for the interviewee to elaborate upon, contradict, or fail to support previous statements.

The interviews are assigned to one of five primary classifications: secure/autonomous, preoccupied, dismissing, unresolved, or cannot classify (CC). Individuals rated as secure describe the positive and negative aspects of their childhood experiences with their parents in an open, balanced, coherent, and consistent manner. They seem to be thinking afresh while the interview is in progress, at times reflecting on their own thinking process. Secure adults generally have a favorable, realistic, and coherent representation of self and are flexible, realistic, and forgiving in interpersonal relationships. Individuals with attachment styles classified as dismissing devalue the importance of attachment relationships or portray them in an idealized fashion with few corroborating concrete examples. As the interview proceeds, inconsistencies usually emerge between vaguely positive generalizations and “leaked” evidence to the contrary. Individuals classified as preoccupied typically speak about their childhood experiences in a confused, incoherent manner with long, grammatically entangled sentences, use of jargon and nonsense words, reversion to childlike speech, and confusion regarding past and present relationships, all of which convey a lack of distance or perspective. Individuals are classified as having an unresolved (for trauma and loss) attachment style when they show lapses in their monitoring of reasoning or discourse when discussing experiences of loss and abuse. These individuals may speak in a moderately coherent manner, but they make highly implausible statements regarding the causes and consequences of traumatic attachment-related events. Because their interviews may have prominent features of either the secure, dismissing, or preoccupied attachment style, these interviews are given a corresponding secondary classification Interviews that do not fall into one of the above three categories are given a CC rating, signifying a more global breakdown in discourse and inconsistent uses of attachment strategies. The first three categories parallel the parent–child attachment patterns originally identified in childhood (the secure, avoidant, and ambivalent) by Ainsworth et al. (1978). The unresolved for trauma and loss category corresponds to the pattern of disorganized–disoriented attachment later described in infants who had been subjected to maltreatment or to frightened or frightening behaviors on the part of parents with histories of trauma and loss experiences about which they themselves remained unresolved (Hesse & Main, 2000).

In another important development, Bartholomew (1990; Bartholomew & Horowitz, 1991) showed that adult attachment, like infant attachment as conceptualized by Main and Solomon (1990), can best be characterized by four rather than three major categories. Bartholomew's key insight was that Main's prototype of the adult avoidant style (assessed in the context of parenting) is more defensive, denial-oriented, and overtly unemotional than Hazan and Shaver's (1987) avoidant romantic attachment prototype, which seems more vulnerable, conscious of emotional pain, and “fearful.” In Bartholomew's four-category interview and self-report classifications of adult attachment styles, both kinds of avoidance, dismissing and fearful, are included.

With this revision of the Hazan and Shaver classification scheme, it became evident to Bartholomew that the four categories could be arrayed in a two-dimensional space, with one dimension being “model of self” (positive vs. negative) and the other being “model of others” (positive vs. negative; see, e.g., Griffin & Bartholomew, 1994). In other words, Bartholomew conceptualized adult attachment styles in terms of the combinations of representational models of self and others that purportedly underlie them. For secure individuals, models of self and other are both generally positive. For preoccupied or anxious–ambivalent individuals, the model of others is positive (i.e., relationships are attractive) but the model of self is not. For dismissing individuals, the reverse is true: the somewhat defensively maintained model of self is positive, whereas the model of others is not (i.e., intimacy in relationships is regarded with caution or avoided). Fearful individuals have relatively negative models of both self and others.¹

Temperament and Attachment

The influence of temperament on attachment security is controversial (Chess & Thomas, 1982; Kagan, 1982; Lamb et al., 1984), but the balance of the evidence suggests that attachment is independent of temperament (Vaughn & Bost, 1999). Temperament has been defined as a behavioral style under direct biological control, rather than a personality style subject to motivational influences. It is thought to be constitutional and derived from genetic origins; whereas attachment behaviors are regarded as highly influenced by environmental factors (e.g., caregiver sensitivity) and largely independent of genetic influences. Initially temperament theorists interpreted attachment behavior as manifestations of temperament. Behavioral differences between infants classified as avoidant and anxious–resistant during the Strange Situation were viewed as manifesting differences in temperament rather than different interaction histories with caregivers (Kagan, 1982, 1985). It has been argued that anxious–resistant infant behavior reflects the high distress temperament of babies who are simply irritable, difficult, reactive, and/or behaviorally inhibited; whereas avoidant infant behavior in the Strange Situation reflects the low distress temperament of babies who are simply precociously independent. According to theories of temperament, secure infant behavior in the Strange Situation reflects moderate distress and easy temperaments.

Early studies simply related parental reports of their infants' temperaments with Strange Situation behavior, usually finding little association between the two (Bates, Maslin, & Frankel, 1985; Bradshaw, Goldsmith, & Campos, 1987; Egeland & Farber, 1984). Other studies examined the relationship between neonatal ratings of temperament (Belsky & Rovine, 1987; Crockenberg, 1981; Crockenberg & McCluskey, 1986; Waters, Vaughn, & Egeland, 1980) or observed temperament (Belsky, Fish, & Isabella, 1991; Thompson & Lamb, 1984). Taken as a whole, the findings from these studies were generally mixed, methodologically limited, and thus inconclusive. For example, Thompson and Lamb (1984) found an association between observed emotional responsiveness, attachment classification, and Strange Situation behavior. However, contingencies between mothers and infant are forming from birth and are established as early as the first week of life; thus observed emotional responsiveness in the infant is most likely influenced by both constitutional factors and environmental factors. Compounding the problem, there are no pure measures of temperament. Parental reports are prone to bias, observational measures can be confounded by the child's experience of care, and assessments at birth or shortly thereafter often provide only limited assessment of temperament domains (e.g., activity level).

Attachment theorists have noted that infants often show different attachment patterns with each parent, presumably because of different relationship histories (Fox, Kimmerly, & Schafer, 1991; Steele, Steele, & Fonagy, 1996). In addition, studies have shown, consistent with the theory, that caregiver sensitivity was related to attachment patterns. Sroufe and Waters (1977) measured changes in heart rate in children during the Strange Situation. They found that all children, regardless of attachment status, showed accelerated heart rates during the separation phases, which remain elevated until reunion with the parent. At reunion, however, there are important differences. Secure infants exhibited a return to baseline heart rate in less than a minute; whereas both anxious–resistant and avoidant children exhibited sustained heart rate accelerations into the reunion phase, despite clear differences in behavior (in the absence of vigorous motor activity). Upon the caregiver's return, avoidant children displayed very little distress, and instead engaged a toy; however, the accelerated hear rate suggested that they were not actually engaged with the toy but only trying to distract themselves (although apparently unsuccessfully). Sroufe and Waters (1977) have interpreted these findings as an indication that these children have a definite affective response, and are not simply indifferent or precociously independence.

A temperament-based theory of attachment might posit that distress-prone infants would develop anxious–resistant attachments. However, a number of investigators have found fearful and shy infants among secure, anxious–resistant, and avoidant groups (Gunnar, Brodersen, Nachmias, Buss, & Rigatuso, 1996; Gunnar, Mangelsdorf, Larson, & Hertsgaard, 1989; Hertsgaard, Gunnar, Erickson, & Nachmias, 1995; Nachmias, Gunnar, Mangelsdorf, Parritz, & Buss, 1996; Spangler & Schieche, 1994; Stevenson–Hinde & Marshall, 1999; van den Boom, 1989). These findings suggest that there is little evidence that distress-prone infants become anxious–resistant babies.

In addition, Gunnar et al. (1989, 1996; Hertsgaard et al., 1995; Nachmias et al., 1996) in a series of studies found that security of attachment moderates the relationship between fearfulness and stress reactivity. They found that fearful insecure–avoidant and fearful insecure–anxious–resistant infants exhibited elevations in cortisol in response to a variety of attachment-related stressors, whereas secure infants did not. Spangler and Schieche (1994) also reported that fearful insecure avoidant and insecure–anxious–resistant infants exhibited elevations in cortisol in response to the Strange Situation, whereas fearful secure infants did not. These findings suggest attachment security moderates the expression of temperament in stressful situations and thus, is a protective factor against elevated cortisol during stressful situations.

Belsky and Rovine (1987), based on the suggestions of Thompson (Frodi & Thompson, 1985; Thompson & Lamb, 1984), grouped infants according to the level of distress displayed in the Strange Situation. Avoidant (A) infants were grouped with secure attachment subclassifications (B1 and B2) infants and contrasted with a group consisting of anxious–resistant and B3 and B4 babies. Belsky and Rovine (1987) found that the A1 through B2 infants were less difficult and expressed less negative affect than the B3 through anxious–resistant subclassifications (C) infants. This finding suggests that temperamental distress is reflected in attachment behavior but does not necessarily determine whether the child is classified as securely versus insecurely attached. Thus, they noted that low distress characterizes two of the secure attachment subclassifications (B1, B2), just as it does the avoidant subclassifications (A1, A2), whereas high distress characterizes the two other secure attachment subclassifications (B3, B4), just as it does the anxious–resistant subclassifications (C1, C2). However, a number of investigators have failed to replicate this finding (Gunnar et al., 1989; Mangelsdorf et al., 1990; Seifer, Sameroff, Dickstein, Keitner, & Miller, 1996; Vaughn, Lefever, Seifer, & Barglow, 1989) and others have found that these molar temperamental groupings lack stability (Barnett, Ganiban, & Cicchetti, 1999).

Moreover, the findings from a number of intervention studies with irritable babies suggest that attachment patterns are independent of temperament. Crockenberg (1981) found increased rates of secure attachment among babies with irritable temperaments as a function of maternal social support. van den Boom (1994, 1995) showed that a brief therapeutic intervention with mothers and their irritable babies resulted in dramatic and enduring increases in secure attachment.

In sum, examining the role of temperament and its association with attachment has been slowed because temperament has been difficult to specify in humans, and there is still no agreement on a conceptual framework for characterizing human temperament. The current consensus based on research suggests that temperament may influence the expression of attachment, but that attachment and temperament is not the same thing (Vaughn & Bost, 1999). That is, attachment behaviors are not merely the manifestations of temperament; however, temperament likely influences the expression of attachment patterns (Belsky & Rovine, 1987; Frodi & Thompson, 1985).

Although temperament does not appear to directly influence attachment security, it may interact with attachment security to increase the risk for BPD. A number of prominent psychological theories of BPD hypothesize the interaction between temperament and environment in the formation of the disorder (Bateman & Fonagy, 2004 Kernberg, 1984; Linehan, 1993). For instance, Kernberg (1984) argues that high levels of constitutional aggression interfere with normative developmental processes of integrating disparate representations. Instead, the high levels of aggression result in a division between positive and negative representations. In attachment terms, high levels of constitutional aggression may result in multiple contradictory internal working models of self and others. Likewise, Gurvits et al. (2000) point out that affective instability may interfere with the ability to develop stable perceptions of self and others. They note that both the specific role of aggression and the more general role of affect lability may make the developmental task of integrating stable representations of self and others more difficult to accomplish. There are a number of important conceptualizations of temperament processes in the development of BPD that suggest the importance of understanding the intersection of attachment and constitutional factors (Depue & Lenzenweger, 2001; Derryberry & Rothbart, 1988; Posner et al., 2003).

Implications of Attachment for BPD

Psychopathology researchers and theorists have begun to understand fundamental aspects of BPD such as unstable, intense interpersonal relationships, feelings of emptiness, bursts of rage, chronic fears of abandonment and intolerance for aloneness, and lack of a stable sense of self as stemming from impairments in the underlying attachment organization (Blatt, Auerbach, & Levy, 1997; Fonagy et al., 1996; Gunderson, 1996; Levy & Blatt, 1999; Yeomans & Levy, 2002). These investigators have noted that the impulsivity, affective lability, and self-damaging actions that are the hallmark of borderline personality occur in an interpersonal context and are often precipitated by real or imagined events in relationships (benign separations may be perceived as rejection, bids for intimacy may be seen as intrusive or engulfing, differences of opinion may be seen as personal attacks). For example, mood lability in BPD patients is often triggered by the misperception of subtle events in the environment (Gurvits et al., 2000; Yeomans & Levy, 2002). Once the mood state is obtained, it can rapidly lead to aggressive, impulsive, self-destructive, interpersonally intrusive, or extremely isolative behavior (Gurvits et al., 2000; Yeomans & Levy, 2002).

These investigators have begun examining the clinical applications of attachment theory both theoretically (Blatt & Levy, 2003; Bowlby, 1988; Diamond, Clarkin, Levine, Levy, et al., 1999; Fonagy, Gergely, Jurist, & Target, 2002; Gunderson, 1996; Holmes, 1996; Levy & Blatt, 1999) and empirically (Dozier, 1990; Dozier, Cue, & Barnett, 1994; Fonagy et al., 1996; Levy et al., in press; Tyrrell, Dozier, Teague, & Fallot, 1999). These authors have begun to delineate how attachment classifications and dimensions contribute to understanding the underlying psychopathology and the quality and nature of the therapeutic alliance, psychotherapy process, patterns of transference and countertransference, and psychotherapy outcome.

From its inception, Bowlby conceptualized attachment theory in both normal and psychopathological development. Bowlby (1973) believed that attachment difficulties increase vulnerability to psychopathology, and can help identify the specific types of difficulties that arise. Bowlby (1977) contended that internal working models of attachment help explain “the many forms of emotional distress and personality disturbances, including anxiety, anger, depression, and emotional detachment, to which unwilling separations and loss give rise” (p. 201). He held that childhood attachment underlies the “later capacity to make affectional bonds as well as a whole range of adult dysfunctions” including “marital problems and trouble with children as well as … neurotic symptoms and personality disorders” (1977, p. 206). Bowlby postulated that insecure attachment lies at the center of disordered personality traits, and he actually tied the overt expression of felt insecurity to specific character disorders. For instance, Bowlby connected anxious ambivalent attachment to “a tendency to make excessive demands on others and to be anxious and clingy when they are not met, such as is present in dependent and hysterical personalities,” and avoidant attachment to “a blockage in the capacity to make deep relationships, such as is present in affectionless and psychopathic personalities” (1973, p. 14). Avoidant attachment, Bowlby postulated, results from the individual constantly being rebuffed in his or her appraisals for comfort or protection, and “may later be diagnosed a narcissistic” (1973, p. 124). Thus, Bowlby not only postulated that early attachment experiences have long-lasting effects that tend to persist across the life span, but are among the major determinates of personality organization and pathology.

The Association of Adult Attachment and BPD

To date, nine studies have examined attachment patterns using the AAI in study groups with patients diagnosed with BPD or with identifiable borderline patients (Babcock, Jacobson, Gottman, & Yerington, 2000; Barone, 2003; Diamond, Stovall–McClough, Clarkin, & Levy, 2003; Fonagy et al., 1996; Levy et al., in press; Patrick et al., 1994; Rosenstein & Horowitz, 1996; Stalker & Davies, 1995; Stovall–McClough & Cloitre, 2003; van IJzendoorn et al., 1997). A few others have examined the relationship between BPD and attachment using AAI-like interviews or methods (Buchheim, George, & Walter, 2003; Salzman, Salzman, & Wolfson, 1997). Two studies have examined attachment using the AAI in study groups of patients with personality disorders, where subgroups of borderline patients have not been reported (Levinson & Fonagy, 2004). A number of studies have examined the relationship between self-reported attachment patterns and BPD (Alexander, 1993; Bender, Farber, & Geller, 2001; Brennan & Shaver, 1998; Dutton, Saunders, Starzomski, & Bartholomew, 1994; Fossati et al., 2001, 2003; Levy, 1993; Levy, Meehan, Weber, Reynoso, & Clarkin, 2005; Nickell, Waudby, & Trull, 2002; Sack, Sperling, Fagen, & Foelsch, 1996; Sperling, Sharp, & Fishler, 1991; West, Keller, Links, & Patrick, 1993). Two studies have examined the relationship between attachment and BPD using clinician-based ratings of patient attachment (Meyer, Pilkonis, Proietti, Heape, & Egan, 2001; Nakash–Eisikovits, Dutra, & Westen, (2002). In a particularly creative design, Handley and Swenson (1989) carried out a case study for a 27-year-old hospitalized woman diagnosed with BPD. They used an observational scheme, which parallels the Strange Situation, and an “acting out” scale to examine the effects of separations from her therapist. These studies, reviewed below in greater detail, have generally found that BPDs and traits are significantly associated with fearful avoidant and preoccupied attachment.

Summary of Findings

Consistent with conceptualizations of BPD patients as insecurely attached, secure attachment in this group is extremely low, especially compared to other groups. Across interview measures, secure attachment ranges from 0 to 30%, usually around 6 to 8%. Across the self-report measures, the rates of secure attachment are also low; although rates have been higher in nonclinical samples (e.g., Brennan & Shaver, 1998). All studies found an inverse relationship between scores on borderline dimensions and secure attachment. Although early interview studies suggested a strong relationship between BPD and preoccupied and unresolved attachments, especially angry preoccupied and fearfully preoccupied, later studies suggests that BPD is not specifically related to one type of attachment pattern. Although most studies show elevated rates of unresolved attachment, rates have ranged from as low as 35%, and are generally in the 50% range, a range that is not much different than other psychiatric disorders (Buchheim, Strauss, & Kachele, 2002; Carlson, Cicchetti, Barnett, & Braunwald, 1989; Espinosa, Beckwith, Howard, Tyler, & Swanson, 2001; Fonagy et al., 1996; Lyons–Ruth, Connell, Grunebaum, & Botein, 1990; Manassis et al., 1994; O'Connor, Sigman, & Brill, 1987; Radke–Yarrow, Cummings, Kuczynski, & Chapman, 1985; Rodning, Beckwith, & Howard, 1991; Seifer, Schiller, Sameroff, Resnick, & Riordan, 1996; Stovall–McClough & Cloitre, 2001; Teti, Nakagawa, Das, & Wirth, 1991; Wallis & Steele, 2001). Additionally, a number of interview studies now suggest that many BPD patients can be diagnosed with dismissing attachment (Barone, 2003; Diamond et al., 2003; Levy et al., in press; Rosenstein & Horowitz, 1996; Salzman, 1988; Stalker & Davies, 1995). Some more recent studies (Diamond et al., 2003; Levy et al., in press; Stalker & Davies, 1995) have also employed the CC category (Hesse, 1996), and this category is also common, ranging in about the 30% range, and this category may be very important for understanding BPD.

With regard to studies using self-report measures, studies have consistently found that borderline personality traits are significantly negatively correlated with attachment security and significantly positively correlated with both fearful avoidant and preoccupied attachment.

Implications of Findings on the Association of Attachment and BPD

The fact that BPD appears not to be specifically related to a particular attachment pattern, combined with the fact that many individuals from nonclinical and normal samples also can be classified in these same patterns, suggests that there may be a range of functioning within each attachment pattern. Consistent with this idea, Levy and Blatt (1999), integrating Blatt's (1995) cognitive–developmental psychoanalytic theory with attachment theory, proposed that within each attachment pattern, there may exist more and less adaptive forms of dismissing and preoccupied attachment. These developmental levels are based on the degree of differentiation and integration of representational or working models that underlie attachment patterns. In terms of PDs, Levy and Blatt (1999) noted that several PDs (i.e., histrionic, dependent, borderline) appear to be focused in different ways, and possibly at different developmental levels, on issues of interpersonal relatedness. They proposed that preoccupied attachment would run along a relatedness continuum from non-BPD individuals who value attachment, intimacy, and closeness to the gregarious who may exaggerate relatedness, to those with a hysterical style, who not only exaggerate closeness and overly value others but also may defend against ideas inconsistent with their desires, to more histrionic individuals, who are overly dependent and easily show anger in attachment relationships, to those with BPD. In contrast, another set of PDs (i.e., avoidant, paranoid, obsessive–compulsive, narcissistic) appear to express a preoccupation with establishing, preserving, and maintaining a sense of self, possibly in different ways and at different developmental levels. Levy and Blatt (1999) proposed that avoidant attachment would run along a self-definitional continuum from individuals without personality disorders who are striving for personal development to those who are more obsessive, to those with avoidant PD, to those with narcissistic PD, and finally, at the lowest developmental levels, to those with BPD and antisocial PD.

This integration allows us to note that the two primary types of insecure attachment, avoidant and anxious–preoccupied, can occur at several developmental levels. Differences in the content and structure of mental representations (or internal working models) distinguishes more and less adaptive forms of avoidant and anxious–preoccupied attachment, thereby bringing a fuller developmental perspective to the study of attachment patterns. Different patterns of attachment not only involve differences in the content of internal working models but also differences in the structure of those models (e.g., degree of differentiation and integration). It may be the structure of these models, more so than the content, that results in different capacities and potentials for adaptation. Thus, within specific attachment styles, internal working models may vary in the degree of differentiation, integration, and internalization (Levy et al., 1998).

Association Between BPD and Unresolved/Disorganized Status

Association Between BPD and the CC Category

In addition to, or instead of, being unresolved for trauma and loss, it is likely that many patients with BPD would be characterized by mental states suggesting the CC status with respect to attachment. Like those categorized as CC, patients with BPD exhibit multiple, contradictory, incompatible, and unintegrated working models, often leading to chaotic and mood-dependent behavior in interpersonal relationships. Hesse (1999) has suggested that those classified in the CC category show a global breakdown of coherent discourse about attachment experiences, whereas individuals classified as dismissing or preoccupied display an insecure but systematic or organized strategy toward attachment experiences. Unresolved individuals show only a local breakdown in the discourse on loss or trauma. Van IJzendoorn (1992) found that 8 of the 11 CC subjects were diagnosed with a PD. Many of the studies examining BPD have not used the CC category (Barone, 2003; Fonagy et al., 1996; Patrick et al., 1994; Rosenstein & Horowitz, 1996; Stovall–McClough & Cloitre, 2003), although two recent studies have (Diamond et al., 2003; Levy et al., in press) found that about 25% of patients met criteria for CC classification and that these individuals were at increased risk for dropping out of treatment. Thus, in addition to showing incoherence of internalized representations of self and others, many BPD patients also show an unintegrated mixture of approach/activating and avoidance/deactivating strategies with regard to attachment relationships.

Developmental Research Related to Attachment and BPD

Studies of Families of Patients With BPD

In several studies, patients with BPD report that their parents were neglectful, uncaring, underinvolved, and had serious psychopathology, including depression and alcoholism (Goldberg, Mann, Wise, & Segall, 1985; Grinker, Werble, & Drye, 1968; Gunderson, Kerr, & Englund, 1980; Links, 1990; Patrick et al., 1994; Trull, 2001a, 2001b; Walsh, 1977; Zanarini, Gunderson, Marino, Schwartz, & Frankenburg, 1989; Zweig–Frank & Paris, 1991); however, in all these studies, data was collected retrospectively through self-report; therefore, recall bias cannot be ruled out.

There have been few studies of the risk of psychopathology in children of mothers with BPD (Crandell, Patrick, & Hobson, 2003; DeMulder, Tarulla, Klimes–Dougan, Free, & Radke–Yarrow, 1995; Espinosa et al. 2001; Hobson et al., 2005; Rutter & Quinton, 1984; Weiss et al., 1996). Feldman et al. (1995) examined the families of mothers with BPD as compared to families of mothers with other PDs. They found that the families of BPD mothers were more unstable, less cohesive and organized, but not less conflicted. Feldman and colleagues suggests that low family cohesion and high instability may affect the parenting capacities of the mother with BPD and place her children at increased risk for their own maladaptive outcomes. This conclusion is consistent with Rutter and Quinton's (1984) findings. They examined the effects of a PD diagnoses on parenting behavior among women with affective disorders. They found that the presence of PDs was an important prognostic indicator of both the parents functioning and the child's behavioral outcome.

A couple of studies had mothers of BPD adolescents also report on their parenting (Golomb et al., 1994; Guttman & Laporte, 2000). Golomb et al. (1994) compared 13 mothers of adolescents with borderline personality with 13 mothers of nonclinical adolescents using an interview measure of maternal empathy. They found that the mothers of the borderline patients provided responses that were blindly coded as less empathic, more egocentric, less differentiated from their daughters', and showing that they view their daughters in “need gratifying” ways. The authors note that the mothers of BPD daughters also reported more environmental stressors, which likely affected their daughters directly and affected their capacity to parent effectively and empathically.

Guttman and Laporte (2000) examined empathy in the families of 27 women with BPD, 28 women with anorexia nervosa, and 27 women without a clinical diagnosis. The daughters and both parents completed questionnaires and interview measures of empathy. BPD patients scored significantly higher than the other two groups on immature empathy and lower on mature empathy. The parents of the BPD patients scored significantly lower on all measures of empathy. On the interview measures, BPD patients and their parents agreed about the relative absence of empathic parenting.

There are a number of recent studies that have examined caregiving behavior directly and prospectively. DeMulder et al. (1995) examined maternal caregiving in affectively ill mothers with comorbid PDs. Affectively ill mothers reported more PD symptoms than did well mothers, and severity of illness was related to higher rates of PD symptomatology. Although PD symptoms were generally related to less engagement and involvement, BPD symptoms were significantly related to increased engagement among bipolar mothers, and there was a trend for decreased engagement among depressed mothers. BPD symptoms were not related to irritability or criticalness for either bipolar or depressed mothers.

Crandell et al. (2003), using Tronick's Still Face procedure, compared mothers with BPD to nonclinical controls. They found that the BPD mothers demonstrated more intrusiveness and insensitivity toward their 2-month-old infants. Similar to adults with BPD, the infants showed less positive affect during and after the Still Face procedure, made more bids to their mother but looked away from mothers more often. In addition, the infants of BPD mothers were more likely to display a dazed look that includes freezing of facial movements and eye contact similar to the freezing that is found in the disorganized attachment pattern in the Strange Situation test. The infants of BPD mothers also showed less positive affect with a stranger afterwards, indicating a possible carryover effect from the maternal interactions. Hobson et al. (2005) found similar findings with BPD mothers when their infants were 12 months old. The authors compared 10 infants of borderline mothers with 22 infants of mothers completely free of any current and past psychopathology. Consistent with predictions, infants of borderline mothers were more likely to be classified as disorganized, were less sociable, and the mothers with BPD were more likely to engage in intrusive behaviors with their infants. Hobson et al. found that 8 out of 10 infants of mothers with BPD were classified as having a disorganized attachment. Interestingly, of the eight disorganized babies, four were assigned a secondary classification of securely attached and the two nondisorganized infants were classified as securely attached. The mothers of the secure nondisorganized babies scored even more deviant than the borderline or control disorganized mothers. These findings generate a number of issues for further study. For example, we have to consider that a subgroup of disturbed infants may not be able to be identified by their strange situation behaviors or that there may be some unrecognized infant behaviors that mimic security, but represent a pseudosecurity.

Weiss et al. (1996) examined psychopathology in the offspring of mothers with BPD. Weiss et al. (1996) examined 21 children of BPD mothers compared with 23 children of non-BPD mothers. The children of BPD mothers had more psychiatric diagnoses, more impulse control disorders, a higher frequency of child BPD, and lower global functioning scores. A general conclusion from these studies is that offspring of BPD mothers are at high risk for psychopathology.

As mentioned earlier, Lyons–Ruth et al. (2005) followed 56 infants and their mothers for 18 years in a longitudinal study examining the development of BPD symptomatology. Mothers of children who later displayed BPD features were significantly more disrupted in their affective communication, particularly through withdrawal. Infant disorganization was unrelated to later BPD features. Later abuse during adolescence did not mediate the relationship between early referral status and later BPD features, although unresolved for that abuse may predict BPD features.

Macfie, Rivas, Engle, Hamilton, and Rathjen (2005) compared 10 children of BPD mothers with 10 mothers without BPD on measures on narrative representation. They found that compared to the children of non-BPD mothers, the narratives of children of BPD mothers were significantly more negative and less positive of mother, had more intrusions of traumatic material, and showed significantly more fear of abandonment.

Danon and Graignic (2003) found that borderline mothers showed an incapacity of modulating behavior after the Still Face episode, therefore heightening noncontingency during reunion play. Infants of borderline mothers displayed more negative emotional expressions, heightened autonomic behaviors before the Still Face, less self-comfort behavior, and more distancing behavior. They tended to respond to situations that require self-soothing by hiccupping and spitting, and using self clasping and touch behavior; whereas, infants of control mothers would yawn, or engage in sucking behavior. Danon and Graignic concluded that BPD mothers seemed less aware of infants needs for emotional regulation both before and after Still Face. Their speech to the babies was characterized as more descriptive than engaging and resulted in heightened noncontingency during reunion play.

In summary, there is substantial evidence to suggest that pathways to BPD may involve heightened risk from chaotic family life, increased stressors on parents, noncontingent interactions, and disrupted communication between caregivers and their children. Of course, these risk factors most likely interact with temperamental expressions of genetic predispositions toward impulsivity, negative affect, and perhaps aggression. The relationship between constitutional factors and environmental factors is probably in relation to each other. Thus, the higher the constitutional disadvantage, the lower the threshold for environmental perturbations to overwhelm the child's capacity to assimilate and accommodate to his or her environment. Conversely, a child who has a low constitutional load may be resilient to greater perturbations. In addition, family stressors may affect the developing child directly and through the effects on caregivers. Finally, the development of multiple, contradictory, and unintegrated internal working models of self and attachment figures may leave these children vulnerable to life's stressors and traumatic experiences.

Recommendations for Future Research

Although borderline personality is one of the most serious psychiatric disorders, relatively little is know about its developmental precursors. Establishing childhood precursors and identifying etiological factors related to the development of borderline personality is a significant research problem. Longitudinal prospective research designs are critical in this regard because they have the potential to provide vital information concerning developmental adaptation and potential causal relationships. There are a number of existing samples, both low and high risk, that have followed infants into young adulthood (Hamilton, 2000; Lyons–Ruth et al., 2005; Main, 1999; Waters, Hamilton, & Weinfield, 2000; Weinfield, Sroufe, & Egeland, 2000). These samples are well characterized in terms of infant attachment and environments. There are other samples of children followed into adulthood that may be relevant for examining precursors of BPD (Shedler & Block, 1990). However, these studies are limited by not being specifically designed to examine BPD. These studies generally have sufficiently large enough numbers of participants to address the questions initially posed in the investigators; however, given the base rate of BPD, these samples may be too small to individually address questions about precursors of BPD. For example, in the Minnesota Longitudinal Study of Children and Parents, only four (2%) of 175 participants met criteria for BPD. One recommendation would include conducting a multisite follow-up of these infants into young adulthood. Constructs that are common across samples, but identified through previous research as pertinent developmental variables, could be assessed at various time points. For example, assessments during the infant assessment can specify temperament, disorganized attachment, atypical caregiver behaviors, caregiver hostile–helpless mental states, and caregiver attachment status, and so forth. Assessments carried out during childhood and adolescence can specify concrete stressors such as parental loss, divorce, adolescent abuse, and substance use. Finally, important adult outcome variables could be measured (BPD symptomatology, love and work functioning, attachment status, unresolved attachment, helpless–hostile states of mind, RF).

Another strategy suggested by Gunderson and Zanarini (1989) and very consistent with a developmental psychopathology perspective is to investigate children exposed to known adversities for BPD over time. These adversities, identified in both retrospective and prospective research, would include the experience of sexual abuse or other traumas, parental impulsive spectrum disorders such as substance abuse, parental recent loss or trauma, domestic violence, depressed mothers, and mothers with BPD. This approach could be broadened to include following up children who display childhood parallels of borderline symptoms. For example, depressed and/or anxious children have been shown to be at risk for developing BPD (Kassen, Cohen, Skodol, Johnson, Smailes, & Brook, 2001). Levy and Clarkin (2005) have found that one-third of their BPD patients engaged in self-injurious behaviors before age 12.

In conclusion, attachment theory provides a useful approach within the developmental psychopathology perspective for conceptualizing BPD. Within attachment theory, BPD is viewed as resulting from a series of successive interactional processes along a developmental path. Although new experiences influence the individual, these later experiences are not independent of preexisting representations but are understood within the context these models. Distorted and impaired representational models may result when perturbations overwhelm the child's organizational capacities to accommodate the experience, thereby compromising the child's development of representational structures (Blatt, 1995). These representational processes interact with constitutional factors that may increase risk; therefore, it has become central to identify how these experiences interact with constitutional factors. These compromised representations can result in multiple, contradictory, and unintegrated representations, which in those patients with BPD, typically oscillate quickly and result in an unintegrated view of self and others and extremely chaotic behavior.

Attachment research has made important advances in the understanding of BPD; however, many questions regarding specific mechanisms for the development of BPD remain unanswered. Nevertheless, an attachment theoretical perspective within a developmental psychopathology framework appears to be a powerful approach to understanding the mechanisms underlying both the interpersonal and intrapersonal difficulties characteristic of BPD.